Co-sponsors

978-93-82288-63-3 Isbn: Online Isbn:978-93-82288-54-1 Print

Proceedings of International Conference on Advances in Computers, Communication and Electronic Engineering

16 - 18 march 2015 Department of Electronics and Instrumentation Technology तमसो मा ज्योित

UN

IVE

RSIT

‫ﻟﻨﻮر‬ ‫ا‬

‫ﻣﻦ اﻟﻈﻠ ٰﻤﺖ‬ ‫اﻟﯽ‬

य गम

Y OF KAS

HM

IR

University of Kashmir, Srinagar, J & K

COMMUNE - 2015

University of Kashmir

Proceedings of 2015 International Conference on

Advances in Computers, Communication, and Electronic Engineering

ISBN (Online): 978-93-82288-63-3 ISBN (Print): 978-93-82288-54-1 Publisher: University of Kashmir, Hazratbal, Srinagar, 190 006,

J&K, India. Publication Date: 16 March, 2015 Editor: Dr. Mohammad Tariq Banday Copyright Notice: © All rights are reserved by the Department of Electronics and Instrumentation Technology, University of Kashmir, Hazratbal, Srinagar, 190 006, J&K, India.

University of Kashmir

Foreword The Department of Electronics and Instrumentation Technology had the privilege of organizing the 2015 International Conference on Advances in Computers, Communication, and Electronic Engineering (COMMUNE-2015), an erudite gathering of learned academicians, scholars, and students from across the globe. This is to acknowledge the determination and efforts made by the faculty, research scholars, students, and administrative staff of the department in organizing such a huge gathering that carried an aura of international scope. The idea to convene COMMUNE-2015 was conceived by the dedicated and young faculty members in the department who always stand up to the occasion for undertaking all sorts of academic and research activities. The title of the conference was selected in a way to ensure participation of academicians and researchers from a number of academic disciplines. To target the anticipated academicians and scholars globally, a dynamic website solely prepared for this purpose was created. The response we received across globe reflected the twin successes, both with regard to the selection of the theme of the conference and the wide publicity it received. Absolute professionalism was maintained in the selection, evaluation, and review process of the articles and research papers. The international academic standards were complemented by the local traditions of hospitality to provide the visiting delegates an ambience of camaraderie and warmth. As the head of the varsity, I certainly feel proud to be part of such an excellent academic activity. Being a first of its kind to be organized by the department of Electronics and Instrumentation Technology, in particular and University of Kashmir in general, COMMUNE2015 will go a long way in shaping the future policy of the Institution in organizing seminars and conferences. It was in this spirit that the organizers are determined to turn COMMUNE into an annual activity. The fruition of COMMUNE-2015 in the form of these proceedings has been prepared with diligence as per the standard norms in academic publishing. Great efforts were put in to prepare these proceedings in time to allow the visiting delegates to take these along at the culmination of the conference. Along with authors, the names of respected reviewers have also been included in the proceedings as a mark of compliment to their efforts. I hope these proceedings will surely serve an excellent reference book for scholars and researchers of various denominations around the globe. I wish all the participants a great time ahead.

Prof. Khurshid Iqbal Andrabi (Honourable Vice-Chancellor, University of Kashmir) Chief Patron, COMMUNE-2015

2015

iii

University of Kashmir

Foreword The 2015 International Conference on Advances in Computers, Communication, and Electronic Engineering (COMMUNE-2015) organized by the Department of Electronics and Instrumentation Technology from 16th to 18th March is a landmark in University of Kashmir’s tradition of bringing together academicians, scholars and learned men from across the world to discuss and deliberate on variegated issues of knowledge. With a sense of satisfaction, I acknowledge the efforts put in, and the enthusiasm shown by various stakeholders of the varsity to organize a conference of such a magnitude. The conference has been conceived with the idea to underscore the scientific information interchange between researchers, developers, engineers, students, and fellow citizens working in and around the world. Through presentations, special talks, panel discussions and networking, COMMUNE-2015 provided an excellent avenue for budding researchers and academicians to discuss with and learn from the established academic community in the field and infact served a motivation for scholars of various denominations to approach the problems at hand from an interdisciplinary perspective. In their pledge to make COMMUNE an annual affair, I promise to extend all kinds of support to the organizers. With a broader aim to promote research and developmental activities in Electronics, Computer Science, and Communication Engineering, I have been given to understand that the conference has attracted the attention of academicians, researchers and learned scholars from numerous institutions, research centres and universities within and outside country. Around 200 quality research papers addressing the pressing issues in the field of Electronics, Computer Science and Communication Engineering, were received by the organizers. With serious efforts, these were peer reviewed and evaluated based on originality, technical and/or research content/depth, correctness, relevance to conference, contributions, and readability. The papers presented and included in the proceedings cover multiple themes and ideas that are currently being addressed world over in the field of Electronics, Computer Sciences, and Communication Engineering. I believe that the proceedings will serve an excellent reference book for the scientific community and certainly will stimulate endeavors of further research.

Prof. Mohamad Ashraf Wani (Dean Academic Affairs, University of Kashmir) Patron, COMMUNE-2015

2015

v

University of Kashmir

Foreword With a sense of accomplishment, I write this foreword for the proceedings of the 2015 International Conference on Advances in Computers, Communication, and Electronic Engineering (COMMUNE-2015) organized by the Department of Electronics and Instrumentation Technology from 16th to 18th March. Seeking academic excellence has been the benchmark of University of Kashmir since its inception and concerted efforts are always being made to organize scholarly interactions in the form of seminars and conferences. COMMUNE2015 has been a right step in that direction and credit must go to its organizers whose hard work, sincerity, and perseverance made it a successful endeavor. The idea underlying COMMUNE-2015 has been to create a platform for academicians, researchers, and scientists from across the globe to present their work on the wide variety of modern day issues in the field of Electronics, Computer Sciences, and Communication Engineering. The presentations, discussions, and talks delivered during the conference were also aimed to motivate the budding research scholars and academicians on their threshold, to imbibe the scientific temperament displayed by men of knowledge present in the institution during these days. COMMUNE-2015 was successful in creating a community of reputed scientists and scholars, upcoming academicians, research scholars and students who have vowed to make it an annual activity. COMMUNE-2015 received wide publicity, especially through the websites exclusively dedicated to it and as such, the conference attracted scholars, scientists, and academicians from all over the globe. In addition, given the theme of the conference that encompasses a host of academic disciplines, as many as 239 research papers were received by the organizers. With absolute professionalism, the papers to be presented were assessed for their relevance to the conference theme, their novelty, and technical correctness, besides peer review. I congratulate the organizers for successfully culminating such a commendable task. I hope the present proceedings will definitely add new facts and details to the various domains of knowledge covered by it and will serve as a reference book globally.

Prof. M. Y. Shah Dean, Faculty of Applied Science and Technology University of Kashmir

2015

vii

University of Kashmir

Foreword The contemporary world is referred to as being in the "age of technology." Twenty years ago, the world had no Facebook, no Google, and no YouTube! There were no Smart-phones, no Bluetooth, and no Wireless Internet. Today, these things are such a part of our lives that we fail to realize that they have not always been there. Human-computer interface has advanced to such an extent that at times, it is difficult to do without it. Technology is as entrenched into our everyday life as never before. The economies of nations have become directly proportional to the advancement in technology, particularly to those pertaining to electronics, communication, and computers. A process of continuous exploration and innovation in these technologies is, thus, vital for being competitive in the rapidly changing technological world. The International Conference on Advances in Computers, Communication, and Electronic Engineering (COMMUNE-2015) held from 16th to 18th March, at University of Kashmir, is a significant effort in this direction. COMMUNE-2015 has provided a platform for researchers, academicians, and engineers to contribute towards exploring new trends and technologies. The objectives of the conference have been to create an avenue for the researchers, to present high-quality research and to be involved in professional interactions for the advancement in electronics, communication, and computers. The present publication, carrying about one hundred full-length research papers, is a valuable contribution as it addresses the most pertinent and upcoming advancement in computers, communication, and electronic engineering. The research articles embodied in this special issue have been shortlisted after a careful review of more than two hundred qualified submissions. I would like to thank all those who have been involved in the preparation of this manuscript, especially those who assisted in the review process. I deeply acknowledge and appreciate the efforts of the Organizer COMMUNE-2015 as well as those of the faculty, students, and the Research Scholars of the Department of Electronics and Instrumentation Technology, University of Kashmir, for their painstaking effort to organize COMMUNE-2015 within a short period of time.

Prof. G. Mohiuddin Bhat Dean Faculty of Engineering & Head of the Department Convener, COMMUNE-2015

2015

ix

University of Kashmir

Editor’s Note With the ever-growing interaction of technology with individual and collective lives in societies across the world, the challenges posed to technologists, scientists and academicians have multiplied manifold. This is especially true of Information and Communication Technology (ICT) that has now become the inevitable part of modern life. There is no denying to the fact that the life of 21st century human beings is inextricably intertwined with the varied manifestations of ICT. However, it is also certain that the ICT revolution that the world is presently witnessing has been successful because of the unprecedented advances made in Electronic Engineering. Advances in Electronic engineering over the years have subtly worked at the back end of this ICT boom. Developments in High performance VLSI devices and circuits, Monolithic Integrated circuits, Silicon Nanoelectronics, to name a few have efficiently contributed towards bringing in effectiveness in ICT. The broadened scope of ICT has now in its folds, issues and challenges concerning design of low power, high efficiency and small chip area VLSI circuits, bio-information, cryptography, information security, digital forensics, data hiding, artificial intelligence, etc. COMMUNE-2015 was organized with the intention of assembling scholars and academicians from around the world to deliberate on these issues concerning Electronics, Communication, and Computers in conformity with the current trends of interdisciplinary research. As such, the theme of the conference was purposefully chosen to ensure participation from scholars affiliated to a wide range of academic and research disciplines. The overwhelming response shown by scientists, academicians, and scholars from within and outside country from Computer Science, Electronics, Physics, Mathematics, Statistics, Linguistics, etc. reflect the diversified research areas subsumed under the theme. COMMUNE-2015 had eleven sessions including a virtual session, eight keynote addresses, and three expert lectures. The keynote address by Professor Chaturvedi described the characteristics of a cognitive radio as an intelligent device that can use side information about its environment to improve spectral utilization. Professor Sarkar in his keynote address shared hefty number of tricks and techniques to overcome current design challenges in building efficient and low power consuming VLSI circuits within minimum chip area. The third keynote speaker, Professor Chaudhury underscored the importance of deep neural network in building a hybrid text recognizer for document image analysis. Professor Ansari in his address highlighted various perspectives of IT ACT 2000 and deliberated upon management of IT security issues at technical and organizational levels. Dr. Kaushik discussed Spintronics based magneto-resistive memories in terms of their architecture, operation and compared them with conventional memories. In his second keynote, Dr. Kaushik highlighted the prospects of a promising interconnect material “Graphene nano-interconnects” and discussed challenges involved therein. Professor Lehal, reflected upon “the Sangam”, a machine transliteration system transforming Perso-Arabi to Hindi Script. Professor Beg, reflected upon advances in Mobile and Wireless Communication technologies such as coding and compression that could permit flawless telepresence wherein elements such as smell, touch, and taste can also be transmitted. Professor Bhat, while highlighting various Government initiatives such as STIP-2013, GIAN and PRIM for promoting innovation

2015

xi

2015 led entrepreneurship summarized the prospects of ‘Make in India’ initiative and challenges involved in its implementation. Mr. Rayees Mohammad Bhat highlighted emerging cybercrimes such as cyber terrorism, spamming, phishing, online theft, credit card frauds, etc. and discussed a few such cases reported from the State of Jammu and Kashmir. Professor A. H. Mir reflected upon checking authenticity of images using watermarking and stenography and detecting forgery through second order statistical approaches. The organizers received 239 full-length papers in total via Easy Chair submission system, which were reviewed by a team of 141 international and national reviewers. Based on the review, 98 papers were selected for presentation in the COMMUNE-2015. Apart from the review process, all the selected papers were strictly checked for plagiarism using the ‘Turnitin’ software and the authors were given the feedback as well. As per international academic standards, some authors were asked to drop the level of quoted work to below 15 percent of the total word count of the camera-ready paper. For as many as 90 selected papers, the COMMUNE-2015 received registrations, camera-ready copies, and online copyright transfers. Camera Ready copy of each paper submitted to the COMMUNE-2015 was checked for formatting errors and were corrected by organizing committee members. To publish the proceedings of a conference of COMMUNE-2015’s magnitude, it takes considerable amount of time and energy. However, by the grace of Almighty Allah and the perseverant efforts and dedication of the team behind COMMUNE-2015, presenters were handed over a copy of the proceedings at the inaugural function. It was indeed a moment of pleasure for the members of organizing committee to have accomplished such an uphill task. The seeds of COMMUNE were sown quite early in 2011, when the idea of convening such a conference was conceived by the current organizers under the chair of late Prof. N. A. Shah (May Allah bless his soul) and COMMUNE-2015 represents its first flower. The future will see many such colourful flowers as it has been decided to organize COMMUNE annually, in sha Allah. As member of the organizing committee and the editor of the proceedings, I would request your cooperation and seek your suggestions in making the COMMUNE more successful in future. Lastly, I would like to express my gratitude to Prof. Khurshid Iqbal Andrabi, the Honourable Vice-Chancellor, Prof. Mohamad Ashraf Wani, Dean Academic Affairs, Prof. Sheikh Javeed Ahmad, Dean Research, Prof. M. Y. Shah, Dean, Faculty of Applied Science and Technology, and Prof. G. Mohiuddin Bhat, Dean Engineering and Head of the Department for their outright support and patronage.

Dr. Mohammad Tariq Banday Editor Department of Electronics and Instrumentation Technology University of Kashmir

2015

Organization Chief Patron: Prof. Khurshid Iqbal Andrabi, Honourable Vice-Chancellor, University of Kashmir. Patron: Prof. Mohamad Ashraf, Dean Academic Affairs, University of Kashmir. Patron: Prof. Sheikh Javeed Ahmad, Dean Research, University of Kashmir. Convener: Prof. G. M. Bhat, Dean Engineering and Head, Department of Electronics and Instrumentation Technology, University of Kashmir. Organizer: Dr. M. Tariq Banday, Sr. Assistant Professor (Coordinator UGC-SAP), Department of Electronics and Instrumentation Technology, University of Kashmir. Advisory Committee: Prof. Khurshid Iqbal Andrabi, Honourable Vice-Chancellor, University of Kashmir. Prof. Mohamad Ashraf, Dean Academic Affairs, University of Kashmir. Prof. Sheikh Javeed Ahmad, Dean Research, University of Kashmir. Prof. Zaffar Ahmed Reshi, Registrar, University of Kashmir. Prof. M. Y. Shah, Dean, Faculty of Applied Science and Technology, University of Kashmir. Prof. Nisar Ahmad Rather, Dean, Physical and Material Sciences, University of Kashmir. Prof. G. M. Bhat, Dean Engineering and Head, Department of Electronics & Instrumentation Technology, University of Kashmir. Prof. S. M. K. Qaudri, Head, Department of Computer Applications and Director, IT&SS, University of Kashmir. Prof. Sharief-ud-din Pirzada, Head, Department of Mathematics, University of Kashmir. Prof. Manzoor Ahmad Malik, Head, Department of Physics, University of Kashmir. Dr. M. A. K. Baigh, Head, Department of Statistics, University of Kashmir. Prof. Ajaz Hussain Mir, Professor, Department of Electronics and Communication Engineering, NIT, Srinagar. Er. A. H. Moon, Director, NIELIT, Srinagar. Dr. Mohammad Ahsan Chesti, Department of Computer Science and Engineering, NIT, Srinagar. Dr. M. Tariq Banday, Sr. Assistant Professor (Coordinator UGC-SAP), Department of Electronics & Instrumentation Technology, University of Kashmir. Prof. Aadil Amin Kak, Department of Linguistics, University of Kashmir. Dr. Basharat Ahmad Want, Associate Professor, Department of Physics, University of Kashmir. Dr. Majid Zaman Baba, Scientist B, Directorate of IT & IS, University of Kashmir. Dr. Farooq Ahmad Khanday, Assistant Professor, Department of Electronics & Instrumentation Technology, University of Kashmir. Dr. Shabir Ahmad Parah, Assistant Professor, Department of Electronics & Instrumentation Technology, University of Kashmir. Dr. Javaid Ahmad Sheikh, Assistant Professor, Department of Electronics & Instrumentation Technology, University of Kashmir.

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

xiii

2015 Dr. Tariq Rashid Jan, Sr. Assistant Professor, Department of Statistics, University of Kashmir. Dr. Musavir Ahmad, Sr. Assistant Professor, Department of Linguistics, University of Kashmir. Dr. Nadeem Ahmad, Sr. Assistant Professor, Department of Library Sciences, University of Kashmir. Er. Abdul Mueed Hafiz, Assistant Professor, Department of Electronics & Instrumentation Technology, University of Kashmir. Er. Rouf ul Alam Bhat, Assistant Professor, Department of Electronics & Instrumentation Technology, University of Kashmir. Mrs. Farhat Roohi, Electronic Engineer, Department of Electronics & Instrumentation Technology, University of Kashmir.

Members of Organizing Committees: Dr. M. Tariq Banday, (Coordinator UGC-SAP), (General and Program Chair) Dr. Farooq Ahmad Khanday, (Finance Chair) Dr. Shabir Ahmad Parah, (Publicity Chair) Dr. Javaid Ahmad Sheikh, (Hospitality Chair) Mrs. Farhat Roohi, (Registration Chair) Er. Rouf ul Alam Bhat, (Accommodation Chair) Er. Abdul Mueed Hafiz, (Accommodation Chair) Mr. Nisar Ahmad Paray Mrs. Muzamil Hassan Mr. Azad Ahmad Shah Mr. Mohamad Rafiq Beigh Ms. Shafiya Afzal Sheikh Mr. Javeed Iqbal Reshi Mr. Nasir Ali Kant Mr. Mohammad Rafiq Dar Ms. Uzma Ms. Tawheed Jan Ms. S. Umira R. Qadri Mr. Farooq Aadil Rather Mr. Reyaz Ahmad Mathangi Mr. Jahangir Ahmad Mr. Mehboob ul Amin Mr. Imran Nazir Beigh Ms. Sakeena Akhtar Ms. Asma Nazir Naqash Ms. Jaipreet Kour Wazir Ms. Farhana Ahad Mr. Nazir Ahmad Mr. Zubair Ahmad Bangi

Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar

xiv

2015 Technical Program Committee Members: Prof. Aadil Amin Kak, Department of Linguistics, University of Kashmir, Srinagar, India. Prof. Abdul Quaiyum Ansari, Department of Electrical Engineering, Jamia Millia Islamia, New Delhi, India. Prof. Aijaz Ahmad, Department of Electrical Engineering, National Institute of Technology, Srinagar, India. Prof. Ajaz Hussain Mir, Department of Electronics & Communication Engineering, National Institute of Technology, Srinagar, India. Prof. Alam Aftab, Department of Physics, Indian Institute of Technology, Mumbai, India. Prof. Anurekha Sharma, Department of Electronic Science, Kurukshartra University, Kurukshartra, India. Prof. Anwar Shahzad Siddiqui, Department of Electrical Engineering, Jamia Millia Islamia (Central University), New Delhi, India. Prof. Carlos Molina, Department of Computer Science, Universidad De Jaen, Spain. Prof. Costas Psychalinos, Department of Electronics & Computers, University of Patras, Rio, Patras, Greece. Prof. Satya Prakash Ghrera, Department of Computer Science and Engineering, Jaypee University of Information Technology, Waknaghat, India. Prof. Ekram Khan, Department of Electronics & Communication Engineering, Aligarh Muslim University, Aligarh, India. Prof. Farooq A. Mir, Department of Law, University of Kashmir, Srinagar, India. Prof. Florina Ungureanu, Faculty of Automatic Control & Computer Engineering, Technical University of IASI, IASI, Romania. Prof. G. Mohiuddin Bhat, Department of Electronics & Instrumentation Technology, University of Kashmir, Srinagar, India. Prof. Mehraj Ud Din Mir, Central University of Kashmir, Srinagar, India. Prof. M. Mustafa, Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India. Prof. M. Salim Beg, Department of Electronics Engineering, Zakir Hussain College of Engineering & Technology, Aligarh Muslim University, Aligarh, India. Prof. M. Shah Alam, Department of Electronics Engineering, Aligarh Muslim University, Aligarh, India. Prof. Mainuddin, Department of Electronics & Communication Engineering, Jamia Millia Islamia, New Delhi, India. Prof. Mairaj-ud-din, Department of Electrical Engineering, National Institute of Technology, Srinagar, India. Prof. Mridula Gupta, Department of Electronics Science, South Campus, University of Delhi, Delhi, India. Prof. Najeeb-ud-din, Department of Electronics & Communication Engineering, National Institute of Technology, Srinagar, India. Prof. Nasib Singh Gill, Department of Computer Science & Applications., M. D. University, Rohtak, Haryana, India. Prof. Nilanjan Dey, Department of Computer Science Engineering, BCET, Durgapur, West Bengal, India.

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

xv

2015 Prof. Paresh V. Virparia, Department of Computer Science, Sardar Patel University, Vallabh Vidyanagar, Gujarat, India. Prof. R. K. Sarin, Department of Electronics & Communication Engineering, National Institute of Technology, Jalandhar, India. Prof. Roohie Naaz Mir, Department of Computer Science Engineering, National Institute of Technology, Srinagar, India. Prof. S. P. Singh, Electronics Engineering, Indian Institute of Technology, Roorkee, India. Prof. Santanu Choudhury, Department of Electronics Engineering, Indian Institute of Technology, Delhi, India. Prof. Seifedine Kadry, Department of Applied Mathematics, American University of the Middle East, Egaila, Kuwait. Prof. Shamim Ahmad Lone, Department of Electrical Engineering, National Institute of Technology, Srinagar, India. Prof. Sharief-ud-din Pirzada, Department of Mathematics, University of Kashmir, Srinagar, India. Prof. S. Naseem Ahmad, Department of Electronics & communication Engineering, Jamia Millia Islamia, New Delhi, India. Prof. Stefan Segla, Department of Mechanical Engineering, Technical University of Kosice, Kosice, Slovakia. Prof. Subir Kumar Sarkar, Department of Electronics & Telecommunication Engineering, Jadavpur University, Kolkata, India. Prof. B. A. Usha, Department of Computer Science Engineering, R. V. College of Engineering, Bengaluru, Karnataka, India. Prof. Vivek Kshirsagar, Department of Computer Science, Govt. Engineering College, Aurangabad, India. Prof. Yudong Zhang, School of Information Science & Technology, Nanjing Normal University, China. Dr. A. K. Daniel, Department of Computer Science & Engineering, M. M. M. University of Technology, Gorakhpur, U.P., India. Dr. Amit Kant Pandit, Department of Electronics & Communication Engineering, Shri Mata Vaishno Devi University, Katra, Jammu & Kashmir, India. Dr. Anwar Sadar, Department of Electronic Engineering, Aligarh Muslim University, Aligarh, India. Dr. Arman Rasool Faridi, Department of Computer Science, Aligarh Muslim University, Aligarh, India. Dr. Athar Ali Moinuddin, Department of Electronic Engineering, Aligarh Muslim University, Aligarh, India. Dr. Atul M. Gonsai, Department of Computer Science, Saurashtra University, Rajkot, India. Dr. Basharat Want, Department of Physics, University of Kashmir, Srinagar, India. Dr. Bharati Harsoor, Department of Information Science & Engineering, Poojya Doddappa Appa College of Engineering, Gulbarga, Karnataka, India. Dr. Brajesh Kumar Kaushik, Department of Electronics & Communication Engineering, Indian Institute of Technology, Roorkee, India. Dr. B. Sharada, Department of Computer Science, University of Mysore, Mysore, India.

Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar

xvi

2015 Dr. Carkis A. De La Cruz Blas, Department of Electric & Electronical Engineering, Public University of Navarre, Pamplona, Spain. Dr. Dharam Veer Sharma, Department of Computer Science, Punjabi University, Patiala, India. Dr. Dharmender Singh Kushwaha, Department of Computer Science & Engineering, Motilal Nehru National Institute of Technology, Allahabad, India. Dr. Faisal Anwer, Department of Computer Science, Aligarh Muslim University, Aligarh, India. Dr. Farooq Ahmad Khanday, Department of Electronics & Instrumentation Technology, University of Kashmir, India. Dr. Giuseppe Serra, School of Computer Science, University of Modena & Reggio, Emillia, Italy. Dr. Gurpreet Singh Lehal, Department of Computer Science, Punjabi University, Patiala, India. Dr. Harsupreet Kaur, Department of Electronic Science, University of Delhi, New Delhi, India. Dr. Hemraj Saini, Department of Computer Science & Engineering, Jaypee University of Information Technology, Wakanaghat, Solan-173234, India. Dr. Hung-Wei Chen, Prime Electronics & Satellites Inc., Taiyuan, Taiwan. Dr. Javaid Ahmad Sheikh, Department of Electronics & Instrumentation Technology, University of Kashmir, India. Dr. Javier Rubio Loyola, Centre of Research & advanced studies, National Polytechnic Institute of Mexico, Cinvestav, Mexico. Dr. Jitendra Agrawal, Department of Computer Science & Engineering, Rajiv Gandhi Proudyogiki Vishwavidyalaya, Bhopal, MP, India. Dr. Kandarpa Kumar Sarma, Department of Electronics & Communication Technology, Gauhati University, Guwahati, Assam, India. Dr. Kandarpa Kumar Sharma, Department of Electronics & Communication Technology, Gauhati University, Guwahati, Assam, India. Dr. Larrey Wen, Institute of Integrated & Intelligent Systems, Griffith University, Southeastern Queensland, Australia. Dr. Lian Wen, School of Information & Communication Technology, Griffith University, Brisbane, Australia. Dr. M. Hanumanthappa, Department of Computer Science, Bangalore University, Bangalore, India. Dr. M. Tariq Banday, Department of Electronics & Instrumentation Technology, University of Kashmir, India. Dr. Majid Zaman Baba, Directorate of IT & SS, University of Kashmir, Srinagar, India. Dr. Malaya Kumar Nath, Department of Electronics & Communication Engineering, National Institute of Technology, Puducherry, Karaikal, India. Dr. Mansaf Alam, Department of Computer Science, Jamia Millia Islamia, New Delhi, India. Dr. Manzoor A. Malik, Department of Physics, University of Kashmir, Srinagar, India. Dr. Mohammad Ahsan Chishti, Department of Computer Science & Engineering, National Institute of Technology, Srinagar, India.

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

xvii

2015 Dr. Mohammad Jawaid Siddiqui, Department of Electronics Engineering, College of Engineering, Aligarh Muslim University, Aligarh, India. Dr. Mohammad Nayeem Teli, Department of Computer Science & Engineering, National Institute of Technology, Srinagar, India. Dr. Mohammad Sarosh Umar, Department of Computer Engineering, Aligarh Muslim University, Aligarh, India. Dr. Mohammad Shukri Salman, Department of Electrical & Electronics Engineering, Mevlana University, Konya, Turkey. Dr. Mohammad Amjad, Department of Computer Engineering, Jamia Millia Islamia (Central University), New Delhi, India. Dr. Monica Mehrotra, Department of Computer Science, Jamia Millia Islamia, New Delhi, India. Dr. Musavir Ahmed, Department of Linguistics, University of Kashmir, Srinagar, India. Dr. Muheet Ahmed Bhat, Department of Computer Sciences, University of Kashmir, Srinagar, India. Dr. Musheer Ahmad, Department of Computer Engineering, Jamia Millia Islamia, India. Dr. Musiur Raza Abidi, Department of Electronics Engineering, Aligarh Muslim University, Aligarh, India. Dr. Navneet Agrawal, Department of Electronics & Communication Engineering, Maharana Pratap University of Ag. & Technology, Udaipur, Rajasthan, India. Dr. Nayan M. Kakoty, Department of Electronics & Communication Engineering, School of Engineering, Tezpur University, Tezpur, India. Dr. Norbert Herencsar, Department of Telecommunications, Brno University of Technology, Brno, Czech Republic. Dr. Nusrat Parveen, Department of Electronics, Islamia College of Science and Commerce, Srinagar, India. Dr. Omar Farooq, Department of Electronics Engineering, Aligarh Muslim University, Aligarh, India. Dr. Prashant M. Dolia, Department of Computer Science & Applications, Maharaja Krishna-Kumar-Sinhji Bhavnagar University, Bhavnagar, Gujarat, India. Dr. P. Thimmaiah, Institute of Technology, Sri Krishnadevaraya University, Anantapur, Andhrapradesh, India. Dr. R. D. Morena, Department of Computer Science, Veer Narmad South Gujarat University, Surat, India. Dr. Rafiqul Zaman Khan, Department of Computer Science, Aligarh Muslim University, Aligarh, India. Dr. Rahmat Widia Sembiring, CEO of Data Mining Research & its Application, University of Malaysia Pahang, Politeknik Negeri Medan, Indonesia. Dr. Raj Senani, Ex-Director, Netaji Subhas Institute of Technology, New Delhi, India. Dr. Rajeev Agrawal, School of Technology, North Carolina A&T State University, Greensboro, USA. Dr. Rajni Mohana, Department of Computer Science & Engineering, Jaypee University of Information Technology, Waknaghat, Solan, India. Dr. Rakesh K. Jha, School of Electronics & Communication Engineering, Shri Mata Vishnu Devi University, Katra, Jammu, India.

Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar

xviii

2015 Dr. Rakesh Vaid, Department of Physics & Electronics, University of Jammu, Jammu, India. Dr. Roshan G. Ragel, Department of Computer Engineering, University of Peradeniya, Peradeniya, SriLanka. Dr. Sajad A. Loan, Department of Electronics & Communication Engineering, Jamia Millia Islamia, New Delhi, India. Dr. Sanjay Jamwal, Department of Computer Science, Baba Ghulam Shah Badshah University, Rajouri, Jammu & Kashmir, India. Dr. Sanjay Tyagi, Department of Computer Science & Applications, Kurukshetra University, Kurukshetra, Haryana, India. Dr. Sanjeev Singh, Institute of Informatics & Communication, University of Delhi, South Campus, New Delhi, India. Dr. Sara Kadry, Software Research Institute, Athlone Institute of Technology, Republic of Ireland. Dr. Shabir A. Parah, Department of Electronics & Instrumentation Technology, University of Kashmir, Srinagar, India. Dr. Shikha Agrawal, Department of Computer Science & Engineering, Rajiv Gandhi Proudyogiki Vishwavidyalaya, Bhopal, MP, India. Dr. Shruti Jain, Department of Electronics & Communication Engineering, Jaypee University of Information Technology, Solan, Himachal Pradesh, India. Dr. Soumik Roy, Department of Electronics & Communication Engineering, Tezpur University, Tezpur, India. Dr. Subhash Chander Dubey, Department of Electronics & Communication Engineering, Govt. College of Engineering & Technology, Jammu, India. Dr. Sunil Kumar Wanchoo, Department of Physics, Shri Mata Vaishno Devi University, Katra, Jammu & Kashmir, India. Dr. Suresh Kumar, Department of Electronic Science, Kurukshartra University, Kurukshartra, India. Dr. Susheel Sharma, Department of Physics & Electronics, University of Jammu, Jammu, India. Dr. Svetlana Vasileva, International University College, Dobrich, Bulgaria. Dr. Syed Zaffer Iqbal, Department of Physics, Government College for Women Nawakadal, Srinagar, India. Dr. T. V. Prasad, Department of Computer Science & Engineering, Chirala Engineering College, Chirala, AP, India. Dr. T. Veera Kumar, Department of Electronics & Communication Engineering, National Institute of Technology, Goa, India. Dr. Tariq Rashid Jan, Department of Statistics, University of Kashmir, Srinagar, Kashmir, India. Dr. Tasleem Arif, Department of Information Technology, Baba Ghulam Shah Badshah University Rajouri, Jammu & Kashmir, India. Dr. Thomas Schlechter, Hardware R & D, Skidata AG, Groedig/Salzburg, Austria. Dr. Utpal S. Joshi, Department of Physics & Electronics, Gujarat University, Gujarat, India. Dr. Varsha Sharma, School of Information Technology, Rajiv Gandhi Proudyogiki Vishwavidyalaya, Bhopal, India. Dr. Vasudha Bhatnagar, Department of Computer Science, University of Delhi, Delhi, India.

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

xix

2015 Dr. Vijay Kumar, Department of Information Technology, Govt. of India, New Delhi, India. Dr. Virender Singh Kunda, Department of Electronic Science, Kurukshartra University, India. Dr. Vishal Goyal, Department of Computer Science, Punjabi University, Patiala, India. Dr. Vivek Chalotra, Department of Physics & Electronics, University of Jammu, Jammu, India. Dr. Werner Hartmann, Siemenns AG, Corporate Technology Research & Technology Centre, PET, Guenther-Schorowsky-Str, Germany. Dr. Zhi-Kai Huang, Nanchang Institute of Technology, Nanchang, China. Dr. Zhong Lin Wang, Department of Electronics & Communication Engineering, Georgia Institute of Technology, Atlanta, GA, USA. Er. Rajandeep Singh, Department of Electronics & Communication Engineering, Guru Nanak Dev University, Regional Campus, Sultanpur, Lodhi, India. Ms. Lafifa Jamal, Department of Computer Science & Engineering, University of Dhaka, Dhaka, Bangladesh. Ms. Maria B. Line, Department of Telematics, Norwegian University of Science & Technology, Trondheim, Norway. Ms. Pragya Dwivedi, Department of Computer Science & Engineering, Motilal Nehru National Institute of Technology, Allahabad, India. Mr. Jatinder Manhas, Department of Computer Science, University of Jammu, Jammu, India. Mr. Asad Mohammed Khan, Department of Computer Engineering, Aligarh Muslim University, Aligarh, India. Mr. Diana Palsetia, Department of Electrical Engineering & Computer Science, Northwestern University, Evanston, USA. Mr. Padma Prasada, Department of Electronics & Communication Engineering, Mangalore Institute of Technology & Engineering, Karnataka, India. Mr. Shibaji Mukherjee, Senior Manager at Oracle, Kinsight Analytics, Bangalore, India. Mr. Suhel Mustajab, Department of Computer Science, Aligarh Muslim University, Aligarh, India. Mr. Suryadip Chakraborty, Department of Computer Science & Engineering, University of Cincinnati, Cincinnati, OH - 45220, USA. Mrs. B. Shanmuga Priya, Department of Computer Science, Sri Ramakrishna College of Arts & Science for Women, Coimbatore, Tamil Nadu, India. Mrs. Shraddha Arya, Department of Computer Science, Sri Guru Gobind Singh College, Chandigarh, India.

Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar

xx

2015

Keynotes and Expert Lectures 

Information Theoretic Perspective on Cognitive Radio Networks [1] Prof. A. K. Chaturvedi, IIT Kanpur.



Design Challenges for Low Power VLSI Circuits [3] Prof. Subir Kumar Sarkar, Department of Electronics and Telecommunication Engineering, Jadavpur University, Kolkata.



Deep Learning for Document Image Analysis [5] Prof. Santanu Chaudhury, Department of Electrical Engineering, IIT, Delhi.



Security Issues in IT Systems and their Management [7] Prof. Abdul Quaiyum Ansari, Department of Electrical Engineering, Jamia, Millia, Islamia, New Delhi.



Spin Transfer Torque based Magneto-resistive Memories [9] Dr. Brajesh Kumar Kaushik, Department of Electronics and Communication Engineering, IIT, Roorkee.



Graphene Based On-chip Interconnects and TSVs: Prospects and Challenges [11]

Dr. Brajesh Kumar Kaushik, Department of Electronics and Communication Engineering, IIT, Roorkee. 

A Perso-Arabic to Indic Script Machine Transliteration Model [13] Dr. Prof. Gurpreet Singh Lehal, Department of Computer Science, Punjabi University, Patiala, India.



Moving Towards “Flawless Telepresence” Systems of the Future [15] Prof. M. Salim Beg, Department of Electronics Engineering, AMU, Aligarh.



Technology Innovation and Diffusion Practical approach towards 'Make in India' [17] Prof. G. Mohiuddin Bhat, University of Kashmir, Srinagar.



Current State of Cyber Crimes in the State of Jammu and Kashmir [19] Mr. Rayees Mohammad Bhat, IPS, SP, Hazratbal, Srinagar.



Detecting Forgery in Images: A Statistical Perspective [21] Prof. Ajaz Hussain Mir, NIT, Srinagar.

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

xxi

2015

Keynote - 1 Information Theoretic Perspective on Cognitive Radio Networks Prof. A. K. Chaturvedi* Indian Institute of Technology Kanpur, India

Keynote Cognitive radios hold tremendous promise to increase the spectral efficiency of wireless systems. We will start with a brief introduction to information theory and cognitive radio networks and then discuss the fundamental capacity limits in such networks. We will characterize a cognitive radio as an intelligent device that can use side information about its environment to improve spectral utilization. © 2015 Published by University of Kashmir, Srinagar. Selection and/or peerreview under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar.

Keywords: Cognitive Radio; Capacity; Degrees of Freedom; Interweave; Overlay; Underlay

*Speaker. Tel.: +91 512 2597613. E-mail address: [email protected]. Keynote/Expert Lecture ID: SEEDS/COMMUNE-001

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

1

2015

Keynote - 2 Design Challenges for Low Power VLSI Circuits Prof. Subir Kumar Sarkar* Department of Electronics and Telecommunication Engineering, Jadavpur University, Kolkata-700 032, India

Keynote Tremendous growth of the VLSI technology has been mainly due to progress of fabrication technology, which allowed systematic scale-down of device feature sizes and exponential growth of the integration level. Continuous device performance improvement is possible probably through a combination of device scaling, new device structure and material property improvement. Due to its small size, their potential integration level is significantly high and its low power operation solves some of the instability and reliability problems. The major challenges for design Engineers are to design new generation products, which consume minimum power, without compromising its performance or achieving minimum chip area. As we approach millennium, power dissipation has become the main design concern in many applications such as wristwatch, laptop, computers, and pace makers although early VLSI design did not consider it. The objective of such applications is minimum power for maximum battery life. Power dissipation is the greatest obstacle for Moore’s law. Modern chips consume power of which about 20% is wasted in leakage through the transistor gates. The traditional means of coping with increased power per generation has been to scale down the operating voltage of the chip but voltages are reaching limits due to thermal fluctuation effects. To save power, several tricks have been considered viz., minimize activity, glitches, effective capacitance, wire length of nodes and use of minimum possible supply voltage constrained by performance needed, design for high speed and then reduce voltage to get the desired speed. © 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar.

Keywords: Low Power; VLSI Circuits; Power Dissipiation; Tricks to Save Power; Glitches; Effective Capacitance

*Speaker. Tel.: +91 3324 572810. E-mail address: [email protected]. Keynote/Expert Lecture ID: SEEDS/COMMUNE-002 Delivered in Joint Session of SEEDS-2015 and COMMUNE-2015

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

3

2015

Keynote – 3 Deep Learning for Document Image Analysis Prof. Santanu Chaudhury* Dhananjay Chair Professor, FNAE, FNASc, FIAPR, Department of Electrical Enginering, I.I.T, Delhi

Keynote Deep networks provide a new paradigm for feature discovery and recognition. We can approach problems of document image analysis in the framework of deep learning. We shall examine use of deep learning for scene text recognition. Next we shall present an architecture for text recognition using deep LSTM. Text recognition involves some initial image processing steps like segmentation of lines and words which can induce error to the recognition system. Without segmentation, learning very long range context is difficult and becomes computationally intractable. Therefore, alternative soft decisions are needed at the pre-processing level. This paper proposes a hybrid text recognizer using a deep recurrent neural network with multiple layers of abstraction and long range context along with a language model to verify the performance of the deep neural network. In this paper we construct a multi-hypotheses tree architecture with candidate segments of line sequences from different segmentation algorithms at its different branches. The deep neural network is trained on perfectly segmented data and tests each of the candidate segments, generating unicode sequences. In the verification step, these unicode sequences are validated using a sub-string match with the language model and best first search is used to find the best possible combination of alternative hypothesis from the tree structure. Thus the verification framework using language models eliminates wrong segmentation outputs and filters recognition errors. © 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar.

Keywords: Feature Discovery; Document Image Analysis; Hybrid Text Recognizer; Deep Neural Network

*Speaker. Tel.: +91 9891 266595. E-mail address: [email protected]. Keynote/Expert Lecture ID: SEEDS/COMMUNE-003 Delivered in Joint Session of SEEDS-2015 and COMMUNE-2015

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

5

2015

Keynote – 4 Security Issues in IT Systems and their Management Prof. Abdul Quaiyum Ansari* Department of Electrical Engineering, Jamia Millia Islamia, New Delhi

Keynote IT security means protecting information and information systems from unauthorized access, use, disclosure, disruption, modification, or destruction and IT Security management is a process of defining the security controls in order to protect the information assets. The issue of security related to IT systems is not only technical but is also a governance and organizational issue. Several agencies have been working simultaneously to resolve this issue both at the technical as well as at the organizational levels. Many standards have been developed to handle this complex problem. Every country has its own IT law. The Indian IT ACT 2000 aims to provide the legal infrastructure for e-commerce in India that is supposed to make a major impact for ebusinesses and the new economy in India. This Key Note will focus on the generic issues of the security challenges and their management as also the various perspectives of the IT Act 2000 as related to the international IT security standards and challenges. © 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar.

Keywords: IT ACT; Information Security; IT Law; IT Security Management

*Speaker. Tel.: +91 9873 824597. E-mail address: [email protected]. Keynote/Expert Lecture ID: SEEDS/COMMUNE-004 Delivered in Joint Session of SEEDS-2015 and COMMUNE-2015

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

7

2015

Keynote - 5 Spin Transfer Torque based Magneto-resistive Memories Dr. Brajesh Kumar Kaushik* Microelectronics and VLSI Group, Department of Electronics and Communication Engineering, I. I. T, Roorkee, Uttarakhand

Keynote Researchers believe spintronics to be one of the most promising technologies to replace the conventional CMOS technology that suffers from severe static leakage beyond 22nm technology node. Spintronics is an emerging technology that exploits an electron's spin orientation and its associated magnetic moment as state variable. It involves the storage of information in terms of non-volatile magnetization state instead of the charge. Thus, the new goal is to develop computing architecture that can normally be off when not in use to prevent static leakage. Moreover, such architecture can be turned on instantly with full performance when required. The primary requisite to achieve non-volatile architecture is non-volatile RAM (NVRAM). Most promising technology to achieve non-volatile RAMs is the emerging spintronics based magneto-resistive memories that switches by spin transfer torque (STT). Spintronics based magneto-resistive memories were revolutionized by the phenomenon of spin transfer torque (STT) effect, first demonstrated by J.C Slonczewski in 1996. After this monumental discovery, spintronics based magneto-resistive memories have evolved considerably in the last decade into their novel form known as spin transfer torque magneto-resistive random access memories (STT MRAMs). STT MRAMs store data as the resistance state of a magneto-resistive device known as magnetic tunnel junctions (MTJs). An STT MRAM cell is composed of two primary components: the "Magnetic Tunnel Junction", which is usually characterized by magneto-resistance and switching current density, and the "Access Device", which allows a given memory cell to be addressed for read or write. This talk will target for a clear understanding of STT MRAMs in terms of architecture, operation and performance comparison with other volatile and non-volatile memory technologies. Moreover, the talk will also focus towards the recent developments and challenges ahead for STT MRAMs. © 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar.

Keywords: Spin Transfer Torque Switching; Magnetic Tunnel Junction; MRAM; Perpendicular Magnetic Anisotropy

*Speaker. Tel.: +91 1332 285662. E-mail address: [email protected]. Keynote/Expert Lecture ID: SEEDS/COMMUNE-005 Delivered in Joint Session of SEEDS-2015 and COMMUNE-2015

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

9

2015

Keynote - 6 Graphene Based On-chip Interconnects and TSVs: Prospects and Challenges Dr. Brajesh Kumar Kaushik* Microelectronics and VLSI Group, Department of Electronics and Communication Engineering, I. I. T., Roorkee, Uttarakhand

Keynote The conventional on-chip interconnect copper material is unable to meet the requirements of future technology needs, since it demonstrates lower reliability with down scaling of interconnect dimensions. Therefore, researchers are forced to find an alternative solution for interconnects. Graphene nano interconnects have been proposed as promising interconnect materials due to their unique physical properties such as higher thermal conductivity, current carrying capability and mechanical strength. Graphene nano interconnects can be classified into carbon nanotubes (CNT) and graphene nanoribbons (GNR). CNTs are made by rolling up of graphene sheet in a cylindrical form and GNR is a strip of ultra-thin width graphene layer. Most of the physical and electrical properties of GNRs are similar to that of CNTs, however, the major advantage of GNRs over CNTs is that both transistor and interconnect can be fabricated on the same continuous graphene layer. Therefore, one of the manufacturing difficulties in formation of perfect metal-nanotube contact can be avoided. On other hand, the GNRs fabricated till-date, have displayed some level of edge roughness. The electron scattering at rough edges reduces the mean free path (MFP) that substantially lowers the conductance of the GNR. This fundamental challenge limits the performance of GNR interconnects. Presently, researchers and industrialists are standing at crossroads where they need to make subtle improvements to make CNTs and GNRs a workable solution for future. The conventional planar integrated circuit (2D) packaging technique has already hit the red brick wall and is almost on the verge of extinction due to limited number of I/O pins and lower bandwidth. The best way to move towards the “More-thanMoore” technologies is 3D IC packaging, where the dies are vertically stacked. The electrical connections between the dies are established by through silicon vias (TSVs). The idea of using CNTs and GNRs as filler material in TSVs has also rapidly gained research interests. Considering the above-mentioned issues, this talk will analyze and compare the performance of CNTs and GNRs for both on-chip interconnects and TSVs applications. © 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar.

Keywords : Carbon Nanotube; CNT; Graphene Nanoribbon; GNR; On-Chip Interconnects; Through Silicon Vias; TSVs

*Speaker. Tel.: +91 1332 285662 E-mail address: [email protected] Keynote/Expert Lecture ID: SEEDS/COMMUNE-006

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

11

2015

Keynote - 7 A Perso-Arabic to Indic Script Machine Transliteration Model Prof. Gurpreet Singh Lehal* Department of Computer Science, Punjabi University, Patiala, India.

Keynote Indian sub-continent is one of those unique parts of the world where single languages are written in different scripts. This is the case for example with Punjabi, spoken by tens of millions of people, but written in Indian East Punjab (20 million) in Gurmukhi script (a Left to Right script based on Devanagari) and in Pakistani West Punjab (80 million), it is written in Shahmukhi (a Right to Left script based on Perso-Arabic). Whilst in speech, Punjabi spoken in the Eastern and the Western parts is mutually comprehensible in the written form it is not. This is also the case with other languages like Urdu and Hindi (whilst having different names, they are the same language but written, as with Punjabi, in mutually incomprehensible forms). Hindi is written in the Devanagari script from left to right, Urdu is written in a script derived from a Persian modification of Arabic script written from right to left. A similar problem resides with the Sindhi language, which is written in a Perso-Arabic script in Pakistan and both in Perso-Arabic and Devanagari in India. Similar is the case with Kashmiri language too. Konkani is probably the only language in India, which is written in five scripts i.e. Roman, Devanagari, Kannada, Perso-Arabic, and Malayalam. The existence of multiple scripts has created communication barriers, as people can understand the spoken or verbal communication, however when it comes to scripts or written communication, the number diminishes, thus a need for transliteration tools, which can convert text written in one language script to another script arises. A common feature of all these languages is that, one of the script is Perso-Arabic (Urdu, Sindhi, Shahmukhi etc.), while other script is Indic (Devanagari, Gurmukhi, Kannada, Malayalam). Perso-Arabic script is a right to left script, while Indic scripts are left to right scripts and both the scripts are mutually incomprehensible forms. Thus, there is a dire need for development of automatic machine transliteration tools for conversion between Perso-Arabic and Indic scripts. Machine Transliteration is an automatic method to generate characters or words in one alphabetical system for the corresponding characters in another alphabetical system. The transformation of text from one script to another is usually based on phonetic equivalencies. We present Sangam, a Perso-Arabic to Indic script machine-transliteration system, which can convert with high accuracy text written in Perso-Arabic script to one of the Indic script sharing the same language. The system has been successfully tested on Punjabi (Shahmukhi-Gurmukhi), Urdu (Urdu- Devanagari) and Sindhi (Sindhi Perso Arabic - Sindhi Devanagari) languages and can be easily extended for other languages like Kashmiri and Konkani text. © 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar.

Keywords : Perso-Arabic; Machine Transliteration; Indic Script; Indian Languages

*Speaker. Tel.: +91 9815 473767. E-mail address: [email protected]. Keynote/Expert Lecture ID: SEEDS/COMMUNE-007

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

13

2015

Keynote - 8 Moving Towards “Flawless Telepresence” Systems of the Future Prof. M. Salim Beg* Department of Electronics Engineering, Z. H. College of Engineering & Technology, A.M.U., Aligarh.

Keynote To be able to communicate and compute anywhere and anytime has been one of the goals for last few decades. Due to the huge investments in R&D in the area of Wireless networks in general, and Mobile Communications and Computing in particular, both in the academia and the industry, there has been tremendous outcome in terms of new technologies being developed, as well as resulting products and services. We have in fact now moved on from terms like ‘Ubiquitous Communications’ or ‘Ubiquitous computing’ to newer areas. The traditional mobile and wireless communication networks combined with multimedia communication gave rise to sending text, audio, image, and video to any person, any time, and at any place. In future, several new systems will be made available to us that will lead to something that may be referred to as “flawless telepresence”. The latter implies technologies dealing with not just traditional multimedia elements like text, audio, image, video but incorporating newer elements like smell, touch, and taste. All this will lead to newer frontiers and a paradigm shift in our concept of both computing and communication. Of course, the expectations of users in terms of Quality of Service is growing higher and higher. Thus, one would expect from the service providers to give networks in future that provide richer quality of service and experience by incorporating newer techniques, systems, and networks. While there will be a lot of contributing and enabling technologies for these ‘flawless telepresence’ systems of future, this talk would concentrate mainly on (i) mobile/wireless communication systems and networks (ii) coding and compression of multimedia information and new multimedia systems. Some of the trends and directions in research in these areas will be covered in the talk. © 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar.

Keywords: Flawless Telepresence; Ubiquitous Communications; Ubiquitous computing; multimedia information; Coding

*Speaker. Tel.: +91 9897 023521. E-mail address: [email protected]. Keynote/Expert Lecture ID: SEEDS/COMMUNE-008

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

15

2015

Expert Lecture - 1 Technology Innovation and Diffusion Practical approach towards 'Make in India' Prof. G. Mohiuddin Bhat* Department of Electronics & Instrumentation Technology, University of Kashmir, Srinagar.

Expert Lecture 'Make in India', an international marketing strategy conceptualized by the Prime Minister of India, Shri Narendra Modi in September, 2014, has been aimed to boost the industrial production and labour intensive manufacturing in India. The objective is to create a job market for a largely unemployed population of 1.2 Billion people in the country, needing about one Million additional jobs every month. The economy of India, presently based on agriculture, cannot be left on the mercy of the unpredictable South-West Monsoons. In order to sustain the rapid growth and alleviate poverty, India rightly needs to harness its potential of 'Make in India'. From the invention of Pencillin to the present day Mobile phone, history is witness to the fact that innovations and inventions have enabled societies to produce more. However, technology innovations can contribute towards productivity only through their application, adoption, and diffusion. Liberal support for technology innovation will enhance entrepreneurship development, which will in turn accelerate the economic growth. Technology innovation and its diffusion are, thus, very crucial towards boosting the manufacturing and service sectors. However, in spite of its large publicly funded science & technology infrastructure, India has not been able to realize its innovative potential. The decrease in the number of indigenous patent applications being filed in India in recent years has raised several questions on the promotion of innovation eco-system in the country. While China topped the global list by filing 5,26,412 Patent applications in the year 2011 (with USA having 5,03,582 patent applications, as runner up), only 42,291 patent applications were filed in India during this period. Realizing that innovation led entrepreneurship development shall promise an economic growth; the Govt. of India has recently taken several initiatives with an innovation agenda. Declaration of 2010-20 as the Decade of Innovation, establishment of National Innovation Council and formulating the Science, Technology & Innovation Policy-2013 (STIP-2013) are indicative of some positive developments in this regard. Further, National level Organizations and programmes like National Innovation Foundations (NIF), Promotion of Innovations among Individuals and MSMEs (PRISM), Grass-root Innovation and Augmentation Network (GIAN) are several other initiatives launched in this direction. Creation of a robust national innovation eco-system is one of the key elements listed in the Science Technology Innovation Policy2013 of the Govt. of India. With a focus on the new initiative of 'Make in India', as conceived by the Prime Minister of India, this article summarizes the possibilities and challenges in the implementation of the initiative through innovation-led entrepreneurship development. The support available for promotion of innovations in the formal and non-formal sectors, as well as the process for protecting innovations through patenting, leading to the diffusion of innovative technologies have been discussed. The recommendations identified in the article shall ensure the flow of technology from academia to the industry, thereby transforming ideas into wealth, and creating a job market in the country.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar.

Keywords: Patent; Innovation; STIP-2013; GIAN; NIF; Make-in-India

*Speaker. Tel.: +91 9906 677322. E-mail address: [email protected]. Keynote/Expert Lecture ID: SEEDS/COMMUNE-009 Delivered in Joint Session of SEEDS-2015 and COMMUNE-2015

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

17

2015

Expert Lecture - 2 Current State of Cyber Crimes in the State of Jammu and Kashmir

Mr. Rayees Mohammad Bhat* IPS, Superintendent of Police, Hazratbal, Srinagar

Expert Lecture It has been often said that the basic nature of crime is theft... that all crime is theft. Only the dimensions of theft change, the paradigms keep shifting and the convoluted course of time shapes the contours. Every innovation in history has brought a flurry of change in human lifestyle and society. Law and crime being barometers of society have naturally witnessed similar changes. Proverbially crime has stayed one-step ahead of the law. In practice, law has always been a subject of supplydemand dynamics; and once in, it has anticipated a larger slew of likely demands and adapted accordingly. The initiation of technological daily life and the rapid movement along the information highway has led to a generational leap in the past decade. It was pretty rare to find a PC let alone one with an Internet connection then. Now it is a sine qua non. We not only have information literally at our fingertips but we can actually “use” that information in the real sense – buying tickets, comparing prices of that car we like, waiting for online sales to buy our favourite jeans – it’s all happening. And, yet it has brought on its attendant problems of that basic nature of crime – theft. Identity theft, password theft, credit card fraud, bank details phishing, spamming, online cheating, skimming, Nigerian scams, MMS “scandals”, Internet pornography, spy cams, misuse of communication devices, cyber terrorism, cyber wars and what not have become common knowledge nowadays. And, J&K has not been left untouched in this highly evolving scenario. All the above crimes have been perpetrated and are occurring here every day. The trends in cybercrimes show a disconcerting increase over the last few years. Not only cybercrime per se, but usual or traditional crimes unsupported by computer resources and communication devices are now almost unheard of! Emulation of modus operandi and learning criminal / anti-social tricks online has also become a trend. And, J&K Police has been using almost all the tools available at the global level to counter the threats posed by this nouveau aspect in crime.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar.

Keywords: Cybercrimes; Cyber Terrorism; Cyber Wars; Identity Theft; Spamming

*Speaker. Tel.: +91 84918 40107. E-mail address: [email protected]. Keynote/Expert Lecture ID: SEEDS/COMMUNE-010

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

19

2015

Expert Lecture - 3 Detecting Forgery in Images: A Statistical Perspective Prof. Ajaz Hussain Mir* Department of Electronics and Communication Engineering, National Institute of Technology, Srinagar, India.

Expert Lecture Images have become inseparable part of our life. Almost in all fields ranging from media to evidence in courtrooms, images play an important role. Whereas digitization of images have opened new vistas in image, processing and analysis techniques that enable to extract concealed information from images that may even be beyond visual perception. However, because of the easy availability of photo editing software’s digitized images have, at the same time, become vulnerable to image tampering. Attacker may tamper the images to mislead the public, distort truth, and destroy someone's reputation without leaving any trace. This puts authenticity of any image in doubt. Although techniques like watermarking and stenography can be used to check authenticity of an image but these techniques are no longer viable for every generated image in view of cost in terms of time and complexity. This limitation is overcome by digital image forensics. We need a reliable forensic technique that is able to act as an evidence to image authenticity. A number of forensic image authenticity techniques have been proposed. These work with varying degree of reliability. In our approach, we base our solution on the hypothesis that tampering may change underlying statistics of an image; though traces left by tampering may not be perceptible. It may be pointed out that a number of image forgery techniques exist. However, to test the proposed technique we have used two most commonly used forgery techniques Copy-Move and Splicing on images taken from two standard databases CASIA and CoMoFoD. To test the proposed hypothesis, efficacy of Grey Level Run Length Method (GLRLM) based on second order statistics has been used to detect forgery in images. The features obtained based on GLRLM have demonstrated the potential of proposed method in detecting image forgeries. © 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar.

Keywords: Forgery in Images; Digitized Images; Digital Image Forensics; Watermarking; Stenography

*Speaker. Tel.: +91 9419 010409. E-mail address: [email protected]. Keynote/Expert Lecture ID: SEEDS/COMMUNE-011 Delivered in Joint Session of SEEDS-2015 and COMMUNE-2015

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

21

2015

Articles Title

Authors

Pages

Author Name Disambiguation using a Mix of Hard and Fuzzy Clustering

Tasleem Arif Rashid Ali M. Asger Ghazi Majid Bashir Malik Atul Kumar Kapil Dev Goyal Sajjad Ahmed Mohammad Ahsan Chishti N.A. Kant F.A. Khanday

29-33

M. Tariq Banday Shafiya Afzal Sheikh Javid Ahmad Rather M. Tariq Banday Tawheed Jan Shah Z.A.Bangi F.A.Khanday M. Tariq Banday Farooq Aadil

52-60

Rakesh Prasher Devi Dass Rakesh Vaid Burhan Khurshid Roohie Naaz Mir

78-81

Information Diffusion Modelling and Social Network Parameters (A Survey) Performance Analysis of DPI Overhead on both Elastic and InElastic Network Traffic: A Delay Sensitive Classification and Inspection Algorithm (DSCI)

Mudasir Wani Manzoor Ahmad

87-91

Ashaq Hussain Dar Zubair Manzoor Shah

92-96

Integrated Tactile and Pointing Interface System using NonInvasive Approach

G. Mohiuddin Bhat Rouf Ul Alam Bhat Uferah Maqbool Fayiqa Naqshbandi Naheeda Reshi Fozia Abid Baba

97-102

A Compound of Negative Binomial Distribution with Two Parameter Lindley Distribution as a Tool for Handling over Dispersion

Adil Rashid T. R. Jan Musavir Ahmed

103-109

Grammatical Structure in the Dependency Framework: A Computational Perspective in Kashmiri

Aadil Amin Kak Sumaya Jehangir Mansoor Farooq Sumaira Nabi

110-114

Confusion Matrix based Suggestion Generation for OCR Errors Hybrid Wireless Mesh Protocol in Static IEEE 802.11s Networks Ultra Low-Voltage, Robust and Integrable/Programmable Neural Network based Design of 2:1 Multiplexer File Tracking System for University of Kashmir: Design Guidelines and Model Implementation Color Image Compression using EZW and SPIHT Techniques A Novel Universal (FNZ) Gate Based Adders in QCA Technology A Study of CMOS Frequency Synthesizers in Short Range Wireless Communication A Comparative Study of InSb, InAs and Si based Nanowire MOSFET Optimizing FPGA based Fixed-Point Multiplier using Embedded Primitive and Macro-support

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

34-39 40-44 45-51

61-65 66-70 71-77

82-86

23

2015 Title

Authors

Pages

Phrase Structure in Kashmiri: A UNL Approach

Aadil Amin Kak Sumaira Nabi Mansoor Farooq Sumia Tariq Adil H. Khan T. R. Jan

115-118

Roshani Gupta Rockey Gupta Susheel Sharma Shah Jahan Wani M. A. Peer K. A. Khan Nusrat Parveen Syed Zaffer Iqbal

126-130

Javeed Reshi M. Tariq Banday F. A. Khanday Nadiya Mehraj Faizan Kitab Zia Malik M. Tariq Banday Sukhdev Singh Dharam Veer Sharma Javaid A. Sheikh Shabir A. Parah Uzma Aijaz Tawseef Farah Sanna Aiman G. Mohiuddin Bhat Liyaqat Nazir Roohie Naaz Mir

143-148

Zahid Ashraf Wani Tazeem Zainab

170-174

Muzamil Ahmad Shameem Yousf Sheikh Nasrullah Shifaa Basharat Manzoor A. Chachoo Jayesh C. Prajapti Ekta Khimani Shivani Raval Navdeep Lata Simpel Rani Jindal Deepti Sharma Rakesh Vaid

175-179

Estimation of Stress-Strength Reliability using Finite Mixture of Exponential and Gamma Distributions Design of XOR Gate using Floating-Gate MOSFET

Cellular Automata: Evolution and Parallel Dimensions

High Impedance First-Order Transadmittance-Mode Allpass Filter using CCII and OTA Sequential Circuit Design using Quantum Dot Cellular Automata (QCA) Performance Evaluation of OLSR, DSR and AODV MANET Protocols

Binarization of Natural Science Images Effect of Convolutional Encoding on Bit Error Rate (BER) for Image Transmission using Multiple Input Multiple Output Orthogonal Frequency Division Multiplexing (MIMO-OFDM) System Over Fading Channels.

FPGA Evaluation of Wave Front Allocator for Crossbar based OnChip Switches The Social Web: Expressive use among the Undergraduate Students of University of Kashmir Free Text Plagiarism Detection using Lexical Database and Document Fingerprinting On Linear Classifiers vs. Hybrid Configuration: An Empirical Study Study and Analysis of Downstream ROF PON using TWDM concept Recognition of Typewritten Gurmukhi Characters Effect of Buried Oxide (BOX) in the Drift Region of a Super Junction MOSFET

24

Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar

119-125

131-137

138-142

149-154

155-160 161-164

165-169

180-184 185-188

189-194 195-199

2015 Title

Authors

Pages

On Parameter Estimation of Erlang Distribution using Bayesian Method under different Loss Functions

Kaisar Ahmad S. P. Ahmad A. Ahmed Suriya Jabeen T. R. Jan

200-206

Richa Gupta Deepika Jamwal Rakesh Vaid Harjeet Kaur Rockey Gupta Susheel Sharma Rajandeep Singh M. L. Singh Sajaad Ahmed Lone Faroze Ahmad Renu Rakesh Vaid

214-217

T-X Family of Gamma-Exponential Distribution and Its Structural Properties Impact of Body Thickness on the Performance of InAs Gate-AllAround Nanowire Field Effect Transistor FGMOS based Log Domain Integrator

Optimization of Thulium Doped Fiber Amplifier for S-Band Multimedia Stream Transmission in Mobile Ad hoc Networks

Review on the Electrical Properties of Ultra-Thin Silicon Oxynitride Films Babu Lal Sharma Design of Frequency Reconfigurable Dual Band Microstrip Patch Girish Parmar Antenna with Directional Patterns

Study the Effect of Diode Area on the Current-Voltage and Capacitance-Voltage Characteristics of Al/n-SnSe2/In Thin Film Schottky Diodes Performance Evaluation of Multiplexer Designs in Quantum-Dot Cellular Automata (QCA) Impact of Scaling Gate Oxide Thickness on the performance of Silicon based Triple gate/Quad gate Rectangular-NWFET Detection of Software Cloning by using Visual Detection Technique Support Vector Machine based Multi-Unit Iris Biometric Verification using GLCM for Feature Extraction FloSwitch Board Design using Multi FPGA

On the Realization of Robust Watermarking System for RGB Medical Images

Realization of a Fragile Medical Image Watermarking System for Content Authentication

Content Centric Networking and Interest Flooding in Communication Networks : A Review

Mithilesh Kumar Baljeet Singh Sinwar R.Sachdeva U. Parihar N. Padha

207-213

218-222

223-225 226-231 232-235 236-239

240-244

M. R. Beigh M. Mustafa

245-249

Deepika Jamwal Richa Gupta Rakesh Vaid Harish Patidar Amit Mishra Shiv Kumar Shoaib Amin Banday Ajaz Hussain Mir

250-254

Mursal Ayub Jagannatham V. V Rajshekhar Shabir A. Parah Javaid A. Sheikh Farhana Ahad G. Mohiuddin Bhat Shabir A. Parah Javaid A. Sheikh Zahid Hussain Syed Mohsin Rohit Agnihotri Kshitij Pathak Prashant Bansod Chetan Chouhan

265-269

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

255-259

260-264

270-274

275-279

280-285

25

2015 Title

Authors

Pages

Comparative Performance Analysis of MEMS Piezo Resistive Pressure Sensors Model Order Reduction of Large Scale Continuous Time Interval System

M. Tariq Banday S. Umira R. Qadri

286-291

Rajesh Bhatt Girish Parmar Rajeev Gupta M. R. Dar F. A. Khanday Khalid Sultan Sajad Ahmad Mir Zubida Habib M. Ikram A. H. Moon Ummer Khan Zaffar Kanth Sheikh Junaid Shah Jahan Wani Zahoor Ahmad Peer Fasel Qadir K. A. Khan Shabir A. Parah Javaid A. Sheikh Nazir A. Loan G. Mohiuddin Bhat Javaid A. Sheikh Jai Preet Kour Wazir Shabir A. Parah G. Mohiuddin Bhat M. Y. Kathjoo F. A. Khanday Majid Bashir Malik M. Asger Ghazi Rashid Ali Tasleem Arif Shabir A. Parah Javaid A. Sheikh Jahangir A. Akhoon G. Mohiuddin Bhat M. Tariq Banday Reyaz Ahmad Mathangi

292-298

Javaid A. Sheikh Shabir A. Parah Sakeena Akhtar G. Mohiuddin Bhat I. N. Beigh F. A. Khanday

352-355

Kamal Deep Garg Ajit Kumar Usha Parihar R. Sachdeva C. J. Panchal N. Padha

362-365

Design of a Fractional Order Ramp Generator Structural and Dielectric Studies of RFeO3 (R=Pr, Eu and Ho)

WSN Based Secure Ambient Intelligent Hospitals

QCA Full Adder Design and Noise Problems

A Blind Watermarking Technique in Spatial Domain Using InterBlock Pixel Value Differencing

On the Study and Performance Evaluation of Multirate Filter

Biomedical Sensor Interfacing Circuitry: A Watch Through Privacy Preserving Data Mining using Fuzzy based Approach

High Capacity Data Hiding using Random Plane Indicator Technique for Color Images

Control of IP Address Spoofing - A Comparative Study of IPv4 and IPv6 Networks Performance Evaluation and Comparison of Speech Compression using Linear Predictive Coding and Discrete Wavelet Transform

0.5V Design of Signal Conditioning Circuit for ECG Signal Retrieval A Survey of Spell Checkers Available for Hindi and Punjabi Effect of Semiconductor Thickness on Al/p-CuIn0.81A10.19Se2 Schottky Diodes

26

Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar

299-304 305-308

309-312

313-320

321-326

327-330

331-335 336-338

339-343

344-351

356-361

366-369

2015 Title

Authors

Pages

Question Classification using Knowledge based Semantic Kernel

Mudasir Mohd Zahid Maqbool Shraddha Arya Indu Chhabra G. S. Lehal

370-374

Deepali Vishal Goyal Ajit Kumar Sunita Gaurav Bharadwaj M. M. Sharma Sanjeev Kumar Sharma Gurpreet Singh Lehal Devi Dass Rakesh Prasher Rakesh Vaid Umrinderpal Singh Vishal Goyal Gurpreet Singh Lehal Rohit Sachdeva Dharam Veer Sharma

380-384

Sofi Shabir Roohie Naaz Mir

411-414

G. Mohiuddin Bhat Rouf Ul Alam Bhat Naazira Badar Malik Rabaie Mushtaq Afzan Hussain Hamadani Manzoor A. Chachoo Farah Fayaz Quraishi Summera Ashraf

415-419

Computational Approaches for Emotion Detection and Classification in Textual Data

Abid Hussain Wani Rana Hashmy

425-428

Achievements and Limitation of the First Machine Translation System to Convert Hindi into Dogri

Preeti Devanand

429-432

Computational Aspect of different Neuroimaging Modalities to Study the Structural and Functional Brain Networks: Analysis and Measure Based on Advanced Statistical and Classical Graph Theory Approach

Vinay Shukla Shrawan Kumar Dharmendar Singh

433-435

Study of the Emergence of Sky Computing

Vivek Chalotra

436-440

Digital Identity called Selfie-Means of Narcissism, Self-Exploration or Entertainment? A Review

Aadil Masood Wani Benish Ali Bhat

441-444

Recognition of Offline Handwritten Devanagari Numerals using Statistical Techniques Sentiment Analysis of Views Written in Gurmukhi Script

Rectangular Patch Antenna using Metamaterial for Multi Band Applications Identification of Clause Boundary in Punjabi Language Parameter Optimization of Ballistic Carbon Nanotube Field Effect Transistor Morphological Analysis of Proper Nouns in Punjabi

Issues in Word Segmentation of Handwritten Text in Devanagari Script Remote Monitoring of Water Pollution in Dal Lake using Wireless Sensor Networks in Realtime Machine to Machine(M2M) Control & Communication for Internet of Things (IoT) Using DTMF

Framework for Web Security Using Multimedia Password Authentication

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

375-379

385-387

388-393 394-397

398-404

405-410

420-424

27

2015 Title

Authors

Pages

Experimental Study of Different Wavelets for Real Time Environmental Monitoring in Wireless Visual Sensor Networks

Umar Farooq Shabir Ahmad Sofi Roohie Naaz Mir

445-450

Search Interfaces of Select Online Public Access Catalogues: An Assessment

Huma Shafiq Sheikh Mohammad Shafi Shazia Rasool Tariq Shafi Zahid Ashraf Wani Samiah Jan Nasti M. Asger Ghazi Muheet Ahmed Butt Majid Zaman Baba Mir Aadil M. Asger Ghazi Vishal Goyal Afshan Amin Khan Shivendra Pandey Joytirmoy Pathak Suresh Kumar Virender Kundu Mamta Nikhil Chauhan Saba Mushtaq Shoaib Amin Banday Ajaz Hussain Mir Anupama Chadha Suresh Kumar

451-455

Padma Prasad Sathisha

486-490

Javaid A. Sheikh Mehboob ul Amin Shabir A. Parah G. Mohiuddin Bhat

491-494

Big Data: A Growing Tide not Hype

English-Kashmiri Machine Translation: Issues and Challenges

A Comparative Analysis of Full Adder Cells in Nano-Scale for Cascaded Applications Synthesis and Characterization of Chemical Bath Deposited CuZnSnS Nano/Microstructures

Verification using Multimodal Biometric Fusion

Extension to the K-Means Algorithm for Automatic Generation of Clusters for Mixed Datasets Implementation of an Embedded Device to Prevent Friendly Fire in Battle Field Improving the Network Capacity and Coverage of 4G-LTEAdvanced Systems using Relaying Technology

28

Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar

456-459

460-464

465-471

472-474

475-479

480-485

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Author Name Disambiguation using a Mix of Hard and Fuzzy Clustering Tasleem Arifa, Rashid Alib, Mohammed Asgerc, Majid Bashir Malikd* a Department of Information Technology, BGSB University Rajouri, Jammu & Kashmir, India Department of Computer Engineering,Aligrah MuslimUniversity,Aligarh, Uttar Pradesh, India c School of Mathematical Science & Engineering, BGSB University Rajouri, Jammu & Kashmir, India d Department of Computer Science, BGSB University Rajouri, Jammu & Kashmir, India b

Abstract Author name ambiguity has long been a problem confronting the effective management of digital literature and digital libraries. Uncertainty about the real authors of a publication sometimes lead to wrong credits to authors or otherwise. Previous studies have tried to solve this problem by using traditional computational techniques. Soft Computing promises to be a good option one can look forward to deal with the problems of uncertainty. In this paper, we present the result of our ongoing work for resolving name ambiguity problem in digital citations. We propose a model that uses both traditional and fuzzy clustering approaches in a two stage framework to solve this problem. The results of our name disambiguation approach which we obtained on DBLP data are very encouraging and we have been able to achieve very good disambiguation performance in comparison to other baseline methods. On an average the values of Precision, Recall and F1 were 94.17, 91.56 and 92.42 percent respectively.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Name Disambiguation; Soft Computing; Fuzzy Clustering; DBLP

1. Introduction The advent of Information Technology has paved the way for proliferation of scientific knowledge and researchers find themselves highly benefited from the use of Information & Communication Technology (ICT) for furthering their research activities (Chang, Huang, 2014). It has been argued (Zhao, Strotmann, 2014) that advances in ICT has led to an increase in research productivity, increased level of research collaborations between researchers geographically far apart from each other, increase in citations, etc. This has also led to accumulation of large amount of bibliographic data in digital libraries like DBLP, CiteSeerX, Microsoft Academic Search, etc. ICT which has made the work of researcher more worthwhile has also compounded the problem for digital libraries by either mixing or splitting the research publications of authors sharing a common name. This is because of the reason that more and more authors with similar names are contributing to scientific knowledge by way of publishing their research work. This is evident from a steep rise in the number of publications in the recent past (Tang, Walsh, 2010). In research publications or bibliographies, the name ambiguity problem arises in two different forms, (a) when same name is expressed in different formats and (b), when different authors express their name in similar ways (Shin et al, 2014). In first case, the ambiguity arises because of not following a uniform naming pattern by an author. This could happen because of different naming conventions by different journals, conferences, book publishers etc. (Han et al, 2003). A case in point is an author Richard Taylor, Professor Emeritus, Information and Computer Sciences, University of California, Irvine. The publications of Richard Taylor appear under six different name variations: Richard N. Taylor; Taylor, R. N.; R.N. Taylor; Richard Taylor; Taylor, R.; and R. Taylor, even on his homepage†, leave aside digital libraries. In second case, the ambiguity arises because of multiple authors sharing a common name (Han et al, 2003). This can happen because of limited number of name options that our parents have while choosing a name for us (Arif et

* Corresponding author. Tel.: +91 9419 182881 E-mail address: [email protected].

ISBN: 978-93-82288-63-3

Tasleem et al/COMMUNE – 2015

al, 2014). In DBLP‡ there are nine different Richard Taylor. Along with the Richard Taylor mentioned in the above example, one Richard Taylor is a senior research fellow at Stockholm Research Institute, one with Institute for Information Policy, College of Communications, Pennsylvania State University, one with University of Houston, etc. These problems have long been impeding the efficient information management and retrieval in digital libraries (Shin et al, 2014). It requires efficient solutions capable of doing correct attribution and classification of publications of ambiguous authors especially when the information available to deal with such a problem is limited and imprecise in some cases. The rest of the paper is organized as follows: in second section, we briefly present the related work; in third section, we present our proposed approach for resolving the name ambiguity problem; in fourth section, we present the experimental results, and in the last section, we conclude the paper. 2. Related Work Efforts for resolving the name ambiguity problem in digital libraries are not new and a number of studies previously have tried to solve the problem broadly using three different techniques: supervised learning (Han et al, 2004; Veloso et al, 2012; Peng et al, 2012), unsupervised learning (Han et al, 2005; Tan et al, 2006; Masada et al, 2007; Soler, 2007; Pereira et al, 2009; Cota et al, 2010) and graphic oriented (Yin et al,2007; Fan et al, 2011). Supervised techniques try to learn a model based on both positive and negative training examples. Han et al. proposed two name disambiguation models, one based on Bayesian probability, and the other on support vector machines. The technique proposed by Veloso et al. uses a supervised rule based classifier. Peng et al. proposed a model based on Web correlations and authorship correlations using a classifier. These methods try to infer the authors of a publication by using various publication attributes like author(s), title, venue, etc. Han et al. (2005) proposed a K-way spectral clustering based name disambiguation mechanism that uses the same kind of information used by Han et al. (2004). The method proposed by Masada et al. uses a two-variable mixture model (by adding two variables), an extension of naïve Bayes mixture model. Another unsupervised model proposed by Soler groups publications iteratively based on the similarity between various publication attributes like author(s), email, title, venue, year of publication, keywords etc. The method proposed by Tan et al. uses a search engine to extract additional information from the Web. On the basis of the information so generated, hierarchical agglomerative clustering (HAC) is used to create clusters of publications. The method proposed by Pereira et al. also obtains additional information from the Web for resolving the author name ambiguity problem. Information is extracted from specific documents on the Web, e.g. CV, by submitting a query to a search engine. The query contains paper title, name of the author and venue. HAC is used to group ambiguous publications which appear on the same Web source. HAC is also used by Cota et al. The clusters are generated in a bottom-up fashion by first fusing them on the basis of similar co-authors, then title of publication and venue of publication. The process is repeated until no more fusions are possible based on the similarity score. The model proposed by Yin et al. applies SVM to weigh different types of linkages used to distinguish authors. In this model what Yin et al. call as DISTINCT, combines two complementary approaches, set resemblance and random walk probability, for measuring similarities between citation records. Another graph theoretic approach, Fan et al. proposed a method called GrapHical framework for name disambiguation (GHOST) using co-authorship information to solve the namesake problem. It first tries to exploit the relationships among publications to construct a graphical model, and solves the namesake problem by serially performing valid path selection, similarity computation, name clustering, and user feedback. GHOST uses only the co-authorship as attribute while excluding all other attributes such as e-mail, publication venue, paper title, and author affiliation, and proposes a novel sophisticated similarity metric to solve the namesake problem. Unsupervised techniques discussed above use hard clustering mechanism, HAC in majority of the cases. None of these approaches make use of fuzzy clustering. To the best of our knowledge no one till date used fuzzy clustering for name disambiguation in digital libraries. The method proposed by us uses a mixture of hard and fuzzy clustering. 3. Proposed Approach Author name disambiguation can be viewed as a classification problem in which it has to be decided whether the publication under consideration belongs to a particular group or not. Classification methods can broadly follow discriminant analysis or cluster analysis technique. Cluster analysis or clustering (commonly known term) is used in those situations where little or no information is available about group structure prior to the classification (Naes, Mavek, 1999). Traditional clustering methods have been used for author name disambiguation in a number of different ways (Arif et al, 2014a). In the proposed approach we use a mix of hard and fuzzy clustering in a two stage clustering framework. In the first stage we use hard clustering framework and in the second we use fuzzy clustering framework. Fig.1 shows ‡ ‡

http://www.ics.uci.edu/~taylor/Publications.htm http://www.informatik.uni-trier.de/~ley/pers/hs?q=richard+taylor

[30]

Tasleem et al/COMMUNE – 2015

the architecture of the proposed system. The bibliographic data for an author name is extracted from DBLP using the methodology shown in fig.2. After extraction this data is supplemented with additional publication attributes obtained in a resource bound manner (Kanani et al, 2007) from WWW using a search engine. We do not go into the details of the extraction of the additional publication features. In first stage we use the clustering process and similarity measures used in (Arif et al, 2014). In second stage, we compute the distance between all the available attributes of a publication with those of the other to combine these distances into a similarity score. This similarity score is used to calculate the value of membership function used for fuzzy clustering step.

Fig.1: Architecture of the proposed system.

Fig.2: Publications Data Extraction from DBLP

Fuzzy or soft clustering allows data elements to belong to more than one cluster simultaneously, and be associated with each cluster with certain membership levels. The degree or grade of membership which can be any value in the range [0, 1] indicates the strength of the association between that data element and a particular cluster. Soft clustering is a process of assigning these membership values, and then using these membership values to assign data elements to one or more clusters. Soft clustering has proved to be beneficial in dealing with uncertainty. There may be certain cases where agglomerative clustering used in first stage may have a good number of clusters with only one citation record. In such a case we use fuzzy clustering to find the relative similarity between clusters having a single publication and the rest by calculating the value of membership function (µ) by using equation (1) as follows:

cos cri , c j 

m

 i, j 



coscri , c r  r 1 k

m

(1)

where cri and Cj are ith publication in a single publication cluster and jth cluster (i ≠ j), respectively, and m is the fuzzy factor. The parameter m determines the “softness” of the clustering solution. If m=0, the degree of membership of a publication with all the remaining clusters is same and when m approaches ∞, the clustering becomes hard clustering (Zhao, Karypis, 2004). In general, the softness of the clustering solution is inversely proportional to fuzzy factor m. In our case, we merge a singleton cluster with any other cluster only if the value of fuzzy membership function is above a threshold. 4. Experimental Results We extracted bibliographic data from publications of ten ambiguous authors indexed by DBLP. The statistics of the dataset used are shown in Table-1. Here, #Publications refer to the number of publication records retrieved from DBLP for the author name listed in a particular record, #Actual Authors to the number of real authors, #Predicted Authors-HC to the number of authors predicted by the proposed approach after hard clustering stage and #Predicted Authors-FC to the number of authors predicted after fuzzy clustering stage.

[31]

Tasleem et al/COMMUNE – 2015 Table.1: Dataset Statistics. Author Name Jim Smith Ajay Gupta Michael Wagner Rakesh Kumar Hui Fang Jie Tang Richard Taylor Paul Jones Robert Fisher Gang Wu

#Publications

#Actual Authors

81 107 232 233 156 227 186 57 191 295

6 11 20 33 21 12 19 11 11 24

#Predicted AuthorsHC 6 14 25 31 25 13 24 13 17 40

#Predicted AuthorsFC 6 13 19 28 23 12 20 13 16 37

The performance of the proposed disambiguation approach used in this study has been shown in terms of percentage Precision, Recall and F1 scores in table-2. Table-2 presents the values of the above metrics for all the authors and the percentage change in the values of these metrics over their respective values obtained in the first stage i.e. hard clustering stage. Although these values may not represent any significant change but whatever they have been able to achieve is quite meaningful for the disambiguation process. In case of Michael Wagner, the fuzzy clustering step has been able to improve the values of precision, recall and F1 by a margin of 3.415, 5.278 and 4.368 percent respectively. On an average the improvements in precision, recall and F1 for all the ten authors is 0.86, 0.00 and 0.40 percent. We considered HAC (Tan et al., 2006), which uses agglomerative clustering and the metadata information is augmented using search engine results in a similar fashion that we used in the first stage, for comparison of disambiguation results. The comparison of the results obtained through the proposed approach with the base line method taken from (Tang et al., 2012) on all three metrics listed above is shown in table-3. The values of precision, recall and F1 obtained through the proposed approach were 94.17, 91.56 and 92.42 percent, respectively. In case of majority of the authors under consideration, the values of all the three metric were more than 90 percent. In case of Gang Wu, the low value of recall was instrumental in bringing down the value of F1 to 67.28. The low value of recall in this case can be attributed to large number of false-negative cases as more than one Gang Wu published in a similar venue. The proposed approach has been able to improve the values of precision and F1 by 21.55 and 13.78 percent respectively over HAC. However, the value of recall decreased by 0.13 percent. Table.2: Values of Precision, Recall and F1 and their percentage change in comparison with results of first stage i.e. hard clustering. Final Results (After Fuzzy Clustering) Precision Recall F1 Jim Smith Ajay Gupta Michael Wagner Rakesh Kumar Hui Fang Jie Tang Richard Taylor Paul Jones Robert Fisher Gang Wu Average

100.00 97.12 100.00 84.51 98.04 98.23 90.45 88.46 95.81 89.09 94.17

100.00 96.19 95.50 89.11 97.40 99.55 93.60 90.20 100.00 54.04 91.56

Change over Hard Clustering (%) Precision Recall F1

100.00 96.65 97.70 86.75 97.72 98.89 92.00 89.32 97.86 67.28 92.42

0.000 0.029 3.415 1.031 0.671 0.442 0.625 0.000 0.549 1.818 0.86

0.000 0.038 5.278 -1.433 -0.636 -0.448 -1.715 0.000 0.000 -1.103 0.00

0.000 0.033 4.368 -0.168 0.015 0.000 -0.525 0.000 0.281 0.000 0.40

Table.3: Comparative F1 Scores (%age) of Representative Authors with HAC

Jim Smith Ajay Gupta Michael Wagner Rakesh Kumar Hui Fang Jie Tang Richard Taylor Paul Jones Robert Fisher Gang Wu Average

Prec.

HAC Recall

F1

Prec.

Our Approach-MSC Recall

F1

92.43 41.88 18.35 63.36 100.00 100.00 80.17 36.36 96.14 97.54 72.62

86.80 100.00 60.26 92.41 100.00 100.00 99.93 80.00 100.00 97.54 91.69

89.53 59.04 28.13 75.18 100.00 100.00 88.97 50.00 98.03 97.54 78.64

100.00 97.12 100.00 84.51 98.04 98.23 90.45 88.46 95.81 89.09 94.17

100.00 96.19 95.50 89.11 97.40 99.55 93.60 90.20 100.00 54.04 91.56

100.00 96.65 97.70 86.75 97.72 98.89 92.00 89.32 97.86 67.28 92.42

[32]

Tasleem et al/COMMUNE – 2015

5. Conclusions Resolving name ambiguity has become important in view of increasing number of publications and widespread usage of digital libraries among the researchers. In this paper we proposed a hybrid clustering mechanism by employing hard clustering in first stage and soft clustering in the second. Experimental results conducted on DBLP dataset are very encouraging as the proposed approach has been able to achieve F1 score of 92.42 percent. By using soft clustering we have been able to deal with the split citation problem to a good extent. In some cases where F1 score were below expectations is due to the fact that more than one authors published with same journals or conferences which lead to distinct clusters being merged based on venue information. We are also of the view that with ever-increasing number of publications with similar authors venue information may not prove to be a good feature for disambiguation purposes. References Arif, T., Asger, M., Ali, R. (2014a). Author name disambiguation using two stage clustering. INROADS-An International Journal of Jaipur National University (Special Issue), ISSN: 2277-4904, Vol.3,no.1, pp-340-345. Arif, T., Ali, R., Asger, M. (2014b). Author name disambiguation using vector space model and hybrid similarity measures. In Proceedings of 7th International Conference on Contemporary Computing-IC3’2014, Noida, India: IEEE. pp: 135-140. Cota, R.G., Ferreira, A.A., Nascimento, C., Gonçalves, M.A., Laender, A.H.F. (2010). An unsupervised heuristic-based hierarchical method for name disambiguation in bibliographic citations. Journal of the American Society for Information Science and Technology, Vol.61, no.9, pp: 1853– 1870. Chang, H.W., Huang, M.H. (2014). Cohesive subgroups in the international collaboration network in astronomy and astrophysics. Scientometrics, Vol.101, no.3, pp: 1587-1607. Fan, X., Wang, J., Pu, X. Zhou, L., LV, B. (2011). On graph-based name disambiguation. ACM Journal of Data and Engineering Quality, Vol.2, no.2, pp: 10. Han, H., Zha, H., Giles, C.L. (2003). A model-based K-means algorithm for name disambiguation. Proceedings of 2nd International Semantic Web Conference, USA. Han, H., Giles, L., Zha, H., Li, C., Tsioutsiouliklis, K. (2004). Two supervised learning approaches for name disambiguation in author citations.” In Proceedings of Joint Conference on Digital Libraries’2004, pp: 296 – 305. Han, H., Zha, H., Giles, C.L. (2005) Name disambiguation in author citations using a K-way spectral clustering method. In Proceedings of Joint Conference on Digital Libraries’2005, pp: 334 – 343. Kanani, P., McCallum, A., Pal, C. (2007). Improving author coreference by resource-bounded information gathering from the web. Proceedings of 20th International Joint Conference on Artificial Intelligence-IJCAI, Hyderabad, India, pp: 429-434. Masada, T., Takasu, A., Adachi, J. (2007). Citation data clustering for author name disambiguation. In Proceedings of 2nd International Conference on Scalable Information Systems. Naes, T., Mevik, B-H. (1999). The flexibility of clusters illustrated by examples. Journal of Chemometrics, 13(-4), pp: 435-444. Peng, H., Lu, C., Hsu, W., Ho, J. (2012). Disambiguating authors in citations on the web and authorship correlations. Expert Systems with Applications, Vol.39, no.12, 10521-10532. Pereira, D.A., Ribeiro-Neto, B., Ziviani, N., Laender, A.H., Gonçalves, M.A., Ferreira, A.A. (2009). Using web information for author name disambiguation. In Proceedings of 9th ACM/IEEE-CS Joint Conference on Digital Libraries’2009, ACM. Shin, D., Kim, T. Choi, J., Kim, J. (2014). Author name disambiguation using a graph model with node splitting and merging based on bibliographic information. Scientometrics, Vol.100, no.1, pp: 15-50. Soler, J. (2007). Separating the articles of authors with the same name. Scientometrics, Vol.72, no.2, pp: 281-290. Tan, Y.F., Kan, M. and Lee, D. (2006). Search engine driven author disambiguation. Proceedings of ACM/IEEE Joint Conference on Digital Libraries (JCDL ’06), pp: 314-315. Tang, L., Walsh, J.P. (2010). Bibliometric fingerprints: name disambiguation based on approximate structure and equivalence of cognitive maps. Scientometrics, Vol.84, no.3, pp: 763-784. Tang, J., Fong, A.C.M., Wang, B., Zhang, J. (2012). A unified probabilistic framework for name disambiguation in digital library. IEEE Transactions on Knowledge and Data Engineering, Vol.24, no.6, pp: 975-987. Veloso, A., Ferreira, A. A., Gonçalves, M. A., Laender, H.F.A., Meira Jr., W. (2012). Cost-effective on-demand Associative Author Name Disambiguation. Information Processing and Management,Vol. 48, no.4, 2012, pp: 680– 697. Yin, X., Han, J., Yu, P.S. (2007). Object distinction: Distinguishing objects with identical names. In Proceedings of IEEE International Conference on Data Engineering, pp: 1242-1246. Zhao, D., Strotmann, A. (2014). The knowledge base and research front of information science 2006–2010: An author cocitation and bibliographic coupling analysis. Journal of the Association for Information Science and Technology, Vol.65, no.5, pp: 995–1006. Zhao, Y., Karypis, G. (2004). Soft clustering criterion functions for partitional document clustering: a summary of results. Proceedings of 13th ACM International Conference on Information and Knowledge Management, New York, USA, pp: 246-247.

[33]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Confusion Matrix Based Suggestion Generation for OCR Errors Atul Kumar, Kapil Dev Goyal* Department of Computer Science, Punjabi University, Patiala, India

Abstract This paper proposes a method for generating suggestion of errors done by OCR system based on confusion matrix for Punjabi language (Gurumukhi script). Confusion matrix is developed from large text Corpus. The proposed method firstly determines the probability of confusion of one character (OCR output) with another character (OCR input) from confusion matrix. For each word of OCR output, number of strings is generated from generated lists of topmost five confused characters for each character of input word and probability scoring of these strings is calculated for ranking. While generating strings, each string is validated with the trigram lexicon. If validated, that string is taken and after generating all the valid strings, lexicon is used for best suggestions. The topmost five words are taken as suggestions. This method has been tested for variety of OCR outputs documents. The system developed also used for Devanagri Script.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: DWT; DCT; Embedded Zero Tree; Set Partitioning in Hierarchical Tree

1. Introduction With the improvement in technology, current OCR technology is rectified but there are many old documents like books, text documents, which are of poor quality results in many errors during OCR processing. The post-processing is the last stage in OCR system. The main purpose of post-processing is to remove errors occurs during the various stages of OCR. Mainly the errors occur between the characters, which have some similar shape structures (like in Gurumukhi script ਥ with ਖ). Correction of these kinds of errors is very difficult one. The main aim of this research work is to develop a system that perform the post-processing based on confusion matrix and generates the spelling suggestions without taking into account the context. The rest paper has been organized as follows: Part 2 discusses previous work in the field of post-processing of OCR; Part 3 explains the proposed research work, Part 4 shows experimental results, Part 5 gives conclusion about research work and Part 6 contains References. 2. Previous Work Post-processors have been extensively used in the past to enhance OCR accuracies. Various OCR post-processing techniques are used from past decades. (Kernighan, et al, 1990) developed a program for spelling corrections based on a Noisy channel which proposes a list of candidate corrections and sort them according to their probability (Kenneth, et al, 1991) developed the new program called spell which took the incorrect word and proposed a list of candidates words along with probabilistic scoring. The performance of system was increased up to 89%. (Kukich, 1992) has done the survey on various kinds of errors and explained various kinds of techniques used in this area. Masaaki NAGATA used the statistical OCR model, an approximate word matching method using character shaped similarity and word segmentation algorithms. The accuracy of 90% to 97.34% has been found. (Kolak, et al, 2005) proposed lexicon free method for error correction and implemented using FSM (finite state machines). Authors have claimed the improvement of about 78% error reduction obtained for low density languages like Igbo. (Bassil, et al, 2012) used Google’s spelling suggestion scheme, which is based on the probabilistic n-gram model for predicting the next word in * Corresponding author. Tel.: +919814877883. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Kumar and Goyal/COMMUNE – 2015

a particular sequence of words. The error rate was dropped up to 3-4% by applying these algorithms. In case of Indian Language script for OCR systems, (Pal, et al, 2000) proposed a technique which was based on morphological parsing using two separate lexicons of root words and suffixes. Authors have claimed the accuracy of 84.27%. Lehal, et al, 2001) developed a shape based postprocessor for Gurumukhi OCR. The recognition accuracy of the OCR without post processing was 94.35%, which was increased to 97.34% on applying the post-processor to the recognized. Sharma et al developed shape encoding based post-processing for Gurumukhi (an Indian script).The accuracy of 4-7% was seen. (Mohan, et al, 2010) proposed a post-processing scheme which use statistical language models at the sub-character level to boost word level recognition results. Authors have claimed the accuracy rate of 92.67% for Malayalam text. (Jain, et al, 2011) proposed used the minimum edit distance technique along with substitution cost as confusion probabilities of characters. Authors have claimed to improve the accuracy of OCR by 33%. 3. Proposed Solution As mentioned in second part, various methods are there to correct the OCR errors. This paper proposes a method to generate suggestions of OCR errors based on confusion matrix. The proposed method firstly creates a confusion matrix from a large corpus text. 3.1.

Confusion matrix

Confusion matrix shows how many times one character is confused with other character. The accuracy is obtained by dividing the no. of correctly recognized characters to the total number of characters images which are actually present in the database. The confusion between character recognition is due to shape similarity of the characters. Handwritten data increases the confusion further. In order to get the accuracy, confusion matrix is obtained from large corpus of data. Table 1 shows the confusion matrix for Gurumuhi script. The first row shows actual characters in OCR input and first column shows the confused characters with which actual characters are confused. Each cell represents how many times particular row character confused with column character. Tacle 1. Confusion Matrix of Gurumukhi Script.

3.2.

ਂ

ਅ

ਆ

ਇ

ਈ

ਂ

136828

32

11

0

2

ਅ

0

37795

32

0

0

ਆ

70

231

87324

1

0

ਇ

2

5

0

44012

2

ਈ

16

0

3

4

26481

Calculation of probabilities from confusion matrix

This step involves the calculation of probabilities i.e. what is the probability of one character confused with other character. The probabilities are calculated from the confusion matrix by using the following formula

pr(y/x ) = num(sub(x,y))/num(x) Where pr is probability, num (sub(x,y)) is number of times x is substituted by y and num(x) is the total number of x. For example suppose ਅ is replaced by ਆ. So from table 2. pr(ਆ / ਅ ) = num(sub(ਅ/ ਆ)/num(ਅ) = 231/38575 (calculated) =0.00598(Result taken from implementation) Table 2. .Confusion matrix showing probabilities of confusion of column character with each row character.

ਂ

ਅ

ਆ

ਇ

ਈ

ਂ

0.98830

0.0008

.0001

0

0.000075

ਅ

0

0.9798

.0003

0

0

[35]

Kumar and Goyal/COMMUNE – 2015

3.3.

Creating Top five confusion Probabilities

Since in confusion matrix, there are many entries, which have very low probabilities as shown in table.2, so we have to ignore those probabilities along with characters. Table 3. Confusion Probability of ਖ

3.4.

Characters ਖ

Confusion Probability 0.846768456192196

ਖ਼

0.0862201061675696

ਥ

0.0452033054233021

ਬ

.00448749521151426

ਂ

0.00311935642751601

Creation of Character Trigram Lexicon

For validating word, generally the method is to look in the dictionary. But the problem is that there are not all the words in the dictionaries. In addition, no forms of verbs are available in dictionaries. For e.g. ਦੀ, ਦਾ, ਦੈ. All these words are not available but if we generate character trigrams then all these words are validated. Steps for creating trigram dictionary are: (i) For each unique word in the dictionary we create character trigrams, e.g.: ਸ਼ਹਿਰ ਸ਼ਹਿਰ

ਰ@@

@@ਸ਼ @ਸ਼ਿ ਸ਼ਹਿ

ਿਰ@

ਹਿਰ

Fig. 1.Character trigrams for ਸ਼ਹਿਰ

We have taken the dictionary of 105629 words. (ii) Each trigram generated above is stored in Binary Search Tree to avoid duplicity. (a) Generate trigram of the word as mentioned earlier. (b) Search for the trigram in the BST. (c) If trigram is found in the BST tree then reject it (d) If trigram is not found add a new node with that trigram. Finally, In-order traversal is applied on the BST tree and traversed nodes are stored in a file. We have applied this method for around 105629 words and 23031 character trigrams are obtained. 3.5.

Generation of words and Creation of Suggestions

When post processing is to be applied, first the character trigram Lexicon (Fig 1.) is loaded into Binary search tree. Also simple dictionary is loaded in binary search tree. The following algorithm is applied for generating suggestions for each word of OCR output: (i) Let S be the source word which is taken from OCR output. (ii) Firstly calculate the length of word. After that dynamically generate number of lists equal to the length of word. (iii) For each character of word S, generate the dynamic array list containing the top five confusing characters along with probabilities. Characters are stored at even positions starting from zero and corresponding probabilities at odd positions.

[36]

Kumar and Goyal/COMMUNE – 2015

(iv) After the lists are generated for each character of source word S, next step is to generate words by combining each character of every list with every character of other lists of source word S and also generating probabilities for the purpose of ranking. 3.5. 1.

Validation Process

Before generating suggestions, we have to validate each word. For each word generated in step (iv)  Generate the character trigrams as shown in figure 1.  Search each trigram in trigram dictionary loaded as Binary search Tree (Fig 2). If all the trigrams of word are present in Binary search tree then word is valid otherwise invalid. After this we used dictionary to remove wrong words. Each word generates above is searched in the lexicon. Dictionary we have taken contains around 105629 words. If the search is successful then that word is placed in final suggestion otherwise deleted. OCR text that is to be corrected

Confusion Matrix

Candidate generation with probability

Lexicon

Candidate validation

No

Character Trigram Lexicon

Validated

Yes Candidate Tree Generation

Top five Suggested Candidates

Corrected Text Fig.2. System Architecture

(v) The valid words are ranked according to probabilities and top five probabilistic valid words are taken as suggestions. For the efficiency of the Post-processing, it is important that the right suggestion is presented at the top position. Otherwise, the efficiency of system will affect. So the suggested words are again checked against Lexicon to remove non words and to improve the positions of suggestions so that correct word can be found at rank 1 or rank 2. 4. Experimental Results We have tested the performance of this post-processing on variety of OCR output documents. Following are the initials taken  Number of distinct words in dictionary = 105629  No. of nodes created for dictionary in Binary search tree=105629  No of character trigrams generated from the dictionary words = 23030

[37]

Kumar and Goyal/COMMUNE – 2015

For correct words we have taken around 3400 words from 40 OCR documents and accuracy is measured 100%. Fig 3 shows results. Here First position shows suggestion at top with probability. Second position represents suggestion at second position and so on. Not found means correct word is not obtained. The vertical bar represents number of words related to following category. Example ਹਿਆ is correct word correctly at tops along with highest probability as shown in table 4. Table 4. ਹਿਆ suggestions along with probabilities

Rank.

Word

Probability

1

ਹਿਆ

0.984526219030382

2

ਹਿਆ

0.000799556482201164

3

ਹਰਆ

0.000535872961475248(

4

ਹਿਆ

0.000484837441334748

5

ਹਿਆ

0.000365754561006916.

Correct words suggestions 4000 3000 2000 1000 0 Correct words

Fig.3. Correct words suggestions for correct words

For incorrect word are taken from same 40 OCR documents and firstly find performance of developed system without applying dictionary. Fig 4 shows results. The donations of bar chart are same as mentioned in previous chart. Example ਪਰੀਹਥਆਢਾਂ is incorrect word. The correct word is at position 4 shown in Table 5 Table 5. ਪਰੀਹਥਆਢਾਂ suggestions along with probabilities

Rank.

Word

Probability

1

ਪਰੀਹਬਆਵਾਂ

0.00105692891130884

2

ਪਰੀਹਬਆਦਾਂ

0.00054927014288491

3

ਪਰੀਹਬਆਚਾਂ

0.00035785782036441

4

ਪਰੀਹਖਆਵਾਂ

0.000238979264328572

5

ਪਰੀਹਖਆਦਾਂ

0.000124193948391227.

[38]

Kumar and Goyal/COMMUNE – 2015

Fig.4. Correct words suggestions for incorrect words

When dictionary is applied, then we have seen the improvement of first position suggestions that are correct words. Also 4th and 5th position suggestions are diminished. The overall accuracy has been improved. Table 5 shows the results. Example ਪਰੀਹਥਆਢਾਂ is incorrect word. When lexicon is not applied, correct word is found at 4 th position as mentioned in Table 4.When dictionary(lexicon) is applied , correct word is at top position as shown in Table 4

Fig.5 Correct words suggestions for incorrect words using Lexicon

5. Conclusion The developed system uses different sources like character trigrams, Lexicon dictionary, and confusion matrix. It corrects non words as well as words which have some meanings. The accuracy rate of OCR results has been improved up to 96.14%.This system does not take care of name inconsistency which can be a subject of further research. References Bassil, Youssef, Alwani. Mohammad, 2012. OCR post-processing error correction algorithm using Google’s online spelling suggestion, Journal of Emerging Trends in Computing and Information Sciences, p, 90-99. Jain. Rupi, Chaudhury.Santanu.2011.Probabilistic Approach For Correction Of Optically-Character-Recognized Strings Using Suffix Tree in Proceedings of Third National Conference on Computer Vision, Pattern Recognition, Image Processing and Graphics ,p, 74-77. Kenneth. W. Church, William. A. Gale, 1991. Probability scoring for spelling correction, Statistics and Computing, Volume 1, Issue 2, p, 93-100. Kernighan. D. Mark, Kenneth. Church. W, Gale. A. William, 1990. A Spelling Correction Program Based on a Noisy Channel Model. AT&T Bell Laboratories 600 Mountain Ave Murray Hill, N.J., USA, p. 205-210. Kolak. Olan, Resnik. Philip, 2005. OCR Post-Processing for Low Density Languages. Computer Science and UMIACS, University of Maryland College Park, MD 20742, p. 867-874. Kukich. K, 1992.Techniques for Automatically Correcting Words in Text in ACM Computing Surveys , Vol. 24, No. 4, p, 377-439. Lehal. S. G, Singh. Chandan, Lehal.Ritu.2001. A Shape Based Post Processor for Gurumukhi OCR, Proceedings of the Sixth International Conference on Document Analysis and Recognition (ICDAR’01) IEEE Computer Society Press, USA , p.1105-1109. Mohan. Karthika, Jawahar. V. C., 2010. A Post-Processing Scheme for Malayalam using Statistical Sub-character Language Models in Ninth IAPR International Workshop On Document Analysis Systems, Boston, MA ,p. 493-500. Nagata.. Masaaki, 1998. Japanese OCR Error Correction using Character Shape Similarity and Statistical Language Model. in Proceedings of the 36th annual meeting on Association for Computational Linguistics -Volume 2 , COLING-ACL, p.922 – 928. Pal.U, Kundu. K. P, Chaudhuri. B.B., 2000. OCR Error Correction of an Inflectional Indian Language Using Morphological Parsing. Journal of Information Science and engineering, p, 903-922 Sharma. Veer. Dharam, Lehal. G. S, Mehta.Sarita.2009. Shape Encoded Post Processing of Gurumukhi OCR, Proceedings of tenth International Conference on Document Analysis and Recognition p. 788-792.

[39]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Hybrid Wireless Mesh Protocol in static IEEE 802.11s Networks Sajjad Ahmeda, Mohammad Ahsan Chishtib* a

Department of Information Technology, National Institute of Technology Srinagar, India Department of Computer Science Engineering, National Institute of Technology, Srinagar

b

Abstract Wireless mesh network (WMN) is an important technology which is being used in next generation wireless networks to solve last mile problem of anywhere, anytime and low cost connectivity . IEEE has developed an extension of IEEE 802.11 Wireless network based on wireless mesh networks. The standardization work is in progress and is called IEEE 802.11s networks. The default routing protocol specified in the draft is Hybrid Wireless Mesh Protocol (HWMP) and airtime metrics as default routing metric. In this paper, author analyzed the performance of modes of Hybrid Wireless Mesh Protocol (HWMP) using Network Simulator -3(NS3).

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Routing protocol; wireless Mesh network; IEEE 802.11s; hybrid Wireless Mesh Protocol; Network Simulator

1. Introduction Wireless local area networks (WLANs) are one of the most popular technologies for providing network services are as compared to their wired counterparts because of ease, low cost deployment and their support for user mobility. WLANs are growing rapidly and are found everywhere. One of the reasons for their widespread use is the unlicensed ISM band over which it operates. However, ISM band is interference limited. Therefore, maximum allowable output power is regulated as described in IEEE 802.11 Standard which limits range of IEEE 802.11 based devices. To extend coverage further single hop communication is replaced with multi-hop communication. The Multihop communication can help extend the transmission coverage area without installing new access points as described in (Seungjoon Lee et al, 2004). To add Multihop capability to existing standard IEEE Task Group “S” is working on a new standard called IEEE 802.11s, which is in draft stage. Few implementations are still available such as OLAP, openMesh and open80211s by cozybits. The amendment being considered is based on Wireless Mesh Networks (WMN). Detailed Survey on WMN can be found in the paper by (Akyildiz at el, 2005). The Main hurdle for wide spread use of wireless communication networks was low data rate provided by wireless communication system as compared to wired networks but with recent advances in wireless communication technology wireless networks are providing high data rates as described in (Kuran et al, 2007). As a result, wireless Internet access is everywhere these days. Network Deployment is carried out by placing wireless access points at well-planned places. As per IEEE 802.11s Wireless LAN Mesh Networking standardization, WMNs additionally simplifies network establishment, administration, and maintenance. It also makes setup of networks more cost effective as mentioned in Raffaele Bruno et al. The properties such as cost effective, easy to deploy, establish, and maintain make WMN based networks ideal for emergency scenarios. As compared to traditional WLANs WMN based networks requires and planning as well as less administration as described in (Camp et al, 2008). Traditionally routing protocols are implemented at network layer to enable multihop forwarding, which makes the approach link layer independent. A recent proposal in wireless networking puts multihop forwarding in the link level, which extends the WLAN functionality of nodes. For better utilization of wireless links additional quality aware metrics which have access to link level parameters of the channel can be implemented at MAC level. Currently, The IEEE Task Group 802.11s is developing a promising standard for mesh networking at the MAC level. This new approach provides a large number of benefits. Such an approach makes it appear as a LAN for layerthree protocols. Implementing multihop routing at network layer need modify TCP\IP layer which is not required if multihop is implemented at link layer. IEEE 802.11s specifies a mandatory routing protocol Hybrid Wireless Mesh *Corresponding author. Tel.: +91 9419 023039. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Ahmed and Chishti/COMMUNE-2015

Protocol (HWMP), and a mandatory route metric for path selection called Airtime Metric. The standard also specifies a framework for implementing new path selection and path metric for future extensions, Guido R Hiertz et al. In IEEE 802.11s networks, routing is implemented at layer 2. Thus, high-level layer are shielded from the wireless mesh network at MAC level. The routing at MAC level is called Path selection (Carrano et al, 2011). The paper is organized as follow: section 2 contains background about the routing protocol hybrid wireless mesh protocol (HWMP), section 3 contains simulation setup details, results are presented in section 4 and conclusion is presented in section 5. 2. Hybrid Wireless Mesh Protocol Hybrid Wireless Mesh Protocol is a routing protocol implemented at layer-2 that is MAC Layer. The protocol is a hybrid of a reactive and a proactive routing protocol. The reactive part of the protocol is a variant of Adhoc On-demand Distances Vector (AODV) Routing protocol called Radio Aware-AODV. The proactive part of the protocol is a tree based protocol. The Combination of two approaches helps in finding an optimal and efficient paths in a wide variety of networking scenarios. Routing or path selection table is stored at MAC layer instead of IP layer. Thus, hiding the Multihop nature of MAC layer from upper layers as described in Chun-Ping Wang et al. HWMP uses airtime metric for link measurement. To differentiate between routing at network layer and routing at layer 2 the term path selection is used instead of routing. A mesh node which participate in path selection is known as Mesh Point (MP). The control messages used by HWMP are Path Request (PREQ), for finding new paths, Path Reply (PREP), packets sent by destination or by those having path to destination, in response to PREQ, Path Error (PERR), packet sent to source when paths are no more available. Route Announcement (RANN) packets are flooded into the network for announcing routes. A special metric field is used to propagate path metric information between Mesh Points(MP).Sequence number are maintained by MPs to avoid infinite loops and are exchanged with other MPs through HWMP control messages. 2.1.

Modes of HWMP

HWMP supports two modes of operations, Reactive Mode and Proactive Mode. The Proactive mode is further divided into three types viz. Simple Proactive Mode, Forced Proactive Mode, Proactive With RANN. In reactive mode a path to destination node is determined as and when needed. No tables are maintained to store paths. To find a route or path, source MP broadcasts a special routing control message called Path Request message (PREQ) with the destination field of the frame is initialized with the destination address and the metric field initialized to 0. The destination MP on receiving PREQ replies with path reply message (PREP). The intermediate nodes may be allowed to reply to the source node depending on whether a particular flag called “destination only” flag is enabled or not. In proactive mode, path to every mode is determined beforehand so that a latency free transmission of data can be commenced immediately. In this mode, a special root is designated as root mesh station which acts as Gateway for external traffic. A root mesh station proactively creates a tree with root mesh station as root of the tree and other mesh station end points of the tree. As a result a path to every mesh station is always available. Periodically root broadcasts Path request message (PREQ), telling every mesh station about the root. There are two proactive modes of building path trees in HWMP, Simple and forced proactive mode. In simple proactive mode, the mesh stations are not allowed to send Path reply message, PPREP, in response to the path request, PREQ, broadcasted by root mesh station. Every mesh station learns about the path to root and update their metrics toward root. As a result a unidirectional paths are created from all nodes to the root node. Suppressing of PREP messages is achieved by setting the Proactive Path Reply flag of the proactive PREQ control message. No proactive paths are available from root node to other nodes in the network. In case root node needs a path to some mesh node reactive approach is used to select path. In forced proactive mode bidirectional paths exists between root and every mesh station. This is achieved by setting path reply flag in the PREQ message forcing every mesh station to reply to the PREPs sent by the root. In effect a bi directional tree is created between root and every mesh station enabling both root and mesh stations to start latency free transmission. In IEEE 80.211s airtime metric is identified as default path metric. Airtime metric is given preference over other metrics as it is an efficient radio-aware metric as described in Michael Bahr et al. It represents the amount of channel resources consumed for transmitting a frame over a particular link. Best path is the path over which a frame occupies transmission medium for less duration as transmission medium in wireless communication is expensive resource. 3. Simulation The Network Simulator-3 (NS-3) was used to study the performance of modes of HWMP. Network Simulator is a discrete-event network simulator with a special focus on Internet based systems. The ns-3-dev version was used which is a development version containing the latest release of this network simulator. Wireshark, patched with mesh patch,

[41]

Ahmed and Chishti/COMMUNE-2015

was used for analysis of captures packet data In NS-3 only two modes, reactive and forced proactive, of HWMP are implemented. The simple proactive mode was implemented by editing code in hwmp-protocol.cc. After doing the changes and compiling the ns-3-dev, the simulation work was started by implementing evaluation scenario. 3.1.

Evaluation Scenario

Scenario used for the evaluation of performance of HWMP models a typical WLAN installed in a campus consisting of two kind of nodes, static and mobile nodes. Static Mesh nodes are fixed at a given position forming a grid structure and acts as infrastructure to provide service to the mobile mesh nodes. The static nodes do not generate data traffic. Mobile nodes are placed randomly in the area.

Fig 1. Simulation Scenario showing static and mobile mesh nodes in NS3

3.2.

Parameters

The Simulation scenario consists of 16 static mesh node acting as routers placed 100m apart to provide a coverage area of 500m X 500m and 30 mobile mesh nodes with RandomWayPoint mobility where speed of node is varied from 1 to 10m/s with a fixed pause of 5s. Each node with IEEE 802.11a physical layer and IEEE 80.2.11s based MAC layer is used for simulation. Path selection protocol used is Hybrid wireless mesh protocol (HWMP) with airtime as path selection metric. Each Traffic flows used for simulations consist of UDP packets with randomly initialised payload data of 512 Bytes. Traffic type is Constant bit rate traffic which generate 10 packets per seconds with random start and stops time. The amount of traffic toward root is initially 0% then it is increased with a step of 20% until all traffic is going toward root i.e. 100% traffic toward the root with an additional measurement at 10% traffic rate. All other traffic is send between other nodes which are randomly chosen. Top left static mesh node in the grid is considered as root node, simulating mesh gateway for external traffic. To take reading each simulation was repeated 20 times with different set of random variable to generate sufficient confidence in the results. Four parameters, initial path requests Path selection overhead ratio, path errors and throughput were used to compare behaviour of the mode of HWMP. The path selection overhead ratio is the ratio of bytes sent by the mesh stations as path selection packets to that of all bytes sent by the mesh stations. Throughput, in kbps, is measured as the application level data that is received by static mesh nodes from source mobile mesh nodes and successfully transferred to the sink mesh node. 4. Results Different modes of HWMP experience different path selection overhead as in fig 2. Reactive mode experience least overhead as compared to proactive modes. Among the proactive mode, overhead experienced by forced proactive mode is more than simple proactive mode. Also the path selection overhead also decreases with higher fraction of root traffic.

[42]

Ahmed and Chishti/COMMUNE-2015

Fig 2. Path selection overhead ratio

Fig 3. Number of Path Errors

Fig 4. Number of initial path requests

Fig 5. Throughput at different Hops

[43]

Ahmed and Chishti/COMMUNE-2015

A Path Error packet, PERR, is generated when a path break Path errors decrease as fraction of traffic toward root increases. Proactive mode generates less number of PERRs than Reactive mode. Forced Proactive mode generated least number of PERRs as in fig 3. For a given mode the number of initial path request remain almost constant with a slight decrease as percentage of traffic toward root increases refereeing fig 4. The reactive mode generated least number of path requests whereas proactive mode generated much more number of path requests. The Throughput, fig 5, is evaluated with all traffic sent toward root. Thus, focusing on evaluating throughput experienced by static router at different hops. Irrespective of HWMP mode, throughput decreases as number of hops increases. The Proactive modes show better performance in a root centric environment. Decrease in throughput is more in case of reactive mode. Throughput of simple is less then forced proactive mode because of unidirectional tree and reactive path setup when data is transferred from root to the nodes. 5. Conclusion After analyzing the results author come to following conclusion: Forced proactive mode performs better than simple proactive mode which in turn performs better then reactive mode. As number of hops increases throughput decreases rapidly irrespective of mode of HWMP. Forced proactive mode performs well in the network condition when 100% traffic is directed toward the root node showing higher throughput and least path errors as compared to other modes. However, network must be able to handle high path selection overhead load. In case most of the traffic is outbound and is destined to or from root node, forced proactive mode is a good choice because of better throughput. While analyzing the throughput of the mesh network it is found that the per hop throughput of IEEE 802.11s network do not scale well as number of hops increase. References Perkins, C., Belding-Royer, E., S. Das, S., 2003. AdHoc On-Demand Distance Vector (AODV) Routing, IETF RFC 3561. Chun-Ping Wang, Brett Hagelstein, Mehran Abolhasan, Experimental Evaluation of IEEE 802.11s Path Selection Protocols in a Mesh Testbed, 4th International Conference on Signal Processing and Communication Systems, ICSPCS, IEEE, pp.1-3, 2010. Guido R Hiertz, Sebastian Max, Rakesh Taori, Feb. 2010. IEEE 802.11s: The WLAN Mesh standard, IEEE wireless communications pp. 104-111. Akyildiz I. F., Wang, X., Sept. 2005.A survey on wireless mesh networks, IEEE Commun. Mag, vol. 43, no. 9, pp. S23–S30. IEEE Computer Society, Oct. 2009. IEEE P802.11s/D3.04 Draft Standard: Wireless LAN Medium AccessControl (MAC) and Physical Layer (PHY) specifications, Amendment 10: Mesh Networking. Camp, J. D., Knightly,E.W., August 2008. The IEEE 802.11s Extended Service Set Mesh Networking Standard,IEEE Communications Magazine, vol. 46, no.8, pp. 120–126. Kai Yang, Jian-feng Ma, Zi-hui Miao.,(2009). Hybrid Routing Protocol for Wireless Mesh Network, International Conference on Computational Intelligence and Security. Kuran M.S., Tugcu T., 2007.A survey on emerging broadband wireless access technologies,Comput. Netw., vol. 51, no. 11, pp. 3013–3046. Michael Bahr, 2007. Update on the Hybrid Wireless Mesh Protocol of IEEE 802.11, IEEE 2007 Siemens Corporate Technology, Information & Communications. OpenMesh,Retrived from url: http://www.open-mesh.com. Raffaele Bruno, Marco Conti, Enrico Gregori, March 2005. Mesh Networks: Commodity Multihop Ad Hoc Networks, IEEE Communications Magazine, pp.123-131. Ricardo C. Carrano, Luiz C. S. Magalhães, Débora C. Muchaluat Saade ,2011. Célio V. N. Albuquerque, IEEE 802.11s Multihop MAC: A Tutorial IEEE Communications Surveys & Tutorials, VOL. 13, no. 1, First Quarter. Seungjoon Lee, Suman Banerjee, Bobby Bhattacharjee, 2004.The Case for a Multi-hop Wireless Local Area Network, IEEE.

[44]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16-18March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Ultra Low-Voltage, Robust and Integrable/Programmable Neural Network based Design of 2:1 Multiplexer N. A. Kant*, F. A. Khanday Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India

Abstract Hardware implementation of neural networks is one of the contemporary areas of research for the scientists. After a careful survey of the open literature, it is found that apart from some digital implementations and the analog implementation of activation functions/basic logic gates, very less work has been reported for the high-order hardware implementations of neural networks. Digital implementations of neural networks use the sequential machines and therefore loose the basic essence of neural networks. Consequently, there has been an increasing interest in the analog implementations of neural networks. In analog implementation of neural networks, focus is on reducing the supply voltage/power consumption, as the complexity of neural network based circuit is often more than the conventional CMOS based design. Besides the technology is also scaled down to reduce the area of the circuit on the chip. This paper outlines the design and simulated performance of ultra low-voltage multiplexer function designed from the neural network using Sinh domain technique. Two different activation functions are used for the design, where the activation functions are also designed using the Sinh-Domain companding technique. The circuit is designed in 0.35µm AMS CMOS process and the simulation results are obtained from Hspice simulation software.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Artificial Neural Network; Hardware Implementation of Neural Networks; Activation Functions; Analog Integrated Circuits; Companding Techniques; Sinh-Domain Technique; Digital Logic Functions; Multiplexers

1. Introduction Artificial Neural Networks (ANNs) being one of the most widely studied topic for research, now from past few decades, has found its applications in vast variety of fields and is gaining more attention with the advancements in the modern technologies. They have been used for different variety of problems like in the areas of pattern recognition, classification, signal processing, image processing, control systems, biomedical engineering etc. by employing their software based designs. We have many examples in open literature showing neural network codes running on von Neumann computers. In spite of many years of studies involving Neural Networks (NNs), their actual implementation for study and their applications have become practical owing to the advancement in programmable hardware of these networks. As the key features of neural networks are parallel processing, continuous-time dynamics, and global interaction of network elements [Hopfield, 1982] and all these can be fully utilized only by its hardware design rather than software approach. Therefore, it is important to look out for the methods by which one can achieve the hardware design of NNs. But then again also to mention that the hardware designs of these networks is going at a dawdling pace and there are still few commercially available neural networks implemented in hardware [Chao-Ming eta al., 2007, Seul and Sung, 2007]. The biggest concern while designing the neural network is the complexity and their training. As the complexity increases the concern shifts towards the power consumption and the voltage of operation. In the aforementioned context, flexibility and power consumption (in order to satisfy a wide range of applications) are the important aspects to be kept into consideration in the hardware implementation of neural networks. The designer must thus look for the design technique to achieve the said goal of low-power low-voltage hardware circuit for neural network. *

Corresponding author. Tel.: +91 9906 937561. E-mail address: [email protected].

ISBN: 978-93-82288-63-3

Khanday and Kant/COMMUNE – 2015

Hardware implementation of ANNs can be achieved through analog or digital means. However, analog realization of the NNs provides a fast and power efficient realization compared to the digital realization whereas the feature of flexibility is added by employing tunable analog-hardware. Now to achieve low-power low-voltage hardware design of these networks companding design is a good choice. Companding design techniques are interesting subclass of analog design with potential for low-voltage operation capability. This is originated from the fact that it is the compressed signals that are processed, which results in the reduction of swing compared with that of the signals in the conventional linear circuit design. NNs can be used to solve any mathematical/engineering problem, multiplexer being one of them. Multiplexer, frequently termed as MUX, an important logic building block which find its application in almost all the digital integrated circuits. MUX find its application in the field of telecommunications and are extremely important, as they are very helpful in reducing network complexity by minimizing the number of communications links needed between two points and in turn reducing the cost. Along all the computing systems, Multiplexer have evolved with time. Each new generation has additional intelligence, and additional intelligence brings more benefits. Few of the accrued benefits are [www.informit.com] a) b) c)

Data compression: The capability to do data compression, which enables us to encode certain characters with smaller number of bits than normally required. The freed capacity can be utilized for movement of other information. Error detection and correction: Data integrity and accuracy are maintained by error detection and correction between the two points that have being connected. Managing transmission resources: The capability to manage transmission resources on a dynamic basis, by introducing the additional features such as priority levels.

Device cost versus line cost is an important issue in communication for transferring signal from one point to the other. We can either provide extremely high levels of service by ensuring everybody always has a continuous and live communications link. This however has a drawback of extremely high cost. The other option we can opt to offset the costs associated with providing large numbers of lines is by instead using devices such as multiplexers that helps in making more intelligent use of a smaller group of lines. The more intelligent the multiplexer, the more perceptively and vigorously it can work on our behalf to dynamically make use of the available transmission resources. In this paper, first time in the literature, a NN based 2:1 MUX design is reported. This has been achieved by employing Sinh-Domain technique. The design has been implemented using two activation functions and important performance parameters have been calculated. The paper is organized as follows: the brief idea about Sinh-Domain companding along with the NN is presented in Section 2. The circuit description of 2:1 MUX along with the simulated results is presented in Section 3. The paper is finally concluded by section 4 and references. 2. Sinh-Domain Companding Technique and NN Design There are a number of techniques which have been used in the literature to achieve the low-voltage, low-power circuit designs. Companding being a technique which is the preferred one for low voltage, low power circuit design due to its inherent advantages of low-voltage requirement, grounded passive component requirement, class AB behavior, electronic tunability etc.. Sinh-Domain companding which is a sub class of companding technique comes in the category of instantaneous companding where the compression of the input current could be performed through the inverse of the hyperbolic sine function realized by translinear loops formed by bipolar transistors in active region or MOS transistors in weak inversion [Serdijn et al., 1999, Katsiamis et al., 2008, Frey, 1996]. The non-linear transconductor cell, which forms the basic building block of sinh-domain technique, is depicted in Fig. 1. This circuit realize the expression given in (1) in the case of the hyperbolic sine output

 vˆ  vˆ IN  i  2 I o sinh IN   nVT

  

(1)

 vˆ  vˆ  i  2 I o cosh IN  IN    nVT 

(2)

(2) in the case for hyperbolic cosine output

(3) in the case for inverted hyperbolic sine output  vˆ  vˆ  i  2 I o sinh IN  IN    nVT 

and (4) in the case for weighted hyperbolic sine output

[46]

(3)

Khanday and Kant/COMMUNE – 2015

 vˆ  vˆ  i  2 KI o sinh IN  IN    nVT 

(4)

where Io is a dc current, VT is the thermal voltage (26mV @ 27oC), n is the subthreshold slope factor (1
ˆIN  , ˆIN  are the voltages at the non-inverting and inverting inputs, respectively.

ANNs are the mathematical models, electronic circuits or the computer codes that perform the task in a manner to mimic the behavior of human neural network (brain). Fig. 2 (a) represents a single neuron cell whose mathematical expression can be given as    2 i Ni .Wi ,1 i out  AF    b   I  o   i 1



  Vˆ   

(5)

and a two layer perceptron shown in Fig. 2 (b) is represented by the expression (6)  2    2 iNi .Wi , j iout  AF  W j ,3 . AF    b j   j 1     i 1 I o 



  Vˆj   b3  



 Vˆ3   

(6)

The Sinh-Domain implementation of the single neuron cell is shown in Fig. 2 (c), where the divider circuit used is given by sinh-domain multiplier [Khanday et al., 2013] and the Activation Function (AF) block used is also designed in sinh-domain technique [Kant et al., 2014]. The activation function is one of the basic building block for NNs and the analog implementation of these functions is a bit difficult. The reason being the involvement of exponentiation and division, both of which are expensive operations. Many researchers have achieved the design of these activation functions using lookup tables, applying piecewise linear approximation, piecewise non-linear approximation [Basterretxea et al., 2004, Basterretxea et al., 2001, Kwan, 1992, Piazza et al., 1993]. But the direct realization of activation functions has a very less literature available [Khodabandehloo et al., 2012, Tabarce et al., 2005, Nedjah et al., Khuder and Husain, 2013, Bogason , 1993, 2011, Lu , and Shi, 2000, Keles et al., 2003, Keles and Yıldırım, 2010]. These implementations however are either using high voltage circuits or floating gate MOSFETs which is a costly technique in itself. Figs. 3 (a) and 3 (b) shows the Sinh-Domain designs for Tanh and Sigmoidal functions [Kant et al, 2014] which find their applications in Hopfield neural, Sigma-pi networks in addition to being employed in multilayer perceptron neural network. The divider circuit used is shown in Fig. 4in order to realize the relationship given by (7)

iout  I 0

I1 I2

Fig. 1. Non-linear transconductor cell (Multiple output).

[47]

(7)

Khanday and Kant/COMMUNE – 2015

(a)

(b)

(c) Fig. 2.(a) Single neuron cell, (b) A two-layer perceptron, (c)Single neuron cell in Sinh-Domain.

3. Circuit Description of 2:1 MUX Design of a 2:1 MUX is achieved using neural networks where the network is designed using neuron cell shown in Fig. 2 (c). The activation functions used for our network are sigmoid and hyperbolic tangent, sigmoid for the output layer neurons while as hyperbolic tangent for other neurons.

(a)

(b)

Fig. 3. Sinh-domain realization of activation functions (a) Tanh, (b) Unipolar Sigmoidal

Fig. 4. Sinh-domain realization of Two-quadrant multiplier/divider [Khanday et al., 2013].

The Sinh-Domain realization of the 2:1 MUX circuit is given in Fig. 5. The select line input is provided to the first neuron which acts as inverter, and to one of the input to N3 neuron. The output of N1 is given to one of the input of N2 neuron whose second input is Data line 1. The inputs to neuron N3 are data line 2 and the second input is select line. N2 and N3 are trained to act as AND gates. Output of N2 and N3 are fed to N4 which is trained to be OR gate. The activation functions used for N1, N2 and N3 is Tanh while as for N4 Sigmoidal function has been used.

[48]

Khanday and Kant/COMMUNE – 2015

Fig. 5. Shin-Domain realization of NN based 2:1 MUX.

4. Simulation Results The sinh-domain activation functions, neurons and the 2:1 MUX were designed with VDD=0.5V, VDC=0.1V and 0.35µm AMS CMOS process technology. The training of the neurons was done in the Matlab environment using back propagation algorithm from where the weights associated with the neurons and values of biases were calculated. The weight associated with N1 is -2.23pAdc and the bias is 5.37pAdc, for N2 and N3 weights are 6.98pAdc, 6.98pAdc and bias is -21.87pAdc and for N4 weights are 10.98pAdc, 12.98pAdc while the bias is -110.87pAdc. The chosen bias scheme (values of different current sources other than the bias current and the weights) in case of the activation functions and for the 2:1 multiplexer network is given in Table 1. Also to mention that the aspect ratio for MOS transistors used in the transconductor cell in Fig. 1 is given in Table 2. The aspect ratio for PMOS transistors applied for designing mirrors so as to realize bias scheme is 55um/1.5um. The obtained output simulation results for the activation functions and 2:1 MUX are given in Fig. 6 and Fig. 7 respectively. The total current source power dissipation of the 2:1 multiplexer network was 213.1359pW while as total voltage source power dissipation was 10.2736nW which establishes the fact that the low-power design of the 2:1 MUX has been achieved. Next, the Monte Carlo analysis was done to check the robustness of the design. For this purpose, Monte Carlo analysis of the circuits were done for a number of N=70 runs assuming 5% deviation (with Gaussian distribution). The statistical graphs of the sigmoidal and tanh activation functions for the same conditions are shown in Figs. 8 and 9 respectively where the upper saturation, lower saturation levels and switching threshold voltage (voltage corresponding to mean of upper and lower saturation currents), which form the three important performance parameters, have been considered for the study. From the obtained results it is clear that the deviations are very small. The Monte Carlo analysis results for the proposed sinh-domain design of 2:1 MUX is shown in Fig. 10. From the results, it can be concluded that the output result of the circuit remains un-altered with respect to the parameter variation i.e. the deviation in parameters due to the variation in conditions has no effect on the performance of the circuit. Thus the proposed design has the important feature of robustness. 5. Conclusion In this paper, low-voltage realization of 2:1 multiplexer function using sinh-domain technique has been introduced. The implemented circuit is then checked for its performance under varying parameter conditions so as to establish its robustness. The simulation results indicate that the realized low-voltage neural network based multiplexer can be well applied with contemporary technology requirements and besides its robustness, in the future, the study can be extended to train the multiplexer circuit so as to achieve more and more intelligent design with the priority aspect also incorporated.

[49]

Khanday and Kant/COMMUNE – 2015 Table 1.The aspect ratio ofMOS transistors employed in the nonlinear transconductor cell in Fig. 1.

Transistor Mp1–Mp4 Mp5–Mp12 Mn1–Mn8

Aspect ratio (um) 35/0.55 58/0.6 21/1

Table 2. Bias scheme for the sinh-domain circuits.

Parameter Io



I03 I032

Value (pA) 1 100 100

Parameter I01 I012

Value (pA) 30 25

(a)

Parameter I02 I022

Value (pA) 30 25

(b)

Fig. 6. Input/output characteristics of activation functions (a) Tanh (b)Unipolar Sigmoidal.

Fig. 7. Simulation results for sinh-domain NN based 2:1MUX.

(a)

(b)

(c)

Fig. 8. Monte Carlo results for sinh-domain Sigmoidal activation function (a) Histogram for threshold, (b) Histogram for upper saturation level, and (c) Histogram for lower saturation level.

(a)

(b)

(c)

Fig. 9. Monte Carlo results for sinh-domain tanh activation function (a) Histogram for threshold, (b) Histogram for upper saturation level, and (c) Histogram for lower saturation level.

[50]

Khanday and Kant/COMMUNE – 2015

Fig. 10. Montecarlo analysis results for 2:1 MUX.

References Hopfield, J. J., 1982, Neural networks and physical systems with emergent computational abilities, Proc. Natl. Acad. Sci. USA., 79, p. 2554. Chao-Ming C., Chih-Min L., Ching-Tsan C., Yeung, D. S., 2007, Hardware implementation of CMAC neural network using FPGA approach, IEEE International Conference on Machine Learning and Cybernetics, 4, p. 2005. Seul J., Sung, K., 2007, Hardware implementation of a real-time neural network controller with a DSP and an FPGA for nonlinear systems, IEEE Trans. Industrial Electronics, 54, p. 265. http://www.informit.com/articles/article.aspx?p=24687&seqNum=6 Serdijn W., et al., 1999, Design of High Dynamic Range Fully Integrable Translinear Filters.”Analog Integrated Circuits and Signal Processing, 19, p. 223. Katsiamis, A. , Glaros K., Drakakis, E., 2008, Insights and Advances on the Design of CMOS SinhCompanding Filters, IEEE Transactions on Circuits and Systems-I, 55, p. 2539. Frey, D., 1996, Exponential state-space filters: a generic current mode design strategy, IEEE Transaction on Circuits and Systems-I, 43, p. 34. Khanday, F. A., Pilavaki, E., Psychalinos, C., 2013, Ultra Low-Voltage, Ultra Low-Power Sinh-Domain Wavelet filter for ECG analysis, ASP Journal of Low Power Electronics, 9, p. 1. Kant, N. A., Khanday, F. A., Psychalinos, C., 2014, 0.5V Sinh-Domain Design of Activation Functions and Neural Networks, ASP Journal of Low Power Electronics, 10, p. 1. Basterretxea, K., Tarela, J. M., Del Campo, I., 2004, Approximation ofsigmoid function and the derivative for hardware implementation of artificial neurons, IEE Proceedings—Circuits, Devices and Systems (2004), 151, p. 18. Basterretxea K., Tarela, J. M., 2001, Approximation of sigmoid functionand the derivative for artificial neurons: Advance in neural networksand applications, WSES Press, Greece (2001). Kwan, H. K., 1992, Simple sigmoid-like activation function suitable for digital hardware implementation, Electronics Letters, 28, p. 1379. Piazza, F., Uncini, A., Zenobi, M., 1993, Neural networks with digital LUT activation functions, Proceedings of International Joint Conference on Neural Networks (IJCNN), Japan, October (1993), p. 1401. Khodabandehloo, G., Mirhassani, M., Ahmadi, M., 2012, Analog implementation of a novel resistive-type sigmoidal neuron, IEEE Transactions on Very Large Scale Integration (VLSI) Systems, 20, p. 750. Tabarce, S., Tavares, V. G., de Oliveira, P. G., 2005, Programmable analogue VLSI implementation for asymmetric sigmoid neural activation function and its derivative, Electronics Letters, 41, p.1. Nedjah, N., Da Silva, R. M., 2011, Analog hardware implementations of artificial neural networks, Journal of Circuits, Systems and Computers, 20, p. 349. Khuder, A. I., Husain, S. H., 2013, Hardware realization of artificial neural networks using analogue devices, Al-Rafidain Engineering, 21, p. 77. Bogason, G., 1993, Generation of a neuron transfer function and its derivatives, Electronics Letters, 29, 1867. Lu , C., Shi, B., 2000, Circuit design of an adjustable neuron activation function and its derivative, Electronics Letters, 36, 553. Keles, F., Avcı, M., Yıldırım, L., 2003, Floating gate MOS transistor based low voltage neuron design and XOR problem implementation, Proceedings of the Third International Conference on Electrical and Electronics Engineering (ELECO’2003), Turkey, December (2003), p. 41. Keles, F., Yıldırım, L., 2010, Low voltage low power neuron circuit design based on subthreshold FGMOS transistors and XOR implementation, Proceedings of the XIth International Workshop on Symbolic and Numerical Methods, Modeling and Applications to Circuit Design (SM2ACD), Tunisia, October (2010), p. 1–5.

[51]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

File Tracking System for University of Kashmir: Design Guidelines and Model Implementation M. Tariq Bandaya*, Shafiya Afzal Sheikha, Javid Ahmad Ratherb a

Department of Electronics and Instrumentation Technology, Univesity of Kashmir, Srinagar, India b Indira Gandhi National Open Univeersity, Regional Center, Srinagar, India

Abstract With an objective to develop and implement a file tracking-system for the University of Kashmir, this paper proposes a project based implementation of File-Tracking System. It explains design, development, and implementation strategy that can be adopted to implement a web-based file-tracking system for University of Kashmir. It reviews existing file flow system in the University of Kashmir, discusses proposed file-tracking system and its working. Through a design outlay, this document gives an insight about its working including user screens, roles of users and reports. It discusses why existing file-tracking systems that are in place at various government organizations are not suitable for the University of Kashmir. The document gives guidelines for successful implementation of the system including employee training and its phased implementation. The system when implemented shall improve efficiency and effectiveness of the existing system, consistency of file records, resource management, and quality of administration. It will establish transparency and accountability and thereby will help to reduce turnaround time, and processing delays of files. The system will pave a way towards adoption of complete e-governance wherein paperless file processing and its tracking is possible. Such a system is already in place at reputed Universities and Institutions.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: File Tracking System; FTS; e-Govrnance

1. Introduction File Tracking System (FTS) also called File Tracking and Monitoring System (FTMS) is a web-based application to monitor the movement of files and receipts and assist in their easy tracking. Its features include generation of receipts and files, updating its status, opening of new files, tracking the movement of files, dispatching letters/files, recording their track, etc. With the emergence of e-Governance wherein among other applications, applications for electronic office such as eOffice application, which besides keeping track of files recorded also replaces physical files by electronic files, limited application of FTS exist. However, in situations where organizations do not possess adequate hardware/software resources or have inadequate trained human resources, FTS still finds application. In contrast to electronic office where all officers and officials must possess adequate computing skills and hardware/software resources, FTS requires minimum hardware/software resources and some trained human resource. For implementing FTS in an organization, no radical change is required in its existing functioning. Further to implement e-Office, several pre-implementation steps are required e.g. unstructured file formats have to be converted into highly structured formats, and adequate hardware and software tools have to be procured. Further, existing files have to be converted into electronic files and all human resource have to be trained in the use of computers. File Tracking System is a step in the direction of realizing the concept of paperless office. Several educational institutions such as Jamil Millia Islamia (JMI, 2010), New Delhi,

* Corresponding author. Tel.: +91 94195 48922. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Banday, et al/ COMMUNE – 2015

NIELIT (NIELIT, n.d.), New Delhi, India have already installed File Tracking Systems. Some of these have developed in-house solutions for e-filing while as others have hired the services of private companies to develop such systems. 2. Background Study National Informatics Centre (NIC) of Government of India has made File Tracking System (FTS) (NIC, n.d.), Centralized File Movement, and Tracking Information System (CeFMaTIS) (NIC, n.d.) to monitor the pendency of receipts and files and assist in their easy tracking. It is based on the Manual of Office Procedure and is a system for Government/Semi-Government offices and public sector organizations. The product supports the complete electronic file movement with encryption of content and digital signature. The FTS and CeFMaTIS have been implemented in over 70 government Departments. DAILY RFID Co. Limited has launched a RFID file tracking system in October 2010, which enables automated gathering and sending of document information. The systems enable users to easily locate, inventory and check in or out file movement. This system integrates RFID labels, multi-tag RFID reader, and necessary software to allow organizations to manage document efficiently. The UHF RFID label on each document permits to store location and time of a document (DRCL, 2010). In addition to these, several private software development companies and open source communities offer file-tracking systems. These include University ERP (Expedien, University ERP, n.d.), Virmati (VSTL, n.d.), 3M-RFID File Tracking System (3M T&TS, 2008), File-Coder (FILETACTICS, n.d.), DFTS- Dolphin File & Document Tracking System (DRPL, n.d.), cuteflow document circulation and workflow system (Haberkern et al, 2009), OpenDocMan™ - Open Source Document Management System (OpenDocMan, 2015), File Tracking & Tracing (ROPARDO S.R.L, n.d.), and Openkm-Open Source Document Management (OpenKM, n.d.).sMost of the currently available web based file-tracking solutions are mainly focused on storing a digital copy of files and monitoring changes to the contents rather than tracking their movement alone between various offices of an organization. Some of these applications are provided as Software as a Service (SAAS) wherein the data and the application is stored on some third party servers. The organization using such services do not have any control on these services. Other open source and freely available solutions are not stable, often lack in community support, and have no technical documentation thus making them unreliable and risky for large organization like universities. Further, the feature set offered in ready-made file-tracking systems is comparable to each other; however, they are not designed to the specific requirements of a particular organization. 3. Existing System The University offices include PG Departments, Directorates, and Research Center offices, which are distributed throughout University campuses. Other offices are Administrative offices, which are mostly in the Administrative block of the University. Every office maintains dozens of files housed within them to maintain records in categorized manner. As depicted in figure 1 below, the system involves several types of entities including users who perform different activates within the system. These include “File”, “Officer”, “Office”, “Official”, “Dispatch Register”, “Dispatcher”, “Employee”, “Student,” and “External User”. Table 1. Entities Involved in the Current System Entity and its Functions

A File is a related collection of data, which may contain information on a single sheet of paper or multiple sheets of papers. As the file passes from one office to another office, officers and officials mark comments, remarks, and decision on it. This may also involve movement of another office file that will help officers and File officials to mark comments, remarks, and decisions on it. Files may be categorized based on their term as well as based on its nature. Based on their term, files may be long-term files that are stored in a particular office and short-term files that are disposed off by merging them with some long-term file after taking decision on them.

Officer

Chamber

Dispatcher

Official

Officers are University Employees such as Vice Chancellor, Dean Academic Affairs. Registrar, Dean Research, Deputy Registrars, Assistant Registrars, and Section Officers. These officers except section officers sit in their office chambers, which are physically isolated from their offices. Often one officer processes files originating from different other offices. The composition of University offices irrespective of their type are housed in a single or multiple rooms. Deans of Various Faculties, Heads of various Departments, Directors of various Centers, etc. are office in-charges of these offices. The office staff includes section officer, head assistants, senior, assistants, junior assistants, dispatchers, etc. Each office has sets of files that are used to record various activities pertaining to that office. Often responsibility and custody of files are distributed among officials working in an office. Dispatcher maintain dispatch and receive registers and peon book. These are specially designed printed with columns wherein some information about the file such as its subject, date of movement, etc. is recorded. A record is added in a dispatch register, every time a file moves in or out of a particular office. These registers also serve as a receipt register. Official

Office In-Charge

Dispatch Registers

[53]

Banday, et al/ COMMUNE – 2015

Entity and its Functions

Employee

University employees include faculty members, officers, officials, etc. who may or may not be an officer. They may create files (notes, applications, proposals, etc.) which are moved from one office to the other office for their disposal. They either submit their files in their native office or other offices of the University for further necessary action. Students including scholars submit applications, grievances, forms, and other documents at various University offices, which are also moved from one office to another office. Currently, such documents are not tracked at University offices. Receipts for their submission is given in rare cases.

Student

External User

External users are persons from outside the University of Kashmir. They submit proposals, tenders, applications, and other files at various University offices for further necessary action. These documents are not tracked, however, receipts of there is given in some cases.

Dispatcher

Official

Submit FILE

Officer

Get Notifications

(e.g. HODs, Deans, Admin Heads, A/D/ Reg)

Submit FILE

Chamber A

Get Notification

Official

Office In-Charge

External User

Officer (e.g. VC, DAA, DR, Reg)

Chamber C

Employee Dispatcher

Official

Officer (e.g. HODs, Deans, Admin Heads, A/D/ Reg) Official

Office In-Charge

Chamber Z

Fig. 1. File movement in the existing system

New files of diverse types are submitted by external and internal users that include University employees, Students, Heads of various Departments, Deans of various Faculties, Directors of various Centers, Coordinators, and Principal investigators of various projects, etc. Irrespective of the type of the office, these files pass from one official/officer to another official/officer within same or different office. A file track within the same office is usually not maintained, however, currently, University of Kashmir has a manual file-tracking system, wherein each office has a dispatcher who manually records track of each file moving in and out of the office. For this purpose, offices such as Departmental offices are maintaining a single dispatch register while as some offices such as administrative offices maintain multiple dispatch registers. A dispatcher in the office records minimum file particulars on the dispatch register, which is carried along with the files to be moved to another office by a peon. The peon delivers the files to the dispatcher of the other office who records these files in his dispatch register and signs the entries on the dispatch register of the sending office. This manual system of file record maintenance is inefficient as it provides very limited capability to track file especially in situation wherein in and out flow of files is huge. The system involves marking of multiple manual entries in dispatch registers for files at each passing office and no information about the status of the file is maintained. Reports about file receive and dispatch transactions, file status, file log, etc. are not possible in the current system. Figure 1 shows the file movement in the current system. 4. Proposed System The proposed file-tracking system will not disturb the existing file movement system; however, file dispatch and receive procedures will be computerized. Electronic records for existing files (Long Term files) as well as new files will [54]

Banday, et al/ COMMUNE – 2015

be created to track their movement. There will be no change in the existing work system. University community as well as external users will be able to get information about the position of their file(s). The proposed file tracking-system will involves all entities of the existing system except dispatch registers, which will be replaced by electronic File Database. In addition to them, it will require “Super Administrator”, “Auditors”, and “Administrators”. Each type of user will be assigned specific roles to maintain the system. To ensure role based security, after deployment of the system, IT experts, advisors, programmers, etc. involved in the design and development of the system will have no control on the functioning of the system. The system will be exclusively controlled by Super Administrators. However, there will be scope of improvement and future modification. Table 2 shown below briefly defines involvement of each entity used in the proposed system. Table 2. Entities involved in the proposed system Entity and its Functions

Files as in the existing system will comprise of single or multiple sheets of papers, which will be moved from one office to another office. The files will have a unique identification number generated at the time of their creation. Records of the existing files (long-term files) stored within File University offices will also be created. The officers of the University such as Vice Chancellor, Dean Academic Affairs. Registrar, Dean Research, Deputy Registrars, Assistant Registrars, will perform their activities as they are performing in the existing systems. However, they can optionally inquire about file status (e.g. files currently put for their Table 2remarks, comments, and disposals). Some of the officers will Officer be involved in the audit of files and administration of the file-tracking system. There shall be no Chamber change in their current method of working or additional burden on them. A list of all offices of the University shall be created in the system. A record of the office files i.e. the long-term files stored in the office shall also be created. This is a list of files along with their purpose. Office in-charges such as Deans of Various Faculties, Heads of various Departments, Directors of various centers, etc. shall function as they are functioning in the current system. However, they will be able to keep track of the files that pertain to their offices. They will also manage dispatchers working in their offices and their profiles. They can optionally create file records, dispatch, and receive priority and confidential files. Dispatchers will record file transactions using the proposed file-tracking system. The Dispatch registers will be replaced by electronic dispatch registers, which will be maintained primarily by the dispatchers. Dispatchers can maintain internal movement of files (movement of files within the office) as well as external movement of files Long-Term File Short-Term File (movement of files from one office to another). They will also be able to maintain the Record DB Record DB file transactions involving officers located in their respective office chambers. University employees who want to submit their files for processing may create file record themselves and submit the file through dispatcher in the concerned office who will in turn send the file for processing and record the transaction. In case the employee submits the file without creating its record, the dispatcher will create its record before further processing it. The employee Employee can online track the file movement and remarks put on them. External users will submit their files to the concerned dispatcher of the office, who will create its record and send it for processing after recording its transaction. A file submission receipt and password will be generated by the system, which will be subsequently used by the user to inquire its status online. External Students will be treated as external users in the proposed system. User Dispatcher

Official

Official

Office In-Charge

Auditors will be appointed by Super Administrator (such as Registrar of the University). Auditors will have access rights to inquire about file movement of every files (or group of files). They will be able to generate diverse types of reports from the system. They will also receive system-generated triggers for files that are moving slow than expected. They will be Auditors able to generate reports about all types of users including super administrators and their (Appointed by S. Admin) activity in the file-tracking system. Complaints and grievances can be resolved easily by administration based on auditor’s reports. Administrators will be appointed by Super Administrator (such as Registrar of the University). The administrators will have full control on the system but their activity will also be recorded and visible to auditors and super administrators. They will be able to perform any operation in Administrators the system. However, their primary responsibility will be to create office in-charges and (FTS Committee Members) maintain their profiles and roles. They will also be able to carry out certain special functions on file records, which other types of users are not, allowed to perform. Super Administrator role is the highest in the system. Registrar, being the in-charge of entire Registry can be the super administrator in the proposed system. He can carry out any operation in the system, however, his primary role in the system will be to appoint and maintain administrators and auditors and set their roles. Vice Chancellor of the University Super Administrator (Registrar) will also be able to log on to the system as super administrator; however, he will primarily use it for monitoring and evaluation.

[55]

Banday, et al/ COMMUNE – 2015

File creation will be done by the University employees including office in-charges of various offices as in the previous system. However, before dispatching the file for processing an electronic record of the same shall be created. In case the submitter is a University employee, the electronic record can be created by the submitter. Otherwise, the dispatcher will create the record and dispatch the file to respective quarter after creating its dispatch record. The file along with other files will be pilled and send physically for processing within same office or to another office. In case the files are send to another office, which has a dispatcher, the files will be received by the dispatcher and file record accordingly modified. At each stage of dispatch and receive appropriate comments will be entered by the dispatcher in the respective column. In case the file is send to an officer/official within the same office, the dispatcher may or may not record this file transaction depending upon the instructions and policy adopted by the University. As the file is processed, it may grow in the number of pages; this will also be updated by the concerned dispatcher. File records will available online for viewing by file submitters who can generate different types of reports depending upon their roles in the system. The functioning of the proposed system is depicted in figure 2 below. Table 3 (given at the end of this paper), lists roles and operations that a particular group of users can carry out in the proposed system. Every user group except external users shall have a user profile that can be updated by the user himself. Users with higher privileges in the system shall be able to manage the users immediately below their privileges. To carry out their role in the new system, higher privileged users can authorize lower privileged ones. They can also block them. Super administrators can administer all user accounts and authorize or block any user; however, they shall ordinarily be required to manage administrator roles only. They will also block external users in case it is required. Administrators shall manage roles of office in-charges who in turn shall manage roles of dispatchers and employees working in their respective offices. Office in-charges will also manage the profile of their respective offices and edit or add long-term or short-term file records.

Dispatcher

Official

Submit FILE

Officer

Receive RECEIPT, ETC.

(e.g. HODs, Deans, Admin Heads, A/D/ Reg)

Submit FILE

Receive RECEIPT, ETC.

Chamber A Official

Office In-Charge

Officer

External User

(e.g. HODs, Deans, Admin Heads, A/D/ Reg)

Chamber B Employee

Long-Term File Record DB

Officer Short-Term File Record DB

Auditors

(e.g. VC, DAA, DR, Reg)

(Appointed by S. Admin)

Chamber C

Administrators (FTS Committee Members)

Officer (e.g. HODs, Deans, Admin Heads, A/D/ Reg)

Chamber D

Super Administrator (Registrar)

Dispatcher

Official

Officer (e.g. HODs, Deans, Admin Heads, A/D/ Reg) Official

Office In-Charge

Chamber Z [56]

Banday, et al/ COMMUNE – 2015

Fig. 2. File movement and file record in the proposed system

In table 4 (given at the end of this paper), list of file record maintenance operations along with users who can perform these operations are listed. File records can be created by office in-charges, dispatchers, or employees who can also modify the same if the file has not already been dispatched. Essentially, files records can be dispatched by dispatchers who can also update dispatch records in case the dispatched file is not yet received. When a file is disposedoff, its record can be updated by dispatchers or office in-charges. Wrong file records entered by employees or dispatchers owing to some mistakes or oversights can be removed by respective office in-charges. Special file record operations can be carried out by administrators. In table 4 a list of file record views/reports is also given. Dozens of reports about file movement, user roles, user activity operations, etc. will be generated by the system. The reports can be viewed online, printed, or downloadable as pdf/excel files. A user will be able to view these reports; however, records of only those files, which pertain to that user, shall be included in these reports. The file activity reports include following reports about short-term and long-term files: a) View File Record, View File Status, View File Log, Find File, Office File Reports, Employee File Record Reports, Conditional Alerts, File Disposal Reports, File Record Delete Reports, Track Complaints, Pending File Reports, File Dispatched Reports, File Received Reports, Daily Register Reports, Incoming/outgoing File Reports, Peon Book Reports, Monthly/Quarterly/Yearly Register Reports, File in Hand Reports, Activity log (Administrator, inCharge, Dispatcher, Employee, Office) Reports, Group/Office wise user reports, etc. The system will also be able to generate other criteria based reports. 5. Prototype Design A model design of the web based file-tracking system was implemented in Microsoft Technologies using ASP.NET with client and server-side programing for web forms, reports and backend processing performed in C#.NET. MS SQL Server was used as a backend database. HTML, CSS and XML, JQuery, and AJAX was used to make the design appealing. In the model design, more than twenty user interaction forms were designed including forms for file creation, file dispatch, and receive. A few reports were also designed to evaluate the report generation functionality of the proposed system. A screenshot of file creation and file dispatch user interaction forms is shown in figures 3 and 4. The web based file-tracking system was tested on all major operating systems including Windows, Linux, and IOS. Besides this it was also tested on Web enabled Mobile Devices. The model design was tested on all major web browsers including Apple Safari, Internet Explorer, Mozilla FireFox, Flox, Chrome, and Opera. The results of the tests were successful across tested operating systems, and browsers. 6. Implementation Strategy The proposed file-tracking system may involve a pilot implementation before a phased implementation in all offices of the University of Kashmir. The details are enumerated in table 5 below. Table 5. Implementation Strategy

Stage

Details a) Training: Training in the use of the web-based file-tracking system shall be provided to one super administrator, one administrator, four office in-charges, eight dispatchers, and ten other University employees.

Pilot Implementation

b) Implementation: The pilot implementation of the proposed file-tracking system shall be hosted on test servers. The users trained for pilot implementation shall be given access to the file-tracking system. The office in-charge and dispatchers shall be divided in two groups who will perform the tasks as per their roles. The pilot implementation will run in four University offices parallel with the existing system. c) Debugging: Any errors found by running the proposed system in its pilot implementation shall be rectified. The system shall be put to another pilot implementation until successful results are achieved.

[57]

Banday, et al/ COMMUNE – 2015

a) Training: Training in the use of the web-based file-tracking system shall be provided to administrators, auditors, office in-charges, and dispatchers in a phased manner that will cover all University offices. A manual describing the use of the file tracking system for all types of users including employees and external users shall be made available online. Phased Implementation

b) Implementation: After the successful implementation of the system in the pilot phase, phased implementation of the system can be undertaken to make the system functional in all University offices. At this stage, the system shall be handed over in Toto to the administration and the developers will have no control or roles to play other than debugging and future up-gradations. c) Up-gradations: The system can be later upgraded to add new required features and to take the system further into a full-fudged e-governance system.

7. Conclusion and Future Scope This paper analyzes the existing file flow system in University of Kashmir and proposes the design, development, prototype and implementation strategy of web based file-tracking system which when implemented shall improve the management, tracking, history and auditing of files. It shall all also increase staff efficiency. It shall offer features such as fast searching, scheduling, reminders, and built-in messaging tools. This paper also discusses some of the existing file management and monitoring solutions, their features and shortcomings. In future, the designed file tracking system may be scaled and enhanced in various ways which can lead the University to paperless e-governance system where in digital files will be used instead of traditional paper based files. Email and or SMS notifications can be sent to File Owners notifying them of the status and progress of their files. Mobile apps can be developed to allow officers to view / work on files remotely as well as allow the file owners to monitor the status of their files. Digital signatures can be added to files for the purpose of security and encryption. References 3MT&TS, 2008. 3M Track and Trace Solutions, URL: http://solutions.3mindia.co.in/wps/portal/3M/en_IN/Track_Trace/home/Products/three/, accessed on 23-01-2015. Haberkern, T., Schäfer, M., Constantin, T., Kesch, F., Usey, C., 2009, CuteFlow, URL: http://www. cuteflow.org, accessed on 23-01-2015. DRCL, 2010. DAILY RFID Co. Limited, url:http://www.pr.com/press-release/216714, accessed on 23-01-2015. DRPL, n.d. Dolphin RFID Private Limited, URL: http://www.dolphinrfid.in/file-tracking-system.htm, accessed on 23-01-2015. Expedien, n.d. University ERP, URL: www.university erp.com/file-movement.shtml, accessed on 23-01-2015. ROPARDO, S.R.L., File Tracking & Tracing, URL: http://www.filetrackingclient.com, accessed on 23-01-2015. FILETACTICS, URL: http://filetactics.com/FileTrackingSoftware.html#.VMd1DNKUfTp, accessed on 23-01-2015. Jamia Millia Islamia, 2010. A Central University, URL: http://jmi.ac.in/upload/publication/cit.updates_Q3-2010_issue_11.pdf, accessed on 23-012015. NIC, n.d. National Informatics Center, CeFMaTIS-Centralized File Movement and Tracking Information System-2.0, url: http://www.nic.in/projects/centralized-file-movement-and-tracking-information-systemcefmatis-20, accessed on 25-01-2015. NIC, n.d. National Informatics Center, URL: http://www.nic.in/projects/file-tracking-system-1, accessed on 25-01-2015. NIELIT, n.d. National Institute of Electronics and Information Technology, URL: http://ftms.nielit.in/, accessed on 23-01-2015. OpenDocMan, 2015. URL: http://www.opendocman.com, accessed on 23-01-2015. OpenKM, n.d. URL: http://www.openkm.com, accessed on 23-01-2015. VSTL, n.d. Virmati Software and Telecommunication, URL: http://www.virmati.com/fts.html?val=1, accessed on 30-01-2015.

[58]

Banday, et al/ COMMUNE – 2015

Kashmir University File Track System Create New FILE Current logged IN user Details FAC-100

Tariq Banday

Electronics & IT

File Number

13200

Submission Type

Internal

/

Office In-Charge

/

ELIT

/

Type Abbr. Subject

Log OUT

2015

External

Subject

Type E-mail of Submitter

E-mail

Type/Select Name of Submitter

Submitter

25/01/2015 11:05AM

Type Subject of the file

File Type

Note

File Priority

High

Confidential

1

File Number of Previous Volume Nl

Volume

Number of Pages

1

Notes Side

Corresponding

1

Yes

Cancel

CREATE

Fig. 3. User interaction form for creation of new file.

Kashmir University File Track System Dispatch File Record Current logged IN user Details FAC-100

Tariq Banday

Electronics & IT

Office In-Charge

25/01/2015 11:33 AM

Dispatch Type Within Same Office

Select an Employee

Choose an Employee from the List

To Another Office

Select an Office

Vice Chancellors Office

Through Employee

Mr. Nadeem Ahmad

Select Files to Dispatch Currently there are 5 Open Files in the Office

Show Files

File Number, Subject, Submitter and Remarks

All

Office

Select

Vice Chancellors Office

13200/ELIT/EQUP-PURC/2015

Note for Purchase of Equipment under MRP Sanctioned by UGC in favout of DR M Tariq Banday

Dr. M. Tariq Banday

Notes Side Pages

REMARKS

1

Corresponding Pages

0

File Submitted for Approval of Vice Chancellor

Vice Chancellors Office

13201/ELIT/COMMUNE/2015 Note for Grant of Approval to COMMUNE-2015 Dr. M. Tariq Banday

REMARKS

Notes Side Pages

1

Corresponding Pages

0

File Submitted for Approval of Vice Chancellor

13202/ELIT/SEEDS/2015

Vice Chancellors Office File Subject retrieved from FLDB

Submitter Name Retrieved from FLDB

REMARKS

Notes Side Pages

1

Corresponding Pages

0

File Submitted for Approval of Vice Chancellor

Selection Statistics 3 Files Selected for Dispatch to Vice Chancellors Office

Cancel

Fig. 4. User interaction form for dispatch of file record.

[59]

DISPATCH

Log OUT

 ×  × × × ×

Update O Detail  ×   × × ×

 ×   × × ×

Add O File Record  ×   × × ×

Update O File Record  ×   × × ×

Add O Details

E-mail O Details

Import & Export OFR

 × × × × × ×

 × × × × × ×

 × × × × × ×

 × × × × × ×

 × × × × × ×

Authorize  × × × × × ×

Block

 × × × × × ×

Other Administration

 ×  × × × ×

 ×  × × × ×

 ×  × × × ×

Reset Password

 ×  × × × ×

Send E-mail

 ×  × × × ×

Authorize

 ×  × × × ×

Block

 ×  × × × ×

Other Administration

 ×    × ×

 ×    × ×

Update Dispatch File R

 ×    × ×

Receive File R

 ×    × ×

Dispose Off File R

 ×   × × ×

Delete File R

 ×  × × × ×

Special File Operations

 ×  × × × ×

Other File Operations

      

View File Record

      

View File Status

      

View File Log

      

     × ×

   × ×  ×

 ×    ×  ×  × × × × ×

Criteria Based Reports

      ×

Conditional Alerts

Employee File Reports

File Disposal Report

      ×

File Delete Report

[60]

 ×   × × ×

  × ×     × × × × × ×

   × × ×     × × × × × × × × × × ×

 ×  × × × ×

   × × × ×

      ×

     × ×

Files Dispatched

     × ×

Files Received

     × ×

Daily Register

     × ×

Incoming Files

     × ×

Outgoing Files

          × × × ×

Peon Book

Dispatcher Activity Log

     × ×

Monthly Register

     × ×

Files in Hand

      ×   × ×  × × × × × × × × ×

Activity Log (Admins)

Activity Log (Dispatcher)

    × × ×

Activity Log (Employees)

    × × ×

Activity Log (Offices)

File Record View and Reports

EX. User MGM

  × ×     × × × × × ×

Remove Track Complaint

Pending Files Report

File Record Maintenance

 ×     ×

Dispatch File R

Note: A  mark indicates an essential operation,  mark indicated an optional operation, and a × mark indicates no access.

 ×     ×

Update File R

Block

Dispatcher & Employee Management

 ×   × × ×

Reset Password

    × × ×

Block

     × ×

Reset password

Office In-charge Management

 ×  × × × ×

Set Roles

Administrator & Auditor Management

 × × × × × ×

Create

Import & Export

Banday, et al/ COMMUNE – 2015

 ×   × × ×

Send E-mail

Table 3. Users and their Roles (User Management)

 ×  × × × ×

Import & Export O

Office & Office File DB Management

      ×

Other Profile MGM

User Profile Management

      

Set Roles

Reset Password

Office File Reports

User Type

      

Forgot PASS E-mail

Note: A  mark indicates an essential operation,  mark indicated an optional operation, and a × mark indicates no access.

      ×

Other Administration

Office Files

Create

Import & Export Find File

Super Administrators Auditors Administrators Office In-Charges Office Dispatchers Employees External Users

Authorize

Group Wise User Reports

Login & Change PASS

Table 4: Users and their Roles (File Record Maintenance and View)

User Type

Super Administrator Auditors Administrators Office In-Charges Office Dispatchers Employees External Users

Send E-mail

    × × ×

Authorized/Blocked Users

Create

Import & Export

Activity Log (In-Charge)

Create Create New File R

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Color Image Compression using EZW and SPIHT Techniques M. Tariq Banday, Tawheed Jan Shah* Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India

Abstract Enormous digitized data produced from sequence of images requires huge storage space, processing speed and transmission time. To increase the processing speed and to decrease the storage space and transmission time, images are compressed with different compression techniques. Compression process is applied to image prior to its storage or transmission. Various image compression techniques have been developed to attain required compression ratios without losing quality of the image and information therein. This paper compares Embedded Zero Tree Wavelet Coding and Set Partitioning in Hierarchical Trees compression techniques by applying them on a color image in two standard resolutions 256x256 pixels and 512x512 pixels. The performance of these Discrete Wavelet Transform based techniques has been compared separately and mutually in terms of bits per pixel, peak signal to noise ratio and mean square error for the same colored image in order to determine the best possible compression technique. The results have been obtained using simulation through Matlab software.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Image Compression; DWT; DCT; Embedded Zero Tree; Set Partitioning in Hierarchical Tree

1. Introduction Unprocessed multimedia data such as images, audio, video, text require huge storage space, transmission time and bandwidth (Gonzalez, Woods, 2004). Handling of such an enormous data can often present difficulties. However, proliferation of digital technology motivated for the need of improved image compression techniques. The two basic principles of image compression technique are irrelevancy and redundancy reduction, so that images can be stored or transmitted using fewer bits. The repetition of pixel or pattern across the image is called redundancy and is reduced either through Transform coding or predictive coding and irrelevancy is defined as that portion of data which is ignored by the human visual system and is reduced through the process of Quantization (Shapiro, 1993). Image compression techniques find applications in security industries, federal government agencies, galleries and museums, retail stores, medical science, etc. (ISO, 1991). Image compression techniques are broadly divided into two types: lossless and lossy technique. Lossless compression techniques are defined as those techniques in which there is no loss of information. It includes bit-plane coding, variable-length encoding, Adaptive dictionary algorithms such as LZW, lossless predictive coding, etc. while as lossy compression techniques are those in which there is some loss of information. It includes lossy predictive coding and Transform coding. However, at low bit rates, transform coding is considered as efficient. For a particular application, a compression technique offers advantages over the other techniques. In general, lossless Image compression techniques provide lower compression ratio and better image quality in comparison to the lossy compression techniques. The remaining paper is organized as follows: section II presents the background work. Section III gives an overview of discrete wavelet transform. Section IV discusses the wavelet based image compression techniques. Section V reports and discusses the results of experiments conducted with SPIHT and EZW coding techniques on two different image sizes followed by the conclusion.

* Corresponding author. Tel.: +91 9797 997886. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Banday and Shah/COMMUNE – 2015

2. Background Study Nagamani and Ananth proposed an image compression technique for high resolution, grayscale Satellite urban images. The proposed technique used discrete wavelet transform together with EZW (Embedded Zero tree wavelet) and SPIHT (Set Partitioning in Hierarchical Trees) coding techniques in order to achieve high compression ratio and better image quality. The compression ratio and peak signal to noise ratio determined using EZW and SPIHT codings have been compared to each other for same set of images. The results obtained showed possibility to achieve higher compression ratio and PSNR (approximately CR of 8 and PSNR of 29.20) for SPIHT coding compared to EZW coding (approximately CR of 1.07 and PSNR of 13.07) for applications related to satellite urban imagery (Nagamani, Ananth, 2011). Ma Tao et al. surveyed multimedia compression and transmission techniques to analyze them for energy efficiency in resource-constrained platform in terms of compression efficiency, memory requirement, and computational load. For image compression three important techniques JPEG (DCT), JPEG2000 (Embedded Block Coding with Optimized Truncation EBCOT), and SPIHT have been discussed. It was concluded that SPIHT is the best choice for energyefficient compression algorithms due to its ability to provide higher compression ratio with low complexity. JPEG2000 (EBCOT) achieved higher compression ratio, which mean better quality than SPHIT, however, complexity of EBCOT tier-1 and tier-2 operations caused intensive complex coding, higher computational load, and more energy consumption for resource-constrained systems (Ma Tao, 2013). Singh et al evaluated DWT-EZW and DWT-SPIHT compression algorithms, based on parameters such as decomposition level, compression ratio, PSNR and compressed size. The results showed that DWT-SPIHT compression technique provided better image quality, PSNR value and compression ratio in comparison to the DWT-EZW technique (Pardeep et al, 2012). Singh et al proposed an effective image compression technique for Lossy Virtual Human Spine image. Two image compression techniques viz; EZW and SPIHT using different wavelet filters has been compared on the basis of compression ratio (CR), peak signal to noise ratio(PSNR), mean square error(MSE) and bits per pixel (BPP) values. Experimental results showed that PSNR in SPIHT increased by a factor of 13-15% as compared to the EZW technique. Further, SPIHT produced a fully embedded bit stream and offered better image quality at low bit rates than EZW (Priyanka, Priti, 2011). 3. Discrete Wavelet Transform Modern image processing applications employ transformation coding wherein correlation between adjacent pixels is exploited to compress images (Wern et al, 2008). The adjoining pixels of the image are predicted with highest degree of accuracy. The transformation is a lossless process and it transforms the correlated information into uncorrelated coefficients. In transformed domain, image information proves to be more competent rather than the image itself. The compression is provided by processing and quantization of the transformed coefficients. Discrete Cosine Transform (DCT) (Watson, 1994) and Discrete Wavelet Transform (DWT) are two extensively used transformation techniques. Because of a number of advantages of DWT such as high compression ratio, better image quality, progressive transmission etc. it has become very popular. Through Discrete Wavelet Transform (DWT), a signal with good resolution can be analyzed in both time and frequency domains using a basic set of functions called wavelets (Yea et al, 2006). Wavelet comes from fact that they integrate to zero. Wavelets tend to be irregular and asymmetric. Wavelet based coding is very robust under transmission errors as compared to the DCT based coding. Through DWT (Mammeri et al, 2012), higher compression can be achieved as it decomposes time domain signal into different frequency bands through a series of high and low pass filters. The mathematical expression for wavelet transform is given by: ∞

𝐹[𝑎, 𝑏] = ∫−∞ 𝑓(𝑋)𝜓 ∗ (𝑎, 𝑏)(𝑋)𝑑𝑥 -- (1) Where * is complex conjugate symbol and ψ is some function. 4. Wavelet Based Image Compression Techniques Embedded Zero Tree Wavelet (EZW) and Set partitioning in hierarchical trees (SPIHT) coding techniques based on wavelet transform are widely used for still image compression. These techniques are explained below: 4.1.

Embedded Zero Tree Wavelet Coding

Embedded Zero Tree Wavelet (EZW) coding, introduced by Shapiro (Shapiro, 1993) is considered as one of the effective and powerful wavelet based lossless image compression algorithm. This algorithm is sometimes called embedded coder because the compression process stops when a desired bit rate is reached. It does not require any training, tables, codebooks or any prior information about the image. EZW algorithm performs quantization through entropy-coded successive approximations, and uses DWT or hierarchical subband decomposition. The performance of

[62]

Banday and Shah/COMMUNE – 2015

this algorithm is based on the self-similarity across these different sub-bands and successive approximation methods. A fully embedded code representing a series of binary decisions is produced from the bit stream produced by EZW algorithms. The following main steps are involved in this algorithm: Initialization: Set the threshold ‘T’ to the smallest power of ‘2’ that is greater than max(m, n) |c(m,n)|/2 , where c(m, n) are the wavelet coefficients. b) Significance map coding: Scan all the coefficients in a predefined way and output a symbol when |c(m, n)|>2. When the decoder inputs this symbol, it sets c(m, n) = ±1.5T. c) Refinement: Refine each significant coefficient by sending one more bit of its binary representation. When the decoder receives this, it increments the current coefficient value by ±0.25T. d) Set T=T/2, and repeat step 2 if more iterations are required (Raja, Suruliandi, 2010). a)

4.2.

Set Partitioning in Hierarchical Trees

Said and Pearlman, developed the Set Partitioning in Hierarchical Trees (SPIHT) wavelet coder in 1996 (Said, Pearlman, 1996). Both SPIHT and EZW use the basic idea of zerotree coding. This coding technique uses three coefficient location lists namely the List of significant Pixels (LSP), List of Insignificant Sets (LIS), and List of Insignificant Pixels (LIP) that contain their coordinates. It involves two steps for each iteration: the sorting pass and the refinement pass. Sorting pass results in the organized lists and the refinement pass does the real coding, which ultimately leads to a fully embedded bit-stream. SPIHT coder is restricted to images having pixel resolution of power 2. It provides higher PSNR with better image quality, low power consumption, compact output bit-stream, less complexity, and an intensive progressive transmission capability. 5. Experiments and Results

60

60

50

50

40

40

PSNR

PSNR

Colored image compression based on EZW and SPIHT encoding techniques has been undertaken using Wavelet Toolbox of MATLAB version 7.11.0 (R2010b) on Windows 7, 64-bit OS, with Intel Core i3 processor having, 2GB of RAM. Two same colored images of Jellyfish with standard pixel resolutions of 512x512 pixels and 256x256 pixels have been used. Both images named Jellyfish1 (512x512 pixels) and Jellyfish2 (256x256 pixels) which originally are in BMP format have been compressed using SPHIT and EZW coding techniques. In order to compress the colored image, the image is separated into three channels. On each channel, discrete wavelet transform and encoding is performed separately using input parameters viz; name of the wavelet and encoding technique. The three separately compressed channels are combined together to get the final output image which is in the compressed form. The results are analyzed, compared to each other, and discussed below. To achieve high compression and good quality so that it can be perceived by a human eye correctly and clearly, compression of both above-mentioned images was undertaken with different values of BPP (bits per pixel). In case of SPIHT encoding, the range for BPP remained from 0.038 (with 10 iterations) to 1.2339 (with 15 iterations) for 512x512 pixels image and 0.1499 (with 10 iterations) to 2.1466 (with 15 iterations) for 256x256 pixels image. These values in case of EZW remained from 0.1268 (with 10 iterations) to 3.5512 (with 15 iterations) and 0.4561 (with 10 iterations) to 4.5082 (with 15 iterations) respectively for 512x512 and 256x256 pixels image. The whole image was compressed for all of the above-mentioned values of BPP. Compression Ratio, PSNR and MSE, were calculated for the two techniques discussed above. PSNR vs. Bits per Pixel (BPP) of 512x512 image and 256x256 image are shown in figures 1(a) and 1(b) below:

30

30 20

20

EZW (256x256 pixels)

EZW (512x512 pixels) 10

10

SPIHT (256x256 pixels)

SPIHT (512x512 pixels) 0

0 0

0.5

1

1.5 2 2.5 3 3.5 BITS PER PIXEL (BPP)

4

4.5

0

5

(a)

0.5

1

1.5 2 2.5 3 3.5 BITS PER PIXEL (BPP) (b)

Fig. 1. (a) PSNR Vs. Bits Per Pixel (BPP) of 512x512 pixels image; (b) PSNR Vs. Bits Per Pixel (BPP) of 256x256 pixels image

[63]

4

4.5

5

Banday and Shah/COMMUNE – 2015

For 512x512 pixels image, both SPIHT and EZW showed PSNR above 35dB while as the BPP varied from 0.038 to 3.5512. Higher compression ratio of 16.4851 was obtained in SPIHT technique. Highest PSNR of 49.9057 was obtained using EZW technique. For 256x256 pixels image, SPIHT and EZW both showed PSNR above 36dB while as the BPP varied from 0.1499 to 4.5082, which is higher, than that for 512X512 pixels image. Higher compression ratio of 7.1707 was obtained using SPIHT technique. On observing the results, it is clear that SPIHT coding provides better image quality and higher compression ratio using less BPP in comparison to the EZW coding techniques. Figures 2(a) and 2(b) show MSE vs. Bits per Pixel (BPP) of 512x512 pixels and 256x256 pixels images. The MSE remained below 17.9538 for both techniques in case of 512x512 pixels image at 0.038 BPP while as in case of 256x256 pixels image; this value remained below 14.4617 at 0.1499 BPP. Further higher compression ratio and MSE of 16.4851 and 17.9538 respectively, at 0.038 BPP for 512x512 pixels image was found using SPIHT technique. EZW (512x512 pixels)

18

SPIHT (512x512 pixels)

15

SPIHT (256x256 pixels)

15

MSE

12

MSE

EZW (256x256 pixels)

18

9

12 9

6

6

3

3 0

0 0

0.5

1

1.5 2 2.5 3 3.5 BITS PER PIXEL (BPP)

4

4.5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

5

BITS PER PIXEL (BPP)

(a)

(b)

Fig. 2. (a) MSE Vs. Bits Per Pixel (BPP) of 512x512 pixels image; (b) MSE Vs. Bits Per Pixel (BPP) of 256x256 pixels image

60

60

50

50

40

40

PSNR

PSNR

Figure 3(a) and 3(b) show comparison between 512x512 pixels and 256x256 pixels image using SPIHT and EZW techniques respectively, on the basis of PSNR Vs. Bits Per Pixel. For 512x512 pixels and 256 x256 pixels image compressed using SPIHT showed higher PSNR value of 47.9547 at 1.2339 BPP and 49.5301 at 2.1468 respectively. A higher compression ratio of 16.4851 at 0.038 BPP was obtained for 512x512 pixels image using SPIHT as compared to the 256x256 pixel where a higher compression ratio of 7.1707 at 0.1499 BPP was obtained.

30 20

30

20 SPIHT (512x512 pixels)

EZW (512x512 pixels)

10

10

SPIHT (256x256 pixels)

EZW (256x256 pixels)

0

0 0

0.5

1

1.5 2 2.5 3 3.5 BITS PER PIXEL (BPP)

4

4.5

5

0

0.5

1

1.5 2 2.5 3 3.5 BITS PER PIXEL (BPP)

4

4.5

5

(b)

(a)

Fig. 3. (a) Comparison of 512x512 pixels and 256x256 pixels image using SPIHT (PSNR vs. Bits per Pixel); (b) Comparison of 512x512 pixels and 256x256 pixels image using EZW (PSNR vs. Bits per Pixel)

For 512x512 pixels and 256x256 pixels image compressed using EZW showed higher PSNR value of 49.9057 at 3.5512 BPP and 50.1811 at 4.5082 respectively. A higher compression ratio of 15.1396 at 0.1268 BPP was obtained for 512x512 pixels image using EZW as compared to the 256x256 pixel where a higher compression ratio of 6.8412 at 0.4561 BPP was obtained. From the above-explained results, it is concluded that SPIHT and EZW for 512x512 pixels image shows outstanding performance in terms of compression ratio.

[64]

Banday and Shah/COMMUNE – 2015

20

20

SPIHT (512x512 pixels) SPIHT (256x256 pixels)

EZW (256x256 pixels)

15

MSE

MSE

15

EZW (512x512 pixels)

10

10

5

5

0

0 0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

0

BITS PER PIXEL (BPP)

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

BITS PER PIXEL (BPP)

(a)

(b)

Fig. 4. (a) Comparison of 512x512 pixels and 256x256 pixels image using SPIHT (MSE vs. Bits per Pixel); (b) Comparison of 512x512 pixels and 256x256 pixels image using EZW (MSE vs. Bits per Pixel)

Comparison of 512x512 and 256x256 image using SPIHT and EZW on the basis of MSE Vs. Bits Per Pixel (BPP) is shown in Figure 4(a) and 4(b). For 512x512 and 256x256 pixels image compressed using SPIHT showed a lower MSE value of 1.0414 at 1.2339 BPP and 0.7246 at 2.1466 respectively. While as for 512x512 and 256x256 pixels image compressed using EZW showed a lower MSE value of 0.6645 at 3.5512 BPP and 0.6328 at 4.5082 BPP respectively. Also it is clear from the Figure 4(a) and 4(b) that SPIHT and EZW coding technique for 256x256 image showed outstanding performance in terms of MSE. Further high compression ratio and MSE was achieved in 512x512 pixels image. 6. Conclusion Application of discrete wavelet transformation with EZW and SPIHT coding to the colored images having pixel dimensions of 512x512 pixels and 256x256 pixels suggested effective compression at nearly no loss of quality of the image. Discrete wavelet transform based SPIHT coding showed outstanding performance in terms of compression ratio, PSNR and MSE. In comparison to the EZW coding, SPIHT coding not only used less BPP but also provided reasonable image quality at higher compression ratios. The results show maximum compression ratios of 16.4851 using SPIHT and that of 7.1707 using EZW for 512x512 pixels and 256x256 pixels images respectively. For both of these approaches, PSNR remained above 35dB, which is sufficient for perceiving the image correctly by humans. In addition, SPIHT coding technique used less time during the execution process. References Abdelhamid, M., Brahim, H., Ahmed, K., 2012. A survey of image compression algorithms for visual sensor networks, ISRN Sensor Networks, p. 1– 19. Gonzalez, R.C., Woods, R.E., 2004. Digital Image Processing, Reading. MA: Addison Wesley. ISO. Nov. 1991. Digital compression and coding of Continous-tone still images, part1, requirements, and Guidelines, ISO/IES JTC1 Draft International Standard 10918-1. Nagamani, K., Ananth, A., G., April -June 2011. EZW and SPIHT Image Compression Techniques for High Resolution Satellite Imageries, International Journal of Advanced Engineering Technology Computer Application, Vol. 2 , Issue 2 , P. 82-86, ISSN:0976-3945. Raja, S., P., Suruliandi, A., 2010. Performance Evaluation on EZW & WDR Image Compression Techniques, IEEE Trans on ICCCCT. Said, A., Pearlman, W., A, 1996. A new, fast, and efficient image codec based on set partitioning in hierarchical trees, IEEE Transactions on Circuits and Systems for Video Technology, Vol. 6, no. 3, p. 243– 250. Shapiro, J.M., 1993. Embedded Image Coding Using Zerotrees of Wavelet Coefficients, IEEE Transactions on Signal Processing, Vol. 41, p. 34453462. Singh, P., Nivedita, Gupta, D., Sharma, S., Sep 2012. Performance Analysis of Embedded Zero Tree and Set Partitioning In Hierarchical Tree, International Journal of Computer Technology & Applications, Vol. 3, p. 572-577, ISSN:2229-6093. Singh, P., Singh, P., Dec 2011. Design and Implementation of EZW & SPIHT Image Coder for Virtual Images, International Journal of Computer Science and Security IJCSS, Vol. 5, Issue 5, Kuala Lumpur, Malaysia, p. 433-442. Tao, M., Hempel, M., Dongming, P., Sharif H., 2013. A survey of energy-efficient compression and communication techniques for multimedia in resource constrained systems, IEEE Communications Surveys &Tutorials, Vol. 15, No. 3, p. 963–972. Watson, A., B., 1994. NASA Ames Research Canter, Image Compression Using the Discrete Cosine Transform, Mathematica Journal, Vol. 4, no. 1, p. 81-88. Wern, C.L., Ang, L.M., Phooi, S.K., 2008. Survey of image compression algorithms in wireless sensor networks, In: IEEE information technology, IT Sim 2008, International symposium, Vol. 4, p. 1–9. Yea, S., A., Pearlman, W., 2006. A Wavelet-Based Two-Stage Near-Lossless Coder, IEEE Transactions on Image Processing, Vol. 15, no. 11, p. 3488-3500.

[65]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

A Novel Universal (FNZ) Gate Based Adders in QCA Technology Z. A. Bangi*, F. A. Khanday Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India

Abstract Quantum-dot Cellular Automata (QCA) is a branch of Nanoelectronics that challenges to create general computation at nano scale by controlling the position of electrons within a quantum dot through Coloumbic interactions. QCA technology has a large potential to provide high density and ultralow power dissipation. These features enable us to develop fast, small and high performance QCA circuits for integration and computation. Among the design techniques, employing either the inverter and majority gate or the universal gates such as And-Or-Inverter (AOI) and Nand-Nor-Inverter (NNI) have been frequently used. Recently a new QCA design technique using a novel FNZ gate has been proposed by the authors. In this paper, this new universal QCA logic gate has been used to implement the half and full adders which enjoy superior performance vis-à-vis the already introduced half and full adders. The functionality of FNZ gate based half and full adder (FA) designs are verified by QCA designer tool where a detailed comparison of full adder with the previously reported designs confirms the reliable performance of the proposed designs.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Quantum dot Cellular Automata; FNZ gate; Nano-electronics; Full Adder; Majority Voter

1. Introduction Quantum-dot Cellular Automata (QCA) (Tougaw et al., 1993) is a nano-electronics digital architecture, in which the information is stored as configurations of electron pairs in quantum-dot arrays. To perform useful computations and build Boolean logic functions, arrays of coupled quantum dots are used in QCA (Lent, 2000) that take the benefit of quantum-mechanical effects to work at high speeds and ultra-low power levels, and drastically decrease the size of digital circuits. The binary values ‘0’ and ‘1’ in QCA can be represented by the position of electrons in quantum dots as compared to conventional digital technologies in which binary values can be represented by ranges of voltage or current. The key benefits of QCA are the exceptionally high logic integration derived from the small size of dots, the remarkably ultra-low power delay product and the simplicity (Amlani, 1997). The simple QCA cell contains four quantum dots coupled by tunnel barriers in a square array. The quantum mechanical tunnelling and Coulomb interaction are the two physical mechanisms for interaction between the quantum dots. Electrons, in quantum dots, cannot leave the cell, but they are able to tunnel between the dots, However, in ground state and in the absence of external electrostatic effect, if two mobile electrons are placed in the cell, Coulomb repulsion will force the electrons to dot on the opposite corners (Lent et al., 1994, Meurer et al., 1993, Amlani et al., 1999). Like the NAND/NOR gate forming the universal one in conventional digital technology, the fundamental QCA logical building blocks are majority gate and inverter as shown in Figs. 1 and 2 respectively. As shown in Fig. 1, based on the polarization status on the three inputs, the majority gate is driven to its lowest energy state which then corresponds to a particular logic function. For example, the majority gate can be used to build AND & OR gates as shown in Fig. 1(b) and 1(c) respectively.

*

Corresponding Author. Tel.:+91 9596 434488 E-mail address: [email protected] ISBN: 978-93-82288-63-3

Bangi and Khanday/COMMUNE-2015

(a)

(b)

(c)

Fig. 1: (a) The Majority Gate. (b) Majority Gate based AND gate. (c) Majority Gate based OR gate.

Fig. 2. QCA based Inverter.

The FNZ universal gate was introduced by authors in (Khanday et al., 2013). In its basic form, FNZ gives the NOT, NAND, NOR and buffer functions. In this paper, a novel FNZ universal gate based 1 bit full adder has been introduced. The proposed circuit is designed in such a way that suitable analysis of radius of effect of different QCA cell and the effect of layer spacing are carried out. The novelty of this paper lies in understanding novel universal FNZ gate based QCA architecture. Design of half adder using the FNZ gate is also proposed and the comparison of various 1-bit full adders are defined and analyzed with proposed design. 2. FNZ Gate The QCA implementation of the universal FNZ gate (Khanday et al., 2013) and its symbol are shown in Fig. 3. The design comprises of 8 cells with three inputs and one output. One of the input is vertically translated to 10nm while as the other two inputs are horizontally translated to 10nm making the whole design to occupy 6084nm2 (0.01μm2). The logic function realized by the Universal FNZ gate is:

F = A' B' + (A  B)C

Fig. 3: The QCA implementation of the universal FNZ gate and its symbol (Khanday et al., 2013).

Fig. 4: Simulation result of the universal gate

[67]

Bangi and Khanday/COMMUNE-2015

3. QCA Full Adders 1-bit full adder is the basic and most important part of any arithmetic and logic units. Other important adder circuits are composed of 1-bit full adders. A 1 bit full adder has three inputs A, B and Cin and two outputs Sum and Carry (Chander et al). The Boolean expression for generating the Sum and Carry are given below;

Sum = XOR(A, B, Cin) Carry = A.B + B.Cin + Cin. A It is necessary to write above equations using majority voter and inverter gates as in conventional QCA design; we just have majority voter and inverter logics. We can implement the logic AND gate and logic OR gate using 3-inputs majority gate as mentioned earlier. But using this method, the number of QCA cells becomes very high but the number of gates and clock used are less when compared. The logic expression of one bit full adder using majority voters and inverters are as under Sum  MV ( MV ( A , B , Cin ), MV ( A , B , Cin ), MV ( A , B , Cin ) Cout  MV ( A , B , Cin ) 4. Proposed Designs Our main objective is to develop an efficient design of QCA based full adder using universal FNZ gate. A high-level block diagram of half and full adder designs using FNZ gate are shown in Fig. 5 in which A, B and Cin are the inputs and Sum and Carry are outputs. The designs include Half Adder and Full adder which are made up of 4 and 9 FNZ gates respectively.

(a)

(b)

Fig. 5: Block Diagram of FNZ gate based (a) Half Adder, and (b) Full Adder.

The logic function realized by the Half Adder using Universal FNZ gate is given by Sum  FNZ ( FNZ ( A , B , 1 ), FNZ ( FNZ ( A , B , 0 ), 0 , 0 ), 1 )

Carry  FNZ ( FNZ ( A , B , 0 ), 0 , 0 ) And logic function of Full Adder is Sum  FNZ ( FNZ ( FNZ ( FNZ ( A , B , 1 ), FNZ ( FNZ ( A , B , 0 ), 0 , 0 ), 1 ), Cin , 1 ); FNZ ( FNZ ( FNZ ( FNZ ( A , B , 1 ), FNZ ( FNZ ( A , B , 0 ), 0 , 0 ), 1 ), Cin , 0 ), 0 , 0 ); 1 ) Carry  FNZ ( FNZ ( FNZ ( FNZ ( FNZ ( A , B , 1 ) FNZ ( FNZ ( A , B , 0 ), 0 , 0 ), 1 ), Cin , 0 ), 0 , 0 ); FNZ ( FNZ ( A , B , 0 ), 0 , 0 ); 1 ) The proposed Half Adder comprises of only 4 FNZ gates with two inputs (A, B) and two outputs (Sum, Carry). Two half adders are connected in such a way that less number of cells are used and hence occupy small area to design new FNZ gate based Full Adder. The final QCA implementation layouts of the proposed Half and Full adders are depicted in Fig. 6. In order to have correct signal propagation, the clock phases are sequentially carried out in the suitable order (0, 1, 2, 3,) so that clock phases are always contiguous to each another. These circuits were designed and simulated using QCA Designer tool. The simulation results of the designs, acquired by the QCA Designer Bistable vector simulation engine, are given in Fig.7.

[68]

Bangi and Khanday/COMMUNE-2015 Table 2: Comparison of the universal FNZ Gate Based Full Adder Design with Previously Reported Designs

Model

Wang et al. 2004

Lent et al. 1997

Hurehanro engra et al. 2007

Cho et al. 2009

Navi et al. 2010

Hashemi et al. 2012

4

No. of Gates

5

8

5

5

(1Five

(3MV +

(5MV +

(3MV +

(3MV +

input MV

2NOT)

3NOT

2NOT)

2NOT)

+ 1MV +

Delay (Clock Phase) Estimated area (µm2)

Proposed Design

3 (1Five input

5

MV + 1MV +

(3MV + 2NOT)

2NOT) No. of Cells

Hanninen et al. 2007

9 (FNZ)

1NOT)

145

192

108

86

88

51

102

74

5

NA

4

3

3

3

8

4

0.17

0.2

0.22

0.1

0.04

0.04

0.1

0.07

(a)

(b)

Fig. 6: QCA implementation of the universal FNZ gate based: (a) Half adder, and (b) Full Adder.

(a)

(b)

Fig. 7: Simulation result of the universal FNZ gate based: (a) Half adder, and (b) Full Adder.

Our main objective was to develop a highly-robust, efficient and less clock delay type half and full adder and the proposed FNZ gate based half and full adder designs are highly robust, fault tolerant, efficient and occupy less area. A comparison of proposed designs with those previously reported in the open literature is given in table 2. A study of this table shows that the proposed design of FNZ gate based full adder uses less area, less clock delay and less number of cell count. Further, the design has expressively smaller maximum wire length which leads to a higher maximum temperature for kink free operations. However, in comparison table, one of the adders (Hashemi et al., 2012) is more efficient in terms of number of cells and area but it uses conventional majority voters and inverters to design full adder which lacks in the signal polarity, power dissipation and fault tolerance. 5. Conclusion This paper describes the novel universal FNZ gate based half and full adders. These adders are optimized for QCA implementation in order to achieve attractive results. The proposed full adder is highly robust, fault tolerant and uses only one type of gate called universal FNZ gate which is more efficient as compared to other universal QCA gates. The

[69]

Bangi and Khanday/COMMUNE-2015

proposed adder offers less cell count, less latency (clock delay), high-speed response, infinitesimal area and ultra low power consumption as compared to some previously reported adders. The proposed half and full adder QCA circuits are simulated and their operations are analyzed using QCA designer bistable vector simulation. References Amlani, Orlov, A. O., Bernstein, G. H., Lent, C. S., Snider, G. L., 1997. Realization of a functional cell for quantum-dot cellular automata, Science, 227, p. 928. Amlani, Orlov, A. O., Toth, G., Bernstein, G. H., Lent, C. S., Snider, G. L., 1999. Digital logic gate using Quantum-dot cellular automata, Science, 284, p. 289. Chander Ravi, Krishna P. Murali, Design of Efficient Hybrid Adder Using QCA, International Journal of Engineering Science Invention, ISSN (Online): 2319 – 6734, ISSN (Print): 2319 – 6726, www.ijesi.org, pp.30-34. Cho et al, 2009. Adder design and analyses for quantum-dot cellular automata, IEEE, Transactions on Computer, vol. 58. Hanninan, I., Takala, J., 2010. Adders Based On Quantum dot Cellular Automata, journal of Signal processing Sytems vol 58 issue 1 pp. 87-103. Haruehanroengra, S., Wang, W., 2007. Efficient Design of QCA Adder Structures, Solid State Phenomena, p. 121. Hashemi, S., Tehrani, M., Navi, K., 2012. An Efficient Quantum-dot Cellular Automata Full Adder, Scientific Research Essays, vol. 7, p.177. K. Navi et al, 2010. A New Quantum Dot Cellular Automata Full Adder, Microelectronics Journal, vol. 41, p. 820. Khanday, F. A., Kant, N. A., Bangi, Z. A., Shah, N. A., 2013. A Novel Universal (FNZ) Gate in Quantum Dot Cellular Automata (QCA), IEEE International Conference on Multimedia, Signal Processing and Communication Technologies (IMPACT-2013), p. 255. Lent, C. S., Amlani, Orlov, A. O., Kummamuru, R. K., Bernstein, G. H., Snider, G. L., 2000. Experimental demonstration of a leadless quantum dot cellular automata cell, Applied Physics Letters, 77, p. 738. Lent, C. S., Tougaw, P. D., Porod, W., 1994. Quantum cellular automata: the physics of computing and arrays of quantum dot molecules, Proceedings of the Workshop on Physics and Computing, IEEE Computer Society Press: Dallas, TX, p. 5. Lent, S., Tougaw P. D., 1997. A Device Architecture for Computing with Quantum Dots, Proceeding of the IEEE, vol. 85, NO.4, pp. 541-557. Meurer, B., Heitmann, D., Ploog, K., 1993. Excitation of three dimensional quantum dots, Physical Review B, vol. 48, p. 11488. Safavi, A. A., Mosleh, B. M., 2013. An Overview of Full Adders in QCA Technology, International journal of Computer Science & Network Solutions, vol. 1, issue 4, pp. 12-35. Tougaw, P. D., Lent C. S., 1993. Lines of interacting quantum-dot-cells: a binary wire, Journal of Applied Physics, vol. 74, p. 6227. Tougaw, W. Porod, Bernstein, G. H., 1993. Quantum cellular automata, Nanotechnology, vol. 4, issue 1, pp. 49-57. Zhang, R., Walus, K., Wang, W., Jullien, G. A., 2004. A method of majority logic reduction for Quantum Cellular Automata., IEEE Trans. Nanotechnology. vol. 3, no. 4, pp. 443–450.

[70]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

A Study of CMOS Frequency Synthesizers in Short Range Wireless Communication M. Tariq Banday, Farooq Aadil* Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India

Abstract CMOS technology is being widely adopted to fulfil the growing demand for low power consumption and low phase noise in wireless communication systems. The technology requires circuits to operate at frequency, which can be as low as several kilohertz to as high as thousands of gigahertz. In submicron technology, digital radio frequency (RF) architectures will have to be adopted to reduce cost, power consumption, and required area of the system. As the minimum feature size decreases, the supply voltage will also decrease. One of the most critical blocks in a wireless transceiver is the frequency synthesizer, which significantly affects its cost, batterylifetime, and performance. CMOS frequency synthesizers are commonly used in 2.4GHz to 5GHz ISM band of short-range wireless communication systems. IEEE standard define the PHY and MAC layers of wireless communication over an action range from 1m to 100m such as Bluetooth (IEEE 802.15.1), ZigBee (IEEE 802.15.4), and Wi-Fi (IEEE 802.11a/b/g/n/ac/ad/r/WG). IEEE 802.11ad/WG operates in 60GHz ISM band. This paper provides a detailed comparison of various CMOS frequency synthesizers (PLL) in Bluetooth, Zigbee, and Wi-Fi wireless communication systems. It compares the performance of recently proposed frequency synthesizers for Bluetooth, Zigbee, and Wi-Fi systems in terms of their frequency range, phase noise, power dissipation, supply voltage, and the type of frequency synthesizer used.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Frequency Synthesizer; Bluetooth; ZigBee; Wi-Fi; Frequency Divides

1. Introduction Complementary Metal Oxide Semiconductor (CMOS) digital integrated circuit is an enabling technology for modern information age, owing to its intrinsic features such as low-power consumption, large noise margins, and ease of design. Considerable reduction in chip size (less than 1mm2) has become possible by transition from millimeter to micrometer scale. Frequency synthesizer is one of the most important blocks in wireless communication systems and therefore, in recent years RF Industrial, Scientific and Medical (ISM) 2.4GHZ to 5GHz short-range wireless communication band has attracted significant research attention. These short-range communication bands are Bluetooth, ZigBee, and Wi-Fi. Bluetooth, governed by IEEE 802.15.1 standard operates in the ISM band of 2.4GHz with data rate of 1Mbps. ZigBee, governed by IEEE 802.15.4 standard is a low-cost, low power, wireless mesh network that operates in any of the three frequency bands viz. 868MHz (868MHz to 868.6MHz), 915MHz (902MHz to 928MHz), and 2.4GHz (400MHz to 2483.5MHz) supporting data rate of 250Kbps. Wi-Fi, governed by IEEE 802.11a, 802.11b and 802.11g standards operates in 2.4GHz and 5GHz ISM bands with data rate of 11Mbps, 24Mbps, 54Mbps respectively. Recently, a new Wi-Fi standard namely IEEE 802.11ac/ad has been introduced that has high QoS and better performance to reach the Wi-Fi hotspots. A frequency synthesizer consists of a reference signal, phase frequency detector, charge pump, low pass filter, voltage controlled oscillator and feedback divider blocks. Various types of frequency synthesizers are being used in different wireless applications. For low power consumption and low phase noise, mainly Integer-N, ADPLL, and fractional-N Sigma delta modulator PLL are used. For reduced die area and wide tuning range, Ring VCO PLL’s are used.

* Corresponding author. Tel.: +91 9797 293440. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Banday and Aadil/COMMUNE – 2015

2. Frequency Synthesizers A frequency synthesizer (FS) is a device capable of generating a set of signals of given output frequencies with very high accuracy and precision from a single reference frequency. It consists of reference source mainly crystal oscillator, phase frequency detector, low pass filter and voltage controlled oscillator (VCO). The voltage-controlled oscillator (VCO) controls the tuning range and phase noise. Frequency synthesis is the frequency changing process whereby a new frequency is derived from a given fundamental frequency by combinations of several operations such as additions, subtractions, multiplications, and divisions. Phase locked loop is widely used in various communication applications such as frequency synthesis in wireless communication, radio computer, clock generation, microprocessors, digital circuits, and disk drive electronics. Frequency synthesizers can be classified by their operating frequency range, phase noise performance, settling time, speed, power consumption, and active die area. CMOS frequency synthesizers are aimed to build low power consumption and high speed wireless systems. These include Direct Analog PLL frequency synthesizer, Integer-N PLL frequency synthesizer, Fractional-N PLL frequency synthesizer, Sigma Delta Modulator (SDM) PLL frequency synthesizer, All Digital PLL (ADPLL) frequency synthesizer, Injection locked loop PLL frequency synthesizer, Ring Oscillator based frequency synthesizer and Delay Locked Loop frequency synthesizer (Farazian et al; 2014). 2.1. Direct Analog Direct analog frequency synthesizer uses analog techniques for frequency synthesis. Analog frequency synthesizers have fast switching and fine resolution when repeated arbitrarily many times and are good for microwave applications. The main drawback of analog frequency synthesizer is its large size and noise, and requirement for excessive hardware, which consumes more power. 2.2. Indirect (Integer-N PLL) The based Integer-N frequency synthesizer consists of PLL with feedback network having integer division ratio. By changing the division ratio frequency at the output of VCO is measured. The main drawback of integer-N PLL is its limited bandwidth, low settling time and more in-band phase noise. 2.3. Fractional-N PLL In order to reduce in-band spurs or noise, the integer division is changed to fractional value, so a large reference can be used to achieve a small frequency resolution and the smallest frequency step can be the fraction of reference frequency. The fractional-N PLL has relatively fast switching and increases the frequency resolution. Fractional–N PLL consists of divider network with fractional division (N/N+1), where N is some integer value. The main drawback of Fractional-N PLL is its fractional spurs at the output of VCO. 2.4. Sigma Delta Modulator PLL Fractional-N sigma delta modulator limits the drawback of Fractional-N PLL’s spurs using fractional divider ratio with random selection of divider value, which increases the speed and noise performance of the system. It also pushes the phase noise associated with the divider from low frequencies to high frequencies. The loop filter filters out the phase noise in high frequencies. 2.5. All Digital PLL In order to provide smooth quantization noise and power consumption performance, the modulator should complete its cycle and return to its starting original state. All digital PLL (ADPLL) is a digital PLL, which increases the phase noise performance and locking time of the PLL. ADPLL are more versatile, efficient, and flexible than other techniques. In ADPLL, VCO is replaced by digital controlled oscillator (DCO), phase frequency detector and charge pump is replaced by digital to time convertor (DTO). Direct Digital frequency synthesizer uses digital techniques for high speed, high frequency resolution, and low jitter. The ADPLL has the input frequency range from 40MHz to 98MHz. Another advantage of ADPLL is its low power consumption. 2.6. Injection Locked Loop The injection locked loop frequency synthesizer uses mesh technology employs multiple PLL’s where output of first PLL is fed to the second PLL. The injection locking is used to control the locking range. It reduces the power consumption and phase noise due to injected signal. When the coupling between two PLL’s is strong and the

[72]

Banday and Aadil/COMMUNE – 2015

frequencies are close enough, a process called injection pulling occurs (i.e. the second oscillator captures the first oscillator). Injection locked loop are used for low power consumption, high speed and small die area. 2.7. Delay Locked Loop The main difference between PLL and the DLL is that internal voltage-controlled oscillator is replaced by a delay line in DLL PLL. A DLL can be used to change the phase of a clock signal (a signal with a periodic waveform) called clock recovery. The delay locked loop cannot generate multiple output frequencies unlike in PLL where division value is just changed to generate the other output frequency. The main component of a DLL is a delay chain composed of many delay gates connected front-to-back. The chain input is connected to the clock that is negatively delayed. A multiplexer is connected to each stage of the delay chain; the selector of this multiplexer is automatically updated by a control circuit to produce the negative delay effect. The output of the DLL is the resulting, negatively delayed clock signal. 2.8. Ring Voltage Controlled Oscillator Ring oscillators are used to provide a wide tuning range than other types of PLL’s. A ring based voltage controlled oscillator consists of odd number of delay elements with output of last stage connected to the input of the first stage. The ring oscillator must have phase shift of 2π and unity gain in order to achieve oscillation. Each delay element must provide a phase shift of π/N, where N is the number of delay stages. The main advantage of Ring VCO based PLL technique is its reduced die area and wide tuning range. 3. Recent CMOS Frequency Synthesizers CMOS frequency synthesizers are widely used in 2.4GHz ISM band of short-range wireless communications ranging from 1m to 100m such as Bluetooth, ZigBee, WI-Fi, and UWB. Bluetooth governed by IEEE 802.15.1 standard operates in the ISM band of 2.4GHz has a data rate of 1Mbps and uses DS-FH/FSK technology. ZigBee is a low-cost, low power, wireless mesh networking standard using IEEE 802.15.4 standard. The low cost allows the technology to be widely deployed in wireless control and monitoring applications, the low power-usage allows longer life with smaller batteries, and mesh networking provides high reliability and more wide range. It operates in four frequency bands 784MHz band in China, 868–868.6MHz (868MHz band in Europe), 902–928MHz (915MHz in USA and Australia), and 400 – 2483. 5MHz (2.4GHz worldwide) band with data rate of 20kb/s for 784MHZ band, 250Kbps for 868MHz and 2.4 GHz bands. The ZigBee Alliance is responsible for the ZigBee wireless technology, which defines network, security and application layers upon the IEEE 802.15.4 Physical, and Media Access Control layers. The main advantage of ZigBee technology is that it has high data rate, and consumes less power. The ZigBee network layer supports both the Star, Mesh and tree network topologies. A Wi-Fi system is the WLAN network with the operating range from 10 meters to 100 meters. IEEE 802.11a/b/g/n are the IEEE standard for Wi-Fi. The data rate for Wi-Fi system is 54Mb/s and it uses OFDM modulation technique. IEEE 802.11n specifically IEEE 802.11ac and IEEE 802.11ad are the latest standards for Wi-Fi. IEEE 802.11ac has been developed to deliver its throughput over the 5 GHz band, affording easy migration from IEEE 802.11n, which also uses 5 GHz band (as well as the 2.4 band). These amendments aim to provide gigabit speed WLAN and use of unlicensed 60 GHz band. A comparison of various CMOS frequency synthesizers based on IEEE 802.11.4 standard (Bluetooth) is given in table 1. Table 1. Comparison of recently proposed CMOS frequency synthesizers for Bluetooth communication

Frequency Synthesizer

Frequency

Input Voltage

Phase Noise

Power Dissipation

Frequency Division

2.4GHZ LC VO using 0.13µm and 0.35µm (Rahim et al, 2010)

(Tuning Range) 1.973 GHz to 2.667GHz

3V vcntl = 0-2.4V

114.8dBc/Hz @1MHz

12mW

Integer- N

Fractional N using 32nm VLSI (Manwatkar, 2014, Padole, 2014)

2.1GHZ for 2.5GHz

1V

Not Necessary

57.4µW

SDM

Integer N Based on Pulse Swallow Topology (Harish, 2014 ; Shaguftha, 2014)

5GHz

1V

Not Necessary

9.7mW

32/33/47/48 Prescalar = 2/3 DAC = /16

ADPLL for Low Energy TX (Pereira, 2013)

Fref=16MHz for 2.48GHz

1.8V

-120dBc/Hz @3MHz

79.27µW DCO = 490µW Total = 605µW

DSDM Divide by 8

1.8V

0.96 mW @1MHz 2.2 mW @5MHZ

Divider Consumes 52mW

32/33/47/48 P= 2/3 Swallow divider with Frequency Resolution of 1 MHz -25MHz

-107.59dBc/Hz @1MHz -133.44dBc/Hz @10MHz

64.51µW

Ring Oscillator with Phase Shift of 2 Pi

2.41GHz to 2.483GHz Pulse Swallow Topology with Multiband Divider 5.14GHz to 5.30GHz (Kumar, 2014, Saraaravanan, 2014) 5.715GHz to 5.815GHz

SERO in 0.18 µm CMOS (Gun Pu et al, 2010)

0.7 to 1.7V (Operating Freq. Range) (Tuning Range) 2.4GHz to 5.43GHz 55.77%

[73]

Banday and Aadil/COMMUNE – 2015

Table 1, shows that the ADPLL frequency synthesizers are used for high speed, wide tuning range low jitter, and very high-resolution. Ring based VCO’s are generally used for clock generating circuits. Due to the wide tuning range and electrically tuned property of ring oscillators, they are used in wireless communication to reduce the power consumption and chip area. The frequency synthesizer implemented using fractional-N sigma delta modulator PLL in 32nm CMOS technology consumes low power of 57.4µW (Manwatkar, 2014; Padole, 2014). The frequency synthesizer implemented using ADPLL (Pereira, 2013) for low energy transmitter uses sigma delta modulator with division ratio of 8. The phase noise of the PLL is reduced to -120dBc/Hz with reference frequency of 16MHz. The Single Ended Ring Oscillator (SERO) frequency synthesizer using 0.18 µm with SERO based ring oscillator has improved phase noise 117.59dBc/Hz @1MHz and -133.44dBc/Hz @10MHz, and power consumption of 64.51µw. The tuning range of the proposed ADPLL is 55.77% (Gun Pu et al, 2010). In table 2, a comparison of recently hardware implemented CMOS frequency synthesizers for Bluetooth communication are compared. Table 2. Comparison of recently proposed hardware implemented CMOS frequency synthesizers for Bluetooth communication

Frequency Synthesizer

Frequency

Input Voltage

Phase Noise

Power Dissipation

Frequency Division

Fractional N SDM using 0.18 µm (Pamarti et al, 2004)

Fref=48 XTL for 2.4GHz

1.8V-2.2V

-121dBc/Hz @3MHz

34.4mW PLL BW= 460MHz

SDMFSK Modulator with fc =2404 +k, k=1,2,…78

Fractional N SDM (Gun Pu et al , 2010)

Fref=40MHz XTL O/P Freq. = 1.8 GHz to 5.8GHz

Not Necessary

-119dBc/Hz @ 1MHz offset from 4.2GHz carrier

55.92mW

SDN=26-28-1 O/p of buffer=1/4,1/2

1.86mX1.8m i.e. 3.348mm2

Low power ADPLL 0.13 µm (Gun Pu et al, 2011)

Fout=2.4GHz DCO Freq. Resolution = 0.14KHz

1.2V

-120.5dBc/Hz @1MHz FOM=-192.58 dBc/Hz

12mW

TDC resolution=1ps DCO tuning range= 58% at 2.4GHZ

0.8mm2

Die Area

6.94mm2

The ADPLL frequency synthesizer (Gun Pu et al, 2011) has been implemented using in 0.13µm technology. It has reduced the die area to 0.8mm2. Implementing synthesizers using ADPLL has also improved the phase noise performance of the system. The PLL consumes 12mW of power with tuning range and die area of 58% and 0.8mm2, respectively. The fractional-N sigma delta modulator frequency synthesizer (Pamarti et al, 2004) has been implemented with control frequency of 2404+k, where k=1, 2, 3….78. This technique has reduced phase noise to -121dBc/Hz at 3MHz offset. In table 3, frequency synthesizers proposed for ZigBee short-range wireless communication, which have been realized in hardware, are compared. Table 3. Comparison of recently hardware implemented CMOS frequency synthesizers in ZigBee short-range wireless communication

Frequency Synthesizer

Frequency

Input Voltage

Phase Noise

Power Dissipation

Frequency Division

Die Area

Integer N PLL using 0.18µm (Mandal, 2008; Bhattacharyya, 2008)

2.4GHz

1.8V

-81.55dBc/Hz @100KHz -108dBc/Hz@1MHz

7.95mW

CML =2/3 CMOS=2/3 Settling Time = 25µs

0.75x0.65 = 0.4875 mm2

Integer N PLL using 0.18 µm (Sahafi et al, 2010)

Fref= 5MHz

1.8V

Not Necessary

410µW

S= 7/8, P = 64 33≤S≤48

36x45 = 1620 mm2

Integer N PLL using 0.18µm CMOS (Chore 2013 ; Honade 2013)

2.4GHz

1.8V

-113.4bBc/Hz @1MHz

3.2mW

Dual modulus Prescalar = 15/16 P=32

1x1mm2

0.25 µm Integer N PLL (Aravind, 2014; Praveena 2014)

Fref=31.2MHz For 2.4GHz Fout = 2GHz

2.5V

-107dBz/Hz @600MHz Offset

23mW

Dual modulus Prescalar = 64/65/128/129 Ref. Divider = ½

1.5mm2

The frequency synthesizer implemented using integer-N PLL in 0.18 µm CMOS (Mandal, 2008; Bhattacharyya, 2008) consumes 7.95mW of power and uses two frequency dividers CMOS and CML with division ratio of 2/3. The PLL has the phase noise of -81.55dBc/Hz at 100MHz and -108dBc/Hz at 1MHz, with reduced die area of 0.487mm2. The frequency synthesizer implemented using integer-N PLL in 0.18 µm CMOS technology (Chore, 2013; Honade, 2013) with the input voltage of 1.8V has phase noise and power consumption of -113dBc/Hz at 1MHz and 3.2mW respectively. The proposed technique uses the dual modulus Prescalar of 15/16 and 7/8; the die area has been reduced to 1mm2 with frequency range of 2.4GHz. Integer-N PLL frequency synthesizer using 0.25µm CMOS technology

[74]

Banday and Aadil/COMMUNE – 2015

(Aravind, 2014; Praveena 2014) has the phase noise and power consumption of -107dBc/Hz at 600MHz and 23mW, respectively. The proposed technique uses the reference frequency of 31.2MHz and output frequency of 2GHz using the dual modulus Prescalar value of 64/65/128/129. The die area in the proposed technique is 1.5mm 2. Table 4 compares recently proposed CMOS frequency synthesizers for ZigBee short-range wireless standard. Table 4. Comparison of recently proposed CMOS frequency synthesizers in ZigBee wireless standard

Frequency Synthesizer

Frequency

Input Voltage

Phase Noise

Power Dissipation

Frequency Division/ Settling Time

Integer N PLL pulse swallow divider (Harish, 2014; Shaguftha 2014)

5GHz

1V

Not Necessary

9.7mW

S=6 Bit 128/129/191/192 P=2/3

Integer N PLL using 0.18 μm (Aravind, 2014; Praveena, 2014)

Fref = 531MHz Fout = 2.4GHz, 5 GHz to 5.825GHz

0.6V

-44.77dBc/Hz

9.6mW

Ring VCO in 0. 18 μm CMOS (Talebi et al, 2013)

2.4 -2.48 GHz

1.8V

-117dBc/Hz @3.5 MHz -119dBc/Hz @10MHz

6.1mW

Injection locked FS using 0.18 µm (Talebi et al, 2013)

2.4- 2.48GHz

-116dBc/Hz @3.5MHz -118dBc/Hz @10MHz

7.3mW Current = 2.2mA

1.8V

P=71, s=1 Freq. division=2400 Lock time = 4µs

Settling time = 25μs

Ring VCO Settling time 28µs

The frequency synthesizer implemented using integer-N PLL in 45nm CMOS technology (Aravind, 2014; Praveena, 2014) has the reference frequency of 1MHz and phase noise of -44.77dBc/Hz. The proposed technique reduced the power consumption of 1.026mW with locking time of 4µs and input voltage supply of 1V. Ring oscillator based PLL uses 0.18μm CMOS technology with wide frequency range from 2.4GH to 2.48GHz and power consumption of 6.1mW (Talebi et al, 2013). The proposed technique has low phase noise of -117dBc/Hz at 3.5MHz and -119dBc/Hz at 10MHz. The frequency synthesizer is implemented using injection locking PLL in 0.18µm CMOS technology (Talebi et al, 2013). The PLL has the improved phase noise of -116dBc/Hz at 3.5MHz and -118dBC/Hz at 10MHz with the input voltage supply of 1.8V. The PLL has settling time of 25µs and consumes 7.3mW of power. The proposed PLL has low settling time and has improved phase noise performance. Table 5 compares various CMOS frequency synthesizers implemented in hardware for Wi-Fi communication system. Table 5. Comparison of recently proposed CMOS frequency synthesizers in Wi-Fi communication (hardware implemented)

Frequency Synthesizer

Frequency

Input Voltage

Phase Noise

Power Dissipation

Frequency Division

Die Area/Frequency Resolution

Pulse swallow topology with multiband divider (Jalil et al, 2014)

Not Necessary

1.8V

Not Necessary

0.96 mW @ 1MHz 2.2 mW @ 5MHZ Divider Consumes 52mW

32/33/47/48 P= 2/3

Swallow Divider with Freq .Resolution = 1 MHz to 25MHz

SDM fractional N PLL (Jalil et al, 2014)

Fref = 40MHz O/P Freq. = 4.6GHz to 5.4GHz

1.2V

-106dBc/Hz @1MHz offset

39mW Lock Time = <13.2µs

Fractional N= 113-143 Loop BW=700MHz

0.48mm2

SDM fractional N 0.18µm (Huang et al, 2010)

Fref =40MHz O/P Freq. = 1.8 GHz to 5.8GHZ

1-2V

-119dBc/Hz @1MHz offset

55.92mW Setting Time <50µs

Divide ratio = 26 to 28-1 Divide form 3.5GHz

1.86mX1.8m = 3.348mm2

The frequency synthesizers use multi modulus dividers to get the more frequency resolution and produce more VCO output frequencies. The frequency synthesizer implemented using CMOS technology with pulse swallow topology using multiband divider and has the voltage supply of 1.8V. The PLL has the low phase noise performance of -0.96 dBc/Hz at 1MHz and -2.2dBc/Hz at 5MHz (Jalil et al, 2014) is. The frequency synthesizer using sigma delta modulator fractional-N PLL has been implemented with output frequency range of 4.6GHz to 5.4GHz and using the fractional-N division ratio 113 to 143. The technique uses 40MHz of reference frequency and has phase noise of -106dBc/Hz at 1MHz offset (Jalil et al, 2014). The proposed technique has the fast locking range. The die area of the present technique is also reduced to 0.48mm2 with input supply of 1.2V. The frequency synthesizer is implemented using sigma delta modulator fractional-N PLL in 0.18µm with tunable output frequency range from 1.8GHz to 5.8GHz and consumes 39mW of power with the settling time of less than 50µs (Huang et al, 2010). The proposed technique also has improved

[75]

Banday and Aadil/COMMUNE – 2015

phase noise of -119dBc/Hz, division ratio of 26 to 28-1 and the die area is 3.348mm2. Table 6 compares recently proposed CMOS frequency synthesizers in Wi-Fi communication. Table 6. Comparison of recently proposed CMOS frequency synthesizers in Wi-Fi communication (hardware implemented)

Frequency Synthesizer

Frequency

Input Voltage

Phase Noise / VCO Sensitivity

Power Dissipation

Frequency Division

Integer N pulse swallow (Harish, 2014; Padole, 2014)

5GHz

1V

Not Necessary

9.7mW

Multi modulus = 32/33/47/48 P=2/3, DAC=/16

Low power ADPLL with high resolution TDC 0.13 µm (Park et al, 2011)

2.4GHz

1.2V

-120dBc/Hz at 1MHz

12mW

DCO Tuning Range = 58%

2.4GHz LC VCO using 0.13 µm and 0.35 µm (Rahim et al , 2010)

Tuning range =1.973-2.667GHz

3V Vcntl = 0 to 2.4V

-114.8dBc/Hz @1MHz

12mW

Ring Oscillator

In-band Noise Floor is -108 dBc/Hz rms Jitter from 1kHz to 10MHz is 0.19ps

36mW

5825MHz clock

Ref Freq = 40 MHz Fractional N digital PLL with TDC Tunable = 5.9GHz to 55nm CMOS (Yao et al, 2011) 8.0GHz

Dual band integer N PLL in 0.18µm (Yu et al, 2011)

2.4 GHz / 5.2GHz

1.8V

Not Necessary

22mW

7/15×P+S P, S are 2-2n (P>S)

Ring VCO in 0.18 µm CMOS process (Jalil, et al, 2014)

2.42GHz

1.5V

-126.62dBc/Hz at 25MHz offset

2.47mW

Tuning range = 0.5 GHz to 2.54GHz

The frequency synthesizers compared in table 6 have improved phase noise and reduced power consumption. Large tunable range frequency synthesizer using ADPLL with delta sigma modulator (Yao et al, 2011) can be used for many applications with wide frequency tunable from 5.9GHz to 8.0GHz. The frequency synthesizer implemented using integer-N PLL and the multi modulus technique with division of ratio of 32/33/47/48 and DAC is 16 (Harish, 2014; Padole, 2014). The proposed synthesizer consumes only 9.7mW of power. The frequency synthesizer implemented using 3-step TDC with active inductor DCO. (Park et al, 2011) has improved phase noise of -120dBc/Hz and dies area of 0.8mm2. The dual CMOS technology of 0.13µm and 0.35µm has been implemented using 2.4GHz LC VCO has tuning range of 1.973 to 2.667GHz from the voltage supply of 3V (Rahim et al , 2010). The proposed synthesizer consumes 12mW of power with the phase noise of -114.8dBc/Hz. Synthesizer has the in-band noise floor of 108dBc/Hz and the power consumption of only 36mW from 40MHz reference frequency. The frequency synthesizer has the wide frequency tunable range of 5.9 to 8GHz (Yao et al, 2011). The proposed frequency synthesizer uses the wide band frequency divider for 2.4/5GHz frequency range with a dual band integer N PLL in 0.18 µm cmos technology (Yu el at, 2011). It consumes 22mW of power from 1.8V power supply. A ring VCO based synthesizer in 0.18µm CMOS technology with improved phase noise of -126.62dBc/Hz and the wide frequency tuning from 0.5GHz to 2.54GHz has also been proposed (Jalil, et al, 2014). 4. Conclusion The average channel bandwidth for Wi-Fi is 22MHz with the normal transmitter power of 15-20dBm. As many as 2007 cell nudes and 14 radio frequency (RF) channels for 2.4GHz band are used. For Bluetooth, the modulation technique used is FHSS for 2.4GHz frequency band and the numbers of currently RF channels are 79 with the normal transmitter power of 0-10dBm. The maximum numbers of cell nodes used in Bluetooth are eight. The modulation scheme used in Zigbee is DSSS with 16 RF channels and the number of cell nodes greater than 65000 for normal transmitted power of -25 to 0dBm. The CMOS PLL has comparatively reduced power consumption and phase noise and therefore has improved the performance of communication systems. The ring oscillators base frequency synthesizers considerably reduces the chip area and power consumption. The SERO based ring oscillator has additional advantage of having lowest power consumption and wide tuning range. The sigma delta modulator has lowest phase noise. The TDC with sigma delta fractional–N modulator used in the survey has improved phase noise and power consumption. Improved performance such as wide tuning range, low power consumption, and reduced die area are possible if the supply voltage is reduced below 1volt and ADPLL with sigma delta modulator PLL technique is realized in hardware.

[76]

Banday and Aadil/COMMUNE – 2015 References Abdul Rahim A., Siti M.H., Nazif Emran F., Norman Fadhil Idham M., Mohamed Razman Y., 2010 IEEE . A low phase noise large tuning range 2.4 GHz LC voltage controlled oscillator, I 978-1-4244-7456/10. Manwatkar, A., V., Padole, V., B., June 2014. Design of 2.5 GHz Phase locked loop using 32nm CMOS technology, International Journal of Engineering and Computer Science, ISSN: 2319-7242. Harish, O., Shaguftha, Sk., 2014. A multiband flexible integer-N divider based on pulse swallows topology, Vol 05, Article 09412, and September International Journal of VLSI and Embedded Systems-IJVES ISSN: 2249 – 6556. Patrick G. Pereira, August 2013. All-Digital Phase Locked Loop for Bluetooth Low Energy Transmitters, Institute of Superior Technology, Technical University of Lisbon Lisboa, Portugal. R. Naveen Kumar, Saraaravanan, V., A., 2014. Design of 2/3 Prescalar Low Power Flexible Multiband Divider, International Journal of Innovative Research of computer and communication Engineering Vol 2, ISSN 2320-9801. Jubayer Jalil, Mamun Bin Ibne Reaz, Siti Sarah Binti Md Sallah, 2014. A low power servo VCO in 0.18μm CMOS Process, ISSN 1330-3651. Sudhakar Pamarti, Lars Janssen and Ian Galton, 2004. A Wideband 2.4-GHz Delta-Sigma Fractional-N PLL with 1-Mb/s In-Loop Modulation, IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 39. Young Gun Pu, AnSoo Park, Joon-Sung Park, and Kang-Yoon Lee, 2003. A Fractional-N Frequency Synthesizer for Cellular and Short Range Multistandard Wireless Receiver, IEEE. Young Gun Pu, AnSoo Park, Joon-Sung Park, and Kang Yoon Lee, 2011. Low-Power, All Digital Phase-Locked Loop with a Wide- Range, High Resolution TDC, ETRI Journal, Volume 33. Debashis Mandal and Bhattacharyya, T., K., 2008. Implementation of CMOS Low-power Integer-N Frequency Synthesizer for SOC Design, JOURNAL OF COMPUTERS, VOL. 3, NO. 4. Ali Sahafi, Jafar Sobhi, Mahdi Sahafi, and Omid Farhanieh, 2011. An Ultra-Low Power Frequency Divider for 2.45GHz Zigbee Frequency Synthesizer, ELECO 7th International Conference on Electrical and Electronics Engineering, 1-4 December, Bursa, TURKEY. Amruta M. Chore, Shrikant J. Honade, L, 2013. Low power fractional-N PLL frequency synthesizer using 45nm VLSI technology, International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering Vol. 2, Issue 4. Aravind, A., Praveena, G., 2014. A low power multiband single phase clock distribution for Wi-Fi applications, International journal of professional engineering studies, volume III/issue3/IJPRES. Fatemeh Talebi, Hassan Ghafoorifard, Samad Sheikhaei, Elias Soleiman. 2013. An Injection Locked Ring VCO with Enhanced Phase Noise for 2.4GHz Band Zigbee Applications, Journal of American Science, ISSN: 1545-1003. Fatemeh Talebi1, Hassan Ghafoorifard1, Samad Sheikhaei2, Sajjad Shieh Ali Saleh2. 2013. A Low Phase Noise Ring-VCO Based PLL Using Injection Locking for Zigbee Applications, Circuits and Systems, 304-315. Meng -Ting Tsai, Ching-Yuan Yang,. 2010. A fast-locking agile frequency synthesizer for MIMO dual-mode Wi-Fi/WiMAX applications, Analog Integrated Circuits Sig Process, 64:69–79, Springer. Deping Huang, Zin Zhou, Wei Li, Ning Li, and Junyan Ren. 2010 IEEE. A fractional- N frequency synthesizer for Cellular and short range multistandard wireless receiver, 978-1-4244-5309-2. Joon-Sung Park, YoungGun Pu, AnSoo Park, and Kang-Yoon Lee, 2011. Low-Power, All Digital Phase-Locked Loop with a Wide-Range, High Resolution TDC, ETRI Journal, Volume 33. Abdul Rahim , A., I., Siti M.H., Nazif Emran F., Norman Fadhil Idham M., Mohammad Razman Y., IEE 2010A low phase noise and large tuning range 2.4GHz LC Voltage –Controlled Oscillator, microwave and nanotechnology , TMR &D innovation centre, Malaysia. Chih-Wei Yao , Li Lin , Nissim, B., Arora H. and Cho, T., 2011. A low spur fractional-N digital PLL for 802.11 a/b/g/n/ac with 0.19 psrms jitter, VLSI Circuits (VLSIC), ISSN: 2158-5601 Print ISBN: 978-1-61284-175-5. Q. Yu*, Y. Liu*, X.P. Yu**, W. M. Lim**, F. Yang*, X. L. Zhang*, and Y. Peng*, 2011. A CMOS Frequency divider for 2.4/5GHz WLAN Applications with a Simplified Structure, JOURNAL OF SEMICONDUCTOR TECHNOLOGY AND SCIENCE, VOL.11, NO.4. Jubayer Jalil, Mamun Bin Ibne Reaz, Mohammad Arif sobhan Bhuiyan, Labonnah Farzana Rahman, and Tae Gyu chang,2014. Designing a RingVCO for RFID Transponders In 0.18µm CMOS Process, The scientific world Journal Volume 2014, Article ID 580385. Farazian, M., et al., 2013. Architectures for Frequency Synthesizers,” 978-1-4614-0490-3_2, © Springer Science and Business Media, New York. Saleh Ali Alomari, Putra Sumari, Alireza Taghizadeh, 2011. A Comprehensive Study of Wireless Communication Technology for the Future Mobile Devices,” European Journal of Scientific Research ISSN 1450-216X Vol 4.

[77]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

A Comparative Study of InSb, InAs and Si based Nanowire MOSFET Rakesh Prasher, Devi Dass, Rakesh Vaid* Department of Physics and Electronics, University of Jammu, Jammu, India

Abstract We have compared Indium Arsenide and Indium Antimonide based nanowire MOSFET with Silicon based nanowire MOSFET (SiNWMOSFET), both gives excellent on and off characteristics than Si-NWMOSFET. The device metrics considered at the nanometer scale are Subthreshold swing (S), Drain Induced Barrier Lowering (DIBL), On and Off current, Carrier Injection Velocity (ʋinj) etc. The results advise us that InSb and InAs channeled Nanowire MOSFET has the highest I on and lowest Ioff values. Besides, InSb NWMOSFET has the highest values for Ion/Ioff ratio, ʋinj, transconductance (gm) and improved short channel effect (S = 59.82 and DIBL = 0.84 both values are very closed to ideal condition). More results such as effect of I ds vs. Vds, Ids vs. Vgs, quantum capacitance vs. Vgs, and gm/Id vs. Vgs for three different NWMOSFETs has been investigated. Results obtained suggest that InSb and InAs nanowire channels appear to be applicable for high performance logic and even low operating power requirements.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Indium Antimonide; Indium Arsenide; Nanowire MOSFET; DIBL Subthreshold Slope; Transconductance

1. Introduction Study of using alternative channel materials for their improved transport properties has considerable interest these days for ballistic nanoscale MOSFETs and NWMOSFETs. New materials in the channel in place of silicon enhance oncurrents, reduce series resistance and delay. III-V materials appear to be very attractive candidates as channel materials for highly scaled n-MOSFETs because of their extremely small transport mass leading to high injection velocity (v inj), low interface trap density (Dit), low inversion charge (Qinv) (Chau et al,2005). However, III-V materials have many significant and fundamental issues, which may prove to be severe bottlenecks to their implementation (Fischetti et al, 1991 ; Laux, 2007). Furthermore, the small direct band gaps of novel channel materials inherently give rise to very large band to band tunneling (BTBT) leakage current compared to Si. Due to the increasing E-fields in the channel and the smaller bandgap in these high mobility materials, the BTBT leakage current can become excessive and can ultimately limit the scalability of high mobility channel materials (Chau et al, 2005; Ju et al, 2005; Datta et al, n.d) Another point to note is that since most permittivity (k), they also suffer from worse short channel effects (SCEs) so that is why there is a need of selecting suitable high µ channel materials. Nanowire FETs are important because they are potentially useful in low-power applications, have high packing densities, maintain high gate sensitivity, are capable of individual or bulk (arrayed) fabrication, are scalable, and have recently been investigated for ballistic carrier transport behavior for potential high-performance electronic devices (Dayeh et al, 2006). While nanowire FETs based on silicon, germanium, gallium nitride, and zinc oxide have all been demonstrated over the past few years, nanowires transistors fabricated using these materials remain complex to fabricate and are still, for the most part, non-scalable. Nanowires can be grown using zinc oxide and ZnO nanowire transistors have been recently reported by Ju et al. (Ju et al, 2005). (Dick et al, n.d), Dick et al. reported good I-V characteristic results for two bottom-gate ZnO nanowire devices. By comparison, indium arsenide (InAs) and indium antimonide (InSb) both are attractive semiconducting materials because they have narrow bandgap (0.35 eV and 0.17 eV resp.) and provide extremely high electron mobility’s (can reach up to 33,000 cm2/V-s (Streetman, 2005) and 77,000 cm2/V-s resp. in bulk material as compared to the electron mobility of silicon at 1350 cm2/V-s (Streetman, 2005)) as mentioned in table 1. InAs based FETs are very promising because they offer higher device speed and extremely low power operation, both key factors when designing nanoelectronic circuits. Planar InAs based FETs have recently been demonstrated with very attractive performance (Thelander et al, 2006). * Corresponding author. Tel.: +91-9419106794. Email: [email protected] ISBN: 978-93-82288-63-3

Rakesh et al/COMMUNE – 2015

Further, InAs is used for batch fabrication of multiple devices due to its ease of fabrication and scalability. Perhaps the most significant advantage, in addition to its electrical characteristics, however, is InAs’s capability for forming heterostructures, which enables the formation of high electron mobility transistors (HEMTs). This is significant because through extremely high-quality epitaxial growth, independent of lattice constants (i.e. “strain relaxation”), we can customize the FET according to our specified design (Radosavlievic et al, 2003). This leads to the possibility of fabrication of heterojunction nanowire FETs. In addition to its attractive material properties, there has been success demonstrated in the growth on InAs and InSb nanowires and their integration in field effect transistors (Hang et al, 2006; Bryllert et al, 2006). In this paper, we have made a comparative study by using different channeled nanowire like InAs and InSb in place of silicon. Various simulation results obtained suggest that InAs and InSb have many advantages thus making it suitable for use as channel material in future nanoscale MOSFETs and NWMOSFETs. Table 1: Properties of channel materials used Material properties Effective Mass Electron mobility(µs) Cm2/V-s Band Gap (eV) Lattice Constant(A0)

Si 0.19 1350 1.12 5.43

InAs 0.028 33000 0.35 6.05

InSb 0.014 77000 0.17 6.47

2. Simulation Details In order to examine the ballistic transport properties of a InAs/InSb channeled NWMOSFET, the simulation and modeling in this paper was achieved through FETToy 2.0 is a numerical simulator which uses a set of Matlab scripts to calculate ballistic I-V characteristics for conventional single- and double-gate geometry MOSFETs, nanowire MOSFETs, and carbon nanotube MOSFETs based on the Natori (or “top-of-the-barrier”) approach (Gilbert et al, 2006) The simulated InSb/InAs-NWMOSFET is shown in Fig.1 and various parameter used in our simulation are shown in table 2. Table 2: Input parameter used for different channel materials S. No. 1 2 3 4 5

Input Parameters DNanowire kins T Vth αG αD

Value 1nm 3.9 300 K 0.32 eV 1.00 0

Fig. 1 Simulated NWMOSFET

3. Results and Discussion 3.1

Ids – Vds Characteristics

Fig. 2(a) show Ids-Vds plots at constant Vgs = 1V and dNW =1nm for different NWMOSFETs. Exact saturation occurs around 0.25 volts to 0.5 volts for all the NWMOSFETs. Drain current for InAs and InSb-NWMOSFETs get saturated almost at of 0.5 volts and have higher saturation current as compared to Si-NWMOSFET. It has been observed that the saturation current increases with the decrease in effective masses in the channel (for InAs/InSb nanowire). This means that the increment in saturation current occurs as we are going for less effective mass channel material device.

[79]

Rakesh et al/ COMMUNE-2015

3.2

Ids – Vgs Characteristics

Fig. 2(b) shows the Ids versus Vgs curves at constant Vgs of 1V and dNW of 1nm for different NWMOSFETs. InAs and InSb shows higher drain current but it requires lower threshold voltage, thus it is not possible to suppress subthreshold effects and quantum confinement cannot be achievable. It means that when we reduce the effective mass of material in the channel (change the channel materials), the current capability of the NWMOSFETs enhances. This fig. shows that InSb nanowire gives maximum drain current when used in channel of MOSFET. 3.3

Quantum capacitance w. r. t. gate voltage

Fig. 3(a) shows that all three NWMOSFETs have similar quantum capacitance verses gate voltage behavior. The device can be operated at quantum capacitance limit when its gate capacitance is considerably higher than quantum capacitance. To know device operation at QCL limit, value of quantum capacitance at inversion, depletion and accumulation regions, and the study of Qc – Vg curves are drawn. For InAs/InSb-NWMOSFETs gives higher quantum capacitance than the SI-NWMOSFETs. InAs and InSb have well defined accumulation and inversion regions with higher threshold voltage due its higher gate capacitance and quantum capacitance. 3.4

gm/Id w. r. t. gate voltage

Fig. 3(b) shows (gm/Id) variations w. r. t. gate voltage at constant drain voltage (V ds = 1V) and dNW of 1nm for different NWMOSFETs. In this graph, we can see that as the V gs increases the gm/Id decreases, in other words, the transconductance of the device (gm) decreases for the current polarization by governing the equation, G m = Id/Vgs. As we know that the maximum performance obtained when the value of gm/Id ratio is the largest. The larger the transconductance, the greater the control over the channel in MOSFET. InSb and InAs nanowire shows higher transconductance efficiency than Si-nanowire as channeled in NWMOSFET.

(a)

(b)

Fig. 2- (a) Ids – Vds curve (b) Ids – Vgs curve for different NWMOSFETs.

3.5

Short Channel Effect in NWMOSFETs

Two short channel effects DIBL and Subthreshold Swing (SS) has been discussed and analyzed. DIBL effect occurs when the barrier height for channel carriers at the edge of the source reduces due to the influence of drain electric field, upon application of a high drain voltage. This increases the number of carriers injected into the channel from the source leading to an increased drain offcurrent. Thus, the drain current is controlled not only by the gate voltage, but also by the drain voltage. A device characterized by steep subthreshold slope exhibits a faster transition between off and on states In fig. 4(a) it is shown that InSb-NWMOSFETs have less DIBL than other NWMOSFET. Fig. 4(b) shows that SS for InSb-NWMOSFET (59.82 mV/dec) approaches to ideal value (i.e. 60 mV/dec).

[80]

Rakesh et al/COMMUNE – 2015

(a)

(b)

Fig . 3: (a) Quantum Capacitance – Vgs curve (b) gm/Id – Vgs curve for different NWMOSFETs.

(a)

(b)

Fig.4- (a) DIBL (b) Subthreshold Slope (SS) for different MOSFETs

4. Conclusion On the basis of various results obtained using simulation approach, it is concluded that InSb-NWMOSFET has highest Ion and lowest Ioff as compared to SI-NWMOSFET. It suggests that InSb has highest transconductance (gm) as compared to other two nanowires based channel. Moreover, the carrier injection velocity of InSb and InAs is maximum as compared to Si. It means, the mobility of charge carrier in both these materials is higher due to which they give higher Ion. Finally, InSb-NWMOSFET has threshold swing of 59.82 mV/dec which is close to the ideal value and DIBL of 0.84 mV/V. Hence, the InSb- NWMOSFET possesses higher drive current, maximum transconductance, maximum quantum capacitance, maximum average velocity, and lower short channel effect than the other channel MOSFETs. In future, it may be used in the manufacturing of the MOSFET devices due to several advantages as quoted above. References Chau, R. et al., 2005. Bench marking Nano Technology for High Performance and Low Power Logic Transistor Applications, IEEE Transactions on Nanotechnology, 4, 153. Fischetti, M. et al., 1991. IEEE Transactions on Electron Devices, 38, 650. Laux, S., 2007. A Simulation Study of the Switching Times of of 22nm and 17nm Gâte Length SOI NFETS on High mobility Substrates and SI, IEEE Transactions on Electron Devices, 54, 2304. Datta, S., Dewey, G., Doczy, M., et al.,n.d. High Mobility Si/SiGe Strained Channel MOS Transistors with HfO2/Tin Gate Stack, IEEE IEDM, 653 Krishnamohan, T., Krivokapic, Z., Uchida, K., Nishi, Y. and Saraswat, K. C., n.d. High Mobility Ultra Thin Strained Ge MOSFETS, on Bulk and SOI with Low Band to Band Tuneling Leakage : Experiments IEEE Transactions on Electron Devices, 53, 990. Dayeh, S. A., Aplin, D. P. R., Zhou, X., Yu, P. K. L., Yu, E. T. and Wang, D., 2006. Small Nano Micro Journal. Ju, S., Lee, K. and Janes, D. B., 2005. Low Operating Voltage Single ZnO Nano Wire FET enabled by Self Assembled Organic gate Nano Dielectrics, Nano Letters, 5, 2281. Dick, K. A., Deppert, K., Samuelson, L. and Seifert, W., n.d. Epetaxial growths and Design of Nano wires, and Complex Nano structures, Journal of Crystal Growth, 297, 326. Streetman, B. G. and Banerjee, S. K., 2005. Solid State Electronic Devices, 540. Thelander, C. et al., 2006. Nano wire based 1D Electronics, Materials Today, 9, 28. Radosavlievic, M., Heinze, S., Tersoff, J. and Avouris, 2003. Drain Voltage Scaling in Carbon Nanotube Transistors, Applied Physics Letters, 83, 2435, Hang, Q., Janes, D. B., Wang, F. and Buhro, W. E., 2006. Sixth IEEE Conference on Nanotechnology, 2, 422. Bryllert, T., Wernersson, L. E., Löwgren, T. and Samuelson, L., 2006. Nanotechnology, 17, S227. Gilbert, M. J. and Ferry, D. K., 2006. Journal of Physics: Conference Series, 38, 134.

[81]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Optimizing FPGA based Fixed-Point Multiplier using Embedded Primitive and Macro-support Burhan Khurshid*, Roohie Naaz Mir Department of Computer Science Engineering, National Institute of Technology, Srinagar, India

Abstract The multiplier circuit is an important component in digital signal processing. The increasing cost of silicon technology has put a lot of pressure on developing dedicated SoC solutions for DSP systems and has typically cornered such solutions to high volume productions only. Recently, FPGAs have been used as an alternative platform for DSP systems as they have the ability to develop suitable circuit architecture in a way similar to SoC systems. Whilst the prefabricated aspects of FPGAs avoid many architectural problems, the ability to create an efficient implementation from a DSP system description remains a highly convoluted problem. This has prompted different FPGA vendors to improve the capacity and flexibility of the underlying fabric by including specialized primitive, macro, and IP support. In this paper, we carry out the implementation of fixed-point multiplier circuit by using these primitives and macro support. Our implementation results indicate reduction in resource usage by at least 45%, increase in speed by at least 35% and reduction in dynamic power dissipation by at least 40% when compared to the implementation utilizing the general FPGA fabric.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords - Fixed Point Arithmetic; ASIC; DSP; FPGA; FPGA Primitives; FPGA Macros

1. Introduction Multiplier circuit is one of the most fundamental components used in various digital signal processing (DSP) applications like convolution, filtering, compression etc.(Narayan et al., 2005, Ashour et al., 2000, Compton et al., 2002). The field of DSP has always been driven by the advancements in scaled very-large-scale-integration (VLSI) technologies. The goal of digital design is to maximize the performance while keeping the cost down (Parhi, 1999). In the context of general digital design, performance is measured in terms of the amount of hardware circuitry and resources required; the speed of execution (throughput and clock rate); and the amount of power dissipated. Depending on the application, there is always a tradeoff between these parameters. This demands for high speed, low area and low power realization of circuits used in these DSP systems (Shanthala et al., 2009). Traditional DSP implementations have focused mainly on processor-oriented solutions where the design process mainly consists of developing the necessary high-level code with some thought given to the underlying hardware (Woods et al., 2008). For high-speed applications some platform-based solutions such as application specific integrated circuits (ASIC) and structural ASICs have been used (Ashour et al., 2000). Recently field programmable gate arrays (FPGAs) have proven to be favored platform for VLSI design engineers. FPGAs offer many advantages over ASIC and programmable systems. The high speed and low power advantage of FPGAs over microprocessors is a sustainable trend for a wide variety of applications (Guo et al., 2004, Stitt et al., 2004). Some other advantages include design modifications postproduction, lower non-recurring engineering (NRE) costs, re-configurable design approach etc. (Tessier et al., 2001), (Todman et al., 2005). There has been subsequent work regarding implementation of multipliers on FPGAs (Shand et al., 1991, Louca et al., 1996). These mainly focus on modifying the multiplier architecture to achieve performance improvement. However, there has been very limited effort in improving the performance by using embedded FPGA resources (Zhou et al., 2008, Gao et al., 2009, Ingemarsson et al., 2012) In this paper, we carry out the implementation of fixed-point multiplier *

Corresponding author. Tel.: +91 9797 875263 E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Khurshid and Mir/COMMUNE – 2015

circuit by using the embedded primitives and macro support that are inherent in modern day FPGAs. This requires detailed information about the FPGA target family that is being used and the primitives that are supported. In our study, different primitives have been used and system functionality has been distributed in a way that utilizes these components efficiently. The study focuses on Spartan-6 FPGA family. Our implementation results indicate reduction in resource usage by at least 45%, increase in speed by at least 35% and reduction in dynamic power dissipation by at least 40% when compared to the implementation utilizing the general FPGA fabric. The rest of the paper is as follows. Section II briefly discusses the fixed-point multiplication. Section III lists the primitives that have been used in this work. A brief description about each primitive is provided. Section IV carries out the synthesis and implementation. Conclusions are drawn in section V and references are listed at last. 2. Bit-parallel multiplication Bit-parallel multipliers process one whole word of the input sample each clock cycle and are ideal for high-speed applications. In this paper, we consider the traditional parallel ripple-carry array multiplier. The operands are assumed to be represented in fixed-point 2’s complement representation. Therefore, an N-bit number Y is represented as: 𝒀 = 𝒚𝑵−𝟏 . 𝒚𝑵−𝟐 𝒚𝑵−𝟑 … … 𝒚𝟏 𝒚𝟎

(1)

The most significant bit yN−1 is the sign bit with 0 denoting a positive number and 1 a negative number. The value of this number is given by: −𝒊 𝒀 = −𝒚𝑵−𝟏 + ∑𝑵−𝟏 𝒊=𝟏 𝒚𝑵−𝟏−𝒊 𝟐

(2)

With 2’s complement, representation a correct result is guaranteed even if there is an intermediate overflow (Parhi, 1999). In parallel ripple-carry array multiplier, the carry is rippled to the adder to the left in the same row (Parhi, 1999). Thus, within a row each adder has to wait for the carry input to perform its computation. In other words, there exists an intra-iteration constraint between any two adjacent adder nodes within a row, assuming there is no pipelining involved. Owing to this ripple-carry nature, the critical paths involved are quite large which limits the speed of multiplication. 3. FPGA Primitives Primitives are the components that make an FPGA. The exact nature of a primitive may vary from family to family. In this section, we briefly describe the primitives that have been used in this study. These belong to the Spartan-6 family. 3.1. BUFG (Xilinx) This design element is a high-fan-out global clock buffer that connects signals to the global routing resources for low skew distribution of the signal. BUFGs are typically used on clock nets as well other high fan-out nets like sets/resets and clock enables. 3.2. FDSE (Xilinx) FDSE is a single D-type flip-flop with clock enable and synchronous set. The synchronous set input, when high, overrides the clock enable input and sets the output high during the low-to-high clock transition. The data is loaded into the flip-flop when set is low and clock enable is high during the low-to-high clock transition. 3.3. CARRY4 (Xilinx) This primitive represents the fast carry logic for a slice. The carry chain consists of a series of four multiplexers and four XOR gates that connect to the other LUTs in the slice via dedicated routes to form functions that are more complex. The fast carry logic is useful for building arithmetic functions like adders, counters, subtractors etc. 3.4. DSP48 (Xilinx) This design element is a versatile, scalable, hard IP block that allows for the creation of compact, high-speed, [83]

Khurshid and Mir/COMMUNE – 2015

arithmetic-intensive operations, such as those seen for many DSP algorithms. Some of the functions capable within the block include multiplication, addition, subtraction, accumulation, shifting, logical operations, and pattern detection. 4. Synthesis and Implementation 4.1. Methodology The implementation in this work targets Spartan-6 FPGA family. Only LX series has been considered as it is apt for general logic applications. The implementation is carried out for an input operand length varying from 4 to 32 bits. The parameters considered are resource utilization, timing and dynamic power dissipation. The design synthesis, mapping and translation are carried out in Xilinx ISE 12.1 and the simulator database is then analyzed for on-chip resources, throughput and timing metrics. Power metrics are obtained using Xpower analyzer. 4.2. Experimental Results As mentioned earlier, synthesis based on the utilization of general FPGA fabric will serve as a standard against which other implementations will be compared. Metrics associated with other implementations are named as per the primitive/macro used. Table 1 gives a comparison of the on chip resources utilized by different designs for an input word-length of 16 bits. The target device is XC6SLX16 from Spartan-6. It is observed that by using various primitives and macro blocks there is a subsequent reduction in the on-chip resources being utilized. This is achieved without having to modify any architectural details. The CARRY4 and DSP48 primitives provide fast carry logic for each row. Their inclusion prominently will affect the timing properties of the structure. However, there is still some reduction in the slices being utilized when compared to the basic structure generated using general FPGA fabric. Further analysis is carried out for varying wordlengths. Table 1. Resource utilization on spartan-6 FPGA On-chip resource No. of Slice Registers No. of slice LUTs No. of occupied slices

General fabric 32 943 361

CARRY4 32 385 143

DSP48 32 168 133

The metrics obtained from the synthesizer database are then plotted as a function of operand word-lengths and are presented in figure 1 and 2. RESOURCE UTILIZATION 500

GENERAL FABRIC

RESOURCE UTILIZATION

GENERAL FABRIC

CARRY4

2000

CARRY4

DSP48

1800

DSP48

1600

1400

400

NO. OF LUTs

NO. OF OCCUPIED SLICES

600

300 200

1200 1000

800 600 400

100

200

0

0

4

8

16

32

4

Fig. 1. Occupied slices for different implementations

8

16

32

WORD-LENGTH

WORD-LENGTH

Fig. 2. LUT utilization for different implementations

The fast carry logic associated with the CARRY4 and DSP48 primitives makes the addition process really fast resulting in reduced route delays. Table 2 provides a comparison of the maximum achievable clock rates post implementation for a word length of 16 bits. The target family is Spartan-6. The structures generated using different primitives tend to have better timing closures in terms of the relationship between an external clock pad and its associated data-in or data-out pad. This is indicated by the offset-in and offset-out metrics from the timing database of the synthesizer. The values are included in the tables and are indicative of the fact that with primitives a better timing behavior is achieved.

[84]

Khurshid and Mir/COMMUNE – 2015

Table 2. Timing analysis on spartan-6 FPGA Timing Parameter Maximum frequency (MHz) Minimum available offset-in (ns) Minimum available offset-out (ns)

General fabric 30.67 5.112 11.727

CARRY4 45.7 2.118 6.782

DSP48 144.38 2.543 2.765

Further analysis is carried out by plotting the maximum achievable speed against the operand word lengths for different structures. The results are shown in figures 3. Finally dynamic power dissipation for different structures is considered. The analysis was done for a constant supply voltage and at maximum operating frequency for each structure. To ensure a reasonable comparison the test vectors provided during post route simulation were selected to represent the worst case scenario for data coming into the multiplier block. Same test bench was used for all the synthesized structures. The design node activity from the simulator database along with the power constraint file (PCF) was used for power analysis in the Xpower analyzer tool. Table 3 shows the power dissipated in various resources for operand length of 16 bits. The target device is Spartan-6. MAX CLOCK FREQUENCY (MHz)

350

SPEED/THROUGHPUT

300

GENERAL FABRIC CARRY4

250

DSP48

200 150

100 50 0 4

8

16

32

WORD-LENGTH

Fig. 3. Throughput comparison for different implementations Table 3. Power dissipation on spartan-6 FPGA Power dissipated (mW) General fabric CARRY4 0.54 0.64 2.34 1.4 4.41 1.16 6.01 4.63 13.3 7.83

FPGA resource Clock Logic Signals I/Os Total

DSP48 2.8 0.87 1.23 2.54 7.44

DYNAMIC POWER DISSIPATED (mW)

Further analysis is carried out by plotting the total dynamic power dissipation as a function of input word-length. The result is shown in figure 4. 60

INFERENTIAL_CODING

POWER DISSIPATION

CARRY4

50

DSP48

40 30 20

10 0 4

8

16

WORD-LENGTH

Fig. 4. Dynamic power dissipation for different implementations [85]

32

Khurshid and Mir/COMMUNE – 2015

5. Conclusions This paper implemented the bit-parallel fixed-point multiplier using different primitives and macros available in modern day FPGAs. The analysis and the experimental results carried out in this paper clearly indicate that a considerable improvement in performance is indeed achievable by using these primitives. A judicious choice of primitives will ensure that a particular performance parameter is enhanced as may be required by any particular application. This paper deliberately ruled out any architectural modification that may be carried out at the top level of the design. The idea was to present a clear-cut analysis that will provide an insight about the performance speed-up that may be achieved by utilizing the huge primitive support provided by FPGA families. References Narayan, G. L. and Venkataramani, B., 2005. Optimization Techniques for FPGA based Wave Pipelined DSP Blocks, IEEE Transc.Very Large Scale Integr. (VLSI) syst., vol. 13, No. 7, pp. 783-792. Ashour, M. A. and Saleh, H. I., 2000. An FPGA Implementation guide for some different types of Serial-Parallel Multiplier Structures, Microelectronics Journal, vol. 31, pp. 161-168. Compton, K. and Hauck, S., 2002. Reconfigurable Computing: A survey of Systems and Software, ACM Computing Surveys, vol. 34, No. 2, pp. 171210. Parhi, K. K., 1999. VLSI Digital Signal Processing Systems Design and Implementation, Wiley. Shanthala, S. and Kulkarni, S. Y., 2009. VLSI Design and Implementation of Low Power MAC Unit with Block Enabling Technique, European Journal of Scientific Research, ISSN 1450-216X, vol. 30, No. 4, pp. 620-630. Woods, R., McAllister, J., Lightbody, G. and Yi, Y., 2008 FPGA-based Implementation of Signal Processing Systems, Wiley. Guo, Z., Najjar, W., Vahid, F. and Vissers, K., 2004. A Quantitative Analysis of the Speed up Factors of FPGAs over Processors, in Proc. Int. Symp. on FPGAs, ACM Press,. Stitt, G., Vahid, F. and Nematbakhsh, S., 2004. Energy Savings and Speed ups from Partitioning Critical Software Loops to Hardware in Embedded systems, ACM Transc. Embedded Comput. Systems, vol. 3, pp. 218-232. Tessier, R. and Burleson, W., 2001. Reconfigurable Computing for DSP: A Survey, Journal of VLSI Signal Processing, vol. 28, pp. 7-27, Kluwer Academic Publisher. Todman, T. J., Constantinides, G. A., Wilton, S. J. E., Mencer, O., Luk, W. and Cheung P. Y. K., 2005. Reconfigurable Computing: Architecture and Design Methods, in IEEE Proc. Comput. Digit. Tech., vol. 152, No. 2. Shand, M., Bertin, P. and Vuillemin, J., 1991. Hardware Speedups in Long Integer Multiplication, Computer Architecture News, vol. 19, No. 1, pp. 106–114. Louca, L., Cook, T. A. and Johnson, W. H., 1996. Implementation of IEEE Single Precision Floating Point Addition and Multiplication on FPGAs, in ACM/SIGDA International Symposium on Field Programmable Gate Arrays, Monterey, CA, pp. 107–116. Zhou, G., Li, L. and Michalik, H., 2008. Area optimization of bit parallel finite field multipliers with fast carry logic on FPGAS, International Conference on Field Programmable Logic and Applications. Gao, S., Khalili, D. A. and Chabini, N., 2009. Efficient Scheme for Implementing Large Size Signed Multipliers Using Multigranular Embedded DSP Blocks in FPGAs, International Journal of Reconfigurable Computing Vol. Article ID 145130, Hindawi Publishing Corporation. Ingemarsson, C., Kallstrom, P. and Gustafsson, O., 2012. Using DSP block pre-adders in pipeline SDF FFT implementations in contemporary FPGAs, 22nd International Conference on Field Programmable Logic and Applications.http://www.xilinx.c

[86]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Information Diffusion Modelling and Social Network Parameters (A Survey) Mudasir Wani*, Manzoor Ahmad Department of Computer Sciences, University of Kashmir, Srinagar, India

Abstract Online Social Networks (OSNs) encompasses various facets like influence, communities, tie-strength, homophily, etc. These attributes have been a keen area of research and exhibit an individual as well as collective behavior. In this article, we enlist and summarize some of these facets and explore their involvement as contributors and byproducts in the process of information diffusion process. This survey will help researchers to get an overview of the existing information diffusion models along different facets of OSNs.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Information Diffusion; Online Social Networks; Influence; Tie Strength, Homophily; Communities; Opinion; User Roles; SpatioTemporal Dynamics; Topics and External Influence

1. Introduction OSNs are changing the view of the communication world, as we know it. Every sect of the society is being influenced by the trending information flow on these OSNs either directly or indirectly. Although these emerging OSNs have shown tremendous growth in terms of speed, scale, and reach, but at the same time, they have also given rise to a new paradox of information flow and control. Social networks are growing exponentially with a highly dynamic nature and very complex structure (Wani, et al, 2013). Researchers, scientists, businesspersons, intelligence agencies, governments, and marketers are interested in understanding the dynamics of information flow over these OSNs. Many Information diffusion models have been proposed, modelled, and analyzed. In this paper, we enlist some facets of OSNs viz influence, tie-strength, homophily, communities, opinion, user roles, spatio-temporal dynamics, topics and external influence. We study how these parameters at times act as a decisive construct and at times act in parallel with the process of information diffusion. The rest of the paper is organized as follows. Section 2 presents a brief overview of the information diffusion model along with selected facets of OSNs. Section 3 presents a brief description of the different models along respective facets of OSNs. Finally, section 4 concludes the paper with the overall effectiveness of discussed OSN facets in the process of information diffusion. 2. Information Diffusion Model A typical information diffusion model comprises of i) information and ii) an online social network. We propose a model under same lines and add the factor of external influence as shown in figure 1. Although the term information is an abstraction of a vast concept, we consider two parameters namely i) information pertaining to different topics and ii) interfering information i.e. competing and cooperating information. As we see in sections 3.1.1 and 3.1.2, these have a notable impact on the process of diffusion. The online social network has users as its nodes. Subsection 3.2 and 3.3 discuss users and network respectively. Finally, we discuss the external influence on the overall diffusion model in subsection 3.4.

* Corresponding Author. Tel.: +91 9622 510781. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Wani and Ahmad/COMMUNE – 2015

INFORMATION DIFFUSION MODEL INFORMATION Different Topics Competing & Cooperating Information

OSN USER Influence

EXTERNAL INFLUENCE

Social Role Opinion NETWORK Tie Strength Homophily Spatio Temporal Community

Fig. 1: The Information Diffusion Model with different constructs.

A decent amount of research has been done in the field of information diffusion in recent years, but it usually focuses on a single or similar parameter of social network. Here, we study and isolate a bunch of related parameters and categorize them to help researchers better understand the existing models. On analysis, we found that many properties of information diffusion process remain unnoticed because of its complexities. In this paper, we bring together some existing individual models to uncover these complex properties. The purpose of this paper is to create a knowledge base for future researchers. We try to add a new perspective to already defined models. 3. Facets of Online Social Networks 3.1

Information

The spread of information is creating new nervous system for our planet (Wani and Ahmad, 2014). Information serves as the central entity in the process of communication and its diffusion over a social network. We consider two aspects while dealing with information viz. different topics and interfering information. With huge OSNs, both diversity and interference has become ubiquitous in information. 3.1.1

Across Different Topics

Behavior of different information diffusion models vary over an OSN. Analyzing this behavior helps in understanding the dynamics of information diffusion (Saito, et al, 2010). They use two models Asynchronous Independent Cascade (AsIC) and Asynchronous Linear Threshold (AsLT), and after learning the model parameters, they identify the correspondence between the topic and the model. The authors conclude with the fact that model selection has a very critical effect on its applicability on the topic; however, the range and speed are not affected by the selected model. Information diffusion process is very dynamic and adapts different pathways for different topics. Information belonging to different topics may diffuse through different paths over the same network. (Romero, et al, 2011) identify two attributes for different topics of information under consideration viz persistence and stickiness. On analysis, they found politically controversial topics to be more persistent i.e. repeated exposure to controversial or contentious information validating complex-contagion process. 3.1.2

Cooperating and Competing Information

The flow of information is not a mutually exclusive process, rather there exist simultaneous multiple information flows in an OSN. These simultaneous multiple information streams interfere with each other either constructively or destructively. A constructive interference resulting into cooperating information streams helps in the process of diffusion and serves as a facilitator. While as a destructive interference resulting into competing information streams, inhibit each other’s prorogation through common network pathways. Multiple information propagate over a network at the same time and interact with each other (Myers, 2012). The authors developed a statistical model showing that cooperating information increased each other’s probability of spread, while as competing information reduced spread probability. Their model was able to learn over these interactions and predict contagion infections with a 400% improvement over any of the standard baseline models. On average, these interactions showed a 71% change in the diffusion probabilities.

[88]

Wani and Ahmad/COMMUNE – 2015

3.2

User

A user acts as both generator and acceptor of the information over an OSN. Various user attributes: personal, social, and behavioral, shape the paths of information diffusion. Many researchers have worked on these user attributes along the lines of information diffusion. We analyze and discuss them in the following subsections. 3.3

Influence

The emotional, behavioral, or reactional effect of a user over any other user in an OSN is quantified as influence. This parameter is very closely related to the process of information diffusion, as it determines the paths of information dissemination over a network. Only when a user x is influenced by user y, the information can flow from x to y, depicted as x  y. In their review, (Shidore, 2014) highlight two key concepts namely influence and relevance in the process of information diffusion. The influencers populate the community with relevant information, thereby satisfying its needs. Their study shows that the individual behavior data can be used to identify the influence relationships present in the community. The various extensions to this attribute are as: 3.3.1

Tracking Influence

In an OSN, there is a big network of users with exposures to new information from multiple sources. Under such circumstances, in order to keep track of the information diffusion process, it is important to know who-influencedwhom (Bakshy et al., 2012). This process of identifying the influence sources is termed as tracking influence. Contrary to it, the authors in (Du & Liang, 2014) argue that a diffusion model is not a prerequisite for learning influence diffusion, instead they propose an efficient maximum likelihood based algorithm to learn influence directly from the cascade data without specifying any diffusion model. 3.3.2

Influential User or Spreaders

A user x is said to be an influential user if there exists a one-to-many relationship i.e. a user x influences many users in the current context. Influential users play an important role in the process of information diffusion as they facilitate the process by easily propagating the information to multiple recipients with a higher probability of its acceptance (Kimura, et al, 2007). Influential spreaders are usually involved in promoting or facilitating the process of information spread (Wang, et al 2013). Identifying the influential spreaders is important to make efficient use of a network for optimal information diffusion (Kitsak, et al, 2010). The authors argue that the most influential spreader does not correspond to the highly connected user(s), instead they identify the influential spreader(s) as the one(s) present at the core of the network as identified by the k-shell decomposition analysis. The distance between any two influential spreaders has a prominent effect on the range of information spread. They provide an optimal design for maximizing information diffusion. 3.3.3

Influence Maximization

The process of increasing or maximizing the users affected by any single user x or a set of users X in order to increase the reach of information is termed as influence maximization. In (Lin, Hu, et al 2014) propose a push-driven cascade (PDC) model for dynamic influence maximization, which uses the push mechanism to maximize influence which in turn structures the information diffusion pathways. (Kempe, et al, 2003) formulate a framework for performance guarantees of algorithms for influence maximization. 3.3.3.1 Persuasion It is the process of persuading someone to adopt to a particular information and propagate the same further into the network. While influence mechanism works on the principal of pull, where the recipient pulls the information from the source, persuasion follows the principle of push, where the source pushes the information towards the recipient. Analyzing the effect of persuasion on the recipients opinion towards information acceptance and diffusion, (Das, 2014) gave a structural model of information diffusion proving the fact that the larger-scale behaviors are consistent with the traditional information cascade models and succeeded in modelling the variance in the structural virility of these cascades. 3.3.4

Users Social Role and Human Activity Patterns

The role of a user in an OSN varies under different circumstances. A same user may serve as information inhibitor or facilitator under different roles. By role, we mean the sum total of the users’ social attributes, which determines his position and part in a social network. Using historical diffusion data for a Gibbs-sampling based algorithm (Yang, et al, 2014) propose a Role-Aware INformation diffusion model (RAIN) that integrates social role recognition and diffusion

[89]

Wani and Ahmad/COMMUNE – 2015

modeling into a unified framework. Based on the roles of the users, their model predicts whether or not the user will repost the information that it is exposed to and the scale and duration of the overall diffusion process. 3.3.5

Opinion Diffusion

Opinion refers to the ideas and perspective of a user in an OSN. The diffusion of information implicitly encompasses within it different opinions. Therefore, opinion diffusion serves as a variant to traditional information diffusion process. In (Li, et al, 2012) the authors analyze a hard-interaction model with an opinion threshold, beyond which a user in the network does not interact with its neighbor. This model investigates how convergence properties of opinion dynamics and characterize the phase transition from a society of radicalized opinions to one of convergent behavior over a network. 3.4

Network

As the information has to eventually diffuse through the network, the dynamics and properties of underlying network are important to consider while modelling the information diffusion process. We identify some network attributes and discuss their effect on the process of information diffusion in the following subsections. 3.4.1

Tie Strength and Connectivity

Tie strength implies the authority over a link between two users. Tie strength together with high connectivity plays an important role in the process of information diffusion. A strong tie-strength between two users results into the information adoption. Depending on the tie-direction i.e. unidirectional or bidirectional, tie strength facilitates the process of information diffusion either in one or both directions respectively. Connectivity refers to the frequency of connections i.e. a highly connected network is the one with a large number of connections\links in comparison to the number of users or nodes. (Kossinets, et al, 2008) define the network back-bone to be the subgraph consisting of edges on which information has the potential to flow the quickest, thereby providing insights into the relationship between tie strength and connectivity in OSNs. 3.4.2

Homophily

Similarity between the users of an OSN is termed as homophily. Similar users tend to adopt information and behavior from each other and hence show a prominent effect on the process of information diffusion. Information propagates differently among the hemophilic users (Choudhury, et al, 2010). The authors combine the effect of homophily and external temporal variables to predict the process of information diffusion. They conclude that information diffusion is significantly affected by homophily. (Halberstam and Knight, 2014) use political communications to analyze the effect of homophily over it. They find that, comparatively with homophily, members of the majority group have more connections and information reaches like-minded users more quickly i.e. hemophilic users are exposed to more information. 3.4.3

Spatio Temporal Dynamics

The dynamics and behavior of network over space and time is referred to as spatio temporal dynamics. Temporal heterogeneity has a diverse effect over the information spreading process, which in turn results into temporal sparsity controlling the slowdown of susceptible-infection spreading process (Perotti, Jo, and Holme, 2014). In their study, (Kossinets et al., 2008) devise a temporal measure namely distance between two users as the minimum time required for information to flow between them. Such temporal measures provide an entirely new perspective and insight into the network that remain hidden otherwise. They identify a sparse sub-graph with highly embedded and long-range edges (acting as bridges) over which the information diffusion would be shortest i.e. quickest in terms of time. (Wang et al., 2013) propose a simpler linear partial differential equation that takes into account both the spatial population density and temporal decay of user interests for the process of information diffusion and identifying influential spreaders. 3.4.4

Community Structure

Communities arise from homophily, where similar users form clusters of tightly connected sub graphs with a definite behavior and characteristics. The structure of a particular community has an impact on the process of information diffusion. Presence of communities in a network can facilitate overall diffusion by enhancing both local and intracommunity spreading (Nematzadeh, et al, 2014). The authors demonstrate an optimal network modularity with maximum information spread and minimum number of early adopters.

[90]

Wani and Ahmad/COMMUNE – 2015

3.5

External Influence

Information diffusion over an OSN is not a closed world process (Guille, et al, 2013). Many external world parameters influence the information diffusion process. Apart from the connections within an OSN, information reaches to us from external out-of-network sources (Myers, et al, 2012). The authors present a model by quantifying external influence and its effect over the information adoption. They discover that there are inconsistencies in the diffusion paths over an OSN, and information jumps at certain points, which can be only attributed to the unobservable external influence on the network. Analyzing a Twitter data set, they confirmed that only 71% of information volume attributed to network diffusion, rest 29% corresponded to the events and factors outside the network. 4. Conclusion From this survey, we came to know that many models have been proposed in the area of information diffusion but with huge diversity. Although we were able to group together some models based on the selected attributes, still these models are problem specific rather than being generic. We saw that the influence attribute is the most widely analyzed and used parameter in the information diffusion models. This finding asserts the fact that influence diffuses implicitly with the information. Finally, we considered and discussed some important parameters like external influence to highlight the importance of some usually neglected parameter(s) in the process of information diffusion. The research area in this particular field remains as an open challenge and in future, a more detailed and comprehensive study m be built upon this categorization, which can be further extended to encompass other parameters and related attributes used in the information diffusion modelling process. References Bakshy, E., Rosenn, I., Marlow, C., & Adamic, L., 2012. The role of social networks in information diffusion. Proceedings of the 21st …. Retrieved from http://dl.acm.org/citation.cfm?id=2187907 Das, A., 2014. Effect of Persuasion on Information Diffusion in Social Networks. Retrieved from research.microsoft.com/pubs/217325/persuasion_2014-05-19.pdf De Choudhury, M., Sundaram, H., John, A., Seligmann, D. D., & Kelliher, A., 2010. “Birds of a Feather”: Does User Homophily Impact Information Diffusion in Social Media?, 31. Computers and Society; Physics and Society. Retrieved from http://arxiv.org/abs/1006.1702 Du, N., & Liang, Y., 2014. Influence function learning in information diffusion networks. Retrieved from http://machinelearning.wustl.edu/mlpapers/papers/icml2014c2_du14 Guille, A., Hacid, H., Favre, C., & Zighed, D., 2013. Information diffusion in online social networks: A survey. ACM SIGMOD Record. Retrieved from http://dl.acm.org/citation.cfm?id=2503797 Halberstam, Y., & Knight, B., 2014. Homophily, Group Size, and the Diffusion of Political Information in Social Networks: Evidence from Twitter (No. 20681). National Bureau of Economic Research. doi:10.3386/w20681 Kempe, D., Kleinberg, J., & Tardos, É., 2003. Maximizing the spread of influence through a social network. … of the Ninth ACM SIGKDD International …. Retrieved from http://dl.acm.org/citation.cfm?id=956769 Kimura, M., Saito, K., & Nakano, R., 2007. Extracting influential nodes for information diffusion on a social network. AAAI, 1371–1376. Retrieved from http://www.aaai.org/Papers/AAAI/2007/AAAI07-217.pdf Kitsak, M., Gallos, L., Havlin, S., & Liljeros, F., 2010. Identification of influential spreaders in complex networks. Nature Physics, 6(November), 6– 11. doi:10.1038/NPHYS1746 Kossinets, G., Kleinberg, J., & Watts, D., 2008. The Structure of Information Pathways in a Social Communication Network, 9. Physics and Society; Data Structures and Algorithms; Data Analysis, Statistics and Probability. Retrieved from http://arxiv.org/abs/0806.3201 Li, L., Scaglione, A., & Swami, A., 2012. Phase Transition in Opinion Diffusion in Social Networks. In Acoustics, Speech and Signal Processing (ICASSP), 2012 IEEE International Conference on (pp. 3073–3076). doi:10.1109/ICASSP.2012.6288564 Lin, S., Hu, Q., Wang, F., & Yu, P., 2014. Steering Information Diffusion Dynamically against User Attention Limitation. Cs.uic.edu. Retrieved from http://www.cs.uic.edu/~slin/assets/Lin_ICDM2014.pdf Myers, S. A., 2012. Clash of the Contagions : Cooperation and Competition in Information Diffusion. ICDM, 12, 539–548. Myers, S., Zhu, C., & Leskovec, J., 2012. Information diffusion and external influence in networks. Proceedings of the 18th ACM SIGKDD …. Retrieved from http://dl.acm.org/citation.cfm?id=2339540 Nematzadeh, A., Ferrara, E., Flammini, A., & Ahn, Y., 2014. Optimal Network Modularity for Information Diffusion, 088701(August), 1–5. doi:10.1103/PhysRevLett.113.088701 Perotti, J. I., Jo, H., & Holme, P., 2014. Temporal Network Sparsity and the Slowing Down of Spreading, 1, 1–4. Romero, D., Meeder, B., & Kleinberg, J., 2011. Differences in the mechanics of information diffusion across topics: idioms, political hashtags, and complex contagion on twitter. Proceedings of the 20th …. Retrieved from http://dl.acm.org/citation.cfm?id=1963503 Saito, K., Kimura, M., Ohara, K., & Motoda, H., 2010. Selecting information diffusion models over social networks for behavioral analysis. Machine Learning and …, (Ic). Retrieved from http://link.springer.com/chapter/10.1007/978-3-642-15939-8_12 Shidore, M., 2014. Review On Finding Relevant Content and Influential Users based on Information Diffusion. Ijergs.org, 2(6), 1023–1025. Retrieved from http://ijergs.org/files/documents/REVIEW-143.pdf Wang, F., Wang, H., Xu, K., Wu, J., & Jia, X., 2013. Characterizing Information Diffusion in Online Social Networks with Linear Diffusive Model. 2013 IEEE 33rd International Conference on Distributed Computing Systems, 307–316. doi:10.1109/ICDCS.2013.14 Wani, M., & Ahmad, M., 2014. Survey of Information Diffusion Over Interaction Networks of Twitter. International Journal of Computer Application, 3(4), 310–313. Wani, M., Alrubaian, M. A., & Abulaish, M., 2013. A User-Centric Feature Identification and Modeling Approach to Infer Social Ties Between OSN Users. In Proceedings of International Conference on Information Integration and Web-based Applications & Services (pp. 107:107–107:114). Vienna, Austria: ACM. doi:10.1145/2539150.2539194 Yang, Y., Tang, J., Leung, C., Sun, Y., & Chen, Q., 2014. RAIN: Social Role-Aware Information Diffusion. Retrieved from http://keg.cs.tsinghua.edu.cn/jietang/publications/AAAI15-Yang-Tang-RAIN.pdf

[91]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Performance Analysis of DPI Overhead on Elastic and In-Elastic Network Traffic: A Delay Sensitive Classification and Inspection Algorithm (DSCI) Ashaq Hussain Dara, Zubair Manzoor Shahb* a NIELIT J&K, Rangreth, Srinagar,, India System Soft Inclusion, Lal Chowk, Srinagar, India

b

Abstract The currently used Deep Packet Inspection (DPI) algorithms lack the ability to perform an early classification of delay sensitive traffic due to limitation of performing packet inspection. This paper approaches the problem of classification and inspection of unencrypted delay sensitive traffic i.e. VoIP by introducing an algorithm that combines the widely known AC (Aho Corasick) algorithm with a novel classifier. The task focuses on the innovation of concatenating the hybrid algorithm consisting of a VoIP classifier with the existing optimized DPI algorithm (Aho-Corasick) which was chosen for optimization for its speedy string matching performance. Our classifier sends the packets to the optimized DPI algorithm (Delay Sensitive Classification and Inspection Algorithm [DSCI]) which searches and processes the packets payloads containing the “SIP” string in two different scenarios where G.711 and G.729 codecs have been used.The first scenario was implemented mainly for the training of our (DSCI) Algorithm which analyzes and then processes the captured traffic, while in the second scenario the DSCI algorithm analyzes and then processes the traffic in real time. Results from both the scenarios were compared with an existing DPI algorithm (Boyer Moore Algorithm). It was found that the DSCI algorithm along with the string matching classified and identified the packets. Its processing speed marginally varied which were in some cases slightly slow while in some cases slightly fast which overall proves it as an efficient solution for deep packet inspection.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords; Deep Packet Inspection; Delay Sensitive Classification and Inspection(DSCI) algorithm; G.711 and G.729 codecs; Aho-Corasick Algorithm; Boyer Moore Algorithm

1.

Introduction

Network traffic identification and categorization of all types of application is important for managing a network and its various other tasks which involves prioritization of network traffic, shaping and policing and diagnosing network traffic. These requirements plays important role when an internet service provider needs to block heavy bandwidth consuming traffic like Peer-to-Peer(P2P) to manage its bandwidth and guarantee performance. Tasks like route provisioning, workload characterization and modelling can be achieved by accurately identifying and classifying network traffic (Yang, 2008). In order to accomplish these requirements, network administrators need an efficient tool for managing network traffic, process every single packet, guarantees all the policies to be met without compromising the security. The conventional techniques for classifying the network traffic works on the port numbers or payload inspection which could easily be bypassed by many applications. Since these techniques are in-efficient to new emerging applications like Voice over Internet Protocol (VoIP) and Peer-to-Peer (P2P), studies shows that by employing the machine learning (ML) techniques helps to identify these types of traffic. Machine learning works on different features set, like Inter-Arrival Time, Packet Size etc. to classify traffic (Heywood, 2011). The most adapted approach for identifying the network traffic relies on inspecting the packet payload for strings that resemble to a particular type of application. However, this solution has a drawback as it requires heavy computational power and furthermore this method is ineffective against encrypted traffic, which is generated by those * Corresponding author. Tel.: +91 9622 422793. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Dar and Shah/COMMUNE – 2015

applications which uses Secure Socket Layer (SSL). The use of network traffic classifiers which works on certain metrics like Packet Inter Arrival Time, Packet Size etc. yields the information about the application related to that network traffic flow. Machine Learning (ML) technique for network traffic detection requires training the classifier first, which involve offline payload detection (Giacomazzi, 2008). 2.

Problem Statement

Need for classifying VoIP traffic can be demonstrated with example of proxy server which is acting as an intermediate device between source and the destination. Some of the advantages of having a proxy server is that it enhances security, guarantees the enterprise policies and stores the resources locally in its cache to reduce the external traffic. However these servers increases security by protecting the hosts from various types of attacks but they can also introduce delays and jitter by creating a bottleneck which can affect the Round Trip Time (RTT) of a voice call which in certain cases is 150-200 ms. and if the delay caused by these servers are greater than the said RTT, the call will fail. Thus, in order to maintain the Quality of Service (QoS), identifying and the classifying of the VoIP traffic in a timely manner is critical for its proper working The identification of traffic flow has attracted great attention of the Internet service providers for traffic shaping and government organizations for traffic monitoring through deep packet inspection in past decade. Some of the main problem this paper aims to address is stated as following: a) The voice traffic is usually delay sensitive and if there is a little delay in it then we can experience a lot of problems in terms of quality of service i.e. delay of the traffic, voice distortion. b) The existing traffic classifier focuses on traffic in general rather than being specific to a particular type of traffic. c) The VoIP application uses different techniques to avoid its detection through bypassing the firewalls, changing the port numbers and uses encryption. 3.

Proposed Algorithm

In this section, the requirements for designing the algorithm and its benefits are discussed. The benefits will justify the efforts of trying to classify and identify the VoIP traffic; process them before any other traffic type, in order to reduce the delay caused by a normal Deep Packet Inspection (DPI) algorithm. 3.1.

Principle for the Design of the Algorithm

To validate the main aim of the research it was necessary to make a principle methodology of choosing two different existing algorithms for the DPI, optimize one of them and then flow the voice traffic through it to see whether it creates some delay or not and if there is some delay created, in which case it is to analyse that whether it`s effect is negligible on the quality of the voice traffic or otherwise. After that another existing DPI algorithm was chosen (without being optimized) to compare it with the previous algorithm results.

Fig 1. DSCI Algorithm

[93]

Dar and Shah/COMMUNE – 2015

3.2.

Algorithm Implementation

The main aim of this paper is to introduce an algorithm that will help in the identification and classification of the VoIP traffic, which will improve the VoIP QoS but may slow down other traffic type. In the DSCI algorithm changes were made according to the requirement. For the purpose of this paper a simple VoIP detection and classification algorithm has been designed which will work on certain conditions to detect the voice packets, send them to the string matching unit for further processing. The DSCI algorithm will process the voice packets first (if detected) and then the normal traffic. 3.2.1.

Classification Criteria

In the DSCI algorithm, the classifier plays a vital role in prioritizing the voice traffic. To make the classifier a number of things were considered. In order to detect the voice traffic the traditional approach of port number, header or the payload inspection will not work because certain VoIP applications use port hopping technique in order to circumvent any firewall or IDS and here port number based traffic detection fails. Payload and header inspection fails where these VoIP applications use encryption which makes it impossible for string matching algorithms to find the strings inside the payload or headers (Ling, 2009).

Fig 2. Classification Criteria

The only working way is to check for the real working parameters like the session ID, Inter Arrival Time, packet size etc. These parameters have been used as they can verify any type of traffic even if it's encrypted. 3.2.1.1. The Session ID Parameter The concept of session ID has been used because when a voice call is being setup, one unique session is being given to that call. The session ID is stored nowhere in any voice packet, but can be identified by five things; source/destination IP address, source/destination port number and the protocol used. This information will provide the session ID (Lee, 2009). 3.2.1.2. Inter Arrival Time Parameter The concept of Inter Arrival time is the unique property of the Constant Bit Rate (CBR) voice codec which in the case of G.711 and G.729 is set to 10ms. 3.2.1.3. Packet Size Parameter Next is the packet size, these voice codecs always generates constant size packet. The G.711 packet size is: Voice payload =160 bytes for 20 ms of voice (default) and 240 bytes for 30 ms of voice, 40 bytes of header information which include 20 bytes for IP, 8 bytes for UDP, 12 bytes for RTP and 14 bytes for Ethernet frame, making the packet size of total 214 bytes. In G.729 the packet size is: Voice payload=20 bytes for 20 ms of voice(default) and 30 bytes for 30 ms, 40 bytes of header information and 14 bytes of Ethernet frame, making the packet size total of 74 bytes (Voice Over IP - Per Call Bandwidth Consumption, 2013). This criterion satisfies the information required to identify the voice packets because no other traffic resembles these features. The uniqueness of voice packets are there size and inter arrival time which is generated by the codecs used. 3.

Testing of the Algorithm

The testing phase was carried out in two different scenarios where Ubuntu 12.10 and Centos 6 operating systems were used for testing the algorithm. However, the algorithm was developed in netbeans IDE 7.3. The tools used for testing of the algorithm are stated here followed by the two scenarios.

[94]

Dar and Shah/COMMUNE – 2015

4.1.

Scenarios for Testing Algorithm

The testing of the algorithm is categorized into two different scenarios. The first scenario was carried out for testing and training of the algorithm to check and make sure that the algorithm is performing according to the specified criteria. After which the second scenario was carried out which captured the live traffic and classified the packets based on the classification criteria parameters which were then processed. For each scenario two sets of algorithms were used. One of the algorithm was optimized which is the ``Aho Corasick`` and the second algorithm used for both scenarios was not optimized was the ``boyer moore``. The resulted processed packets from both the algorithms were then compared for analysis purpose. 4.1.1.

Scenario No 01

In this scenario three virtual machines were used in which one of the virtual machines based on the Centos operating system acted as a server where the asterisk server, DNS server and Wireshark were installed, the second and third virtual machine was based on the Win 7 operating system in which the client software Xlite was installed. The dial plans were configured on the asterisk server and calls were placed between two client machines using Xlite, meanwhile the Wireshark on the server successfully captured the traffic generated between the source and destination client machines. The captured traffic was saved for analyses by the optimized algorithm. This scenario was carried out to confirm and check whether the optimized algorithm was able to detect and process the traffic based on the parameters identified for the classification criteria which was completed successfully. 4.1.2.

Scenario No 02

The tool used for testing in this scenario was SIPp which was installed in Ubuntu 12.10 on three virtual machines. One of the virtual machines acted as the user agent client, the second as the user agent server and the third virtual machine was acting as a router. To generate the calls the UAS server was initiated by typing the following command. ``./sipp –sn uas`` which put the server into the listening mode. Meanwhile on the client`s side the command used was ``./sipp –sn uac_pcap ``. This command will generate SIP traffic. The optimized algorithm as explained above was executed on the router which captured the live traffic in real time with the help of jNetPcap which classified the traffic based on the classification criteria. The next step was the processing of the traffic which was then sent for the deep packet inspection algorithm which in this case was the AhoCorasick. The algorithm then generated a graph which showed the number of processed packets along with their time taken to process them within the particular time scale. 4.

Results and Discussions

Two voice codecs were used in these test environments which were G.711 and G.729. Each of the packets of these codecs was processed through both the algorithms and the resulted processing time of the packets were analysed. It is to mention that in both the scenarios mentioned in section 4; only 1000 packets were processed as shown in graphical representation of the results. In some of the cases the DSCI algorithm detected and processed the packets faster in comparison to the existing algorithm.

Fig 3. Packet Processing Time of Scenario 1

Fig 4. Packet Processing Time of Scenario 2

[95]

Dar and Shah/COMMUNE – 2015

5.1.

Limitations

Based on the result analysis it is proven that the DSCI algorithm is much efficient in terms of traffic identification and classification which is lacking in the existing algorithm however there still are some of the limitation which exists in the DSCI algorithm: a)

The DSCI algorithm is limited to the scanning properties like packet size and inter-arrival time of only two codecs which are G.711 and G.729.

b) The deep inspection criteria in the DSCI algorithm is only limited to searching for the string of sip only which can result in the compromise of information security. c) 5.

The DSCI algorithm cannot detect and classify encrypted traffic only.

Conclusion and Future Work

This research focuses on the demonstration of a novel approach towards the detection and classification of VoIP traffic. The DSCI algorithm optimized and implemented was designed based on the specific classification criteria which consist of session ID, packet size and inter-arrival time along with the introduction of the concept of QUEUE. The results of both the Boyer Moore Algorithm and DSCI algorithm implemented were compared against each other which showed that although despite a slightly fast processing speed of the existing algorithm which is limited to string matching only, in general the DSCI algorithm is much efficient and in some cases faster than the Boyer Moore Algorithm To overcome the limitations of the algorithm in the future it is planned to investigate the properties of other voice codecs i.e.G.722, G.723, G.726, G.728 such as their packet size and inter-arrival time and include these properties as the requirement criteria for further optimization of the algorithm. Meanwhile to provide the deep inspection of other traffic types along with measures to avoid any security threads it is planned to design and develop a data base which will include the strings of normal traffic, other voice traffics and malicious codes. The DSCI algorithm will compare these strings with the database and it will not only help in securing the voice traffic but will drop the packets of other traffic types having any malicious code in its strings and to further enhance a robust security to detect and classify encrypted data it is planned to introduce the machine learning algorithms mainly Adaboost and RIPPER algorithm (Yildirim, 2010). References Ai-min Yang, S. Y. J., (2008). A P2P Network Traffic Classification Method Using SVM . Guangzhou, Changsha, Zhuzhou, Guangdong, China, November 18-21, 2008. Zincir-Heywood, R. A., (2011). Is machine losing the battle to produce transportable signatures against VOIP traffic. Halifax, Nova Scotia, Canada. Giacomazzi, G. V., (2008). Performance Evaluation of a Machine Learning Algorithm for Early Applicatio Identification . Milano, Italy. Chen Ling, Z. K., (2009). Fingerprints in the spectrum: spectral analysis and detection of VoIP traffic. Beijing, China. Chang-Yong Lee, H.-K. K.-H.-W.-C. (2009, March). A VoIP Traffic Monitoring System based on NetFlow v9. Seoul, Korea. Voice Over IP - Per Call Bandwidth Consumption, Feb 2 (2006). Retrieved March 26 (2013) from url: http://www.cisco.com/en/US/tech/tk652/tk698/technologies_tech_note09186a0080094ae2.shtml. Taner Yildirim, D. P., (2010). VoIP Traffic Classification in IPSec tunnels. Melbourne, Australia.

[96]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Integrated Tactile and Pointing Interface System using Non-Invasive Approach G. Mohiuddin Bhat, Rouf Ul Alam Bhat*, Uferah Maqbool, Fayiqa Naqshbandi, Naheeda Reshi , Fozia, Abid Baba Department of Electronics and Instrumentation Technolog,University of Kashmir,Srinagar,India.

Abstract The Current trend in research and development is to create technologies that are fully immersive and non-invasive. The same applies to the input/output devices used in computing technology and its variants. The virtual devices already in place have by far removed the problems that we face with conventional input devices like keyboard but most of the techniques used are invasive in nature. Hence, a need is felt to design and develop a single integrated non-invasive tactile and pointing interface device. The proposed setup is able to convert any surface into a fully functional keyboard cum mouse, the hardware of which is integrated well within the computing system e.g. laptop, desktop etc. The device has been designed around an array of sensors that can take tactile, gesture and positional information of hands and map them as cursor information on the computer display or key position information of keyboard for alphanumeric and control entry.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Arduino; IR sensors; Keyboard; Mouse; Non-invasive; Virtual

1. Introduction Computers have become a vital part of our lives these days. The interaction between a human and computer has increased .The interaction and thus operability of a computer by humans changed from the manually controlled switches to commands entered via a piece of black box called QWERTY keyboard. The journey of having a more user friendly, natural, and interactive experience did not stop there. A graphical user interface (GUI) controlled via a handheld small device called mouse made the interaction of humans with computers easier and a bit natural. With the ever rapidly growing technology and with a growing taste of miniaturization of gadgets and devices, new interactive technologies were born. Alternatives in the form of handwriting recognition, speech recognition, lack the accuracy and convenience of a full-blown keyboard. Speech input has an added issue of privacy. The virtual devices have obvious advantages over the conventional input devices. The virtual environment setup is able to convert any surface into a fully functional input device like a keyboard or a mouse. In the virtual Keyboard instead of a keyboard with many keys, an image of a keyboard is projected on any surface. The Virtual mouse is a replacement of a material mouse by nothing. It takes care of the shortcomings existing technologies by removing the need of a handheld device. The ultimate goal of this project is to implement a real time gesture and sound based Two-in-One virtual mouse and keyboard application. A tap denotes the pressing of a key and thus through suitable logic conversion the character is printed on the monitor. The virtual mouse consists of a linear set of arrays that through a fed logic are able to trace the path of the finger moved in front of it. Hence, this idea is precisely implemented to map the hand movements to an equivalent on screen cursor motion. The virtual mouse uses the same hardware as in the conventional ball mouse but replaces the need of moving a device to just the motion of hands in the air.

*

Corresponding author Tel.:+91 9086 422036. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Bhat et al/COMMUNE – 2015

2. Related Work The Virtual tactile and pointing devices has been implemented in a number of different forms like on screen keyboards or touch detection on projection screens most of which are invasive. Virtual keyboard designs based on 3-D optical ranging and CCD cameras have been designed as they are based primarily on image processing. The elaborate research done by Kölsch, M. and Turk, M. Highlights a variety of virtual keyboards in different forms, such as gloves, rings, hand gestures based and projection based devices. In Matsuiet al., a special 3-D camera, or two 2-D cameras are used. Additionally a pattern projector is used for projecting the keyboard. The Virtual keyboard designed in Andre et al. makes use of a single CCD camera. Even more significant is the work presented in andrew et al., where a shadow based analysis is used to acquire depth information from a 2-D image. The device also makes use of a high-power infrared light source and a camera with an infrared filter. A proposed virtual keyboard implementation method is the shadow analysis by Y. Adajaniaet al.. Range information provided by two or more cameras greatly improves the accuracy of touch detection. In Moet al.’s SmartCanvas in Mo et al., one camera is used to track the finger’s trajectory and other one is placed parallel to the surface, to detect whether the finger touches the surface.Korkaloet al. use multiple cameras, which are placed on the side of the LCD display with their optical axis parallel to the screen, to detect touch events and determine the position of touch on the screen. The shadows of the fingers can be used to recognize touch action. Liet al. put forward a Gaussian mixture model to detect shadows. Wilsonet al. and Chanet al. apply infrared llumination and a camera to capture the hand and estimate the touch by exploiting the shadow cast by the fingers. The extraction of the fingertip is simple, accurate, and robust, but auxiliary equipment is needed in the system. Recently, depth sensing cameras have become very popular in detecting touch. Benkoet al. and Wilsonet al. discuss the use of the depth-sensing camera to enable freehand interaction on surface. Similarly, Harrisonet al. and Murugappan et al. propose using Microsoft Kinect to obtain the distance between the fingertipand the surface.The detection of a button’s distortion can be carried out using edge detection. In computer vision and image process-ing, edge detection is a mature technology. Approaches for implementing first- and second-order digital derivatives for the detection of edges in an image are very effective Gonzalez et al. Another proposed model by Jun Huet al. is IPS, an interactive projective system, that merely consists of a projector and a mono-camera which maps by homography and extracting region of interest and uses distortion detection, and touch judgment. All of these methods are either fully invasive or make use of image processing methods to implement either keyboard design or the mouse interface at a time.The basic disadvantage with image processing methods is that it might behave erracticaly in poor or varying lighting conditions.Moreover,the use of more then one camera for touch detection,depth detection is cumbersome and non-ergonomic. 3. Overview of the Proposed System

Fig. 1: Block diagram of Implemented Design

As shown in Figure 1. the design consists of an Infrared distance sensor based finger position recognition system which determines the position of fingers and maps them into values of alphanumeric text elements, the value of which is fed through keyboard ISR, a position and tap detection using microphone sensor array which is a sound processing based differential sound intensity sensor that takes and maps them into graphical coordinate values of computer, a finger position recognition which locates the position of the finger using sensor array and moves the cursor on the screen accordingly, a sensor Module that serves as the eye of the Keyboard, a perception technology wherein the Sensor Module operates by locating the user's fingers in 3-D space and tracking the intended keystrokes, or mouse movements. Mouse tracking and keystroke information is processed and can then be output to the host device via a USB or other interface. Another part of the design is an IR light source & pattern projector to project a full-sized computer keyboard onto almost any surface, and disappears when not in use.

[98]

Bhat et al/COMMUNE – 2015

4. A Prototype System

Fig. 2. Implemented Prototype

4.1. The Virtual Keyboard A typical Keyboard consists of various keys which when pressed generate a particular sequence of bit stream. The bit stream is converted into a corresponding alphanumeric and control function logic, as understood by the computer’s CPU, by the keyboard controller. 4.1.1. Design Consideration Internally a keyboard consists of two thin sheets of insulators separated by another thin insulating sheet. Conducting paths or lines are present on both the sandwiching sheets. As a key is pressed externally particular lines are brought into contact at that point where the sandwiched insulating sheet is punched. These lines form a set of characters called the Scan matrix which is supervised by the keyboard controller. Upon pressing a key the Microprocessor determines the coordinates of the pressed key and a particular Scan-Code is generated by the keyboard encoder IC. This code is then transmitted to the computer’s main CPU. In this way CPU comes to know as to which Key was pressed. The Keyboard driver converts the Scan-Code into the corresponding character. In our design this mechanical setup is replaced by a fully electronic sensor and audio detection circuit. The mode of feeding the input to the encoder IC is changed only, which now being wireless sensing rather any wired connections. The rest of the processing mechanism of these pressed symbols remains same as done by the keyboard IC as the IC module is retained in the design. 4.1.2. Working The overall working of the proposed keyboard design is given as under  The Projector projects the keypad on the surface.  Four distance sensors are used to detect the position of the finger. The distance sensor array used is GP2D12 which is an array of IR sensors. Whenever an IR beam is interrupted by a finger it is reflected back to the detector .Using the method of triangulation, the sensors produce a corresponding output voltage. These sensors continuously send the data input to the Arduino analog pins. An internal program converts this voltage to a corresponding distance and locates the position of the finger on the keyboard.  Whenever a tap is made by our finger it is detected by the tap detection circuit. It is a Preamplifier Circuit that consists of a condenser microphone connected to a preamplifier circuit that boosts the input signal and converts it into a digital output. The sound detection and filtering range is set by changing the value of a Gain-Set resistor. The Pre-amplifier circuit is designed using LM324 quad op-amp IC. The circuit produces an output of one (high voltage) only in a particular range of input sound level. The tap by the finger of the user falls in such a range and is thus able to get detected. The background noise or any other moderate sound is rejected by the tap detection circuit and an output of zero is produced. The tap generates a corresponding control signal to the Arduino Processor and an interrupt service routine for the keyboard module is executed.  The Arduino Mega 2560 upon being interrupted by the tap detection circuit runs an internal program for keyboard. The Program procures the position information of the finger using the input from the sensor module and activates accordingly a particular CMOS CD4006 switch control signal.

[99]

Bhat et al/COMMUNE – 2015





Each CD4066 IC has 4 CMOS switches with four different controls for controlling each of the four switches. The inputs to each of these switches are taken from the upper and lower sections of the keyboard encoder IC. Depending on the position of the finger on the projected virtual keyboard a particular control signal is activated. The switch for which the control signal is pulled high is closed. For each key pressed a different combination of the input lines going to encoder IC are joined by the CD4066 switch and hence the tapped character is displayed on the monitor screen.

4.2. The Virtual Mouse 4.2.1. Design Consideration Looking inside a conventional ball reveals that a ball remains in touch with two rollers .These rollers are set in a way to determine the motion along the rectangular coordinates. A punched wheel remains attached with each roller. The wheel moves in between a light detector circuit. As the wheel rotates, it repeatedly interrupts the light beam from the emitter reaching to the detector. So measuring the no. of times a wheel rotates determines how far the motion has occurred. The detector circuit is analyzed and it is observed that a pair of square waves are generated at both the wheel detectors .The phase between the square waves generated at each wheel determine the direction of motion and the frequency of the generated waves determine the speed of the cursor motion. In conventional mouse, the operation of movements is carried by thin wheels one of which detects motion along y-axis and the other detects motion along xaxis. In case of Virtual mouse design the operation of the mouse is controlled by Arduino UNO 328 & Mega 2560 .The Arduino controls the operation based on the inputs it receives from the sensor module and through the software sends necessary control signals (the square waves) to the x- inputs and y-inputs of the detector circuits The Inputs are fed to the mouse via a controlled inverter circuit i.e. an XOR IC. The mouse remains attached to the computer via a USB port. 4.2.2. The Working The following steps are involved in the working of our proposed prototype of the virtual mouse  In order to move the cursor on screen we placed our finger in front of the sensor module on any surface. The sensor in front of which the finger is placed detects the presence of our finger and forwards the information to the Arduino Mega.  As we move our finger in a particular direction, different sensors are interrupted. The direction of motion hence can be sought by analyzing the way in which the sensors are being interrupted.  If the direction of motion is towards x-axis then the square wave pulse is generated at pins 12 and 13 of Arduino UNO. The output of pin 12 is directly given to the x-input of the mouse and that of 13 is given to XOR gate along with the control input from pin 6 of Arduino UNO. If the delayed input lags the normal xinput, the cursor moves in +X direction otherwise it moves in –X direction.  Similarly, Outputs for the Y direction detector circuit of the mouse are generated at same pins of Arduino Mega. Another XOR gate is controlled by pin 7 of the same processor board. The logic at these pins causes a motion of the cursor along y direction  However, if the motion is at an angle, simultaneous square waves, and their delayed versions are generated by both the processors and fed to the y and x detector circuits of the mouse. This causes a similar motion of the cursor on the screen. The Programs for generation of square waves is already written in the Arduino Processors. The keyboard module is activated upon a tap and the mouse functioning is halted for that moment. However, the halt is too small to be perceived. Hence we observe a smooth operation upon transition from mouse to keyboard or vice-versa.

Fig. 3. Tap Detector output at various stages Fig 4.Sensor Output Variation with distance of finger

[100]

Bhat et al/COMMUNE – 2015

5. Results and Discussion 5.1. Tap Detection Circuit The Tap created a voltage greater than 2.8V and hence was detected .Rest of the environmental noise was rejected by the tap detector setup. Table 1 shows the output of the tap detector as per figure 3. Table 1. Output of Tap detector to varying sound intensity voltages levels

S. No

Voltage as per the sound intensity

1 2

<2.8V >=2.8V

Digital Output corresponding to Preamp output Voltage 0V(Keyboard Inactive) 5V(Tap Detection/Valid Keyboard Input)

5.2. Sensor Module The Output of the sensor varied with the distance of the obstacle as shown in the figure 4 .It can be seen that the differentiable range was greater than 10 cm. Hence the Sensor was calibrated using Voltage= (5/1024)*Analog Reading, Range= (6787 / (Voltage-3))-4 5.3. Virtual Mouse Table 2 summarizes the behavior of the virtual mouse and hence the on screen cursor according to the gesture of hands in front of the sensor Table 2. Summary of Virtual Mouse behaviour Motion of hands

Cursor Y direction Movement

Cursor X direction movement

Stationary (in front or away from sensor array) Towards or away vertically from one sensor

No Motion Up Or Down Respectively

Along horizontal direction(all sensors covered)

No motion

At an Angle w.r.t. the sensor array In between two sensors(stationary)

Up or down No Motion

No Motion No Motion Left or Right as per the direction of motion of hand Right or Left Minute Oscillations horizontally

The oscillations were taken care of by widening the separation gap between the adjacent sensors. Virtual Keyboard   

We designed the keyboard with 12 input characters as given in the table 3. Whenever we tapped on any of the characters the same character was written on the screen editor. Keeping the finger at the same place didn’t make multiple entries but it required multiple taps to enter a same character continuously multiple no. of times. The delay between tapping and writing of the character was just a few milliseconds. Hence it was almost as quick as the conventional keyboard.

6. Conclusion The Implemented design was able to successfully function both as an independent virtual mouse and virtual keyboard not interfering with each other’s working. Taking into account the disadvantages and problems associated with the proposed design some changes can further be made .The future of this technology depends upon the future developments in virtual technology. The proliferation and use of this technology will be immense in the future. References Matsui, N. & Yamamoto, Y. A New Input Method of Computers with One CCD Camera: Virtual Keyboard. Keio University, 2000. Kölsch, M. and Turk, M. Keyboards without Keyboards: A Survey of Virtual Keyboards, Workshop on Sensing and Input for Media-centric Systems, Santa Barbara, CA, June 20-21, 2002 H. Du, T. Oggier, F. Lustenburger and E. Charbon, “A virtual keyboard based on true-3D optical ranging,” Proc. British Machine Vision Conference (BMVC), Oxford, pp. 220-229, Sept. 2005.

[101]

Bhat et al/COMMUNE – 2015 Andrew D. Wilson, PlayAnywhere: a compact interactive tabletop projection-vision system, Proceedings of the 18th annual ACM symposium on User interface software and technology, October 2326, 2005, Seattle, WA, USA Z. Mo, J. P. Lewis, and U. Neumann, “SmartCanvas: A gesture-driven intelligent drawing desk system,” in Proc. ACM IUI, 2005,pp. 239–243. H. Benko and A. Wilson, “DepthTouch: Using depth-sensing camera to enable freehand interactions on and above the interactive surface,” in Proc. IEEE Workshop ITS, vol. 8, 2009. C. Harrison, H. Benko, and A. D. Wilson, “OmniTouch: Wear-able multitouch interaction everywhere,” in Proc. ACM UIST, 2011,pp. 441–450 C. Harrison, D. Tan, and D. Morris, “Skinput: Appropriating the body as an input surface,” inProc. ACM CHI, 2010, pp. 453–462. S. K. Kane, D. Avrahami, J. O. Wobbrock, B. Harrison, A. D. Rea,M. Philipose, and A. LaMarca, “Bonfire: A nomadic system for hybrid laptoptabletop interaction,” in Proc. ACM UIST, 2009, pp. 129–138. A. D. Wilson, “PlayAnywhere: A compact interactive tabletop projection-vision system,” inProc. ACM UIST, 2005, pp. 83–92. L.-W. Chan, H.-T. Wu, H.-S. Kao, J.-C. Ko, H.-R. Lin, M. Y. Chen, J. Hsu, and Y.-P. Hung, “Enabling beyond-surface interactions for interactive surface with an invisible projection,” in Proc. ACM UIST, 2010, pp. 263–272. A. D. Wilson, “Using a depth camera as a touch sensor,” inProc. ACM ITS, 2010, pp. 69–72. R. C. Gonzalez and R. E. Woods, Digital Image Processing. Reading, MA, USA: Addison-Wesley, 1992. O. Korkalo and P. Honkamaa, “Construction and evaluation of multi-touch screens using multiple cameras located on the side of the display,” in Proc. ACM ITS, 2010, pp. 83–90. S. Murugappan, N. Elmqvist, and K. Ramani, “Extended multitouch: Recovering touch posture and differentiating users using a depth cam-era,” inProc. ACM UIST, 2012, pp. 487–496. W. J. Li, L. M. Wu, and C. Liu, “Research of hand gesture recognition in multitouch projector-camera system,” Adv. Mater. Res.,vol. 588, pp. 1184– 1187, Nov. 2012. Jun Hu, Guolin Li, Xiang Xie, Zhong Lv, and Zhihua Wang,” Bare-fingers Touch Detection by the Button’s Distortion in a Projector–Camera System Y. Adajania, J. Gosalia, A. Kanade, H. Mehta, Prof. N. Shekokar,” Virtual Keyboard Using Shadow Analysis

[102]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

A Compound of Negative Binomial Distribution with Two Parameter Lindley Distribution as a Tool for Handling over Dispersion Adil Rashida*, T. R. Jana , Musavir Ahmedb a Department of Statistics, University of Kashmir, Srinagar, India Department of Linguistics, University of Kashmir, Srinagar, India

b

Abstract The present paper introduces in a new count data model with three parameters which is obtained by compounding a negative binomial distribution with two parameter Lindley distribution. The new model so obtained generalizes several compound distributions that has been previously discussed in literature and can be considered as an alternative for data analysis with overdispersed count data. Some properties of the proposed count data model such as factorial moments, mean, variance, standard deviation and coefficient of variation has also been obtained. In addition to this estimation of parameters of the proposed model has also been discussed. Finally the usefulness of the model is carried out on a sample of count data. The results show that proposed count data model offers a better fit as compared to other count data models.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Negative Binomial Distribution; Two Parameter Lindley Distribution; Compound Distribution; Factorial Moments, Over-Dispersion and Count Data.

1. Introduction The analysis of count data has received maximum attention of researchers from the last one decade. In count data the observations always take non-negative integral values. Poisson distribution is a standard model for the analysis of count data with the assumption of equality of mean and variance. However, in practice sometimes count data shows dispersion in that case negative binomial distribution (NBD) can be used efficiently to model the count data claims instead of Poisson distribution. For modelling of claims count data that are over-dispersed and heavy tailed compound distributions provides a better fit when compared with classical Poisson and NBD. From the last few years, researchers have focused much attention on compounding mechanism to construct suitable, flexible, and alternative models to fit observed count data. Sankaran (Sankaran, 1970) introduced Poisson-Lindley distribution by compounding Poisson distribution with one parameter Lindley distribution to fit count data. The compounding of NBD with inverse Gaussian distribution was introduced by Gomez et al. (Gomez et al, 2008). Zamani and Ismail (Zamani, Ismail, 2010) constructed a new mixed distribution by compounding NBD with one parameter Lindley distribution that can be considered as an alternative to model count data claims. Adil Rashid and Jan obtained several compound distributions for instance, (Adil, Jan, 2013) a compound of zero truncated generalized negative binomial distribution with generalized beta distribution, (Adil, Jan, 2014a) they obtained a compound of Geeta distribution with generalized beta distribution and (Adil, Jan, 2014b) by using the same compounding mechanism they constructed a compound of Consul distribution with generalized beta distribution. Most recently again they (Adil, Jan, 2014c) obtained a mixture of generalized negative binomial distribution with that of generalized exponential distribution which contains several distributions as special cases and they proved that this particular model is better in comparison to others when it comes to fit observed count data set. Based on the same compounding mechanism we shall introduce a new count data model which is obtained by compounding NBD with two parameter Lindley distribution (TPLD) because Shankar and Mishra (Shankar, Mishra, 2013) proved TPLD distribution is a better model than one parameter Lindley distribution for analyzing survival, *

Corresponding author. Tel.:+91 8803 892549. E-mail address:[email protected] ISBN: 978-93-82288-63-3

Rashid et al / COMMUNE-2015

waiting time and grouped mortality data and hence it is natural to consider a compound of NBD with TPLD instead with one parameter Lindley distribution which was previously considered by Zamani et al. (Zamani et al, 2010) and this particular model will be symbolized by NB-TPL r ,  , . 2. Material and methods A discrete random variable is X said to have a NBD with parameters r, p if its probability mass function is given by

 r  x  1 r x  p q f1 ( x)    x  r  0 and 0  p  1 . The first two

where, binomial distribution are respectively given by

E ( x)  r

(1) moments about zero and factorial moment of order k of a negative

q1  r q  (r  k  1)! q q E( x 2 )  r [ k ] ( x)  2 p (r  1)! p k p ; and

A continuous random variable is said to have a TPLD

f 2 ( x, , ) 



2

 

 ,  if its probability density function is given by

1  x e  x ; x  0 ,   0 ,    

2

  1 , the two parameter Lindley distribution (2) reduces to one parameter Lindley distribution reduces to exponential distribution with parameter  . For

2.1

and at

  0 it

Definition

X |  be a random variable having NBD r, p  e  where the parameter  is itself treated as another random variable following two parameter Lindley distribution TPLD  ,  then determining the distribution that results from marginalizing over  is known as a compound (mixture) of negative binomial distribution with that of two parameter let

Lindley distribution this compound negative binomial distribution has three parameters and will be denoted by NB TPL

r , ,  .

Theorem 1: The probability mass function for the compound of NB-TPL

 r  x  1  2  f ( x; r ,  , )    x    Where, x  0,1,2.... ,` r ,  ,  0 Proof: Since X

x

 x

  j  (1) j 0

 

j

r , ,  model is giving by

 r   j    r    j 2 

   

|  ~ NBD(r, p  e  ) and  ~ L ( , ) distribution then using definition 2.1 we get 

f ( x ; r ,  , )   f1 ( X |  ) f 2 ( ; , ) d

(3)

0

 r  x  1  2   f ( x : r ,  , )   (1  e  ) x e r 1   e  d   x   0   r  x  1  2 x  x  j      (  1 ) e  ( r   j ) 1   d      x     j 0  j  0 r    j     r  x  1  2 x  x     (1) j     2   x     j 0  j   (r    j )  Where

x  0,1,2.... r,  ,  0

[104]

(4)

Rashid et al /COMMUNE – 2015

2.2

Special cases of NB-TPL

Case (i): For

r , , 

model

  1, TPLD  ,  reduces to one parameter Lindley distribution L   and a compound of NBD with   1 in (4)  r    j  1   r  x  1 x  x       (1) j  , x  0,1,2.... r ,  0 f ( x : r , )  2    1  x  j 0  j  ( r    j )  

one parameter Lindley distribution will be obtained by simply substituting 2

The above compound probability mass function was obtained by Zamani et al.( Zamani et al, 2010) Case (ii): For

  0,

TPLD

 , 

reduces exponential distribution and a compound of NBD with exponential

  0 in it.   r  x  1 x  x   1     (1) j  f ( x : r , )     , x  0,1,2.... r,  0 ( r    j )  x  j 0  j   

distribution is followed from (4) when we put

The probability function displayed above was obtained by Willmot et al. (Willmot et al, 1981) Case (iii): For r  1, NBD reduces to geometric distribution and a compound of geometric distribution with TPLD

 ,  is followed from (4) when we put f ( x :  , ) 

Case (iv): For r  1 and distribution

0.0

0.4

pmf

0.8

0.0 0.2 0.4 0.6 0.8

1

2

3

4

5

6

0

No. of claims

pmf 0

1

2

fig (c):

Fig 1(a): Plot of pmf for NB-TPLD

3

4

5

1

2

fig (b):

0.0 0.2 0.4 0.6

fig (a):

3

4

5

6

No. of claims

0.0 0.1 0.2 0.3 0.4

pmf

   j  2   2 x  x   (1) j  , x  0,1,2....   0  2    1 j 0  j   (  j  1) 

  0 in (4), we get a compound of geometric distribution with exponential distribution x    j  1   x f ( x, )      (1) j  , x  0,1,2.... ,  0 2 j 0  j   (  j  1) 

0

pmf

1    j     2 x  x   (1) j  , x  0,1,2....  ,   0  2     j 0  j   (  j  1) 

  1 in (4), we get a compound of geometric distribution with one parameter Lindley

f ( x, )  Case (v): For r  1 and

r  1 in it.

6

0

No. of claims

1

2

fig(d):

3

4

5

6

No. of claims

r  20 ,   20 ,   100 , fig 1(b): Plot of pmf for NB-TPLD r  7,   20 ,   150 ,

Fig 1(c): Plot of pmf for NB-TPLD

3. Factorial moments of NB-TPL

r  22 ,  30 ,  50 , fig 1(d): Plot of pmf for NB-TPLD r  9 ,  10 ,  10

r , ,  model





In this section we will obtain factorial moments of NB-TPL r ,  , model and also study some of its important mathematical properties such as mean, variance, and standard deviation through its factorial moments.

[105]

Rashid et al / COMMUNE-2015

X |  ~ NBD (r , p  e  ) , where  follows TPLD(  ,  ) then factorial moments of order k of a compound of NBD with TPLD(  , ) model can be obtained by using the expression k  ( X )  E mk  X |   (5) If

Where

mk  X |   is the factorial moment of NBD.

Theorem 2: The

k th order factorial moment of NB-TPL r ,  ,  model is given by the expression

  j  k    2 (r  k  1)! k  k    (1) j   2     (r  1)! j 0  j   ( j  k  )  Where x  0,1,2..., for r ,  and   0 th Proof: It is known that the k order factorial moment of NBD r, p  is

k  ( X ) 

(r  k  1)!(q) k mk  X |    (r  1)! p k Thus, k

th

order factorial moment of NB-TPL

 k  ( X )  

(r  k  1)!  2 (r  1)!   



 (e



 1) k 1   e  d

0

k  (r  k  1)!    (1) j   (r  1)!    j 0  j  k



e

 ( j  k  )

1   d

0

  j  k   k  (r  k  1)!     (1) j   2  (r  1)!    j 0  j   ( j  k  )  x  0,1,2..., for r,  and   0 2

 k  ( X ) 

3.1

p  e 

(r  k  1)! E (e   1) k (r  1)!

2

Where

r, , model is obtained from (5), if we let

k

Factorial moments of some special cases of NB-TPL

Case (i) If distribution

 k  ( X ) 

(6)

r, ,  model

  1 in (6) we get k th order factorial moment of a compound of NBD with one parameter Lindley

 j  k  1 (r  k  1)!  2 k  k     (1) j   2  , x  0,1,2..., for r and   0 (r  1)!   1 j 0  j  ( j  k   )  

  0 in (6) we get k th order factorial moment of a compound of NBD with exponential distribution (r  k  1)! k  k  (1) j   k  ( X )  , x  0,1,2..., for r and   0  (r  1)! j 0  j  ( j  k   )

Case (ii): For

Case (iii): For r  1 and   1 in (6) we get with one parameter Lindley distribution

k th order factorial moments of a compound of geometric distribution

 j  k   1 k! 2 k  k   x  0,1,2..., for   0   (1) j  k  ( X )   2 ,   1 j 0  j   ( j  k )  Case (iv): For r  1 and   0 in (6) we get with exponential distribution

 k   (1) j k  ( X )   k!    j 0  j   ( j  k   ) k

 , 

k th order factorial moments of a compound of geometric distribution

x  0,1,2..., for   0

[106]

Rashid et al /COMMUNE – 2015

3.2

Properties of NB-TPL

r, ,   model

For k  1 in (6) we get mean of NB-TPL

[1] ( X )  E ( X )  r



r, ,  model

 2     (   )(  1) 2 2



( 7)

[1] ( X )  r 1 For k  2 ,3 in (6) we have [ 2] ( X )  r (r  1) 2 ; [3] ( X )  r (r  1)(r  2) 3

E( X 2 )  [ 2] ( X )  [1] ( X )  r (r  1) 2  r1

(8)

E( X 3 )  [3] ( X )  3 2  21  (r  1)(r  2) 3  3r (r  1) 2  r1

(9)

V ( X )  r (r  1) 2  r1 (1  r1 ) r (r  1) 2  r 1 (1  r 1 )

 ( X )  r (r  1) 2  r1 (1  r1 ) ; CV ( X ) 

r 1

where

 2 (  1) 2 (    2)  2 2 (  2) 2 (    1)  2  2        1 and , 2 (   )(  1) 2 (   )(  1) 2 (  2) 2  2 (  1) 2 (  2) 2 (    3)  3 2 (  1) 2 (  3) 2 (    2)  3 2 (  2) 2 (  3) 2 (    1) 3   1 4. (   )(  1) 2 (  2) 2 (  3) 2 1 

4. Parameter Estimation In this particular section the estimation of parameters of the proposed model will be discussed in detail by using two different methods. 4.1 Method of moments The method of moments is based on equating sample moments with the corresponding population moments to





estimate unknown population parameters. Therefore, in order to estimate the parameters of NB-TPL r,  ,  model we need to equate first three sample moments with corresponding population moments thus we have from (7), (8) and (9)

m1  r1 ; m2  r (r  1) 2  r1

and

m3  (r  1)(r  2) 3  3r (r  1) 2  r1

n

1  x i is the ith sample moment. n x 0 The solution of r ,  and  may be obtained by solving above equations simultaneously. Where mi 

4.2

Maximum Likelihood Estimation (MLE)

In this section, estimation of parameters of NB-TPL The likelihood function of NB-TPL

r, ,  model via MLE method will be discussed.

r, ,  model is given by

n  r  x  1   L(r ,  ,  )    x i 1     2

x

 x

  j  (1) j 0

 

j

 r   j    r    j 2 

   

and the corresponding log likelihood function is given by

 xi  x  n n  r  xi 1   2n log   n log(   )   log   i (1) j log L( r ,  ,  )  £ ( r ,  ,  )   log    x   i 1  i 1  j  0  j  i  Differentiating (10) on both sides respectively w.r.t r ,  and

[107]

 we get

 r   j     r   j 2 

   (10)  

Rashid et al / COMMUNE-2015

 x  x      (1) j 1  r    j  2   3  j   n n   j 0    r    j    £ (r ,  ,  )   (r  xi )  n (r )   x r  r   j     x i 1 i 1  j   (1)  2      r    j     j 0  j 

(11)

 x  x      (1) j 1  r    j  2   3    r    j    n   2n n j 0  j    £(r ,  ,  )      x      i 1   r   j     x j   (1)  2      r    j     j 0  j  x    x   (1) j       2  n   n j  0  j   r    j   £(r ,  ,  )         i 1  x  x   r   j  j    (  1 )     r    j 2  j 0 j    

(12)

       

(13)

These three derivative equations cannot be solved analytically, as they need to rely on Newton-Raphson (NR) method which is the best technique for solving equations iteratively. Zamani et al.(2010) estimated the parameters of NB-L r,   distribution from the data given in Table 1 numerically by using NR method and obtained rˆ  4.63 and

ˆ  23.55,

r,  

but here our aim is to fit the same type data set by proposed NB-TPL r,  ,   model and because NB-L

distribution is a special case of our model, therefore the only problem remains here, is to estimate

 which can

be estimated from (7) by substituting rˆ , ˆ in it. 5. Results and Discussion Here, we illustrate the application of the proposed model by fitting it to the real data set taken from Klugman et al. (Klugman et al, 2008). The data which appears in the first two columns of below given table provides information of observed counts of the number of accidents on automobile insurance policies. It is clear that given data is overdispersed since sample variance (0.288) is greater than the sample mean (0.214). Therefore, the given data must be fitted by some over-dispersed model. For this reason, we strongly believe that proposed model is suitable for them. By comparing the fitted distributions in Table 1, based on p value it is quite clear and evident that NB-TPL is a better model when it comes to fit the over-dispersed count data with large number of zeroes.

r, ,  model

Table 1: Number of accidents 0 1 2 3 4 5 6 7

Observed Frequency 7840 1317 239 42 14 4 4 1

8+

0

Total

9461

Parameter Estimation Chisquare Estiimates and DF p value

Poisson 7638.30 1634.61 174.90

12.5 0.7  0  1 3.2  0  0   0  9461

Fitted Distribution NBD 7843.70 1290.20 257.30 54.50

11.8 2.6   0.6  15.3 0 .2   0.1  9461

NBD-L 7853.6 1287.4 247.6 54.2 13.2

NBD-TPL 7878.08 1271.95 241.22 49.64 12.5

3 .5  1.0  5 0 .3  0.2

3.30  3.94  7.6  0.28 0.09

9461

  0.214

r  0.70

r  4.63

  0.76

  23.55

293.80, DF=2

865,DF=2

6.99,DF=3

P value=0

P value = < 1%

P value=0.07

[108]

9461

r  4.63

  23.55   0.52 4.21, DF=2 P value=0.12

Rashid et al /COMMUNE – 2015

6. Conclusion In this paper we have proposed a new model by compounding NBD with TPLD which contains several compound distributions as its special cases. Furthermore, we have derived several properties of proposed model such as factorial moments, mean, variance and coefficient of variation. In addition to this, parameter estimation of the proposed model has also been discussed. Finally, we have compared efficiencies of NB-TPL r,  ,  model with Poisson, NB and NB-L distribution by using chi-square goodness of fit test and based on the p value it has been shown that NB-TPL model offers a better fit as compared to Poisson, NB and NB-L distribution.

r, , 

References Adil, R and Jan, T. R., 2014c. A mixture of generalized negative binomial distribution with Generalized exponential Distribution.J. Stat. Appl. Pro., 3(3) 451-464. Adil, R and Jan, T. R., 2013. A compound of zero truncated generalized negative binomial distribution with the generalized beta distribution, Journal of Reliability and Statistical studies, 6(1), 11-19. Adil, R and Jan, T. R., 2014a. A compound of Geeta distribution with the Generalized Beta Distribution, Journal Of Modern Applied Statistical Methods, 13(1), 278-286. Adil, R and Jan, T .R., 2014b. A compound of Consul distribution with the Generalized Beta Distribution, Journal of Applied probability and Statistics Methods 9, (2), 17-24. Gomez, D. E., Sarabia, J. M and Ojeda, E. C., 2008. Univariate and Multivariate Versions of the Negative Binomial-Inverse Gaussian Distributions with Applications, Insurance Mathematics and Economics 42, 39–49. Kulgman, S. A., Panger, H. H and Willmot, G. E., 2008. Loss Models from data to decision, 3rd Edn. John Wiley and sons, USA, 101-159. Panger, H. H and Willmot, E., 1981. Finite sum evaluation of the negative binomial exponential model, Astin Bulletin 12, 133–137. Sankaran, M., 1970 The Discrete Poisson-Lindley Distribution. Biometrics, 26, 145-149. Shanker, R. Mishra, A., 2013. A two parameter Lindley distribution, Statistics in Transition new Series., 14(1) 45-56. Zamani, H. and Ismail, N., 2010. Negative Binomial–Lindley Distribution And Its Application, J. Mathematics And Statistics., 1, 4–9.

[109]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Grammatical Structure in the Dependency Framework: A Computational Perspective in Kashmiri Aadil Amin Kak*, Sumaya Jehangir, Mansoor Farooq, Sumaira Nabi Department of Linguistics, University of Kashmir, Srinagar, India

Abstract The present paper will attempt to provide a fresh look at the structure of Kashmiri from the Dependency grammar perspective. Furthermore, what is important is to understand the increasing role of Dependency Grammar in Machine Translation, also observed in Indian Languages. The present study can be a step in this direction.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: V2 phenomenon; Word order; Dependency; Valency; DG Parsing

1. Introduction Kashmiri is the language spoken in the greater region of Kashmir. It has more than 5 and a half million speakers (Lewis et al, 2015). Kashmiri is a Dardic language (Grierson, 1915), classified with Shina, under a separate group in the Indo Aryan family. Kashmiri is an inflectionally rich language with pronominal clitics and has a comparatively flexible or free word order, e.g. Table 1. Kashmiri word order examples ra:man

khʲov

batᵻ

ram-ERG

Eat-PST

rice

S ‘Ram ate food’ batᵻ

V

O

khʲov

ra:man

rice

Eat-PST

ram-ERG

O ‘Ram ate food’

V

S

Also like German, Yiddish, and Icelandic, Kashmiri is a V2 language. In some cases the main verb can be present at the end of the sentence, e.g. Table 2. Kashmiri V2-structure example bᵻ

chus

batᵻ

kheva:n

I-NOM

be-PST

food

eat-PR

S ‘I am eating food’

V-aux

O

V-main

These types of structural modifications are hard to capture when a machine is used for parsing a language like Kashmiri. DG framework can be used to develop parsers that can capture such modifications in a better way as compared to the constituency frameworks. * Corresponding author. Tel.: +91 9906 048302 E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Kak et al/COMMUNE – 2015

2. Dependency Grammar and Dependency Framework The origin of dependency grammar can be traced back to the work of Panini, a Sanskrit Grammarian, who talked of the verb and the relations that the other constituents have with it as Karakas, in his grammar Astadhyayi (Katre, 1987). In the modern times, the concept was brought to light by the work of Tesniere (1959). His work has been translated by linguists like Kubler et al (2009), who have also contributed a lot to the notion of dependency grammar and its role in parsing. In the Indian context, Dependency Treebanks and Parsers have been devised for four languages: Urdu, Hindi, Telugu and Bengali. A preliminary attempt has also been made to devise a Dependency Treebank for Kashmiri (Bhat, 2012). Dependency grammar is a framework based on the idea that syntactic structure consists of lexical items linked together by binary asymmetric relations or dependencies. Unlike constituency grammars there are no phrasal nodes in dependency framework. DG is not structure dependent like the other grammatical frameworks. It rather depends upon the semantic relations. The dependency relation exists between a superior term and an inferior term, or we can say that it is a relation between modifier and modified (the modified term being the head and the modifier being the complement or adjunct). The superior term is also called head, governor or regent. The inferior term is also called dependent, modifier or subordinate. Verb is taken to be the head of the sentence, as it is in Constituency Grammars, and the other words are dependents on it. An important concept in dependency grammar is that of Valency. Valency is the ability of the verb to take arguments. The main idea is that verb imposes requirements on its syntactic dependents that reflect its interpretation as a semantic predicate. e.g. Consider the verb dʲun ‘give’. It is capable of taking three arguments with it: subject, direct object, and indirect object. Table 3. Three Arguments of Verb dʲun to ‘give’ ra:m-an

dʲut

ʃa:m-as

kalam

ram-ERG

give-PST

shyam-DAT

pen

S ‘Ram gave a pen to Shyam’

V

Ind O

DO

Therefore, dʲun ‘give’ has three valencies. 3. Dependency Parsing Parsing or syntactic analysis can be defined as the process of analyzing the structure of a natural language or computer languages, conforming to the rules of a formal grammar. Within the domain of computational linguistics, this term is used to refer to the analysis of the sentences or constructions or structures of a language by a machine. Dependency parsing means the computational implementation of syntactic analysis based on dependency relations. This grammatical model is suited in languages that have a free word order and the languages in which morphology and syntax tend to overlap i.e. a word including its morphemes can act as a phrase or even a complete sentence e.g. /khʲom/ ‘I ate’ (a single word forms a complete sentence), /tim patʃᵻ/ ‘the ladies walked’ (the syntactic and morphological information is given by the word /patʃᵻ/). Some examples: Table 4. Syntactic Analysis of ‫تماشہ‬

‫اون‬ٚ ‫رامن‬ ‫اون تماشہ‬ٚ ‫رامن‬ ram-an

on

tama:shᵻ

ram-ERG

bring-PST

toy

‘ram brought a toy’

Fig. 1. Dependency analysis of

‫اون تماشہ‬ٚ ‫رامن‬ v-main ‫اون‬ٚ obj(k2)

sbj (k1)

‫تماشہ‬

‫رامن‬

(a simple sentence)

[111]

Kak et al/COMMUNE – 2015

Table 5. Syntactic analysis of

‫مودُر بٲت‬ ٚ ‫گیوو‬ ٚ ‫لۄکٹۍ ُشرۍ‬ ‫مودُر بٲت‬ ٚ ‫گیوو‬ ٚ ‫لۄکٹۍ ُشرۍ‬

lɔkᵻʈʲ

ʃurʲ

ɡʲov

modur

bəːt

small

child-NOM

sinɡ-PST

sweet

sonɡ

‘a small child sang a sweet song’

Fig. 2. Dependency analysis of

‫مودُر بٲت‬ ٚ ‫گیوو‬ ٚ ‫ۄکٹۍ ُشرۍ‬ v-main ‫گیوو‬ ٚ obj(k2)

sbj(k1)

‫ُشرۍ‬

‫بٲت‬ n-mod

n-mod

‫مو ُدر‬ ٚ

‫لۄکٹۍ‬

(Sentence showing noun modifiers)

Table 6. Syntactic analysis of

‫رام اوس بتہ کھٮ۪ وان‬ ‫رام اوس بتہ کھٮ۪ وان‬

ra:m

o:s

batᵻ

khewa:n

ram-ERG

bring-PST

rice

eat-CONT.ASP

‘Ram was eating rice’ Fig. 3. Dependency Analysis of ‫کھٮ۪ وان‬

‫رام اوس بتہ‬

v-main ‫کھٮ۪ وان‬ sbj(k1)

fraɡof obj(k2)

‫اوس‬

‫بتہ‬

‫رام‬

(Sentence showing V2 phenomenon)

Table 7. Syntactic analysis of

‫حیم تہ سلما ٔگیہ سالس‬ ٖ ‫ٔر‬ ‫حیم تہ سلما ٔگیہ سالس‬ ٖ ‫ٔر‬

rahi:m

tᵻ

salma:

gaji

sa:las

rahim-NOM

and

Salma-NOM

go-PST

party-DAT

‘Rahim and Salma went to the party’

[112]

Kak et al/COMMUNE – 2015

Fig. 4. Dependency analysis of

‫حیم تہ سلما ٔگیہ سالس‬ ٖ ‫ٔر‬ v-main ‫ٔگیہ‬ obj(k2)

sbj(k1)

‫سالس‬

‫تہ‬

ccof

ccof

‫سلما‬

‫رحیم‬ ٖ

(a sentence showing coordination) Table 8. Syntactic analysis of

‫ ل‬ٛ‫ یا کی‬ٚ‫موہنَن کھی‬ ‫ ل‬ٛ‫ یا کی‬ٚ‫موہنَن کھی‬

mohan-an

khe-ja:

ke:l

mohan-ERG

eat-PST.INT

banana

‘Did Mohan eat banana?’

Fig. 5. Dependency analysis of

‫ ل‬ٛ‫ یا کی‬ٚ‫موہنَن کھی‬

v-main ‫ یا‬ٚ‫کھی‬ obj(k2)

‫ ل‬ٛ‫کی‬

sbj(k1)

‫موہنَن‬

(yes no type question)

Table 9. Syntactic analysis of

ٛ ‫چیز‬ ٛ ‫چیز‬ ‫پسند‬ ٖ ‫ اکھ‬ٛ‫ یِس کھس ِہ بی‬ٚ‫پسند تہ بی‬ ٖ ‫أ ِکس کھس ِہ اکھ‬

ٛ ‫چیز‬ ٛ ‫چیز‬ ‫پسند‬ ٖ ‫ اکھ‬ٛ‫ یِس کھس ِہ بی‬ٚ‫پسند تہ بی‬ ٖ ‫أ ِکس کھس ِہ اکھ‬

əkis

tʃhu

akh

tʃiːz

pasand

tᵻ

beyis

tʃhu

bʲaːkh

tʃiːz

pasand

one-DAT

be-PR

one

thinɡ

like

and

other- DAT

be-PR

other

thinɡ

like

‘one person likes one thing and other person likes another thing’

[113]

Kak et al/COMMUNE – 2015

ٛ Fig. 6. Dependency analysis of ‫پسند‬

ٛ ‫چیز‬ ‫چیز‬ ٖ ‫ اکھ‬ٛ‫ یِس کھس ِہ بی‬ٚ‫پسند تہ بی‬ ٖ ‫أ ِکس کھس ِہ اکھ‬

v-main ‫تہ‬ ccof

ccof

k2 ‫کھس ِہ‬ k4

‫أ ِکس‬

K2

‫چیز‬ ٖ ‫اکھ‬

k2

‫کھس ِہ‬ k4

pof

K2

ٛ ‫پسند‬

‫ یِس‬ٚ‫بی‬

‫ اکھ چ ٖیز‬ٛ‫بی‬

pof

ٛ ‫پسند‬

(co-ordination between two sentences) 4. Conclusion The present study is an attempt to explore the usefulness of dependency framework in Kashmiri, which may be used for the purpose of parsing. Since there are no phrasal nodes in dependency grammar, this model can be used to capture various structural diversities of Kashmiri more efficiently than the previous frameworks. Speech recognition and text retrieval are the most important areas in today’s commercial applications and should be fast and consistent. The dependency approach can be useful for this kind of language processing. References Bhat, S. M., 2012. Introducing Kashmiri Dependency Treebank. Proceedings of Workshop on Machine Translation and Parsing in Indian Languages (MTPIL-2012), IIT Bombay, p.53-60. Bhatt, Rajesh, 1995. “Verb Movement in Kashmiri,” in “U. Penn Working Papers on Linguistics, 2. Grierson, G A, 1915. Linguistic Classification of Kashmiri, Indian Antiquary 44, p.257-270. Koul, Omkar Nath and Kashi Wali, 2006. Modern Kashmiri Grammar. Springfield: Dunwoody press. Kubler, Sandra, Ryan McDonald and Joakim Nivre, 2009. Dependency Parsing. New Jersey: Morgan and Claypool Publishers. Nivre, Joakim, 2006. Inductive Dependency Parsing. Dordrecht: Springer Publications. Lewis, M. P., G. F. Simons, and C. D. Fennig (eds.). 2015. Ethnologue: Languages of the World, Eighteenth edition. Dallas, Texas: SIL International. Tesnière, L. 1959. Éléments de syntaxe structurale. Paris: Klincksieck. Katre, Sumitra M. 1987. Astadhyayi of Panini. Austin: University of Texas Press.

[114]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Phrase Structure in Kashmiri: A UNL Approach Aadil Amin Kak*, Sumaira Nabi, Mansoor Farooq, Sumia Tariq Department of Linguistics, University of Kashmir, Srinagar, India

Abstract Universal Networking Language (UNL) is a computer language created to represent and process information across language barriers (Uchida et al, 2001). The main purpose of UNL is to represent an interpretation of a given utterance rather than for the purpose of translation. The present paper is an attempt to provide linguistic resources to the Enconverter for the purpose of UNLisation of Kashmiri corpus. The UNLised text can be converted to any language by the Deconverter. Till date, very little has been done regarding the development of MT (Machine Translation) for Kashmiri language and it is expected that the current work will be an important attempt in that direction. An MT system based on logical relations appears very promising, and applying this to Kashmiri should deliver good results. Keywords: Universal Networking Language; UNL Knowledge Base; UNLisation; UW’s, Relations; Attributes; Semantic network

1.

Introduction

Universal Networking Language is a computer language created to represent and process information across language barriers (Uchida et al, 2001). UNL is basically a knowledge representation language i.e. it is used to represent information conveyed by natural languages (Cardeñosa et al, 2009). UNL represents an interpretation of a natural language text and not its translation. UNL expressions have no ambiguity as is the case with natural languages. UNL provides an infrastructure for machines to handle what is meant by natural languages, and can also be used for the purpose of translation. UNL expressions must be semantically complete in order to be understandable to machines. Even though being a language for machines, UNL has all the components of a natural language. There are two fundamental movements in UNL: a)

UNLisation and Nlisation:UNLization is the process of representing the information conveyed by NL into UNL, and; b) NLisation is the process of generating a natural language document out of UNL. 2.

Expression of Information by UNL

In the UNL approach, information extracted from natural language text is represented in the form of a semantic network (UNL graph), composed of three distinct semantic entities i.e. universal words, relations and attributes. UW’s are the nodes of the UNL graph and can represent simple or compound concepts. Attribute labels express additional information about the UW’s that appear in a sentence. It includes tense, number, aspect, represent information on the role of the node in the UNL graph as in the case of @entry that indicates the main node of the graph. Relations formerly known as “links” are labeled arcs connecting a node to another node in a UNL graph (Martins, 2002). Consider the example below: ‫ندان ِکرکٹ‬ٛ ‫چھے ِگ‬ ‫سیما‬ ٚ ٖ

*Corresponding author. Tel.: +91 9419 055376. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Kak et al/ COMMUNE-2015

Table 1. Syntactic representation of

‫ندان ِکرکٹ‬ٛ ‫ے ِگ‬ ٚ ‫سیما چھ‬ ٖ

Siːmaː

tʃe

ɡindaːn

kirkʌt

sima

be-PR

play-CONT

cricket

‘sima is playing cricket’

Play (icl >do)

obj

agt

Seema(icl>person)

Fig. 1. UNL representation of ‫ِکرکٹ‬

@entry.@present.@progress

Cricket (icl>game)

‫ندان‬ٛ ‫چھے ِگ‬ ‫سیما‬ ٚ ٖ

In this graph “Play (icl >do)”, Seema (icl>person) and Cricket (icl>game) are UW’s . agt (agent) and obj (object) are relations and @entry.@present.@progress are attributes. UNL Expression: [UNL] agt(play(icl>do)@entry.@present.@progress, seema(iof>person)) obj(play(icl>do)@entry.@present.@progress,cricket(icl>game)) [/UNL] 3.

A Brief Description of UNL system

UNL system consists of UNL (which involves UW’s, Relations and Attributes), Language servers (enconverter and Deconverter) and Basic tools. With access to the internet one can convert NL to UNL and viceversa. 4.

UNLisation Framework of Kashmiri phrases

Enconverter is the core software in the UNL system. Enonverter converts a given sentence in natural language to an equivalent UNL expression. Enonverter converts a given sentence to UNL expression by accessing word dictionary and interpreting Analysis Rules. The Enconverter transforms all natural language sentences to UNL expressions using a dictionary and a set of grammar rules of the respective language. Enconverter is a language independent parser which can deal with various languages using respective dictionaries and enconversion rules. ENCONVERTER

NL

Universal word Dictionary

UNL EXPRESSION

UNL kB

Analysis Rules

Fig. 1. Components of Enconverter

[116]

Kak et al/COMMUNE – 2015

Using a NL corpus

Formulation of Kashmiri Dictionary in IAN

Rules for enconversion of Kashmiri corpus to UNL in IAN

To any language with its own De converter

UNL

Fig. 2. Scheme of Enconversion

The format of entries in a word dictionary is as: [HW]{ID}”UW”(ATTR….); [‫"}{ ]خوبصورت‬beautiful"(LEX=J, POS=ADJ) ; [‫"}{ ]جان‬John"(LEX=N, POS=PPN, GEN=MCL, NUM=SNG) ; [‫""}{]نزدیک‬ (LEX=P, POS=PRE,rel=plc,att=@near); ٖ [‫( ""}{]تہ‬LEX=C, POS=CCJ,rel=and); Table 2. UNLization of some Kashmiri phrases Kashmiri phrases

Analysis Rules

UNL Format

‫کتاب‬ /kita:b/

(%a,N,pos=NOU):=(%a);

Book

‫اکھ کتاب‬ /akh kita:b/

(%a,D,@indef)(%b,N):=(%b,+att=%a);

book.@indef

‫ِکتابہ‬ /Kita:bɨ/

(%a,N,PLR):=(%a,-PLR,+@pl);

book.@pl

‫سارے ِکتابہ‬ /Saːreː kitabɨ/

(%a,D,POS=QUA,@all) (%b,N,POS=NOU,NUM=PLR):=(%b,-PLR,+att=@all);

book.@all

‫ٹیبلس پٮ۪ ٹھ ِکتاب‬ /ʈeːblas peʈh kitaːb/

(%a,N,POS=NOU,NUM=SNG,CAS=DAT) (%b,P,POS=PRE,rel=plc,@on) (%c,N,POS=NOU,NUM=SNG) :=plc(%c;%a,+att=%b,-DAT); (%a,D,POS=DEM,@other)(%b,N):=(%b,+att=%a);

Plc (book,table.@on)

‫ اکھ ِکتاب‬ٛ‫بی‬ /bʲaːkh/

[117]

book.@other

Kak et al/ COMMUNE-2015

5.

Kashmiri phrases

Analysis Rules

UNL Format

‫نزدیک ِکتاب‬ ٖ ‫ٹیبلس‬ /ʈeːblas nazdiːk kitaːb/

(%a,N,POS=NOU,NUM=SNG,CAS=DAT) (%b,P,POS=PRE,rel=plc,@near) (%c,N,POS=NOU,NUM=SNG):=plc(%c;%a,+att=%b,-DAT);

Plc (book,table.@near)

‫جان تہ میری‬ /jaːn tɨ meːriː/

(%a,N,POS=PPN,CAS=NOM)(%b,C,POS=CCJ) (%c,N,POS=PPN,CAS=NOM):=and(%c;%a);

and(mary,john)

Conclusion

The present paper is a part of a larger study involving Kashmiri in the UNL set up. The paper is an attempt to frame Analysis Rules and word dictionary to be used by the Enconverter for the purpose of UNLisation of Kashmiri corpus. Once the text is transformed to UNL, it can be converted to any language by Deconverter using respective language’s Dictionary and Analysis Rules. References Cardenosa, Jesús, Alexander Gelbukh, and Edmundo Tovar (Eds) 2005. Universal Networking Language: Advances in Theory and Application. Instituto Politecnico Nacional, Centro de Investigation e. computation: Mexico. Cardeñosa, Jesús, Carolina Gallardo Pérez, Adriana Toni, 2009. “Multilingual Cross Language Information Retrieval: A New Approach,” Universal Networking Language: Present Status and Perspectives Workshop, in conjunction with CSIT 7th International Conference. Yerevan. Prashanth. K., 2010 web. Semantics Extraction from Text. Diss. Indian Institute of Technology, Bombay. Martins, Ronaldo, 2002 web. UNL Developers’ Guide. Vol. 2. UNDL Foundation: Geneva. < http://www.ronaldomartins.pro.br/unlx/Attributes.pdf> Uchida, H. and M. Zhu, 1998. “The Universal Networking Language (UNL) Specification Version 3.0”, Technical Report, United Nations University, Tokyo. Uchida, H., M. Zhu and T.C.D. Senta, 2005. Universal Networking Language, UNDL Foundation: Geneva. UNLweb. 2010. UNDL Foundation.

[118]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Estimation of Stress-Strength Reliability Using Finite Mixture of Exponential and Gamma Distributions Adil H. Khan*, T. R. Jan Department of Statistics, University of Kashmir, Srinagar, India

Abstract The term “stress- strength reliability” refers to the quantity 𝑃(𝑋 > 𝑌), where a system with random strength X is subjected to a random stress Y such that a system fails, if the stress exceeds the strength. In this paper Stress –Strength reliability is considered where various cases have been considered for stress (Y) and strength (X) variables viz., the strength follows finite mixture of one parameter exponential and one, two parameter Gamma distributions and stress follows one and two parameter exponential distribution. The general expressions for the reliabilities of a system are obtained. Special cases are also discussed. At the end, results are illustrated with the help of numerical evaluations.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Exponential Distribution; Gamma Distribution; Reliability; Stress; Strength

1. Introduction Now a day’s industrial world has daily been facing with new systems required high level of safety. In this regard, great attentions have been attracted by safety issues in the recent years. It is reported that system importance has a direct relationship with system safety. Thus, there is an increasing need to work on system safety leading reliability theory. There are appliances (every physical component possesses an inherent strength) which survive due to their strength. These appliances receive a certain level of stress and sustain. However, if a higher level of stress is applied then their strength is unable to sustain and they break down. Suppose Y represents the ‘stress’ which is applied to a certain appliance and X represents the ‘strength’ to sustain the stress, then the stress-strength reliability is denoted by R= P(Y
ISBN: 978-93-82288-63-3

Khan and Jan /COMMUNE – 2015

function of Generalized Possion distribution and Generalized Geometric distribution. The expression for P{X1 , X2 , … . Xk ≤ Y} has been obtained with X’s and Y following Possion distribution. Adil H. khan and T.R Jan (khan, Jan, 2014c) obtained Bayes estimators of the parameters of the Consul, Geeta and Size-biased Geeta distributions and associated reliability function. 2. Stress-Strength Model The term stress is defined as: It is failure-inducing variable. It is defined as stress (load) which tends to produce a failure of a component or of a device of a material. The term load may be defined as mechanical load, environment, temperature and electric current etc. The term strength is defined as: The ability of component, a device or a material to accomplish its required function (mission) satisfactorily without failure when subjected to the external loading and environment. Therefore, strength is failure-resisting variable. The variation in ‘stress’ and ‘strength’ results in a statistical distribution and natural scatter occurs in these variables when the two distributions interfere. When interference becomes higher than strength, failure results. In other words, when probability density functions of both stress and strength are known, the component reliability may be determined analytically. Therefore 𝑅 = 𝑃(𝑌 < 𝑋) is reliability parameter R. Thus R= P(Y
𝑥

𝑅 = 𝑃(𝑌 < 𝑋) = ∫ ∫ 𝑓(𝑥, 𝑦) 𝑑𝑥 𝑑𝑦 −∞ −∞

Where, 𝑃(𝑌 < 𝑋) is a relationship which represents the probability, that the strength exceeds the stress and 𝑓(𝑥, 𝑦) is joint pdf of X and Y. If the random variables are statistically independent, then 𝑓(𝑥, 𝑦) = 𝑓(𝑥) 𝑔(𝑦) So that 𝑥

∞

𝑅 = ∫ ∫ 𝑓(𝑥) 𝑔(𝑦) 𝑑𝑥 𝑑𝑦 −∞ −∞

(2.1)

𝑋

𝐺𝑌 (𝑋) = ∫ g(y)dy ∞

−∞

𝑅 = ∫ 𝐺𝑌 (𝑥)𝑓(𝑥)𝑑𝑥 −∞

Where, f(x) and g(y) are pdf’s of X and Y respectively In this model, we propose the following assumptions 1) The random variables X and Y are statistically independent. 2) The values of stress and strength are non-negative. In addition, we have considered the following cases 1. Mixture of one parameter exponential strength and two parameter exponential stress. 2. Mixture of one parameter gamma strength and exponential stress. 3. Mixture of two parameter gamma strength and exponential stress. 4. Mixture of two parameter gamma strength and mixture of exponential stress. A finite mixture of probability density function with k-components can be represented the form 𝑓(𝑥) = 𝑝1 𝑓1 (𝑥) + 𝑝2 𝑓2 (𝑋) + ⋯ + 𝑝𝑘 𝑓𝑘 (𝑥) 𝑘

where,

𝑝𝑖 > 0,

𝑖 = 1,2, … , 𝑘 𝑎𝑛𝑑 ∑ 𝑝𝑖 = 1 𝑖=

3. Reliability Computation Case I: Mixture of one parameter exponential strength and two-parameter exponential stress. Let X be the strength of k-components which follows mixture of one parameter exponential distribution with pdf 𝑓𝑖 (𝑥, 𝜆𝑖 ) and Y be the stress which follows two parameter exponential distribution with pdf 𝑔(𝑦, 𝛼, 𝜆), where 𝑘

𝑓𝑖 (𝑥, 𝜆𝑖 ) = 𝑝𝑖 𝜆𝑖 𝑒

𝜆𝑖 𝑥

; 𝑥 > 0, 𝜆 > 0, 𝑝𝑖 > 0 𝑖 = 1,2, … , 𝑘 𝑎𝑛𝑑 ∑ 𝑝𝑖 = 1 𝑖=1

[120]

Khan and Jan /COMMUNE – 2015

𝑔(𝑦, 𝛼, 𝜆) = 𝜆𝑒 −𝜆(𝑦−𝛼) ∶ 𝑦 > 𝛼 ≥ 0, 𝜆 > 0 For two components k = 2, we have 𝑓(𝑥, 𝜆1 , 𝜆2 ) = 𝑝1 𝜆1 𝑒 𝜆1𝑥 + 𝑝2 𝜆2 𝑒 𝜆2𝑥 ; 𝑥, 𝜆, 𝑝1 , 𝑝2 > 0, In addition, if X and Y are independent, then from (2.1) reliability R is given by

𝑝1 + 𝑝2 = 1

∞ 𝑥

𝑅2 = ∫ ∫(𝜆𝑒 −𝜆(𝑦−𝛼) )(𝑝1 𝜆1 𝑒 𝜆1𝑥 + 𝑝2 𝜆2 𝑒 𝜆2𝑥 ) 𝑑𝑥𝑑𝑦 0 𝛼 2

∞

𝑅2 = ∑ 𝑝𝑖 𝜆𝑖 [∫ 𝑒 𝜆𝑖𝑥 (1 − 𝑒 −𝜆(𝑥−𝛼) )𝑑𝑥 ] 𝑖=1

𝛼

2

𝑅2 = 𝜆 ∑ 𝑝𝑖 𝑖=1

𝑒 −𝛼𝜆𝑖 𝜆 + 𝜆𝑖

For three components k = 3, we have𝑝1 𝜆1 𝑒 𝜆1𝑥 + 𝑝2 𝜆2 𝑒 𝜆2𝑥 + 𝑝3 𝜆3 𝑒 𝜆3𝑥 𝑓(𝑥, 𝜆1 , 𝜆2 , 𝜆3 ) = 𝑝1 𝜆1 𝑒 𝜆1𝑥 + 𝑝2 𝜆2 𝑒 𝜆2𝑥 + 𝑝3 𝜆3 𝑒 𝜆3𝑥 ; 𝑥, 𝜆, 𝑝1 , 𝑝2 , 𝑝3 > 0, 𝑝1 + 𝑝2 + 𝑝3 = 1 In addition, if X and Y are independent, then from (1) reliability R is given by ∞ 𝑥

𝑅3 = ∫ ∫(𝜆𝑒 −𝜆(𝑦−𝛼) )(𝑝1 𝜆1 𝑒 𝜆1𝑥 + 𝑝2 𝜆2 𝑒 𝜆2𝑥 + 𝑝3 𝜆3 𝑒 𝜆3𝑥 ) 𝑑𝑥𝑑𝑦 0 𝛼

3

∞

𝑅3 = ∑ 𝑝𝑖 𝜆𝑖 [∫ 𝑒 𝜆𝑖𝑥 (1 − 𝑒 −𝜆(𝑥−𝛼) )𝑑𝑥 ] 𝑖=1

𝛼

3

𝑅3 = 𝜆 ∑ 𝑝𝑖 𝑖=1

𝑒 −𝛼𝜆𝑖 𝜆 + 𝜆𝑖

In general, for k-components, we have 𝑓(𝑥, 𝜆1 , 𝜆2 , … , 𝜆𝑘 ) = 𝑝1 𝜆1 𝑒 𝜆1𝑥 + 𝑝2 𝜆2 𝑒 𝜆2𝑥 + ⋯ + 𝑝𝑘 𝜆𝑘 𝑒 𝜆𝑘𝑥 ; 𝑥, 𝜆, 𝑝1 , 𝑝2 , … , 𝑝𝑘 > 0, 𝑝1 + 𝑝2 + ⋯ + 𝑝𝑘 = 1 𝑘

and

𝑅𝑘 = 𝜆 ∑ 𝑝𝑖 𝑖=1

𝑒 −𝛼𝜆𝑖 𝜆 + 𝜆𝑖

Special case: When 𝛼 = 0, two-parameter exponential distribution reduces to one parameter distribution. Therefore, 𝑅𝑘 when strength X follows mixture of exponential distribution and stress Y follows exponential distribution is given by 𝑘

𝑘

𝑖=1

𝑖=1

1 𝑝𝑖 𝜆𝑖 𝑅𝑘 = 𝜆 ∑ 𝑝𝑖 =1−∑ 𝜆 + 𝜆𝑖 𝜆 + 𝜆𝑖 (See (Umamaheswan, Sandhya, 2013)) Case II: Mixture of one parameter gamma strength and exponential stress. Let X be the strength of k-components which follows mixture of one parameter gamma distribution with pdf 𝑓𝑖 (𝑥, 𝛽𝑖 ) and Y be the stress which follows exponential distribution with pdf 𝑔(𝑦, 𝜆), where 𝑘

𝑒 −𝑥 𝑥 𝛽𝑖−1 𝑓𝑖 (𝑥, 𝛽𝑖 ) = 𝑝𝑖 ; 𝑥 > 0, 𝛽𝑖 > 0, 𝑝𝑖 > 0 𝑖 = 1,2, … , 𝑘 𝑎𝑛𝑑 ∑ 𝑝𝑖 = 1 Γ(𝛽𝑖 ) 𝑔(𝑦, 𝜆) = 𝜆𝑒 −𝜆𝑦 ∶ 𝑦 > 0, 𝜆 > 0 For two components k = 2, we have 𝑒 −𝑥 𝑥 𝛽1 −1 𝑒 −𝑥 𝑥 𝛽2 −1 𝑓(𝑥, 𝛽1 , 𝛽2 ) = 𝑝1 + 𝑝2 ; 𝑝1 + 𝑝2 = 1 Γ(𝛽1 ) Γ(𝛽2 ) In addition, if X and Y are independent, then from (2.1) reliability R is given by ∞ 𝑥

𝑅2 = ∫ ∫(𝜆𝑒 −𝜆𝑦 ) (𝑝1 0 0

𝑒 −𝑥 𝑥 𝛽1−1 𝑒 −𝑥 𝑥 𝛽2−1 + 𝑝2 ) 𝑑𝑥𝑑𝑦 Γ(𝛽1 ) Γ(𝛽2 )

[121]

𝑖=1

(3.1)

Khan and Jan /COMMUNE – 2015 ∞

2

𝑝𝑖 𝑅2 = ∑ [∫(1 − 𝑒 −𝜆𝑥 )𝑒 −𝑥 𝑥 𝛽𝑖 −1 𝑑𝑥 ] Γ(𝛽𝑖 ) 𝑖=1

0

2

𝑝𝑖 (1 + 𝜆)𝛽𝑖

𝑅2 = 1 − ∑ 𝑖=1

For three components k = 3, we have 𝑓(𝑥, 𝛽1 , 𝛽2 , 𝛽3 ) = 𝑝1

𝑒 −𝑥 𝑥 𝛽1−1 𝑒 −𝑥 𝑥 𝛽2−1 𝑒 −𝑥 𝑥 𝛽3−1 + 𝑝2 + 𝑝3 ; 𝑝1 + 𝑝2 + 𝑝3 = 1 Γ(𝛽1 ) Γ(𝛽2 ) Γ(𝛽3 )

In addition, if X and Y are independent, then from (2.1) reliability R is given by ∞ 𝑥

𝑅3 = ∫ ∫(𝜆𝑒 −𝜆𝑦 ) (𝑝1 0 0

𝑒 −𝑥 𝑥 𝛽1−1 𝑒 −𝑥 𝑥 𝛽1−1 𝑒 −𝑥 𝑥 𝛽3−1 + 𝑝2 + 𝑝3 ) 𝑑𝑥𝑑𝑦 Γ(𝛽1 ) Γ(𝛽1 ) Γ(𝛽3 ) ∞

3

𝑝𝑖 𝑅3 = ∑ [∫(1 − 𝑒 −𝜆𝑥 )𝑒 −𝑥 𝑥 𝛽𝑖 −1 𝑑𝑥 ] Γ(𝛽𝑖 ) 𝑖=1

0

3

𝑝𝑖 (1 + 𝜆)𝛽𝑖

𝑅3 = 1 − ∑ 𝑖=1

In general, for k-components, we have 𝑒 −𝑥 𝑥 𝛽1−1 𝑒 −𝑥 𝑥 𝛽2−1 𝑒 −𝑥 𝑥 𝛽𝑘 −1 𝑓(𝑥, 𝛽1 , 𝛽2 , … , 𝛽𝑘 ) = 𝑝1 + 𝑝2 + ⋯ + 𝑝𝑘 ; 𝑝1 + 𝑝2 + ⋯ + 𝑝𝑘 = 1 Γ(𝛽1 ) Γ(𝛽2 ) Γ(𝛽𝑘 ) 𝑘

and

𝑅𝑘 = 1 − ∑ 𝑖=1

𝑝𝑖 (1 + 𝜆)𝛽𝑖

Case III: Mixture of two parameter gamma strength and exponential stress. Let X be the strength of k-components which follows mixture of two parameter gamma distribution with pdf 𝑓𝑖 (𝑥, 𝜆𝑖 , 𝛽𝑖 ) and Y be the stress which follows two parameter exponential distribution with pdf (3.1), where 𝑘

𝜆𝑖 𝛽𝑖 𝑒 −𝜆𝑖𝑥 𝑥 𝛽𝑖 −1 𝑓𝑖 (𝑥, 𝜆𝑖 , 𝛽𝑖 ) = 𝑝𝑖 ; 𝑥 > 0, 𝜆𝑖 , 𝛽𝑖 , 𝑝𝑖 > 0 𝑖 = 1,2, … , 𝑘 𝑎𝑛𝑑 ∑ 𝑝𝑖 = 1 Γ(𝛽𝑖 ) 𝑖=1

For two components k = 2, we have 𝑓(𝑥, 𝜆1 , 𝜆2 , 𝛽1 , 𝛽2 ) = 𝑝1

𝜆1 𝛽1 𝑒 −𝜆1 𝑥 𝑥 𝛽1−1 𝜆2 𝛽2 𝑒 −𝜆2𝑥 𝑥 𝛽2 −1 + 𝑝2 ; 𝑝1 + 𝑝2 = 1 Γ(𝛽1 ) Γ(𝛽2 )

In addition, if X and Y are independent, then from (1) reliability R is given by ∞ 𝑥

𝑅2 = ∫ ∫(𝜆𝑒 −𝜆𝑦 ) (𝑝1 0 0

𝜆1 𝛽1 𝑒 −𝜆1 𝑥 𝑥 𝛽1−1 𝜆2 𝛽2 𝑒 −𝜆2𝑥 𝑥 𝛽2 −1 + 𝑝2 ) 𝑑𝑥𝑑𝑦 Γ(𝛽1 ) Γ(𝛽2 ) ∞

2

𝑝𝑖 𝜆𝑖 𝛽𝑖 𝑅2 = ∑ [∫ (1 − 𝑒 −𝜆𝑥 )𝑒 −𝜆𝑖𝑥 𝑥 𝛽𝑖−1 𝑑𝑥 ] Γ(𝛽𝑖 ) 𝑖=1

0

2

𝑅2 = 1 − ∑ 𝑖=1

𝑝𝑖 𝜆𝑖 𝛽𝑖 (𝜆 + 𝜆𝑖 )𝛽𝑖

For three components k = 3, we have

𝜆1 𝛽1 𝑒 −𝜆1𝑥 𝑥 𝛽1−1 𝜆2 𝛽2 𝑒 −𝜆2𝑥 𝑥 𝛽2 −1 𝜆3 3 𝑒 −𝜆3 𝑥 𝑥 𝛽3−1 + 𝑝2 + 𝑝3 Γ(𝛽1 ) Γ(𝛽2 ) Γ(𝛽3 ) ; 𝑝1 + 𝑝2 + 𝑝3 = 1 In addition, if X and Y are independent, then from (1) reliability R is given by 𝑓(𝑥, 𝜆1 , 𝜆2 , 𝜆3 , 𝛽1 , 𝛽2 , 𝛽3 ) = 𝑝1

∞ 𝑥

𝑅3 = ∫ ∫(𝜆𝑒 −𝜆𝑦 ) (𝑝1 0 0

𝜆1 𝛽1 𝑒 −𝜆1𝑥 𝑥 𝛽1−1 𝜆2 𝛽2 𝑒 −𝜆2𝑥 𝑥 𝛽2 −1 𝜆3 3 𝑒 −𝜆3𝑥 𝑥 𝛽3−1 + 𝑝2 + 𝑝3 ) 𝑑𝑥𝑑𝑦 Γ(𝛽1 ) Γ(𝛽2 ) Γ(𝛽3 ) 3

∞

𝑝𝑖 𝜆𝑖 𝛽𝑖 𝑅3 = ∑ [∫ (1 − 𝑒 −𝜆𝑥 )𝑒 −𝜆𝑖𝑥 𝑥 𝛽𝑖−1 𝑑𝑥 ] Γ(𝛽𝑖 ) 𝑖=1

0

[122]

Khan and Jan /COMMUNE – 2015 3

𝑅3 = 1 − ∑ 𝑖=1

In general, for k-components, we have 𝑓(𝑥, 𝜆1 , … , 𝜆𝑘 , 𝛽1 , … , 𝛽𝑘 ) =

𝑝1 𝜆1 𝛽1 𝑒 −𝜆1 𝑥 𝑥 𝛽1−1 𝑝2 𝜆2 𝛽2 𝑒 −𝜆2𝑥 𝑥 𝛽2−1 𝑝𝑘 𝜆𝑘 𝑘 𝑒 −𝜆𝑘 𝑥 𝑥 𝛽𝑘 −1 + + ⋯+ ; Γ(𝛽1 ) Γ(𝛽2 ) Γ(𝛽𝑘 ) 𝑝1 + 𝑝2 + ⋯ + 𝑝𝑘 = 1 𝑘

and

𝑝𝑖 𝜆𝑖 𝛽𝑖 (𝜆 + 𝜆𝑖 )𝛽𝑖

𝑅𝑘 = 1 − ∑ 𝑖=1

𝑝𝑖 𝜆𝑖 𝛽𝑖 (𝜆 + 𝜆𝑖 )𝛽𝑖

Case IV: Mixture of two-parameter gamma strength and mixture of exponential stress. Let X be the strength of k-components which follows mixture of two parameter gamma distribution with pdf 𝑓𝑖 (𝑥, 𝜆𝑖 , 𝛽𝑖 ) and Y be the stress which follows two parameter exponential distribution with pdf 𝑔(𝑦, 𝜆𝑗 ), where 𝑓𝑗 (𝑥, 𝜆𝑗 , 𝛽𝑗 ) = 𝑝𝑗

𝜆𝑗 𝛽𝑗 𝑒 −𝜆𝑗 𝑥 𝑥 𝑗−1 Γ(𝛽𝑗 )

2𝑘

; 𝑥 > 0, 𝜆𝑗 , 𝛽𝑗 , 𝑝𝑗 > 0 𝑗 = 1,2, … , 𝑘 𝑎𝑛𝑑 ∑ 𝑝𝑗 = 1 𝑘

𝑗=𝑘+1

𝑔𝑖 (𝑦, 𝜆𝑖 ) = 𝑝𝑖 𝜆𝑖 𝑒 −𝜆𝑖𝑦 ∶ 𝑥 > 0, 𝜆𝑖 > 0 𝑎𝑛𝑑 ∑ 𝑝𝑖 = 1 𝑖=1

For two components k = 2, we have 𝑔(𝑦, 𝜆1 , 𝜆2 ) = 𝑝1 𝜆1 𝑒 −𝜆1𝑦 + 𝑝2 𝜆2 𝑒 −𝜆2𝑦 ; 𝑝1 + 𝑝2 = 1 𝑓(𝑥, 𝜆3 , 𝜆4 , 𝛽3 , 𝛽4 ) = 𝑝3

𝜆3 𝛽3 𝑒 −𝜆3 𝑥 𝑥 𝛽3−1 𝜆4 𝛽4 𝑒 −𝜆4𝑥 𝑥 𝛽4−1 + 𝑝4 ; 𝑝3 + 𝑝4 = 1 Γ(𝛽3 ) Γ(𝛽4 )

In addition, if X and Y are independent, then from (1) reliability R is given by ∞ 𝑥

𝑅2 = ∫ ∫(𝑝1 𝜆1 𝑒 −𝜆1𝑦 + 𝑝2 𝜆2 𝑒 −𝜆2𝑦 ) (𝑝3 0 𝛼

4

2

𝑅2 = ∑ ∑ 𝑗=𝑖+2 𝑖=1

𝑝𝑖 𝑝𝑗 𝜆𝑗 𝛽𝑗 Γ(𝛽𝑗 ) 4

𝜆3 𝛽3 𝑒 −𝜆3𝑥 𝑥 𝛽3 −1 𝜆4 𝛽4 𝑒 −𝜆4 𝑥 𝑥 𝛽4−1 + 𝑝4 ) 𝑑𝑥𝑑𝑦 Γ(𝛽3 ) Γ(𝛽4 )

∞

[∫ (1 − 𝜆𝑖 𝑒 −𝜆𝑖𝑥 )𝑒 −𝜆𝑗 𝑥 𝑥 𝛽𝑗 −1 𝑑𝑥 ] 0

2

𝑅2 = 1 − ∑ ∑ 𝑝𝑖 𝑝𝑗 𝑗=𝑖+2 𝑖=1

𝜆𝑗 𝛽𝑗 𝛽𝑗

(𝜆𝑖 + 𝜆𝑗 )

For three components k = 3, we have 𝑔(𝑥, 𝜆1 𝜆2 , 𝜆3 , ) = 𝑝1 𝜆1 𝑒 −𝜆1𝑦 + 𝑝2 𝜆2 𝑒 −𝜆2𝑦 + 𝑝3 𝜆3 𝑒 −𝜆3𝑦 ;

𝑝1 + 𝑝2 + 𝑝2 = 1 𝜆3 𝛽3 𝑒 −𝜆3𝑥 𝑥 𝛽3−1 𝜆4 𝛽4 𝑒 −𝜆4𝑥 𝑥 𝛽4−1 𝜆5 𝛽5 𝑒 −𝜆5𝑥 𝑥 𝛽5−1 𝑓(𝑥, 𝜆3 , 𝜆4 , 𝜆5 , 𝛽3 , 𝛽4 , 𝛽5 ) = 𝑝3 + 𝑝4 + 𝑝5 ; Γ(𝛽3 ) Γ(𝛽4 ) Γ(𝛽5 ) 𝑝3 + 𝑝4 + 𝑝5 = 1 In addition, if X and Y are independent, then from (1) reliability R is given by ∞ 𝑥

𝑅3 = ∫ ∫(𝑝1 𝜆1 𝑒 −𝜆1𝑦 + 𝑝2 𝜆2 𝑒 −𝜆2𝑦 + 𝑝3 𝜆3 𝑒 −𝜆3𝑦 ) (𝑝3 0 0

+ 𝑝5

𝜆3 𝛽3 𝑒 −𝜆3 𝑥 𝑥 𝛽3−1 𝜆4 𝛽4 𝑒 −𝜆4𝑥 𝑥 𝛽4−1 + 𝑝4 Γ(𝛽3 ) Γ(𝛽4 )

𝜆5 𝛽5 𝑒 −𝜆5𝑥 𝑥 𝛽5 −1 ) 𝑑𝑥𝑑𝑦 Γ(𝛽5 ) 6

3

𝑅3 = ∑ ∑ 𝑗=𝑖+2 𝑖=1

𝑝𝑖 𝑝𝑗 𝜆𝑗 𝛽𝑗 Γ(𝛽𝑗 ) 6

∞

[∫ (1 − 𝜆𝑖 𝑒 −𝜆𝑖𝑥 )𝑒 −𝜆𝑗 𝑥 𝑥 𝛽𝑗 −1 𝑑𝑥 ] 0

3

𝑅3 = 1 − ∑ ∑ 𝑝𝑖 𝑝𝑗 𝑗=𝑖+2 𝑖=1

In general, for k-components, we have

[123]

𝜆𝑗 𝛽𝑗 𝛽𝑗

(𝜆𝑖 + 𝜆𝑗 )

Khan and Jan /COMMUNE – 2015

𝑔(𝑥, 𝜆1 , … , 𝜆𝑘 ) = 𝑝1 𝜆1 𝑒 −𝜆1𝑦 + 𝑝2 𝜆2 𝑒 −𝜆2𝑦 + ⋯ + 𝑝𝑘 𝜆𝑘 𝑒 −𝑘𝑦 ;

𝑝1 + 𝑝2 + ⋯ + 𝑝𝐾 = 1 𝑓(𝑥, 𝜆𝑘+1 , … , 𝜆2𝑘 , 𝛽𝑘+1 , … , 𝛽2𝑘 ) 𝜆𝑘+1 𝛽𝑘+1 𝑒 −𝜆𝑘+1𝑥 𝑥 𝛽𝑘+1−1 𝜆𝑘+2 𝛽𝑘+2 𝑒 −𝜆𝑘+2 𝑥 𝑥 𝛽𝑘+2−1 = 𝑝𝑘+1 + 𝑝𝑘+2 +⋯ Γ(𝛽𝑘+1 ) Γ(𝛽𝑘+2 ) 𝜆2𝑘 𝛽2𝑘 𝑒 −𝜆2𝑘 𝑥 𝑥 𝛽2𝑘 −1 + 𝑝2𝑘 ; 𝑝𝑘+1 + 𝑝𝑘+2 + ⋯ + 𝑝2𝑘 = 1 Γ(𝛽2𝑘 ) ∞ 𝑥

𝑅𝑘 = ∫ ∫(𝑝1 𝜆1 𝑒 −𝜆1𝑦 + 𝑝2 𝜆2 𝑒 −𝜆2 𝑦 + ⋯ + 𝑝𝑘 𝜆𝑘 𝑒 −𝑘𝑦 ) (𝑝𝑘+1 0 0

+ 𝑝𝑘+2

𝜆𝑘+1 𝛽𝑘+1 𝑒 −𝜆𝑘+1𝑥 𝑥 𝛽𝑘+1−1 Γ(𝛽𝑘+1 )

𝜆𝑘+2 𝛽𝑘+2 𝑒 −𝜆𝑘+2 𝑥 𝑥 𝛽𝑘+2−1 𝜆2𝑘 𝛽2𝑘 𝑒 −𝜆2𝑘 𝑥 𝑥 𝛽2𝑘 −1 + ⋯ + 𝑝2𝑘 ) 𝑑𝑥𝑑𝑦 Γ(𝛽𝑘+2 ) Γ(𝛽2𝑘 ) 2𝑘

𝑘

𝑅𝑘 = 1 − ∑ ∑ 𝑝𝑖 𝑝𝑗 𝑗=𝑖+2 𝑖=1

𝜆𝑗 𝛽𝑗 (𝜆𝑖 + 𝜆𝑗 )

𝛽𝑗

4. Numerical Evaluation For some specific values of the parameters involved in the expression 𝑅𝑘 (𝑘 = 2), we have evaluated the system reliability for different cases of Exponential, Gamma distributions from their expression obtained in case 1, 2, 3 and 4. Mixture of one parameter exponential strength and two parameter exponential stress Table 1 𝜆 0.8 0.8 0.8 0.8 0.8 0.8 0.8

𝜆1 = 𝜆2 0 0.2 0.4 0.6 0.8 1 1.2

𝛼 0.2 0.2 0.2 0.2 0.2 0.2 0.2

Table 2 𝑅 1 0.7686 0.6154 0.5068 0.4260 0.3638 0.3146

𝜆 0.8 0.8 0.8 0.8 0.8 0.8 0.8

𝜆1 = 𝜆2 0.3 0.3 0.3 0.3 0.3 0.3 0.3

Table 3

𝛼 0 0.2 0.4 0.6 0.8 1 1.2

𝑅 0.7272 0.6849 0.6450 0.6074 0.5720 0.5387 0.5074

𝜆 0 0.2 0.4 0.6 0.8 1 1.2

𝜆1 = 𝜆2 0.3 0.3 0.3 0.3 0.3 0.3 0.3

𝛼 0.2 0.2 0.2 0.2 0.2 0.2 0.2

𝑅 0 0.3767 0.5381 0.6278 0.6849 0.7244 0.7534

Mixture of one parameter gamma strength and exponential stress Table 4

Table 5

𝝀

𝜷𝟏 = 𝜷𝟐

𝑹

𝝀

𝜷𝟏 = 𝜷𝟐

𝑹

0.7 0.7 0.7 0.7 0.7 0.7 0.7

0 0.2 0.4 0.6 0.8 1 1.2

0 0.1006 0.1912 0.2726 0.3459 0.4117 0.4709

0 0.2 0.4 0.6 0.8 1 1.2

0.7 0.7 0.7 0.7 0.7 0.7 0.7

0 0.1198 0.2098 0.2803 0.3373 0.3844 0.4241

Mixture of two parameter gamma strength and exponential stress Table 6

Table 7

Table 8

𝝀

𝝀𝟏 = 𝝀𝟐

𝜷𝟏 = 𝜷𝟐

𝑹

𝝀

𝝀𝟏 = 𝝀𝟐

𝜷𝟏 = 𝜷𝟐

𝑹

𝝀

𝝀𝟏 = 𝝀𝟐

𝜷𝟏 = 𝜷𝟐

𝑹

0 0.2 0.4 0.6 0.8 1 1.2

0.4 0.4 0.4 0.4 0.4 0.4 0.4

0.7 0.7 0.7 0.7 0.7 0.7 0.7

0 0.2471 0.3844 0.4734 0.5365 0.5839 0.6210

0.3 0.3 0.3 0.3 0.3 0.3 0.3

0.6 0.6 0.6 0.6 0.6 0.6 0.6

0 0.2 0.4 0.6 0.8 1 1.2

0 0.0778 0.1497 0.2159 0.2770 0.3333 0.3852

0.3 0.3 0.3 0.3 0.3 0.3 0.3

0 0.2 0.4 0.6 0.8 1 1.2

0.5 0.5 0.5 0.5 0.5 0.5 0.5

1 0.3675 0.2440 0.1835 0.1471 0.1229 0.1055

[124]

Khan and Jan /COMMUNE – 2015

Mixture of two parameter gamma strength and mixture of exponential stress Table 9

Table 10

Table 11

𝝀𝟏 = 𝝀𝟐

𝝀𝟑 = 𝝀𝟒

𝜷𝟑 = 𝜷𝟒

𝑹

𝝀𝟏 = 𝝀𝟐

𝝀𝟑 = 𝝀𝟒

𝜷𝟑 = 𝜷𝟒

𝑹

𝝀𝟏 = 𝝀𝟐

𝝀𝟑 = 𝝀𝟒

𝜷𝟑 = 𝜷𝟒

𝑹

0 0.2 0.4 0.6 0.8 1 1.2

0.2 0.2 0.2 0.2 0.2 0.2 0.2

0.6 0.6 0.6 0.6 0.6 0.6 0.6

0 0.3402 0.4827 0.5647 0.6192 0.6587 0.6888

0.2 0.2 0.2 0.2 0.2 0.2 0.2

0 0.2 0.4 0.6 0.8 1 1.2

0.6 0.6 0.6 0.6 0.6 0.6 0.6

1 0.3402 0.2159 0.1585 0.1253 0.1036 0.0883

0.2 0.2 0.2 0.2 0.2 0.2 0.2

0.6 0.6 0.6 0.6 0.6 0.6 0.6

0 0.2 0.4 0.6 0.8 1 1.2

0 0.0559 0.1086 0.1585 0.2055 0.2500 0.2919

5. Conclusion In this paper Stress –Strength reliability is considered where various cases have been considered for stress (Y) and strength (X) variables viz., the strength follows finite mixture of one parameter exponential and one, two parameter Gamma distributions and stress follows one and two parameter exponential distribution. From Table 1, it may be observed that reliability decreases when both the stress parameters are kept constant and strength variable is increased, in table 2, reliability increases if one of the stress variable 𝛼 is increased (𝜆 fixed) and strength variables 𝜆1 , 𝜆2 are constants. Similarly, from table 3, reliability increases with increase in stress variable 𝜆 (𝛼 constant) and strength being constant. Table 4 and Table 5 exhibits that reliability increases in both the cases i.e. if we increase the stress or strength variables reliability increases. For example, in Table 4 if we increase the strength variables from 0 upto 1.2 the reliability increases from 0 upto 0.4709. Table 6 reveals that if we increase the stress and keep the strength constant reliability increases from 0 upto 0.6210. Again, with increase in one of the strength variable (𝛽1 = 𝛽2 ) keeping stress constant, reliability increases, but in table 8 reliability decreases when stress is constant and strength variable (𝜆1 = 𝜆2 ) is increased. Table 9 exhibits that if we increase the stress and keep strength constant them the reliability of the system increases. Similarly, in table 11 reliability increases but this time stress is constant and one of the strength variables 𝛽3 = 𝛽4 (𝜆3 = 𝜆4 , constant). And in table 10 reliability decreases with increase in strength variable and keeping stress constant, for example, if we increase 𝜆3 = 𝜆4 from 0 upto 1.2 reliability decreases from 1 upto 0.0883. References Awad, A. M. and Gharraf, M. K., 1986. Estimation of (P(Y < X) in the Burr case: A comparative study, Commun. Statist. Simul. Comp., 15(2), p-389403. Beg, M. A.and Singh, N., 1979. Estimation of P(Y < X) for the pareto distribution, IEEE Trans. Reliab., 28(5), p- 411-414. Chaturvadi, A., Tiwari, N. and Kumar, S., 2007. Some remarks on classical and Bayesian reliability estimation of binomial and Poisson distributions, Statistical papers, 48, p-683-693. Chaturvedi, A., and Tomer, S. K., 2002. Classical and Bayesian Reliability estimation of the negative binomial distribution, Jour. Applied Statist. Sci., 11(1), p-33-43. Church, J. D. and Harris, B., 1970. The estimation of reliability from stress strength relationships, Technometrics, 12, p-49-54. Gogoi J., and Borah, M., 2012. Estimation of Reliability for Multi-Component Systems Using Exponential Gamma and Lindley Stress-Strength Distributions, Journal of Reliability and Statistics Studies, 5(1), p-33-41. Gogoi, J., Borah, M., and Sriwastav, G. L., 2010. An Interference Model with Number of Stresses a Poisson Process, IAPQR Transactions, 34(2), p139-152. Khan, Adil H., and. Jan, T. R., 2014a. Estimation of Multi Component Systems Reliability in Stress-Strength Models, Journal of Modern Applied Statistical Methods. 13(2), Article 21. Khan, Adil H., and. Jan, T. R., 2014b. Reliability Estimates of Generalized Poisson Distribution and Generalized Geometric Series Distribution, Journal of Modern Applied Statistical Methods. 13(2), Article 20. Khan, Adil H., and. Jan, T. R., 2014c. On estimation of reliability function of Consul and Geeta distributions, International Journal of Advanced Scientific and Technical Research. 4(4), p-96-105. Kotz, S., Lumelskii, Y. and Pensky, M., 2003. The Stress-Strength Model and its Generalizations: Theory and Applications, World Scientific Publishing, Singapore. Krishnamoorthy, K., Mukherjee, Shubhabarata and Guo, Huizhen, 2007. Inference on Reliability in two parameter exponential stress-strength model, Metrica, 65(3), p-261-273. Kundu, D. and Gupta, R.D., 2005. Estimation of P[Y < X] for generalized exponential distribution, Metrika, 61, p-291–308. Kundu, D. and Gupta, R.D., 2006. Estimation of R = P[Y < X] for Weibull distributions, IEEE Transactions on Reliability, 55, p-270–280. Raqab, M. Z. and Kundu, D., 2005. Comparison of Different Estimators of P[Y < X] for a Scaled Burr Type X Distribution, Communications in Statistics-Simulation and Computation, 34, p-465–483. Maiti, S. S., 1995. Estimation of Pr{X ≤ Y } in the Geometric case, Jour. Indian Statist. Assoc., 33, p- 87-91. Sandhya, K. and Umamaheswari, T. S., 2013. Estimation of Stress- Strength Reliability model using finite mixture of exponential distributions, International Journal of Computational Engineering Research, 3(11), p-39-46. Woodward, W. A.and Kelley, G. D., 1977. Minimum variance unbiased estimation of P(Y < X) in the normal case, Technometrics, 19, p-95-98.

[125]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Design of XOR Gate using Floating-Gate MOSFET Roshani Gupta*, Rockey Gupta, Susheel Sharma Department of Physics & Electronics, University of Jammu, Jammu, India

Abstract This paper presents the design of XOR gate using floating-gate MOSFET that is widely used technique for the design of mixed-mode circuits due to its unique feature of threshold voltage tunability through a bias voltage. The performance of CMOS XOR gate has been compared with FGMOS based XOR gate. It has been observed that by varying the bias voltages in FGMOS, the transient characteristics of XOR gate can be altered that results in less propagation delay and energy delay product as compared to CMOS XOR gate. The performance of these circuits has been verified through PSpice simulations carried out using level 7 parameters in 0.13 µm CMOS technology with a supply voltage of 1V.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Floating Gate MOSFET; CMOS; XOR Gate; Transient Characteristics; Propagation Delay; Energy Delay Product

1. Introduction The challenge of designing high performance low voltage and low power digital circuits is increasing due to the scaling down of CMOS technology and the increasing demand for portable electronic equipments. The general trend in CMOS technology is to make the devices smaller and smaller to increase the density and speed of digital circuits. It is also common to reduce the thickness of the gate oxide in order to increase the driving capability of the transistor. In addition, the thickness reduction implies that the supply voltage must be decreased to avoid excessive electric field in the devices as indicated by Chandrakasan et al., 1992 or Gonzalez et al., 1997 or Fayomi et al., 2004. The speed of conventional digital integrated circuits is degrading on reducing the supply voltage for a given technology. To fulfill these requirements, there is a need of development of new integrated circuits that have low voltage supply requirement, without any degradation in the performance. Yan et al., 2000 in his paper proposed various techniques for implementing low voltage and low power applications like Sub-threshold operation of MOSFET, Bulk-driven MOSFET, Level shifter technique, Self-cascode MOSFET, Floating-gate MOSFET and Quasi-floating-gate MOSFET. The Floating-gate MOSFET (FGMOS) proves to be a suitable element for low power applications as it allows programmability of threshold voltage and results in lowering of the threshold voltage below its conventional value. In this paper, we have employed floating-gate MOSFET (FGMOS) to implement XOR gate and compared its performance with its CMOS counter part. The performance of these circuits has been verified through PSpice simulations carried out using level 7 parameters in 0.13 µm CMOS technology with a supply voltage of 1 V. 2. Floating-Gate MOS Transistor Floating-Gate MOS transistor (FGMOS) is a modified form of conventional MOSFET where extra capacitances are introduced between the conventional gate and the multi-input signal gates as shown in Fig. 1. By applying a bias voltage on one of the input gates, the threshold voltage of FGMOS can be changed. FGMOS can be fabricated using a standard CMOS process by electrically isolating the gate of a standard MOSFET, so that there are no resistive

* Corresponding author. Tel.: +91 9796 264186. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Gupta et al/ COMMUNE– 2015

connections to its gate. A number of secondary gates or inputs are then deposited above the floating-gate (FG) which are electrically isolated from it and are only capacitively connected to FG. Since FG is completely surrounded by highly resistive material, so for DC operation, FG acts as floating node according to (Villegas, 2006), (Gupta et al., 2010), (Hasler et al., 2001), (Anand et al., 2013), (Keles et al., 2009) or (Murthy et al., 2011). VD CGD

V1

CGB

C1 FG

V2

VB

C2 CGS

VN CN

VS

Fig. 1 Floating-gate MOSFET

3. XOR Gate The XOR circuit is basic building block in various circuits, especially arithmetic circuits such as adders and multipliers, compressors, comparators, parity checkers, code converters, error-detecting or error-correcting codes and phase detectors according to (Keles et al., 2010) or (Wairya et al., 2012). The XOR (exclusive-OR) gate acts in the same way as the logical "either/or". In XOR gate, the output is high only if either, but not both of the inputs are one and the output is low if both inputs are one or if both inputs are zero i.e. the output becomes high if the inputs are different and low if the inputs are same. The circuit for CMOS XOR gate is shown in Fig. 2. VDD

A'

M1

A

M2

B

M3

B'

M4

Vout A

M5

A'

M6

B

M7

B'

M8

Fig. 2 XOR Gate

The performance of XOR gate can be characterized through its transient response which is a plot of input and output voltage with respect to time. Transient response is important to determine the maximum speed at which the device can be operated. It is measured between the 50% transition points of the input and output waveforms and is given by Hodges et al., 2005 or Razavi, 2008 as:

tP 

t plh  t phl

(1)

2

Where tplh defines the response time of the gate for a low to high output transition and tphl refers to the response time for a high to low output transition. The circuit of XOR gate has been simulated to obtain its transient characteristics by selecting W/L of p-channel MOSFETs as 2.6 μm/0.13 μm and n-channel MOSFETs as 1.3 μm/0.13 μm with the supply voltage of 1 V as shown in Fig. 3. From the simulation results, the propagation delay of XOR gate is calculated to be 0.3 ns.

[127]

Gupta et al/ COMMUNE– 2015

Vout (Volts)

1.2 1 0.8 0.6 0.4 0.2 0 0

1

2

3

4

5

Time (ns) Fig. 3 Transient characteristics of XOR gate

4. XOR Gate using FGMOS For high-speed digital circuits, logic gates must introduce a minimum amount of delay when inputs change. In order to enhance the speed of XOR gate, it is important to minimize the propagation delay. The propagation delay of XOR gate can be reduced by implementing the circuit using floating-gate MOS transistors (FGMOS) as shown in Fig. 4. The circuit is similar to CMOS XOR gate except that extra capacitances are introduced between the conventional gate and the input signal gate. The bias voltages Vbp and Vbn provide tunability to the threshold voltages of p and n-channel FGMOS transistors respectively. VDD

A'

C1 M1

A

C2 C3

Vbp B

A

B

M2 C6 C7

M3 C4 C9

B'

A'

C10 C11

Vbn

C5

M5 M7

C12

M4 C8

C14 C15 B'

Vout

C13 M6 M8 C16

Fig. 4 XOR gate using FGMOS

The simulation of circuit shown in Fig. 4 has been performed to obtain its transient response at different values of Vbp and Vbn with supply voltage of 1 V. Fig. 5 and 6 shows how propagation delay varies with bias voltage. In Fig. 5, bias voltage of p-channel FGMOS transistors (Vbp) is varied from 0 V to 1 V, while keeping bias voltage of n-channel FGMOS transistors (Vbn) fixed at 1 V. Similarly in Fig. 6, Vbn is varied from 0 V to 1 V, while keeping Vbp fixed at 0 V and output voltage (Vout) is obtained with respect to time. 1.2

Vout (Volts)

1 Vbp=0V Vbp=0.2V Vbp=0.4V Vbp=0.6V Vbp=0.8V Vbp=1V

0.8 0.6 0.4 0.2 0 0

1

2

3

Time (ns) Fig. 5 Transient response of XOR gate using FGMOS at different Vbp

[128]

4

5

Gupta et al/ COMMUNE– 2015 1.2

Vout (Volts)

1 Vbn=0V

0.8

Vbn=0.2V

0.6

Vbn=0.4V

Vbn=0.6V

0.4

Vbn=0.8V 0.2

Vbn=1V

0

0

1

2

3

4

5

Time (ns) Fig. 6 Transient response of XOR gate using FGMOS at different Vbn

Now, from the transient responses shown in Figs. 5 and 6, we have calculated the propagation delay at different bias voltages. The variation of propagation delay as a function of bias voltage is shown in Fig. 7. 0.6

Delay (ns)

0.5

Vbn

Vbp

0.4 0.3 0.2 0.1 0 0

0.2

0.4

0.6

0.8

1

Bias voltage (Volts)

Fig. 7 Propagation delay at different values of Vbp and Vbn

From the above results, it has been found that as we go on increasing the bias voltage of p-channel FGMOS transistor from 0 V to 1 V propagation delay increases from 0.18 ns to 0.57 ns, where as increasing bias voltage of nchannel FGMOS transistor from 0 V to 1 V reduces propagation delay from 0.39 ns to 0.18 ns. Thus, the appropriate selection of bias voltages of n and p-channel FGMOS transistors decreases the propagation delay, thus enhancing the operating speed. Now, the comparative transient characteristics of CMOS and FGMOS XOR gate have been obtained by selecting same W/L of M1, M2, M3 and M4 as 2.6 μm/0.13 μm and M5, M6, M7 and M8 as 1.3 μm/0.13 μm while keeping bias voltages of p and n-channel FGMOS transistors fixed i.e. Vbp = 0 V and Vbn = 1 V with supply voltage of 1 V and are shown in Fig. 8. From the simulation results, it has been found that FGMOS based XOR gate has propagation delay of 0.2 ns which is less as compared to CMOS XOR gate (tp = 0.3 ns).

Vout (Volts)

1.2 1

0.8

CMOS XOR

0.6

FGMOS XOR

0.4

0.2 0 0

1

2

3

4

5

Time (ns)

Fig. 8 Comparative transient response of XOR gate using CMOS and FGMOS

Now, the values of propagation delay obtained from the transient responses of XOR gate using CMOS and FGMOS has been used to calculate the energy delay product (EDP) at different values of supply voltage shown in Fig. 9. Energy delay product represents the trade off between power and the speed. Thus, FGMOS based XOR gate will exhibit more speed and dissipate less power as compared to CMOS XOR gate.

[129]

EDP (Js)*10 -23

Gupta et al/ COMMUNE– 2015

1.4 1.2 1 0.8 0.6 0.4 0.2 0

CMOS XOR FGMOS XOR

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

1

VDD (Volts) Fig. 9 Comparative EDPs of XOR gate using CMOS and FGMOS

From the graph shown above, it has been found that energy delay product varies with supply voltage and XOR gate implemented using FGMOS is better since the energy delay product is lower than CMOS XOR gate. 5. Conclusions In this paper, we have discussed the transient characteristics of XOR gate using CMOS as well as FGMOS. The performance of CMOS XOR gate has been compared with FGMOS. We have observed that variation in bias voltages of p and n-channel FGMOS transistors results in less propagation delay and energy delay product as compared to CMOS XOR gate. The performance of these circuits has been verified through PSpice simulations carried out using level 7 parameters in 0.13 µm CMOS technology with a supply voltage of 1V. References Chandrakasan, A. P., Sheng, S., Brodersen, R. W., 1992. Low-Power CMOS Digital Design, IEEE JSSC 27, p. 473. Gonzalez, R., Gordon, B. M., Horowitz, M. A., 1997. Supply and Threshold Voltage Scaling for Low Power CMOS, IEEE Journal of Solid-State circuits 32, p. 1210. Fayomi, C. J. B., Sawan, M., Roberts, G. W., 2004. Reliable Circuit Techniques for Low Voltage Analog Design in Deep Sub micron Standard CMOS: A Tutorial, Analog Integr. Circuits Signal Proc. 39, p. 21. Yan, S., Sinencio, E. S., 2000. Low voltage analog circuit design techniques: A Tutorial, IEICE Trans. Fundamentals E00-A, p. 1. Villegas, E. R., 2006. Low power and Low voltage circuit design using FGMOS transistor, IET Circuits, Devices and Systems series 20. Gupta, M., Pandey, R., 2010. FGMOS based voltage-controlled resistor and its applications, Microelectronics Journal 41, p. 25. Hasler, P., Lande, T. S., 2001. Overview of floating-gate devices, circuits and systems, IEEE Transactions on Circuits and Systems II: Analog and Digital Signal Processing 48, p. 1. Anand, A., Mandal, S. K., Dash, A., Patro, B. S., 2013. FGMOS based low-voltage low-power high output impedance regulated cascode current mirror, International Journal of VLSI design & Communication Systems 4, p. 39. Keles, S., Kuntman, H. H., 2009. Four Quadrant FGMOS Multiplier, Proceedings of ELECO’: The 6th International Conference on Electrical and Electronics Engineering 2, p. 45–48. Murthy, P. H. S. T., Chaitanya, K., Krishna, M. M., Rao, M., 2011. FTL based 4Stage CLA Adder Design with Floating Gates, International Journal of Computer Applications 17, p. 1. Keles, F., Yildirim, T., 2010. Low voltage low power neuron circuit design based on subthreshold FGMOS transistors and XOR implementation, 11th International Workshop on Symbolic and Numerical Methods Modeling and Applications to Circuit Design, p. 1. Wairya, S., Nagaria, R. K., and Tiwari, S., 2012. Comparative Performance Analysis of XOR XNOR Function based high-speed CMOS Full Adder Circuits for low voltage VLSI Design, International Journal of VLSI design & Communication Systems (VLSICS) 3, p. 221. Hodges, D. A., Jackson, H. G., Saleh, R. A., 2005. Analysis and Design of Digital Integrated circuits, MC Graw-Hill. Razavi, B., 2008. Fundamentals of Microelectronics, John Wiley and sons.

[130]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Cellular Automata: Evolution and Parallel Dimensions Shah Jahan Wania*, M. A. Peera, K. A. Khanb a

Department of Computer Sciences, University of Kashmir, Srinagar, India b Govt. Degree College, Beerwah, Budgam, India

Abstract Cellular Automata rules producing evolution type phenomena have been used for a wide range of applications. Various models have been designed and explored for different applications. Although the strength of its parallelism has been felt by various researchers but its exploration for applications will not minimize the hardware but also maximize the optimum strength of processors. Our present study was intended to identify the additive 2D Cellular Automata linear rules on the quality of pattern evolution and the periodic parallelism utilization. We have made an analysis of 2DCA linear game of life (GOL) rule in Neumann neighborhood pattern evolution and observed pattern multiplication in the process. The results achieved will not only minimize the required hardware for parallel channel creation but also expand the microcomputer processing horizon.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Cellular Automata; Boundary Conditions; Computer Simulation; Patterns Grnrration; Introduction

Introduction Von Neumann and Stanislaw Ulam introduced cellular lattice in late 1940s as a frame work for modeling complex structures capable of self-reproduction (Von, 1966). Cellular Automata is based on a concept of dividing space into a regular lattice structure of cells where each cell can take a set of ‘n’ possible values. The value of the cell change in discrete time steps by the application of rule R that depends on the neighborhood around the cell. The neighborhood can be along a line, in a plane or in space. Cellular Automata (CA) model is composed of a universe of cells in a state having neighborhood and local rule. With the advancement of time in discrete steps the cell changes its value in accordance to the state of its neighbors. Thus the rules of the system are local and uniform. There are one- dimensional, two-dimensional and three-dimensional CA models. In one-dimensional CA the cells are like a linear canvas and the values of the canvas cells change due to application of a local rule in discrete advancing time steps. In two-dimensional CA the cells form a canvas plane and the changes take place in two dimensions while as in three dimensional CA volumetric changes take place by the application of local rule with advancement of time. As image is two dimensional data matrix, here we use 2DCA model, where cells are arranged in a two dimensional canvas matrix having interaction with neighboring cells. The central space represents the target cell (cell under consideration) and all spaces around represent its eight nearest neighbors. The structure of the neighbors mostly discussed and applied include Von Neumann neighborhood and Moore neighborhood, are shown in figure 1. In Von Neumann neighborhood, four cells are positioned at the orthogonal positions of the target cell (a i,j) while as Moore neighborhood is extension of Neumann structure with additional four cells placed diagonally at the four corner positions. For simplicity Von Neumann neighborhood cells can be termed as orthogonal neighbors and the additional cells by Moore can be called as corner neighbors.

*

Corresponding author. Tel.: +91 9086 897920. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Wani et al/ COMMUNE-2015

ai,j-1 ai-1,j

Target Cell

ai,j

ai+1,j

ai,j+1 Von Neumann Neighborhood

ai-1,j-1 ai,j-1

ai+1,j-1

ai-1,j

ai+1,j

Target Cell ai,j

ai-1,j+1 ai,j+1

ai+1,j+1

Moore Neighborhood Fig. 1 The two dimensional Cellular Automata in general are represented by the equation (I) as given below: [ai,j] t+1 = R[ ai, j , ai, j+1 , ai+1, j , ai, j-1 , ai-1, j ] t

--(I)

For Additive Cellular Automata the implementation of the famous totalistic rule in Von Neumann and Moore neighbourhoods, the respective representative equations can be written as follows: [ai,j] t+1 = XOR[ ai, j , ai, j+1 , ai+1, j , ai, j-1 , ai-1, j ] t --(II) [ai,j] t+1 = XOR[ ai, j , ai-1, j-1 , . . . . . . , ai+1, j+1 ] t --(III) Since the exploring worksheet/canvas is practically limited, researchers (Norman, 1986; Wolfram, 1985; Jarkko, 2012; Choudhury, et al) have defined some boundary conditions to facilitate the protection of data overflow outside the edges of the worksheet. On the basis of the applied boundary conditions the cellular automata have been divided into three main categories, briefly defined as follows: 1.1. Null Boundary Cellular Automata (NBCA) Under null boundary conditions the extreme edge cells are having zero values. For 1DCA the extreme right cell and the extreme left cell are considered to be having a value of binary zero ’0’ 1.2. Periodic Boundary Cellular Automata (PBCA) Under periodic boundary conditions the canvas is considered to be folded so that the extreme cells are taken to be adjacent to each other. For 1DCA the extreme right cell is considered to be adjacent to extreme left cell and the extreme left cell is considered to be adjacent to extreme right cell. 1.3. Intermediate Boundary Cellular Automata (IBCA) Under intermediate boundary conditions the left neighbor of the leftmost cell is regarded as the second right neighbor and right neighbor of the rightmost cell is considered as the second left neighbor. The linear additive 2-dimensional cellular automata attracted a number of researchers who have applied the rules for various applications in industry and research. The most important among such applications is the VLSI design that uses

[132]

Wani et al/COMMUNE – 2015

the cellular automata under the periodic boundary conditions. We have earlier made use of rules under null boundary conditions for cryptography and graphical translations (Wani, et al, 2014; Fasel, et al, 2012). The status value of the target cell at time t+1 depends on its own status value and the status value of cells in the Moore neighbourhood at time t, where t is earlier time than t+1. A number of research studies have been carried out by Stephen Wolfram and Norman H. Packard who categorized these rules under the discussed neighbourhoods into general, symmetric and totalistic. The famous example of 9-neighborhood totalistic cellular automata is John Horton Conway’s ‘Game of Life’. Use of the Game of Life rule in Von Neumann neighborhood has also been reported for pattern generation on multiple geometrical shapes by (Wani, et al, 2013) Various studies have also been carried out by Pabitra Pal Choudhury et al., who classified the cellular automata rules in Moore neighbourhood by assigning the rule values to different cells as shown in figure 2. The rules are generated by the interaction of target cell with itself and with the 8-neighbors around it. These nine rules are said to be basic or fundamental rules and group rules are derived from their combination, Group 2 are rules generated by addition of two basic rules, Group 3 by the combination of three basic rules, Group 4 by the combination of four basic rules, Group 5 by the combination of five basic rules and so on. Group 9 rule (only rule in the group) is the combination of all basic rules. All the combinations are additive (i.e. EX-OR operation).

64 128 32 1

256 2

Target Cell

16

8

4

Fig. 2: Examples: (Linear CA)

Rule 3 = Rule 2  Rule 1 Rule 11= Rule 8  Rule 2  Rule 1 Rule 15 = Rule 8  Rule 4  Rule 2  Rule 1

(Group 2) (Group 3) (Group 4)

Using this sort of configuration of patterns in 1DCA (Makoto, Leon, 2009; Radu, et al, 2006), we have reported (Wani, et al, 2013) 2DCA for cryptographic applications where the plaintext is converted to pattern based cipher. The cipher on the receiving end can be converted back to plaintext using the same CA rule in the forward iterations. This technique of generating ciphers has advantages of high cracking immunity due to wide range of rule possibilities and low hardware cost of implementation using VLSI technology. According to criteria of applying cellular automata rules to a group of data in any neighborhood, the cellular automata have been divided into two types: i) Uniform Cellular Automata ii) Non-Uniform Cellular Automata Uniform CA also known as Linear CA is where rule is applied uniformly on a data matrix of cells. All the cells in matrix get operated with the same rule. Non-Uniform CA also known as Hybrid CA is one in which all the cells of the matrix have their own local rule that may be different from the rule applied to other cells of that matrix. 2. Methodology For the experimentation purpose, we have used a null matrix of (129×129) elements and binary image of the seed was loaded at its centre. As the CA totalistic rule was applied it started generating patterns. The iteration loop control and the boundary conditions were also applied according to the following algorithm:

Label

Start Load a Null Matrix Input Binary Seed Start Iteration Counter Apply CA Rule Decrement Iteration Counter Loop to Label on Counter Condition Output Resulting Pattern Stop

[133]

Wani et al/ COMMUNE-2015

Start Load Seed Control Counter Boundary Condition Apply CA Rule No

Counter = 0 Yes Pattern Out Stop

Flowchart 3. Pattern Generation We have studied the pattern generation using totalistic cellular automata Game of Life rule (Rule 170) in Neumann neighborhood for the different orientations and have achieved results that could lead to various field applications for interdisciplinary research. In this new exploration we have used various geometrical orientations of the input seed and boundary conditions for the generation of the patterns in evolution as well as what can be termed as boundary invasion. The different orientations used range from simple regular shapes to boundary line matrices and deliberately introduced boundary defects. Different seed shape exploration generated the following pattern results on the MATLAB simulations under null boundary conditions and boundary invading conditions. 3.1. Pattern Generation Using Boundary Defects In this study, we have made the use of above methodology to observe the effect of null boundary on the world of full life i.e. universal matrix of ones. The dark or null boundary is a rectangle of zeros. Some resulting pattern generation references are indicated in Table 1 below: Table 1. Boundary Defect Patterns

T=0 (Dark World)

T=1 (Null Boundary)

T=10

T=21

T=29

T=37

T=41

T=47

T=53

[134]

Wani et al/COMMUNE – 2015

3.2. Pattern Generation Using Single Seed Boundary Invasion In this study same methodology is used to observe the effect of null boundary with a single central live seed on the world of full life i.e. universal matrix of ones. The dark or null boundary is a square of zeros with every side having a central seed of one. Some resulting pattern generation references are indicated in Table 2 below: Table 2. Single Seed Boundary Defect Patterns

T=0 (Dark World)

T=1 (Null Boundary)

T=10

T=21

T=29

T=37

T=41

T=51

T=59

3.3. Pattern Generation Using Single Seed Evolution and Boundary Invasion Moving ahead with the above methodology here we observe the effect of null boundary with a single central live seed at the centre of a world of full life i.e. universal matrix of ones. The dark or null boundary is a square of zeros. A single live seed is introduced at the centre of the matrix. Some resulting pattern generation references are indicated in Table 3 below:

[135]

Wani et al/ COMMUNE-2015 Table 3. Single Seed Evolution and Boundary Invasion Patterns

T=0 (Dark World)

T=1 (Null Boundary)

T=10

T=21

T=29

T=37

T=41

T=51

T=59

3.4. Pattern Generation Using Star Shaped Data Block Here we introduce a multi seed data block with the above methodology to observe the effect of evolution with the application of same rule. The evolution results in a periodic multiplication of the data block in various orientations. Some resulting pattern generation references are indicated in Table 4 below: Table 4. Multi Seed Evolution Patterns

T=0

T=16

T=24

T=32

[136]

Wani et al/COMMUNE – 2015

T=47

T=48

4. Discussion and Conclusion The results are definitely not only showing a bidirectional interaction that can be useful in studying physical processes under environmental conditions but also a powerful parallelism. The power of the parallelism can be utilized to minimize the hardware complexities.

References Choudhury, P. P., et al, Dec. 2010. Classification of CA Rules based on their Properties., IJCC. Fasel Qadir, Ahmad, P. Z., Wani, S. J., Peer M. A., December-2013. Quantum-Dot Cellular Automata: Theory and Applications, IEEE Conference on Machine Intelligence and Research Advancement (ICMIRA-2013), pp. 540-544. Fasel Qadir, Shah, J., Peer, M. A., and Khan, K. A., July 2012. Replacement of Graphic Translations with Two-Dimensional Cellular Automata, Twenty Five Neighborhood Model” IJCEM , pp. 33-39. Jarkko Kari, April, 2012. Universal Pattern Generation with Cellular Automata, Theoretical Computer Science, vol. 429, Makoto Itoh and Leon, O., 2009. Chua, Difference Equations for Cellular Automata, International Journal of Bifurcation and Chaos, Vol. 19 No.3, 805-830. Norman, H. Packard and S, 1986. Wolfram. Two Dimensional cellular Automata: J. Stat. Phys, vol. 38. Radu V., Craiu and Thomas, C., Lee, M., July 2006. Pattern Geeration Using Likelihood Interference for Cellular Automata IEEE Transactions on Image Processing vol. 15 No. 7. Shah, J. W., Fasel Qadir , Khan, K. A., and Peer, M. A., 2013, Dec. Springer Conference ICICIC Global Pattern Generation Using 2D Cellular Automata on Multiplr Geometrical Shapes. Von Neumann, J, 1966. Theory of Self-Reproducing Automata: University of Illinois Press. Wani, S. J., Khan, K. A., and Peer, M. A., April, 2014. 2D- Cellular Automata Linear Rules for Cryptography Based on Pattern Evolution. International Journal of Advanced Research in Computer Science Engineering and Information Technology vol 2 no.3. Wolfram, S., , 1985. Twenty Problems in the Theory of CA, Physica Scripta, pp. 170-183, vol T9.

[137]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16-18March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

High Impedance First-Order Transadmittance-Mode Allpass Filter Using CCII and OTA Nusrat Parveena*, Syed Zaffer Iqbalb a

Department of Electronics, Islamia college of Science and Commerece, Srinagar, India b Department of Physics,Government Women’s College, Nawakadak Srinagar, India

Abstract A novel transadmittance-mode (TAM) first-order allpass (AP) filter using a single second-generation current conveyor (CCII+), operational transconductance amplifier (OTA), one grounded resistor, and capacitor is presented. The input is voltage signal and output is current signal. The circuit has advantage of having input and output impedances high, thus facilitate cascading without additional devices. The phase angle, in addition to frequency of the applied signal, can also be adjusted electronically through the bias current of OTA without disturbing realizability condition. PSPICE simulation confirms the theoretical results

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Transadmittance Filter; High Impedance; Current Conveyor; Operational Transconductance Amplifier

1. Introduction Allpass filter is most important building block, used in analog signal processing. Due to potential advantages, these filters have received much attention among the circuit designers. AP filters are generally employed for introducing frequency dependent phase shift while keeping the amplitude of the input signal constant over the desired frequency range. Other areas of application of these active circuits are realization of oscillators and high-Q filters (T. Comer et al, 1997, Atti. Abuelma et al-1998, R. K. Rajeev et al, 2014, Toker . A. et al, 2000, A. Soliman, 1997). The active devices that have been used for the realizations of first-order all-pass circuits include operational amplifiers (OP-AMP), secondgeneration current conveyor (CCII), current feedback Op-Amps (CFOA), operational transconductance amplifier (OTA) and four-terminal floating nullor (FTFN). Several voltage and current-mode first-order all-pass filters are available in the literature (Higashimura M, 1990, Fukui Y et al,, 1999, O. Cicekoglu ety al, 2000, Khan, I. A and S. Maheshwari, 2000, Cam U.,et al, 2000, Khan,I. A and S Maheshwari, 2001, Pandey Paul. N, S.2006, Horng, J. et al, 2004, A Toker and S. Özoguz et al, 2001). A survey of technical literature reveals that there are only a few transadmittance-modes and/or Transimpdance-mode (TIM) filters (Toker A et al, 2001, Abuelma’atti M.T, Minaei, 2004, Shah, N. A et al 2004, Shah N. A et al, 2005, U Cam et al, 2005, Cam U, 2004). On the other hand, such type of filters are useful as an interface connecting a voltage-mode circuit to a current-mode circuit. Traditionally the current/voltage signal is converted to voltage/current signal and then processed. However, a TIM/TAM filter will perform this conversion and filtering simultaneously lending reduction in hardware, which in turn save chip area, thus resulting economically viable topologies. An application of transadmittance-mode filters for the realization of modern base-band receiver block of radio system can be found in (Toker A et al, 2001). Two first-order allpass filters one is TIM and other in TAM is reported in literature (Cam U et al, 2004, Cam U, 2005). The TIM employs one operational transresistance amplifier with the floating passive components and TAM uses one dual output current conveyor third generation (CCIII), two floating resistors, one grounded resistors and a grounded capacitor. The main purpose of this paper is to introduce a new transadmittance-mode first-order all-pass filter employing a single CCII+, one OTA, and each of grounded resistor and capacitor. Although the proposed circuit employs one more active component than the above mentioned first-order allpass filter, the proposed circuit provide the features of high input and output impedance *

Corresponding author. Tel.: +91 9419 426556. E-mail address: [email protected]

ISBN: 978-93-82288-63-3

Parveen and Syed/COMMUNE – 2015

and using only two grounded passive components. Since the TAM is the interface circuit used between voltage-mode and current-mode circuit thus it is essential for the TAM that input as well as output impedance should be high. The phase angle can be electronically adjusted through the bias current of OTA without disturbing the realizability condition Rg1= 1. Moreover, the circuit employs a grounded capacitor and resistor, besides offering electronic tunability feature, which is beneficial for IC technology (Horng, J et al, 2004). PSPICE simulation confirms the workability of the filter. 2.

The Proposed Circuit

Symbolic representation of second-generation current conveyor is shown in Fig 1(a), and is a all-round active device with versatile port relations given in the matrix form as under:  I y  0 0 0 V y  V   1 0 0  I   x   x   I x  0 1 0 V z 

(1)

OTA, circuit symbol is given in Fig. 1(b) and its micro modal in Fig. 1(c) is differential voltage controlled current source (DVCCS) and is characterized by the following port relation

(V   V  )  I o gm

(2)

where gm is the transconductance gain of OTA, used to electronically control the phase angle, is given by

gm 

I bias 2VT

where Ibias is the bias current and VT is the thermal voltage. Routine analysis of the proposed TAM filter shown in Fig. 1 yields the following transfer function:

Io 1  Vin g1

g2 ) g1 RC g (s  2 ) C

(s 

(3)

The realizability condition is g1R1 =1. The pole frequency is given by

o 

g2 C

(4)

and phase angle is given by

 (s)   2 tan 1 C / g 2 

(5)

Eq.(2) shows that the circuit realizes TAM first-order allpass transfer function. Moreover, from Eq.(4) phase can be controlled by g2 in addition to C and frequency of applied signal without disturbing realizability condition.

2. CCII Non-Idealities Taking the tracking errors of the CCII into account, the CCII is characterized by the following matrix

0 0 V y  I y   0 V     s  0 0  I x   x   I x   0   s  0 V z 

(6)

Where (s) and (s) represent the frequency transfers of the internal current and voltage followers of the CCII, respectively. They can be approximated by the following first-order functions (A Fabre et al, 1993)

[139]

Parveen and Syed/COMMUNE – 2015

 s  

And

 s  

0 1 s

0

0 1 s

0

where 0 = 0.9914,  = 3.8  109 rad/sec 0 = 0.9999,  = 6.48  109 rad/sec. If the proposed circuit is designed for the frequency much less than the corner frequency then, (s) and (s) take the form (s) = 1- i and (i << 1) and (s) 1- v and (v << 1) i and v respectively represent the current and voltage tracking errors. Using the non-ideal port relation the transfer function of the proposed circuit is given by

Io   g 1 Vin

s

g2 C

 1    1      R1 g1   g s 2 C

(7)

From Eq. (7) it is clear that the tracking errors have no effect on pole frequency. The passive sensitivities are small and are given as 3.

S g20  S C,0  1

Simulation Results

To verify the theoretical analysis, the first order all-pass filter was simulated with PSPICE program. The figure 1 was designed for a phase shift of 900 at centre frequency of 159.2 KHz. The designed values are as g 1 = g2 =1 mS, R = 1 k and C = 1nF. CCII was implemented using AD844 of analog Devices. The CA3080 OTA macro model shown in figure (1C) with Ri = 100 k, Ro = 70M, Ci = 2:6 pF, Co =3:6 pF was used (WU. J, 1994) the output current was taken along 1  resistor (RL). Figure 3 shows the phase frequency response while as Fig. 3 depict the magnitude frequency response. The figure 4 shows the variation of phase with g 2. It is clear from Fig. 4-Fig. 5 that the circuit performs well and the results are in close conformity with the theoretical results.

Fig. 1(a) Symbol of CCII+

Fig 1(b) Symbol of OTA

Fig.1(c) Micro Model of OTA

[140]

Parveen and Syed/COMMUNE – 2015

Fig. 2 Proposed TAM AP filter

Fig. 3 Phase frequency response of proposed filter

Fig.4 Magnitude frequency response of proposed filter

4. Conclusion A new TAM first-order AP filter employing a single CCII, one OTA, one grounded resistor and a grounded capacitor is presented. The phase angle, in addition to frequency of the applied signal and grounded capacitor, can also be adjusted by the transconductance gain of OTA without disturbing realizability condition. The proposed circuit has both high input and output impedance (z-terminal of CCII) thus facilitating cascading without additional devices. PSPICE simulation confirms the theoretical results.

[141]

Parveen and Syed/COMMUNE – 2015

References: D.T. Comer, Comer, D.J, Gonzalez, J.R. 1997, A high frequency integrable band-pass filter configuration. IEEE Trans. Cir. and Systems-II. 41, p 856–860. R. K.Rajeev, Y. P. Surya , S. Shubham , Paul K Sajal, 2014, Active Comb Filter Using Operational Transconductance Amplifier, Active and Passive Electronic Components 115-587932 M. T. Abuelma’atti, Al-qahtani, M. A. 1998, A new current controlled multiphase sinusoidal oscillator using translinear current conveyors, IEEE Trans. Cir. and Systems II. 45 ,881–885 Toker, A, Ozoguz, S. Cicekoglu, O. Acar, C. 2000, Current-mode all-pass filters using current differencing buffered amplifier and new high-Q bandpass filter configuration. IEEE Trans. Cir. and Systems II. 47 949–954. Soliman, A. M., 1997, Generation of current conveyor based all pass filters from OpAmp based circuits. IEEE Trans. Cir. and Systems II. 44, 24– 330. M. Higashimura, and Y. Fukui, 1990, Realization of current mode all pass networks using current conveyor. IEEE Tran. Cir.and Systems 37 660–661. Cicekoglu, O. kuhtman, H. Berk, S, 1999, All pass filters using a single current conveyor.” Int. J. Electronics, 86 , 947–955. Khan, I. A. Maheshwari, S.: 2000, Simple first order all pass section using a single CC II.” Int. J. Electronics, 87 303–306. U. Cam, O. Cicekoglu, M. Gulsoy, and H Kuntman, “New voltage and current mode first-order all-pass filters using single FTFN.” Frequenz, 54 (2000),177–179 Maheshwari, S, Khan,I. A., 2001, Novel first order all-pass sections using a single CCIII.. Int. J. Electronics, 88, 773–778. Pandey, N. Paul, S. K.: 2001All-pass filters based on CCII_ and CCCII-. Int. J. Electronics, 91, 485–489. Horng, J. W. Hou, C. L. Change, C. M.; Chung W. Y.; Liu, H. L.; Lin, C. T.: 2006, High output impedance current-mode first-order allpass networks with four grounded components and two CCIIs. Int. J. Electronics, 93, 613–621. Toker, A. Özoguz, S. 2004, Novel All-Pass Filter Section Using Differential Difference Conveyor. AEU Int. J. Electronics, and Communication 58 , 153–155. Toker, A. Cicëkoglu, O. Ozcan, S. Kuntman, H, 2001, High-output-impedance transadmittance type continuous-time multifunction filter with minimum active elements. Int. J. Electronics, 88 1085-1091 Abuelma’atti M.T. Bentrcia A.; Al-Shahrani, S. M. 2004, A novel mixed-mode current-conveyor-based filter. Int. J. Electronics, 91,191–197. Minaei, S. Topcu G. Cicekoglu, O.2005, Low input impedance trans-impedance type multifunction filter using only active elements. Int. J. Electronics 92 385–392. Shah, N. A. Iqbal,S. Z. Parveen, B.2004, SITO high output impedance transadmittance filter using FTFNs. Analog Integrated Circuits and Signal Processing, 4087-89. Shah, N. A. Iqbal,S. Z. Parveen, B.2005, Lowpass and Bandpass Transadmittance Filter using Operational Amplifier Pole. AEU Int. J. Electronics and Comm., 59, 410-412. Cam, U. Cakir, C. Cicekoglu, O. 2004, Novel Transimpedance Type First-Order All-Pass Filter Using Single Otra. AEU Int. J. Electronics and comm., 58, 296–298. Cam, U.2005 , A New Transadmittance Type First-Order Allpass Filter Employing Single Third Generation Current Conveyor. Analog Integrated Circuits and Signal Processing, 43(97–99. Fabre, A. Saaid, O. Barthelem, H., 1993 On the frequency limitation of the circuits based on the second generation current conveyor. Analog Integrated Circuits and Signal Processing, 7113-129 WU, J, 1994 Current-mode high-order OTA-C filters. Int. J. Electronics, 761115-1120.

[142]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Sequential Circuit Design Using Quantum dot Cellular Automata (QCA) Javeed Iqbal Reshi*, M. Tariq Banday, Farooq A. Khanday Department of Electronics and Instrumentation Technology, University of Kashmir, Srinaga, India

Abstract Conventional technologies have lot of limitations when scaled to a nano-level. Several alternative technologies have been proposed as solutions in the open literature. Quantum dot Cellular Automata (QCA) is one such technology, which is thought to be a perfect replacement of Complementary Metal Oxide Semiconductor (CMOS) technology for digital designs. Therefore, many researchers have given their efforts to realize the various combinational logic functions. Till date, some sequential circuits have also been reported in QCA which include a D Flip flop and a serial-in-serial-out (SISO) shift register. In this paper, a novel design of D flip flop has been introduced. The proposed design of D Flip Flop has then been employed to implements three shift registers, which include the SISO one as well. The proposed QCA designs enjoy the features of small area, superior performance factors in respect of noise, circuit stability, and low power dissipation as compared to those already reported in the open literature. The operation of QCA circuits was verified by QCA designer tool.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords- CMOS; Quantum Cellular Automata; QCA; D-Flip Flop; Serial-In-Serial-Out; SISO; shift register

1. Introduction The physical limitations of CMOS as the technology is scaled down have made scientists to conduct extensive researches in nanotechnology, as a solution for future generation integrated circuits (ICs). Gordon Gordon Moore observed in early 1960’s that the number of transistors that could be embedded in a specific area on a chip would increase exponentially and double every two years. This has been the challenge in transistor technology ever since. (Liu, 2006, Askari, 2008). This trend in this scaling will come to stop due to the physical limits of solid-state electronics. The heat generated by the billions of transistors cannot be dissipated quickly to prevent the damage Shrinking transistors has been the major trend to achieve circuits with fast speed, high densities and low power dissipation. However, when scaling comes down to submicron level, many problems occur. Physical limits like quantum effects and nondeterministic behavior of small currents and technological limits such as high power consumption and design complexity may hold back the further progress of microelectronics using conventional circuit scaling (Liu, 2006, Navi, 2010) Thus, to continue trends of scaling and increasing performance, other technologies need to be explored. As an alternative to CMOS-VLSI, researchers have proposed an approach of computing with quantum dots, the quantum cellular automata (QCA) for low power and high speed applications. QCA exploit the quantum effects that come with a small size and its basic element is quantum dot cell where the logic state is encoded as the position of electrons with in a cell. This novel idea was first proposed by Dr. Craig Lent in 1993(Lent, 1993). Recent papers show that QCA can achieve high density, fast switching speed, at room temperature as well (Cho, 2007).

*

Corresponding author. Tel.: +91 9419 436575. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Reshi et al/ COMMUNE-2015

The aim of this paper is to present the design strategy and analysis of QCA sequential logic circuits. The proposed design for sequential circuits is based on the novel D flip-flop. The prime aim is to increase the circuit density by minimizing the use of cells in each block. The proposed QCA structures have been designed and verified using the QCADesigner tool (bi-stable approximation). 2. Quantum-Dot Cellular Automata (QCA) The fundamental unit of QCA is the QCA cell created with four quantum dots positioned at the vertices of a square (lent, 1993). Two mobile electrons are loaded in the cell that can move to different quantum dots in the QCA cell by means of electron tunneling. The electrons can only occupy the corners of QCA cell as a result of Coulombic repulsion thereby resulting the two polarizations (i.e., +1 or -1). as shown in Fig. 1a. The tunneling outside the cell is not allowed because of high potential barrier (Omar, 2007). When the two cells are placed together the Coulombic interaction between the electrons forces the cell to acquire a particular polarization.(either +1 or -1) The majority gate is the fundamental QCA circuit which is used to implement the primitive logic functions (Tougaw, 1994). The logic equation for a majority gate is M ( A, B, C)  AB  BC  AC and can be implemented by the arrangement as shown in Fig. 1(b). By fixing the polarization of one the input as logic “1” or “0”, we can obtain an OR gate or an AND gate respectively.

a.b  M (a, b,0) a  b  M (a, b,1)

(1) (2)

Fig. 1(c) shows the QCA wire which is formed by placing the QCA cells consecutively. In a QCA wire, the signal propagates from input to output as a result of the electrostatic interactions between cells. There are two types of QCA wires ;( 1) 900 QCA wire wherein the signal propagates down the wire and the same output is obtained at different stages depending upon the polarization of the input cell. (2) 450 QCA wire in which the propagation of signal alternates between the two polarization states. Another important structure in a QCA is the inveter as shown in Fig. 1(d).

Fig.1 Primitives of the QCA

2.1. QCA Clock All QCA circuits require a clock to synchronize and control information. It also provides the power to the circuit (Sahni, 2008). The QCA computation is controlled by a four phase clock signal as shown in Fig. 2.

Fig.2. Four Phases of Clock

This clocking scheme in QCA consists of four phases (Snider, 2006, Kim, 2007, Lent, 1997): Switch, Hold, Release and Relax, as shown in fig.2. The phase difference between the various clock zones is 900. During the Relax phase, the electrons are pulled into the middle dots, and the cell is said to be in “null” state. During the Switch phase, the interdot barrier is slowly raised and pushes the electrons into the corner dots, and hence the cell attains a definite polarity under the influence of its neighbors. In the Hold phase, barriers are high and a cell retains its polarity and acts as input to the neighboring cells. Finally in the Release phase, barriers are lowered and the electrons are pulled into the middle dots so the cell loses its polarity and the cell attains the null state again. Thus during Relax and release phase the QCA cell remains in unpolarized state, while the QCA cell attains a particular polarization during the switch and hold states. The clock signals for QCA circuits are generated through an electric field, which is applied to the cells to either raise or lower the tunneling barrier between dots within a QCA cell. This electric field can be supplied by either by CMOS wires, or carbon nano tubes.

[144]

Reshi et al/ COMMUNE-2015

In a clocked QCA circuit, the information is transferred and processed in a pipelined fashion. All cells within the same clock zone are allowed to switch simultaneously, while cells in different clock zones are isolated. The QCA binary wire in Fig. 3 can be used to illustrate this clocking scheme. When a fixed polarization input is applied the subarray1 enters in to switch state and attains a fixed polarization depending on the polarization of input cell. Then, subarray1 enters the hold phase and retains the polarity as attained during the switch phase. At this time, subarray2 starts switching and the subarray3 is in the relaxed state, but it will not influence the computational state of subarray2. During the next phase, subarray1 is moved to a release phase; subarray2 is in the hold state and provides the input to subarray3 (that is in the switch phase). Thus, information transfers take place in a pipelined fashion as shown in Fig. 3, and there is a 90 phase shift from one clock zone to the next.

Fig.3. Signal Propagation though a Binary wire.

2.2. Proposed QCA Implementation of D-Flip flop The QCA D Flip-flop (DFF) can be simply constructed by a QCA binary wire with four clocking zones as shown in Fig. 4(a). The Latching can be accomplished trough timing by using four phases of clock assignment; however this implementation does not have the provision of separate clock control signal. In order to overcome this limitation a clock control signal is incorporated in the schematics as shown in Fig. 4(b). The logic equation of the D Flip-flop can be represented as: 

Q(t )  Clk .D  Clk .Q(t 1)

(3)

Fig. 4(b) & 4(c) shows the schematics and proposed QCA implementation of the D-Flip flop

(a)

(b)

(c)

Fig. 4. (a) Un-clocked QCA implementation of D-Flip flop; (b) Schematics of clocked D-Flip flop; (c) proposed QCA implementation of D-Flip flop

The designs essentially require multi-phased clocking mechanism for the synchronization and proper information flow. In this design each of the majority gates are constructed using the three different clocking zones. The important feature of this design is that 450 wire is used which provides both the original signal and its complimented output thus eliminating the explicit use of inverter. The simulation result of the proposed D-Flip flop is shown in Fig. 5. The simulations were carried out using the QCADesigner v2.3 bi-stable vector engine with the following parameters; cell size = 20nm; No. of samples = 28800; Convergence tolerance = 0.001000; Radius effect = 65.000000nm; Relative permittivity = 12.900000; Clock high = 9.800000e-022; Clock low = 3.800000e-023; Clock shift = 0; Clock amplitude factor= 2.000000; Layer separation = 11.500000; Maximum iterations per sample = 100.

Fig. 5. Simulation Result of D- Flip flop

[145]

Reshi et al/ COMMUNE-2015

The performance of the proposed D-Flip flop is compared with the already existing circuits available in the literature and is given in Table.1. Table 1. Comparison of various D-Flip flops Circuit Cell Count

Cell Area (µm2)

Total Area(µm2)

Latency(in Clock cycles

References

D-Flip flop

35

0.0113

0.0272

1

(Mustafa. M, 2014)

D-Flip flop

36

0.0116

0.0262

1

(Ahmad, Firdous, 2014)

D-Flip flop

43

0.0139

0.0453

1.25

(Sarkar, Tama., 2013)

D-Flip flop

48

0.0155

0.0356

1

Hashemi, Sara., 2012

D-Flip flop

68

0.0220

0.0680

1.5

Vetteth, A., 2003

D-Flip flop

33

0.0106

0.0210

1.25

Proposed

From Table 1, it is evident that the proposed design is efficient in terms of cell count, cell area and the total area. The latency of the proposed design would have been equal to 1 but increasing it to 1.25 gives more stability (when complex sequential circuits are designed) than the other designs of the table. These designs are less stable as complex sequential circuits are designed otherwise the latency would have been more. 2.3. Proposed QCA implementation of Shift Registers The proposed D-Flip flop can be used as a basic primitive for designing the shift registers. In this paper, three basic shift registers i.e. Serial Out shift register (SISO), Serial in Parallel Out (SIPO) and Parallel in Parallel Out (PIPO), were designed. Fig. 6. Shows the QCA implementation of the SISO. The serial data input is applied to the left most D-Flip flop and the output of each D-Flip flops is connected to the data input of the next stage in the chain resulting in a circuit that shifts the bit array stored in it by one position, at each transition of the clock input. The simulation result of the SISO shift register is shown in Fig. 7.

Fig. 6. QCA Implementation of Serial In Serial Out (SISO) Shift Register.

Fig.7. Simulation Result of SISO Shift Register.

This kind of the shift register finds the extensive use in digital communication circuits where we need to design a linear feedback circuit for generating the PN sequences. Another application of this kind of the circuit is in the delay circuits where we need to provide a delay from an input to the output. Fig. 8 shows the QCA implementation of SIPO shift register. This kind of the shift register is also designed with the proposed D-Flip flop. In this shift register the input is applied to the serial Data in of the first D-Flip flop and the output is obtained in parallel fashion from the different stages of the Flip flops in the chain. The output of the first stage is also applied to the data input of the second stage and so on. This kind of shift register is used where we need to have a serial to parallel conversion. The simulation result of the SIPO shift register is depicted in Fig. 9.

[146]

Reshi et al/ COMMUNE-2015

Fig..8. QCA implementation of Serial In Parallel Out (SIPO) Shift Register.

Fig. 9. Simulation result of SIPO Shift Register.

Fig. 10 shows the QCA implementation of the PIPO shift register. The design is almost similar to that of the Fig. 6 and 8. In this kind of shift register the inputs are applied to the individual stages of the D-Flip flops thus instead of having a single input there are 4 input and the outputs are also obtained from the individual D-Flip flops. The simulation result of the PIPO shift register is shown in Fig. 11.

Fig. 10. Simulation Result of Parallel in Parallel Out (PIPO) Shift Register.

Fig. 11: Simulation Result of the PIPO Shift Register. The comparisons of the various types of shift registers are mentioned in the Table 2. Table 2. Comparison of Shift Registers Latency(in Clock cycles)

Circuit

Cell count

Cell Area

Total Area

SISO Shift Register

195

0.23

0.1632

SISO Shift Register

204

0.24

0.1827

4

(Ahmad, Firdous, 2014)

SISO Shift Register

191

0.22

0.1415

4.25

Proposed

SIPO Shift Register

191

0.21

0.1415

1.25

Proposed

PIPO Shift Register

173

0.19

0.1266

1.25

Proposed

[147]

4

References (Mustafa. M, 2014)

Reshi et al/ COMMUNE-2015

Only the comparison of SISO shift register is available in the open literature. The proposed designs of all the types of shift registers are fed through a common clock control signal which is obtained with a 45 0 wire. Such a kind of approach has the advantage of providing the original signal and its complimentary output thereby remove the need of using an explicit inverter. The other feature of this kind of wire is that it provides a strong polarity output as compared to 900 wires. The SISO shift register reported by (Mustafa, 2014) and (Ahmad, 2014) uses the conventional 900 wire ideally in which we need to restrict the single clock zone to 10 cells otherwise the output will not be correct as the signal propagation will be weak. The other issue in both the designs is the uneven clocking. Both the implementations are restricted to only three bits and they will not yield the desired output when higher order structures needs to be designed. These limitations are overcome in the proposed SISO shift register thus making the design more reliable and efficient. This paper also discusses QCA implementation of SIPO and PIPO shift registers which, to the best knowledge of authors, are reported first time. Conclusion In this paper the design methodology and the testing scheme for proposed design of QCA based D-Flip flop (DFF) and the various shift registers has been discussed. The proposed designs are efficient in terms of area, delay and cell count and fault tolerance. The Simulations of the proposed designs were carried out using bi-stable vector engine of QCADesigner tool. The proposed QCA based shift registers have the unique feature of modularity and hence can be extended to any high order design or even may be used as basic building block of a general purpose nanocomputers/processors. The proposed designs as well as verification methodology will therefore prove significant in designing complex sequential circuits.

References Liu, M., 2006. Robustness and Power Dissipation in Quantum-Dot Cellular Automata. PHD thesis, Notre Dame University, Indiana. Askari, M., M. Taghizadeh, Kh. Fardad, 2008. Digital Design using Quantum-Dot Cellular Automata (A Nanotechnology Method), in Proc. International Conference on Computer and Communication Engineering, pp: 952-955. Navi, et al., 2010, A new quantum-dot cellular automata full-adder, Microelectronics Journal. Lent C. S. and Tougaw P. D., 1993, Lines of interacting quantum-dot cells: a binary wire, Journal of applied Physics, vol. 74, pp: 6227-33. Askari, M., M. Taghizadeh, Kh. Fardad, 2008. Design and Analysis of a Sequential Ring Counter for QCA Implementation, in Proc. International Conference on Computer and Communication Engineering, pp: 933-936. Cho, H., E.E. Swartzlander, 2007, Adder Designs and Analyses for Quantum-Dot Cellular Automata, IEEE Transactions on Nanotechnology, 6(3), pp: 374-383. Omar Paranaiba, et al., 2007, Neural Network Simulation and Evolutionary Synthesis of QCA Circuits, IEEE Transactions on Computers, 56(2), pp: 191-201. Tougaw P. D, and Lent C. S., 1994, Logical devices implemented using quantum cellular Automata, Journal of Applied Physics, 75(3), pp: 18181825. Sahni, 2008, Nanocomputing, Tata McGraw-Hill. Snider, G., et al., 2006, Implementation of Quantum-dot Cellular Automata, ICONN 2006, pp: 544-547. Niemier, M. T., at al., 2000, A design of and design tools for a novel quantum dot based microprocessor, In 27th Ann. Design Automation Conference, pp: 227–232. Kim, K., at el., 2007, The Robust QCA Adder Designs Using Composable QCA Building Blocks, IEEE Trans. On Computer-Aided Design of Integrated Circuits and System, 26(1), pp: 176-183. Lent, C. S., P.D. Tougaw, 1997. A device architecture for computing with quantum dots, Proc. IEEE, vol. 85, pp: 541–557. Huang, J., M. Momenzadeh, F. Lombardi, 2007, Analysis of missing and additional cell defects in sequential quantum-dot cellular automata, VLSI journal (40), pp: 503–515. Mustafa, M., Beigh, M R., 2014 “Novel Llinear feedback shift register design in quantum dot cellular automata”, Indian Journal of Pure & Applied Physics, p 203. Ahmad, Firdous. etal., 2014 “A novel idea of pseudo-code generator in quantum-dot cellular automata (QCA)”, Int. J. Simul. Multisci. Des. Optim, p 5. Sarkar, Tama., 2013 “Design of D Flip-Flip Using Nano-Technology based Quantum Cellular Automata”, International Journal of Soft Computing and Engineering (IJSCE). Hashemi, Sara., Nevi Keivan., 2012 “New Robust QCA D flip flop and memory structures” , Microelectronic Journal Vetteth, A etal., 2003 “quantum-dot cellular automata of flip-flops” ATIPS Laboratory 2500 University Drive, N.W., Calgary, Alberta, Canada

[148]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Performance Evaluation of OLSR, DSR and AODV MANET Protocols Nadiya Mehraja*, Faizan Kitabb, Zia Malika, M.Tariq Bandaya a

Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India b Department of Computing and Mathematical Science, University of Greenwich, London, UK

Abstract Mobile Ad-hoc NETwork (MANET) is an autonomous network providing wireless (radio signal) transmission without any centralized infrastructure unlike other wireless networks that employ use of some base station or access point. Three type of MANET protocols that are common for routing in such networks are proactive, reactive and hybrid. The most popular protocols are AODV, DSR, and TORA, which are reactive routing protocols, OLSR that is a proactive routing protocol, and GRP, which is a hybrid protocol. This paper compares three routing protocols viz. Optimized Link State Routing (OLSR), Dynamic State Routing (DSR) and Ad-hoc On Demand Distance Vector (AODV) in terms of delay, load, and throughput. Experimentation with these protocols through OPNET Modeller 17.1.A.PL2 (Build 11124 64-bit) have shown that AODV and OLSR protocols outperform DSR protocol. The AODV protocol outperformed others in throughput whereas the OLSR outperformed others on load balance, and delay. Further, OLSR protocol showed a consistent performance throughout the timeline and the DSR protocol showed substantial performance in less dense network with less node mobility.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords : MANETs; OLSR; DSR; AODV; OPNET

1. Introduction Owing to the flexible connectivity and ease in accessibility, techniques to interconnect devices and share information among them are considerably changing from wired to wireless networks. (Tadeusz, et al, 2006). Wireless networks (Kumaran, et al, 2007) involve various nodes, which communicate with each other through some wireless interface. In recent times sensor network, ad-hoc mobile networks, cellular networks, and satellite networks, which can exist as either infrastructure wireless networks (centralized) or infrastructure less wireless networks (decentralized) have received extensive research attention. The decentralized networks, commonly known as Mobile Ad-hoc NETworks (MANETs), offer self-governing system wherein routers and hosts communicate wirelessly with other nodes eliminating the need for static nodes and link configurations. Standards such as Bluetooth and Wi-Fi, and emergence of portable and handheld devices such as laptops, PDAs, and smartphones have made MANETs a remarkably hot topic. Unlike networks such as GSM and WLAN, where hosts can be mobile but routers are always fixed, a MANET permits peer-to-peer multi-hop networking with both nodes and routers in-motion. MANETs have applications in areas such as defense, under water or at an earthquake hit spot where router connectivity may change frequently. By establishing alternative connections inside hotspot cells, ad hoc network rules out its dependence upon fixed topologies, fixed neighbours, fixed relationship between IP address and locations (Tanenbaum, 2010). The key requirement for a MANET is node-to-node communication. MANET freely switches to some distinct path, changes its direction and hence adapt new connections with other devices. Due to the absence of pre determined topology or central control in MANETs, it is not possible to to handle packets by using traditional routing protocols such as OSPF, RIP and EIGRP. MANET uses AODV, DSR and OLSR protocols that respectievely work on the priciples of Distance Vector Routing, Implicit Source Routing and Link State Routing. On Demand Distance Vector (AODV) uses an on-demand methodology for finding routes only when required by a source node for transmitting data. It engages destination sequence numbers to identify the most recent path *

Corresponding Author: Tel: +91-9906829933 Email address: [email protected] ISBN: 978-93-82288-63-3

Mehraj et al COMMUNE-2015

(Perkins et al, 2003). Dynamic Source Routing (DSR) is an on-demand protocol for wireless mesh networks intended to restrict the bandwidth expanded by control packets in ad hoc wireless networks by eliminating the periodic table-update messages prerequisite in the table-driven approach. On the other hand Optimised Link State Routing (OLSR) protocol operates as a table driven, proactive protocol, i.e it keeps on exchanging topology information with other nodes of the network regularly (Clausen, et al, 2009). 2. Related Work Josh Broch et al studied DSDV, TORA, DSR and AODV routing protocols on a 50-node multi-hop wireless ad-hoc network using NS2 simulator (Josh, et al, 1998). The study concluded that for many scenarios with different metrics, DSR protocol performed better than other protocols. Further, AODV protocol eliminated the drawback of overhead due to source routing as was experienced in the case DSR with all the positive points staying intact. Al-Maashri, A. and Ould-Khaoua examined DSR, AODV, OLSR using NS2 simulator. The model was set under burst traffic conditions at node mobility of 20m/s (Ould-Khaoua, 2006). The study showed that the use of DSR was effective in terms of throughput, end-to-end delay and packet delivery ratio) when mobility of nodes remained below 10m/s. The OLSR showed very poor performance when tested under self-similar traffic and high node mobility. The AODV showed low end-to-end delay, average performance and stability in many conditions. Gowrishankar et al, studied AODV and OLSR, which are the two prominent MANET routing protocols using NS2 tool (Gowrishankar, et al, 2007). The study concluded that AODV requires less computational power to handle the network in comparison to the OLSR protocol. This performance of the protocol remained best with static traffic where total number of destination and source pairs was small for each host. The OLSR protocol proved efficient for dense networks where mobility of nodes was high and traffic was sporadic. Gibson studied four MANET routing protocols OLSR, AODV, TORA and DSR under different scenarios of traffic, node mobility and network size (Gibson, 2008). The OLSR protocol showed satisfactory performance in most scenarios that degraded with the increased network traffic and node mobility. The performance of AODV protocol remained better for medium sized network even with high network traffic. The study concluded that the DSR protocol is suitable for small networks where node mobility is low. Mahajan and Chopra examined AODV, OLSR and TORA protocols for 900 seconds using OPNET 14.5 simulator in a wireless/GSM voice channel network of size 2.5X2.5kms having 15 and 30 nodes. The study concluded that the overall performance of OLSR remained better in small and large networks (Mahajan, Chopra, 2013). 3. Performance Evaluation 3.1.

Experimental Setup

Most of the works that have evaluated efficiency of MANET protocols have used NS2 simulator, however, to take the advantage of OPNET software tool in handling networks with large number of nodes, this study uses OPNET simulator for comparing efficiency of three competing MANET protocols. Performance in terms of delay, network load and throughput of OLSR, DSR, and AODV MANET protocols have been studied under the three different scenarios of MANET network whose important parameters are given in table 1. Table 1: Network parameters used in model implementation C1

C2

C3

C4

C5

C6

Scenario I Basic

25

1500m X 2000m

10m/s

Voice

5 sec

Scenario II High Node Density

40

1500m X 2000m

10m/s

Voice

Scenario III High Node Mobility

25

1500m X 2000m

50m/s

Voice

C7

C8

C9

C10

C11

C12

C13

1024 Random Bytes

600 Sec

IEEE 802.11e

Wireless Physical Layer

Fragment

11 mbps

5 sec

1024 Random Bytes

600 Sec

IEEE 802.11e

Wireless Physical Layer

Fragment

11 mbps

5 sec

1024 Random Bytes

600 Sec

IEEE 802.11e

Wireless Physical Layer

Fragment

11 mbps

Column Legends: C1- Network Scenario, C2- Number of Nodes, C3- Network Area, C4- Mobility (Random Waypoint), C5- Data type, C6Packet Inter-Arrival Time, C7- Packet Size, C8- Destination IP Address, C9- Simulation Time, C10- Wireless LAN, C11- Network Interface, C12- Large Packet Processing, C13- Data Rate.

The network area in all three scenarios was set to 1500mX2000m and performance has been studies for 600 seconds. The individual parameters in the experimental implementation of the three studies MANET protocols were set to values as given in tables 2, 3 and 4.

[150]

Mehraj et al/COMMUNE – 2015

Table 2: OLSR metric parameters and their corresponding values

Willingness

Hello Interval

TC Interval

Neighbour Hold Time

Topology Hold Time

Duplicate Message Hold Time

Default

2.0 sec

5.0 sec

6.0 sec

15.0 sec

30.0 sec

Table 3: DSR metric parameters and their corresponding values

RequestTable Size

Maximum RequestTable Identifier

64 nodes

16

Non Initial Propagation Request Request Period Timer

Maximum Request Period

10 sec

0.5sec

Gratuitous Route Reply Timer

0.03 sec

Maximum Buffer Size

1 sec

Maintenance Max Maintenance Hold-Off Maintenance Nod Time Retransmission

50 pkt

0.25 sec

2

0.5

Table 4: AODV Parameters with corresponding values

Route Request Retries

Route Request Rate Limit

Hello Interval

Route Error Rate Limit

Time Out Buffer

5

10 pkt/sec

(Uniform) (1,1.1) sec

10 pkt/sec

2

Node Traversal Time

0.04 sec

Net Diameter

35

Allowed Active Hello Route Loss Time Out

2

3

Local Repair

Enabled

4. Results The delay recorded with respect to time in three scenarios i.e. scenarios I (basic), scenarios II (high node density), and scenarios III (high node mobility) is plotted in figures 1(a), 1(b) and 1(c).

(a)

(b)

(c) Fig. 1: Delay (in sec) vs time (in sec) in (a) Scenario I, (b) Scenario II and (c) Scenario III

In all three scenarios, delay in case of OLSR remained very low in comparison to other two protocols. The DSR and AODV protocols differed slightly when compared to each other. This difference was due to periodic updates in OLSR network wherein routes are determined and stored in nodes and therefore, no delay occurs at the time of transmitting [151]

Mehraj et al COMMUNE-2015

data packets. Further, due to proactive nature of OLSR, least packet delivery delay occurred in OLSR whereas for DSR and AODV more delay occurred, which approximately remained almost same for both. This justifies their reactive nature. AODV network experience most delay relative to other protocols. Irrespective of the network conditions, with the use of AODV, the delay remained highest. Table 5 shows the average delay recorded for three compared protocols for three different network scenarios. Table 5: Average delay measured with respect to time. Average Delay (in seconds) Protocol OLSR

Scenerio I

Scenerio II

Scenerio III

0.0005

0.0004

0.0007

DSR

0.018

0.017

0.014

AODV

0.12

0.0121

0.022

On comparing the delay in all three scenarios, least average delay of 0.0007seconds was recorded for OLSR protocol in Scenario III where node mobility was 50m/s and node density was 50 nodes. The highest average delay of 0.12 seconds was recorded for AODV in Scenario I where node mobility was 10m/s and node density was 25 nodes. The load recorded with respect to time in three scenarios i.e. scenarios I (basic), scenarios II (high node density), and scenarios III (high node mobility) is plotted in figures 2(a), 2(b) and 2(c).

(a)

(b)

(c) Fig. 2: Load (bits/sec) vs time (in sec) in (a) scenario I, (b) scenario II, and (c) scenario III

In all three scenarios, load for DSR protocol remained very low in comparison to other two protocols. The OLSR and AODV protocols differed on load slightly when compared to each other. This difference was due to source routing nature of DSR. In DSR, updates are only made when needed in network. This reduces the network overhead. Further, high delay was experienced in AODV, and medium in case of OLSR. From the graphs it is observed that the load fluctuates more in AODV protocol based network throughout the time line where as it is more or less same for DSR and OLSR protocols.Table 6 shows the average load recorded for three compared protocols for three different network scenarios. Table 6: Average load measured with respect to time. Average Load (bits per sec) Protocol OLSR DSR AODV

Scenerio I

Scenerio II

Scenerio III

600 305 1711

935 208 2421

805 304 1607

[152]

Mehraj et al/COMMUNE – 2015

On comparing the load in all three scenarios it was found that maximum average load of 2421 bits/sec was recorded for AODV protocol based network in Scenario II which has an increased node density of 40 nodes and the minimum average load of 208 bits/sec was recorded for DSR protocol based network in scenario II. The throughput recorded with respect to time in three scenarios i.e. scenarios I (basic), scenarios II (high node density), and scenarios III (high node mobility) is plotted in figures 3(a), 3(b) and 3(c).

(a)

(b)

(c) Fig. 3: Throughput (bits/sec) vs time (in sec) in (a) scenario I, (b) scenario II, and (c) scenario III

In all three compared scenarios, throughput for AODV protocol based networks was higher in comparison to DSR and OLSR protocol based networks.DSR protocol failed in comparison to the other two protocols whereas OLSR performed at moderate levels. The performance of OLSR is almost similar to AODV in high node density networks. The DSR protocol almost stops working when more nodes are in the network suggesting that it cannot handle dense MANETs. If all the three performance metrics are considered, OLSR performs with consistency. Table 7: Average throughput with respect to time. Average throughput (bits per sec) Protocol OLSR DSR AODV

Scenerio I 13750 465 18877

Scenerio II 33072 48 39405

Scenerio III 14885 464 20255

On comparing throughput in all three scenarios, the maximum average throughput of 39405 bits/sec was recorded for AODV based networks in high node density. The minimum average throughput of 46 bits/sec was recorded for DSR based networks in scenario II. 5. Conclusion The study considering low as well as very high mobility concludes that performance of DSR protocol is consistent with increase in mobility but degrades when the number of nodes increases. This is in conformity with (Gibson, 2008) however; it differs from (Sumit, 2013) because results are influenced by various network parameters such as network size, node density, and node mobility. With AODV protocol maximum throughput was achieved even when mobility of nodes is increased by five times and the density of nodes is increased by 60%. The OLSR protocol offered a consistent performance throughout the simulation time providing an acceptable delay and load throughout varied scenarios of increased node density and mobility. The DSR protocol performed well for delay and load but lagged far behind AODV and OLSR protocols in throughput. This was due to source route information carried by packet headers in DSR protocol based networks. However, its performance in small networks with low mobility remained considerable.

[153]

Mehraj et al COMMUNE-2015

References Al-Maashri, A., Ould-Khaoua, M., November 2006. Performance analysis of MANET routing protocols in the presence of self-similar traffic, In Proceedings of the 31st IEEE Conference on Local Computer Networks, 14-16 2006, pages pp. 801-807, Tampa, Florida, USA. Broch, J., Maltz, D. A., Johnson, D. B., Hu, Y. C., Jetcheva, J., 1988. A Performance Comparison of Multi-Hop Wireless Network Routing Protocols, Proceedings of the Fourth Annual ACM/IEEE International Conference on Mobile Computing and Networking (MobiCom'98), Dallas Texas, USA, pp. 25-30. Gowrishankar, S., Basavaraju, T. G., Singh, M., .Subir Kumar Sarkar, 2007. Scenario based Performance Analysis of AODV and OLSR in Mobile Ad Hoc Networks. Proceedings of the 24th South East Asia Regional Computer Conference, November 18-19, 2007, Bangkok, Thailand. Perkins, C., Belding-Royer, E., Das, S., July 2003. Ad hoc On-Demand Distance Vector (AODV) Routing. IETF. RFC 3561. tools.ietf.org/html/rfc3561. Retrieved 2010-06-18. In Nokia Research Center, University of California, Santa Barbara and University of Cincinnati. Shiva Kumaran, R., Rama Shankar Yadav, Karan Singh, March 2007. Multihop wireless LAN. HIT haldia. International Journal of Computer Science and Security (IJCSS), Volume 1, Issue 1, CSC Journals, Kuala Lumpur, Malaysia. Sumit Mahajan, Vinay Chopra, 2013. Performance Evaluation of MANET Routing Protocols with Scalability using QoS Metrics of VOIP Applications, International Journal of Advanced Research in Computer Science and Software Engineering, vol. 3, no. 2, pp. 150-156, 2013. Tadeusz, A., Wysocki., Arek Dadej., Beata J., Wysocki., 2006. Advanced Wired and Wireless Networks”, Springer Science & Business Media, 17Jan-2006. Tanenbaum, A. S., 2010. Computer Networks, fifth ed. Printice Hall, Upper Saddle River, NJ. Thomas Clausen, Phillippe Jacquet, Adjih, C., Laouiti, A., Minet, P., Muhlethaler, P., Qayyum, A., Viennot, L., 2009. The optimized routing protocol for MANETs: protocol specification. Of ICIS 2009 2nd International Conference on Interaction Sciences: Information Technology, Culture and Human, Seoul, Korea, 24-26 Nov, 2009, pp, 412-417. Wilford Gibson LOL, 2008. An Investigation of the Impact of Routing Protocols on MANETs using Simulation Modelling, Masters Degree Thesis, Master of Computer and Information Sciences, Auckland University of Technology, New Zealand.

[154]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Binarization of Natural Science Images Sukhdev Singha*, Dharam Veer Sharmab a

Department of Computer Science, Multani Mal Modi College,Patiala, India b Department of Computer Science,Punjabi University Patiala, India

Abstract The text extraction from images is the prominent area in digital image processing. It has number of applications like Navigation system for visually impaired persons, Automatic text extraction from document images, and Number plate detection to enforcement traffic rules, etc. The Binarization is a key process in any text extraction system from images. The Binarization is carried out to convert grayscale or colored image into Binary images (Black/White) images. Where one color represents background and other represents foreground. The various algorithms are available in the literature for Binarization of documents images and only few algorithms are available for natural scene images. The present study includes twelve well known algorithms which are previously used for document binarization. These algorithms are test on natural scene images to find out suitability of these algorithms for binarization of natural scene images.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Binarization; Niblack Binarization technique; Souvola Binarization technique; Morphological Binarization; Bernsen’s Binarization; Feng’s Binarization; Sliding WindowMean Thresholding; Sliding Window Median Thresholding

1.

Introduction

Binarization is carried out to convert colored or grey scale image to monochromatic or binary images. The binarization algorithms are classified into two categories (Ezaki et al, 2004) based on discontinuity or similarity of grey values. The algorithms, using discontinuity, segment an image based on abrupt changes in grey level, whereas algorithms using similarity are based on thresholding, region growing, region splitting and merging. In case of thresholding algorithms, the conversion is based on finding a threshold grey value and deciding whether a pixel having a particular grey value is to be converted to black or white. Usually within an image the pixels having grey value greater than the threshold is transformed to white and the pixels having grey value lesser than the threshold is transformed to black. The document binarization is different from natural scene binarization on the basis of complexity, shapes in background, foreground text and size, orientation and illumination of text. 2.

Binarization Techniques

In the current study we have applied well known techniques on natural scene image which are discussed one by one below: 2.1. Otsu Binarization Otsu Binarization (Vishwanatha, Jawahar, 2001) technique is Global thresholding technique in which single threshold value is used to Binarized the whole Image. The algorithm assumes that the image is to represent into two classes of pixels i.e. foreground and background. The optimum thresholding value is calculated such that intra-class variance is minimal.

* Corresponding author. Tel.: +91 9814 610887. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Singh and Sharma/COMMUNE – 2015 Table1. Otsu Binarization

Original Image

Otsu’s Binarization

Original Image

Otsu’s Binarization

2.2. Niblack Binarization Technique Niblack (Bhattacharya et al, 2009) is local Binarization method based on sliding window in which thresholding is computed on each and every pixels of the image. In this technique local mean and standard deviation is calculated within rectangular region to find out thresholding value for central pixel surrounded by neighborhood pixels. The thresholding is decided by the following formula. T(x,y)=m(x,y)+K* s(x,y), where m(x,y) and s(x,y) are the average of intensity values in local area of window and standard deviation for pixel intensity values of the window. Table2. Niblack Binarization

Niblack’s Method

Original Image

Original Image

Niblack’s Method

2.3. Souvola Binarization Technique Sauvola’s Binarization (Yoon et al, 2009) is a locally adaptive thresholding of grey scale image. According to the algorithm, local mean and standard deviation are calculated on WxW window around a pixel. Let us consider T threshold for pixels of image and Mw (i ,j),σw(i,j) are considered as local mean and standard deviation on local window of size WxW for pixel location (i,j). The value of T can be calculated as Tw,k(i,j)= Mw(i,j) *(1+ K (σw (i,j) /R – 1)) Where, Mw (i,j) is called local mean for pixels in window of size WxW, σw(i,j) represents standard deviation for pixels in window of size of size WxW, R,k are constant parameters(also called control Factors) to control sensitivity to noise. Table3. Sauvola’s Binarization

Original Image

Sauvola’s Method

Original Image

Sauvola’s Method

2.4. Bernsen’s Binarization Bernsen’s Binarization (Sezgin, Sankur, 2004) introduced local thresholding method. In this method we compute local minimum and maximum for neighborhood around each pixel f(x,y) ϵ [0,L-1] and use median of two as the threshold for each pixel in image with reference of size of window. The sliding window is mask of size NxN centered

[156]

Singh and Sharma/COMMUNE – 2015

pixel is coincide with the pixel in focus. We keep on slide window from one pixel location to another pixel position in the image. T(x,y)= (Plow +Phigh)/2 , If Phigh - Plow ≥ L T(x,y)= GT , If Phigh - Plow < L Where Plow is low value of gray pixel value in the NXM size window, Phigh is low value of gray pixel value in the NXM size window. GT stands for Global Thresholding using otsu’s Binarization. G(x,y)=(Fmax(x,y)+Fmin(x,y))/2 B(x,y)= 1 if f(x,y)
Original Image & Binarized Image

2.5. Feng’s Binarization Feng’s Binarization (Pavlos et al, 2008) is based on local thresholding in which in place of single sliding window two local windows are used. The values of local mean(MeanLocal-m)and minimum gray value (MinGray-M) are calculated in the primary local window with reference to pixels located in the neighborhood. The values of standred deviation are calculated in the both the Windows. The following formula is used to calculate threshold: Tfeng =(1- α1)*m+α2*(S/Rs)*(m-M)+ α3*M

Where α2=k1*(S/Rs)ϒ , α3=k2*(S/Rs)ϒ, value of ϒ=2,

S is Standard deviation within local windows standard deviation of secondary window, m is local mean within primary window, M is minimum gray value within primary window. Table5. Feng’s Binarization

Original Image & Binarized Image

2.6. Sliding Window Mean Thresholding Sliding window mean thresholding (Chitrikala, Manjula, 2010) is performed by using calculating mean of gray scale value within local window. The following equation is used to find local mean using sliding window. pixel={ Pixel value > Local Mean-C ? Object: Background}, usually c=0 or any value to control noise level.

[157]

Singh and Sharma/COMMUNE – 2015

Table6. Sliding window mean thresholding

Original Image & Binarized Image

2.7. Sliding Window Median Thresholding Sliding window median thresholding (Xiangrong, 2004) is performed by using calculating mean of gray scale value within local window. The following equation is used to find local mean using sliding window. Pixel= {Pixel value > Local Median-C? Object: Background}, Usually c=0 or any value to control noise level. Table7. Sliding window Median thresholding

Original Image & Binarized Image

2.8. Sliding Window MidGray Thresholding Sliding window median thresholding (Stathis et al, 2008; Ntirogiannis et al, 2012) is performed by using calculating mean of gray scale value within local window. The following equation is used to find local mean using sliding window. Pixel= {Pixel value > (Local Maximum+Median)/2 –C ? Object: Background}, usually c=0 or any value to control noise level. Table 8. Sliding window MidGray thresholding

Original Image & Binarized Image

2.9. Wolf’s Binarization Technique Wolf’s Binarization technique (Solihin, Leedham, 1999; Kim et al, 2002) is implemented by calculating stranded deviation and mean within local window and over whole image. The following equation is used to find the threshold over local window. Twolf=(1-K)*m+K*M+K*S/R(m-M) Where k=0.5(Fixed), M- Minimum Gray value, m-mean local, M-mean global, S-Standard Deviation local, R Standard Global

[158]

Singh and Sharma/COMMUNE – 2015 Table 9. Wolf’s thresholding

Binarization of Natural scene Images using Wolf’s Binarization Original Image & Binarized Image

2.10. White and Rohrer’s Binarization White and Rohrer’s Binarization (Li, Kwok, 2004) technique is used to calculate threshold value for Binarization of image by using bias parameters as multiple to pixel value under consideration. The concept of sliding window is used to find out mean value of neighborhood pixels around centered pixel. mlocal(x,y)
Original Image & Binarized Image

2.11. T.R.Singh and et. al ’s Binarization T.R Singh (Poignant et al, 2012; Ntirogiannis et al, 2008) introduced local thresholding computation by calculate local mean and mean deviation by using following equation. T(x,y)=m(x,y)[1+K*(δ(x,y)/(1-δ)-1)] Where δ (x,y)=I(x,y)-m(x,y), k is a bias and k € [0,1] Table10. T.R Singh ‘s thresholding

Original Image & Binarized Image

3. Analysis Parameters Comparison of these binarization algorithms prove to be difficult task since there is no efficient way to compare the results. In literature several papers are found that use precision and recall analysis. But in case of natural scene image, the background is too complex that it is not easy to separate object from it. There is lot of text like objects like tree leaf, branches, window, etc. In current study the following methods are used (1) Visual Observation and Noise Detection Parameters (Minetto et al, 2010; Pan et al, 2011; Chandraskaran et al, 2012) MSE: (Mean Square Error) , SNR(Signal to Noise Ratio), PSNR(Peak Signal to Noise Ratio) Where the PSNR value is most important which looks at how many pixels in the text image differ from Ground truth image values and find quantity of the pixels. Higher the value of PSNR indicates better result. The more than two thousand images are captured using Nikon D3200 Digital Camera. The images covered different location that contains sign board, notice board, banners, and text written on wall and vehicles in city Patiala. The text present in these boards is in Gurmukhi Script. The following is the result analysis of image of Bus: [159]

Singh and Sharma/COMMUNE – 2015 Table11. Analysis of thresholding techniquies Method Otsu Niblack Souvola Morphological Bernsen Feng Sliding Mean Median MidGray Wolf White&Rohrer T.R.S.

4.

AQE 135.689 114.420 90.597 85.527 228.964 140.892 238.590 223.0454 280.633 142.253 356.109 277.228

SDQE 197.194 171.962 161.548 169.585 194.922 199.067 201.560 187.797 189.704 197.388 158.328 185.397

MSE 57297.410 42662.983 34305.802 36074.220 90419.378 59478.553 97552.029 85017.161 114742.888 59198.257 151882.32 111227.908

SNR -8.445 -5.496 -3.315 -3.8184 -13.007 -8.818 -13.766 -12.391 -15.389 -8.7716 -18.1937 -15.078

PSNR 12.2512 15.200 17.380 16.878 7.6892 11.877 6.9299 8.30530 5.306 11.924 2.5027 5.6180

Detail Analysis and Conclusion

The twelve algorithms from literature has been considered for present research work and their performance has been evaluated using Visual Observation and Mathematical noise and Error calculation methods such as MSE AQE,SDQE,SNR and PSNR. The following observations have been noticed to find out best suitable algorithm. The Otsu’s method is good Global Binarization method for image where background is uniformly illuminated. The Niblack’s, Sauvola’s are good Binarization techniques which dynamic in nature but suitable size of window and parameters are required. The Feng's Binarization with two sliding window gave good result with 15 x 15 size windows but we need to determine empirical parameters. The TR Singh’s method is gives good result even for small size of window for example w=7 whereas. Acknowledgement We take this opportunity to express our gratitude to University Grants Commission (UGC), New Delhi for providing financial assistance in the form of Minor Research Project to complete the present research work. References Ezaki Nobuo, Marius Bulacu, Lambert Schomaker, 2004. Text Detection from Natural Scene Images: Towards a System for Visually Impaired Persons, in the proceedings of 17th International Conference on Pattern Recognition, Cambridge ,UK, vol. 2, pp.683-686. Yoon Hou-sub, Hong-Chang Lee, Jaeyeon Lee, 2009. Automatic Number Plate Detection for Korean Vehicles”, in the proceedings of the 2nd International Conference on Interaction Sciences: Information Technology, Culture and Human, Korea, pp.1191-1195. Sezgin M., Sankur B., 2004. Survey over Image Thresholding Techniques and Quantitative Performance Evaluation, Journal of Electronic Imaging, pp. 146-165. Pavlos Stathis, Ergina Kavallieratou, Nikos Papamarkos, 2008. An Evaluation Technique for Binarization Algorithms, Journal of Universal Computer Science, vol. 14, pp. 3011-3030. Bhattacharya, Parui U., Mondal S.K., 2009. Devanagari and Bangla Text Extraction from Natural Scene Images, in the proceedings of 10th International Conference on Document Analysis and Recognition, Barcelona, Spain, pp. 171-175. Vishwanatha Kaushik, Jawahar C. V., 2001. Detection of Devanagari Text in Digital Images using Connected Component Analysis, in the proceedings of the National Conference on Document Analysis and Recognition, Mandya, India, pp.41-48. Chitrakala Gopalan, Manjula D., 2010. Sliding window approach based Text Binarisation from Complex Textual images”, International Journal on Computer Science and Engineering, pp. 309-313. Xiangrong Chen, Alan L., Yuille, 2004. Detecting and Reading Text in Natural Scenes, in proceeding IEEE, pp.67-74. Solihin Y., Leedham C.G., 1999. Integral ratio: A New Class of Global Thresholding Techniques for handwriting Images, Trans Pattern ecognition and Machine Intelligence, pp.761-768. Li, S., Kwok, J. T., 2004. Text extraction using edge detection and morphological dilation, in proceedings of International Symposium on Intelligent Multimedia, Video, and Speech. pp. 330-333, IEEE. Poignant, J., Besacier, L., Quenot, G., Thollard, F., 2012. Text detection in videos to person identification in the proceedings of International Conference on Multimedia and Expo. pp. 854-859. Minetto, R., Thome, N., Cord, M., Fabrizio, J., Marcotegui, B., 2010. A multiresolution system for text detection in complex visual scenes, in proceedings of International Conference on Image Processing (ICIP), pp.3861-3864. Pan, Y. F., Hou, X., Liu, C. L., 2011. A hybrid approach to detect and localize texts in natural scene images, IEEE Transactions Image Processing, pp.800-813. Chandrasekaran R., Chandrasekaran R.M., Natarajan P., 2012. Text Localization and Extraction using Support Vector Machine and morphological functions, IEEE, pp.55-60. Uddin,Sultana, M.Rahman, Busra, 2012. Extraction of text from Scene Image using Morphological based approach, IEEE, pp.876-880. Kim S., Kim D., Ryu Y., Kim G., 2002. A Robust License-Plate Extraction Method Under Complex Image Conditions, in proceedings of International Conference on Pattern Recognition, Vol. 3, pp. 216-219. Ntirogiannis, K., Gatos, B., Pratikakis, I., 2008. An objective evaluation methodology for document image binarization techniques in proceedings of 8th IAPR Workshop on Document Analysis Systems. Stathis, P., Kavallieratou, E., Papamarkos, N., 2008. An Evaluation Technique for Binarization Algorithms. J. Univers. Comput. Sci.,pp. 3011–3030. Ntirogiannis, K., Gatos, B., Pratikakis, I., 2012. Performance Evaluation Methodology for Historical Document Image Binarization, IEEE Trans. Image Process. 22(2), pp. 595–609.

[160]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16-18March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Effect of Convolutional Encoding on Bit Error Rate (BER) for Image Transmission using Multiple Input Multiple Output Orthogonal Frequency Division Multiplexing (MIMO-OFDM) System Over Fading Channels Javaid A. Sheikh*, Shabir A. Parah, Uzma, Aijaz, Tawseef, Farah, Sanna, Aiman, G. Mohiuddin Bhat Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India

Abstract Multiple Input Multiple Output Orthogonal Frequency Division Multiplexing (MIMO-OFDM) has been the topic of interest for various researchers throughout the world owing to the ever increasing applications of high speed wireless communication. MIMOOFDM came into existence to overcome the various flaws suffered by cellular and other wireless communication in view of the multipath fading. OFDM is a very competent block modulation scheme which supports multimedia communication at higher data rates with elimination of Inter Symbol Interference (ISI) and Inter Carrier Interference (ICI). In addition to this it can adjust more number of users and thus leads to better spectral usage. MIMO on the other hand is the antenna configuration which exploits the diversity property. MIMO is used in conjunction with OFDM to achieve all the possible gains of both the technologies. The performance of any communication system is determined in terms of various parameters including Bit Error Rate (BER). BER gives the error rate by comparing the received data with transmitted data. This paper focuses on the BER of the MIMO-OFDM system over multipath fading channel, which has been developed in MATLAB. Convolutional encoding as a method of BER improvement has been used and discussed in this paper. The entire work has been carried out on an image signal making it relevant to the modern communication world.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Orthogonal Frequency Division Multiplexing; Multiple Input Multiple Output; Spatial multiplexing; Antenna Diversity; Trellis Structure; Convolutional Encoding

1. Introduction The field of wireless communication is facing pressing demands from all quarters of life for enhancement in the data rates with equally good quality of service. In order to meet all these demands there has been evolution on larger scale in techniques including modulation schemes as well as antenna arrangements. Of various available options MIMO-OFDM scheme seems to be strong candidate for the said features. Orthogonal Frequency Division Multiplexing (OFDM) can be considered as the modulation as well as multiplexing technique. In OFDM the basic problems like Inter Symbol Interference (ISI) and Inter Carrier Interference (ICI) are mitigated by sending the data, to be transmitted, in parallels there by increasing the data rates without actually decreasing the time duration. Thus reduces the ISI problem. MIMO technology on the other hand is related to the antenna configuration. Using more than one antenna at both the ends of the communication (wireless) channel is known as Multiple Input Multiple Output Technology (Javaid et al., 2012). MIMO technology exploits the multipath propagation which otherwise is a problem in conventional communication system. MIMO creates the channel matrix unlike the previous concept of channel vector. MIMO configuration also increases the capacity without increasing the power and bandwidth.

*

Corresponding author. Tel.: +91 9419 090554 E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Sheikh et al/COMMUNE – 2015

2. MIMO-OFDM Systems The OFDM modulation in combination with MIMO system leads to a strong system that has the capability to effectively eliminate ISI and fading and fulfils the need for high throughput. The enhanced throughput of MIMO is basically the translation of increased channel capacity with the help of appropriate coding that precedes transmission. The coding process effectively adds extra bits to the transmitted data that helps the receiver in deciding whether errors occurred during transmission. In MIMO-OFDM system the data is mapped into modulation symbols. An NxM matrix is formed through the serial to parallel conversion of N successive modulation symbols, where N and M represent the FFT size and the number of OFDM symbols in each slot, respectively. The OFDM symbols are then space time encoded for each row using two successive symbols for Almouti’s scheme and four successive symbols for the quasi orthogonal scheme. A preamble is inserted at the beginning of each slot for channel estimation. The size of the preamble should be as small as possible to avoid the serious reduction of transmission efficiency. To this end, we assign one OFDM symbol for preamble and encode the preamble in the frequency domain. The figure 1 given below shows the MIMO-OFDM transmitter and receiver. “The upcoming IEEE 802.11n wireless local area networks (WLANs) standard uses MIMO architecture along with OFDM and LDPC coding. The standard has the ability to provide a stunning throughput of up to 600 Mbps as opposed to 54 Mbps provided by the older, non-MIMO IEEE802.11a standard” (Upena, 2009).

Fig. 1. MIMO-OFDM Transmitter

3. Channel Coding Coding theory deals with transmission of data over noisy channels by adopting various source and channel coding/decoding schemes. The subject of coding theory was first proposed by Claude Shannon in 1948 with the publication of his paper, “A mathematical Theory of Communication.” He also worked on entropy, source coding and capacity of the channel, etc. Afterwards, Huffman and Hamming also worked on this theory. Due to the growth of digital communication, coding theory has grown into a discipline intersecting mathematics and engineering with application to almost every area of communication, such as satellite, cellular telephony, Internet, CD recording, and data storage. “Channel coding refers to the class of signal transformations designed to improve communication performance by enabling the transmitted signals to better withstand the effects of various channel impairments, such as noise, interference and fading” (Ahmed, 2010). Some of the error correction techniques are error detection, error correction, data acknowledgment and data resend. Among these the most commonly used is the error correcting codes. There are three main classes of error correcting codes: Block codes (which include Hamming codes, Bose-ChaudhuryHocquenghem(BCH) codes and Reed Solomon codes), Convolutional codes and Turbo codes( which include Block/product turbo codes(BTC or PTC), Convolutional turbo codes(CTC)).These codes are easy to generate and their hardware complexity is also low as compared to other error control methods which have been mentioned above. All the mentioned coding techniques have their own field of application and advantages. In our simulation work convolutional encoding with soft viterbi decoding algorithm has been used. 4. Convolutional Encoding Convolutional encoding as mentioned earlier belongs to the class of feedback based channel coding techniques. “A convolutional encoder is a discrete linear time invariant system. Every output of an encoder can be described by its own transfer function, which is closely related to a generator polynomial” (Upena, 2009). Convolutional encoder generates [162]

Sheikh et al/COMMUNE – 2015

the code word using two parameters: the code rate (k/n), which is described as a ratio of the number of bits into the said encoder (k) to the number of output bits (n) and the constraint length. The constraint length, K, denotes the "length" of the convolutional encoder, i.e. how many k-bit groups are provided to the encoder to get the output symbols. Convolutional codes can be generated mathematically with the help of generator polynomial. Convolutional code generation can be represented using a state diagram also. To convolutionally encode the data we pass the information sequence through a finite state shift register with L memory elements each holding one input bit. Unless otherwise mentioned, all memory registers start with 0 value. The encoder consists of modolu-2 adders, and n generator polynomial, one for each adder. The figure 2 given below shows the schematic of convolutional encoder. An input bit mj is fed into the rightmost register. Using the generator polynomials and the existing values in the remaining register, the encoder outputs n bits. Then, bit shift, shifts all the register values to the left (m j shifts to mj-1, mj-1 shifts to mj-2) and next input is fed. If there are no left over input bits, the encoder continues output until all registers have returned to the zero state. There are various algorithms for decoding the data of which Viterbi algorithm is mainly used. The viterbi decoding algorithm was discovered and analyzed by Viterbi in 1967. The Veterbi decoder is either the hard decision viterbi decoding or soft decision viterbi decoding. The performance of soft decision viterbi decoding has been found to be more satisfactory but has the disadvantage of system complexity where as hard decision viterbi decoding is simple in its construction. In this paper soft decision viterbi decoding has been used for achieving better performance.

Fig. 2. Convolutional Encoder

5. Simulation and Results The proposed work is concerned with the design and development of real time wireless communication system for transmission of image (bmp image) over fading channel using MIMO-OFDM. The MATLAB programme developed prompts the user for various input values required for the simulation including the input image, the size of the IFFT, number of carriers to be used, modulation scheme, amplitude clipping value, Signal to noise ratio (SNR). After having received the necessary inputs the programme reads data from an input (image) file and forms an a-by b matrix where a and b is the height and width (in pixels) of the image respectively The data read is then transformed into serial data which is then encoded using convolutional encoding with Trellis structure .The encoded data, as per the PSK order chosen by the user, is arranged into the corresponding symbol size. The data is then put into different frames and consequently becomes the desired baseband signal for OFDM modulator, which has been implemented by using Inverse Fast Fourier Transformer (IFFT).After the necessary signal processing the data is fed to Orthogonal Space Time Block Encoder (OSTBC), which encodes the data using orthogonal space time block code. “This encoder maps the input symbols block-wise and provides the concatenated output code-word matrices in the time domain “(Abdulraham, 2011). At the receiving end, the reverse of all the processes, that were performed at the transmitter, are used to get the original data back, of course with some modifications that will be there due noise that gets added due to the channel used in the simulation .First of all the encoded data is received by “OSTBC combiner which combines the input signal from all the receivers and channel estimate signal to extract the soft information of the symbols encoded by OSTBC encoder” (Abdulraham, 2011). Thereafter the OFDM demodulator demodulates the data frame wise as was modulated by the OFDM modulator. Maintaining the coherence the data is then decoded for convolutional encoding and in this simulation work the decoding is done by soft decision viterbi decoder because of its above mentioned advantage. The programe then contains certain image processing steps to get back the image in its original dimensions and format. Demodulated data is compared to the original baseband data to find the total number of errors. Dividing the total number of errors by total number of demodulated symbols, the bit-error rate (BER) is found (Paul, 2010). The bit error rate has been calculated with and without using convolutional encoding to study the effect of coding. The results obtained are given below. The results show a considerable decrease in BER with channel coding like convolutional coding.

[163]

Sheikh et al/COMMUNE – 2015

Fig. 3. Image used for transmission

Fig. 4. Image recieved after transmission

Fig. 5. BER plot for BPSK scheme

6. Conclusion and Future Work Owing to the various applications and advantages of MIMO-OFDM systems, a successful attempt has been made in this paper to develop the MIMO-OFDM system model with considerable options of making changes in it to see the effect of various parameters on its performance. BER, being the important parameter of performance check, was studied in this work and for its reduction channel coding (convolutional encoding) has been used. The results obtained are satisfactory although authors are contended that it is open for further improvement. The results shown have been obtained on logical image (512x512), with 2048 IFFT size and 1024 number of carriers using BPSK modulation at 20 SNR. References Javaid A. Sheikh, Shabir A. Parah, G. Mohiuddin Bhat “Orthogonal Variable Spreading Factor (OVSF) based image Transmission using Multiple Input Multiple Output Orthogonal Frequency Division Multiplexing (MIMO-OFDM) System” IEEE sponsored International Conference on Communications Devices and Intelligent systems” Jadavpur University, Kolkatta (in press). Upena Dalal, “Wireless Communication”, Oxford University press, 2009. Ahmed B.M Abdurrhman, “Error Resilient Video Communication Using High Level M-QAM”, Ph.D. thesis, Department of Computing, University of Bradford, 2010. Dr. Abdulraham Ikram Siddiq,’’Variable length cyclic prefix OFDM using Multipath delay tracking”, Tikrit Journal of Engineering Sciences, June 2011. Paul Guanning Lin,’’OFDM simulation in MATLAB”, Graduate dissertation, California Polytechnic State University, 2010. www.mathworks.com

[164]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

FPGA Evaluation of Wave Front Allocator for Crossbar based On-Chip Switches Liyaqat Nazir*, Roohie Naaz Mir National Institute of Technology, Srinagar, India.

Abstract Multiprocessor system-on-chip (MP-SoC) are emerging as an important trend for SoC design. Network on chip (NoC) has been proved to be efficient in handling challenging ultra-high data rates in such systems. NoC uses many high performance internetworking protocol routers that are based on crossbar switch matrix. A scheduling algorithm is used to configure the crossbar switch, deciding the order in which the packets in NoC will be routed. The scheduling algorithm must be able to keep up with the high switching rates presented by the ultra-high link rates in SoC. This paper discusses the synthesis and implementation of the Wrapped wave front allocator on varying port density request vectors. The paper will help Network-on-chip router designers to scrupulously choose the appropriate scheduler for their design.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Network-On-Chip; Arbiter; Scheduling; FPGA; Crossbar Switch; Router; Wave Front Allocation.

1. Introduction Multi-core system on chip-based systems has provided an important design challenge for SoC designers. Moreover in order to have a complete system-on-chip integration, a significant amount of reduction in design techniques and topologies is required as the current SoC does not scale to such dimension and complexity (Banerjee, A. etal 2007). Network-on-chip has been proposed as a solution to such a problem (Sudhir N. Shelke et al 2014). NoC is basically a design paradigm that has attracted lot of attention by providing higher bandwidth and higher performance architectures for communication on chip. NoC can provide simple and scalable architectures if implemented on reconfigurable platforms (Osterloh, B. et al 2008). Network on chip offers a new communication paradigm for system on chip (SoC) design (Maher, A. et al). Many processing elements of SoC are connected through Network-on-chip (NoC) routers which are arranged in some regular fashion such as Mesh, linear, torrous, 2D, 3D type of topologies (W. J Dally et al 2003) as shown in figure 1. To achieve high performance router should provide high bandwidth and low latency (Ahmaed, A. B. et al 2010). Although the performance of the NoC is normally seen by its throughput, which is defined by the network topology, router throughput and the traffic load at the network (Anderson, T. et al). The router throughput is determined by the critical path of the data path units in the router and the efficiency of control path units (Karol, M.et al, 1988; McKeown, N. et al, 1996; Yuan-Ying Chang et al, 2013; Phanibhushana, B. et al, 2011). The data paths of the on-chip router comprises of buffers, VC and switching fabric and the control paths of on-chip communication routers are largely composed of arbiters and allocators (J. Guo et al, 2005). Allocators are used to allocate virtual channels and to perform matching between a groups of resources on each cycle (Dally, W. J and Brain Towels, 2003; Mckeown, N., 1995; Lee, K. et al, 2003). As the flits/cells arriving at the fabric contend for access to the fabric with cells at both input and output A switch allocator must perform matching between the input units and the output ports of the router so that at most one flit from each input port is selected and at most one flit destined to each output port is selected (Gupta, P. et al, 1999).

*Corresponding Author. Tel.: +91 9796 353284. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Liyaqat and Mir/COMMUNE – 2015

This work, therefore, focuses on the implementation of wrapped wave front allocator with varying port density request vectors. The port density we selected in this work is 9-bit, 12-bit and 16-bit request vector. In this paper we carried out implementation and synthesis of various allocation algorithms used in allocators. The rest of the paper is organized as follows. Section 1 gives the introduction. Section 2 gives some background about Wavefront allocators used in NoC router. Section 3 talks about synthesis and implementation. Section 4 provides the conclusion and future scope of the work carried out and finally references are listed in the end.

Fig. 1. Various Topologies of Network-on-Chip.

2. Wavefront Allocator Wavefront allocation (WFA) arbiter is a high performance arbitration scheme for crossbar switches which requires a two dimensional arbitration that incorporates a rotating priority design to Incorporate fair arbitration. The algorithm distributes the arbitration computing over all of the crossbar fabric nodes. Any poor allocation of resources results in long transmission crossbar’s resources. The wavefront allocator, unlike the separable allocators arbitrates among requests for inputs and outputs simultaneously. The structure of the 3×3 wavefront allocator is shown in figure 2. The wavefront allocator works by granting row and column tokens to a diagonal group of cells, in effect giving this group priority. A cell with a row (column) token that is unable to use the token passes the token to the right (down), wrapping around at the end of the array. These tokens propagate in a wavefront from the priority diagonal group, hence the name of the allocator (Jose, G. et al, 2003). If a cell with a request receives both a row and a column token, either because it was part of the original priority group or because the tokens were passed around the array, it grants the request and stops the propagation of tokens. To improve fairness, the diagonal group receiving tokens is rotated each cycle. However, this only ensures fairness. In an n×n arbiter, diagonal group k contains cells xij such that (i+j) mod n = k. Thus, for example, in the 3×3 allocator of Figure 2, priority group 0, selected by signal p0, consists of cells (cij) x00, x13, and x22. Because each diagonal group must contain exactly one cell from each row and from each column, all wave front allocators must be square. A non-square wave front allocator can be realized, however, by adding dummy rows or columns to square off the array. The priority groups in an allocator need not be diagonals as long as they each contain one element of each row and column. When the cell is a member of the priority group, signal priority is asserted, which generates both a row token xpri and a column token ypri via a set of OR gates. If the cell is not a member of the priority group, row tokens are received via signal xin and column tokens via yin. If a cell has a row token xpri, a column token ypri and a request.

Pri

Yin

Xout

Xin Cij

Yout Fig. 2. (a) Wavefront Allocator (b) Individual Node of Wavefront Allocator.

[166]

Liyaqat and Mir/COMMUNE – 2015

3. Synthesis and Implementation 3.1.

Methodology

The implementation of this work is targeted for Virtex 5 FPGA family. Only VLX series has been considered as it is appropriate for general logic applications. The implementation is carried out for input request vector of 9-bit, 12-bit and 16-bit corresponding to the 9-bit, 12-bit and 16-bit grant output as shown in figure 3. The parameters considered are resource utilization, timing and average power dissipation. Resource utilization is considered in terms of on chip FPGA components used. Timing refers to the clock speed of the design and is limited by the setup time of the input and output registers, clock to the output time associated with the flip flops, propagation and routing delay associated with critical paths, skew between the launch and capture register. Timing analysis is carried out by providing appropriate timing constraints. The average power dissipation mainly composes of dynamic power dissipation besides negligible static power dissipation. In order to provide a fair comparison of the results obtained the test bench is provided with same clock period and same input statistics in all the topologies implemented. The constraints related to the period and offset are duly provided in order to ensure complete timing closure. The design synthesis, implementation and mapping have been carried out in Xilinx ise 12.4 (Xilinix Inc). Power metrics are measured using Xpower analyzer. The simulator data base is observed to obtain clock period and operating frequency of the implemented design. 3.2.

Experimental Results

Resource utilization is actually FPGA resource utilization and Table 1 gives the comparison of the resource utilization of various port densities of request vector of WFA. The various resources utilized and the total resource utilization are plotted in figure 4. It can be seen that allocation algorithm implemented using 16-bit p0 p0 00

10

20

01

11

21

02

12

22

03

13

23

00

10

20

20

01

11

21

21

02

12

22

22

03

13

23

23

p1 p1 00

10

20

01

11

21

02

12

22

p2 p2

p3

p3

Fig. 3. (a)3×3Wavefront Allocator (b) 4×3Wavefront Allocator (c) 4×4Wavefront Allocator.

Request vector port density utilizes the maximum on chip resources because it requires more logic as compared to other two designs while as the 9-bit vector based allocation algorithm utilizes the least on chip resources because of least hardware involved. Table 2 provides a comparison of maximum achievable clock rates post implementation for a 12-bit input request. Table 2 gives the timing comparison in the different request vector variants of the WFA. Finally static and dynamic power dissipation for different allocation algorithms is considered. The static power dissipation in an FPGA consists of device static power dissipation and the design static power dissipation. Design static power is very high in percentage of dynamic power dissipation. The dynamic power dissipation is function of input voltage V 2, the clock freq (fclk), the switching activity (α), load capacitance (CL) and number of elements used. The analysis was done on a constant supply voltage and at maximum operating frequency for each structure. To ensure a reasonable comparison, diverse input test vectors in post route simulation were selected to represent worst case switching activity. The physical constraint file and design node activity file captured during the post route simulation were used for power analysis in Xpower analyzer tool. Table3. Shows the power dissipated in WFA for various port densities considered. The power dissipated in the clocking resources varies with the clocking activity (clock frequency) as provided in the PCF. The power-delay product and the area delay product measured for the various cases are plotted in figure 4. Table 1. Resource Utilization Comparison for Different Input Port Density for WFA Algorithm. On Chip Resources Wavefront Allocator

No. Slice Registers

No. Slice LUTs

No. of Occupied Slices

3×3 WFA

45

60

80

4×3 WFA

27

36

48

4×4 WFA

17

24

28

[167]

Liyaqat and Mir/COMMUNE – 2015

Table 2. Timing Comparison for Different Input Port Density for WFA Algorithm

Wavefront Allocator 3×3 WFA 4×3 WFA 4×4 WFA

Timing Parameter Min available offset-in (ns 5.197 4.896 4.900

Max Frequency (MHz) 497.76 720.981 301.568

Min available offset-out (ns 8.782 8.924 8.892

Table 3. Power Dissipation for different input port densities of WFA algorithm.

FPGA Resource Clock logic Signals I/Os Dynamic Quiescent Total

180 160 140 120

3×3 WFA 4.34 0.15 0.19 7.79 12.47 3458.89 3471.36

Power Dissipation (mW) 4×3 WFA 4.48 0.18 0.25 8.29 13.20 3458.93 3472.13

4×4 WFAs 4.62 0.29 0.30 16.69 21.90 3459.38 3481.28

Number of slice registers

25

No of Slice LUT's

Dynamic Power Dissipation

No. of occupied slices

20

total resources

100

15

80

10

60 40

5

20

0

0 3 3 WFA

4 3 WFA

3×3 WFA

4 4 WFA

4×3 WFA

4×4 WFA

35 600

30

Area delay product

500

Power delay …

25 400

20 300

15 200

10 100

5

0

0

3×3 WFA

4×3 WFA

3×3 WFA

4×4 WFA

4×3 WFA

4×4 WFA

Fig. 4. Various Parameter plots for WFA (a)Resource Utilization (b) Power dessipation (c) Power delay product (d) Area delay product.

4. Conclusion and Future work This paper carried out the performance analysis of WFA having input request with various port densities. The implementation was targeted for Viertex 5 FPGA device. The resulting structures showed differences in the way of using resources available in the target FPGA device. The 4×4 based WFA uses resources extensively and has more power delay product and largest area delay product metric. On the other hand 3×3 based WFA uses lesser resources than any other algorithm considered and also consumes least power dissipation. To sum up, a judicious trade-off between area, power and throughput parameters, and the intended application will determine the correct approach for implementing the allocation algorithms of a On-chip router. Since the area delay product and the power delay product of the design targeted for FPGA platforms are considered as important parameters. Hence implementation of above allocation algorithm for on-chip communication architectures, for FPGA based platforms revealed that 4×4 based WFA algorithm as reasonably appropriate among the ones considered, for the high speed low power Network-on-chip (NoC) communication architectures. Currently we are implementing allocation algorithms using normal coding techniques. We intend to implement allocators using lookup table (LUT) based approach. This would further decrease the resources and increase the performance of the architecture. We intend to design NoC router having a varying input request vector as 8-bit, 12-bit, 16-bit, 32-bit up to 64-bits. The router latency, throughput and other performance parameters can be thus determined.

[168]

Liyaqat and Mir/COMMUNE – 2015

References Banerjee A., Mullins, R., and Moore, S., 2007. “A power and Energy Exploration of Network-on-chip Architectures,” First International Symposium on Network-on-chip. Princeton, NJ. Sudhir N. S., Pramod B., Patil 2014. “Low-latency, Low-Area Overhead and High throughput NoC Architecture for FPGA based computing system,” International Conference on Electronics Systems, Signal Processing and Computing Technologies. Nagpur, India. Osterloh, B., H. Michalik, B. Fiethe, K.Kotarowski, “SoCWire: A Network-on-Chip Approach for Reconfigurable System-on-Chip Designs in Space Applications,” in proc of NASA/ESA Conference on Adaptive Hardware and Systems, pp 51-56, june 2008. Maher, A., Mohhamed, R., Victor, G., “Evaluation of The Scalability of Round Robin Arbiters for NoC Routers on FPGA,”7 th International symposium on Embedded Multicore/Manycore System-on-chip, pp61-66,2013. Ahmaed, A. B., Abdallah, A. B., Kenichi, K., “Architecture and Design of Efficient 3D Network-on-Chip (3D NoC) for custom multicore SoC,” in International confrence on Broadband, Wireless Computing, communication and Application, FIT,Fukuoka, Japan, Nov 2010. Anderson, T., Owicki, J., Saxe, and Thacker, C., “High speed switch scheduling for local area networks,” ACM Trans. Comput. Syst., vol. 11, no. 4, pp. 319–352, Nov. 1993. Karol, M.,and Hluchyj, M., “Queueing in high-performance packetswitching,” IEEE J. Select. Areas Commun., vol. 6, pp. 1587–1597, Dec. 1988. McKeown, N., Anantharam, V,. and Walrand, J., “Achieving 100% throughput in an input-queued switch,” in Proc. IEEE INFOCOM ‘96, San Francisco, CA, pp. 296–302. Chang, Y.Y., Huang, Y.S.-C., Poremba, Narayanan, M., Yuan, V., Xie King, C., “Title TS-Router: On maximizing the Quality-ofAllocation in the On-Chip Network,” in IEEE 19th International Symposium on High Performance Computer Architecture (HPCA2013), pp 390-399, Feb 2013. Phanibhushana, B., Ganeshpure, K., Kundu, S., “Task model for on-chip communication infrastructure design for multicore systems,” in proc of IEEE 29th International Conference on Computer Design (ICCD), pp 360-365, oct 2011. J. Guo, Yao, J., Bhuyan, L., “An efficient packet scheduling algorithm in network processors,” in proceedings of 24th Annual Joint Conference of the IEEE Computer and Communications Societies ,pp 807- 818, march 2005. Dally, W. J, Towels, B., Principles and Practices of Interconnection Networks, Ist ed. Morgan Kaufmann publications, 2003. Mckeown, N., “Scheduling algorithms for input buffered cell switches,” Ph.D thesis, University of Calfornia at Berkely,1995. Lee, K., Lee, S., Yoo, h., “A Distributed Crossbar Switch Scheduler for On-Chip Networks,” in Custom Integrated Circuits Conference, 2003. Proceedings of the IEEE ,pp 671-674, Sept. 2003. Gupta, P., McKeown, N., “ Designing and implementing a Fast Crossbar Scheduler,” in proc of Micro, IEEE, vol 19, no 1, pp 20-28, Feb 1999 McKeown, N., “The iSLIP Scheduling Algorithm for Input-Queued Switches” IEEE/ACM transactions on Networking, vol 7, no,2, april 1999. Ouyang, J., Xie, Y,. “LOFT: A High Performance Network-on-Chip Providing Quality-of-Service Support, ” in 43rd Annual IEEE/ACM International Symposium on Microarchitecture (MICRO),pp 409-420, Dec 2010. Jose, G., Delgado-Frias, Girish, B. R., “A VLSI Crossbar switch wih Wrapped Wave Front Arbitrarion,” IEEE transaction on circuits and systems-I: Fundamental theory and applications,Vol 50, No 1,jan 2003. http://www.xilinx.com.

[169]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

The Social Web: Expressive use among the Undergraduate Students of University of Kashmir Zahid Ashraf Wani, Tazeem Zainab* Department of Library and Information Science, University of Kashmir, Srinagar, India

Abstract The web 2.0 has greatly transformed the approach of using the web. Now days, teenagers take refuge in different tools and services of the Web 2.0 and are of view that virtual networking and allied tools give them space to give vent to their ideas, frustrations and creativity. In this milieu, present study has made an endeavour to gauge use and impact of different Web 2.0 tools on personality development of the students in terms of socialisation and expression. The study explores the impression of different categories of Web 2.0 tools on the day-to-day activities of the students.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Web 2.0; Expressive Web; SNS; Privacy Concern; Blogs and Youth

1. Introduction Inherently web is a global medium. From a group work instrument for CERN scientists it has turned into a universal info cosmos with billions of users, and off late evolved as a social podium. The nonstop varying developments have steered the Web to arrive in a new version Web 2.0. The Web 2.0 has gained considerable force in the last epoch. The influence of Web 2.0 principles and technologies has driven a burst of information and media content on the Web, and individual and corporate adoption of the technologies continues to rise. In the words of Anderson, Web 2.0 is a collective term for a set of web tools and services that increase users’ communication abilities. In addition, he states that Web 2.0 technologies are not actually technologies as such, but services constructed by using the building blocks of different technologies and open standards that support the Internet. These tools or services include blogs, wikis, and multimedia sharing services, RSS feeds, podcasting and content tagging services (Anderson, 2007). Kataria and Anbu reveal that maximum people call web 2.0 as an experience. Some people describe it as “warm” web or “hot” web, “warm” as it offers a sort of friendliness to the users—a form of ‘interactivity’ to the users. It is “warm” for it offers platform discourse. They further say that sharing dominates all aspect of Web 2.0 (Kataria, Anbu, 2009). McManus describes Web 2.0 as “The Web as Platform” relying on individual perception; i.e. for marketers, the Web is a platform for communications, for journalists the it is a platform for alternate media, for corporate people the it is a platform for business, for geeks the Web is a platform for software development, etc. (McManus, 2001). Singel reports Mayfield, CEO of software WIKI Solutions Company: ‘‘Web 1.0 was commerce. WEB 2.0 is people’’ (Singel, 2005). Safran et al. highlights that Web 2.0 services make it feasible to sustain critical and analytical thinking enable intuitive and associational thinking and back analogical thinking through easy access to comprehensive information and opinions (Safron et al, 2007). In addition, these tools and services might be effective in guiding case studies due to their cooperative nature based on experiential learning approach (Huang, Behara, 2007). Various studies show growing and swift adoption of web 2.0 tools and services, particularly SNS viz. Facebook & Twitter. The research studies reveal that 80-90% of college students possess an account on these SNSs (Ribiere, 2010). The Web 2.0 is having a dramatic effect on how people work, communicate, and collaborate. It opens an entirely different world of social interconnectivity in which educators, researchers, professionals, and students alike can easily communicate and share with each other for life-like collaborative knowledge construction. Using Web 2.0 tools users can nowadays effortlessly learn from and *

Corresponding Author. Tel.: +91 9086 668935.

E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Wani and Zainab / COMMUNE-2015

with industry, business, professional organisations, communities, and within an ever-expanding diversity of social, cultural, political networks (Wenger, 2004). Klamma et al. recommend that Web 2.0 services could enable and improve lifelong learning experience by linking students in cooperative settings with fading frontiers around the globe (Klamma et al, 2007). People are involved in a varied array of technologies-based informal education at home and in the groups by unceasingly collaborating with others in virtual networked settings (Selwyn, 2007). 2. Literature Review There has been a major impact of technology on the societies. Each change has affected the relation between people. Technology has a vast impact on the growth of nations, on the livelihood of people, and who owns what. Technologies are the reason of people being rich and more social. A decent amount of research has revealed different Web 2.0 tools & services have great influence on people. According to Goddard technology offers an interactive, visually persuasive, motivating realistic environment to develop the knowledge and skills needed in multicultural atmosphere (Goddard, 2002). Anderson in his study says that Web 2.0 is a collective term for a set of web-based tools and services that increase users’ communication abilities (Anderson, 2007). Abram comments that Web 2.0 is about novel human facets of interactivity. It is pertaining to dialogues, social networking, & focus on self at the same time (Abram, 2013). Various studies show growing and swift adoption of web 2.0 tools and services, particularly SNSs viz. Facebook & Twitter. The research studies reveal that 80-90% of college students do possess an account on these SNSs (Ribiere, 2010). Scheneckberg is of view that the speed of acceptance of Web 2.0 services is quite high as these are easy to use and intuitive, and allow the direct and instant online publication and dissemination of user content. (Schneckenberg, 2009). Klamma et al. again recommend that Web 2.0 services could enable and improve lifelong learning experience by linking students in cooperative settings with fading frontiers around the globe (Klamma et al, 2007). Ribiere et al. believe that social grooming; self-representation; Privacy concern of users; user’s experience including pleasure, curiosity& fun, Identification and self-expression, Surprise and serendipity on social networks are some of the major fields of interest among the researchers (Ribiere, 2010). Web 2.0 services empower users to link to and cooperate with others, with varied interactions (Selwyn, 2007). Chapman and Lahav conducted research on SNSs from diverse nations and ethos. The findings reveal the dissimilarities in terms of the users’ objectives, distinctive patterns of selfexpression, and collective interaction approaches (Chapman, Lahav, 2008). Hampton in his study revealed that people are progressively using the expressive web using modes that supplement or advance their offline sociality (Hampton, 2007). The expressive web has been growing fast, a course often pronounced in the general press as the escalation of social computing. Studies indicate that these technologies have been embraced as a tools of social networking and social collaboration for growing numbers of people and communities (Haythornthwaite, 2005). The Web 2.0 for many reasons which is different for every user group motivated and enabled to put their profiles likes and dislikes, ideas and thoughts before the world wide community. This trend especially among young generation is very popular. 3. Problem The expressive web use represents the exercise and activities of technologically facilitated sociality, employing the Web to do & accomplish social relations, self-staging, open demonstration, collective vigil, & the evolving, sustenance and advancing of societal bonds. The Web 2.0 has become the prime service of internet to achieve expressive web use especially among generation X. Now a days, teenagers take refuge in different tools and services of the Web 2.0 and are of view that virtual networking and allied tools give them space to give vent to their ideas, frustrations and creativity. In this milieu present study has made an endeavour to gauge use and impact of different Web 2.0 tools on personality development of the students in terms of socialisation and expression. 4. Objectives Main objectives of the study are a) To gauge if Web 2.0 tools bring sense of excitement, fun and enjoyment among the students. b) To measure the effect of prominent web 2.0 tools on social grooming of students. c) To analyze how prominent web 2.0 tools influence the personal efficiency and privacy concern of students. 5. Scope The scope of the study is confined towards analysing the impact of expressive Web 2.0 tools viz. SNS and blogs on the undergraduate students of various UG departments of University of Kashmir viz., Department of Law, Department of Management, Department of Pharmacy and Department of Electronics and Technology.

[171]

Wani and Zainab / COMMUNE-2015

6. Methodology The methodology for present study comprised of following stages: a) STAGE-I: A review of relevant literature was conducted to evolve and refine the methodology and tools to satisfy the objectives. Accordingly, methodology for present study was refined to make findings more scientific and logical. b) STAGE-II: Experience and knowledge gained during Stage-I helped study to evolve a comprehensive questionnaire. The information was collected on the use and satisfaction with different web 2.0 tools and services among the students for which 5 point Likert scale was employed. c) STAGE-III: Thereby, a questionnaire was framed administrated among undergraduate students of Law (LLB), Pharmacy (B. Pharma), Electronics and Technology (B. Tech) and Management (IMBA) departments of University of Kashmir. In order to make the study manageable and systematic, proportionate manageable stratified random sampling was applied. A total of 100 questionnaires were administrated in four departments of University of Kashmir viz., Department of Law, Department of Management, Department of Pharmacy and Department of Electronics and Technology. Out of 100 questionnaires administrated 92 were responded whereas 8 students did not respond. Among 100 questionnaires, 38 questionnaires were distributed in Department of Law in which 35 responded, out of which 26 were males and 9 were females. In the Department of Management 32 questionnaires were distributed with cent per cent response rate among in which 27 were females and 5 males. In the department of Electronics & Technology and Department of Pharmacy, a total of 16 and 13 questionnaires were distributed and 13 and 12 questionnaires were received back respectively. Amongst the respondents, the Department of Electronics and Technology had a total of 9 males and 4 females while as in Department of Pharmacy there were 4 male and 8 female respondents. Table 1 offers a lucid picture. Further, a clear representation of the male and female respondents, for gauging the total impact of Web 2.0 on both the genders has been made in the study. It doesn’t take into account the independent results of both genders in order to avoid the discriminative factors that may influence the results. Table 1: Number of Students Randomly Selected for the Study Gender Males Females Total

B.A.LLB 26 (74.3) 9 (25.7) 35

I-MBA 5 (15.6) 27 (84.4) 32

B.Tech. 9 (69.3) 4 (30.7) 13

B. Pharma 4 (33.3) 8 (66.7) 12

Total 44 (47.8) 48 (52.1) 92

Figures in parenthesis indicate percentage

7. Analysis & discussion 7.1

Excitement, Fun and Enjoyment

When the students were asked which Web 2.0 tools they use for excitement, fun and enjoyment 30.4% agree that SNS are the main tools for pleasure and fun, while small number (16.3%) strongly agree. Besides, 10.8% apiece disagree or strongly disagree with this notion. As for as blogs are concerned majority agree (27.1%) that use of blogs bring excitement, fun and enjoyment in their life, while as 13% students strongly agree with the statement. Second largest number of students (23.9%) have taken a neutral position. Almost one third of students disagree (10.8%) to strongly disagree (19.5%) that blogs are source of excitement and fun. Table 2: Excitement, Fun, and Enjoyment

S N S

B L O G S

Strongly Disagree Disagree Neutral Agree Strongly Agree Not Responded Strongly Disagree Disagree Neutral Agree Strongly Agree Not Responded

LLB 6(17.1) 7(20) 8(22.8) 9(25.4) 5(14.4) 0(0) 2(5.1) 5(14.4) 9(25.7) 11(31.4) 4(11.4) 4(11.4)

IMBA 2(6.2) 3(9.3) 6(18.7) 11(34.3) 4(12.5) 6(18.7) 2(6.2) 3(9.3) 8(25) 9(28.1) 3(9.3) 0(0)

Figures in parenthesis indicate percentage

[172]

B. Tech 1(7.6) 0(0) 1(7.6) 3(23) 4(30.7) 4(30.7) 1(7.6) 2(15.3) 2(15.3) 2(15.3) 2(15.3) 4(30.7)

B.Pharma 1(8.3) 0(0) 3(25) 5(41.6) 2(16.6) 1(8.3) 1(8.3) 0(0) 3(25) 3(25) 3(25) 2(16.6)

Total (% age) 10(10.8) 10(10.8) 18(1.9) 28(30.4) 15(16.3) 11(11.9) 6(19.5) 10(10.8) 22(23.9) 25(27.1) 12(13) 10(10.8)

Wani and Zainab / COMMUNE-2015

7.2

Social Grooming

Social grooming talent is a competitive benefit that further develops and grooms individual’s personality, repute and accumulation of societal assets. The actions people involve in on SNSs mimic the behaviour of social grooming. It was found that majority of students are of the view [agree (28.2%), strongly agree (11.9%)] that SNS help in social grooming. However, reasonable number remains neutral (22.8%) on the issue. Whereas small number of students disagree (9.7%) to strongly disagree (10.8%)] with the perception that SNS helps in social grooming. Majority of students agree (28.2%) to strongly agree (19.5%) that blogs help in social grooming. Whereas 20.6% have remained neutral, besides 29.3% students did not respond to this question. Table 3: Social Grooming Aspects

S N S

B L O G S

Strongly Disagree Disagree Neutral Agree Strongly Agree Not Responded Strongly Disagree Disagree Neutral Agree Strongly Agree Not Responded

LLB 2(5.1) 8(22.8) 13(37.1) 10(28.5) 2(5.1) 0(0) 1(2.8) 5(14.4) 9(25.7) 13(37.1) 1(2.8) 6(17.1)

IMBA 6(18.7) 1(4.4) 5(15.6) 9(28.1) 6(18.7) 5(15.6) 1(4.4) 6(18.7) 8(25) 6(18.7) 2(6.2) 9(28.1)

B.Tech 0(0) 0(0) 2(15.3) 2(15.3) 0(0) 9(69.2) 0(0) 0(0) 1(7.6) 4(30.7) 0(0) 8(61.5)

B.Pharma 2(16.6) 0(0) 1(8.3) 4(33.3) 3(25) 2(16.6) 0(0) 1(8.3) 1(8.3) 3(25) 3(25) 4(33.3)

Total (% age) 10(10.8) 09(9.7) 21(22.8) 26(28.2) 11(11.9) 16(17.3) 02(4.3) 12(13) 19(20.6) 26(28.2) 06(19.5) 27(29.3)

Figures in parenthesis indicate percentage

7.3

Personal Efficiency

Among the main objectives of employing information technology is the capability to surge the efficacy of the people who use it. It is believed that this facet is among the main reasons for the acceptance and utilization of web 2.0 services for expressive & influential purpose. A progressive impact on individual efficacy will give impetus to the embracing speed of web 2.0 services. When the students were asked which Web 2.0 tool helps them to increase their personal efficiency it was found equal number of students (1.e. 19.5%) agree or remain neutral, while 13% disagree and 14.3% strongly disagree with this belief. Besides, 25% of students did not respond to the question. In case of blogs 19.5% remained neutral while 22.8 % agree & 13% strongly agree that blogs help in personal efficiency of the students. Table 4: Personal Efficiency

S N S

B L O G S

Strongly Disagree Disagree Neutral Agree Strongly Agree Not Responded Strongly Disagree Disagree Neutral Agree Strongly Agree Not Responded

LLB 6(17.1) 4(11.4) 13(37.1) 8(22.8) 1(2.8) 2(5.1) 3(8.5) 4(11.4) 12(34.4) 9(25.7) 2(5.1) 5(14.4)

IMBA 6(18.7) 6(18.7) 3(9.3) 7(21.8) 3(9.3) 7(21.8) 2(6.2) 3(9.3) 4(12.5) 5(15.6) 8(25) 8(25)

B.Tech 0(0) 1(7.6) 1(7.6) 2(15.3) 0(0) 9(69.2) 0(0) 1(7.6) 0(0) 3(23) 0(0) 9(69.2)

B.Pharma 1(8.3) 1(8.3) 1(8.3) 1(8.3) 3(25) 5(41.6) 0(0) 1(8.3) 2(16.6) 4(33.3) 2(16.6) 3(25)

Total (%age) 13(14.3) 12(13) 18(19.5) 18(19.5) 07(7.6) 23(25) 05(5.4) 09(9.7) 18(19.5) 21(22.8) 12(13) 25(30.4)

Figures in parenthesis indicate percentage

7.4

Privacy Concern

The moment question of dissemination private information pertaining to one’s personal life, it is generally assumed the online privacy concern is a hurdle to the embracing of web 2.0 tools and services. Nevertheless, evolving trust and increasing trend of sharing personal information cannot be undermined unless people have developed certain degree of reliability on online security. We have no reason not to concede that in near future the web 2.0 services and tools will become an essential part of daily life with robust online privacy protection. Since, web 2.0 is a fairly novel idea in IT usage, it is expected that online privacy will remain as hurdle for web 2.0 embracing for some time to come. When it was asked to indicate privacy concern of users with select web 2.0 tool like SNS, it was revealed an reasonable number of users agree (20.6%) to strongly agree (7.6%) that users do have their doubts and apprehensions about SNS privacy, while 18.4% users remained neutral on the issue. Furthermore, around one third of users disagree (16.3%) to strongly disagree (14.3%) with this notion, besides 22.8% of users did not respond to the question. As for as blogs are concerned one third of users disagree (20.6%) to strongly disagree (9.7%) with the perception that blogs are susceptible to online privacy infringements, while small number of users agree (13%) to strongly agree (5.4%) with the ideas that blogs are

[173]

Wani and Zainab / COMMUNE-2015

prone to privacy infringements. Table 5: Privacy Concern

S N S

B L O G S

Strongly Disagree Disagree Neutral Agree Strongly Agree Not Responded Strongly Disagree Disagree Neutral Agree Strongly Agree Not Responded

LLB 6(17.1) 6(17.1) 11(31.4) 7(7.3) 3(8.5) 2(5.1) 6(17.1) 6(17.1) 11(31.4) 6(17.1) 2(5.1) 4(11.4)

IMBA 6(18.7) 7(21.8) 1(4.4) 7(21.8) 3(9.3) 8(25) 2(6.2) 10(31.5) 5(15.6) 4(12.5) 2(6.2) 9(28.1)

B.Tech 0(0) 1(7.6) 1(7.6) 2(15.3) 0(0) 9(69.2) 0(0) 2(15.3) 1(7.6) 0(0) 0(0) 10(76.9)

B.Pharma 1(8.3) 1(8.3) 4(33.3) 3(25) 1(8.3) 2(16.6) 1(8.3) 1(8.3) 3(25) 2(16.6) 1(8.3) 4(33.3)

Total (%age) 13(14.3) 15(16.3) 17(18.4) 19(20.6) 07(7.6) 21(22.8) 09(9.7) 19(20.6) 20(21.7) 12(13) 05(5.4) 27(29.3)

Figures in parenthesis indicate percentage

8. Conclusion There is a favourable impact of Web 2.0 individual factors like personal efficiency, social grooming, excitement and enjoyment. The study found increased degree of acceptance of web 2.0 tools and technologies among students. These factors do have a robust effect on the expressive use of the web 2.0 applications. Web 2.0 tools and technologies are not only used for fun, but are considered useful in terms personal efficiency and social grooming. On the other hand, privacy concern pertaining to Web 2.0 tools especially Social Networking Sites (SNS) is seen negatively affecting the expressive use of these applications. The study reveals that students are reasonably concerned about their online privacy. References Abram, S., 2013. Web 2.0, Library 2.0, and Librarian 2.0: Preparing for the 2.0 World. OneSource,2(1).Retrieved from url: http://www.imakenews.com/sirsi/e_article000505688.cfm?x=b6yrqlj,b2rpqhrm. Anderson, 2007. What is Web 2.0? Ideas, Technologies and Implications for Education. Retrieved from url: http://www.jisc.ac.uk/media/documents/techwatch/tsW070/co/oen.odt. Chapman, C. N., Lahav, M., 2008. International Ethnographic Observation of Social Networking Sites, Italy, Florence. Goddard, M., 2002. What do we do with these computers? Reflections on technology in the classroom. Journal of Research on Technology in Education, 35 (1), 19-26. Retrieved from url: http://connection.ebscohost.com/c/articles/8559540/what-do-we-do-these-computers- reflectionstechnology-classroom. Hampton, K. N., 2007. Neighborhoods in the network society: the e-Neighbors study. Information, Communication & Society, 10(5), 714 – 748. DOI: 10.1080/13691180701658061. Haythornthwaite, C., 2005. Social networks and Internet connectivity effects. Information, Communication & Society, 8(2), 125–147. DOI: 10.1080/13691180500146185. Huang, C.D. , Behara, R.S., 2007. Outcome-driven experiential learning with Web 2.0. Journal of Information Systems Education, 18(3), 329-36. Retrieved from url: msmcactech.wikispaces.com/file/view/Outcome_Driven.pdf. Kataria, S., Anbu, K. J.P., 2009. Applications of Web 2.0 in the Enhancement of Services and Resource in Academic Libraries: An Experiment @ JIIT University Noida, India. Retrieved from url: http://crl.du.ac.in/ical09/papers/index_files/ical-98_130_287_1_RV.pdf . Klamma, R., Chatti, M.A., Duval, E., Hummel, H., Hvannberg, E.H., Kravcik, M., Law, E., Naeve, A. & Scott,P. (2007). Social software for life-ling learning. Journal of Educational Technology and Society, 10(3), 72-83. DOI:/10.1.1.108.5134. McManus, D.A., 2001. The Two Paradigms of Education and the Peer Review of Teaching. Journal of Geoscience Education, 49(5), 423-34. Retrieved from url: http://serc.carleton.edu/resources/13154.html. Ribière, V.M., Haddad, M., Wiele, P.V., 2010. The impact of national culture traits on the usage of Web 2.0 Technologies. VINE, 40(3), 334 – 361. DOI: 10.1108/03055721011071458. Safran, C., Helic, C., Gutl, C., 2007. E-Learning practices and Web 2.0. Proceedings of ICL 2007, 1-8. Villach, Austria. DOI=10.1.1.161.8553. Schneckenberg, D., 2009. Web 2.0 and the empowerment of the knowledge worker. Journal of Knowledge Management, 13(6), 509-520. Retrieved from url: http://lpis.csd.auth.gr/mtpx/km/material/JKM-13-6a.pdf. Selwyn, N., 2007. Web 2.0 applications as alternative environments for informal learning – a critical review. Retrieved from url:https://www1.oecd.org/edu/ceri/39458556.pdf. Singel, R., 2005. Are You Ready for Web 2.0.? Retrieved from url: http://www.wired.com/science/discoveries/news/2005/10/69114. Wenger, E., 2004. Learning for a small planet: a research agenda. Retrieved from url: http://www.ewenger.com/research.

[174]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Free Text Plagiarism Detection using Lexical Database and Document Fingerprinting Muzamil Ahmad*, Shameem Yousf, Sheikh Nasrullah Department of Information Technology, Central University of Kashmir, Srinagar, India

Abstract Plagiarism can be referred as the multi-dimensional problem which has influenced different domains of our society like academics, research, art, multimedia etc. Plagiarism detection is the fight back of unveiling the true nature of the content and decides whether it is plagiarized or not. In this paper, we present the system of plagiarism detection by integrating different aspects of detection process like use of lexical databases, document fingerprinting. The use of lexical database in our system helps in dealing with the problem of Synonymy i.e., the cases of plagiarism where words are replaced with Synonyms. We also introduced the concept of Variable Window Document Fingerprinting (VWDF) so as to yield better comparison rate. Finally we simulate our system on predefined tasks and corpus in order to yield the performance of the system and present the statistical information in this paper.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Plagiarism Detection; Document Fingerprinting; Parsing; Cloning; Variable Window

1. Introduction Plagiarism can be defined as the immoral and illegal authorship of the other persons work as one’s own, either whole or a part of it. In the past the more closely you imitate your master, was reflected as the step towards perfection, but the time has changed now & the imitation is referred to as the act known as Plagiarism, in other words which can be regarded as the intellectual theft of anything. This theft can exist in any form like text, ideas, art, manufacturing, multimedia etc. Due to the vast domain of the problem definition it means that the solution for this problem is out of scope for now, but researchers from different platforms are working in order to develop a system which can help the society and academicians in order to curb the menace. Due to the advancement in computer science, much of the literature has been digitized in order to avail it to the masses for the better utilization, but this availability of information in the form of web has a dark spot within it i.e. anybody present in the cyberspace can access it and then mould it and prepare another plagiarized copy of this original document. Plagiarism detection is the process of identifying the plagiarized text. The task of detection can be accomplished in two ways i.e. Manual detection and Computer aided detection. Manual Detection: The process of detecting plagiarism cases manually by checking the suspicious documents against the source documents by the subject matter experts is the reliable way of detection but needs lot of time and effort in order to produce the desirable outputs. Computer Aided Detection: The process of identifying plagiarism cases with the help of computer and a set of algorithms that run over the hardware in order to accomplish the task is quite good and can perform effectively. Although there doesn’t exist a complete plagiarism detector which can handle all the cases of plagiarism. Document Fingerprinting: Document Fingerprinting is the method of mapping an arbitrary large amount of data into the collection of integers. Typically, a fingerprint is generated by selecting substrings from the text and applying

* Corresponding author. Tel.: +91 9622 757507. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Ahmad et al/COMMUNE – 2015

a mathematical function to each selected substring. This process is similar to hashing function. The process of document finger-printing is shown in Fig.1.

Fig.1 Document Fingerprinting Process

1.1.

Related Work

The plagiarism detection is quite complex. Various different technologies and concepts have been used for detection purposes. The forms of the plagiarism and the challenges in automatic plagiarism detection have been discussed by (Clough et al., 2003). Plagiarism detection can be done in two ways i.e. intrinsic as well as extrinsic. The intrinsic plagiarism detection systems have been developed by (Eissen et al., 2006, Su´arez et al., 2010 and Stamatatos, 2011). The methodology of document fingerprinting has been used by (Heintze, 1996). Winnowing was introduced by (Schleimer et al., 2003) as a local method for document fingerprinting and help in detecting plagiarism. (Eissen et al., 2006) have analysed the Stylometric variations using syntactic and Parts Of Speech (POS) features and structure of the document for plagiarism detection. Extrinsic plagiarism detection methodologies suggest of checking a particular document against a given set of documents known as corpus. (Bravo-Marquez et al., 2010) suggests of chopping the document into segments and passing it to a search engine as a query in order to check originality. The use of n-gram approach was explored by (Lyon et al., 2001). It gives flexibility in detecting those segments that have been reworded. The other variations of this include the tri-gram approach by (Barr´on-Cede˜no et al., 2010), which was found effective. 2. Methodology The detection process is a complicated task and depends on lot of variables like the obfuscation strategy employed to create plagiarised text, document retrieval, semantics etc. In this paper, we integrate different approaches and aspects of detection in order to yield a better detection domain. In this section we discuss the methodology used to detect the plagiarism. The approach of detection as shown in Fig.2 is divided into three phases: 



Pre-processing: This phase forms the ground of the detection process. In this phase, relevant documents are retrieved from the corpus, which are then transferred to interim buffer for parsing and sanitization with the help of Natural Language Processing (NLP) tools. Intermediate-Processing: After retrieving the sanitized document its clones are generated so as to enhance the probability of detection. Finally, fingerprints are generated using Variable Window Document Fingerprinting (VWDF).

Fig.2 Proposed Model



2.1.

Post-Processing: Finally fingerprints are passed to the comparison function, which decides whether the document is actually plagiarised or not, and the result files are generated. Parsing

In this stage we parse the retrieved documents in order to extract the parts of speech information. The Stanford Part of Speech Tagger is used for parsing, and part of speech is assigned to each word. Table 1 contains POS of some words.

[176]

Ahmad et al/COMMUNE – 2015

Table 1 Parts of Speech Information

2.2.

Word

POS

Move

NN

Quickly

RB

In

IN

This

DT

world

NN

Sanitization

Based on the tagged data passed by the parser, we begin the stage of cleaning the data. The basic steps involved in cleaning the data are:  Stop Word Removal: The frequently occurring words like the, of etc. are removed.  Noun/Pronoun Replacement: Since using the Nouns and Pronouns cannot be counted as plagiarism, we replace noun with NN and pronoun with PN to simplify the processing.  Verb/ Adverb/ Adjective Retrieving: The synonyms of verb, adverb and adjective of documents are retrieved using WordNet† dictionary. 2.3. Cloning In this phase, the clones of retrieved document are generated by replacing the verbs, adverbs and adjectives in it with their respective synonyms. Table 2 shows an example of cloning. Table 2 Example of Cloning

2.4.

Original Sentence

Clone 1

Clone 2

Clone 3

he is a beautiful boy

he is a handsome boy

he is a lovely boy

he is a gorgeous boy

Fingerprint Generation

After retrieving the cleaned data, the fingerprints are generated using Variable Window Document Fingerprinting (VWDF). In normal document fingerprinting window size for calculating fingerprint is kept fixed but in VWDF while creating the fingerprint of the document window size is determined by the length of the word. The example for fingerprint generation using VWDF is as given below: VWDF Example:   

2.5.

Input a sanitized segment of document e.g. ”Insanity : doing the same thing over and over again and expecting different results.”‡ Generate Variable sized windows from the segment e.g. “insanity, doing, the, same, thing, over, and, over, again, and, expecting, different, results.” Generates fingerprints local to the windows e.g. ”(109,8), (105,5), (107,3), (105,4), (107,5), (111,4), (102,3), (111,4), (102,5), (102,3), (107,9), (105,9)”

Comparison

The fingerprint of the suspicious document is now compared with the retrieved Corpus documents and it is determined in this function whether the document is actually a case of plagiarism or not. The comparison using the document fingerprinting approach is shown in Fig.3.

† http://wordnet.princeton.edu/ ‡

Quote by Albert Einstein

[177]

Ahmad et al/COMMUNE – 2015

Fig.3 Comparison in Fingerprints of Documents

3. Experiments In this section, we describe the details of the resources used and the experimental setup. We performed our tests on the data compiled by (Clough et al., 2011). These documents were written by native and non-native authors. The corpus contains the four types of obfuscated documents:     3.1.

Near Copy Light Revision Heavy revision Non-Plagiarism

Settings

Each document was processed against all the retrieved/ related documents that were in the (.txt) format. All processing, including the pre-processing, intermediate-processing and post-processing were carried out under computer control i.e. there was no human interaction with the PDS to carry out the detection process. 3.2.

Threshold Determination

Based on the testing data we were able to fetch the threshold values for the different kind of obfuscations as per mentioned in the corpus. 4. Results This section gives the results of the detection process for a particular document against the retrieved set of documents in different obfuscation categories. Table 3 Results, Accuracy, and Threshold Native

Non-Native

Obfuscation Strategies

#Docs

Threshold

#Errors

Accuracy

#Docs

Threshold

#Errors

Accuracy

NON

24

25

8

66.67%

14

25

3

78.57%

HEAVY

12

30

3

75%

7

30

1

85.71%

LIGHT

12

40

3

75%

7

40

1

85.71%

CUT

12

50

2

83.33%

7

50

2

71.42%

The Table 3 shows the statistical results of the test carried out on the documents. The Fig.4 shows the average results of all the tasks in terms of detection percentage that were generated by our system.

[178]

Ahmad et al/COMMUNE – 2015

100

Native Non Plag Non Native Non Plag

80

Native Heavy

60

Non Native Heavy Native Light

40

Non Native Light Native Cut

20

Non Native Cut

0 Fig.4 Results of Average Case of All Tasks

Precision: is the fraction of retrieved documents that are relevant to the search. Recall: is the fraction of the documents that are relevant to the query that are successfully retrieved. The Table 3 shows the values of precision and recall against the given tasks. In this case recall always comes as 1.0 because of the reason that the documents retrieved in our case, are actually the relevant documents which we generate for our experimentation. Table 3 Precision and Recall Values Task Number

Precision

Recall

1.

0.5

1.0

2.

0.61904764

1.0

3.

0.7

1.0

4.

0.65

1.0

5. Conclusion and Future Work The task of plagiarism detection is quite a vast and difficult field; the area needs people or researchers from different backgrounds in order to devise a heuristic model. It requires intermingling different technologies to attain the best possible results. The best possible solution to the problem can be if we could any how filter the data over the web so that we could create different ordered sets that are relatively linked together in some fashion. This research document presents the basic ideology on implementation of how we can integrate lexical databases in generating different version of the same document and which can be later used to detect the plagiarism. We can also focus on the fuzzy hash functions in order to yield the better results. Since we know the problem set is very vast and lot of work needs to be done, in this field like we could incorporate efficient search, indexing and retrieval techniques, in addition to that databases could be integrated to store the fingerprints of the documents for the faster execution. References Clough, P., and D. O. I. Studies, 2003. Old and new challenges in automatic plagiarism detection, National Plagiarism Advisory Service, http://ir.shef.ac.uk/cloughie/index.html, p. 391–407. Eissen, S., M., Zu., Stein, B., 2006. Kulig, M., Plagiarism detection without reference collections, GfKl, ser. Studies in Classification, Data Analysis, and Knowledge Organization, R. Decker and H.-J. Lenz, Eds. Springer, p. 359–366. Su´arez, P., Crist´obal, J., C., G., Villena-Rom´an, J, 2010. A plagiarism detector for intrinsic plagiarism - lab report for pan at clef 2010, CLEF (Notebook Papers/LABs/Workshops), M. Braschler, D. Harman, and E. Pianta, Eds. Stamatatos, E., 2011. Plagiarism detection using stopword n-grams, JASIST, vol. 62, no. 12, p. 2512–2527. Heintze, N., 1996. Scalable document fingerprinting, IN PROC. USENIX WORKSHOP ON ELECTRONIC COMMERCE, Schleimer, S., Wilkerson, D., S., Aiken, A., 2003. Winnowing: Local algorithms for document fingerprinting, SIGMOD Conference, p. 76–85. Bravo-Marquez, F., L’Huillier, G., R´ıos, S., A., Vel´asquez, J., D., Guerrero, L., A., 2010. Docode-lite: A meta-search engine for document similarity retrieval, KES (2), ser. Lecture Notes in Computer Science, R. Setchi, I. Jordanov, R. J. Howlett, and L. C. Jain, Eds., vol. 6277. Springer, p. 93–102. Lyon, C., Malcolm, J., Dickerson, B., 2001. Detecting short passages of similar text in large document collections, Proceedings of the Conference on Empirical Methods in Natural Language Processing, p. 118–125. Barr´on-Cede˜no, A., Basile, C., Esposti, M., D., Rosso, P., 2010. Word length n-grams for text re-use detection, CICLing, ser. Lecture Notes in Computer Science, A. F. Gelbukh, Ed., vol. 6008. Springer, p. 687–699. Clough, P., and Stevenson, M.,2011. Developing a corpus of plagiarised short answers, Language Resources and Evaluation, vol. 45, no. 1, pp. 5–24.

[179]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

On Linear Classifiers vs. Hybrid Configuration: An Empirical Study Shifa Basharat*, Manzoor. A. Chachoo Department of Computer Science,University of Kashmir, Srinagar, India

Abstract With the proliferation of internet and social media, sentiment analysis has become an important research area. Sentiment analysis also referred to as opinion mining, is a text categorization task that determines the contextual polarity of the text e.g. whether a particular product review is positive/negative or whether a customer is satisfied/dissatisfied and so on. The text classification task can be accomplished via machine learning techniques or lexicon based techniques. This paper focuses on two basic machine learning algorithms namely Naive Bayes and logistic regression for text classification and the results show that both the algorithms used in combination (i.e. a hybrid configuration) can improve the classification effectiveness in terms of various evaluation measures such as accuracy, precision, recall, fallout and f-measure.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Gaussian Naive Bayes; Logistic Regression; Machime Learning; Machine Learning Algorithms; Text Categorization

1. Introduction With the proliferation of internet and the increased availability of documents in digital form, there is an increasing need to organize them and with this increasing need, the automatic text classification has witnessed a booming interest in the past decade. Text categorization (a.k.a. text classification) is the task of assigning predefined categories to freetext documents. It can provide conceptual views of document collections and has important applications in the real world. For example, news stories are typically organized by subject categories (topics) or geographical codes; academic papers are often classified by technical domains and sub-domains; patient reports in health-care organizations are often indexed from multiple aspects, using taxonomies of disease categories, types of surgical procedures, insurance reimbursement codes and so on. Another widespread application of text categorization is spam filtering, where email messages are classified into the two categories of spam and non-spam, respectively (Yang, Joachims, 2008). The most popular approach to TC, is the ML paradigm, according to which a general inductive process automatically builds an automatic text classifier by learning, from a set of pre-classified documents, the characteristics of the categories of interest (Sebastiani, 2002). In ML based techniques, two sets of documents are needed: training set and test set. A training set is used by an automatic classifier to learn the differentiating characteristics of documents and a test set is used to check the efficiency of a particular classifier (Prabowo, Thelwall, 2009; Vohra, Teraiya, 2012). A number MLAs have been adopted for text classification. Broadly, the machine learning algorithms can be classified into supervised MLAs and unsupervised MLAs. This paper focuses on the two basic supervised MLAs namely naive Bayes and logistic regression and the objective is to compare the efficiency of these algorithms. This paper is organized as follows: Section 2 describes the naive Bayes and logistic regression algorithm. Section 3 provides an overview of various performance measures used for comparison. Section 4 describes the experimental procedure including the details of data used, implementation of algorithms, and the results. Section 5 presents the conclusion.

* Corresponding author. Tel.: +91 9419 049791. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Basharat and Chachoo/ COMMUNE-2015

2. Different Classification Approaches Used 2.1.

Naive Bayes classifier

Naive Bayes, a generative classifier learns a model of joint probability, p(x,y) of the inputs x and the label y, and makes their predictions by using Bayes rule to calculate p(y|x), and then picking the most likely label y (Ng, Jordan, 2002).This means that given a feature vector table, the algorithm computes the posterior probability that the document belongs to different classes and assigns it to the class with the highest posterior probability, assuming that the features are conditionally independent, given the class (Varela, n.d).Naive Bayes classifier can be used for discrete as well as continuous inputs. In this paper, however, we focus on a binary classification problem with continuous inputs. When dealing with continuous data, a typical assumption is that the continuous values associated with each class are distributed according to a Gaussian distribution (NBC, 2014). In the Gaussian distribution, for every continuous attribute in the training data, we compute the mean and variance of the attribute in each class. Then probability density of any value given a class can be calculated by the following equation.

p ( x  v | c) 

 ( v  c ) 2

1 2 c

2

e

2 c 2

(1)

where: x is a continuous attribute, v is any value of x associated with class c, σc2 is the variance of the values in x associated with class c, and µc the mean of the values in x associated with class c 2.2.

Logistic Regression

Logistic regression is a special case of linear regression and is preferred in situations/problems where the dependent variable is a dichotomy because using linear regression for a binary response variable violates the following assumptions:  The response variable(Y') will range from -∞ to ∞ i.e. Y’ε [-∞,∞]  Normality of errors  Homoscedasticity Such violations give rise to the danger of misleading significance tests. Logistic regression measures the relationship between a categorical dependent variable and one or more independent variables(not necessarily) continuous, by using probability scores as the predicted values of the dependent variable (LR, 2014).Logistic regression can either be binomial (dependent variable is binary) or multinomial(dependent variable can take more than two values).However, in this paper we will be using binomial logistic regression because we are dealing with a binary classification problem. In a logistic regression model the probability of a dependent variable is given by the expression (Kleinbaum, Klein, 2010):

P

1 1 e

(  0   i xi )

(II)

where, β0 refers to slope, βi are unknown model parameters to be estimated from the data, xi independent variables, e represents the base of a natural algorithm whose value equals 2.71828 and P represents probability given by the following:

1 if P 0 if 2.3.

P  0.5 P  0.5

Hybrid Classification

Hybrid classification means applying classifiers in a sequence thereby forming a chain of classifiers. A chain of three classifiers can be represented as: C1 C2 C3 where Ci represents the different classifiers used. If the first classifier fails to classify the given instances of the problem, the unclassified instances are passed to the next classifier in the chain and so on until there are no more unclassified instances or all the classifiers in the chain have been used. The hybrid configuration used in this paper is: Logistic Regression Naive Bayes

[181]

Basharat and Chachoo/ COMMUNE-2015

3. Evaluation Metrics Used Table 1 shows the contingency table or confusion matrix for the problem in question. Table 1. A contingency table Relevant True Positives(TP) False Negatives(FN)

Diabetic Non-Diabetic

Non-Relevant False Positives(FP) True Negatives(TN)

Here, TP refers to the number of patients rightly identified as diabetic, FP refers to the number of patients who have been identified as diabetic but are not actually diabetic, FN refers to the number of patients who have been identified as non-diabetic but are actually diabetic and TN refers to the number of patients rightly identified as non-diabetic The following evaluation metrics have been used in this paper:

1.

RECALL(R): Percentage of diabetic patients that are identified.

R 2.

PRECISION (P): Percentage of relevant diabetic patients that are identified.

P 3.

TP  TN TP  FP  TN  FN

ERROR (E): Percentage of diabetic and non-diabetic patients that are wrongly identified. E

5.

TP TP  FP

ACCURACY (A): Percentage of diabetic and non-diabetic patients that are correctly identified.

A 4.

TP TP  FN

FP  FN TP  FP  TN  FN

FALLOUT (F): Percentage of diabetic patients that are wrongly identified.

F 6.

FP FP  TN

A COMBINED MEASURE F :(Fβ): It is the weighted average of precision and recall. F 

(  2  1)  P  R 2 P R

where

2 

(1   )



α ε [0, 1] and thus β2 ε [0,∞] However, the most frequently used is 'balanced F' where β = 1 which makes F equal to the and R i.e.

F1 

harmonic mean of P

2 P R PR

4. Experiment 4.1. Data To evaluate the effectiveness of various algorithms, we obtained data from "Pima Indians Diabetes Database". [182]

Basharat and Chachoo/ COMMUNE-2015

4.1.1

Data Set Information

Several constraints were placed on the selection of these instances from a larger database. In particular, all patients here are females at least 21 years old of Pima Indian heritage. The database consists of eight attributes plus the class attribute, the details of which are given in table 2: Column 1 refers to the name of the attribute in the data set, column 2 gives the type of attribute and column 3 refers to the details of a particular attribute. Table 2. Data Set Information

4.2.

Attribute name

Attribute type

Attribute information

Preg

Continuous

Number of times pregnant

Plas

Continuous

Plasma glucose concentration

Pres

Continuous

Diastolic blood pressure

Skin

Continuous

Triceps skin fold thickness

Insu

Continuous

2-hour serum insulin

Mass

Continuous

Body mass index

Pedi

Continuous

Diabetes pedigree function

Age

Continuous

Age

Class

Categorical

Diabetic/Non-Diabetic

Experimental Procedure

From a total of 768 instances of diabetes database, we use 500 instances for training the algorithm(training data) and the remaining 268 instances for testing the effectiveness of algorithms(test data).To implement naive Bayes we first segment the training set by class(diabetic/non-diabetic) and then calculate the mean and variance of each attribute in each class. From this mean and variance we calculate for each instance of the test data the probability densities pd 1, pd2,..., pd8 of values v1, v2,..., v8 for attributes x1, x2,..., x8 as shown in equation (I).Next we calculate the prior probability of each class and from the prior probability and probability densities we calculate the posterior probability of each class as shown:

posterior1  prior (C  diabetes )  pd1  pd 2      pd 8 posterior2  prior (C  nodiabetes )  pd1  pd 2      pd 8 The instance will be placed in the class with greater posterior probability. To implement logistic regression we use the logistic function already discussed in equation (II) of this paper which takes following form for the problem in question:

P

1 1 e

(III)

(  0  1 x1   2x2 ... 8 x8 )

We used least square estimation to estimate the value of βi where we choose the value of βi that minimizes the sum of squared residuals.Next we determine the value of P by plugging in the values of β i(already calculated) and the values of xi from the database using equation (III) and classify the instances by the following rule:

if  1( Diabetic ) P 0( Non  diabetic ) if 4.3.

P  0.5 P  0.5

Results

Table 3 compares naive Bayes, logistic regression and hybrid classifier in terms of recall, precision, accuracy, error, fallout and F1.The results indicate that naive Bayes outperforms logistic regression by a slight margin but the hybrid configuration completely outperforms both the naive Bayes and logistic regression with an accuracy of 92.9%. Table 3. Evaluation Results Attribute name

Recall

Precision

Accuracy

Error

Fallout

F1

Naïve Bayes

0.5588

0.7215

0.75

0.25

0.1325

0.6298

Logistic Regression

0.4412

0.7258

0.7239

0.27

0.1624

0.5488

Hybrid

0.8137

1

0.9291

0.0709

0

0.8973

[183]

Basharat and Chachoo/ COMMUNE-2015

4.4.

Discussion

Of the machine learning algorithms trialed in this study, apart from the hybrid configuration,naive bayes performs slightly better than the logistic regression. In real world scenarios, the results listed in Table 3 suggest that one should prefer naive bayes over logistic regression.However, Ng and Jordan (Prabowo, Thelwall, 2009),show that for several data sets logistic regression outperforms GNB when many training examples are available, but GNB outperforms LR when training data is scarce.In 2009 John Halloran (Halloran, 2009) stated that it is a safe bet to choose LR if NB is outperforming LR by a slight margin because LR will outperform NB within a relatively small amount of new training examples.Thus none of these algorithms performs better than the other in all situations and the choice of the algorithms depends largely on the type of data. Vohra and Teraiya (Vohra and Teraiya, 2012) stated in there study that no classifier outperforms the other classifier.They need each other to achieve best performance. 5. Conclusion and Future Scope As already discussed, no classifier can perform well in all situations, so it is better to switch to hybrid configuration where each classifier contributes to other classifier to achieve a high level of accuracy as shown in Table 3.In this paper GNB performs well both individually and in the hybrid configuration because the data used fulfills the independence feature assumption. If however, the data has highly correlated features then one can choose some other classifier over GNB to contribute to hybrid configuration. As part of future work, we can add more classifiers to the hybrid configuration chain thereby increasing the effectiveness of the resulting configuration. References Yang, Y. and Joachims, T., 2008. Text Categorization, Scholarpedia, vol. 03, pp. 4242. Sebastiani, F., 2002. Machine learning in automated text categorization, CSUR, vol. 34, no. 1, pp. 1-47. Prabowo, R. and Thelwall, M., 2009. Sentiment analysis: A combined approach, Journal of Informetrics, vol. 03, pp. 143-157. Vohra, S. and Teraiya, J., 2012. A Comparative Study of Sentiment Analysis Techniques, Journal of Information, Knowledge and Research in Computer Engineering, vol. 02, pp. 31. Ng, A. and Jordan, M., 2002. On discriminative vs. generative classifiers: A comparison of logistic regression and naive Bayes, Advances in neural information processing systems, vol. 14, pp. 841. Varela, P., Martins, A. and Figueiredo, M., An Empirical Study of Feature Selection for Sentiment Analysis, 9th Conference on Telecommunications, Castelo Branco, Portugal. NBC (Naive Bayes classifier) Wikipedia, 2014. The Free Encyclopaedia. LR (Logistic regression) Wikipedia, 2014. The Free Encyclopedia. Kleinbaum, D. and Klein, M., 2010. Logistic Regression: A Self-Learning Text, 3rd ed. Halloran, J., 2009. Classification: Naive Bayes vs Logistic Regression.

[184]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16-18March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Study and Analysis of Downstream ROF PON using TWDM concept Jayesh C. Prajapti*, Ekta Khimani, Shivani Raval Department of Electronics and Communication Engineering, U. V. Patel College of Engineering, Ganpat University, Kherva, Gujarat, India

Abstract Radio over fiber technology is becoming increasingly important now days for wireless communication area in order to support the ever-growing need of data traffic volume. In such systems, optical fiber is used as an ideal medium to transmit modulated optical signals by microwave signals to the remote sites of network. Due to offered distinguished low loss and high wide bandwidth characteristics, optical fibers are used with PONs where they serve as high capacity wireless distribution nodes. With possible deployment of the FTTH project, a PON can be used to overcome the challenges of fixed users. Suggested combined TWDM technology also offers increased capacity with reduced cost. In this paper, we have shown simulation of basic RoF link with QAM modulation technique using Optisystem 13 software, which can be is proven the best technology to be used with OFDM PON networks in future.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Radio Over Fiber; Orthogonal Frequency Division Multiplexing; Passive Optical Network; Gigabit Passive Optical Network; Radio Access Point; Fiber to the Home; Wavelength Division Multiplexing; Time Wavelength Division Multiplexing

1. Introduction 1.1.

ROF

Radio Over Fibre (RoF) provides better reliability, coverage network, and more secure for users. RoF is one of demanding technology in terms of wired and wireless networks. RoF is used to enhance the capacity and bandwidth for wireless signals over long distance. RoF system consists of a Central Site (CS) and a Remote Site (RS), which is connected by an optical fiber network shown by figure 1. RoF technology provides simplified RAP, which includes optoelectronic conversion devices, an amplifier and the antenna. Radio signal processing functions such as multiplexing, frequency up-conversion, carrier modulation are carried out at the Central Base Station (Kasim et al, 2008). The RF signals processing functions at the Central Base Station allow system operation and maintenance, equipment sharing, dynamic allocation of resources. If RoF network is used in GSM network, then the CS will work as Mobile Switching Centre (MSC) and the RS as the base station (BS).

*

Corresponding author. Tel.: +91 9824 549615. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Prajapti et al/COMMUNE-2015

1.2.

PON

PON is a combination of (Optical Distribution Network) ODN, optical line termination (OLT), multiple optical network units (ONU). PON is generally suits to managing protocols, transmission of convergence layer and physical medium depend layer. PON uses point-to-multipoint i.e. tree topology which carries frames of data between multiple optical network units (ONU) and optical line termination (OLT) via a passive optical splitter. The component elements of PON architecture are shown in figure 2.

Optical-access networks have been implemented to remove the access-network bandwidth problem. However, by adopting PON technology we can provide truly cost-effective solution. Long-reach optical-access networks introduce reduction of significant cost, direct connection of the customer to the core network, by passing metro network. 2. Access technologies 2.1

TDM PON

With future requirements for next-generation optical access, current NG-PON1 based on previous TDM-PONs standards are not enough to satisfy the current demand for a new revolution of technology. 40 Gb/s TDM-PON standards is a starting point towards higher capacity. The main issue with 40 G TDM-PON for NG-PON2 is difficulty in involving more users in the feeder fiber due to limited power budget, the availability and maturity of the components needed being low and finally the cost issue being more impactful for high-speed transmitter and receiver. Moreover, the limited reach is due to chromatic dispersion, which limits the transmission distance. 2.2

WDM PON

Table. 1

Parameters Splitting loss [dB] SNR penalty Bit rate transparency Open architecture and unbundling Support Fault finding (OTDR)

WDM-PON 3 to 5 dB ( filter loss) No Unlimited

TDM-PON 16 to 17 dB (32 user splitter loss) N4 ~ N5 ~ 10 Gb/s

Simple

Hard

Simple

Hard

WDM PON provide separate wavelength channel in the operation of transmission. However, this needs a point-topoint link from source to destination this approach will lead the WDM-PON to be impacted by high cost and poor resource utilization. To avoid the cost of WDM-PON or to remove the static wavelength is when several XG-PONs are ombined and separated by multiplexer and de-multiplexer. 2.3

OFDM PON

OFDM uses a large number of carriers which are closely-spaced with each other orthogonally to carry data traffic. Since the data rate carried by each subcarrier is low, the duration of each symbol is relatively large. Thus, the intersymbol interference can be efficiently reduced in a wireless multipath channel. In OFDM PON, cheaper electronic devices are used instead of costly optical devices, and ASIC-based DSP and AD/DA also reduce equipment costs (Ansari et al). OFDM-PON can be combined with WDM to further increase the bandwidth provisioning, and has therefore become a competitive technology for NG-PON2.

[186]

Prajapati et al/COMMUNE – 2015

Fig.3.

OFDM PON exhibits the advantages like enhanced spectral efficiency, it avoids use of costly optical devices.it allows use of cheaper electronic devices and converged wire line and wireless access Long-reach optical access suffers from the problem of high fiber chromatic dispersion. The OFDM modulation scheme can help address the chromatic and polarization-mode dispersion in optical links. Fig-3 shows RoF with QAM modulation. The Q-factor about 2.962 and low BER of 6.8e^-98 is obtained for QAM which is better than all other modulation techniques used .The resultant eye opening is also better. OFDM system gives better result with QAM modulation technique. 2.4

TWDM PON

TWDM is the combination of TDM and WDM technologies. Having advantage of higher capacity it is used in PON. Key features of such technology are (Bindhaiq et al, 2014) It support capacity from 128 Gbit/s up to 500 Gbit/s per feeder fibre, It support ONU i.e. (customer) from 256 up to 1024 per feeder fibre and the working range can be extended up to 20 to 40 km. 3. Performance of PON ROF link with TWDM As shown in figure 4, 32 users were provided service using PON ROF with TWDM concept .Here signals from various transmitters were multiplexed using WDM multiplexer. we have used 2.5Gbps transmission data rate which is the latest one used in GPON. Then we shifted PON signals in to optical domain. Later on they were transmitted over fiber maximum up to having length of 25 Km. We have passed them through EDFA with 5m length. Then at receiver side we have splitted it through power splitter of 1:4 and at ONU side, then we have detected optical signal with PIN photo detector and passed through filter then results observed through BER analyser. In this design, sequence length is 256 and samples per bit are 64 (Abdullah et al, 2013). We have observed that PSK modulation technique is best for 2.5 GPON Architecture (Dane et al, 2013).

Fig 4: Simple PON ROF link with TWDM

[187]

Prajapti et al/COMMUNE-2015

4. Simulation Results and discussion

Fig.5 Eye diagram of ROF PON basic downlink

Fig.6 Q factor output

From fig.5 we can analyze the output of basic PON ROF downlink. Then we can show that we are getting Q factor is about 6.71 and BER is about 10^-13 order which is low comparable to standard BER value. We can see the Q factor out put in the fig 6. We can see the Q factor out put in the above diagram. Maximum Q factor is about 6.74. Eye opening is very good so we can say that we can get the high quality of signal at the output. 5. Conclusion Using Optisystem13 software our simulation results shown that output of the system provides better quality of signals. Use of TWDM technology with ROF PON has shown enhanced capacity of such network in term of received signal quality with reduced of overall cost of implementation of actual system. Transmission of OFDM signals through such ROF PON with TWDM technology can give better possible results. References Salem Bindhaiq , AbuSahmah M. Supa'at, Nadiatulhuda,Zulkifi Abu BakarMohammada, Redhwan Q Shaddad , Mohamed A. Elmagzoub, Ahmad Faisal;” Recent development on time and wavelength-division multiplexed passive optical network (TWDM-PON) for next-generation passive optical network stage 2 (NG-PON2)” ; optical switching and networking;s.d 2014. Abdullah O. Aldhaibani, S. Yaakob, R.Q. Shaddad, S.M. Idrus, M.Z. Abdul Kadi,A.B. Mohammad;”2.5 Gb/s hybrid WDM/TDMPON using radio over fiber technique”,optic 2013. Abhishek Dixit, Bart Lannoo, Didier Colle, Mario Pickavet;” Wavelength Switched Hybrid TDMA/WDM (TWDM) PON: a Flexible Next Generation Optical Access Solution”, 2012, IEEE. Zhang Chunlei, Geng Ling, Zhang Pengtu;”An Overview of Integration of RoF with PON”; International Conference on Computer Application and System Modeling; 2010 IEEE. Priya Dane, Hemani Kaushal;” Characterization of RoF GPON Performance for Different Modulation Schemes”, 2013 IEEE. Norazan Mohd Kasim ,”Recent Trends in Radio Over Fiber Technology”, Penerbit Universiti Teknologi Malaysia, 2008 Ansari N, Zhang J, “Media access control and resource allocation for next generation passive optics networks” from www.springer.com.

[188]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Recognition of Typewritten Gurmukhi Characters Navdeep Lataa*, Simpel Rani Jindalb a

Department of Computer Engineering, Malout Institute of Management and Information Technology, Malout, India b Department of Computer Engineering, Yadavindra College of Engginering, Talwandi Sabo, India

Abstract The field of Document Analysis and Recognition (DAR) is mushrooming day by day and Character recognition is an important subject of this field. Ample work is done for various scripts particularly in case of English but in case of Indian scripts, the research is limited to printed and handwritten text of fine documents. As far as degraded documents such as typewritten document and background problem are, concerned, little work has been reported. In this paper, we have recognized isolated typewritten Gurmukhi characters in which we have considered various types of typewritten degradations such as broken, heavily printed and shape variant characters by using various feature extraction and classification techniques. This process is carried out in two stages. In first stage, we are extracting the features of typewritten Gurmukhi characters by using structural, statistical features and their combination like Zoning, Transition features, Distance Profile features and Neighbour pixel zone etc. for generating feature sets. In second stage, feature sets are used for recognizing typewritten Gurmukhi characters by using SVM and K-NN classifiers. Maximum accuracy of 96.78% has been achieved by using SVM classifier.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: OCR; Typewritten Document; Feature Extraction; Zoning; Classifer

1. Introduction OCR is the process of converting scanned document into a computer processable format (Lehal and Singh, 2000). This process is divided into five steps viz. Pre-processing, Segmentation, Feature Extraction, Classification, and PostProcessing. Pre-processing is representing the scanned image in binary format for processing the image for recognition. It reduces noise and distortion, removes background problem to much extent and also enhance the image and performs skeletonizing of the image thereby simplifying the processing for the rest of the stages. In Segmentation stage document is segmented into its subcomponents. Segmentation means dividing something into different categories by separating the logical parts like text from graphics etc. Here Segmentation means separation of characters, words, lines which directly affects the recognition rate of the script. Feature extraction in character recognition system is used to extract the feature of a character that uniquely identifies the character. It does so by analyzing its shape and comparing its features against a set of rules stored on OCR engine that distinguishes each character. Classification perform the major step as it uses the feature extracted in pervious stage as an input to identify the text segment according to already decided rules. In Post-Processing stage, refining process is carried out over the above result in order to recognize the character/word using context. It is responsible for the output of the best solution and is often implemented as a set of techniques that rely on character frequencies, and other context information (Lehal and Singh, 1999). How typewritten documents are different from printed? On the analysis of these typewritten dataset, we concluded the following facts: In typewritten document, each character appeared considerably stronger or fainter than its neighbours in contrast to other printed documents.  Many typewritten documents survived only as carbon copies of the originals produced on a thin paper, which has prominent texture. (Pletschacher et.al , 2009)

*

Corresponding author. Tel.: +91 9888 373733 E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Lata and Jindal/ COMMUNE-2015

  

Printed documents produced results entirely in an ordinary sequential manner whereas, typewritten documents faces a problem. Many typewritten documents produced touching character problem. Typewritten Historical documents were affected by problems of ageing and repeated use resulting in discolouration, strains, background problem, etc.

2. Degradations In 20th Century, typewriters were essential tools for recording the written word. Even in the beginning of 21st Century Professional writers and Government offices used typewriters conspicuously. Typewritten documents are a kind of degraded documents. It is a mechanical device with keys. When the keys are pressed characters are printed on a paper. Typically one character is printed per key press. The machine prints the characters by making ink impressions of elements resembling the pieces of cast metal type. An example of typewritten document in Gurmukhi script has been shown in Fig. 1.

Fig. 1. Isolated character typewritten document.

In this paper, we have considered the three problems related to typewritten Gurmukhi characters i.e. broken character problem, heavy printing problem, and shape variance (Cannon et.al, 1999; Kumar, 2008). 2.1.

Broken character problem:

Broken character is composed of several connected components, i.e. a set of black pixels that are contiguous (Nagy et.al., 1999). Main reasons of broken characters are Poor Printing, Low-resolution scanning, and threshold errors. Broken characters pose many problems in recognition system as written below and shown in Fig. 2(a):1. 2. 3.

Each connected component cannot be treated as single complete character. Simple merging of small components is not feasible since it is not clear beforehand which component belong to which character. Incorrectly segmented characters are not likely to be correctly recognized.

(a)

(b)

(c)

Fig. 2. (a) Broken character problem (b) Heavy printed isolated characters (c) Shape variance problem

2.2.

Heavy Printing Problem

In typewritten document heavy printed occurs due to extra pressure on keystrokes, hard pressing the keys, extra ink on ribbon, and double strokes. Generally, the darkness of the character results in heavily printed document and shape distortion as shown in Fig. 2(b). However, characters are easily isolated, heavy print can distort their shapes, making them unidentifiable (Nagy et.al., 1999). These problems arises difficulty in feature extraction due to extra black pixels, which leads to shape distortion, touching character problem, and zoning problem. 2.3.

Shape Variance Problem

Shape variance means distortion in original character shape. Here distortion means size variance, skewing problem, and shape changes due to heavy printing. These distortions make it difficult to recognize the character. Fig. 2(c) shows some problem related to shape variance. 3. Feature Extraction Feature extraction can be defined as extracting the most representative information from the raw data, which minimizes the within class pattern variability while enhancing the between class pattern variability. For this purpose, a set of features are extracted for each class that helps distinguish it from other classes while remaining invariant to characteristic differences within the class. A good survey on feature extraction methods for character recognition can be found (Trier et.al., 1996).

[190]

Lata and Jindal/ COMMUNE-2015

3.1.

Projection Histogram Features (F1-F4)

Characters are represented by projecting the pixel values count in various directions. This representation creates onedimensional signal from a two dimensional image which can be used to represent the character image (Siddharth et.al., 2011). Projection histogram is a count of number of pixels in specified direction in a line. In our work, we used threedirection viz. Horizontal, Vertical and Diagonal. We have computed the Projection histograms by counting the number of foreground pixels in a particular direction. In horizontal histogram, these pixels are counted row wise, i.e. for each row. In vertical histogram, the pixels are counted column wise. In diagonal-1and diagonal-2 histogram the pixels are counted by left diagonal wise and right diagonal wise respectively. These types of features are used by G.S. Lehal for extracting various left, right, top, and bottom profiles of Gurmukhi characters in (Lehal and Singh, 1999). (Siddharth et.al., 2011) used projection histograms for extracting features of Gurmukhi characters and used projection histograms on 3X3 pattern as depicted in Fig. 3.

(a)

(b)

(c)

(d)

Fig. 3. Projection Histograms (a) Horizontal Histogram (b) Vertical Histogram (c) Diagonal-1 Histogram (d) Diagonal-2Histogram (Nagy et.al., 1999)

3.2.

Zoning Density Features (F5)

The frame containing the character is divided into several overlapping and non-overlapping zones (Sharma and Jain, 2010). Extracted character image is segmented into windows which is the best method for handling size variance character-set. We have created 16(4 X 4) zones of our 60 X 60 sized sample images by horizontal and vertical division. By dividing the number of foreground pixels in each zone by total number of pixels in each zone, i.e. we can obtain the density of each zone. Thus, we obtained 4 X 4 zoning density features, i.e. total number of features vector for one character is sixteen (16) as shown in Fig. 4(a).

(a) Fig. 4.(a) 4* 4 Zoning of characte

3.3.

1

1

1

1

1

1

1

0

1

0

1

1

2

1

2

0

1

0

1

1

2

1

2

0

1

1

1

1

1

1

1

0

(b)

(c)

; (b) 4×4 Neighbour pixel zone image, (c) Neighbour pixel computation

Neighbour Pixel Zone (F6 – F7)

Foreground pixels '0' are considered as the base of the zone. Every pixel of Fig. 4(c) is computed by taking into account the number of foregrounding pixel attached to it from all the sides in Fig. 4(b) of the same size, e.g. the first pixel of Fig. 4(b) is having only one count of foregrounding pixel that is why it is being assigned value 1. Further, we compute the average number of neighbours in each zone by two methods: (1) average with respect to size of zone and (2) average with respect to total number of black pixel in a zone. 3.4.

Transition Features (F8-F9)

In this structural feature, total numbers of transitions from background pixel ‘1’ to foreground pixel ‘0’ in the vertical and horizontal directions are noted. The transition feature used here is mostly similar to that proposed by (Gadar et.al., 1997) To calculate transition information, image is scanned from left-to-right, top-to-bottom. Feature vector size is 120, i.e. 60 feature vectors for each direction. 3.5.

Background Directional Distribution Features (BDD) (F10)

In this directional distribution of neighbouring background pixels to foreground pixels is being computed in 8 direction i.e. D1-D8. The masks is being used for calculation distribution in each direction. The foreground pixel 'X' in centre is considered to calculate background directional distribution values. We compute directional distribution value for foreground pixel in direction d1. By adding, the corresponding mask values of neighbouring background pixels and obtain all the other direction value similarly for each foreground pixel. At last similar directional distribution values are summed up in each zone and compute 8 directional distribution feature values for each zone (Siddharth et.al., 2011).

[191]

Lata and Jindal/ COMMUNE-2015

Fig. 5. Masks for 8 different direction for computing directional distribution

3.6.

Distance Profile Features (F11-F15)

Profile counts the number of pixels (distance) from bounding box of character image to outer edge of character. This distance traced here can be vertical, horizontal or radial. Fig. 6 shows the distance profile features of character ‘ the profiles of four sides, i.e. left, right, top, and bottom can be used. Left and right profiles are traced by horizontal traversing of distance from left bounding box in forward direction and from right bounding box in backward direction respectively to outer edges of character. Similarly, top and bottom profiles are traced by vertical traversing of distance from top bounding box in downward direction and from bottom bounding box in upward direction respectively.

Fig. 6. Distance Profile Features of character ‘

3.7.

Performance Analysis of Structural and Statistical Features

Data is being collected from various typewriters and scanned at 600 dpi on Epson L210 scanner and extracted isolated character of image size 60x60 in bmp format. Feature vector are the values, which are used for recognizing isolated typewritten Gurmukhi characters. The techniques (individual or combinations) and there number of feature vector for each technique is as follows: Table 1. Feature Vector Size according to Feature Extraction Techniques S.No.

Feature extraction Technique

Vector

S.No.

Feature extraction Technique

Vector

1

Projection Histogram (F1-F4)

358

8

PH+ ZD+TF+BDD+DP (F17)

862

2

Neighbourhood pixel zone (F5-F6)

32

9

PH+NPZ +TF+BDD+DP (F18)

878

3

Zoning density (F7)

16

10

PH+NPZ+ZD +BDD+DP (F19)

774

4

Transition feature (F8-F9)

120

11

PH+NPZ+ZD+TF +DP (F20)

766

5

Background Directional Distribution (F10)

128

12

PH+NPZ+ZD+TF+BDD (F21)

654

6

Distance Profile (F11-F15)

240

13

PH+NPZ+ZD+TF+BDD+DP(F22)

894

7

NPZ+ZD+TF+BDD+DP (F16)

536

4. Classification In OCR, classification stage assigns labels to character images on the base of features extracted and the relationships among them. In simple terms, this part of OCR which recognizes individual characters and returns the output in machine editable form. 4.1.

Support Vector Machines (SVM)

SVM are based on statistical learning theory that uses supervised learning (Burges, 1998). In supervised learning, a machine is trained instead of being programmed, to perform a given task on a number of input-output pairs. According to this concept, training means choosing a function which best describes the relation between the inputs and the outputs. A practical guide for SVM and its implementation is available at (Burges, 1998) (Hsu, 2003). We have considered two kernel parameters that are linear and polynomial kernel (Siddharth et.al., 2011).

[192]

Lata and Jindal/ COMMUNE-2015

4.2.

K- Nearest Neighbour (K-NN) classifier

K-NN classifier relates unknown pattern to the known according to some distance or some other similarity function by using the instance based learning. It classifies the object by majority vote of its neighbour. Because it considers only neighbour object to a particular level, it uses local approximation of distance function. 5. Results and Dıscussıon Table 2: Recognition Accuracy by using (a) SVM (b) K-NN (c) Summarization of (a) & (b) table Rec. Acc. (%age) K=0 Structural Features Projection Histogram (F1-F4) 62.14 Transition Feature (F8-F9) 72.50 Distance Profile (F11-F15) 22.67 Statistical Features Zoning Density (F7) 30.17 Neighbourhood Pixel Zone (F5-F6) 25.71 BDD (F10) 24.82 Combination of Structural & Statistical Features NPZ+ZD+TF+BDD+DP (F16) 16.07 PH+ ZD+TF+BDD+DP (F17) 14.28 PH+NPZ +TF+BDD+DP (F18) 14.28 PH+NPZ+ZD +BDD+DP (F19) 14.28 PH+NPZ+ZD+TF +DP (F20) 34.82 PH+NPZ+ZD+TF+BDD (F21) 13.39 PH+NPZ+ZD+TF+BDD+DP (F22) 14.28 Feature extraction

K=1 92.50 74.64 84.82 53.21 46.78 52.5 58.75 64.82 64.82 64.82 96.78 59.10 64.82

(a) Feature (F1-F4) (F8-F9) (F11-F15) (F7) (F5-F6) (F10) (F20) (F21) (F22)

k=1 79.6 58.3 60.1 41.2 24.2 33.5 78.3 36.6 39.4

Recognition Accuracy (%age) k =3 k =5 k =7 k =9 77.3 77.5 75.8 74.8 55.8 60.1 59.2 61.7 60.3 59.1 59.6 60.5 42.8 42.8 42.6 42.3 24.4 25.7 25.1 25.5 35.5 35.7 35.3 35.3 78.9 80.3 79.1 79.2 37.5 37.5 37.5 37.1 39.4 40.7 39.2 39.2

k=11 73.9 59.6 59.1 41.6 26.2 33.4 80 35.5 37.5

(b) Classifiers SVM K-NN

Structural 92.50 79.64

Features Statistical 53.21 42.86

Combined 96.78 80.36

(c) The results of SVM, k-NN classifier using various kinds of structural and statistical features and their various options have been tabulated in Tables 2(a, b) that shows the recognition accuracy using SVM, k-NN with different values of k on different sets of structural features. Table 2(a) shows that maximum recognition accuracy of 96.78 % with kernel value '1' is achieved over features 'F20' using SVM classifier with feature vector size 766. Table 2(b) shows maximum accuracy of 80.3% at k = 5 of Feature 'F20' from structural, statistical and various combinational features. From Table 2(c) one can see that maximum recognition accuracy has been achieved with SVM i.e. 96.78% for typewritten Gurmukhi characters. Table 3 shows confusion matrix of feature ‘F20’result obtained by applying SVM. Table 3: Confusion Matrix of feature 'F20' by applying SVM Char

%age of Recognized character

Accuracy 100%

(93.75%)

(6.25%)

93.75%

(93.75%)

(6.25%)

93.75%

12.5%)

5

(6.25%)

93.75%

5 (93.75%)

[193]

Lata and Jindal/ COMMUNE-2015

Char

%age of Recognized character 5

5 93.75%

Average %age

6. 6.1.

Accuracy

93.75%

75

25

75%

75

25

75% 96.78%

Conlusion and Future work Conclusion

This paper presents recognition of isolated typewritten Gurmukhi characters in which two stages of OCR has been carried out in detail that are feature extraction and classification. Major problems undertaken for recognition are broken character, heavy printed, and shape variance characters. The structural, statistical and there combinational features selected for recognition of typewritten Gurmukhi characters are used such as neighbour pixel zone, zoning density, distance profile, BDD and transitions feature. SVM and K-NN have been applied for classification purpose. By analysing result and discussion section, we have concluded that SVM shows best recognition rate of 96.78% over feature ‘F20’ for isolated typewritten Gurmukhi characters. 6.2.

Future Work

We believe that this paper has opened up many areas of research in typewritten Gurmukhi documents that can be explored in future. The work can also be extended to recognition of other typewritten Indian scripts containing these kinds of degradations and recognition of these documents can be enhanced by creating a complete Optical Character Recognition system as little work has been proposed. References Lehal, G. S., & Singh, C.(2000) "A Gurmukhi script recognition system.", Proceedings. 15th International Conference In Pattern Recognition IEEE, vol. 2, pp. 557-560. Lehal, G. S., & Singh, C.(1999) "Feature extraction and classification for OCR of Gurmukhi script." VIVEK-BOMBAY, no. 2 pp: 2-12. Pletschacher, S., Hu, J., & Antonacopoulos, A. (2009) "A new framework for recognition of heavily degraded characters in historical typewritten documents based on semi-supervised clustering." 10th International Conference In Document Analysis and Recognition, pp. 506-510. Cannon, M., Hochberg, J., & Kelly, P. (1999) "Quality assessment and restoration of typewritten document images." International Journal on Document Analysis and Recognition 2, no. 2-3 pp: 80-89. Kumar, M. (2008). "Degraded Text Recognition of Gurmukhi Script." PhD diss., Thapar University Patiala. Nagy, G., Nartker, T. A., & Rice, S. V. (1999) "Optical character recognition: An illustrated guide to the frontier." In Electronic Imaging, pp. 58-69. Trier, Ø. D., Jain, A. K., & Taxt, T. (1996) "Feature extraction methods for character recognition-a survey." Pattern recognition 29, no. 4 pp: 641662. Siddharth, K. S., Jangid, M., Dhir, R., & Rani, R. (2011) “Handwritten Gurmukhi Character Recognition Using Statistical and Background Directional Distribution Features”, (IJCSE), 3 no. 6, pp. 2332-2345. Sharma, D., & Jain, U. (2010). "Recognition of Isolated Handwritten Characters of Gurumukhi Script using Neocognitron." International Journal of Computer Applications 10, no. 8, pp : 10-16. Gader, P. D., Mohamed, M., & Chiang, J. H. (1997). “Handwritten word recognition with character and inter-character neural networks”, IEEE Transactions on Systems, Man, and Cybernetics-Part B: Cybernetics, 27 no. 1, pp. 158-164. Burges, C. J. (1998). “A tutorial on support vector machines for pattern recognition”, Knowledge Discovery and Data Mining, 2 no. 2, pp. 121-167. Hsu, C. W., Chang, C. C., & Lin, C. J. (2003). "A practical guide to support vector classification." National Taiwan U., www. csie. ntu. edu. tw/cjlin/papers/guide/guide. pdf .

[194]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Effect of Buried Oxide (BOX) in the Drift Region of a Super Junction MOSFET Deepti Sharma , Rakesh Vaid* Department of Physics and Electronics, University of Jammu, Jammu, India

Abstract In this paper, we propose a novel Superjunction (SJ) MOSFET with buried oxide (BOX) in the drift region to reduce the onresistance and to address the tradeoff between specific on-state resistance (Ron) and breakdown voltage (Bv). The proposed device is simulated using a 2-D numerical device simulator. The results show an improvement in the breakdown performance as compared to the conventional SJMOSFET due to a reduction in the vertical electric field. The effect of Buried Oxide (BOX) in both the n and ppillars of a Super Junction MOSFET is also investigated. In addition, the proposed device NBOX1-SJMOSFET with BOX in n-pillar shows a linear relation between the Bv and Ron as compared to the conventional device. Various results obtained reveal that the device having BOX in the n-pillar gives the better results. The breakdown voltage has been increased by about 20% in NBOX1SJMOSFET leading to improvement in various other performance parameters as compared to proposed device with BOX in the ppillar and conventional device.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Carbon Superjunction ; Breakdown Voltage ; Buried Oxide ; On-Resistance ; SJMOSFET; Device Simulator.

1. Introduction The Superjunction (SJ) concept was introduced to achieve high breakdown voltage (Bv) and high current densities along with low on-resistance (Ron) to improve square law relation between Ron and Bv in the conventional power MOSFETs to a linear relation [Ron  Bv] as shown by (Deboy et al., 1998). Various SJ device configurations have been studied to improve the relation between Bv and Ron such as Super junction VDMOS, Trench Gate Super junction MOSFET etc discussed by (Onishi et al., 2015), (Chen et al., 2012), (Zhu et al., 2013), (Zhu et al., 2013), and (Lin et al., 2009). In this paper, the concept of buried-oxide has been used in SJMOSFET to enhance the Bv and to reduce Ron. New concepts have been introduced to directly obtain an insulating layer underneath the silicon surface by implantation of oxygen ions in a silicon wafer by SIMOX (Separation by Implanted Oxygen) by (Izumi et al., 1978) or by direct bonding of two oxidized silicon wafers by BESOI (Bond and Etch back SOI) by (Maszara et al., 1988), (Mishima et al., 1989). We have studied the effect of BOX (buried oxide) in the n and p-pillars of SJMOSFET with the help of a 2-D device simulator PISCES-II. The device dimensions are chosen to simulate a typical device structure. These simulations are aimed at understanding the device physics through various electrical quantities like potential distribution, electric field distribution, and electron concentrations etc. in different regions of the device both in on/off states. The I-V characteristics were compared for different devices for various gate and drain voltages. The effect of BOX in the n and p-pillars of the SJMOSFET was investigated. Various results obtained showed that the device having BOX in the npillar gives the best results. The current density is minimum and the specific on-resistance is slightly maximum for this case; however the breakdown voltage increases with BOX in the n pillar. We also demonstrated that the relation between Ron and the Bv is improved in NBOX1-SJMOSFET as compared to the conventional device. The breakdown voltage has been increased by about 20% in NBOX1-SJMOSFET leading to improvement in various other performance parameters e.g. current, potential etc.

* Corresponding Author. Tel.: +91 9419 E-mail address: [email protected]

106794.

ISBN: 978-93-82288-63-3

Sharma and Vaid /COMMUNE – 2015

2. Device Description Fig.1. shows the typical cros-section of proposed NBOX1-SJMOSFET. All dimensions of the proposed NBOX1SJMOSFET structure are listed in Table 1 taken on the basis of literature by (Sharma et al., 2011).

Fig. 1. Schematic view of Proposed NBOX1-SJMOSFET

The proposed work is based on the superjunction structure that has been analyzed rigorously by (Chen et al., 1998). We can propose the fabrication of NBOX1-SJMOSFET with BOX (buried oxide in the drift region with a structural modification in the n-pillar of the drift region i.e. by inserting BOX) and PBOX1 that is based on literature review. For our work, we choose to fabricate BOX in our device by using separation-by-implanted oxygen process in specific areas where we selectively create the BOX. Table 1: Design Parameters used for NBOX1-SJMOSFET Simulation. Parameter

Value

WP

2.0m

WN

2.0m

tox

0.08m

tepi

16m

Nn-

5.01015 cm-3

Nn+

1.01019 cm-3

Np

1.01017 cm-3

tBOX

2.8m

lBOX

1.6m

3. Results and Discussion In this section, we have simulated and discussed NBOX1, PBOX1, and the conventional device (SJMOSFET that is without BOX) and performed numerical simulations for their current-voltage characteristics, breakdown performance etc. In the blocking state of a SJMOSFET, the charge is counter balanced by exactly the same amount of charge of the opposite type. Fig. 2(a) & (b) are showing the 2-D electron concentration in the structures of various simulated SJMOSFETs at Vds=18.2V.

[196]

Sharma and Vaid /COMMUNE – 2015

(a)

(b)

Fig.2. Simulated electron concentration at Vds=182V in the structures of (a) PBOX1-SJMOSFET (b) NBOX1-SJMOSFET

Fig. 3 shows the comparison of breakdown plot of various devices. From fig. 3, we can see that the breakdown voltage of the conventional device is about 182.0V and the breakdown voltage of the NBOX1-SJMOSFET is about 219.2V. The improvement in the breakdown voltage performance of the proposed NBOX1-SJMOSFET device is about 20% whereas this percentage is less in case of PBOX1 that is only 6.37% as compared to conventional device. Fig. 4(a) and (b) are showing the 2-D current flow lines in the structures of various simulated SJMOSFETs at Vds=40V.

Fig.3. Comparison of breakdown performance of NBOX1, PBOX1 & Conventional device (Vgs=0V).

(a)

(b)

Fig.4. Simulated 2-D current flow lines at Vds=40V in the structures of (a) PBOX1-SJMOSFET (b) NBOX1-SJMOSFET.

In the simulated device NBOX1, the vertical electric field along the direction of the current flow is reduced when cut at the edge of the n- pillar near the junction i.e. at x=2µm voltage as shown in fig.5 (a) and (b) due to the SOI–RESURF

[197]

Sharma and Vaid /COMMUNE – 2015

(reduced surface field) effect explained by (Kanechika et al., 2005). It is due to the fact that BOX is introduced in the npillar of SJMOSFET that leads to reduction in electric field. The vertical electric field along the direction of the current flow in PBOX1 is also reduced but this enhancement is more in case of NBOX1 as can be seen from the comparative plot shown in Fig.5 (b) and (c) of NBOX1, PBOX1 and Conventional device at off state. Fig. 6 (a) and (b) shows the drain and transfer characteristics of various simulated devices. It can be seen that conventional device is having higher drain and transfer current in comparison to other devices and current is less in case of NBOX1. It is due to the fact that BOX is introduced in n-pillar which has blocked the path of current and it leads to reduction of current in NBOX1.

(a)

(b)

Fig.5. Magnitude of electric field of various simulated devices along the direction of current flow near the junction i.e. at 2µm at (a) Vds=182V (b) Vds=40V.

(a)

(b)

Fig.6. Comparison of typical (a) Ids-Vgs (b) Ids-Vds characteristics of NBOX1, PBOX1 & Conventional devices Table 2: Breakdown Voltage (Bv) & specific on- resistance (Ron.A) values of all devices. Device

Bv (Volts)

Ron.A(mΩcm2)

Conventional

182.0

0.0569

PBOX1

193.6

0.0409

NBOX1

219.2

0.0425

Table 2 is showing values of breakdown voltage and specific on-resistance for various devices. From there we can conclude that Ron.A for conventional device is more as compared to NBOX1 and PBOX1 which showed the improvement of breakdown voltage in NBOX1. The improvement in breakdown voltage also seems to be depending on length of BOX (lBOX). So, we can conclude from various results that BOX in n-pillar showed improvements in various parameters as compared to PBOX1 & conventional device. We demonstrated that the use of buried oxide in the n-pillar of SJMOSFET leads to the decrease in the vertical electric field along the direction of current flow. The main difference in the structure of SJMOSFET (conventional) and various other devices resulted in the improvement in the breakdown [198]

Sharma and Vaid /COMMUNE – 2015

performance as well as electric field but this improvement is more in case of BOX (Buried-Oxide-In-Drift-Region) layer in the n-pillar of the SJMOSFET i.e. NBOX1. 4. Conclusions We have demonstrated that in NBOX1-SJMOSFET, introducing BOX in the n-pillar of SJMOSFET results in improvement in breakdown voltage due to reduction in the vertical electric field. Using 2-D numerical simulations, we have shown that in the proposed devices NBOX1 has higher breakdown voltage (Bv), lower specific on-resistance (Ron.A), higher switching speed as compared to the conventional device. This work indicates that nearly 20% improvement in the breakdown performance of the device can be achieved as compared to the conventional as well as other simulated devices, PBOX1. References Deboy, G., Marz, M., Stengl, J. P., Strack, H., Tihanyi, J., Weber, H., 1998. A new generation of high voltage MOSFETs breaks the limit line of silicon, Proc. IEDM Tech. Dig., San Francisco, CA, USA, p.683. Onishi, Y., Hashimoto, Y., 2015. Numerical analysis of specific on-resistance for trench gate superjunction MOSFETs, Japanese Journal of Applied Physics 54, p.024101. Chen, J., Sun, W., Zhang, L., Zhu, J., Lin, Y., 2012. A Review of Superjunction Vertical Diffused MOSFET, IETE Technical review 29, p.44. Zhu, J., Zhang, L., Sun, W., Qian, Q., Ma, W., Yang, Z., Lu, S., 2013. Analysis of the electrical characteristics of 600 V-Class electron irradiated fast recovery Superjunction VDMOS, Solid-State Electronics 80, p.38. Zhu, J., Yang, Z., Sun, W. F., Qian, Q. S., Xu. S., Yi, Y. B., 2013. High-voltage superjunction VDMOS with low reverse recovery loss, Electronics Letters 49, p.219. Lin, W. C., and Jun, S., 2009. An oxide filled extended trench gate super junction MOSFET structure, Chinese Physics B 18, p.1231-01. Izumi, K., Doken, M., Ariyoshi, H., 1978. C.M.O.S. devices fabricated on buried SiO2 layers formed by oxygen implantation into silicon, Electronics Letters 14, p.593. Maszara, W. P., Goetz, G., Caviglia, A., McKitterick, J. B., 1988. Bonding of silicon wafers for silicon-on-insulators, J. Appl. Phys. 64, p.4943. Mishima, H., Yasui, T., Mizuniwa, T., Abe, M., Ohmi, T., 1989. Particle-free wafer cleaning and drying technology, IEEE Trans. Semiconductor Manufacturing 2, p.69. Sharma, D., Vaid, R., 2011. Effect of Drift Region Doping and Column Thickness Variations in a Super Junction Power MOSFET: a 2-D Simulation Study, Journal of Nano- and Electronic Physics 3, p.1112. Chen, X. B., Mawby, P. A., Board, K., Salama, C. A. T., 1998. Theory of a novel voltage sustaining layer for power devices, Microelectronics Journal 29, p.1005. Kanechika, M., Kodama, M., Uesugi, T., Tadano, H., 2005. A Concept of SOI RESURF Lateral Devices with Striped Trench Electrodes, IEEE Trans. on Electron devices 52, p.1205.

[199]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

On Parameter Estimation of Erlang Distribution using Bayesian Method under different Loss Functions Kaisar Ahmada*, S. P. Ahmada and A. Ahmedb a Department of Statistics, University of Kashmir, Srinagar, India Department of Statistics, Aligarh Muslim University, Aligarh, India

,b

Abstract: In this paper, Erlang Distribution is introduced. Bayesian method of estimation has been employed to estimate the parameters of Erlang Distribution using Jeffrey’s and extension of Jeffrey’s priors under different loss functions. The classical Maximum Likelihood Estimator is also obtained.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Erlang distribution; Jeffrey’s and Extension of Jeffery’s prior; loss functions

1. Introduction The Erlang distribution is a continuous probability distribution with wide applicability, primarily due to its relation to the exponential and Gamma distributions. The Erlang distribution was developed by A. K. Erlang (1909) to examine the number of telephone calls that could be made at the same time to switching station operators. This distribution can be expressed as waiting time and message length in telephone traffic. If the duration of individual calls are exponentially distributed then the duration of succession of calls is the Erlang distribution. Bhattacharyya and Singh (1994) obtained Bayes estimator for the Erlangian queue under two prior densities. Haq and Dey (2001) addressed the problem of Bayesian estimation of parameters for the Erlang distribution assuming different independent informative priors. Suri et al. (2009) used Erlang distribution to design a simulator for time estimation of project management process. Damodaran et al. (2010) obtained the expected time between failure measures. Further, they showed that the predicted failure times are closer to the actual failure times. The pdf of an Erlang variate is

k

y k 1e y for y  0, k  N and   0 (k  1)! Where  and k are the rate and the shape parameters, respectively. Such that k is an integer number. f ( y;  , k ) 

(1.1)

2. Materials and Methods There are two main philosophical approaches to statistics. The first is called the classical approach which was founded by Prof. R. A. Fisher in a series of fundamental papers round about 1930. In this approach the unknown parameters are taken as fixed. The alternative approach is the Bayesian approach which was first discovered by Reverend Thomas Bayes. In this approach, parameters are treated as random variables and data is treated fixed. Recently this approach has received great attention by most researchers. Rahul et al. (2009) have discussed the application of Bayesian methods. An important pre-requisite in this approach is the appropriate choice of priors for the parameters. Very often, priors are chosen according to ones subjective knowledge and beliefs. * Corresponding author. Tel.: +91 9797 246287 E-mail address: [email protected] ISBN: 978-93-82288-63-3

Ahmad et al/COMMUNE – 2015

The other integral part of Bayesian inference is the choice of loss function. No specific analytical procedure is determined to choose the appropriate loss function. A number of symmetric and asymmetric loss functions used by various researchers; see Zellner (1986), Ahmed et al. (2013) and S.P. Ahmad and Kaisar Ahmad (2013) etc. 2.1 Maximum likelihood estimation Let y1 , y2 ,...,y n be a random sample of size n having probability density function (1.1), and then the likelihood function is given by ( ) nk L( x;  , k )  (k  1)!

n

y

k 1 i



n

 yi i 1

e

(2.1.1)

i 1

The log likelihood function is given by n



log L( x;  ,  )  nk log   k  1

log yi  

i 1

n

y

i

 log(k  1)!

(2.1.2)

i 1

Differentiating above equation w.r.t.  and equating to zero.

ˆ 

nk (2.1.3)

n



yi

i 1

3. Bayesian Method of Estimation Posterior Density under Jeffrey’s Prior

3.1

Let y1 , y2 ,...,y n be a random sample of size n having pdf (1.1) and the likelihood function (2.1.1). Jeffrey’s prior for  is given by g ( ) 

1



(3.1.1)

By using the Bayes theorem, we have

 

1  | y  L( y |  ) g ( )

(3.1.2)

Using (2.2.1) and (3.1.1) in (3.1.2), we get

 

nk 1

( ) (k  1)!

1  | y 





 1  | y  Knk1e

y

k 1



e

i

n

 yi i 1

i 1



n

 yi i 1

(3.1.3)

 .

Where K is independent of

   K

and

n

 yi   i 1  nk n



nk

Using the value of K in (3.1.3), we have n   nk1   yi    e i 1    1  | y    nk    





n

 i 1

 yi   

nk

        

[201]

(3.1.4)

Ahmad et al/COMMUNE – 2015

3.2 Posterior Density under Extension of Jeffrey’s Prior: Let y1 , y2 ,, y n is a random sample of size n having pdf (1.1) and the likelihood function (2.1.1). The Extension of Jeffrey’s for  is given by 1 (3.2.1) g ( )  2 c



Using the concept of Bayes theorem, it follows

 

 2  | y  L( y |  ) g ( )

(3.2.2)

Using (2.2.1) and (3.2.1) in (3.3.2)

 

2  | y 



nk2c

( ) (k  1)!



n

y

i 1

e

i

n

 yi

i 1

nk 2c

 2  | y  K

Thus

k 1





n

 yi i 1

e

(3.2.3)

nk  2c 1

 n   yi    i 1  K (nk  2c  1)



and

By using the value of K in (3.2.3), we have n nk  2c 1     nk2c   yi  n  i 1    e y   i     i 1   2  | y    (nk  2c  1)        







(3.2.4)

4. Bayesian Estimation by using Jeffrey’s Prior under Different Loss Functions

 

Theorem 4.1:- Assuming the loss function L p ˆ,  , the Bayesian estimator of the rate parameter  , if the shape parameter k is known, is of the form

ˆ p 

nk (nk  1)    



n

 y  i

i 1



 

Proof: - The risk function of the estimator  under the precautionary loss function L p ˆ,  is given by the formula 

 

2

(ˆ   ) R ˆ   1  | y d ˆ

 

(4.1.1)

0

Using (3.1.4) in (4.1.1), we get

  nk1  2    (ˆ   )   R ˆ   ˆ  0   

 

nk

   yi yi  e i 1  i 1  nk n



On solving (4.1.2), we get

[202]

n

   d    

(4.1.2)

Ahmad et al/COMMUNE – 2015



1 nk (nk  1) 2nk R ˆ  ˆ   2 ˆ  n    n   yi   yi       i 1   i 1 





Minimization of the risk with respect to ˆ gives us the optimal estimator

nk (nk  1)

ˆ p 

   

n

 i 1

(4.1.3)

 yi   

 

Theorem 4.2:- Assuming the loss function l A ˆ,  , the Bayesian estimator of the rate parameter  , if the shape parameter k is known, is as follows

ˆ A 

(nk  c1 )  n   yi     i 1 



 

Proof: - The risk function of the estimator  under the Al-Bayyati’s loss function LA ˆ,  is given by the formula 

 

 

R ˆ  c1 (ˆ   ) 2 1  | y d

(4.2.1)

0

On substituting (3.1.4) in (4.2.1), we have

  nk1      c1 ˆ 2 ˆ R    (   )  0   

 

nk

   yi yi  e i 1  i 1  nk n



n

   d    

(4.2.2)

Solving (4.2.2), we get



R ˆ  ˆ2

(nk  c1 )  n  nk  yi     i 1 



Minimization of the risk with respect to

ˆ A 

c1

ˆ



(nk  c1  2)  n  nk  yi     i 1 



c1  2



2ˆ(nk  c1  1) c 1

 n 1 nk  yi     i 1 



gives us the optimal estimator

(nk  c1 )  n   yi     i 1 

(4.2.3)



 

Theorem 4.3:- Assuming the loss function Ll ˆ,  , the Bayesian estimator of the rate parameter  , if the shape parameter k is known, is given as

      nk a  ˆl  log1  a   n  yi          i 1  



 

Proof: - The risk function of the estimator  under the linex loss function Ll ˆ,  is given by the formula

[203]

Ahmad et al/COMMUNE – 2015 

   exp(a(ˆ   ))  a(ˆ   ) 1  | y d

R ˆ 

(4.3.1)

1

0

Using (3.1.4) in (4.3.1), we have



  nk1     ˆ ˆ exp( a(   ))  a(   )  1     

  

R ˆ 

0



nk

   yi yi  e i 1  i 1  nk n



n

   d    

(4.3.2)

On solving (4.3.2), we get c

 n 1  y i  exp( aˆ )   a(nk  1) i 1  R ˆ    aˆ  1 nk n  n      nk yi a  yi      i  1   i 1  









Minimization of the risk with respect to ˆ gives us the optimal estimator

      nk a  ˆl  log1  a   n  yi        i  1   

(4.3.3)



5. Bayesian Estimation by using Extension Jeffrey’s Prior under different Loss Functions

 

Theorem 5.1:- Assuming the loss function L p ˆ,  , the Bayesian estimator of the rate parameter  , if the shape parameter k is known, is obtained as

(nk  2c  1)( nk  2c  2)

ˆ p 

   

n

 i 1

 yi   

 

Proof: - The risk function of the estimator  under the precautionary loss function L p ˆ,  is given by the formula 

 

2

(ˆ   ) R ˆ   2  | y d ˆ

 

(5.1.1)

0

Using (3.2.4) in (5.1.1), we have nk 2c 1 n  n   yi   nk2c  i 1  yi e 2     (ˆ   )  i 1   ˆ R   (nk  2c  1) ˆ 0   

 



   d    

Solving (5.1.2), we get



1 R ˆ  ˆ  ˆ

(nk  2c  3)  n  (nk  2c  1) yi     i 1 



2



2(nk  2c  2)  n  (nk  2c  1) yi     i 1 

[204]



(5.1.2)

Ahmad et al/COMMUNE – 2015

Minimization of the risk with respect to ˆ gives us the optimal estimator

(nk  2c  1)( nk  2c  2)

ˆ p 

(5.1.3)

 n   yi     i 1  Remark: - Replacing c=1/2 in (5.1.3), the same Bayes estimator is obtained as in (4.1.3).



 

Theorem 5.2:- Assuming the loss function l A ˆ,  , the Bayesian estimator of the rate parameter  , if the shape parameter k is known, is derived as

ˆ A 

(nk  2c  c1  1)  n   yi     i 1 



 

Proof: - The risk function of the estimator  under the Al-Bayyati’s loss function LA ˆ,  is given by the formula 

 

 

R ˆ  c1 (ˆ   ) 2 2  | y d

(5.2.1)

0

By using (3.2.4) in (5.2.1), we have nk  2c 1 n  n   yi   nk2c  i 1  yi e     i 1   c1 ˆ 2 ˆ R    (   )  (nk  2c  1) 0   



 

   d    

(5.2.2)

Solving (5.2.2), we get



R ˆ  ˆ2

(nk  2c  c1  1)  (nk  2c  1)  

n

 i 1

Minimization of the risk with respect to

ˆ A 

c

1 yi   

ˆ



(nk  2c  c1  3)  (nk  2c  1)  

n

 i 1

c 2

1 yi   



2ˆ(nk  2c  c1  2)  (nk  2c  1)  

n

 i 1

c 1

1 yi   

gives us the optimal estimator

(nk  2c  c1  1)  n   yi     i 1 

(5.2.3)



Remark: - Replacing c=1/2 in (5.2.3), the same Bayes estimator is obtained as in (4.2.3).

 

Theorem 5.3:- Assuming the loss function Ll ˆ,  , the Bayesian estimator of the rate parameter  , if the shape parameter k is known, is given by

      (nk  2c  1)  a  ˆl  log 1  a   n  yi          i 1  



 

Proof: - The risk function of the estimator  under the linex loss function Ll ˆ,  is given by the formula 

   exp(a(ˆ   ))  a(ˆ   )  1  | y d

R ˆ 

2

0

[205]

(5.3.1)

Ahmad et al/COMMUNE – 2015

Using (3.2.4) in (5.3.1), we have nk  2c 1 n  n   yi   nk2c  i 1  yi e     i 1    ˆ ˆ ˆ R   exp( a(   ))  a(   )  1  (nk  2c  1) 0    On solving (5.3.2), we get

  



(5.3.2)

nk2c 1

 n  exp( aˆ ) yi     i 1  R ˆ  nk2c 1 n   a  yi    i 1  







   d    

 aˆ 



a(nk  2c  2) 1  n  (nk  2c  1) yi     i 1 



Minimization of the risk with respect to ˆ gives us the optimal estimator

      ( nk  2 c  1 ) a  ˆl  log1  a   n  yi          i 1  

(5.3.3)



Remark: - Replacing c=1/2 in (5.3.3), the same Bayes estimator is obtained as in (4.3.3). 6. Conclusion In this paper, we have obtained Bayesian estimate of rate parameter of Erlang Distribution using Jeffrey’s and Extension of Jeffrey’s prior under symmetric and asymmetric loss functions. In addition, classical MLE is established. References Ahmad, S.P. and Kaisar Ahmad ,2013. Bayesian Analysis of Weibull Distribution Using R Software Australian Journal of Basic and Applied Sciences, 7(9): 156-164. ISSN 1991-8178. Ahmed A, Ahmad. S.P and Reshi J.A 2013. Bayesian analysis of Rayleigh distribution; International Journal of Scientific Research and Publications vol. 3, issue 10, 217-225. Bhattacharyya, S.K. and Singh, N.K., 1994. Bayesian estimation of the traffic intensity in M/Ek/1 queue Far. East, J. Math. Sci., 2, p. 57-62. Damodaran, D., Gopal, G. and Kapur, P. K. 2010. A Bayesian Erlang software reliability model. Communication in Dependability and Quality Management, 13(4), p. 82-90. Erlang, A. K. (1909). The theory of probabilities and telephone conversations. Nyt Tidsskrift for Matematik B, 20(6), 87-98. Haq, A., & Dey, S. 2001. Bayesian estimation of Erlang distribution under different prior distributions. Journal of Reliability and Statistical Studies, 4(1), 1-30. Rahul, G. P., Singh and O.P. Singh, 2009. Population project OF Kerala using Bayesian methodology .Asian J. Applied Sci., 2: 402-413. Suri, P. K, Bhushan, B., & Jolly, A. 2009. Time estimation for project management life cycles: A simulation approach. International Journal of Computer Science and Network Security, 9(5), 211-215. Zellner, A., 1986, Bayesian estimation and prediction using asymmetric loss function. Journal of American Statistical Association exponential distribution using simulation. Ph.D Thesis, Baghdad University, Iraq. 81, 446-451.

[206]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

T-X Family of Gamma-Exponential Distribution and Its Structural Properties Suriya Jabeen* and T. R. Jan Post Graduate Department of Statistics, University of Kashmir, Srinagar, India

Abstract In T-X distribution, two random variables X, the transformer, and T, the transformed are used to develop a new distribution. A new distribution namely Gamma-Exponential Distribution (GED) is proposed which is generated from family of T-X distributions. Different properties of the resulting distribution are obtained. The limiting behaviour, moments, skewness, kurtosis, Mean deviation, Hazard function, Survival function, Fisher’s information matrix are provided. Parameter estimation of the gamma-exponential distribution by the maximum likelihood method is also proposed.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: T-X family; Limiting behaviour, Hazard function, Survival function, Moments, Mean deviation, Estimation, Fisher’s information matrix

1. Introduction The events occurring in time is usually described by exponential distribution , the distribution is frequently used as life time distribution. Numerous authors have derived various generalizations of the distribution. Gupta and Kundu (1999, 2001 a, b) introduced the two-parameter generalized exponential distribution can be used in analyzing many life time data. Researchers developed and studied new and more flexible distribution. Eugene et al (2002) used the betagenerated distribution with support between 0 and 1 as the only generator. Later Alzaatreh (2013a) developed a new technique by taking the PDF of any continuous distribution as the generator.Alzaatreh et al (2013a) presented the cumulative distribution function (cdf) of the Transformed-Transformer family or T-X family as follows; Gx  

 log1 F  x 

 r t dt

1.1

0

Where r(t) is the probability density function (pdf) of random variable T defined over [0,∞) and F(X) is the cumulative distribution function of random variable X. Using (1.1), and taking X as a continuous random variable the probability density function can be written as

g x  

f x  r log 1  F x  1  F x 

1.2

g x   hx r H x 

Equation (1.2) shows that the T-X family of distributions has associated with the hazard functions and each generated distribution can be considered as a weighted hazard function of the random variable X. If

T ~   ,   , then

* Corresponding author. Tel.: +91 9906 898018. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Jabeen and Jan/COMMUNE – 2015

t

 1 r t    t  1e    

,t  0

Using (1.2), the pdf of gamma-X family is



g x     

 f x log1  F x 1  F x 1

 1

1



1

1.3

Parameters estimation of gamma-Pareto distribution was considered by Alzaatreh (2012). Alzaatreh (2013c) studied various structural properties of gamma X family specialised their results on gamma-normal distribution. The problem of estimation of paremeters of weibull-Pareto distribution by the method of modified maximum likelihood was considered by Alzaatreh (2013b) . In this paper we defined the Gamma-Exponential distribution (GED) and its properties as moments, moment generating function, limiting behaviour, skewness, kurtosis, Fisher ’s information matrix are discussed.

2. The Gamma-Exponential Distribution (GED)

If X is a exponential random variable with density function

f x   e x

, x 

F x  1  e x

and cdf

g x  

1   1 x    x e   



Putting

g x  

then (1.3) results in

; x   ,  ,  ,  0

2.1

then (2.1) becomes   c,

1

 1

x e

 c

x c

2.2

; x   ,  ,  ,  0

g(x) follows the gamma-exponential distribution with parameters α and c. If α=1 in (2.2), the gammaexponential distribution leads to the exponential distribution with parameters 1/c and if c=1 it leads to the gamma distribution with parameter α. The cdf of the gamma-exponential distribution results as

Gx  

1

 

 

x c

2.3

  ,  t

Where

  , t    u  1e u du

is the incomplete gamma function.

0

The survival function of the GED is given as;

R x   1  G  x   1 

  ,x   

2.4

and the hazard function of the GED can be obtained as

[208]

Jabeen and Jan/COMMUNE – 2015

1

 1



x e g  x   c h x      ,x  R x  1   h x  



 1

x c

;x 

x c

2.5

x e  c      ,x 

3. Properties of the Gamma-Exponential Distribution

Gamma-exponential distribution has some relations with other distributions. Using the transformation technique these relations can be obtained. Let Y be a random variable with

parameters (α,c), then using the transformation

X=θe probability density function of gamma-exponential distribution is obtained. The limiting behaviours of the gamma-exponential PDF is given in the following theorem. Theorem.( a): The limit of the gamma-exponential density function is zero when x goes to infinity, and when the limit is given by: Y

 1

0, 1  lim g x   lim hx    , x0 x0 c , Proof.

lim g x   x 

1

 c 

x 0,

3.1

 1  1 lim

x  1

x 

e

x c

Applying L,Hopital,s rule, it reduces to

lim g x   lim x

x

  1!  0 1

c 1

e

3.2

x c

If α>1, then, (3.2) goes to zero, if α<1, it goes to infinity and if α=1, it reduces to 1/c. Theorem. (b): The limit of the hazard function for the gamma-exponential distribution when x goes to infinity is given by:

0, 1  lim hx    , x c ,

  1; 3.3

  1;   1.

Proof: we have

g x  x Rx 

lim hx   lim

x

g x  x  1  G  x 

 lim

Since

lim g x   0 , it can be shown that lim hx   0 by using L,Hopital,s rule. x

x

4. Moments of GED The sth non-central moments of GED are given by

[209]

Jabeen and Jan/COMMUNE – 2015 

 

x

 1 x s 1e c dx   c   0

E Xs 

4.1

x  t , (4.1) reduces to c

Letting

c s s     

E X s  

 

4.2

E X    c is the mean of the gamma-exponential distribution. The central moments shown by Alzaatreh (2012), expresses that for any random variable, it can be shown as s 4.3 s s k s

 

E  X        1  sk E X k k 0  k 

Using (4.2) & (4.3), the central moments for the gamma-exponential random variable X can be obtained as; s 4.4 k    s s k s

E  X     c s    1  sk k 0  k 

 

Using (4.4), the variance of GED is given as 2 2 k     2  c 2    12k  2k   k 0  k 

 2  c2

4.5

  3   2  3  2 3    

1 

 1 

4    1

4.6

2

1

2

2  2 

3 2

  4   3   2  4  6 2  3 4      

2

  1  2  3  4  1  2  6 2   1  3 3

4.7



The above equations (4.6) & (4.7), are the skewness and kurtosis of the gamma-exponential distribution. 5. Mean Deviations of GED By the definition 

5.1

D   2G   2 xg x dx 0 M

DM     2  xg x dx

5.2

0

Now the integral

[210]

Jabeen and Jan/COMMUNE – 2015

m

 xg xdx  o

m

x

5.3

 1 x e c dx   c   0

on simplification the (5.3), becomes



m



m

 xg xdx      1, c 

5.4

0

Using (2.3) and (5.4) , the mean deviation from the mean and the mean deviation from the median for the gammaexponential distribution can be written as

      2    ,       1,  c    c  D    

 2  M  DM    1     1,  c      6. Moment Generating Function of GED

M x t   E e



tx

   e g x dx tx

0



 e 0

Let

tx

1

 1



6.1

x c

x e dx  c

dz 1  x  t   z , then dx  1  c    t c 

The integral in (6.1) can be written as



M x t  



c 1     t  c 



z

 1  z

e dz

0

 z 



1   c   t  c 

6.2



Equation (6.2), is moment generating function for gamma-exponential distribution. 7. Parameter Estimation for GED The log-likelihood function of (2.2) is given by

  xi   1  n n  1  i 1  log L , c   log  x e c     i   c  i 1    n

[211]

Jabeen and Jan/COMMUNE – 2015

n

n

x

i 1

c

log L , c   n log    n log c    1 log xi 

i 1

i

7.1

Differentiating (7.1), with respect to α and c gives n

 n    n log c   xi i 1

7.2

n

xi n  i 1   2 c c

7.3

Setting (7.2) and (7.3) to zero, we obtain the following MLE of

ˆ

and

cˆ 7.4

n

 ˆ    log cˆ  cˆ 

 xi i 1

n 7.5

x ˆ

8. Fisher’s Information Matrix of Gamma-Exponential Distribution Appling log on b/s in (2.2), we have

log g  , c    log c  log      1log x 

x c

8.1

Differentiating (8.1), with respect to α and c, we get

 log g  , c    log c      log x  where        

2 2 log g  , c          2 

 1  c c   x c



c



c2

 1  c c

2  2x  2 3 2 c c c Taking expectations on both sides of the equations, we get 2  2  2 I 1,1   E  2 log g  , c          I 1,2   E   log g  , c   1     c  c

 2  1 I 1,2   E  log g  , c    c  c [212]

Jabeen and Jan/COMMUNE – 2015

 2   I 2,2   E  2 log g  , c   2  c  c Now, the Fisher’s information matrix of gamma-exponential distribution is given by

  2   E  2 log g  , c     I  , c      2   E  log g  , c     c 1  2        c I  , c    1    c c2

 2   E log g  , c    c  2     E  2 log g  , c    c       

8.2

References Alzaatreh A., Carl Lee & Felix F. (2013a). “A new method for generating families of continuous distributions”. METRON. 71. 63-79. Alzaatreh A, Famoye F, Lee C. (2013b) Weibull-Pareto distribution and its applications. Commun stat Theory Methods 42:1-19. Alzaatreh A, Famoye F, Lee, C. (2012). Gamma-Pareto distribution and its applications. J Mod Appl Stat Methods, 11:78-94. Alzaatreh A, Famoye F, Lee, C. (2013c).The gamma-normal distribution: Properties and applications. Computational Statistics and Data Analysis, 5665,1-13. Gupta, R.D., Kundu, D., (1999). Generalized exponential distribution. Australian and New Zealand Journal of Statistics 41,173-188. Gupta, R.D., Kundu, D., (2001a). Exponentiated exponential distribution: an alternative to gamma and weibull distributions. Biom. J. 43, 117-130. Gupta, R.D., Kundu, D, (2001b).Generalized exponential distributions: different methods of estimation. J. Stat. Comput. Simul. 69, 315-338.

[213]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Impact of Body Thickness on the Performance of InAs Gate-AllAround Nanowire Field Effect Transistor Richa Gupta, Deepika Jamwal and Rakesh Vaid* University of Jammu, Department of Physics and Electronics, Jammu, India

Abstract As the classical MOSFET is reaching its scaling limits, alternative devices are required to be investigated. Among the various categories of SOI devices proposed, GAA nanowire transistor is one of the novel nanoelectronic devices that can significantly overcome these MOSFET limitations. Infact, nanowire transistors with high mobility III-V compound semiconductors are drawing interest for high speed and low power electronics. In the present work, the performance limits of GAA InAs NWFET have been explored by considering the substrate body thickness. The device metrics considered at nanometer scale are the drive current, leakage current, transconductance, switching speed, charge density and Energy band profiles. It has been revealed that InAs-NWFET exhibits high current driving capability, excellent charge density, higher transconductance and better gain. Thus, these devices are suitable to be used for future high performance and low power nanoelectronic applications. Keywords: Gate All Around (GAA); InAs; Nanowire FET (NWFET); transconductance and short channel effects (SCEs).

1.

Introduction

Conventional CMOS technology is facing scaling challenges owing to reduced gate control over the channel, high leakage currents, increased short-channel effects (SCE’s). Therefore, the development of new designs like Gate-AllAround (GAA) nanowire transistors have made it possible for the Moore’s law to be satisfied even in the next decade. Si NWFETs were the first to be tested and fabricated. However, with the advent of new nanowire fabrication techniques, newer and better materials are drawing interest within the semiconductor world (Gupta et al, 2014). It has been reported that GAA Nanowire transistor provides excellent electrostatic control over the channel surrounded by the conducting gate. Simulation analysis has revealed that GAA configuration exhibits an excellent performance due to considerable effects of short channel relative to the other structures (Sharma, Akashe, 2014). Quasi-one dimensional semiconducting structures are expected to reduce the short-channel effects and drain induced barrier lowering (DIBL) and Sub-threshold slope ~ 60mV/decade. However, due to lack of doped source and drain contacts in these devices their performance deteriorates. Gate all-around (GAA) MOSFETs have drawn considerable attention as compared with double-gate and tri-gate (Sharma, Akashe, 2013). Use of III–V compound semiconductors in this device structure really opens the opportunity for the future CMOS applications. Among various III–V semiconductor materials, Indium arsenide (InAs) is considered one of the most promising materials for n-channel enhancement MOSFETs because of its high electron mobility and unique band alignment (Conrad et al, 2013). Basically, the tomorrow’s need of the semiconductor industry would be achievable only due to the advent of unconventional materials and non-planar structures. Both of these aspects are innate in GAA InAs Nanowire transistors making them a promising platform for the future high-performance transistors (Dayeh et al, 2007; Thelander, et al, 2008; Thelander et al, 2008; Bessire, 2011). Moreover, one of the critical goal in scaling is to obtain ballistic behavior of the device where carriers can transport through the channel without undergoing any scattering events. Ballistic devices are really attractive as they possess minimal resistive voltage drop in the channel (Chuang, 2013). In this regard, we have performed the detailed characterization of InAs NWFETs by taking into consideration the body thickness parameter.

*

Corresponding author. Tel.:+91 9419 106794. E-mail address: [email protected] ISBN: 978-93-82288-63-3

Gupta et al/ COMMUNE-2015

2.

Device Structure

Gate-All-Around Nanowire transistor is surrounded by several gates, which results in more effective suppression of the leakage cuurent. In the “on-state”, these devices allow enhanced current per area. These advantages lead to lower power consumption and enhanced device performance (Tocci, 2010). The GAA Silicon nanowire FET studied here is illustrated in Fig. 1 and the parameters used in simulation are shown in Table 1.

Fig. 1: Device structure of GAA InAs Nanowire transistor

In this structure, the nanowire is used as a channel, which is surrounded by an oxide layer and is finally surrounded by a metal contact. This metal contact serves as the gate terminal. Hence, the current flows through the nanowire or is pinched off under the control of the voltage on the gate-electrode, which surrounds the nanowire (Gupta et al, 2014). To examine the ballistic transport properties of GAA NWFET, the simulations in the paper has been achieved through the Multi-gate Nanowire FET tool, which is basically a numerical simulator. The simulator features include: Effective mass theory, Uncoupled mode space non-equilibrium Green’s function (NEGF), Poisson-transport self-consistent calculation and quantum ballistic transport (Shin, 2007). Quantum mechanical approach has been used for GAA nanowire transistor based on non-equilibrium Green’s function, Schrodinger’s equation and transport equation. Table 1: Parameters used in simulation Parameters Source/Drain length Gate length Oxide width Channel length

Symbol Lsd Lg Wox Lc Tbox

Buried Oxide thickness

3.

Value (nm) 10 10 1 10 1

Results and Discussion

3.1.

Transfer Characteristics

Fig. 2 shows the variation of InAs body thickness on the transfer characteristics of InAs-NWFET. Increase in the body thickness directly leads to enhanced on-state current. This is because of the fact that the numbers of charge carriers are directly related to the thickness of the substrate.

Fig. 2: Variation of Drain current and Gate voltage for various body thicknesses

[215]

Gupta et al/ COMMUNE-2015

Simulation studies have been performed by keeping channel length at 10nm and insulator thickness at 1nm. As the substrate becomes thicker, the availability of number of modes for the electrons to populate in the channel undergoes a significant increase. Moreover, the electrons have more volume to travel in, leading to the reduction of the surface scattering effects that the nanowires otherwise possess as they have very large surface to volume ratio. 3.2.

Energy band Profile and Charge density profile

Fig. 3 (a) shows the variation of Energy with respect to the position along the length of the channel at oxide thickness = 1nm. From the graph, it is evident that 40nm body thickness requires lower energy relative to the other values.

Fig. 3: (a) Conduction band profile (b) Mid-Channel Charge density profile for various body thicknesses

From the Fig. 3(b) it is clear that 40nm substrate thickness has the highest charge density in comparison to all the thicknesses being studied. Moreover, the charge density exactly lies between 10-20nm, which states that the maximum conduction in the device occurs at this range. 3.3.

Variation of Ion and Ioff versus InAs Body thickness

Fig. 4(a) and 4(b) shows the variation of off-current and on-current with respect to body thickness InAs NWFET.

Fig. 4: Variation of (a) Leakage Current (Ioff ) and (b) Drive Current (Ion) for various body thicknesses

A maximum drive current of 1013μA was obtained at a body thickness of 40nm; while drive current of 120μA was obtained at a body thickness of 5nm. Therefore, there occurs a noteworthy increase in on-current as the body thickness is increased. Moreover, the device has mo re control over the gate at 40nm body thickness. The leakage current also raises on ever-increasing the channel body thickness. This can be explained on the basis of degradation of the control of gate over the channel. As discussed above, as the body thickness is increased, the channel volume increases while the oxide thickness is kept constant at 1nm. As a result, the reduction in the gate capacitance per is experienced. This reduction in gate control makes the Drain Induced Barrier Lowering (DIBL) more prominent as the InAs body thickness is increased. Additionally, as the channel length is reduced to 10nm level, gate control over the channel is reduced and

[216]

Gupta et al/ COMMUNE-2015

there is an increase in the source-drain tunneling of electrons. The leakage current is 215 x 102 pA at 5nm channel body thickness. However, it increases by a factor of 775 x 105 pA at 40nm, which corresponds to leakage current that is 10 3 times higher. Such a large increase in leakage current is quite undesirable and is prone to degrade the device performance. 3.4.

Variation of gm and Ion/Ioff versus InAs Body Thickness

Fig. 5(a) shows the variation of the switching speed of the device w.r.t. the InAs body thickness. From the Fig. 5(a), it is quite clear that InAs NWFET shows higher ON/OFF ratios at the lower body thickness in comparison to the higher values of body thicknesses.

Fig. 5: Variation of (a) Switching speed Ion/Ioff and (b) Transconductance (gm) for various body thicknesses

The primary reason for the lower Ion/Ioff values at higher TInAs is mainly due to the degraded sub-threshold characteristics namely the higher Ioff values and lower sub-threshold slopes for the InAs NWFET at higher body thicknesses. Figure 5(b) shows the variation of transconductance w.r.t. different values of body thicknesses. A transconductance value of 120µS at 5nm has been obtained in comparison to a transconductance value of 1013 µS at 40nm. This indicates that the device transconductance has a reasonably strong dependence on the body thickness. 4.

Conclusion

We have analyzed the effect of InAs Body Thickness on the electrical characteristics of GAA InAs NWFET in detail. It has been shown that GAA InAs nanowire FET has excellent current–driving capability and hence better transconductance at body thickness equals to 40nm. Charge density profiles and Energy band profiles are found to possess excellent characteristics at 40nm. The main problem being encountered was basically the higher leakage current problem at 40nm relative to 5nm InAs body thickness due to the degradation of sub-threshold slope. Thus, it is important here to reveal that the performance limits of the device parameters should be kept in mind for achieving excellence in fabrication.

References Gupta, R., Dass, D., Prasher, R., and Vaid, R., 2014.“Study of gate all around InAs/Si based nanowire FETs using simulation approach”, International Conference on Signal Propagation and Computer Technology, Ajmer, p. 557. Sharma, A., and Akashe, S., 2014.“Analyze the tunneling effect on Gate-All-Around Field transistor”, International Journal of Advanced Science and Technology 63, p. 9. Sharma, A., and Akashe, S., 2013.“Performance Analysis of Gate-All-Around Field Effect Transistor for CMOS Nanoscale Devices”, International Journal of Computer Applications 84, p. 44. Conrad et al., 2013.“Performance and Variability Studies of InGaAs Gate-all-Around Nanowire MOSFETs”, IEEE Transactions on Device and Materials Reliability 13, p. 489. Dayeh et al., 2007.“High Electron Mobility InAs Nanowire Field-Effect Transistors”, Small 3, p. 326. Thelander, C., et al., 2008.“Vertical Enhancement-Mode InAs Nanowire Field Effect Transistor with 50-nm Wrap Gate”, IEEE Electron Device Letters 29, p. 206. Thelander et al., 2008.“Development of a Vertical Wrap-Gated InAs FET”, IEEE Transactions on Electron Devices 55, p. 3030. Bessire, C. D., 2011.“Trap-Assisted Tunneling in Si-InAs Nanowire Heterojunction Tunnel Diodes”, Nano Letters 11, p. 4159. Chuang, S., 2013.“Ballistic InAs Nanowire Transistors”, Nanoletters 13, p. 555. Tocci, G., 2010.“Performance estimation and Variability from Random Dopant Fluctuations in Multi-Gate Field Effect Transistors: A Simulation Study”, M.S. Thesis, Sweden. Shin, M., 2007.“Efficient simulation of silicon nanowire field effect transistors and their scaling behavior”, Journal of Applied Physics 101, p. 024510.

[217]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

FGMOS based Log Domain Integrator Harjeet Kaur*, Rockey Gupta, Susheel Sharma Department. of Physics and Electronics, University of Jammu, Jammu, India

Abstract This paper presents a FGMOS based low-voltage log domain integrator. The proposed integrator is implemented with the MOS transistors working in weak inversion region. Therefore, we have discussed the sub-threshold behavior of MOSFET where it functions as trans-linear device and exhibits non-linear characteristics with low currents. The integrator building blocks such as compressor, non-linear integrator, and expander are also described with their mathematical analysis. The magnitude and phase response of integrator is also presented. The cut-off frequency of integrator has been found to be 292.2 kHz that dissipates power of 1.23E-04 watts. The behavior of these circuits has been verified through PSpice simulations using level 7 parameters in 0.13 μm technology with supply voltage of 1V.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Compressor; Expander; Trans-Linear loop; Differential Integrator; Log Domain Filter

1. Introduction Analog filter design based on log-domain technique is suitable for at low supply voltages operation with low power dissipation. In this technique, low-level, non-linear device characteristics are exploited to implement systems, which possess linear relationship between signals at input and output, but internal processing of signals occurs in non-linear mode. The internal variables may be chosen to be logarithmically and exponentially related so that the overall operation becomes linear. Thus, a filter based on nonlinear internal processing of signals involving logarithmic and exponential functions is referred to as log-domain filter. This technique employs the naturally occurring exponential relationship between the collector current and base emitter voltage of bipolar junction transistor (BJT) or drain current and gate voltage of a MOSFET operating in weak inversion region. Therefore, log domain filtering is considered as a method of designing analog continuous-time filters with a linear transfer function between input and output terminal variables whereas the signals at all nodes inside the circuit are assumed non-linear. That is, log domain filters are internally nonlinear and externally linear filters. The principle of operation of log domain circuits is based on instantaneous companding technique where the signals in current form with large dynamic range are compressed logarithmically during transformation into voltages and later expanded exponentially when they are converted back into current form as indicated by Masry et al., 2000. Further, log domain technique is well suited for increasing the dynamic range besides enabling operation of circuits at very low supply voltages as indicated by Frey, 1993 and Yang, et al. 1996. The application of FGMOS in log-domain filters results in simplification of trans-linear loop besides reduced number of internal nodes as indicated by Minch B.A. 2001. Since bulk terminal is not used in trans-linear equation, therefore, circuit instability can be avoided and programming of filter parameters can be incorporated. 2. Floating-Gate MOSFET The structure of floating-gate MOSFET (FGMOS) is similar to a conventional MOSFET expect that its multiple input gates are capacitively connected to the conventional gate which becomes floating being embedded in insulator. The conventional gate (floating-gate) is not accessible anymore for signal application rather it is controlled indirectly

* Corresponding author. Tel.: +919419383815 E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Kaur et al/COMMUNE – 2015

through capacitance coupling by applying signals at control gates. The structure of multi-input n-channel floating-gate MOSFET is shown in Fig. 1. V1 V2

Floating gate

VN

Drain

Source n+

n+ P-substrate

Fig. 1 Structure of n-inputs FGMOS

2.1

Characteristics of n-channel FGMOS

The circuit for two input n-channel FGMOS for obtaining the drain and transfer characteristics is shown in Fig. 2. The input voltage (Vin) is applied at gate through C1 and the drain current (IDS) is obtained from the drain of the MOSFET (M1). The bias voltage (Vbias) is applied through C2 which provides tunability to the conventional threshold voltage (VT) of M1. Therefore, VT for the MOSFET adjusts itself to a new value VT,eff which is expressed as:

VT ,eff 

VT  Vbias k 2 k1

where k1 

(1)

c1 and 𝑘 = 𝐶2 , C and C are the capacitances between floating-gate and control gates, C is the sum of 1 2 T 2 𝐶𝑇 cT

capacitances between control gates and floating-gate, capacitances between floating-gate and drain, capacitances between floating-gate and source, and capacitances between floating-gate and bulk . VDD ID

C1 Vin

M1 Vbias C2

Fig. 2. Two-input n-channel FGMOS

From Eq. (1) we find that VT,eff will be less than VT if we select Vbias, k1 and k2 properly. Thus, we have been able to get a MOSFET where VT,eff is lower than normal VT. The drain characteristics of n-channel FGMOS is shown in Fig.3, which resembles with that of the conventional MOSFET. 100

50

80

Drain Current (μA)

Drain current, Ids (μA)

90 Vin=.4V

70

Vin=.6V

60 50

Vin=.8V

40

Vin=1V

30 20

Vbias=0V

40

Vbias=.1V 30

Vbias=.2V Vbias=.3V

20

Vbias=.4V Vbias=.5V

10

10 0

0 0

0.2

0.4

0.6

Drain to source voltage,Vds (V )

0.8

Fig3. Drain characteristics of n-channel floating-gate MOSFET

1

0

0.2

0.4

0.6

0.8

1

Input Voltage (Volts)

Fig. 4. Transfer characteristics of n-channel floating-gate MOSFET

The transfer characteristics of n-channel FGMOS are shown in Fig.4. These are the curves between drain current (IDS) and input voltage (Vin) at different values of bias voltage (Vbias). From Fig. 4, it has been observed that as we go

[219]

Kaur et al/COMMUNE – 2015

on increasing Vbias from 0 V to 0.5 V, effective threshold voltage of n-channel FGMOS goes on decreasing from 0.9 V to 0.4 V. 3. Log Domain FGMOS based integrator The block diagram of Log domain integrator is shown in Fig.5, which includes compressor, non-linear integrator and expander. Expander

Iin

Compressor

VX NLB

IX=Iout

Vin C VBin

VB

Fig .5 Block Diagram of Log Domain Integrator

In Log domain filter, input signal is compressed by compressor, processed by non-linear integrator and then expanded at the output to restore its original linearity. The resulting small internal voltage swings lead to higher operating frequencies, lower distortion, and lower internal interference in the circuit. The general state space equation of a linear integrator can be written as 𝑥̇ = 𝛼𝑥 + 𝜂𝑥𝑖𝑛 (2) where 𝛼 and 𝜂 are constants while 𝑥 and 𝑥𝑖𝑛 are state variable and input, which are expressed with two new variables y and u through exponential relationship as given below

x  k y e y & xin  k in e

u

(3)

The linear expression, eqn. (2) is converted into non linear function of u and y

𝑦̇ = 𝛼𝑘𝑦 𝑒 𝑦 + 𝜂𝑘𝑖𝑛 𝑒 𝑢 𝑦̇ = 𝛼 + 𝐾𝑒 𝑢−𝑦 𝜂𝑘 𝐾 = 𝑘 𝑖𝑛

(4)

𝑦

Where, y is proportional to the voltage difference between its terminals. 3.1

Compressor circuit analysis

Compressor circuit is shown in Fig.6 which converts input variable current into voltage that is related to it in a I logarithmic way as shown in eqn. (6). in

Vin

M1

Fig.6.Compressor circuit

This voltage can be generated by a MOSFET with one of its inputs connected to its drain. Because of the aforementioned compression function, this device will be referred to as compressor. Drain current is VGS

I D  I Do e

nUT

 v DS

(1  e UT )

(5)

where n = 1/κ =1+ C dep / Cox is the sub-threshold slope factor, I DO

W 2   O COX VT is a parameter which depends on L

source-to-bulk voltage and threshold voltage. Taking log on both sides of eqn.(5), we get

I  Vgs  nUT ln  D   I DO 

(6)

[220]

Kaur et al/COMMUNE – 2015

3.2

Expander circuit analysis

Expander circuit generates the state variable, which in this case is the current, IX .The expander circuit is implemented by a FGMOS where constant voltage, VB and capacitor are connected to the two inputs of FGMOS as shown in Fig.7. Ix Vx VB

M1

C

Fig.7 Compressor circuit

Now the circuit diagram of integrator as indicated by Villegas, 2006 is shown in Fig. 8(a). In Fig.8(a) FGMOS is used as compressor while in Fig. 8 (b) conventional MOSFET is used as compressor. FGMOS based integrator occupies more chip area ( aspect ratio : 91/.13 µm) as compared to conventional MOS based integrator (aspect ratio : 10.4/.13 µm). Comparison of FGMOS based Integrator with different compressor blocks is shown in Table 1. The second block performs a nonlinear integration whose output is the ratio between the state and input current. The output signal is finally obtained by taking the voltage at the integration node and processing it through the expander. This block is implemented by sourcing instead of sinking current to the integrating capacitance. This is done just by using a conventional current mirror to copy and invert the sign of the current IB delivered by M3 and connecting its output together with an inverted version to the integrating capacitance as indicated by Villegas et al., 2003.or Fox et al.,1999.or Villegas et al., 2001.or Villegas et al., 2004.

Iout

VDD VDD

VDD

IA

Iin

VBin1

VA

VBin

IB

VDD IA

Iin

VA VBin

M4 M3

M3 M1

Iout

VDD

IB

VDD

M1

C

M2

C

M2

Fig.8(a and b) Circuit diagram of integrator 1

0

0

-10

-2

Phase, degrees

Current gain, dB

-1

Vbias=0V

-3 -4

Vbias=0.2V

-5

Vbias=0.4V

-6 -7

Vbias=0.6V

-8

Vbias=0.8V

-9

10

Vbias=0V

-30

Vbias=1V

-40 -50 -60 -70 -80

Vbias=1V

-10

-20

-90 100

0.01

1000

Frequency, kHz

0.1

Frequency,MHz

1

Fig.9 Magnitude response of log-domain integrator

Fig.10 Phase response of log-domain integrator

Table 1 Comparison with different compressor circuits

Table 2 Effect of Vbias on bandwidth

Parameters M1 M2 M3 M4 f-3dB SupplyVoltage

Conv.MOSFET 10.4 μm .78/.13 μm .78/.13 μm .78/.13 μm 209.25kHz 1V

FGMOS 91/.13μm .78/.13 μm .78/.13 μm .78/.13 μm 296.89kHz 1V

Vbias 0V 0.2 V 0.4 V 0.6 V 0.8 V 1V

[221]

Io 9.81 μA 10.36 μA 10.79 μA 11.17 μA 11.47 μA 11.72 μA

f-3dB 265 kHz 292.2 kHz 321.2 kHz 336.8 kHz 353.13kHz 364.4 kHz

10

M4

Kaur et al/COMMUNE – 2015

The log-domain integrator shown in Fig. 8(b) has been simulated for level 7 PSpice parameters for 0.13 µm technology with supply voltage 1V and Iin current is 10µA. Capacitor chosen is of value 0.2 pF and bias current is of 1µA. The magnitude response of log domain integrator is shown in Fig. 9 The phase response of log domain integrator degrades from 0 degree to -90 degree as shown in Fig.10.The simulation results shows that pass band gain is 0dB and 3dB cut-off frequency is 292.2 kHz and dissipates dc power of about 2.82E-04 Watts. It has been observed that as we go on increasing Vbias from 0V to 1V, bandwidth goes on increasing from 265 kHz to 364.4 kHz as shown in Table2 and phase also varies at different bias voltages as shown in Fig.10 5. Conclusions A FGMOS based integrator with different compressor blocks for low voltage applications has been presented. Conventional MOSFET and FGMOS based compressor are used in integrator. The proposed integrator circuit employs the weak inversion region of the transistors with low supply voltage and offers improved frequency response for low current values. Mathematical analysis of FGMOS, its drain and transfer characteristics are also presented. Different building blocks with their mathematical analysis have also been explained. Comparison of conventional MOSFET and FGMOS based compressor in integrator is also presented. The frequency and phase response of log-domain integrator has also been presented using PSpice parameters 0.13μm technology. The simulation results shows that as we vary bias voltage of FGMOS from 0V to 1V bandwidth also varies from 265 kHz to 364.4 kHz and dissipates dc power of about 2.82E-04 Watts.

References Masry,E.I.E.,Wu, J.,2000. Low voltage micropower log domain filters, Analog Integrated Circuits Signal Process. 22, p. 209 Yan, S. & Sanchez-Sinencio, E., 2000. Low Voltage Analog Circuit Design Techniques: A Tutorial, IEICE Trans. Fundamentals vol.E83-A. Minch, B.A., 2001. Multiple input translinear element log domain filters, IEEETrans.CircuitsSystem..48, p.29 Yang, F., Enz, C. and Ruymbeke, G. Van., 1996. Design of low-power and low voltage log-domain filters, Proc. ISCAS, vol. 1, pp. 117-120, Atlanta. Frey, D. R., 1993. Log domain filtering: an approach to current mode filtering Proc. Inst. Elec. Eng. pt. G, vol. 140, no. 6, pp. 406-416. Villegas, R. E., & Barnes, H., 2003. Solution to trapped charge in FGMOS transistors. Electronics Letters, 39, p. 1416. Fox R. M. and Nagarajan, M., 1999. Multiple operating points in a CMOS log-domain filter, IEEE Trans. Circuits Syst. II 46, p. 705. Villegas, E.R., Rueda,A. and A. Yúfera, 2001. A 1.25-V FGMOS filter using translinear circuits in Proc. IEEE Int. Symp. Circuits and Systems, p. 61. Villegas,E.R., Yúfera,A. and Rueda, A., 2004. A 1-V Micropower Log-Domain Integrator Based on FGMOS Transistors Operating in Weak Inversion, IEEE journal of solid-state circuits 39. Villegas, E. R, 2006. Low Power and Low Voltage Circuit Design with the FGMOS Transistor, IET Circuits, Devices, and Systems Series 20.

[222]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Optimization of Thulium Doped Fiber Amplifier for S-Band Rajandeep Singh*, M. L. Singh Department of Electronics Technology, Guru Nanak Dev University, Amritsar

Abstract S-Band is the candidate for the bandwidth enhancement outside C-band and Thulium doped fiber amplifier (TDFA) is an enabling technology for the S-band WDM systems. In this paper, optimization of 1064nm pumped Thulium doped fiber amplifier has been discussed. Performance of Thulium doped fiber amplifier has been explained in terms of gain and noise figure with respect to other parameters, which are signal power, pump power, TDFA length, and signal wavelength.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: TDFA; WDM; S-Band

1. Introduction The S-band is very promising band for future WDM systems, to increase the bandwidth of conventional C–Band WDM systems. The other alternatives to S-band are O, E and U band. But S-band has an edge over other bands as in O and E bands doped fiber amplifiers (Praseodymium-doped glass fiber amplifier) has gain regions but it yields very low gain (Whitley, 1994). In addition, the total accumulated dispersion in S-band system can be minimized with the new technique as presented in our previous work (Singh et al, 2014). O and E-Band for these bands Raman amplifier is also not as attractive because the Raman pump wavelength lies in high attenuation region. Er3+-doped fiber amplifiers (EDFA) are being very actively introduced in 1.5-pm band transmission systems (Sun et al, 1998). The conventional EDFA and gain shifted EDFA amplifiers provide gain in wavelengths ranging from 1530nm to 1610nm (Kani et al, 1999). .Thulium doped fiber amplifier is main amplifier in S-band WDM system, Thulium doped fiber can be pumped by many pumping configurations one of them is pumping with high power 1064 nm pump (Watekar et al, 2007) 2. Simulation Setup In this paper Optisystem 13.0 simulator has been used for simulations, a continuous wave laser at 1460nm is used as the transmitter and a pump laser with 1064nm wavelength is used. A pump couples is used to couple the transmitted signal and the pump signal and the composite signal is then fed to the Thulium doped fiber amplifier component. The pumping scheme used here is co directional pumping so the counter directional input port is grounded. To visualize the gain and noise figure of the amplifier a dual port WDM analyzer is used.

CW Laser Pump Laser

Dual port WDM analyzer Coupler

TDFA

Fig..1 Simulation Setup

* Corresponding author. Tel.: +91 9417 143244. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Singh and Singh/COMMUNE – 2015

Simulations have been carried out 4 times. In first simulation length of TDFA is varied from 1m to 14m as shown in table.1, other parameters have been fixed. In second simulation, the transmitted signal power has been varied from 0dBm to -30 dBm in steps of -10 dBm. In the third simulation pump, power has been varied from 100mW to 1000 mW in steps of 100mW. In forth simulation the transmitted signal wavelength has been varied from 1460 nm to 1530 nm (entire S-band) in steps of 10 nm. For each simulation, the parameters have been listed in table 1 below. For each simulation gain and noise figure have been plotted for varied parameter. Table.1: Simulation parameters

Parameter

Simulation 1

Simulation 2

Simulation 3

Signal Wavelength

1460nm

1460nm

1460nm

Signal power Pump wavelength

-20dBm 1064nm

0dBm to -30dBm 1064nm

Pump power

1000mW

1000mW

-20dBm 1064nm 100mW to 1000mW

TDFA Length

1m,4m,6m,8m,10m, 12m,14m

6m

6m

Simulation 4 1460nm to 1530nm -20dBm 1064nm 1000mW 6m

3. Results For the first simulation, the gain vs varied TDFA length and Noise figure VS. TDFA length for fixed other parameters as per table 1, the plots are as shown in figure 2. From the figure 2 it is clear that the gain increases with the TDFA length almost linearly as length is increased from 1m to 6m. At TDFA length 6m gain is maximum (8.8dB) and at TDFA length 1 m the gain is minimum (< 3dB).

Fig.2 Gain vs TDFA length

Fig.3 Noise figure vs. Thulium doped fiber

In figure 3 noise figure vs. TDFA length is plotted, noise figure firstly increases from 1.6 dB to 2.92 dB, when TDFA length is increased from 8m from 14m then the noise figure becomes almost constant. Case .2 In second case the Thulium doped fiber amplifier length is fixed at 6m and other parameters have been varied as given in table 2, from figure 4 it is clear that gain is not dependent on signal power, this is an important finding.

Fig. 4: Gain vs. signal power

Fig. 5: NF vs. signal power

Figure 5 shows the noise figure dependence on input signal power, noise figure increases slightly with increase in input signal power level. The variation is more above -10 dB input power. In figure 6 gain variation has been plotted for varied pump power from 100 mW to 1000 mW, The gain increases linearly with increase in pump power ,pump power upto 200 mW do not yield any gain , maximum gain (near 8 dB) has been observed at 1000 mW pump power. At pump powers more than 1000mW the gain could be more but the life and reliability of fiber decreases for very high pump powers, ,so in this study maximum pump power is limited to 1000mW.

[224]

Singh and Singh/COMMUNE – 2015

Fig. 6: Gain vs. Pump power

Fig. 7: Noise figure vs. pump power

Figure 7 shows the variation of noise figure with the varied pump power, noise figure increases when pump power is increased from 100mW to 300 mW ,then the noise figure decreases sharply as pump power is increased. In last 4 th case the gain and noise figure variation for the entire S-band i.e 1460nm to 1530nm has been calculated. In the figure 8 gain variation with respect to signal wavelength for optimized parameters has been plotted, the maximum gain has been achieved at 1470nm (~9 dB), as the wavelength increases the gain decreases almost linearly, Above 1510 nm the TDFA shows –ve gain that is its attenuating the signal.

Fig. 8: Gain vs. Wavelength

Fig. 9: Noise figure vs. Wavelength

In figure 9 noise figure for varied wavelength has been plotted the noise figure is below 5dB level for wavelengths ranging from 1460-1510 nm, above 1510 nm the noise figure increases sharply and becomes unacceptable. 4. Conclusion In this paper optimization of 1064nm pumped Thulium doped fiber amplifier for S-band has been discussed. The results have been analyzed in terms of gain and noise figure for varied signal power, pump power, TDFA length, and signal wavelength, It has been observed that the maximum gain is achieved at 6m TDFA length, although noise figure of TDFA increases as the length of fiber increases but still the noise figure is very low (<3dB) for 6m TDFA fiber. Gain of TDFA remains same for all input powers ranging from -30 dB to 0 dB. Gain of TDFA increases linearly as the pump power increases maximum gain is achieved at 1000mW. Also for high pump power, the noise figure is very low. Maximum gain is achieved at 1470 nm and gain decreases as wavelength increases positive values of gain have been obtained up to 1510 nm. References Whitley, T., 1994. Progress toward a practical 1.3 micron fiber amplifier, in Proc. ECOC., FGenze, Italy, pp. 939-946. Singh, R., Singh, M. L. and Singh, J., 2014. 160 Gbps S-band WDM transmission system over 1400 km using ITUT-652a and ITUT-655 fibers with low dispersion accumulation, Optik - International Journal for Light and Electron Optics, Volume 125, Issue 3, Pages 905-907 Sun, Y., Sulhoff, J. W., Srivastava, A. K., et al., 1998. A gain-flattened ultra wide band EDFA for high capacity WDM optical communications systems, in Proc. Eur. Conf. Optical Communications,Madrid, Spain, p. 53. Kani, J., Hattori, K., Jinno, M., Kanamori, T. and Oguchi, K., 1999. Triplewavelength- band WDM transmission over cascaded dispersion-shifted fibers, IEEE Photon. Technol. Lett., vol. 11, pp. 1506–1508, Nov. 1999. Watekar, P. R., Ju, S.M. and Han, W.T., 2007. Analysis of 1064 nm pumped Tm-doped silica glass fiber as 1470 nm amplifier, J. Lightw. Technol., vol. 25, no. 4, pp. 1045–1052.

[225]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Multimedia Stream Transmission in Mobile Ad hoc Networks Sajaad Ahmed Lonea*, Faroze Ahmadb a Department of Computer Science and Engineeing, IUST, Awantipora, J&K Department of Electronics and Communication Engineeing, IUST, Awantipora J&K

b

Abstract Multimedia is known as highly voluminous data, generally have high bandwidth, high bit rate requirement, and stringent delay constraint, whereas achievable bandwidth and bit rate for multimedia transmission on each link is limited in mobile ad hoc networks. Because of the bandwidth and bit rate constraints of mobile ad hoc networks in this paper, multiple path transmission technique is applied between source and destination to overcome these constraints. Most of the coding and transmission schemes for multimedia stream transmission employ layered coding that is multimedia stream is split into base layer and one or more enhancement layers. The base layer is transmitted with highest priority with strong error protection while as enhancement layer with low priority and with low control bits. Therefore, to transmit the base layer only on the paths which are better in terms of low end-to-end delay and packet loss. This work proposed a new scheme for multipath selection, which takes into consideration the end-to-end delay and packet loss of the path at the time of transmitting the packets. In this scheme the active probing packets are not used to measure the QoS of the multiple paths, instead QoS is measured at the receiver and the report of the best paths is send to the source after certain intervals of time.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Multimedia; MPEG4; QoS; Multipaths; RBMS; Traffic Allocator; Active Probing; Sender Based Approach; Round Robin Scheduling

1. Introduction Mobile ad hoc network is a dynamically reconfigurable system of wireless nodes with no infrastructure and connected by wireless links the union of which forms an arbitrary graph (Corson and Macker, 1999). There is no mobility restriction on the mobile nodes, which can reorganize themselves in a rapid and unpredictable fashion topology. Message transfer in mobile ad hoc networks can either take place between two nodes that are within transmission range or indirectly through multiple hops. The nodes which act as intermediate nodes in the data transfer process must be willing to participate in communication until successful message transfer has been accomplished, and the failure of such an event will result in message getting lost or interpreted as an unproductive congestion. The nodes in a mobile ad hoc network are free to move arbitrarily, and thus topology may change frequently and rapidly at unpredictable times. Wireless links in mobile ad hoc networks have significantly lower capacity and are of low bit rate as compared to wired network and therefore transmission of voluminous data like multimedia is a very challenging task. In addition to high bit rate transmission, multimedia needs low end-to-end delay. To fulfill the above requirements for multimedia data transmission in these networks some mechanism has to be adopted to make multimedia transmission feasible in these networks. One characteristic of mobile ad hoc networks is that there are many paths available between source and destination. Thus, a mechanism that takes advantage of this multitude of paths between source and destination is bound to perform better. Moreover selecting a single path at any time for a specific connection, a better scheme will be would be always distribute the information among multiple paths, possibly with some information of the QOS of the path. The mechanism that has been used in this paper is as 1) Discovery of multiple paths between source and destination 2) Selection of subset of paths based on some goodness measure like end to end delay and packet loss * Corresponding author. Tel.: +91 9419 458660. E-mail address: [email protected].

ISBN: 978-93-82288-63-3

Lone and Ahmad/COMMUNE – 2015

3) Transmission on these available paths. To overcome the constraints mentioned above, multiple paths between source and destination are used which aggregates the available bandwidth of all the available paths to the destination. The transmission of multimedia packets is generally on priority basis which means the signal is split into base layer and one or more enhancement layers (Gogate et al, 2002). The base layer is transmitted with high priority with strong error protection and needs to be transmitted on low end to end delay and low packet loss paths. The enhancement layer is transmitted on the paths with lower quality and discarded in case of congestion. The objective is that the quality of multimedia signal at the receiver should be acceptable even if only one enhancement layer is received. To judge the quality of the path Receiver Based Multipath Selection (RBMS) protocol is proposed in this paper which takes into consideration the QoS of the path while scheduling the multimedia packets on multiple paths. Existing algorithms for multipath selection have not proven better, as these algorithms do not take into consideration the QoS of the path and gives equal priority to all the packets. The rest of the paper is organized as follows: Section 2 focuses on the existing approaches for multipath selection. Section 3 discusses about Receiver Based Multipath Selection algorithm (proposed). Section 4 presents simulation. Section 5 presents performance results. Section 6, 7 presents conclusion and future work of the proposed work. 2. Related Work Multipath routing establishes multiple network paths between pair of routes to provide more efficient load balancing and higher performance paths as compared to unipath routing. In practice, multipath routing is implemented via equal cost multipath as specified in (Moy, 1998) an extension of OSPF that establishes multiple paths with identical hop count. Once multiple routes are established, the traffic allocator at the transmitting node needs a policy to determine how to allocate individual packets to the paths. Round robin allocation of packets along paths is the most commonly deployed policy due to simplicity. A second policy is to divide the traffic into hash function applied to the source and destination pair possibly includes port no and protocol Id (Hesselbach et al, 2005). However this policy require computationally complex per packet flow and does not guarantees into a desired traffic splitting ratio as flow rates are known. These two splitting polices ignore the relative quality of the paths. Besides these algorithms for the multipath splitting there are some algorithms that take into account the quality of the path but they have other disadvantages. The reference picture selection algorithm for splitting traffic (Lin et al, 2001) uses a splitting policy depending upon the reference of the good path i.e. path on which positive acknowledgement is received is used as a reference for the next frame. This mechanism of scheduling has been used to send the coded frames on separate paths. This mapping of frame to a path depends upon the available bandwidth. The disadvantage of this mechanism is that it cannot be used when number of paths is large. The Opportunistic Multipath Scheduling (Cetinkaya and Knighty, 2004) is one of the best scheduling policies proposed, but disadvantage of this splitting policy is that it uses active probing technique for the measurement of the quality of the paths. The splitting of the traffic is done according to the end to end delay and queue size of the path. The Opportunistic Multipath Scheduling algorithm schedules the packet over multiple paths at the source splitter. The objective of this algorithm is to minimize the average delay of the multipath traffic by exploiting the time varying conditions while also satisfying the route splitting weights. By taking into consideration the disadvantages of the above-mentioned splitting policies, this paper proposes a new approach for multipath selection known as Receiver Based Multipath Selection (RBMS) algorithm, which takes into consideration the quality of the path at the time of transmission and at the same time does not use any active probing technique for the path measurement. Another feature of the proposed scheduling algorithm is the receiver-based approach for the path quality measurement, which will reduce the overload of the sender. 3. Receiver Based Multipath Selection Protocol (RBMS) Model 3.1

Overview of the Receiver Based Multipath Selection.

The RBMS protocol has been proposed for multimedia stream transmissions. Most of the multimedia stream transmission schemes employ a layered coding technique over wireless channels. With these schemes, the multimedia streams are split into a base layer and one or more enhancement layers. The base layer is transmitted with highest priority and with strong error protection, while as the enhancement layer is transmitted with fewer control bits and is simply discarded in case of congestion in the path. In order to transmit the base layer on the best paths that is the paths having low end to end delay and packet loss RBMS algorithm has been proposed. In this approach, the QoS of the multiple paths, which map to the recent network, is calculated by the receiver in terms of end-to-end delay and packet loss and the report about the QoS of the paths is sent to the source periodically. The source node upon receiving report uses only the paths, which are having low end-to-end delay and packet loss. In Receiver Based Multipath Selection is an attempt to overcome the disadvantages of the Round Robin and Hash based scheduling algorithms, which does not take QoS of the path into consideration while transmitting.

[227]

Lone and Ahmad/COMMUNE – 2015

3.2

Detailed Description of Receiver Based Multipath Selection

The traditional protocols like OSPF and UDP select the shortest path for the packet transfer, and hence cannot guarantee the optimum path for transferring the packet. These protocols do not select the best paths among the available paths from sender to the receiver. Different packets in a multimedia stream often need to be assigned a different priority. The UDP and OSPF treat all the packets with same priority and same QoS. In RBMS algorithm, several optimum paths are selected to transfer the multimedia packets to resolve the above disadvantage. Transferring all the packets on one path will cause more delay and congestion on that path and will result in into hotspots and bottlenecks in the network. Because of this, the quality of the path will decrease. In the proposed protocol, load balancing is achieved i.e. several best paths are selected to transfer different packets according to the quality of the path. For multimedia streams, one can select best paths to transfer the base layer, while less efficient paths for enhancement layer in mobile ad hoc network. This will enhance the QoS as important layers the are provided with the best paths. In sender-based approach, the server records the information to determine the optimum path to the receiver. For normal transmission, there will be sometime other nodes connected, and this will increase the sender node’s overload. So by using the proposed algorithm most of the functionalities of the path QoS are performed by the receiver node. In most protocols, if the sender and receiver want to probe the network for the QoS service, they must introduce additional control packets. In the proposed protocol, neither the sender nor the receiver needs to use additional packets for testing the traffic of the network. The main idea of the protocol is to select several optimum paths instead of one path to transfer the packets. Receiver based approach is used in the protocol in order to reduce the overhead incurred by the sender node. The traffic information of the path is detected by the receiver from the multimedia packets sent by the source node. Because the media packets includes the sent packet time and packet sequence number, the receiver has the ability to know the packet delay and packet loss in the path. The feedback information is sent from the receiver to the sender using the RTCP protocol. However in between after some intervals of time Round Robin scheduling is used so as to consider the paths for the transmission which were less efficient in the past and at present are better in terms of end to end and packet loss. The basic architecture of our scheme is shown in figure 1. A scenario is considered with multimedia sender and multimedia receiver node in the mobile ad hoc network. There are multiple paths between the sender and the receiver and these are characterized by the varying parameters like delay, bandwidth, and packet loss. In normal scenarios (like OSPF), packets will be routed by along the shortest path. Therefore, different metrics must be used to find alternate better paths. It is proposed to find these paths by measuring the end-to-end delay and packet loss experienced by the path 3.3

Architecture of Receiver Based Multipath Selection

The proposed architecture of the Receiver Based Multipath Selection Protocol is shown in Figure 1.On the sender side the multimedia source is fed to the MPEG4 encoders, which has a number of content-based functionalities. Each scene in the is defined in one background and one or more foreground audiovisual objects. These audiovisual objects are encoded and the encoded streams are transmitted on different paths by the traffic allocator according to the quality of path. The Receiver Based Multipath selection algorithm is implemented in the traffic allocator block as shown on figure 1. In our implementation, the traffic source for MPEG4 was taken from library of MPEG4 encoded movies from university of Berlin website, containing the MPEG4 frames (Fitzek and Reisslein, 2001). Therefore, in this paper no focus will be on the encoding and decoding part of the proposed architecture. For transmitting the MPEG4 multimedia streams on multiple paths, the sender uses the feedback received from the receiver node about the QoS of the various paths in the network.

Fig1. System Architecture for proposed Receiver Based Multipath Selection

On the receiver, side on receiving the multimedia packets the receiver calculates the end-to-end delay and packet loss of the paths, selects several optimum paths and the report of these optimum paths is send to the sender using RTCP

[228]

Lone and Ahmad/COMMUNE – 2015

protocol. 3.4

Protocol Implementation

In Receiver Based Multipath Selection Protocol, the receiver will record previous multimedia packets sent by the source node. These packets will be delivered by RBMS protocol in the form of MyProtoSend, which is implemented on the source node. After receiver receives a certain number of packets, which is assumed to be 500 for our simulation, it maintains a data structure to store the statistics. These statistics include the number of packets received on a particular path, the average delay and the number of packets lost on that path. Using this data, the receiver finds the optimum paths. These paths are then assigned weight and this information about the paths and the receiver sends their corresponding weight to the sender using feedback. The weights are assigned using following function: Weighti =f (Sj, Dj, Pli) for packets j= 1, 2, 3….n on path i=1, 2...m where “f” is the function to calculate the weight according to the average delay (D) and the packet loss (PL). The sender can now distribute the packets along the paths depending upon the feedback received. The overall regular interval, the receiver would be able to update the weights of the specified paths and find new better paths if any. This procedure is repeated after regular intervals in order to dynamically adapt to the changing network conditions. Thus, the proposed approach will adapt to the network conditions and give a better performance by sending larger proportion of the traffic on the least congested or least delay paths. 3.5

Design Issues for Receiver Based Multipath Selection The various issues related to the designing of Receiver Based Multipath Selection are as follows 1) Path selection: How the sender selects several optimum paths from the receiver feedback. In the proposed scheme the receiver records the previous packets and sends the feedback to the sender and sender selects optimum paths using the feedback 2) Selection Policy Metric: According to the packet delay and packet loss along the path. Packet delay can be calculated from the packet send time and receive time. Packet loss can be derived from packet sequence number 3) Feedback Mechanism: Using the RTCP to report the new optimum information to the server 4) Type of method(Static ,Dynamic): we use the dynamic method to let the receiver report to the sender periodically using RTCP protocol

4. Simulations The work in this paper uses NS-2.29 for the implementation and simulation of RBMS with the wireless extension by the CMU Monarch group (NS-2) which includes the modeling of an 802.11wireless LAN. It implements network protocols like TCP, UDP, RTP, RTCP traffic source behavior like FTP, Telnet CBR, VBR route queue management mechanism such as DropTail, RED, and CBQ. NS-2.29 has also support for various routing protocols like DSR, AODV, and TORA DSDV etc. For the simulation, we have used DSR as the multipath routing protocol. NS-2.29 is powerful tool, which has the support of developing own protocols. 4.1

Mobility Model

The random waypoint mobility model is used to model mobility in a rectangular field. The field configuration used was 500 x 500 m with 20 nodes. Here each node whether node transmitting multimedia or normal CBR packets starts its journey from random location to the random destination with a randomly chosen speed. Once the destination is targeted, another random destination is targeted at a pause. 4.2

Traffic Source

For the entire simulation, two traffic sources CBR for normal packet transmission and for trace files of the movie Jurassic Park available at the University of Berlin for multimedia transmission are used [8]. This trace file contains the encoded MPEG4 frames; the number of frames available in this trace file is 65 million frames and contains the following format  Frame Id  Frame Type [I,P,B]  Frame Length 4.3 Transmission Policies used in Simulation The simulation of multimedia traffic involves a network of 20 nodes, out of which 4 nodes are transmitting at different intervals of time. In case of both simulations Round Robin and Receiver Based Multipath Selection with

[229]

Lone and Ahmad/COMMUNE – 2015

multimedia traffic same policy has been used that is node number 4 transmitting multimedia traffic (Jurassic Park Movie), while others 1,2 and 3 node number transmitting CBR traffic. The other three nodes except multimedia transmitting node are used to see the performance of Receiver Based Multipath Selection when other nodes in the network are transmitting and sharing paths in the mobile ad hoc network. In the simulation to see the performance or Receiver Based Multipath Selection with CBR traffic all the four nodes are transmitting the CBR traffic. 4.4

Performance Metrics

For evaluating, the simulation results following parameters are used: End-to-End Delay: It is the difference between the times when packet is sent by the source to the time when receiver receives it Throughput: It is the amount of data from a source to a destination processed by the protocol for which throughput has to be measured for instance TCP, UDP, and MAC protocol. Packet Loss: Packet loss is the ratio of no of packets send and the number of packets received 5. Performance Results Intensive simulation with different parameters has been carried out in this paper to see the performance of Receiver Based Multipath Selection Algorithm. The performance results have been analyzed and its comparison is made with the Round Robin Multipath Selection algorithm. The parameters used for the comparison used are throughput, end-to-end delay, and packet loss. The comparison of the performance of the both the algorithms are made by using both multimedia and normal CBR traffic. The comparative analysis is shown below. 5.1

Comparative Analysis Round Robin and Receiver Based Multipath Selection with Multimedia Traffic.

End To End Delay (ms)

1) End to End Delay Figure 2 shows the performance of the four transmitting nodes in the mobile ad hoc network of 20 nodes. The node 1, 3 and 4 transmit the CBR traffic of 512 Kb of packet size while as the node 2 transmits the multimedia traffic (Jurassic Park Movie). In case of Receiver Based Multipath Selection have got low end to end delay in case of the node 2 transmitting the multimedia traffic as only paths with low end to end delay and packet loss is used for the transmission. 800 700 600 500 400 300 200 100 0

ROUND ROBIN RBMS

1

2

3

4

Node Number

Fig 2. Performance End-To-End delay, Round Robin Vs RBMS Multimedia Traffic

Throughput(Kbps)

. 2) Throughput 700 600 500 400 300 200 100 0

ROUND ROBIN RBMS

1

2

3

4

Node Number

Fig 3. Performance Throughput Round Robin Vs RBMS Multimedia Traffic

The throughput of all the four nodes transmitting at different intervals of time during simulation is shown in Figure 3. The performance in case of throughput has been increased as compared to the Round Robin Scheduling algorithm.

[230]

Lone and Ahmad/COMMUNE – 2015

This is because only the optimum paths are used for transmission in case of both multimedia and CBR traffic in case of Receiver Based Multipath Selection. 3) Packet Loss The analysis Figure 4 shows no packet loss in case of the nodes transmitting the CBR traffic using Receiver Based Multipath Selection. However in case of node 2 has got the high packet loss as it is transmitting the multimedia traffic Jurassic Park movie trace file. This trace file is too big having 65 million frames because of this heavy traffic most of the packet loss occurred in node 2

PacketsLost

600 500 400

ROUND ROBIN

300

RBMS

200 100 0 1

2

3

4

Node Number

Fig 4. Performance packet loss , Round Robin Vs RBMS Multimedia Traffic

6. Limitations Receiver Based Multipath Selection is based on the assumption that the multiple paths between source and destination are discovered. The overhead incurred in assigning weights may not be justifiable for the small sessions in which case short single path session may rove to be better solutions. Receiver Based Multipath Selection approach can not prioritize the different packets used in the multimedia , base and enhancement layer packets but can transmit all the packets on best paths whether it is of high priority or not. 7. Conclusion and Future Work In this paper, a new proposal for the multipath selection has been proposed. The proposal provides the protocol for the “Multimedia Stream Transmission in Mobile Ad hoc Networks”. Receiver Based Multipath Selection has shown significant improvement as compare to the Round Robin scheduling in case of end to end delay, packet loss and throughput when CBR and Multimedia is taken as traffic source in the simulation. The main motive to develop Receiver Based Multipath Selection is to avoid active probing technique for the quality of path measurement and to avoid transmitting node overload, as the measurement of the quality of path is done at receiver in case of Receiver Based Multipath Selection. The advantages of using Receiver Based Multipath Selection are: a) it can reduce some of the limitations of the existing scheduling algorithms like Round Robin and hash, b) uses receiver based approach for to reduce the sender nodes overload, and c) does not use any additional for probing the network. Receiver Based Multipath Selection as explained does not prioritize the different packets, although we have developed this protocol for the for multimedia stream transmissions using multiple paths. So prioritizing the multimedia packets is the main feature that has be incorporated in this protocol to make it perform better for multimedia stream transmission. References S.Corson and J Macker “Mobile Ad hoc Networking (MANET): Routing Protocol Performance Issues and Evaluation Consideration” IETF RFC2 January 1999 http://wwwietf.org/rfc/rfc2501.txt. Nitin Gogate Doo-Man Chung Shivadra S.Panwar “Supporting Image and Video Application in an Multihop Radio Environment Using path Diversity and Mutiple Description Coding” 1051 –8215 @ 2002 IEEE J. Moy,”OSPF Version 2” 1998 Internet RFC 2328 Xavier Hesselbach,Ramon Fabreget,Banjamin Baran “Hashing based Traffic Partitioning in a Multicast-Multipath MPLS network Model” 2005 ACM 1-5593-008-6/05/0010 Shunan Lin, Shiwen Mao, Yao Wang, Shivendra Panwar “A Reference Picture Selection Scheme for Video Transmission over Ad-hoc Networks using Multiple Paths” in proc IEEE ICME Japan Aug 2001 page 96-99 Coskun Cetinkaya, Edward W Knighty “Oppertunistic Traffic Scheduling Over Multiple Network Paths” IEEE INFOCOM 2004 NS-2 Reference Manual from www.isi.edu Frank H.P Fitzek and Martin Reisslein “MPEG-4 and H.263 Video Traces for Network Performance Evolution” IEEE Network Vol 5, no 6, pp 40-54 November 2001.

[231]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Review on the Electrical Properties of Ultra-Thin Silicon Oxynitride Films Renu, RakeshVaid* Department of Physics and Electronics, University of Jammu, India

Abstract This review paper focuses on the research work done on the electrical characteristics of ultra-thin Silicon oxynitride films and investigates the effect of too high concentration of nitrogen on the quality of silicon-oxide dielectric. The high concentration of nitrogen also stops boron diffusion from poly-Si gate. It also has been explored that with the increase in refractive index, the dielectric constant increases and the charge-trap density decreases with the increase in oxygen concentration in the SiON film. Silicon oxynitride films can be grown via direct nitridation in nitrogen, which involves two different rapid thermal process (RTP) approaches: either by the oxidation of Silicon nitride (Si 3N4) or by the nitridation of silicon oxide (SiO2) under nitrogen flow. It has been found that the ultrathin silicon oxynitride films grown by the latter method exhibits excellent leakage characteristics. Based on these results it has been revealed that Silicon oxynitride has excellent potential to replace alternative gate dielectrics for the applications in MOS devices, ultra-thin film transistors, and non-volatile memory devices.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Silicon Oxynitride; High-k; Dielectric Films; Nitridation; MOS

1. Introduction Silicon has been used in MOS technology for many decades because of the better qualities of its native oxides such as low leakage current, low interface state density (1.4×1012 states per cm2) and good thermal stability (Garg et al, 2006). In order to accomplish faster switching speed and high device density in advanced semiconductor technology, the transistor gate length will keep on reducing in size and reaches sub-20nm by 2015. This results in reduction of gate length and gate oxide thickness to 1.5nm and ultimate oxide scaling down to EOT (Equivalent oxide thickness) of approximately 0.5nm is required (Khairnar, Mahajan, 2013). As the thickness of SiO2 shrinks below 1.5nm, it suffers from its physical limitations due to gate dielectric leakage currents and reliability requirements (Kim, 2008). The shrinkage of device dimension without corresponding reductions in supply voltage leads to very high electric fields near the drain of the device (Dennard et al, 1974). Strong electric fields results in physical damage, which leads to worsening of device reliability. With the decrease in SiO2 film thickness, its blocking power decreases and therefore, the penetration of impurities (boron) into the gate dielectric becomes more prominent. Thus there is a need for alternate channel materials that can enhance channel mobility beyond the physical limits of Si based MOS devices (Chandra et al, 2011). From the relation, COX=εOX /tOX, oxide thickness is inversely related to oxide capacitance therefore oxide capacitance can be increased instead of oxide thickness to meet the scaling requirements. Because C = k εA/t where k is the dielectric constant, A is the Area of the capacitor and t is the thickness, larger dielectric constant k is needed to increase the overall capacitance – that’s where the high-k dielectric materials come into play. High-k dielectrics are crucial for small power-consumption, small leakage and high performance logic devices for future generation. As a good replacement for SiO 2, the high-k materials ought to have numerous advanced features apart of high-k value such as they should be chemically stable with Si substrate and the gate electrode and they should be thermally stable at temperatures no less than 5000C. Furthermore, they should have good quality interface properties with the Si substrate so that the structure can have low interface trap density, high channel mobility, low oxide trap density, large bandgap, and large band offset energies. SiON has been used as a substitute for SiO2 because of its high dielectric permittivity and low density of surface states. It is a thermally stable structure with a high crystallization temperature and stable on silicon (Wong, Iwai, 2006). The structure of stable oxynitride is analogous to the SiO2 that is each silicon atom is coordinated with four atoms with any grouping of oxygen and nitrogen (Fig 1). It also has a comparatively better dielectric/Si interface in comparison with other high-k contenders. It has the

*Corresponding author. Tel.: +91 9419 106794. E-mail address: [email protected].

ISBN: 978-93-82288-63-3

Renu and Vaid/ COMMUNE – 2015

properties like large band gap (5.9–8.9 eV), greater conduction band offset (>2.1 eV) and the high dielectric constant (3.9-7.8) which are adequate for MOS device applications.

Fig 1. Silicon oxide and silicon oxynitride have similar bonding structure (Wong, Iwai, 2006)

In addition, by differing O/N ratio the band energy of SiON can be tailored between 5 and 9 eV (Konofaos, 2004; Fainer et al, 2005). The incorporation of nitrogen reduces the generation of defects thus causing a reduction in trap formation, which is formed by hot electron injection (Cartier et al, 1994a; Buchanan et al, 1994; Cartier et al, 1994b). In the past, NH3 was used to introduce N into films but hydrogen incorporation causes electron trapping thus requiring reoxidation to reduce its effect (Saks et al, 1994). Silicon Oxynitride have been shown to have superior electrical properties over the silicon dioxide when it comes to interface traps, breakdown voltage strength and boron penetration resistance (Liu et al, 1997; Soleimani et al, 1995; Doyle et al, 1995; Liu et al, n.d; Lu et al, 1996; Tobin et al, 1994).

2. Related Work J. H. Liao et al. (Liao et al, 2009) studied the relationship between both the physical and electrical characteristics of SiON films and the refractive index. The single wafer rapid thermal process (RTP) modules were used for low-pressure chemical vapour deposition (LPCVD) of the SiON films with various refractive indices. XPS and FTIR analysis was performed to investigate the bonding configurations of different SiON films. The relationship between the ratio of the dielectric constant and the refractive index of the SiON films shown in Fig 2 indicates that the ratio of dielectric constant is directly proportional to the refractive index. As a charge-storage layer, the SiON has a lower trap density and the reduced Coulombic repulsion between the trapped charges which improves the charge retention performance of SONOS devices. In addition, the nitrogen in the SiON layer reduces the silicon-dangling bonds at the interface and results in a decreased current leakage, which improves the reliability of MOS devices. Z. H. Lu et. al. (Lu et al, n.d) studied the interaction of Si (100) surface with N2 under rapid thermal process condition by X-ray photoelectron spectroscopy and it was found that the oxynitride formed by the nitridation of SiO 2 revealed excellent leakage current results (shown in Fig 3) which explores its suitability as a replacement for SiO2 in deep submicron MOS device applications. The Capacitance-Voltage characteristics of various dielectrics shown in Fig 4 shows that 32 Å SiON film grown by oxidation of nitride exhibits a flat band voltage of –0.35 V, which is lesser than SiO2 (-0.75). Due to lower threshold voltage, the oxynitride can be an attractive contender for low power CMOS devices. D. Brazzeli et al. (Brazzeli, 2001) explores the relationship between the boron barrier diffusion and the high-k dielectric quality. The insitu steam generation technique (ISSG) and the nitridation by NO or N 2O was used for the growth of thin oxides. The comparison was made between the three different nitridation processes: N 2O and NO after ISSG step and NO on native oxide layer followed by ISSG oxidation. It was observed that the sample with NO after ISSG exhibits lowest interface state density, which is due to the larger N concentration at the Si/SiO 2 interface due to the stronger Si-N bond with respect to Si-H bond (Fukuda et al, 1996; Alessandri et al, 1997). But the same sample exhibits oxide degradation due to increased nitrogen concentration. Thus, there is a trade off between the boron barrier diffusion and the quality of oxide.

Fig 2. Relationship between the ratios of dielectric constant index of the SiON films (Liao et al, 2009)

Fig 3. Current-Voltage characteristics of various and the refractive Dielectrics (Lu et al, n.d)

[233]

Renu and Vaid/ COMMUNE – 2015

.

Fig 4. Capacitance-Voltage characteristics of various dielectrics (Lu

et al, n.d)

S. Hwang et al. (Hwang et al, 2007) deposited thin film of silicon oxynitride by using plasma assisted N 2O oxidation in an inductively-coupled-plasma chemical-vapour deposition (ICP-CVD) reactor and studied the effect of the presence of a high concentration of nitrogen in a silicon-oxide layer under the different process conditions. Ultra-thin films with improved capacitance-voltage characteristics was obtained at the constant flow rate of nitrous oxide (N 2O) of 2.5 sccm, 10m Torr working pressure, 150 W RF power, 350 0 C substrate temperature and a 10 min deposition time. It was found that a long processing time or a high RF power damages the Si surface. The results reveal the suitability of oxynitride films in high quality ultra-thin-film transistors and non-volatile memory devices. 3. Results and Analysis Fig 5 shows the comparison between the high-frequency capacitance-voltage characteristics of simulated nSi/SiO2/Pt and n-Si/SiON/Pt MOS capacitor for the dielectric thickness of 1.5nm at different frequencies. The simulation was done with the help of PADRE based simulation tool. We have taken the value of dielectric constant of SiON at 5.9 and SiO2 at 3.9. For the nitrogen rich oxide the graph shows increase in capacitance at both the frequencies of 10 KHz and 1 MHz as compared to SiO2.This is due to the fact that capacitance is directly related to the dielectric constant and inversely related to the oxide thickness. The graph shows shift in C-V characteristics for SiON because at lower frequencies the inversion state forms and inversion capacitance increases due to the capacitance response of minority carries and the traps and at high frequency the traps are not only localized but also distributed in the dielectric nearer to the interface and non equilibrium behaviour is exhibited by the interface states. The calculated flat band voltage of SiON is 1.7V which is lower than SiO2 (2V) that makes it suitable for NVM devices. Fig 6 shows the position versus energy graph for the SiO2 and SiON dielectrics. Due to the presence of nitrogen in the oxynitride film there is a change in the band gap. Thus the band alignment is directly related to the concentration and distribution of nitrogen in the silicon oxide film. ×10

-6

1.5

3.5

SiON SiO2

1.0

SiON

2.5

0.5 Position (µm)

2

Capacitance (F/cm )

3.0

2.0 1.5

10kHz 10KHz 1MHz 1MHz

1.0

SiO2

Conduction band

0.0

-0.5

Valence Band

0.5

-1.0 0.0 -3

-2

-1

0

1

2

3

0

Voltage (V)

1

2

3

4

5

Energy (eV)

Fig 5. High-Frequency Capacitance-Voltage Characteristics

Fig 6. Energy-Band diagram of SiO2 and SiON

[234]

Renu and Vaid/ COMMUNE – 2015

4. Conclusion Silicon oxynitride film has advantages over silicon oxide and other high-k materials. There are various methods to grow nitrogen rich oxide film but the oxynitride film grown by direct nitridation of SiO2 yields improvement in interface quality, stops boron diffusion from the polysilicon gate and excellent leakage results. The incorporation of nitrogen also affects the band alignment and the quantity of nitrogen in the film can be directly controlled by the time of exposure to nitrogen and the operating temperature. Thus silicon oxynitride has excellent potential to substitute silicon oxide dielectric and can be utilized as a tunnel oxide in metal floating gate non-volatile memory devices, ultrathin film transistors and future MOS applications. References Alessandri, M., Clementi, C., Crivelli, B., Ghidini, G., Pellizzer, F., Martin, F., Iwai, M., Ikegawa, H., 1997. Microelectronics engineering 36, p. 211. Brazzeli, D., Ghidini, G., Crivelli, B., Zonca, R., Bersani, M., 2001. High quality thin oxynitride by RTP annealing of in situ steam generation oxides for flash memory applications, Solid state electronics 45, p.1271. Buchanan, D. A., Marwick, A. D., DiMaria D. J., Dori, L., 1994. Hot-electron-induced hydrogen redistribution and defect generation in metal-oxidesemiconductor capacitors, Applied physics 76, p. 3595. Cartier, E., Buchanan, D. A., Dunn, G. J., 1994a. Atomic hydrogen-induced degradation of re-oxidized nitrided silicon dioxide on silicon, applied physics letters 64, p. 901. Cartier, E., DiMaria, D. J., Buchanan, D. A., Stathis, J. H., Abadeer, W. W., Voilersten, R. R., 1994b. IEEE Transaction electron devices 99, p. 1234. Dennard, R. H., Gaensslen, F. H., Yu, H., Rideout, V. L., Bassons, E., Leblance, A. R.,1974. Design of ion-implanted MOSFETs with very small physical dimensions, IEEE Journal of solid state circuits 9 p. 256. Doyle, B. S., Soleimani, H. R., A. Philipossian, A.,1995, Simultaneous growth of different thickness gate oxides in silicon CMOS processing, IEEE Electron devices letters 16, p. 301. Fainer, I., Kosinova, M. L., Macimovsky, A. E. A., Rumyantsev Yu. M., Kuznetsov, F. A., Kesler, V. G., Kirienko, V. V., 2005. Study of the structure and phase composition of nano crystalline silicon oxynitride films synthesized by ICP-CVD”, Nuclear instruments and methods in physics research section A 543, p. 134. Fukuda, H., Endoh, T., Nomura, S., 1996, Proceedings of the third international symposium on physics and chemistry of SiO2 and SiO2-Si interface, 15. Garg, R., Misra,D., Swain, P. K., 2006. Ge MOS Capacitors with Thermally Evaporated HfO 2 as Gate Dielectric, Journal of the electrochemical society 153, p. F29. Hwang, S., Jung, S., Jang, K. S., Lee, J. I., Park, H., S. K. Dhungel S. K., Yi, J., 2007, Properties of the ultra-thin silicon-oxynitride films deposited by using plasma-assisted N2O oxidation for semiconductor device applications, Journal of the Korean physical society 51, No. 3, p. 1096. Khairnar, A. G., Mahajan, A. M., 2013. Effect of post-deposition annealing temperature on RF-sputtered HfO2 thin film for advanced CMOS technology, Solid-state sciences 15, p.24. Kim, H.S., 2008. A study of HfO2-based MOSCAPs and MOSFETs on III-V substrates with a thin germanium interfacial passivation layer, Ph.D. thesis, The University of Texas, Austin. Konofaos, N., 2004. Electrical characterization of SiON/n-Si structures for MOS VLSI electronics. Microelectronics Journal 35, p. 421. Liao, J. H., Hsieh, J. Y., Lin, H. J., Tang, W. Y., Chiang, C. L., Lo, Y. S., Wu, T. B., Yang, W. L., Yang, T., Chen, K. C., Lu, C. Y., 2009. Physical and electrical characteristics of silicon oxynitride films with various refractive indices, Journal of physics D: Applied Physics 41, p.1. Liu, C. T., Ma, Y., Becerro, J., Nakahara, S., Eaglesham, D. J., Hillenius, S. J., 1997. Light nitrogen implant for preparing thin-gate oxides, IEEE Electron devices letters 18, p. 105. Liu, C. T., Ma, Y., Cheung, K. P., Chang, C. P., Fritzinger, L., Becerro, Luhan, H., Vaida, H. M., Colonell, J., Kamgar, A., Mionr, J. F., Muirray, R.G., Lai, W. Y. C., Pai, C. S., Hillenius, S. J., Symposium on VLSI technology digest of technical papers IEEE, Piscataway, NJ, 1 W6) 18. Lu, H. C., Gusev, E. P., Gustafsson, T., Garfunkel, E., Green, L. M., Brasen, D., Feldman, L. C., 1996. High resolution ion scattering study of silicon oxynitridation, Applied physics letters 69, p. 2713. Lu, Z. H., Khoueir, A., Ng, W. T., Tay, S. P., Growth of ultrathin nitride on si (100) by rapid thermal N2 treatment. S. V. J. Chandra, M. Jeong, Y.C. Park, J. W. Yoon and C. J. Choi, 2011. Effect of Annealing Ambient on Structural and Electrical Properties of Ge Metal-Oxide-Semiconductor Capacitors with Pt Gate Electrode and HfO2 Gate Dielectric, Materials transactions 52, p. 118. Saks, N. S., Ma, D., Fleetwood, D. M., Twigg, M. E., 1994. Characteristics of oxynitrides grown in N2O, in silicon nitride and silicon dioxide thin insulating films, Electrochemical society proceedings series 94- 16, p. 395. Soleimani, H. R., Doyle, B. S., Philipossian, A., 1995, “Formation of Ultrathin Nitrided SiO2 Oxides by Direct Nitrogen Implantation into Silicon,” Journal of electrochemical society 142, p. L132. Tobin, P. J., Okada, Y., Ajuna, S. A., Lakhotia, V., Feil, W. A., Hedge, R., 1994. Journal of applied physics 75, p. 1811. Wong, H., Iwai, H., 2006, “On the scaling issues and high-k replacement of ultrathin gate dielectrics for nanoscale MOS transistors”, Microelectronic engineering 8, p. 1867.

[235]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Design of Frequency Reconfigurable Dual Band Microstrip Patch Antenna with Directional Patterns Babu lal Sharma*, Girish Parmar, Mithilesh Kumar, Baljeet Singh Sinwar University College of Enginnering, Rajasthan Technical University, Kota, Rajasthan, India

Abstract The frequency reconfigurable microstrip patch antenna with reflector at the bottom has been given here. To overcome the challenges of dual-frequency and frequency reconfigurable operation, a new scheme of dual band frequency reconfigurable has been proposed and thereafter the design of a dual band microstrip patch antenna with frequency reconfigurable operations has been suggested. For switching the frequency and for enhance the directivity of all the frequency bands, three diodes and reflector have been used, respectively.. The antenna has compact size of 80  60  3.06 mm3 including the ground plane. In the proposed design, there is a microstrip patch antenna with a slot in ground plane. The antenna is designed on Taconic RF-35 substrate (εr= 3.5) of thickness (H) 3.04 mm. The proposed structure is simulated using CST (Computer Simulation Technology), which is electromagnetic (EM) simulation software. The simulated return loss for all the bands are less than -10 dB at different frequencies.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Dual-Band; Microstrip Line; Patch Antenna; Frequency Rconfigurable Slot

1. Introduction Reconfigurable antenna is capable to reconfigure its characteristics such as frequency, pattern, bandwidth, and polarization to adapt to the environment. Moreover, frequency reconfigurable antennas are used to reduce the size of front end system. Thus, frequency reconfigurable antenna can support many applications depending on its operating frequency. The idea of reconfigurable antenna was proposed in 1930 as reported by Haupt, R. L. and M. Lanagan (2013). DiFonzo et al. reported that a pattern reconfigurable antenna was designed in 1979 for satellite communication. From 1998 to present day, many reconfigurable microstrip antenna has been designed for different frequency bands as well as for different applications (Pourziad et al.,2013, Jusoh et al.,2013, Majid et al., 2012). The proposed antenna is used to reconfigure at five different combinations depending upon the switching of the diodes. The effective length (electrical length) of an antenna plays a major role in determining the operating frequency. There are many technologies to achieve frequency reconfiguration. The effective length of an antenna can be modified using electronic switches. Nowadays, electronic switches, such as RF MEMS, FETs and PIN diodes, are commonly used for frequency reconfiguration as demonstrated (Huda A. Majid et. al, 2014). Although PIN diode is used in many designs because it has good performance such as low cost and ease of fabrication. PIN diodes are used in the ground plane of antenna to switch the frequency bands. A similar method for slot antenna is also reported. There are also some other methods of switching which were reported in many works. All the methods have some advantage as well as some disadvantage but PIN diode have acceptable performance. So, In this work, frequency reconfiguration is accomplished by three PIN diodes. 2. Antenna Design The proposed antenna is described in this section with dimensional view. Figure 1(a) demonstrates the front view of the design with a slot inserted at the patch. This antenna is designed on Taconic RF-35 with a permittivity of 3.5 and substrate *

Corresponding author. Tel.: +91 9509 160038. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Sharma et al/COMMUNE – 2015

thickness of 3.04. The patch is designed using conventional formulas of rectangular microstrip patch antenna and inset feeding is used to feed the antenna. The dimension of the antenna is tabulated in Table 1 and all the parameters are in millimeter. For the best return loss at the desire frequency, proper impedance matching is required which has been achieved by inset feed. A reflector is used for increasing the directivity. This reflector is placed 33 mm far from ground plane of antenna. There is also a slot in ground plane in which three PIN diodes are placed for switching the frequency. Three diodes have the eight possibilities of switching out of them five are explained in the results. Figure 1 (b) shows the side view of antenna and Figure 1 (c) shows the bottom view of antenna. Positions of all the three diodes are shown in Figure 1 (c). The antenna and the reflector has the area of L1×L2 mmz2 and L10×L11 mm2, respectively. A 50 ohm impedance line is also designed in order to offer perfect impedance matching. Table 1. Dimensions of Proposed Antenna Parameters L1 L2 L3 L4 L5 L6 L7

Dimension (in mm) 80 60 18.3 9 13 29 4

Parameters L8 L9 L10 L11 L12 L13

(a)

Dimension (in mm) 1.5 1 102 122 8.5 6.4

(b)

36 3 D3 D2 D1 PIN Diodes (c) Fig. 1. Geometry of proposed antenna (a) front View (b) Side view (c) back View

3. Results and Discussions Time domain solver of CST Microwave studio is used to simulate the proposed design. After simulation of design the reflection coefficients are obtained and radiation patterns for directivity have also been calculated. S- parameters (reflections coefficient) and radiation patterns are shown in Fig. 2-7. Figure 2 shows the return loss of antenna at different switching configuration of diodes as shown in Table 2. Each switching configuration has the dual band operation and return loss below -10 dB which shows that there is good impedance matching. The operating frequencies for five configuration of switching are 2.26, 5.70 GHz, 3.28, 4.86 GHz, 3.37, 4.93 GHz, 3.36, 4.94 GHz, 3.49, 5.00 GHz, respectively.

[237]

Sharma et al/ COMMUNE-2015

Table 2. Switching configuration and operating frequencies of proposed antenna. Sr. No. 1 2 3 4 5

Diode 3 0 0 0 1 1

Diode 2 0 0 1 0 1

Diode 1 0 1 1 1 1

Frequency 1 2.26 3.28 3.37 3.36 3.49

Directivity (in dBi) 8.8 7.2 7.6 7.3 7.1

Frequency 2 5.70 4.86 4.93 4.94 5.00

Directivity (in dBi) 7.5 8.7 9.2 9.8 9.1

Fig. 2. Simulated S- parameters of the proposed antenna with different switching configurations

(a) Fig 3. Radian Pattern for (a) 2.26 GHz (b) 5.70 GHz

(b)

(a) Fig. 4. Radian Pattern for (a) 3.28 GHz (b) 4.86 GHz

(b)

(a) Fig.5. Radian Pattern for (a) 3.37 GHz (b) 4.93 GHz

(b)

[238]

Sharma et al/COMMUNE – 2015

(a) Fig.6. Radian Pattern for (a) 3.36 GHz (b) 4.94 GHz

(b)

(a) Fig.7. Radian Pattern for (a) 3.49 GHz (b) 5.00 GHz

(b)

4. Conclusions A frequency reconfigurable antenna having slot in ground plane as well as in patch with directional radiation pattern has been designed using CST Microwave studio. It has been observed that the frequency reconfigurability can be achieved by three PIN diodes as a switch in the ground slot of the antenna. Using these PIN diodes, five different frequency bands can be reconfigured. The antenna gives a good directional radiation patterns due to reflector, which is placed at the back of the proposed antenna, at all frequency bands. The proposed antenna has good impedance matching at every band. In near future, one can reconfigure polarization and other properties of antenna using different switching methods, which can also be taken up by future research workers in this area. References Huda A. Majid, Mohamad K. A. Rahim, Mohamad R. Hamid, and Muhammad F. Ismail, 2014, "Frequency Reconfigurable Microstrip Patch-Slot Antenna with Directional Radiation Pattern", Progress In Electromagnetics Research, Vol. 144, 319-328. Haupt, R. L. and M. Lanagan, 2013, "Reconfigurable antennas," IEEE Antennas and Propagation Magazine, Vol. 55, No. 1, 49-61. Gardner, P., P. S. Hall, and J. Kelly, 2008, "Reconfigurable antennas for cognitive radio: Requirements and potential design approaches," Institution of Engineering and Technology Seminar on Wideband, Multiband Antennas and Arrays for Defence or Civil Applications, 89-94. DiFonzo, A., D. Karmel, and P. Atia, 1979, "Multiple shaped beam reconfigurable satellite antenna," IEEE Antennas and Propagation Society International Symposium, Vol. 17, 457-460. Capone, E. and R. Pelaca, 1986, "Advanced satellite multifeed reconfigurable antenna with very good cross-polar and scan performances," IEEE Antennas and Propagation Society International Symposium, Vol. 24, 723-725. Chang, B. C. C., Y. Qian, and T. Itoh, 1999, "A reconfigurable leaky mode/patch antenna controlled by PIN diode switches," IEEE Antennas and Propagation Society International Symposium, Vol. 4, 2694-2697. Majid, H. A., M. K. A. Rahim, M. R. Hamid, and M. F. Ismail, 2012, "Frequency and pattern reconfigurable Yagi antenna," Journal of Electromagnetic Waves and Applications, Vol. 26, Nos. 2-3, 379-389.

[239]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Study the Effect of Diode Area on the Current-Voltage and Capacitance-Voltage Characteristics of Al/n-SnSe2/In Thin Film Schottky Diodes R. Sachdeva*, U. Parihar, N. Padha University of Jammu, Department of Physics and Electronics, , Jammu, India

Abstract SnSe2 thin films of the thickness of 300nm were deposited at a substrate temperature of 473K by thermal evaporation technique on Indium coated (thickness of 200 nm) glass substrate, which serves as back ohmic contact to the Schottky Diode. A layer of Aluminum (Al) of the thickness 200 nm was deposited on n-SnSe2 thin films, which formed rectifying contacts with n-SnSe2. Al/nSnSe2/In Schottky diodes having different areas viz 0.6 mm2, 0.9 mm2 and 9 mm2 were fabricated using suitable masks. The diodes were subsequently undertaken for the Current-Voltage (I-V) and Capacitance-Voltage (C-V) behaviour at room temperature. The data obtained thus used for calculating the diode parameters such as ideality factor (η), barrier height (ɸbo), series resistance (Rs) and breakdown voltages (VBR). The studies were then undertaken to investigate the effect of change in diode area on the current transport behaviour of the Al/n-SnSe2/In Schottky diodes. From the analysis of the experimental data, it has been observed that barrier heights (ɸbo) increase (from 0.63 eV-0.68 eV) and ideality factors (η) decrease (from 5.15-3.31) with decrease in diode area. Further, the breakdown voltage (VBR) of the undertaken Schottky diodes was however found to increase from 0.52-1.21V with decrease in diode area. Thus smaller area diodes show better parameters such as breakdown voltage, ideality factor and barrier height as compared to large area diodes. The area dependent C-V characteristics of the undertaken Schottky diodes were also studied at room temperature. The Schottky barrier height deduced from I-V characteristics at room temperature has been observed lower than that obtained from the C-V characteristics.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Ideality Factor; Barrier Height; Series Resistance; Breakdown Voltage

1. Introduction The metal-semiconductor contact has been a subject of many investigations because of its existence in electronic circuits; importance in advanced VLSI or ULSI technologies and a fundamental interest to understand the behaviour of Schottky barriers(S. M. Sze; Rhoderick E. H). In the present investigation, attempt has been made to study the effect of diode area on the I-V and C-V characteristics. Further, Al/n-SnSe2/In Schottky diodes of different diode areas i.e 9 mm2, 0.9 mm2 and 0.6 mm2 have been formed on the In coated glass substrates by using suitable masks. The I-V and CV data thus obtained have been analyzed to calculate various diode parameters viz ideality factor (η), barrier height (ɸbo), series resistance (Rs) and breakdown voltage (VBR). The current transport phenomenon across the schottky barrier diode has been widely studied and various attempts been made to understand its behaviour but its complete description is still a debatable issue. Several factors such as the surface states, interface states, defects & dislocations on the semiconductor surface, image force lowering, barrier inhomogenities, field emission as well as presence of interfacial layer between metal and the semiconductor have been found contributing in the current transport behavior of MS contacts (M.S. Tyagi). There are in general two types of MS contacts which are either ohmic or rectifying (schottky).

* Corresponding author. Tel.: 9419185073 E-mail address: [email protected] ISBN: 978-93-82288-63-3

Sachdeva et al/COMMUNE – 2015

2. Experimental Details SnSe2 thin films of the thickness of 300nm were deposited at a substrate temperature of 473K by thermal evaporation technique on Indium coated (thickness of 200 nm) glass substrate which serves as back ohmic contact to the Schottky Diode. A layer of Aluminum (Al) of the thickness 200 nm was deposited on n-SnSe2 films which formed rectifying contacts with n-SnSe2. Al/n-SnSe2/In Schottky diodes having different areas viz 0.6 mm2, 0.9 mm2 and 9 mm2 were fabricated using suitable masks. Fig. 1(a,b & c) shows the structure of Al/n-SnSe2 Schottky diode fabricated on Indium coated glass substrates and its schematic view.

(

(

Fig. 1 (a) Structure of Al/n-SnSe2 Schottky diode fabricated on Indium coated glass substrates (b) schematic view and (c) images of the different areas of Al/n-SnSe2 Schottky diodes

The room temperature Current-Voltage (I-V) as well as Capacitance-Voltage (C-V) measurements of the undertaken Schottky diodes were measured using a computer interfaced setup comprising a programmable Keithley Source Meter (model-2400) and Precision programmable LCR meter (Agilent make 4284A). The setup was established through interfacing of these instruments using LabVIEW software (National Instruments, U.S.A). This setup was successfully used for the measurements of I-V and C-V characteristics of the Schottky diodes at different temperatures and frequencies. 3. Result and Discussions 3.1

Forward Current-Voltage (I-V) Characteristics

In the undertaken Al/n-SnSe2 Schottky diodes, the experimental data have been analyzed on the basis of thermionic emission (TE) theory. According to which the current flow over a uniform metal-semiconductor interface at a forward bias V  3kT/q can be expressed as [5]   qV      1 I  I s exp   KT   

(1)

where Is is the saturation current and Js; the saturation current density, defined as Js 

Is

A

 q   A**T 2 exp   bo   kT 

(2)

The quantities, A is the diode area, A**; the effective Richardson constant (= 18 A K -2 cm-2 has been used in the present work (Werner J. H)), T; the measurement temperature in Kelvin, k; Boltzmann’s constant (=1.38 ×10-23 J/K), q; the electron charge (=1.6 × 10-19C), V; the forward applied voltage, bo ; the zero-bias barrier height and RS, the series resistance of the diode respectively. The ideality factor η included in equation (1) has been introduced to describe the deviation of the experimental I-V data from the ideal TED mechanism (Tyagi M.S) expressed as: (3) dV  q  n   kT  d ln( I / I s ) The zero bias barrier heights ( bo ) of the Al/n-SnSe2 Schottky diodes have been determined from the equation (2) on the basis of thermionic emission and diffusion theory. The voltage across the diode can be expressed in terms of the total voltage drop across the diode as well as the series resistance. This is accounted by replacing the voltage V by (VIRs) in equation (1), therefore, equation (1) becomes   qV  IRS    (4)   1 I  I s exp   KT   

[241]

Sachdeva et al/COMMUNE – 2015

Fig. 2 shows the room temperature current-voltage (I-V) characteristics of different sized Al/n-SnSe2 schottky diodes wherein it has been observed that slopes increase and there is a shift towards higher voltage side with decreased area i.e. larger the Schottky contact area higher is the current passing through the sample (M.S. Tyagi). 1

0.1

2

ln J (A /mm )

0.01 2

9 mm 2 0.9 mm 2 0.6 mm

1E-3

1E-4

1E-5 0.0

0.5

1.0

1.5

2.0

Voltage (V)

Fig. 2 Forward current-voltage characteristics of Al/n-SnSe2 Schottky diodes with different areas

Various diode parameters viz., η and

bo

extracted from the linear fitting of the I-V data has been presented in Table

1, it is observed that ideality factor increases and barrier height decreases with increase in the schottky diodes area. It is attributed to the fact that as device area increases, effects of the defects on the surface and other factors at the interface would increase and hence, cause more deviations from the ideal current transport behaviour. Table 1 Schottky Diode parameters for different areas

η

0.6 3.31 0.68

Diode area (mm2) 0.9 3.85 0.65

9 5.15 0.63

Rs() Vbr(V)

20K -1.12

10K -0.81

1.79K -0.52

Diode Parameter

bo (eV)

Series resistance (Rs) has been important parameter of schottky diode which however cannot be evaluated using linear fitting of the I-V plot. A method to extract the series resistance Rs of ideal Schottky diodes (η = 1) was first proposed by Norde. For cases where η > 1, Sato and Yasumura had modified Norde's approach to extract the values of η, bo and Rs from the forward 1-V data of the schottky diode. The approach when applied to Schottky diode requires two experimental 1-V measurements conducted at two different temperatures and the determination of the corresponding minima to the modified Norde's function (Yakuphanoglu F. and Senkal F). In undertaken Schottky diodes, the linear region is very small (Refer Fig. 2), thus linear fitting method is not much suitable and an alternate technique developed by S. Cheung was used to determine the values of η, bo and Rs from a single I-V measurement using equations given hereunder (Cheung S. K. and Cheung N. W): d (V )  H ( J )  Rs AJ   bo (6) (5) and  Rs AJ  d (ln J )  -2

J (A cm )

-2

J (A cm ) 0.0010

0.0015

0.0020

0.0025

0.0030

0.0035

0.0040

0.19

0.32

0.0002 0.0003 0.0004 0.0005 0.0006 0.0007 0.0008 0.0009 0.0010 2.59

3.50 3.48

0.17

3.46 3.44

3.40 0.24

3.38

0.16

H vs J

2.58

d(V)/d(lnJ) vs J Linear Fit

2.57

 = 3.85 b= 0.65 eV

2.56

0.15

2.55

0.14 2.54 0.13 2.53

3.36 0.22

(a) Diode area = 9 mm

H (Volts)

3.42

0.26

H (Volts)

d(V)/d(lnJ) (Volts)

0.28

0.18

H vs J d(V)/d(lnJ) vs J Linear Fit  = 5.15 b= 0.63 eV

d(V)/d(lnJ) (Volts)

0.30

0.12

2

3.34

(b) Diode area = 0.9 mm

2

2.52

0.11 0.20

3.32 0.0010

0.0015

0.0020

0.0025

0.0030

0.0035

2.51 0.0002 0.0003 0.0004 0.0005 0.0006 0.0007 0.0008 0.0009 0.0010

0.0040

-2

J (A cm )

-2

J (A cm )

-2

J (A cm ) 0.0000 0.0005 0.0010 0.0015 0.0020 0.0025 0.0030 0.0035 2.48 0.30 H vs J 0.28 0.26

2.44 2.42 2.40

0.22

2.38

0.20

2.36

0.18

2.34

0.16

H (Volts)

d(V)/d(lnJ) (Volts)

0.24

2.46

d(V)/d(lnJ) vs J Linear Fit  = 3.31 b= 0.68 eV

2.32

0.14

2.30 (c) Diode area = 0.6 mm

2

0.12

2.28

0.10

2.26 0.0000 0.0005 0.0010 0.0015 0.0020 0.0025 0.0030 0.0035 -2

J (A cm )

Fig. 3 Plots of d(V)/d(lnJ) versus J and H(J) versus J of Al/n-SnSe2 Schottky diodes of different areas (a) 9 mm2 (b) 0. 9 mm2 and (c) 0.6 mm2

[242]

Sachdeva et al/COMMUNE – 2015

Using η value determined from Eq. (5), a plot of H (J) vs. J provides a straight line with y-axis intercept giving η bo while the slope of this plot provides a second determination of R s which can be used to check the consistency of this approach. Thus, two different plots corresponding to equations (5) as well as (6) of the J-V data obtained from one measurement can determine all the three key diode parameters η, bo and Rs. Fig. 3 shows d(V)/d(lnJ) versus J as well as H(J) versus J plots for schottky diodes with three different areas. The parameters extracted from the linear fitting of ln(I) versus V plots using thermionic emission equation as well as those extracted from the Cheung’s method are summarized in Table 2. However, it can be seen that there is relatively small difference in the values of the η, bo from the downward curvature region of forward bias I-V characteristics and linear portion of the same characteristics. Table 2 Comparison of diodes parameters extracted from the linear fitting of ln(I) versus V plots using thermionic emission equation and those extracted by using Cheung’s method

Diode Area(mm2 ) 0.6 0.9 9 3.2

Linear Fitting using TE equation Barrier Ideality Ideality Height Factor Factor ( bo ) eV (η) (η) 3.42 4.18 5.46

0.72 0.70 0.67

Barrier Height ( bo ) eV

3.31 3.85 5.15

0.68 0.65 0.63

Cheung’s Method Series Resistance SeriesResistance (Rs) (KΩ) (Rs) (KΩ) Using Eq. 6.13 Using Eq. 6.15 1.79 10.31 20.14

1.92 10.91 21.85

Reverse Current-Voltage (I-V) Characteristics

The reverse bias current-voltage (I-V) characteristics of different area Al/n-SnSe2 Schottky diodes at room temperature have been shown in Fig. 4 where it has been observed that initially the current change is negligible upto a certain voltage and increases fast above this point for all the three different area diodes.

(a)

(b)

Fig. 4 Reverse biased I-V Characteristics of Al/n-SnSe2 schottky diodes with different diode areas measured at Room Temperature

However, in all the cases the knee point is not well defined and possesses a soft breakdown phenomenon. The breakdown voltage has been defined as a voltage corresponding to a reverse current of few micro-amperes. It has, however been observed that at room temperature, the breakdown voltage (VBR) increases with decrease in diode area Table 1. 3.3

Capacitance-Voltage (C-V) Characteristics

The C–V relationship is applicable to intimate metal contacts on uniformly doped semiconductor materials (schottky barrier) and can be rewritten as [4]

2VR  Vo  1  2 C q s N A A2

(7)

where VR is the reverse bias voltage, Vo ; the built-in potential or diffusion potential, which is usually measured by extrapolating the C−2-V plot to the V-axis. The zero-bias barrier height ( bo ) from the C–V measurement can be measured as

 bo Vd  Vn 2

(8)

The variation of 1/C -V curve at 1 MHz frequency for the schottky diodes of different areas have been illustrated in

[243]

Sachdeva et al/COMMUNE – 2015

Fig. 5. The donor concentration ND and the zero-bias barrier height

bo

at 1 MHz for the Al/n-SnSe2 schottky diodes

-2

have been calculated from the experimental C -V characteristics. The value of ND was found to vary in the range 1.50 x 1018- 1.64 x 1018 m-3 for the schottky diodes of different areas. 2

9 mm 2 0.9 mm 2 0.6 mm

1.85

1.75

18

-2

1/C (1x10 F )

1.80

2

1.70

1.65

1.60

1.55 -2.0

-1.5

-1.0

-0.5

0.0

0.5

Voltage (volts)

Fig. 5 Plot of 1/C2versus V of Al/n-SnSe2 Schottky Diode 1 MHz frequency having different areas 9 mm2 , 0.9 mm2 and 0.6 mm2 Table 3 Parameters of diodes extracted from C-V data

Diode Parameter

bo

(eV)

-3

18

ND (cm )x(10 )

0.6

Diode area (mm2) 0.9

9

0.88 1.50

0.81 1.60

0.71 1.64

The capacitance–voltage (C-V) measurements have also been used to measure the barrier height which found to vary in the range 0.71-0.88 eV as shown in Table 3. The value of barrier heights bo obtained from C-V measurements are higher than the same obtained from the I-V measurements. The barrier heights deduced from the two techniques are always different due to the nature of I-V and CV measurements. The discrepancy between the CV and  IV can be explained on the basis that barrier heights discrepancies over the interface are caused by inhomogenities such as non-uniformity in the interfacial layer thickness and distributions of the interfacial charges(Yakuphanoglu F. and Senkal F). On the other hand the barrier height obtained from the C-V method includes an average value of the schottky barrier heights of the patches available in the contact. 4. Conclusion Al/n-SnSe2 Schottky diodes with three different areas (0.6 mm2, 0.9 mm2, 9 mm2) have been fabricated on In coated glass substrates. It has been observed that the shift of the slope towards higher voltage side increases with decrease in area. It has also been observed that barrier height ( bo ) increases from 0.63 eV-0.68 eV and ideality factor (η) decreases from 5.15-3.31 with decrease in diode area. The breakdown voltage (VBR) of the undertaken Schottky diodes was found to increase from 0.52-1.21V with decrease in diode area. It is found that smaller area diodes have higher breakdown as compared to big sized area diodes. The possible reason for these variations has been attributed to the fact that as device area increases, the effect of ‘interface defects’ and several other factors would increase and add up causing deviations in the ideal current transport behaviour. The area dependent C-V characteristics of the undertaken Schottky diodes were also studied at room temperature. The schottky barrier height deduced from I-V characteristics at room temperature has been observed lower than that obtained from the C-V characteristics. This may be due to the reasons that I-V analysis includes the ‘image force lowering’ and ‘dipole lowering effects’, ‘tunneling’ and leakage currents as already been reported for similar diodes by several authors. References S. M. Sze, “Physics of Semiconductor Devices” (Wiley, New York, 1981). Rhoderick E. H., Metal Semiconductor Contacts (Clarendon, Oxford, 1978). M.S. Tyagi, “Introduction to Semiconductor Materials and Devices” (John Wiley & Sons, New York, 2008). M.S. Tyagi, “Introduction to Semiconductor Materials and Devices” (John Wiley & Sons, New York, 1991). Martinez-Pastor J., Segura A., J.L. Valdes and Chevy A., J. Appl. Phys. 62 539 (1987). Werner J. H., Appl. Phys. A47, 291 (1988). Tyagi M.S., “Introduction to Semiconductor Materials and Devices” (John Wiley & Sons, New York, 2008). M.S. Tyagi, “Introduction to Semiconductor Materials and Devices” (John Wiley & Sons, New York, 1991). Norde H., J. Appl. Phys. 50 5052 (1979). Sato K. and Yasumara D Y., J. Appl. Phys. 58 3655 (1985). Cheung S. K. and Cheung N. W., Appl. Phys. Lett. 49 85 (1986). Yakuphanoglu F. and Senkal F., Polym. Adv. Technol. 19 1882 (2008).

[244]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Performance Evaluation of Multiplexer Designs in Quantum-Dot Cellular Automata (QCA) M. R. Beigh*, M. Mustafa Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India

Abstract Quantum-dot Cellular Automata (QCA) is one of the promising successors to conventional CMOS technology. QCA offers advantages such as low power consumption, high operating speed, high throughput, and deep pipelining. This technology allows the implementation of logic circuits using quantum devices (quantum dots) instead of the traditional devices such as transistors, diodes, and resistors. The basic building block in this technology is the QCA cell that is used to obtain not only logic gates but also interconnection wires. Various digital logic designs are already available in the literature presently. This paper evaluates the performance of various implementations of QCA based multiplexer designs and proposes novel layouts with better performance parameters. The designs are compared in terms of area and latency. The simulations have been carried out using the QCADesigner tool.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Nanocomputing; Quantum-dot Cellular Automata; Digital logic; Multiplexer; QCADesigner

1. Introduction Quantum-dot cellular automata (Lent et al., 1993) is a promising future technology for transistorless nanoelectronic computing which rely upon the novel design concepts. The QCA cells have features on the very low nanometer scale, much smaller than the smallest transistor. Many studies have reported that QCA for implementation of logic circuits can achieve high device density, extremely low power consumption, and very high switching speed (Lent et al., 2003). A device architecture based on QCA cells provides the opportunity to break away from transistor-based logic. This paradigm exploits the quantum effects that come with small size. In this new paradigm, the basic logic element is no longer a current switch but a small array of quantum dots. A QCA cell consists of four quantum dots arranged in a square pattern (Orlov et al., 1997). In this computing paradigm, the fundamental logic primitive is a 3-input majority gate. The logic element, a majority gate has an odd number of binary inputs and a binary output. Majority logic helps in the implementation of digital operations based on the principles of majority decision. Thus the output of a majority gate is a logical 1 when the majority of inputs are at logic 1, and logical 0 when majority of inputs is logic 0. The majority gates along with the binary inverters helps in the implementation of digital functions as per the logic design applications (Beigh et al., 2013; Mustafa and Beigh, 2013; Mustafa and Beigh, 2014). In this paper, we propose novel implementations of QCA based 2:1 and 4:1 multiplexer designs. A detailed comparison with regard to various characteristics of these designs is also presented. The paper has been organized in four sections. The first and second section provides the necessary brief review of QCA fundamentals. The third section presents the multiplexer designs available in the literature and proposed area efficient designs. A novel QCA based multiplexer designs are presented at the end of third section. The advantages of the proposed structures have been summed up as conclusion in the fourth section.

*

Corresponding author. Tel.: +91 9622 741766. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Beigh and Mustafa /COMMUNE – 2015

2. QCA Basics Quantum-dot Cellular Automata emerged as a new paradigm, beyond current switches to encode binary information. QCA encodes binary information in the charge configuration within a cell. Coulomb interaction between cells is sufficient to accomplish the computation in QCA arrays—thus no interconnect wires are needed between cells. No current flows out of the cell so that low power dissipation is possible (Timler and Lent, 2002). 2.1.

The Basic QCA Device

QCA cells perform computation by interacting coulombically with neighboring cells to influence each other’s polarization. Fig.1shows a high-level diagram of a four-dot QCA cell. In this arrangement, four quantum dots are positioned to form a square. Quantum dots are small semiconductor or metal islands with a diameter that is small enough to make their charging energy greater than kBT (where kB is Boltzmann’s constant and T is the operating temperature in degrees Kelvin). It is reported that in future, they will shrink to regions within specially designed molecules (Isaksen and Lent, 2003). In this case, they will trap individual charge barriers (Taugaw and Lent, 1994; Lent and Taugaw, 1997). Exactly two mobile electrons are loaded in the cell. By means of electron tunneling, they can move to different quantum dots in the QCA cell. Coulombic repulsion will cause the electrons to occupy only the corners of the QCA cell. This will result in two specific polarizations as shown in Fig. 1.

Fig 1. QCA cell polarization and representation of binary 1 and binary 0

For an isolated cell there are two energetically minimal equivalent arrangements of the two electrons in the QCA cell, denoted by cell polarizations P = +1 and P = -1 representing a binary 1 and a binary 0 respectively. It is also worth noting that there is an unpolarized state as well. In an unpolarized state, interdot potential barriers are lowered which reduces the confinement of the electrons on the individual quantum dots. Consequently, the cells exhibit little or no polarization and the two-electron wave functions delocalize across the cell (Lent and Taugaw, 1997). The fundamental QCA logical circuit is the three-input majority gate shown in Fig. 2 (Taugaw and Lent, 1994). With the majority gate, computation is performed by driving the device cell i.e., cell 4 in Fig. 2(a) to its lowest energy state. This condition will take place when it attains the polarization of the majority of the three input cells.

Fig. 2. (a) The fundamental QCA logic device- the majority gate; (b) The circuit symbol.

2.2.

The QCA Clock

In contrast to the standard CMOS clock, the QCA clock has more than a high and a low phase. All the QCA circuit proposals require a clock signal. The clock signal synchronizes and controls information flow. It actually provides the power to run the circuit. QCA computation is carried out by controlling the tunneling with a four phase “clock” signal as shown in Fig. 3. The clocking of QCA is achieved by controlling the potential barriers between adjacent quantumdots (Snider et al., 2006; Toth and Lent, 1999). The clock used in QCA consists of four phases namely hold, release, relax, and switch. It is reported that the lag between adjacent phases is 90°. In other words the clock changes phase when the potential barriers that affect a group of QCA cells are raised or lowered or remain raised or lowered.

[246]

Beigh and Mustafa /COMMUNE – 2015

Fig 3. The 4 phases of QCA Clock

When the inter-dot barrier is raised, and the QCA cell settles down to one of the two ground polarization states (as influenced by its neighbors), the phase is known as switch phase, During the hold phase, the inter dot barrier is held high, preventing electron tunneling and maintaining the current ground polarization state of the QCA cell. The inter dot barriers are lowered during the release and relax phases and the excess electrons gain mobility. In these two phases, a QCA cell remains in an unpolarized state. Generally, the polarization of a QCA cell is determined when it is in its switch phase by the polarizations of its neighbors that are in switch and/or hold phases. The unpolarized neighboring cells in release and relax phases will have no effect on determining the state of the QCA cell (Kim et al., 2007). It has been reported that the clock signals can be generated by CMOS wires embedded below the QCA plane through an induced electric field (Lent and Taugaw, 1997). In Fig. 3 a QCA binary wire is used to illustrate this clocking system. The information transfers in a pipelined fashion and after every 4 time steps, it is possible to put a new value onto a QCA wire (Huang et al., 2007). It is evident from Fig. 3 that there is a 90 degree phase shift from one clock zone to the next. 3. Multiplexer Design Multiplexer is an important part in many variant circuit designs, which include implementation of signal control systems and memory circuits. It allows us to choose one of the inputs and transfer it to the output. The characteristic equation of 2:1 multiplexer is given by equation (1) below, where A and B are the two inputs and S is the select line used to select one between the two inputs. Output (Y) = A.S + B.S’ 3.1.

(1)

Previously Presented Multiplexers

There are various implementations of multiplexers in QCA already present in the literature. In the present work various QCA multiplexer designs have been studied and a new efficient design is proposed. The proposed multiplexer design is compared with the layouts presented in (Kim et al., 2007; Teodosio and Sousa, 2007; Hashemi et al., 2008; Tehrani et al., 2011) in terms of cell count, area and complexity. The multiplexer implementations presented in the previous work are shown in Fig. 4. As mentioned previously, many QCA logic designs including multiplexer can be implemented using majority gate as a fundamental building block. The equivalent expression for output variable of a multiplexer based on majority function (M) is given by equation (2) below. Output (Y) = M [M (B, S, 1), M (B, S, 0), A] (2) Fig. 4(a) depicts the multiplexer presented in (Kim et al., 2007). The three multiplexer designs, based on the majority gate function, presented in (Teodosio and Sousa, 2007), (Hashemi et al., 2008) and (Tehrani et al., 2011) are shown in Fig. 4(b), 4(c) and 4(d), respectively. The layouts shown in Fig. 4(b) and 4(c) are constructed in three layers; however, the design shown in Fig. 4(c) is smaller and therefore more efficient in terms of area and complexity.

(a)

(b) [247]

Beigh and Mustafa /COMMUNE – 2015

(c)

(d)

Fig. 4. Multiplexer designs already available in the literature (Kim et al., 2007; Teodosio and Sousa, 2007; Hashemi et al., 2008; Tehrani et al., 2011).

3.2.

Proposed Multiplexer Design

The proposed 2:1 QCA multiplexer design is shown in Fig. 5(a). This design consists of just 25 cells covering an area of 0.03 μm2. The proposed multiplexer has been implemented on a single layer in QCADesigner without any crossover. A 4:1 multiplexer based on the proposed 2:1 multiplexer is shown in Fig. 6(a). This 4:1 multiplexer consists of three blocks of 2:1 MUX and is implemented in two stages. It comprises of 132 cells with an area of 0.24 μm2. The larger multiplexers such as 8:1 MUX, 16:1 MUX can also be constructed using proposed 2:1 multiplexer design. The proposed QCA designs were simulated with default parameter values in QCADesigner 2.0.3. Table 1 gives the comparison of proposed 2:1 MUX design with the designs already proposed by various authors. It is evident from Table 1 that the proposed design is efficient in terms of area and cell count.

Fig. 5: QCA Layout and simulation results of proposed 2:1 MUX

Fig. 6. QCA Layout and simulation results of proposed 4:1 MUX

[248]

Beigh and Mustafa /COMMUNE – 2015

Table. 1. Comparison of multiplexer designs Area (μm2)

MUX design(2:1)

No. of layers

Cell count

Multiplexer presented in (Kim et al., 2007) Fig. 4(a)

1

88

0.14

Multiplexer presented in (Teodosio and Sousa, 2007)

3

46

0.08

3

36

0.06

1

62

0.12

1

25

0.03

Fig. 4(b) Multiplexer presented in (Hashemi et al., 2008) Fig. 4(c) Multiplexer presented in (Tehrani et al., 2011) Fig. 4(d) Proposed Multiplexer Fig. 5(a)

4. Conclusion Multiplexer with its wide range of applications is an important and fundamental element in most commonly used electronic circuits. Many QCA implementations of multiplexers have already been proposed. This paper presents novel and efficient designs of 2:1 and 4:1 QCA multiplexers with minimum complexity and cell count. The proposed multiplexers have been simulated using QCADesigner with default parameters and tested in terms of complexity (number of layers, cell count) and area. The main aim of this work is to propose the layouts with minimum cell count in order to maximize the circuit density and focus on simpler and area efficient layouts. As it is evident from the comparison table (Table 1) the proposed multiplexer has some advantages over the previous designs in terms of complexity and area. References Lent, C. S., Tougaw, P. D., Porod, W., Bernstein, G. H., 1993. Quantum cellular automata, Nanotechnology 4, p. 49 Lent, C. S., Snider, G. L., Bernstein, G., Porod, W., Orlov, A., Lieberman, M., Fehlner, T., Niemier, M., Kogge, P. 2003. Quantum-dot Cellular Automata in “Electron Transport in Quantum Dots” Jonathan P. Bird, Editor. Springer US, p. 397. Orlov, A.O., Amlani, I., Bernstein, G.H., Lent, C.S., Snider, G.L., 1997. Realization of a Functional Cell for Quantum-Dot Cellular Automata, Science 277, p. 928. Beigh, M. R., Mustafa, M., Ahmad, F., 2013. Performance Evaluation of Efficient XOR Structures in Quantum-dot Cellular Automata (QCA), Circuits and Systems 4, p.147. Mustafa, M., Beigh, M. R., 2013. Design and Implementation of Quantum Cellular Automata Based Novel Parity Generator and Checker Circuits with Minimum Complexity and Cell Count, Indian Journal of Pure and Applied Physics 51, p. 60. Mustafa, M., Beigh, M. R., 2014. Novel Linear Feedback Shift Register Design in Quantum-dot Cellular Automata, Indian Journal of Pure and Applied Physics 52, p. 203. Timler, J., Lent, C.S., 2002. Power Gain and dissipation in Quantum-dot Cellular Automata, Journal of Applied Physics 91, p. 823. Isaksen, B. and Lent, C. S., 2003. “Molecular quantum-dot cellular automata”, Third IEEE Conference on Nanotechnology, IEEE-NANO vol. 1,. San Francisco, CA Tougaw, P.D., Lent, C.S., 1994. Logical devices implemented using quantum cellular automata, Journal of Applied Physics 75, p. 1818. Lent, C. S., Tougaw, P. D., 1997. A device architecture for computing with quantum dots, Proceedings of the IEEE 85, p. 541. Snider, G., Orlov, A., Lent, C.S., Bernstein, Lieberman, G., M., Fehlner, T., 2006. “Implementation of Quantum-dot Cellular Automata”, ICONN 2006, p. 544. Toth, G., Lent, C. S., 1999. Quasiadiabatic switching for metal-island quantum-dot cellular automata, Journal of Applied Physics 85, p. 2977. Huang, J., Momenzadeh, M., Lombardi, F., 2007. Analysis of missing and additional cell defects in sequential quantum-dot cellular automata, INTEGRATION, VLSI journal 40, pp. 503. Kim, K., Wu, K., Karri, R., 2007. The Robust QCA Adder Designs Using Composable QCA Building Blocks, IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 26, p.176. Teodosio, T., Sousa, L., 2007. "QCA-LG: A tool for the automatic layout generation of QCA combinational circuits," Norchip, 2007, p.1, Aalborg, Denmark. Hashemi, S., Azghadi, M.R., Zakerolhosseini, A., 2008. “A novel QCA multiplexer design”, International Symposium on Telecommunications, p.692, Tehran. Tehrani, M.A., Safaei, F., Moaiyeri, M.H., and Navi, K., 2011. Design and implementation of Multistage Interconnection Networks using Quantumdot Cellular Automata, Microelectronics Journal 42, p. 913. Walus, K., Dysart, T.J., Jullien, G.A., Budiman, R.A., 2004. QCADesigner: a rapid design and Simulation tool for quantum-dot cellular automata, IEEE Transactions on Nanotechnology 3, p. 26.

[249]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Impact of Scaling Gate Oxide Thickness on the performance of Silicon based Triple gate/Quad gate Rectangular-NWFET Deepika Jamwal, Richa Gupta, Rakesh Vaid* Department of Physics and Electronics, University of Jammu, Jammu, India

Abstract In this paper, we have made a comparative study of triple gate and quad gate Si based Rectangular-NWFET by scaling gate oxide thickness on its device performance in terms of device metrics viz. transfer characteristics, output characteristics, drive current (Ion), leakage current (Ioff), switching speed (Ion/Ioff), subthreshold slope (SS).We concluded that the conductivity and switching speed of quad gate Si-NWFET is larger than triple gate Si-NWFET by reducing oxide thickness and has an improvement over the leakage current, which means that gate has an efficient control over the channel. Further, SS for the triple gate approaches to 75.01mv/decade compared to its quad gate counterpart with 66.68mv/decade and highly improved drain induced barrier lowering (DIBL) which is likely to approach its ideal value by reducing the oxide thickness. Thus, quad gate configuration is better option for enhancing the performance of rectangular Si-NWFET at smaller value of oxide thickness.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords : Rectangular Si-NWFET; OMEN; Oxide Thickness; Drive Current; Leakage Current; Drain Current; Subthreshold Swing

1. Introduction According to the literature review, Si nanowire with 3D structures provides way to new applications viz. transistors, memories, sensors, light-emitting diodes and photovoltaic cells and Si nanowire FET with gate-around configuration has an ability to suppress the leakage current compatible for ultra- small CMOS devices (Iwai, 2009). Literature regarding experimental studies of Si nanowire FETs has been published which reveals on-current many times higher than conventional MOSFET (Bangsaruntip et al, 2009; Bidal et al, 2009; Bidal et al, 2009; Boykin et al, 2007; Ernst et al, 2006; Ernst et al, 2006; ITRS, 2009; Iwai, 2009; Kim,et al, n.d; Luisier, Schenk, 2008). With regard to ITRS 2009, conventional MOSFETs might disappear and 3D multiple gate (or Fin type) structures will come into play because of its tendency to shrink the short-channel effects (SCE) (ITRS, 2009). It has been given in literature that FinFETs will switch to nanowire FETs, because it has an efficient control over SCE and with the introduction of Si nanowire FETs, good compact models are essential but it is quite challenging job to develop compact model because drain characteristics of Si nanowire FETs are affected by bandstructure which is very sensitive to the nanowire diameter, cross-sectional shape, crystal orientation, mechanical stress and interface states. 2. Device Structure Typical device structure that OMEN can handle consists of circular nanowire FET and rectangular nanowire FET. Rectangular nanowire FET is further divided into triple gate and quad gate NWFETs as shown in Fig.1.In OMEN Nanowire FET, Nanowire is used as a channel, which is finally surrounded by a metal contact. The metal contact serves as a gate terminal. The source and drain contacts are surrounded by the insulating material i.e. SiO 2. The current from source to the drain is turned on and off by the voltage applied to the gate. Since the gate in nanowire is surrounding the channel, it controls the electrostatics of the channel much more efficiently than conventional MOSFET (Mehrotra et al,

* Corresponding author. Tel.: 9419106794. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Jamwal et al/COMMUNE – 2015

2009). Nanowire in nanowire field-effect transistors consists of any cross-section shape (i.e. square, circular, triangular, hexagonal,) and gate architecture may be (all-around, single, double, or triple), and transport direction consists of (<100>, <110>, <111>). Since the oxide layer that surrounds the channel do not participate in the calculation of transport behaviour, the electrons are confined in the channel and do not penetrate into the oxide (Luisier, Klimeck, 2008).

Fig. 1. Structure of Rectangular (a) triple gate and (b) quad gate NWFET

We can use different semiconductor as a channel materials, viz. Si, Ge, GaAs, InAs, AlAs, or SiGe has been employed. Ternary alloy semiconductors like InGaAs and AlGaAs are treated either as virtual crystals or as atomistically and randomly disordered entities (Boykin et al , 2007).The simulation tool (OMEN) which we have used in our work is a two and three-dimensional Schrodinger- Poisson solver based on the sp3d5s∗semi-empirical tightbinding method (Boykin et al, 2004.). We can obtain carrier and current densities by injecting electrons and holes at different momentum and energies into the device and then solving the resulting set of equations in the Wave Function (WF) or in the Non-equilibrium Green’s Function (NEGF) formalism (Luisier et al, 2006). In OMEN, the OBCs are obtained from a shift-and-invert procedure which is at least one order of magnitude faster than the usual approaches (Luisier, Schenk, 2008.). In this paper, we have a comparative study of silicon based triple gate and quad gate Rectangular-NWFET by scaling gate oxide thickness (T ox). Table 1: Parameters used for simulation Parameters

Symbols

Values

Drain Voltage

Vds

0.6 V

Channel Length

Lc

15 nm

Drain Length

Ld

10 nm

Source Length Channel Concentration

Ls Ch conc.

10 nm 1010cm-3

3. Results and Discussions In this section, by using OMEN Nanowire as a simulation tool, various results pertaining to the scaling of gate oxide thickness (Tox) can be described as follows: 3.1.

Doping Profile of the Device Versus Width for different Oxide Thickness

In Fig. 2, the variation of electron density with respect to the width for different values of oxide thickness (T ox) is shown. The curves for different values of oxide thickness overlaps with each other but still we can see the minute difference within the curves i.e. electron doping density is high for T ox= 1nm<1.5nm <2nm <2.5nm and least electron doping density is available for T ox=3nm.Thus, electron doping density (i.e. electron concentration/unit area) is high as we go on reducing oxide thickness (T ox) or vice versa which has been further verified by the equation:

(1)

[251]

Jamwal et al/COMMUNE – 2015

Fig. 2. Doping profile of the device with respect to the width for various oxide thicknesses

3.2.

Variation of Ids Versus Vgs for 1nm Oxide Thickness

In Fig. 3, the variation of drain current with respect to gate voltage at 1nm oxide thickness (Tox) for triple gate and quad gate Rectangular NWFET is shown. As it is acknowledged that at a lower value of Tox, conductivity gets surged. So, for transfer characteristics we made a comparison between triple and quad gate at T ox=1nm and examined that the drain current goes high for quad gate contrast to triple gate for same Tox because as the number of gate increases, its control over the channel increases. Thus, quad gate configuration has enhanced transfer characteristics.

Fig. 3. Variation of Ids Vs Vgs at Tox= 1nm in triple gate and quad gate Rectangular NWFET

3.3.

Variation of Ids Versus Vds for 1nm Oxide Thickness

In Fig. 4, the variation of drain current with repect to drain voltage at 1nm oxide thickness (Tox) for triple gate and quad gate Rectangular NWFET is shown. Refering to the explanation given in the section 4.2, we select Tox=1nm. From the graph, we can notice that the drain current goes high for quad gate compared to the triple gate and then saturates for both the configurations. Since as the number of gates increases, current driving capability of the device enhances. Thus, with the increase in the number of gates, quad gate configuration has highly improved drain characteristics.

Fig. 4. Variation of Ids Vs Vds at Tox= 1nm in triple gate and quad gate Rectangular NWFET.

3.4.

Variation of On Current (Ion) versus Oxide Thickness (Tox)

In Fig. 5, the variation of On current with respect to different values of Oxide thickness (T ox) for triple gate and quad gate Rectangular NWFET is shown. From the graph, it is clearly visible that On current remain high for quad gate Rectangular NWFET indicated by black curve than for triple gate depicted by grey curve at same value of Oxide thickness and it continue to increase as we go on reducing the value of oxide thickness (Tox). Thus, we can say that the value of On current (drive current) enhances as we switch on from triple gate to quad gate NWFET.

[252]

Jamwal et al/COMMUNE – 2015

Fig. 5. Variation of Ion Vs Tox in triple gate and quad gate Rectangular NWFET.

3.5.

Variation of Off Current (Ioff) versus Oxide Thickness (Tox)

In Fig. 6, the variation of Off current with respect to different values of Oxide thickness (T ox) for triple gate and quad gate Rectangular NWFET is shown. From the graph, it is clearly visible that Off current goes low for quad gate Rectangular NWFET indicated by black curve than for triple gate depicted by grey curve at same value of Oxide thickness and it continue to decrease as we go on reducing the value of oxide thickness (Tox). Thus, we can say that the value of Off current (leakage current) shrinks as we switch on from triple gate to quad gate NWFET.

Fig. 6. Variation of Ioff Vs Tox in triple gate and quad gate Rectangular NWFET.

3.6.

Variation of Ion/Ioff versus Oxide Thickness (Tox)

In Fig. 7, the variation of Ion/Ioff with respect to different values of oxide thickness (Tox) for triple gate and quad gate Rectangular NWFET is shown. From the graph it can be shown that the value of I on/Ioff goes high for quad gate Rectangular NWFET indicated by black curve than for triple gate depicted by grey curve at same value of Oxide thickness and also the factor increases with the low value of T ox which is desirable. Thus, switching speed of the device increases as we switch on from triple to quad gate NWFET with the reduction in oxide thickness (T ox).

Fig. 7. Variation of Ion/Ioff Vs Tox in triple gate and quad gate Rectangular NWFET

3.7.

Variation in Subthreshold Swing for triple gate and quad gate

In Fig. 8, the variation of subthreshold swing with respect to different oxide thickness is shown. From the graph, we can see that the value of SS gets lowered for quad gate compared to its triple gate counterpart at each value of oxide thickness. As we go on reducing the value of T ox, SS for triple gate shrinks from 91.17 to 75.01 mv/decade where as SS for quad gate approaches to its ideal value from 83.34 to 66.68 mv/decade i.e. SS gets lowered by 12.49 % in quad gate configuration. Thus, quad gate NWFET shows least value of SS at lower T ox which is desirable.

[253]

Jamwal et al/COMMUNE – 2015

Fig. 8. Variation of SS Vs Tox in triple gate and quad gate Rectangular NWFET Table 2: Comparison of device metrics of triple gate and quad gate rectangular NWFET S.No Device metrics

Triple gate

Quad gate

Rectangular NWFET

%age

Rectangular NWFET

1. Drive current(Ion)

1.26E-05

1.54E-05

22.22

2.Leakage Current(Ioff)

2.92E-11

2.19E-11

25.00

3.Switching Speed(Ion/Ioff)

4.30E+05

7.03E+05

63.48

4.Subthreshold Slope(SS)

75.01

66.68

12.49

4. Conclusion At this juncture, we can say that no other full band and atomistic simulator offer quantum simulation at an atomistic basis. Thus, by using OMEN nanowire as a simulation tool, we have made a comparative study of triple gate and quad gate Rectangular NWFET by scaling oxide thickness (T ox).It has been observed the conductivity and switching speed of quad gate Si-NWFET is larger than triple gate Si- NWFET by reducing oxide thickness and reduced leakage current which is an improvement over CNTs and planer MOSFETs. Further, SS gets lessened in quad gate to 66.68 mv/decade compared to the triple gate with 75.01 mv/decade and improved DIBL. Thus, quad gate configuration is better option for enhancing the performance of rectangular Si-NWFET at smaller a value of oxide thickness. References Boykin, T. B., Klimeck, G. and Oyafuso, F., 2004. Valence band effective-mass expressions in the sp3d5s∗ empirical tight-binding model applied to a Si and Ge parameterization, The University of Alabama in Huntsville, USA: Phys. Rev. B, vol. 69, pp.115201. Bangsaruntip, S., Cohen, G. M., Majumdar, A., et al., 2009. High performance and highly uniform gate-all-around silicon nanowire MOSFETs with wire size dependent scaling, IEDM, pp. 297–300. Bidal, G., Boeul, F., Denorme, S.S., et al., 2009. High velocity Si-nano-dot: a candidate for SRAM applications at 16 nm node and below, Symp VLSI Tech, pp. 240–241. Boykin, T. B., Luisier, M., Schenk, A., Kharche, N., and Klimeck, G., 2007.The electronic structure and transmission characteristics of disordered AlGaAs nanowires, IEEE Trans. on Nanotech. vol. 6, pp. 43-47. Ernst, T., Dupre, C., Isheden, C., et al., 2006. Novel 3D integration process for highly scalable nano-beam stacked-channels GAA (NBG) Fin FETs with HfO2/TiN gate stack,” IEDM, pp. 997–1001. Ernst, T., Duraff ourg, L., Dupre, C., et al., 2008. Novel Si-based nanowire devices: Will they serve ultimate MOSFETs scaling or ultimate hybrid integration, San Francisco, USA: IEDM, pp.745–749. ITRS 2009 Ed: http://www.itrs.net/Links/2009ITRS/Home2009.htm. Iwai, H., 2009. Roadmap for 22 nm and beyond Microelectron Eng, Tokyo Institute of Technology, Japan: Elsevier, pp.1520–1528. Kim, S.G., Mehrotra, S. R., Haley, B., Luisier, B., and Klimeck, G., Network for Computational Nanotechnology (NCN), Electrical and Computer Engineering **First Time User Guide to OMEN Nanowire**https://nanohub.org/resources/5359. Luisier, M. and A. Schenk, A., 2008. Atomistic Simulation of Nanowire Transistors, Journal of Computational and Theoretical Nanoscience, vol. 5, pp. 1031-1045. Luisier, M. and Klimeck, G., 2008. Purdue e pubs, OMEN an atomistic and full-band quantum transport simulator for post-CMOS nanodevices, paper#142. Luisier, M. Klimeck, G., Schenk, A., and Fichtner, W., 2006. Atomistic Simulation of Nanowires in the sp3d5s∗ Tight-Binding Formalism: from Boundary Conditions to Strain Calculations”, ETH Zurich, Switzerland: Phys. Rev. B, vol. 74, pp.205323. Sato, S., Kamimura, H., Arai, H., et al., 2010. High-performance Si nanowire FET with a semi gate-around structure suitable for integration, SolidState Electronics, vol.54, pp. 925–928. Singh, N., Lim, F.Y., Fang, W.W., et al., 2006. Ultra-narrow silicon nanowire gate-all-around CMOS devices: Impact of diameter, channelorientation and low temperature on device performance, IEDM, pp.383–386. Suk, S.D., Lee, S.Y., Kim, S.M., et al., 2005. High performance 5 nm radius twin silicon NWMOSFET (TSNWFET): fabrication on bulk Si wafer, characteristics, and reliability, Washington, USA: IEDM, pp.717–720. Tachi, K., Casse, M., Jang, D., et al., 2009. Relationship between mobility and high-k interface properties in advanced Si and SiGe nanowires, IEDM, pp.313–316. Tian, Y., Huang, R., Wang, Y., et al., 2007. New self-aligned silicon nanowire transistors on bulk substrate fabricated by Epi-free compatible CMOS technology: Process integration, Experimental characterization of carrier transport and low frequency noise, IEDM, pp. 895–899.

[254]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Detection of Software Cloning by using Visual Detection Technique Harish Patidar*, Amit Mishra, Shiv Kumar Department of Information Technology, TIT Bhopal, India

Abstract Modern age is based on program re-usability, about 10% to 30% of software programs contain repeated code, and the primary goal of software Program is to provide best quality software in the limited period. This project use visual detection technique to find the code clone. Here use the two different programs at different location, trace duplication. Application is web that is found out. First program is called sample (source) program and second program is target (test) program. First exact clone (type-I) clones are find with normalization and filter method. For other types of clones (Near-miss), token is generated and token is converted in to flag. This project also detects type-2, type-3 and type-4 clone with the help of Visual detection technique find level of Plagiarism. In this technique source and target program code convert into token. Each source token is compare with target token, it code is same then flag is zero otherwise clone. Software plagiarism is removed by this method. It is imported to measure the original quality of software.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keyword:-Code clone; Web server application; Visual detection; Token and flag

1. Introduction In computer, a program code is used as Duplicate code in another program by copy and paste (Roy, Cordy, 2007; Rainer Koschke, 2012; Davide Falessi, 2013), this duplicate code is called clone code. The working of clone code is similar as originally code. For example if we want to make a clone of a person, we create clone body, voice, face, hair. Clone is similar to person in body, voice, face and hair but habit and mind power is not match means clone is functionally same but logically different. Mainly four types of cloning: (Chanchal Roy, 2010; Bellon et al, 2007).   



Type-1.(a) Difference whitespace. (b) Changing in Comments. Type-2.(a) Renaming of identifiers. (b) Renaming of operator. Type-3.(a) Modification code. (b) Add new lines. (C) Delete some lines. Type-4.(a) Reordering Statements. (b) Control displacements.

Automatic process of finding duplicate in source code (Chanchal Kumar Roy, James R. Cordy, 2007) is called clone detection. In two programs clone codes are similar (Salwa Abd-El-Hafiz, 2007; Rysselberghe, 2010) in following manner.  

 

Code character-to-character similar. Code character-to-character similar with comments and white. Codes are token-to-token same. Codes are functionally same.

* Corresponding author. Tel.: +91 9617 201628. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Patider et al/ COMMUNE-2015

  1.1.

Codes are logically same. Identifier and variable are same

How Duplicates are Created

Copy and paste is the best way to create the clone code. In object oriented programming code reuse property is apply, level of piracy is increase by this property. Functionality is the way to create cloning. Some codes are functionally identical but to change in declaration of variable (Aaron Bloomfield, 2005; Schugerl et al, 2012). Mainly personal used software is implemented to duplicated code, level of piracy is increase day by day because code re-usability is do in maximum software. Reusable code is error free and reduces programming time (Chanchal Roy, 2010). Find the several reusable codes, calculate the actual programming effort. Two duplicated codes are functionally same but nature and characteristics are difference. Behaviour (Object oriented nature) different. 2. Literature Survey “A Metrics-Based Data Mining Approach for Software Clone Detection” It is metrics based, metrics are collected for all functions in the software system. A data mining algorithm, fractal clustering, is then utilized to partition the software system into a relatively small number of clusters. It is only find out type-1,type-2 and type-3 clones ,type-4 is not detected. “Detecting Software Clones Using Association Rule Mining”, detection of software clones is based on frequent item set of data mining. First step is data reprocessing, which find structural clones. The second step is to find frequent item sets using Apriori algorithm, Find out the clones and method clones by clustering. This algorithm is find type-1 and type-2 clones. “Evolution of Near-Miss Clones” investigates in which ways the evolution of near-miss clones differs from the evolution of identical clones. By analysing seven open source systems we draw comparisons between identical and near-miss clones. Evolution of near-miss is complex technique and not to find each sub- type of clones. D.Gayathri Devi and Dr. M Punithavalli, 2013 this paper study is based on software detection with the help of association mining. Finding frequent item of clone data, firstly preprocessing method is applied. The second way to apply Apriori algorithm, Find out the frequent item set. This method is only finding type-2 clone. It is time consuming method. Comparison of frequent item set is complex method. Rainer Koschke, 2012, explain the suffix tree for scalable comparison. It is improvement of index based technique. Suffix tree is used for compare similar data type. Calculation is done between inner system and subject system. Suffix tree is generated of each token. It is not handle type-3 clone, because tree is change in change in code. In this case construction and comparison is complex. Shaheen Khatoon and Azhar Mahmood, 2011 Studies about clone detection tool like CC Finder, Davide Falessi,2013 it is a detection tool , find copy past data in particular platform. Doppel code, Christopher Forbes, Iman Keivanloo, Juergen Rilling, 2012, it is differentiate global and local clone. Find selective type data. Some other mining technique like (Salwa K, Abd-El-Hafiz, 2012) clustering, association, Patterned matching method (Saman Bazrafshan, 2012), used finding duplicity code. Survey report show that all technique not finding minimum number of code clone. We find some problem in survey, which we try to remove in this paper and finding the better solution of clone detection. 3. Problem Identification Many clone detection method to find simple cloning, variation in code is not find-out properly. At present time most of clone detection technique to detect only copy past code. Base paper (Rainer Koschke, 2012), technique is only find Type-1 and Type-2 code clone. Accuracy level range is 30-40%. Traditional methods (Salwa K, Abd-El-Hafiz, 2012) do not solve the problem of server side clone detection and not to provide application of web. In object-oriented application (Schugerl, 2012; Saman Bazrafshan, 2012), code reuse property is used inheritance, polymorphism, encapsulation, and those features to growing up the cloning. Logical change in code is not detected by traditional method. Most of the method not to support multi language program code detection. If the base code is error full then it’s all clone code is contain bug. In old approach long code size clone is not measure at accurately. It is also find the misuse of reusability property in software field. New programming effort is reduce and quality of code programming is down. 4. Objective In this paper, automatic clone detecting technique is used, which is based on visual detection technique to provide maximum number of copy-past. Firstly present technique for structural clones, by lexical analysis. It is to find type-1, type-2 & type-3 clone detection. Data stored into. Detection technique is based on C++ and GUI. Data evaluated the performance by analysing structural clones found in software systems. On detecting code clones of code fragments, it [256]

Patider et al/ COMMUNE-2015

saves comprehension time and space. It is believe that this technique is scalable and useful. On detecting code clones, the quality of code is improved. This tool detects a significant amount of code clones. Identity and subsequent uniform of simple clones is helpful in software management. Its main goal is identify clones and quantify the amount of similarity present. 5. Proposed Methodology In this Approach Visual detection method (Token based and Normalization based techniques) used to find out quantity of copy-past. Data stored into SQL/MYSQL. Detection technique is based on C++ and GCC compiler. Clone detection method to find code clone between two program, one is sample program other is server related web application. In this way it can find (Forbes et al, 2012) the maximum similarity between two programs. Comparison between the two programs is show the level of originality. Two programs which are belong to two different addresses, to compared at the one platform and find the level of cloning. 5.1.

Proposed Block Diagram

Firstly take source and target programs, normalized the both codes and remove white space, comments. Then filtration is second steps for removing regular language key and header files, then compare. Tokenization: In the tokenbased approaches (Chanchal Kumar Roy, James R. Cordy, 2007; Filip Van Rysselberghe, Serge Demeyer, 2010), each line of the source code is converted into tokens with the help of lexical rule of the programming language. First, each line of source files is converted into I/P

Local Program

Call server web application

Tokenization

Code converts Into Token

Search Comparison token of both program No clone

Compare

Flag = 0

Flag generated

Flag = 1 Find Clone

Match

Fig. 1. Block diagram of Near miss clone detection

Tokens by a laxer and the tokens of all source lines are then concatenated into a single token sequence. Example: Input Source Code, If (p>=q) q=20; Generated Token is: Flag-based Techniques: This is search based technique (Rainer Koschke, 2012), token as input to symbol that is used for matching similarity, after search value of flag is zero that means no cloning, if value of flag is one then clone code is find out. Traverse & Select methods are used to find the appropriate item set. Then flag is used as input for visual detection, to find the code clone. Flag search period is depend on flag variable which is contain limited time depend on user. 5.2

Algorithm for Near miss clone detection

Step1: As an input to take sample program. Step2: Take target program.

[257]

Patider et al/ COMMUNE-2015

Step3: normalize both the source codes by removing the comment and white space step4: filtrate the source code obtained after normalization removing the elements of the code such as void main (), header files, getch () etc. Step5: Tokens are generated both programs code. Step6: i) Compare tokens between both programs. Compare all source tokens with target tokens and find the new token which not present in source. ii) Some token which is not find in target program. iii) Find change in token sequence in source to target tokens. iv) Change in control statement (for, while) with respect to source to target. Step7: Flag is generated. If (Flag = 1); Code clones are finding then generate chat wizard and level of piracy. Else Flag =0; No cloning. Step8: stop 5.3 Flowchart of Near miss clone detection

Start

Input Machine Take Local Program

Put Target

Normalized and Filtration (comments , header file)

Normalized and Filtration (comments, header files,

Token generate

Token generate

Comparison between token (search for similarity)

Flag generated

False (Flag=1) IF (Flag = 0)

True

Show clone

)

program

No clone

Stop

Fig 2. Flow chart of visual detection technique

[258]

Patider et al/ COMMUNE-2015

6. Conclusion Existing methods for evaluating (and comparing) clone detection tools suffer from several limitations. Visual Detection presented an evaluation technique that uses Near-Miss Token generation method to accurately measure of clone detection. The Visual Detection is based on Near-Miss method that is used to synthesize artificial clones by mimicking developers’ typical editing activities in clone creation. The Visual Detection is capable of evaluating and comparing recall of clone detection tools for C++ language and clone types with no need for manual intervention. In future accuracy can increase by a fitness function creation as well as this technique is limited due to single language, further it can improved by adding other language library also. References Devi D. G. and Punithavalli, M., Jan. 2013. Detecting Software Clones Using Association Rule Mining , Volume 3. Shaheen Khatoon and Azhar Mahmood, 2011. An Evaluation of Source Code Mining Techniques, Eighth International Conference on Fuzzy Systems and Knowledge Discovery (FSKD), Page No.234-246. Chanchal Kumar Roy and James R. Cordy, 2007. ClonesA Survey on Software Clone Detection Research, PageNo.1-115. Christopher Forbes, Iman Keivanloo and Juergen Rilling, 2012. Doppel-Code: A Clone Visualization Tool for Prioritizing Global and Local Clone”, IEEE 36th International Conference on Computer Software and Applications, PageNo.366-368. Rainer Koschke, 2012. Large-Scale Inter-System Clone Detection Using Suffix Trees”, 16th European Conference on Software Maintenance and Reengineering, pp. 309-318. Yuehua Zhang, Ying Liu, Lingling Zhang and Yong Shi, 1998. A Data Mining Based Method Detecting Software Defects in Source Code” in ICSM ’98: Proceedings of the International Conference on Software Maintenance, Page No. 368–377. Alexander Breckel, 2012. Error Mining: Bug Detection through Comparison with Large Code Databases, Zurich, Switzerland, page No.175-178. Salwa K. and Abd-El-Hafiz, 2012. A Metrics-Based Data Mining Approach for Software Clone Detection, 36 th International Conference on Computer Software and Applications, PageNo.35-42. Yoshihito Higo and Shinji Kusumoto, 2012. How Often Do Unintended Inconsistencies Happen? Deriving Modification Patterns and Detecting Overlooked Code Fragments, 28th IEEE International Conference on Software Maintenance (ICSM) Bellon, S., Koschke, R., Antoniol, G., Krinke, J. and Merlo, E., 2007. Comparison and evaluation of clone detection tools,” IEEE TSE, vol. 33, no. 9, pp. 577-591. Filip Van Rysselberghe and Serge Demeyer, 2010. Evaluating Clone Detection Techniques”, PageNo.1-12. Chanchal Roy, 2010. A Mutation / Injection-based Automatic Framework for Evaluating Code Clone Detection Tools”,The 9th CREST Open Workshop Code Provenance and clone Detection. Piatetsky-Shapiro, Gregory , Discovery, analysis, and presentation of strong rules, in Piatetsky-Shapiro, Gregory; and Frawley, William J.; eds, 1991. Knowledge Discovery in Databases, AAAI/MIT Press, Cambridge. Aaron Bloomfield, 2005. Scanning, ppt 1-32. Wu Zhifei ,Wang Tie, Zhang Qinghua, GaoTingyu and Li Hongfang,, 2010. Research on Generating Detector Algorithm in Fault Detection, page 23-26. Philipp Schugerl, Juergen Rilling and Philippe Charland, 2012. Reasoning about Global Clones, 35th IEEE Annual Computer Software and Applications Conference, PageNo.486-491. Davide Falessi, 2013. Empirical Principles and an Industrial Case Study in Retrieving Equivalent Requirements via Natural Language Processing Techniques, IEEE TRANSACTIONS ON SOFTWARE ENGINEERING, VOL. 39, NO. 1, Page No.18-42. Taeho Kwon and Zhendong Su, 2011. Modeling High-Level Behaviour Patterns for Precise Similarity Analysis of Software, 11th IEEE International Conference on Data Mining, DOI 10.1109/ICDM.2011, 104, Page No.1134-1139. Saman Bazrafshan, 2012. Evolution of Near-Miss Clones”, IEEE 12th International Working Conference on Source Code Analysis and Manipulation, Page.No.74-83.

[259]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Support Vector Machine based Multi-Unit Iris Biometric Verification using GLCM for Feature Extraction Shoaib Amin Banday*, Ajaz Hussain Mir Department of Electronics and Communication Engineering, National Institute of Technology Hazratbal Srinagar, India

Abstract Among all the biometric traits that are used for the authentication purposes in the state of the art security systems, Iris gives a superb recognition performance. The reason being that iris has naturally a very complex textural pattern which is almost perfectly established at a very tender age. Other reason being that Iris is stable i-e., it pattern does not change with age. This work is an extension of our previous work wherein we have proposed and implemented a multi-unit iris biometric system using GLCM for feature extraction and Euclidean Distance measure for classification. Here in this work we have implemented the same with a more powerful and versatile classifier Support Vector Machine (SVM). In this paper we have fused matching scores obtained by the use of SVM from left and right iris of a person using the gray level co-occurrence matrix (GLCM) for textural feature extraction. From this proposed fusion methodology, we have found that there is a slight enhancement in the overall accuracy of the biometric authentication system compared to our previous one. The performance is evaluated using Specificity, Sensitivity and the accuracy over a total subject number of 110 which are randomly selected from CASIA-iris-V4 thousand database.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Grey Level Co-Occurrence Matrix ; Multi-Unit Biometric Fusion; Sensitivity; Accuracy; Specificity, Recognition Rate; Support Vector Machines.

1. Introduction Biometrics is the use one’s physiological or behavioral characteristic for the purpose of identification (one-to-many comparison) and verification (one-to-one comparison). Biometric technology is used in a vast number of applications, ranging from small ones (like providing attendance of the employees in a small company or organization) to massive ones (like making sure that there is no compromise on the integrity of a huge number of voter registration database). The advantage of deploying a biometric system will surely depend on the application so as to provide increased security, increased convenience, reduced fraud or delivery of enhanced services etc. Regardless of any logics for implementing a biometric system, there are commonly two elements: 1) The large number of advantages and benefits from the implementation of the biometric systems are derived from the fact that they use a high level of certainty regarding an individual’s identity. 2) These benefits in turn leads directly or indirectly to cost savings and minimized risk of financial losses for an organization. The biometric systems utilizing only single biometric trait (e.g., fingerprint, iris hand geometry, face, handwriting, vein pattern, signature, voice, nailbed, gait etc) for recognition are known as Unimodal biometric systems. Unimodal biometric systems however face some stern limitations like unaccepted error rates, thus degrading both accuracy and performance of biometric recognition system. Other limitations like noise incorporation at the image acquisition time, ie., at sensor level, non-universality, spoof attacks, intra class variations, and distinctiveness were presented in the techniques given in (L.Wang and C.G Leedham; Wang et al; Watanabe at al; Ding et al). The use of single biometric trait for developing a biometric system makes the biometric system more vulnerable to the spoof attacks and thus * Corresponding Author. Tel.: +919796752852. E-mail address: [email protected] ISBN: 978-93-82288-63-3

Banday and Mir/COMMUNE – 2015

surely giving room for the identity compromises. These limitations posed by unimodal biometric systems can however be addressed by fusing information from two different biometric traits like fingerprint with face, fingerprint with iris, face with iris, signature with fingerprint, face with gait, iris with voice etc. Such biometric systems are known as multimodal biometric systems. The use of multiple modalities usually supresses the weaknesses of an individual modality and fuses the strengths of these modalities. Most biometric-based authentication systems consist of mainly four different modules as shown in fig.1.viz 1) Sensor unit 2) Feature extraction unit 3) Matching unit and 4) Decision unit. The image acquisition module is used to acquire the biometric data using a specific type of a sensor, the next module pre-processes the acquired data using various algorithms and mathematical tools followed by the extraction of a discriminative representation of the acquired data. This extraction of discriminative and subjective specific information is known as feature extraction. The features extracted corresponding to a particular person are stored in the database. The matching unit compares input features of the test subject to the stored templates and outputs a matching score corresponding to every template stored in the database. The matching can be done by numerous data classification techniques. Based on the matching score obtained, the decision module either outputs an “accept” or “reject”. The fusion in multimodal systems can be performed at four different levels (fig 1): sensor, feature, matching and decision. The sensor (Yang et al) and the feature levels (Ross et al), (M. Faundez-Zanuy) are referred to as a premapping fusion while the matching score (Cheng Lu et al), (Ribaric S et al) and the decision levels (Anil J., Arun R ), (Heikki A., Elena V) are referred to as a post-mapping fusion (C. Sanderson, K. K. Paliwal). In pre-mapping fusion, the biometric data from the two modalities is fused either immediately after the input data is acquired (sensor level fusion) or after the characteristic features from the two modalities have been extracted (feature level fusion).In other words the fusion is one before the classification while in post mapping fusion techniques the fusion is done after the classification as in matching level fusion and decision level fusion. Pre-mapping schemes which include fusion at the sensor and the feature levels have not been frequently used for the purpose of information fusion in biometrics that often whereas post-mapping schemes which include fusion at the match score, rank and decision levels are used more often. The amount of information that is available for fusion decreases as we move from sensor unit towards the decision unit in a biometric system, the post mapping fusion has grabbed major attention.

Pre-mapping

Post-mappin

Fig. 1: General Architecture Biometric System and variious levels of fusion .

The Fusion of multiple units of iris has been implemented in this paper which is a modification to our previous paper (Banday et al) where we have formed a biometric system that made use of Euclidean distance as a classifier. Here in this work, we have made a successful attempt by making use of a Support Vector Machine (SVM) classifier. The main advantage of using an SVM classifier is that it is more powerful than most of the state of the art data classifiers. The other advantage being that multi- unit biometric fusion require single sensor for left and right iris image acquisition. The feature extraction algorithm required is also the same for the two thus minimizing the complexity of the overall system. The use of different biometrics for fusion demand scores normalization which is not the case here, thus adding more to the simplicity of the system. This paper is organized as follows. Section 2 gives an overview of gray level co-occurrence matrix for iris feature extraction and SVM classifier. The proposed fusion framework is explained in section 3. Section 4 shows the experimental results obtained from proposed fusion scheme. Conclusion is given in section 5. 2. Overview of GLCM for Feature Extraction and SVM for Classification The verification of a subject using a biometric system is basically a recognition or classification problem which can be solved by the proper use of the features extracted from the object under consideration. The more relevant and discriminant features from the objects result in a better recognition rates and classifications. The feature [261]

Banday and Mir/ COMMUNE-2015

extraction aims to locate features which significantly represent the various regions of an object. The feature may be local or global giving distinguishable shapes, intensities and textures. The features that we have extracted in this work are based on the texture of an image. Since the modalities that we are using for fusion are left iris and right iris, the use of texture proved to be a better option because of the textura lly rich structure of the iris. Texture is a group of statistical measurements that are calculated from an image to quantify the visual texture, also giving the spatial dependence and arrangement of a region locally with the other. The gray tone spatial d ependence was first used by Julesz for texture classification. Haralick Rosenfeld and Troy proposed two - dimensional spatial dependence of gray tones in a co- occurrence matrix. There are numerous methods that have been proposed for quantifying these sets of textures in image like Gabor filters, co-occurrence matrix, and wavelet analysis etc. Among these methods for determining the texture, GLCM based measurement is the most famous, common and effective statistical method. Haralick first proposed GLCM for texture descriptions in the 1970s. Because of its sound potential, elegant performance it is still remains one of the popular methods until today. GLCM is a second order statistics method which describes the spatial interrelationships of the gray tones in an image (Banday et al). It contains elements that are counts of the number of pixel pairs, which are separated by certain distance and at some angular direction. Typically, GLCM is calculated in a small window (the soft texture of iris makes it more appropriate for feature extraction), which scans the whole image. Before texture estimation, we normalize the GLCM and let GLCM represent probabilities instead of pixel counts. The co -occurrence matrix is constructed by the joint probability density function between the gray level tones which gives the spatial relationship among any two points in the image [16]. It is denoted by P(i,j,d,θ), where i and j give ith line and jth column of co -occurrence matrix respectively, d is the distance between any two points and θ is the direction. Normalization involves dividing by the total number of counted pixel pairs. There are eight texture features based on GLCM as studied by Haralick. These are correlation, entropy, contrast, dissimilarity, Homogeneity, Angular Second Moment, inverse difference moment, GLCM mean (μi, μj), and variance (i2, j2). The classifier used for the classification purpose in this work is Support Vector Machine (SVM) which was introduced by Vapnik (1995) and Cortes and Vapnik (1995). The reason being its competitive performance in the field of pattern classification. It is a versatile pattern classifier that is trained to separate the input feature vectors into two classes (in case of two class classification) with an optimal separating hyperplane. The input set of vectors are said to be optimally separated by the hyperplane, if they are separated without any error and the distance between the closest vector to the hyperplane is maximum.

Fig. 2. Choosing an optimum hyperplane

The Fig. 2 shown above gives three hyperplanes viz: H1, H2 and OH to separate the two classes. All the three hyperplanes are able to separate the two classes but OH (Optimal Hyperplane separates the two classes in a way that it maximizes the margin between the vectors (samples) and the hyperplane thus minimizing the chances of making an error in the classification. 3. Proposed Methodology The iris database used in this work is CASIA-iris-V4 database. This database of iris consist of both left and right irises of different ages. As shown in Fig. 3, the left and the right iris images of a same subject are pre-processed and the Gray Level Co-occurrence Matrix (GLCM) are computed from these images. Haralick gives a textural feature extraction using 9 features as mentioned in the earlier section. After these 9 features have been computed (correlation, entropy, contrast, dissimilarity, Homogeneity, Angular Second Moment, inverse difference moment, GLCM mean (μi, μj), and variance (i2, j2)) at four angles of rotation (0o, 45o, 90o and 180o) from the both iris images in parallel, we reduced the dimensionality of the feature vector by omitting the features that do not contribute significantly. So from a feature vector of size 36, we use only 12 features for forming a template in the database corresponding to a particular subject. The feature vector after the dimensionality reduction consist of Correlation, Contrast and Homogeneity over four angles of rotation. The features extracted from the subjects are divided into two groups: data for training SVM and data for testing SVM. In the training phase SVM is trained that means this phase is used to form an [262]

Banday and Mir/COMMUNE – 2015

optimum hyperplane between any two different classes. In the testing phase, test image (iris) is similarly processed to obtain a textural feature vector. This feature vector of the test image is classified by already trained SVM into its respective class. The classification accuracy from each is taken matching score1 and matching 2. These scores are fused by the sum rule of fusion and a final decision (verification) is made on account of the pre-defined threshold which determines whether a test sample was a genuine one or an imposter one. Left Iris

Right Iris

Fig 3. Proposed Methodology for multi-unit iris fusion using SVM for classification

4. Results and Discussion Since fusing multiple biometric modalities or combining multiple units of same biometric modality is a good way of improving the parameters like Recognition Rate, Sensitivity, Specificity, FAR and FRR. The results obtained in this work show that fusion of multiple units (left and right iris) of the same biometric trait help us achieve much better results in terms of the above parameters compared to the results obtained from any Unimodal systems. Here in this section we have showed the effect of using a better and powerful classifier in the success of a biometric system. The Receiver Operating Characteristic (ROC) parameters like sensitivity and specificity show the overall accuracy of the proposed system. The accuracy is just slightly better than what was achieved in our last work where we had used Euclidean distance as classifier. The Specificity and the Sensitivity are calculated using the confusion matrix. The proposed biometric system obtains an accuracy of 99.1 in comparison to 98.71 of the previous work. Table 1 shows the recognition rates obtained when unimodal left iris, right iris, bimodal left and right iris with Euclidean distance is used. The table shows a clear but little improvement in the recognition rate using GLCM with SVM. Sensitivity = True Positive / True Positive + False Negative = Correctly Selected / Correctly Selected + Mistakenly Rejected = 55/55+0=1 Specificity = True Negative/ True Negative + False Positive = Correctly Rejected / Correctly Rejected + Mistakenly Selected = 55/55+1=55/56=0.9821 Accuracy = (True Positive +True Negative)/ (True Positive+ False Positive + True Negative + False Negative) = (Correctly Selected + Correctly Rejected)/ (Correctly Selected + Mistakenly Selected + Correctly Rejected + Mistakenly Rejected) = (55+55)/(55+0+55+1) =0.9910 = 99.10%

[263]

Banday and Mir/ COMMUNE-2015

Table 1. Recognition rates versus Modalities

Modality Left Iris (unimodal) Right Iris (unimodal) Fused left and right iris (bimodal) Proposed Method

Threshold 30

Recognition Rate 94.6% [17]

30

93.2% [17]

30

98.71% [17]

Linear Kernel Function

99.1%

5. Conclusion In this paper we have proposed GLCM based textural feature extraction for biometric fusion of multi-unit iris at matching score level using Support Vector Machines (SVM) for classification. The matching score is first calculated separately for left and right iris. The recognition rate is found to be 94.6 % and 93.2 % for left and right iris recognition system separately. The matching score obtained from left iris and right iris after SVM is implemented and then combined using sum rule of fusion and the results are compared with Unimodal left iris recognition system and right iris recognition system on the basis of recognition rate. The recognition rate after fusion of left and right iris matching scores using SVM is 99.1 % when a linear kernel function is used. The experiments are carried on CASIA-iris-V4 thousand database and the GLCM feature extraction algorithm and SVM implementation is done in MATLAB 7. References L.Wang and C.G Leedham, “A Thermal Hand Vein Pattern Verification System”, In Lecture Notes in Computer Science, Springer, 2005. Zhongli Wang, Baochang Zhang, Weiping Chen, Yongsheng Gao, “A performance Evaluation of Shape and Texture based methods for vein recognition”, DOI 10.1109/CSIP.2008.106 – 978-0- 7695- 3119/08- IEEE – 2008. Watanabe, M., Endoh, T., Shiohara, M, and Sasaki, S., “Palm Vein Authentication Technology and Its Applications”, In Biometric Consortium Conference, USA, 2005. Ding, Y., Zhuang, D. and Wang, K. “ A study of Hand Vein Recognition Method”, IEEE International Conference Mechatronics and Automation, pp. 2106- 2110, 2005. C. Sanderson, K. K. Paliwal. Information Fusion and Person Verification Using Speech and Face Information. IDIAP-RR, pp.02-33, 2003. F. Yang, M. Paindavoine, H. Abdi, D. Arnoult. Fast Image Mosaicing for Panoramic Face Recognition. Journal of Multimedia, Vol. 1(2), pp.14-20, 2006. A. Ross, K. Nandakumar, A. K. Jain. Handbook of Multibiometrics. Springer, 2006. M. Faundez-Zanuy. Data fusion in biometrics. IEEE Aerospace and Electronic Systems Magazine, Vol. 20, pp.34-38, 2005. Cheng Lu, Jisong Wang, Miao Qi, “Multimodal Biometric Identification Approach Based on Face and Palmprint ”,in Proc. Of Second International Symposium on Electronic Commerce and Security, vol 2, pp 44 – 47,2009. Ribaric S, Ribaric D, Pavesic N, “Multimodal biometric user-identification system for network- based applications”, in IEEE Proc. of Vision, Image and Signal Processing, Vol 150, Issue 6, pp 409 – 416,20 03. Anil J., Arun R., Score normalization in multimodal biometric systems, Pattern Recognition 38, 2005, pp. 2270-2285. Heikki A., Elena V., Soft biometrics combining body weight and fat measurements with fingerprint biometrics, Pattern Recognition Letters 27, 2006, pp.325-334. B. Julesz, “Visual pattern discrimination,” IRE T M In~form, Theory, vol. 8, no. 2, pp. 84-92,Feb.1962. R. M. Haralick, “A texturecontext feature extraction algorithm 1241, Mar. 1973. for remotely sensed imagery,’’ in Roc I971 IEEE Decisim and ConfrolConf. (Gainde, FL), pp. 650-657,Dec. 15-17,1971. A. Rosenfeld and E. Troy, “Visual texture analysis,” Tech. Rep.562-569, May 1.971.Atso in Confmnce Record for Symposium on Feature Extra-70116, University of Maryland, College Park, MD, June 1970.tion and Selection in Pattern Recognition, Argonne, IL, IEEE Publication 7OC-51C, Oct. 1970,pp. 115-124. Haralick, R.M., Shanmugam K. and Dinstein I., “Textural Features for Image Classification,” IEEE Transactions on Systems, Man and Cybernetics, SMC 3(6), pp.610-620, 1973. Banday, S.A.; Mir, A.H.; Khursheed, F., "Multi-unit iris biometric fusion using gray level co-occurrence matrix features," Advanced Electronic Systems (ICAES), 2013 International Conference on , vol., no., pp.225,229, 21-23 Sept. 2013.

[264]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

FloSwitch Board Design using Multi FPGA Mursal Ayuba, Jagannatham V. V b, Rajshekharc* a

Model Institute of Engineering and Technology ( MIET), Jammu, J&K, India b National Aerospace Laboratory (NAL), Bangalore, India c Dayananda Sagar College of Engineering, Bengaluru, India

Abstract The Flosolver Lab was started in 1986 to “Build a parallel computer for fluid dynamics applications”. Since then, different generations of parallel systems (Mk1 to Mk8) and inter-processor communication devices have been built. The communication between the processing elements is very important in parallel processing system. FloSwitch is the communication device, designed and developed indigenously to meet the requirements. The present high communication reconfigurable FloSwitch is Virtex 5 FPGA based to meet the application requirement. There is always scope of improvement of the communication devices. Here new design of FloSwitch is taken up to improve its communication speed, data handling of the parallel system (big data) with more reliable & efficient way and to cut down the no. of hops in the big system for the data transfer. To meet the requirement of the enhanced FloSwitch four FPGAs (Virtex-6) are used. The design will have the more processing elements (64) connected to the one device compared to the earlier device with 16 processing elements. This will bring down the number of hops between the processing elements. The DPM on board will be replaced by Block RAM resources of the four FPGAs , thereby leading to lower power consumption, data transfer rate increase and reduced size for a multicore chip.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: FPGA; Block RAM; DPM; Xilinx; Floswitch.

1. Introduction Starting from the jolting bullock cart of yesteryears to supersonic planes of the 21st century the face of science and technology is not the same .Same is the case with parallel computing technology moving leaps and bounds in recent years. Data-Intensive Computing is an application which uses data in the parallel approach to process large volumes of data typically in terabytes, petabytes in size. High computational Background Study applications which spent most of their execution time to computational requirements will have small volumes of data. Whereas computing applications which require large volumes of data and spent most of their processing time to I/O and manipulation of data are known as high communication (Flosolver Team (a)). Earlier electronics designs were based more on the design concept & its feasibilities. Power requirement was as the part of the design. Current design trends are portable devices with highperformance and low power. Designing the low power boards to improve the performance is extremely challenging & demanding. Communication protocol is a formal description of the digital message formats and the rules for exchanging those messages in or between computing systems. Protocols may include signaling, authentication and error detection and correction capabilities to reduce the size of digital designs (xilinx.com). The industry trend over the last few years has been to move towards the use of high speed serial protocols for data transmission. A digital serial signal uses fewer pins to transmit high-speed data by increasing the clock rate at which the signals are sent (Flosolver Unit (a)). Communication network is the most demanding area in the upcoming technology development. Every day the revolution is taking to its new height to meet the demand in the market. This project mainly deals with design of FloSwitch using four low power Virtex-6 FPGA’s. ON board DPM’s will be replaced with inbuilt BLOCK RAM of FPGA. Replacing on board DPM’s with inbuilt RAM’s of FPGA will use less power (Havinga, 2000). Communication speed will be increased as the internal RAM’s access time is very less compare to the on board DPM’s (Flosolver Unit

* Corresponding author. Tel.: +91 9906436856. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Mursal et al//COMMUNE – 2015

(b)). In this design 64 optical links can be connected to the FloSwitch. As in the earlier Design only it supports to the 16 optical links. Here it will have support of four FloSwitch capabilities in one device. Intra cluster connectivity will can also be the optical link. This new concept aims to reduce the number of hops between the processing elements, so that processing speed increases. This may lead to reduce the area covered by four FloSwitches to just one since it generates the same number of optical links as by four FPGA’s. 2. Block Diagram 2.1.

Existing Design

The design of this proposed floswitch is based upon some limitations of the existing switch. So it is quite natural to first deal with the existing system and know wherein the changes are to computing intensive applications. Here the numbers, which force to think of, power consumption in the big systems. In the existing system of Flosolver MK8(Flosolver Team(b)), FloSwitch is designed using Virtex 5 FPGA (XC5VLX110T, FF1759), DPM (IDT70v658S) and optical links.Flosolver Mk8 is mainly in the form of 128 clusters. Each Cluster is of 8-processor system consists of 4 dual processor server boards with PCI based add-on card(Flosolver Team(b)) which is connected to the FloSwitch through 64 bit parallel bus for intra-cluster communication

Fig. 1. Existing system interface

Such 128 clusters are linked via 16 Flo-optilinks (optical link) of FloSwitch for inter cluster communication (Flosolver Team (b)). So the FloSwitch is the prime communication device across the Mk8 super computer.

[266]

Mursal et al//COMMUNE – 2015

2.2. Proposed Design While the multi FPGA chips are generally used for high intensive data handling calculations, will essentially enhance speed , reduce power as compared to earlier floswitches, and reduce the number of hops .This is achieved by employing Virtex 6 FPGAFPGA (XC6VLX550T, FF1759 ), using FPGA block memory instead of external DPM will reduce the data transfer rate, flexibility in the design and size of the board. Four FPGA's are arranged in a reconfigurable pattern in a manner that each FPGA can connect to every other FPGA via address, data bus and control signals. 3. Reconfigurable FloSwitch Expediency The proposed design consists of four Virtex-6 FPGA’s connected in a manner that each FPGA connects to every other FPGA via address (32bit) and data(64bit) bus in one to one. Here the design of the board is in such a way that all FPGA can read or write with other FPGA’s in parallel. This lends the design of the FloSwitch flexiblele in terms of use of Virtex-6. Internal resources such as block RAM etc. are used for the memory to to all other connected FPGA’s. In the existing design each cluster has one FloSwitch which has 4 parallel links for intra-cluster communication via PCI card interface and 16 optical links for inter-cluster communication.

Fig2. Existing Existing FloSwitch Design

Thus each cluster connects to next 16 clusters in the MK8 ,128 cluster supercomputer design .In the proposed design 80 optical links are available for both inter-cluster and intra-cluster communication. This entails that each cluster can connect to 80 processing elements (PE’s), As a result more processing elements can have the communication using one FloSwitch, it is like a single cluster. This will eventually reduce the number of hops that information carrying from one server other server, thereby leads to an improvement in the communication system as a whole. As shown in Fig1 a single rack consists of four dual core servers (XEON processor). Intra-cluster communication is done in parallel, while serial optical links are used to connect to corresponding clusters. The present system of FloSwitch can be upgraded with DPM's inside the FPGA [Virtex-6]. This will enhance the speed and reduce the area of the board while reducing the access time of the data from memory twice.This will be carried out while considering power reduction schemes.The power section will be addressed to improve further by bringing down the total power in the FloSwitch design. By using low power GTX optical transceivers, low voltage SRAM and PROM of the order of FPGA used considerable amount of power is reduced thereby enhancing the performance of the

[267]

Mursal et al//COMMUNE – 2015

system. Fig.1shown above explains how in the existing system intra-cluster and inter-cluster communication exists via parallel connectors (cables) and optical links respectively.

Fig. 3. Proposed FloSwitch Design.

4. Advantages over Existing Designs While in the proposed design, the multicore FloSwitch can be connected in two configurations. First configuration proposes the use of FloSwitch as a centralized device with all the processing servers surrounding it. This means that all the eighty optical links coming out from the FPGAs will be used to connect to the servers directly. This will lead to great improvement in speed of the FloSwitch design as one to one communication between large number of servers will be possible , which was never possible in the earlier design. The second configuration as shown in fig4 entails the use of one FloSwitch in a manner that, it can firstly connect to sixteen other FloSwitches via optical links and also with sixty four servers directly. This is one of the option to show the inter cluster connectivity with one hop. For example Server 1 of first cluster to server 63 of the sixteenth cluster can be communicated in one hop. Here the total number of servers connected using sixteen Floswitches will be 1024. Total processing elements will be 2048. Earlier the sixteen FloSwitches can connect one to one was 64 server board as 128 processing elements. Therefore, over all there is a gain of 16 times more PE’s can be connected. 4.1. Reduce the Number of Hops In the parallel computing scheme, we define a hop as the communication from one PE to other PE. In computer networking, a hop represents one portion of the path between source and destination. When communicating over the Internet, for example, data passes through a number of intermediate devices (like routers) rather than flowing directly over a single wire. Each such device causes data to "hop" between one point-to-point network Connection and another. While in the case of parallel computers, each FloSwitch encountered leads to a hop and a delay since the proposed design aims to replace four switches by just one, it entails reduction in number of hops encountered also.

[268]

Mursal et al//COMMUNE – 2015

Fig. 4. Proposed Floswitch interface with 64 servers’ direct links and 16 other floswitches.

4.2. DPM removal and Block RAM: In the existing design, one FloSwitch has 8 DPMs occupying a reasonable amount of space on the board. Four such switches carry 32 DPMs. The proposed design will use block RAM of the Virtex-6 FPGA instead, thereby reducing area of the board and also reducing the power consumed by the DPMs (Jagannatham, et al, 2013). Power consumed by one DPM is measured to be 1650 Mw. If we replace 32 DPMS, we are in fact saving 52.8 Watt of power consumption. This no. are useful as we grow system to a big parallel machine. Here the main advantage of the FloSwitch will be to have a one-type communication link as serial weather it is intra cluster or inter cluster communication This design has been under taken to study the feasibility of the communication and the advantage over making a big system for big data and intense communication applications. Table 1 Comparative analysis of Proposed and Existing Designs COMPONENTS

EXISTING SYSTEM 12.217 W(Virtex-5)

PROPOSED SYSTEM 13.645 W(Virtex-6)

SDRAM FLASH MEMORY SFP

12.17 × 4 = 48.6 W 528 mW 49.5 mW (SPI) 990mW

13.645 ×4= 54.6 W 63mW 54 mW (BPI)

TRANSCEIVERS CLOCKING CIRCUIT

990 × 64= 64 W 115.5 mW

FPGA

792mW 792× 64= 50 W 45 mW

5. Acknowledgement I would like to thank the Flosolver Unit, NAL, who cooperated with me in this endeavour. Our vote of thanks would be incomplete without the mention of Mrs. Rajalakshmy Sivaramakrishnan, HOD Flosolver Unit, and NAL. In addition, the college authorities deserve a special mention for they believed that we could go through with the new design. References Flosolver Team, NAL. A study of FPGA modules on FPGA based Floswitch, NAL PDFS 1009. Flosolver Team, NAL. Preliminary performance analysis of Flosolver Mk-8Flosolver, by PDFS 1017. Flosolver unit’s project report titled. Design of High Speed Communication Floswitch for Intra cluster communication using Parallel Connectivity. Flosolver Unit’s project report titled. Design of High Speed communication Floswitch for Intra cluster communication using Serial Connectivity. Havinga, Paul J.M., Smit, Gerard J.M., 2000. Design techniques for low power systems, Journal of Systems Architecture. 46(1), pp, 1-21. xilinx.com/support/documentation/data_sheets/ds150.pdf Jagannatham, V.V., and Rajalakshmy Sivaramakrisnan, 2013. Flosolver Unit’s project report titled: Power analysis of Low power Virtex-6 FPGA based ommunication Floswitch Design. VoISSN: 2278-0181.

[269]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

On the Realization of Robust Watermarking System for RGB Medical Images Shabir A. Parah*, Javaid A. Sheikh, Farhana Ahad, G.M. Bhat Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India.

Abstract Digital medical images when transferred over an unsecure medium like internet must be highly secured for the sake of tele-diagnosis and tele-medicine. In this paper a robust digital watermarking technique has been proposed for color digital medical images. The medical image is firstly divided in two parts viz. Region of Interest (ROI) and Region of non-Interest (RONI). Watermark has been embedded in Region of Non-Interest (RONI) in spatial domain so as to ensure that critical diagnostic data is not fiddled with. The watermark has been embedded in various 8 × 8 blocks using difference between the two predefined pixels of a given block to modify their relation. The experimental results show that the perceptual quality of the watermarked images is better. Further the scheme has been shown to be robust to various image processing attacks like Salt and Pepper, Gaussian noise, JPEG Compression, Sharpening and Rotation.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Tele-Diagnosis; Tele-Medicine; Spatial Domain; Transform Domain; ROI; RONI

1. Introduction In Medical Information Systems (MIS), paper based health records are replaced by Electronic Health Records (EHRs). Although, EHRs have many benefits like overall cost reduction, better health care facility, easy record keeping and flexibility. Still then five security issues; integrity, availability, authentication, confidentiality and non-repudiation are of huge concern. This has led to use of digital watermarking techniques in MIS. Digital watermarking has the potential to act as a significant tool for major security issues related to EHRs and especially digital medical images. The medical images can be classified in two different regions; Region of Interest (ROI) and Region of Non-Interest (RONI). Digital watermark is embedded in RONI keeping in consideration that the ROI has to be preserved with actual information, as tele-diagnosis is based on observing the ROI. In medical images reversible digital watermarking, a special kind of digital watermarking provides the copyright protection (Parah, et al., 2013). Reversible digital watermarking is highly robust watermarking technique. The main requirements of a watermarking scheme are robustness, imperceptibility, payload and security. Robust watermarking scheme are resistant to attack and provide high security whereas fragile watermarks are easily distorted indicating the tamper. For authentication applications fragile algorithms are preferred. Thus generally a tradeoff has to be worked for a particular problem (Shabir et al., 2012). Higher capacity can be achieved at the expense of imperceptibility and robustness/fragility (Shabir et al., 2014). EHR hiding watermarks demand high capacity, authentication watermarks identifies the source of the image whereas integrity watermarks and tamper detection watermarks are concerned with error correction codes. The literature survey of medical image watermarking system reveals that different types of data such as patients’ data; source information, indexing, and authentication information are embedded as watermarks. It is also noticed that many techniques lacks tamper detection and localization mechanism, which is one of the requirement that seeks attention in medical image watermarking. In the works of (Sudeb et al., 2013), a blind fragile watermarking algorithm has been proposed. Least Significant Bit (LSB) data embedding technique along with encryption and data compression technique have been used for different gray scale modalities. Baisa and Suresh, 2012 have proposed non blind robust digital watermarking technique for 512×512 medical images which has high computational complexity. The embedding takes place in transform domain. *

Corresponding Author. Tel: +91 9596 529991. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Parah et al/COMMUNE – 2015

For improving the quality of watermarked images (Lin et al., 2012) have reported a watermarking technique based on transform domain embedding. The scheme has been tested for gray scale images only. Chunhua et al (Chunhua, et al., 2012) have proposed a robust embedding scheme for multiple watermarks for medical image in transform domain. Even though image perceptibility is good but computational cost for implementing the scheme is high. Koushik et al (Koushik et al., 2012) proposed a spatial domain watermarking technique based on majority logic algorithm for gray scale medical images. The multiple copies of the same watermark is embedded. Various spatial domain techniques have been compared with their proposed version. Jingbing, et al. have presented a robust medical image watermarking system in transform domain (Jingbing, et al., 2011). The security of scheme has been enhanced using different keys. This system however requires multiple-key management. Most of the reported work on medical image watermarking uses gray scale images as cover media. Color images (such as endoscopic image, etc.) are now-a-days being frequently used for medical diagnosis. In such a scenario there is a tremendous need for color image watermarking. This paper proposes a robust RGB (color) image watermarking system. Rest of paper is organized as follows. Section 2 explains the proposed work. Results and discussions have been presented in section 3. The paper concludes in Section 4. 2. Proposed work Classical spatial domain embedding for ordinary gray scale or color images uses whole image for data hiding purpose. When dealing with the medical images the sensitive, most informative portion of the image; that is, Region of Interest (ROI) should remain untouched. The proposed scheme uses RONI for embedding watermark in various constituent planes of color image. The proposed scheme is implemented using 8×8 block wise DCT. The block size can be less but it has less opportunity for compression. The higher block size has very high computational complexity and has very low additional energy compaction as compared to 8×8. The overall system function has been explained in two subsection viz. watermark embedding and extraction. 2.1

Watermark Embedding The proposed watermark embedding flowchart is shown in Fig. 1. The algorithm for embedding is as under. 1. Read Cover Image 2. Separate constituent Red, Green and Blue Planes.

Fig.1. Watermark embedding flowchart.

3. 4. 5. 6. 7.

Divide each plane in 8x8 blocks. Find ROI and RONI of blue and green plane. For every RONI 8x8 block find difference between pixel P1{(5,2) th pixel} and P2{(4,3)th pixel} of every 8x8 block (Any other pixel of the same block can be taken). For embedding bit 0, relative difference is used to make P1>P2. For embedding bit 1, relative difference is used to make P1
[271]

Parah et al/COMMUNE – 2015

8. Step 6 and step 7 continues till all the watermark bits are embedded. 9. After embedding Concatenate ROI and RONI for obtaining constituent planes. 10. Obtain watermarked color image from concatenate RGB planes. 2.2

Watermark Extraction

For the extraction of the digital watermark from the watermarked image, the color watermarked image is separated into its three planes; red, green and blue plane. The RONI of each region is obtained. The two pixels (P1 = (5, 2) th pixel and P2 = (4, 3)th pixel) of each non-zero, non over lapping 8×8 block are calculated. And if the P1 >P2 then bit 0 is extracted else bit 1 is extracted. The separated bit sequence is moulded to actual logo size. 3. Experimental Results The experimental results have been computed using MATLAB R2014a platform for 512×512×3 medical images. The reason for choosing medical images is because of the importance of the tele-diagnosis and telel-medicine for remote analysis of patient. The digital watermark used for copyright protection is of the size 64×64. The mask for the ROI for medium medical images is created by means of Color Thresholder application of MATLAB R2014a. The image quality is established on objective analysis by calculating Peak Signal to Noise Ratio (PSNR) between original medical image and watermarked and over attacked image. This in turn is calculated by measuring Mean Square Error (MSE). Formula (1) and (2) is used for computing PSNR. The robustness of the proposed scheme is observed by inspecting Bit Error Rate (BER) and Normalized Cross-correlation (NC) between embedded watermark and extracted watermark for various attacks. Formula (3) is used for calculating BER and 4 for NC. MSE =

1 𝑀𝑁

𝑁 2 ∑𝑀 𝑙=1 ∑𝑘=1(𝑥𝑙,𝑘 − 𝑥′𝑙,𝑘 ) (2𝑛−1)2

PSNR = 10log⁡

𝑀𝑆𝐸

(255)2

= 10log⁡

𝑀𝑆𝐸

(1) dB

(2)

In the above formulae; M, N is the dimensions of the original image and the watermarked image; (x pixel value of original image and (x’j, k) is the (j, k)th pixel value of watermarked image. BER = NCC =

1 𝑀𝑁

𝑁 ⁡[∑𝑀 p=1 ∑q=1 wm(𝑝, q) ⊕ wme(𝑝, q)]

N ∑M p=1 ∑q=1 wm(𝑝,q)×wme(𝑝,q)

l, k)

is the (l, k)th

(3) (4)

N 2 ∑M p=1 ∑q=1 wm(𝑝,q)

In above formulae; M, N is the dimensions of the original logo and extracted logo; wm (p,q) is the (p,q) th pixel of original watermark and wme (p,q) is the (p,q) thpixel of the extracted logo. The robustness and perceptual analysis of the proposed technique is presented below.

3.1

Robustness Analysis

The robustness of the proposed system has been checked by subjecting watermarked images obtained from the system to various attacks like, salt and pepper noise, Gaussian noise, JPEG compression and rotation etc. Fig. 2 to Fig. 8 show the results obtained for different attacks. The robustness of the proposed scheme is compared with (Koushik et al.) for different attacks. It is clear from the Fig 2.

Fig. 2. BER comparision

[272]

Parah et al/COMMUNE – 2015

(a)

(b)

(c)

(d)

Fig. 3. (a)original medical image; (b) watermarked image [PSNR=44.1250 dB]; (c) original watermark; (d) extracted watermark [BER=0.00; NCC=1.0000]

(a)

(b)

(c)

(d)

Fig. 4. (a)original medical image; (b)Noise: Salt and Pepper(d=0.01) [PSNR= 23.2648dB]; (c) original watermark; (d) extracted watermark [BER= 0.0054; NCC= 0.9983]

(a)

(b)

(c)

(d)

Fig. 5. (a)original medical image; (b)Noise: Gaussian Noise(v=0.001) [PSNR= 31.4007dB]; (c) original watermark; (d) extracted watermark [BER= 0.0313; NCC= 0.9916]

(a)

(b)

(c)

(d)

Fig. 6. (a)original medical image; (b)Noise: JPEG compression(Q=90) [PSNR= 49.4291 dB: CR =29.5607]; (c) original watermark; (d) extracted watermark [BER=0.0400; NCC= 0.9761]

(a)

(b)

(c)

(d)

Fig. 7. (a)original medical image; (b)Noise: Sharpening [PSNR= 30.5140dB]; (c) original watermark; (d) extracted watermark [BER= 0.00: NCC=1.0000]

(a)

(b)

(c)

(d)

Fig. 8. (a)original medical image; (b)Noise: Rotation(5 o) [PSNR= 42.6560 dB]; (c) original watermark; (d) extracted watermark [BER=0.0017; NCC= 0.9980]

[273]

Parah et al/COMMUNE – 2015

3.2

Imperceptibility Analysis.

The perceptual transparency of the watermarked images obtained in the proposed scheme has been compared with that of has been compared with that Koushik, et. al. The comparison results have been shown in Table 1. It is evident from the table that the average PSNR of the watermarked images as obtained in our scheme is about 2.5 dB more than that of Koushik et.al. Table 1: Perceptual transparency comparison

Image Foot Brain Knee Hand Skull Average

Koushik, et. al. PSNR (dB) BER 40.8229 0.0633 41.2924 0.0556 42.2900 0.0446 41.4684

Proposed Scheme PSNR (dB) BER 44.1894 0 44.0752 0 43.4390 0 43.6432 0 44.1250 0 43.8943

The various image quality indices obtained when watermarked image is attacked are presented in Table 2. Table 2: Image indices obtained after various attacks Image used Skull Skull Skull Skull Skull Skull

NOISE ADDED No noise Salt and Pepper (d=0.01) Gaussian Noise (v=0.001) JPEG (Q=90) JPEG (Q=80) Sharpening Rotation (5o)

PSNR (dB) 44.1250 23.2648 31.4007 49.4291 48.5274 30.5140 42.6560

BER 0.0000 0.0054 0.0313 0.0400 0.1299 0.0000 0.0017

NCC 1.0000 0.9983 0.9916 0.9761 0.8944 1.0000 0.9980

4. Conclusion A robust medical image watermarking scheme based on RONI embedding in constituent bit planes of RGB image has been proposed in this paper. The embedding is done in spatial domain based on relative pixel magnitude relation in 8x8 block. 512×512 color images have been used for testing and comparing the proposed scheme. The spatial domain embedding used in the proposed technique ensures its hardware implementation at low cost. The experimental results reveal that our scheme is capable of producing good quality watermarked images besides being robust to various image processing attacks like sharpening, Salt and pepper noise, Gaussian noise and rotation etc. References Shabir, A., Parah, Javaid, A., Sheikh, G., Mohiuddin, Bhat, “On the realization of a Spatial Domain Data Hiding Technique based on Intermediate Significant Bit Plane Embedding (ISBPE) and Post Embedding Pixel Adjustement (PEPA)”, Proceedings of IEEE International Conference on Multimedia Signal Processing and Communication Technologies- IMPACT 2013 (AMU, Aligargh-23- 25 Nov. 2013) pp 51- 55. Shabir, A., Parah, Javaid, A., Sheikh, G., Mohiuddin, Bhat, “On the realization of a Secure, high capacity data embedding technique using joint topdown and down-top embedding approach,”, Computer Science and Engineering, 2012, vol:49, pp 10141-10146. Shabir, A. P., Javaid, A. S., A. M. Hafiz, G., M., Bhat, “A secure and robust information hiding technique for covert communication”,International Journal of Electronics, 2014, Taylor& Francis, pp. 1-13 Sudeb, D., Malay, K., K, “Effective management of medical information through ROI-lossless fragile image watermarking technique”, Computermethods and programs in biomedicine, 2013, page no. 662–675. Baisa L. G., Suresh N. M., “ROI Based EmbeddedWatermarking of Medical Images for Secured Communication in Telemedicine” World Academy of Science, Engineering and Technology, Vol:6 ,2012-08-21. Lin Gaoi, TiegangGaol, GuoruiShengi ,YanjunCaoi,LiFani, “A New reversible watermarking scheme based on integer DCTfor medical images”, Proceedings of the 2012 International Conference on Wavelet Analysis and Pattern Recognition, Xian, 15-17 July, 2012. Chunhua Dong, Jingbing Li, Yen-wei Chen, “A DWT-DCT Based Robust Multiple Watermarks for Medical Image” 978-1-4577-0911, 2012, IEEE Jingbing Li, Wencai Du, Yong Bai, Yen-wei Chen, “3D DWT- DCT Based Multiple Watermarks for Medical Volume Data Robust to geometrical Attacks”, 978-1-4577-0321-8, IEEE. Koushik, P., Goutam, G., Mahua, B., “A Comparative Study between LSB and Modified Bit Replacement (MBR) Watermarking Technique in Spatial Domain for Biomedical Image Security”, International Journal of Computer Applications and Technology (2278 - 8298) Vol: 1, 2012, pp. 3039.

[274]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Realization of a Fragile Medical Image Watermarking System for Content Authentication Shabir A. Parah*, Javaid A. Sheikh, Zahid Hussain, Syed Mohsin Department of Electronics and Instrumentation Technology, University of Kashmir, Srinaagar, India

Abstract Due to the unprecedented development of Electronic-health (E-health) sector, watermarking of medical images now a day’s is attracting a considerable amount of interest from the research community. Content authentication of medical images is a very important issue while transmitting a medical image for diagnostic purposes. In this paper, a fragile watermarking technique is proposed to manage the security and genuineness issues of a medical image. A Watermark of size 64×64 pixels has been embedded along the borders (Region of Non Interest) of a gray scale medical image of size 512×512 pixels. To enhance the security of the system RC4 stream cipher has been utilized for scrambling the watermark. The experimental results reveal that the proposed technique is capable of providing high quality watermarked images besides being completely fragile to different image processing attacks. A comparative study of proposed technique with an existing one shows that our technique is suitable for content authentication.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Fragile Watermarking; Spatial Domain; RC4 Encryption; Region of Interest

1. Introduction Due to the headway in field of communications and image processing, patients are treated remotely by the specialist, which includes the methodology of exchanging the medical images and reports between different wellbeing units. Watermarking which was basically devoted to multimedia documents has been found to be a suitable candidate for the field of Telemedicine due to many attractive features associated with it. Digital watermarking is a data hiding technique of inserting a secret data (watermark) into a host media. The main requirements of any data hiding technique are robustness, imperceptivity, capacity and security (Shabir et al., 2014a). The secret data can be embedded into a host media in two domains: spatial and transform domain. Spatial domain methods are easy to implement and have low computational cost. Transform domain methods are more complex as compared to their spatial domain counterpart but are more robust than spatial domain methods (Shabir et al., 2014b). The medical image watermarking can find applications in the following area as reported in (Memon. et al, 2008).   

Videoconferencing/teleconferencing of doctors relating to particular patient details. Multidisciplinary exchange of consultative ideas between various medical experts to discuss ailments and surgical therapies. For distant education of medical practitioners.

However the aforementioned reasons require additional persistence towards image realness (accessibility, unwavering quality, secrecy) (Coatrieux et al, 2007). To encourage imparting and distantly treating of medical pictures in a protected manner, watermarking guarantees attractive characteristics. This technique allows unending relationship of picture substance which checks its perceptuality by inculcating changes in the picture pixel values, regardless of its format (Coatrieux et al, 2006). With a specific end goal with respect to robustness of a watermarked medical image, three types of watermarking have been reported; Robust watermarking (Cox et al, 1997), Fragile watermarking * Corresponding author. Tel.:+91 9596 529991. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Parah et al. / COMMUNE-2015

(Raymond, 1999) and Semi-fragile watermarking (Hazem, 2014).The proposed paper work is principally concerned with the fragile watermarking. A digital watermark is called "fragile" in the event that it neglects to be distinguishable after the smallest change (attack).In our work we decrease the complexity by meeting expectations in the region which is of our least concern. We scramble the Watermark by utilizing a RC4 cipher to make it more secure. The scrambled watermark has been embedded in the border pixels of the cover image. The embedding in the borders of the image influences the image quality by a small amount, consequently giving us high Peak Signal to Noise Ratio (PSNR). More often, it is alluring to implant information outside the Region of Interest (ROI) to give better assurance without compromising on the diagnostic data. The method fills for both the needs of medical image verification and also we don’t need to trade off on the diagnostic information. 2. Literature Survey Diverse gatherings of authors have reported various medical image watermarking strategies. Few authors have worked in the spatial domain watermarking, in which the watermark is inserted into the host image by specifically adjusting a set of pixel values without bringing on clear changes in the host image. While as, now a day’s a large portion of the work is confined to other domain called the transform domain where the watermark is embedded into the picture by altering the transformed coefficients (Shabir et al., 2014c). (Wong, Memon, 2001) have proposed watermarking schemes for image authentication and verification. This technique can locate any adjustment made to the picture and demonstrate the particular areas that have been altered. The scheme is complex and offers a low quality watermarked images. (Archariya et al, 2003) proposed a watermarking technique where Electronic Patient Records (EPR) is embedded in the medical images. Disadvantage of this scheme is that watermark embedding is done without taking care of ROI, which is symptomatically a critical range in medical images. (Baisa, Suresh, 2012) proposed a ROI based watermarking technique for medical images. Watermarked image of low quality is obtained with this technique. An ROI based Medical watermarking scheme, wherein watermark has been embedded in transform domain has been reported in (Neha, Sanjay, 2014). The watermark is scrambled by RSA Technique and Discrete Wavelet Transform methodology is utilized for installing the encrypted information for improving information security. The scheme however is computationally complex due to use of both DWT and RSA. 3. Proposed Scheme Any medical image on account of clinical conclusion comprises of two sections, region of diagnostic information called as Region of Interest (ROI) and region of non-diagnostic information referred to as Region of non Interest (RONI). A typical ROI and RONI scenario is shown in Fig. 3.1. The proposed strategy chooses the RONI for inserting the watermark to guarantee the respectability of the ROI and not to trade off with the determination estimation of medical image. Before embedding, watermark is encrypted using RC4 cipher technique so as to ensure that an adversary cannot counterfeit the system. The encrypted watermark is embedded in the border pixels of the host image along the borders in the spatial domain. In order to ensure the fragility of the watermark, the watermark bits are embedded in Intermediate Significant Bits (ISB) in border pixels.

Fig. 3.1: Region of interest & region of non interest of a medical image

The proposed scheme consists of four broad steps: 1. 2. 3. 4.

Encryption of the watermark. Embedding scrambled watermark along the borders (RONI) of the host image. Extraction of watermark. Decryption of extracted watermark.

The extracted watermark is contrasted with the original watermark known to the receiver for subjective confirmation. For target approval, BER is processed for checking watermark uprightness. 3.1.

Encryption of Watermark

The encryption of the watermark enhances the security of the system. RC4 stream cipher (William, 2005), has been used for encryption of the watermark because of its simple structure and effective feature. After applying RC4 technique the watermark is randomised and is embedded in the host image.

[276]

Parah et al. / COMMUNE-2015

3.2.

Embedding the Encrypted Watermark

The embedding scheme has been depicted in Fig. 3.2. The cover image has been divided into ROI and RONI. The scrambled watermark is embedded into RONI (border pixels). The encrypted watermark, which is 4096 bits (64×64) in size, is embedded into appropriately chosen border pixels. In our scheme 2048 pixels have been selected and their first and second ISB’s of every pixel have been replaced by the encrypted watermark bits. When embedding is completed in RONI, it is then fused with rest of the image (ROI), thus resulting in a watermarked image.

Fig.3.2: Block diagram of embedding process

3.3.

Extraction of the watermark

The extraction is the converse of the inserting methodology. The extraction process yields us the encrypted watermark bits which must be decrypted by the assistance of a key which was used for the encryption process by the sender. 3.4.

Decryption of the watermark

The extracted watermark bits are decrypted by the receiver using the same cipher key as was used by the sender. With regard to decryption, the cipher text is XORed with the key stream used for encryption. 4. Results and Discussions The proposed technique has been implemented using MATLAB R2010a software working on Windows platform. The scheme has been evaluated in terms of objective image indices like, PSNR, and Bit Error Rate (BER). The PSNR and BER are defined as follows: 𝑃𝑆𝑁𝑅 = 10 𝑙𝑜𝑔10

2552 𝑁 2 𝑀×𝑁 ∑𝑀 𝑖=1 ∑𝑗=1[𝑎𝑖𝑗−𝑏𝑖𝑗]

dB

(1)

where 𝑎𝑖𝑗 refers to the value of the pixel in the host image and 𝑏𝑖𝑗 is the pixel value of the watermarked image at the same position and 𝑀 × 𝑁 is the size of the cover image. 𝑀

𝐵𝐸𝑅 =

𝑁

𝑤 ∑ 𝑤 [𝑊(𝑖,𝑗)⊕𝑊 ′ (𝑖,𝑗)] ∑𝑖=1 𝑗=1

(2)

𝑀𝑤 ×𝑁𝑤

where W(i, j) and W ′ (i, j) are inserted and removed watermark respectively with size Mw × Nw . The lesser the BER, better is quality of extracted watermark. The proposed scheme has been tested for gray scale medical image of size 512x512 pixels. The designed system has been subjected to various image processing attacks like filtering, compression, sharpening, rotation etc. to evaluate its fragility. The medical image SKULL has been used to demonstrate the subjective quality as obtained in our proposed scheme. Fig 4.1(a) and 4.1(b) shows the original image and the watermark for, skull, respectively. From Fig. 4.1(c) it can be seen outwardly that the nature of the watermarked medical picture has barely any change resulting in a high PSNR of 60.71dB.

[277]

Parah et al. / COMMUNE-2015

(a)

(b)

(c)

(d)

Fig. 4.1: (a) Original Medical Image (b) Original Watermark (c) Watermarked Image (d) Extracted Watermark

Fig. 4.2 shows the performance of the scheme to various image processing attacks. Watermarked image and the extracted watermark have been shown for different attacks.

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

Fig. 4.2: (a) Embedded Medical Image under median filtering attack [3x3] followed by extracted Watermark. (b) Image under Salt & Pepper attack (d=0.01). (c) Image under Histogram Equalization attack. (d) Image under Gaussian Noise attack (variance=0.001). (e) Image under JPEG Compression attack. (f) Image under Low Pass Filtering attack. (g) Image under Sharpening Attack. (h) Image under Rotation Attack (5 degree)

The PSNR and BER computed for different attacks as obtained in the proposed scheme are shown in the Table 4.1. It is clear from the observed BER values from Table that proposed technique is highly fragile to various attacks and hence a suitable candidate for content authentication. Table 4.1: Test Result for a Gray Scale Medical Image of Skull (512x512) Pixels. Attacks No Attack Median Filtering (3x3) Salt & Pepper (0.01) Histogram Equalization Gaussian Noise (0.001) JPEG Compression (90) Low Pass Filtering Sharpening Rotation (5o)

PSNR in dB 60.71 37.78 25.25 13.78 39.93 45.26 34.68 26.11 16.90

[278]

BER 0 0.3450 0.0049 0.4028 0.4836 0.4128 0.4614 0.5112 0.4985

Parah et al. / COMMUNE-2015

4.1.

Comparative Study

The proposed results have been compared with (Baisa et al, 2012) and the comparison results have been shown in Table 2. It is clear that our method is capable of producing high quality watermarked images compared to reference under comparison. Table 4.2.Comparison Results. Image

PSNR (Proposed Method)

PSNR (Baisa et al )

Brain CT

57.80

48.53

CT Scan (Head Slice)

59.08

42.52

Angiogram

58.19

48.46

5. Conclusion A fragile medical image watermarking technique has been proposed in this paper. The cover image is divided into ROI and RONI. A set of pixels is selected from RONI, and encrypted watermark (using RC4 algorithm) is inserted into the ISB’s of the selected pixels. Embedding scrambled watermark enhances the security of the system. The experimental results reveal that the system is capable of producing high quality watermarked images without losing critical diagnostic information due to RONI embedding. This can be appreciated from the fact that average value of PSNR obtained for medical images is approximately 58dB. Further, the attack analysis of the proposed scheme shows that it is completely fragile to all the image processing attacks performed on the watermarked image. As such the system is capable of authenticating the content. References Shabir, A., Javaid. A., Abdul, M., Ghulam, M., 2014a, A secure and robust information hiding technique for covert communication, International Journal of Electronics, p. 1-14. Shabir, A., Javaid. A., Ghulam, M., Abdul, M., 2014b, Data Hiding in Scrambled Images: A New Double Layer Security Data Hiding Technique, Computers and Electrical Engineering 40, p. 70-82. Nisar, A. M., Gilani, S. A. M., 2008, NROI watermarking of medical images for content authentication, IEEE Conference, Karachi, Pakistan, December 23–24. Coatrieux, G., Montagner, J., Huang, H., Roux, C., 2007, Mixed reversible and RONI watermarking for medical image reliability protection, 29th IEEE International Conference of EMBS, Cite International, Lyon, France, August 23–26. Coatrieux, G., Lecornu, L., Sankur, B., Roux, C., 2006, A review of image watermarking applications in healthcare”, Proceedings of IEEE-EMBC Conference, New York, p. 4691-4694. Cox, I. J., Kilian, J., Leighton, L. T., Shamoon, T., 1997, Secure spread spectrum watermarking for multimedia, Trans. Image Processing, p. 1673– 1687. Hazem, M., Al-O., 2014, Semi-fragile watermarking for gray scale image authentication and tamper detection based on an adjusted expanded-bit multiscale quantization-based technique, Journal of Visual Communication and Image Representation, p. 1064-1081. Raymond, B. W., Edward, J. D., 1999, Fragile watermarking using the VW2D watermark, Security and Watermarking of Multimedia Contents, International Conference on Security and Watermarking of Multimedia contents, April 9, p.204. Shabir, A., Javaid. A., Ghulam, M., 2014c, A Secure and Efficient Spatial Domain Data Hiding Technique based on Pixel Adjustment". American Journal of Engineering and Technology Research 14, p. 33-39. Wong, P.W., Memon, N., 2001, Secret and public key image watermarking schemes for image authentication and ownership verification, IEEE Trans.Image Process 10(10), p.1593-1601. Archariya, U.R., Subhanna, P. B., Satish, K., Lim, C. M., 2003, Transmission and storage of medical images with patient information, Journal of Computers in Biology and Medicine. Vol: 33, p. 303–310. Baisa, L. G., Suresh, N. M., 2012, ROI based embedded watermarking of medical images for secured communication in Telemedicine, World Academy of Science, Engineering and Technology, Vol: 6, August 21. Neha, S., Sanjay, K. M., 2014, ROI based medical image watermarking with zero distortion and enhanced security, I.J. Modern Education and Computer Science, 10, p. 40-48. William, S., 2005, The RC4 Stream Encryption Algorithm, Copyright.

[279]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Content Centric Networking and Interest Flooding in Communication Networks: A Review Rohit Agnihotria*, Kshitij Pathaka, Prashant Bansodb, Chetan Chouhana a

Mahakal Institute of Technology, Behind Air Strip Dewas Road, Ujjain, India b SGSITS, 23 Park Road , Indore, India

Abstract The Content Centric Networking (CCN) is a newer architecture to deal with the data networking, specifically changing the traditional delivery pattern of data from the generalized host-centric networking to content-centric networking with a changed approach. The Denial-of-Service (Dos) attack is now a days a serious matter of concern. Dos attack is root cause for congestion and host inaccessibility. Since the content-centric approach relies on content request messages (Interests packets) and its routing state information stored within routers. Sequentially if this routing state information minimizes, it results to Dos issue of flooding of such content request messages or Interest Flooding Attack (IFA). Thus, the objective of this paper is to analyze IFA and its mitigation techniques for resolving congestion to an extent.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Content Centric Networking; Denial of Service; Interest Flooding Attack

1. Introduction Today in the changing era, vast number of users is using various platforms for accessing the data over the internet using a host centric request-response model. The basic idea of host-centric networking can be changed in way, such that accessing the content can be made easier by CCN architecture CCN Project (2009). The CCN started as a research project at Palo Alto Research Centre (PARC) in 2007 and the first released software CCNx 0.1 came in 2009. Till the date latest version is CCNx 1.0 is available. It is a project in which various implementations are being done and which is also known as Information Centric Networking (ICN) or Named Data Networking (NDN) Jacobson et al. (2009b) and various modifications are planned for future. The basic concept of CCN is to give a relaxation to users in accessing the remote data or “Content Object” by simply naming the content itself and not by the machine, addressing mechanism implemented in the host centric approach. The key difference between CCN and host centric approach is of content chunks only shown below in Fig 1. Jacobson et al. (2009b). There are two basic types of packets as interest and data as depicted in Fig 4. Jacobson et al. (2009b).The broadcast of Interest is possible. The data request from user in general known as Interests packet or Interests so that the node having the data can reply to the requester via data packet. At the time of request, the data been transferred to the requester so that intermediate routers can share the data. The IFA is a serious concern of Dos in the CCN architecture. Thus, creation of unnecessary request for the source from the adversary (malicious user) can cause the congestion and network-flooding scenario. Thus, the situation becomes more vulnerable when the interests are being denied at the router itself. Despite of secure and validated naming mechanism for CCN still the concerns e.g. IFA, cache poisoning, cache privacy and other such exists. The various key advantages are (1) content based naming mechanism (2) No need to secure the channel of the data or the link apart from the current requirement (3) Efficient Computation and need of memory can be fulfilled within the network since the concept of multicaching exist so that multiple nearby caches can be associated for productive memory usage pattern (4)The use of efficient multipath routing in network is also possible (5) Duplication of packet discarded and flow balance achieved also Retransmission of interest packet for which the data not arrived (6) Consumer can have timeouts for the satisfaction of interest within particular time. Thus the next sections of this paper are as follows the architecture of CCN and its * Corresponding author. Tel.: +91 7312 343238. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Agnihotri et al/COMMUNE – 2015

features in Section II. Interest Flooding Attack in Section III also various interest flooding mitigation techniques in Section IV, summary and previous related work in Section V. In Section VI we conclude with some future issues.

Fig.1. Protocol Specification of CCN

2. CCN Overview 2.1.

Naming Framework

The name mechanism comprises of various components having many sequence of bytes. The naming here is just as Uniform Resource Identifier name segment, which carries two things label for segment and name component for identification purpose and value for such labels which gives scheme name as Labeled Content Identifier (lci) for unique identification. The CCN names can vary from human readable ones to non-human readable format. For human readable name, the latest type used is UTF-8 encoding. For e.g. the CCN name in human readable format can be as – /abc.com/CCN/Conference.pdf/…, where “/” identifies the root and subfolders as below in Fig.2. by Jacobson et al. (2009b). Exact name matching is done while previous version was having prefix matching for naming schemes, signing data in Fig. 3 by Jacobson et al. (2009b). Global name/Organization name/ content name/type e.g. - /abc.com / CCN/Conference.pdf/... 7

5 abc.com

3 CCN

1 Conference.pdf

Fig. 2. Naming in human readable and binary encoding formats

2.2.

Fig. 3. Signing Mechanism naming associated with Key-id

Content Object

The Content Object comprises of Name and its Payload. The other details in the Content Object can be a set of validation algorithm schemes, cryptographic signature/s algorithms, Message Authentication Code (MAC) and Message Integrity Check (MIC).The basic validation of the interest packets can be done by Key-id for the authentication mechanism. Thus if an interest is having Name associated with the valid Key id we can easily identify the Content Object requested matching the validation pattern of the Name and Key-id. The ContentObjectHash implementation is easy hashing mechanism using SHA-256 for Content Objects, which adds to security. 2.3.

CCN Node Structure

The CCN node model is just very much same to IP node model. However, here Interest packets being issued for the request of the content and the reverse path is followed for the re-routing, if the content is matched at a node then request not forwarded to upcoming faces and directly data being sent to its requester. Data structures as in Fig. 5: Jacobson et al. (2009b). Content Store Name Packets

Pending Interest Table Interest Requesting Faces

[281]

Forwarding Information Base Prefix Faces

Agnihotri et al/ COMMUNE-2015

Face 1

Face 2

Face N Fig. 4.Classification of IFA

Fig. 5.Classification of IFA

 Content Store (CS) is the structure used as a buffer memory and for the storage of previous named indexes. Also used for the content caching. Same as buffer memory in IP network.  Forwarding Information Base (FIB) is used for forwarding the request towards the original source having the content. It contains the name prefixes and there various faces and list which can give content for the prefixes. It checks for duplicate Interests and discards them. Thus no duplicate request can exist.  Pending Interest Table (PIT) is used for the storage of pending interests and to have the idea of the interest sent through the corresponding node in upstream so that when the data has to be delivered to the requester it can be delivered at the correct path. Also other various terminologies can be associated here as timeout, delay etc. The basic nature of Content Object and CCN Node Structure explained here in brief only. 3. Interest Flooding Attack The two major data structures as CS and PIT is very much useful to the routers or node but correspondingly it can be used against the source from the adversary i.e. creating a huge number of interest packets to slow down the working of the router and flooding of the pending interest at routers which can cause to Dos for the interest packets generated from the legitimate users. Thus the unique strategy of CCN/NDN here is depleted in way to achieve the temporal Dos from the adversary. It can also cause the new interest packets to be dropped at the router level and this is known as the Interest Flooding or IFA in Fig. 6. Afanasyev et al. (2013), Choi et al. (2013).

Fig. 6.IFA Mechanism

Fig. 7.Classification of IFA

There are two types of flooding attacks mainly as follows FIB-based and the Broadcast based as above in Fig. 7 Tang et al. (2013). FIB based scenario is very much similar as explained above that the interests are being forwarded but by the longest prefix match in FIB and such interests are routed very much close to the data producer since a burst of interest can cause the PIT to be full. Also for the impacting target the adversary can throw such interest which is been targeted as similar to prefix names but when the prefix matching does not match these fake request gets only stored at the PIT but also get removed after the timeout period .This type of flooding causes PIT entry to be delayed and causes congestion. The request been re-routed to the reverse path of interest packet as explained earlier so fake request can’t get corresponding replies. And thus identification of fake request has been possible. In Broadcast based scenario if the

[282]

Agnihotri et al/COMMUNE – 2015

prefix matching of names of interest and in FIB not possible in any case as assumption then it results in broadcasting the interest packet to the associated network nodes Tang et al. (2013). 4. Interest Flooding Mitigation Techniques The various Interest flooding mitigation techniques been discussed here as we require to focus on the work of the flooding mitigation corresponding reviews been specified. The Pushback Mechanism is being applied for the attacks or the fake request name patterns as suspected to the router. After identification we can send a pushback message towards the downstream and other routers so that to block such interest patterns and also to limit the forward Interest pattern at the attack timing. Basic mechanism of pushing the request is to send the fake interest back to adversary or the nearby location where attack been detected in Gasti et al. (2013). There can be three types of IFA based on the interest generation mechanism (1) Static (2) Dynamically generated interest (3) Nonexistent. Basically the dynamic behavior of the interest can cause an issue of network bandwidth depletion since the basic in network caching mechanism Lee, Nakao (2013) is affected directly so the interest packets increased with a higher risk. Poseidon used for the purpose to countermeasure the IFA. Thus this method is proposed implementing the algorithm approach at routers for the identification of the various anomalies and to mitigate the attack patterns in a way to have a proper router mechanism as in Compagno et al. (2013). Also the pushback as “alert message” of interest packets being done to avoid the false request messages from the adversary. It differentiates the alert message from the content message. Thus the alert messages are not being checked in PIT and such message contains the time-stamp, payload and the detailed information used of the adversaries. Also it is being realized for the local and collaborative techniques for IFA mitigation. Also in PIT the malicious interests can be decoupled after detection of the attack as in Wang et al. (2013b) by disabling PIT Exhaustion. When the victim is targeted for the name prefixing and different suffixes been chosen as interest the results are not been searched in cache as they do not match the whole content rather the interests been used as malicious content to be rooted in the network. Thus also if an adversary wants to affect the source working it must generate that much no. of packets which can harm the whole source and impairing the overhead of congestion in it quoted by Lauinger (2010). Thus the major concept if there is sufficient interval of time then the router can detect such malicious interest as the fraction of name prefixes and if this fraction or time interval is high then the router can easily detect the false interest and further such interest can be dropped. The other interest mitigation techniques exists as the token bucket with the per interface fairness approach thus to apply a fairness scheme for the admitting the interest packets in the router at every interface and issuing equal aspect of entry for the malicious and legitimate interests but in this method the lacking of proper interest sequence also increased the DoS to a extent having huge no. of the false interest. Then method of intelligent attack mitigation implemented. Since the basic idea of adversary is to just root the interest packet in the network and increase the congestion which will automatically result in flooding of requests. But as adversaries are not at all interested for the content retrieval thus the adversaries can be identified by maintaining the statistical approach of listing such adversaries by Afanasyev et al. (2013). The other such mechanism implemented as the Satisfaction based interest-acceptance and the Satisfaction based pushback in Afanasyev et al. (2013). The only key difference in these two is that the first one applies the feature as the independent decision been taken at each router as to drop the request or forward it. Thus probability of legitimate interest being dropped is increased and again influencing the malicious interest rates may occur. But in the Satisfaction based pushback mechanism actually the pushback of the malicious interest been done towards the adversary by identifying the good and bad interest packets implanting the explicit token limit for the each interface pattern and thus actually implemented a way to control the IFA in greater perspective by Afanasyev et al. (2013). If the CCN FIB size is reduced it can also help in the prevention of the IFA and also it can minimize IFA to an extent. The comparative analysis of the Native beacon Mode Algorithm (NBMA) based on the discrete simulations with the various algorithms as(1) Non Cluster Mode Algorithm (NCMA) (2) Cluster Mode Algorithm (CMA) (3) Terrain Information table (TINT)-CMA Algorithm justifies the different approaches and used methodologies. Since the FIB entries are preloaded for the purpose of beacon advertisements in NBMA but in NCMA the preloading of the entries in FIB avoided to overcome the unnecessary advertisements since the content can be of huge memory the higher gateways shall be of efficient higher size Ekambaram, Sivalingam (2013) .The variation with CMA is clustering concept which is applied for the maintaining the details of the content at cluster level. But significant increase in memory size is required at cluster since logically the memory also increases. Since the higher number of packets generate through a cluster to all corresponding related node/s, the interest packets can be flooded in setup phase thus causing the congestion till the transfer phase of the content chunks. The variation in the TINT-CMA Algorithm thus maintaining the terrain information for the respective data at the extra higher gateways or backbone routers, by just having the preferred route only is maintained for the various clusters under a terrain increasing the memory requirement but additionally serving the better performance for the control of the IFA by Ekambaram, Sivalingam (2013).Thus identifying if interest can be searched in TINT table and then the routing decision is performed thus overcoming the CMA approach where each interest been routed to the all the routers associated. Thus memory consumption is not an issue but improvement of performance, avoidance of the unnecessary interest is achieved. But prefix compression technique can cause the extra overhead in TINT-CMA by Ekambaram, Sivalingam (2013).

[283]

Agnihotri et al/ COMMUNE-2015

5. Summary and Related Work We studied the key concept of CCN and its variation with the IP network. The attacks existing in the CCN architecture as IFA as Dos and its mitigation techniques, which can improve the networks overall performance and security of the name prefixes and content from Smetters, Jacobson (2009). The security at the router implemented by CCN and various data structures at CCN Node by Jacobson et al. (2009b). Also prefix compression can be still a greater issue to root the unnecessary interest Ekambaram, Sivalingam, 2013). The various attack methods as prefix hijacking, reflection attacks, looping are DoS attacks listed in Choi et al. (2013) not only be easily prevented but basically these does not affects the CCN architecture directly as in comparison to IP network but the IFA from Compagno et al.(2012), cache poisoning Gao et al. (2006), privacy of content Xie et al. (2012) and cache in Acs et al. (2013), Deng et al. (2008) are the real threats to CCN that cause depletion of the network bandwidth as well as causing threat as prefix compression in Lauinger (2010), Ekambaram, Sivalingam (2013). The different mechanism of Pushback, Interest mitigation, Poseidon given in the Afanasyev et al. (2013), Choi et al. (2013), Compagno et al. (2013), Ekambaram, Sivalingam (2013).The other such threats as packet flooding Jun et al.(2011), the latest work of ease is the Voice-over IP (VOIP) through CCN Jacobson et al. (2009a). The cache based privacy and accessing issue still exist thus the concept of user-assisted in network caching giving the centralized and distributed approach in Lee, Nakao (2013), Lee et al. (2014) and its analytical approach of performance in the in-network Kim, Yeom (2013), Scenarios of Multicaching thus implementing the features of multicasting as well as caching by Katsaros et al. (2010), collaborative caching, cache privacy in Wang et al.(2013a) and increasing cache robustness in Xie et al. (2012).The entire summary of various approaches and remarks been mentioned in the Table.1. 6. Conclusion and Future Work We have analyzed the key concept of the Content Centric Network and the Dos attacks into it. Mainly the concentrated areas are Interest Flooding Attack and mitigation techniques. Although the idea of CCN is new at present but the advantages of it are still better in comparison with IP network. Few advantages are Redundancy Control, Flow Control, Data Signature schemes. Despite of all these advantages still the interest flooding issue exist at a lesser aspect but having various issues exist as experiments (Camara et al., 2012), (Loo, Aiash, 2014) and to work in the future. Firstly the pushback mechanism unnecessary consumes the network bandwidth to traverse the malicious interest at the reverse path rather dropping such interest there can add to the performance. Although CCN is having the key features better than IP but still it has to be implemented still in a huge aspect to totally replace the IP network. Thus we hope the concept CCN will be a prominent achievement for the future internet and its security. Table 1. Various Studied Approaches and Methods S. No. 1. 2. 3. 4. 5.

Main issue IFA and Countermeasures In NDN Identifying Interest Flooding In NDN DoS and DDoS in Named Data Networking Poseidon-Interest Mitigation Techniques Security and Scalability in CCN

6.

Interest Flooding Reduction In CCN

7. 8. 9. 10.

Voice over-IP(VOIP) Multicaching for improved Cache performance Collaborative caching NDN Interest Flooding Attacks and Countermeasures

Remarks IFA risks and its solutions Attack issues identification Various Dos attack methods A solution over IFA Security issues and impact Algorithms for IFA mitigation TINT-CMA Concept VoIP In CCN Memory use In CCN Collaborative Cache use IFA issues

References Acs et al., 2013. Cache Privacy in Named-Data Networking, IEEE 33rd International Conference on Distributed Computing Systems. Afanasyev et al., 2013. Interest Flooding Attack and Countermeasures in Named Data Networking ,IEEE. Camara et al., 2012. Experimentation with ccn, INRIA, Planete-Project, Presentation. Content Centric Networking (CCN) Project., 2009. ., Sep 2009, last visited on 22nd Feb 2015. Choi et al.,2013. Threat of DoS by Interest Flooding Attack in Content-Centric Networking, Information Networking (ICOIN), 2013 International Conference ,IEEE. Compagno et al., 2012. NDN Interest Flooding Attacks andCountermeasures,www.acsa admin. org/2012/program/posters/poster05 .pdf. Compagno el al., 2013. Poseidon: Mitigating Interest Flooding DDoS Attacks in Named Data Networking, IEEE. Deng et al., 2008. Pollution attacks and defenses for Internet caching systems, Computer Networks, vol. 52, no. 5, pp. 935–956. Ekambaram, Sivalingam., 2013. Interest Flooding Reduction in Content Centric Networks, IEEE 14th International Conference on High Performance Switching and Routing. Gao et al.,2006. Internet cache pollution attacks and countermeasures, in ICNP. IEEE Computer Society, 2006, pp. 54–64. Gasti et al., 2013. DoS & DDoS in Named Data Networking, Computer Communications and Networks (ICCCN), 22nd International Conference, IEEE ,July 30 2013-Aug. 2 2013. Jacobson et al., 2009. VoCCN: voice-over content-centric networks, in ReArch ’09: Proceedings of the 2009 workshop onRe-architecting the internet. New York, NY, USA: ACM, 2009, pp. 1–6.

[284]

Agnihotri et al/COMMUNE – 2015

Jacobson et al., 2009. Networking named content, in: Proceedings of ACM CoNEXT, www.named-data.net/publications. Jun et al., 2011. Ddos flooding attack detection through a step-by-step investigation, Networked Embedded Systems for Enterprise Applications, 0:1– 5. Katsaros et al., 2010. MultiCache: An overlay architecture for information-centric networking, INFOCOM IEEE Conference on Computer Communications Workshops. Kim, Yeom., 2013. Performance analysis of in-network caching for content-centric networking, Computer Networks 57 (2013) 2465–2482 , ELSEVIER . Lauinger,2010. Security & scalability of content-centric networking, Master Thesis, TU Darmstadt, 2010. Lauinger et al., 2012. Privacy Risks in Named Data Networking: What is the Cost of Performance?, ACM SIGCOMM Computer Communication Review , Volume 42, Number 5. Lee, Nakao., 2013. User-assisted in-network caching in information-centric networking, 14 August 2013, Computer Networks, ELSEVIER . Lee et al., 2014. SCAN: Content discovery for information-centric networking, ELSEVIER . Loo, Aiash, 2014. Challenges and solutions for secure information centric networks: A case study of the NetInf architecture, Journal of Network and Computer Applications , ELSEVIER. Smetters, Jacobson., 2009. Securing network content, PARC, Tech. Report. Tang et al., 2013. Identifying Interest Flooding in Named Data Networking, IEEE International Conference on Green Computing and Communications and IEEE Internet of Things and IEEE Cyber. Wang et al., 2013. Intra-AS Cooperative Caching for Content-Centric Networks, August ICN’13. Wang et al., 2013. Decoupling malicious Interests from Pending Interest Table to mitigate Interest Flooding Attacks, Globecom Workshops (GC Wkshps), IEEE. Xie et al., 2012, Enhancing Cache Robustness for Content-Centric Networking , INFOCOM, IEEE Proceedings .

[285]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Comparative Performance Analysis of MEMS Piezo Resistive Pressure Sensors M. Tariq Banday, S. Umira R. Qadri* Department of Electronics & Instrumentation. Technology, University of Kashmir, Srinagar, India

Abstract Micro Electro Mechanical Systems (MEMS) has been recognized as one of the capable technologies of the 21 st Century as it has the potential to update both industrialized and customer goods. With the advances in surface and bulk micromachining, microelectronics based on silicon has led to the development of MEMS pressure sensors. Pressure sensors measure mechanical deformation caused in a diaphragm when it experiences stress by the application of differential pressure. Among micro machined pressure sensors, Piezo resistive Pressure Sensors (PPS) are simplest to fabricate as they can be easily integrated with microelectronics circuits. Most common diaphragm shapes for MEMS pressure sensors are square, rectangular, and circular. This study through analysis with reverence to deflection and stress emphasizes the distinction among the diaphragm shapes of PPS and recognizes the best characteristic diaphragm shape. This comparative study identifies the change in performance and suitability of MEMS pressure sensors for various applications in accordance with certain geometrical parameters such as Poisson’s Ratio, Young’s Modulus in Gigapascals (Gpa), Length and Thickness under a uniform applied pressure from 0 Megapascals (Mpa) to 1 Megapascals (Mpa) in the interval of 0.1 Megapascals (Mpa). The results show that the circular shape is very effective for the characterization of MEMS pressure sensors.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: MEMS; PPS; Young’s Modulus; Poisson’s Ratio; Diaphragm Shape

1. Introduction Micro Electro Mechanical Systems (MEMS) promisingly drive the world to the next technological revolution. MEMS based devices offers significant features over conventional macroscopic devices, which offer small size leading to reduced cost, excellent mechanical properties of silicon comparable to steel, benefits from the sophisticated designing, processing and packing technology developed for the IC industry and easy Integration with IC circuitry to produce systems on a chip (Bryzek et al., 2006). Among various sensing devices developed and commercialized by different academic institutions and industries, MEMS based pressure sensors have gained immense importance as they find applications in everyday life involving sensing, monitoring and controlling pressure. These sensors represent 60 to 70 percent of the marketplace among various MEMS devices. Irrespective of biomedical, automobile, civilian, aerospace, defense, oceanography, or domestic applications, MEMS pressure sensors are becoming necessary, which existed as strain gauges in their primitive form for over several decades with the detection of exceptional mechanical properties of silicon (Lin and Yun, 1998). Piezo Resistive pressure sensors use diffused or implanted resistors usually on the surface of a thin silicon diaphragm to measure various parameters such as blood pressure in biomedical field (Najafi et al., 1990). As pressure is applied, the diaphragm deforms and the resulting strain impacts the carrier mobility along with the number density known as piezoresistive effect which describes the change in electrical resistance that occurs when an external force is applied to a semiconductor. Typically, diaphragms are the mechanical structures, which are simplest and appropriate for use as a pressure-sensing component in both traditional as well as MEMS based pressure sensors. Substantial research works have been emphasized on micro machined diaphragm based piezo resistive pressure sensors during recent years (Kung and Lee, 1992; Zhang et al., 1996; Clark and Wise, 1979). PPS are used in diverse

*Corresponding author. Tel.: +91 9596 434477. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Banday and Qadri / COMMUNE-2015

applications including pressure gauges, pressure switches, tire pressure meters, process control, and automobile parts (Shaikh et al., 2008). 2. Related Work Bahadorimehr et al. signed a framework using Finite Element Method (FEM) analysis with lntellisuite software and showed which kind of diaphragm among different diaphragm shapes (square, rectangular, and circular) with what type of parameters can be selected for a particular MEMS device. Highest deflection has been shown by a circular diaphragm with lowest stress on its ends when compared with other diaphragms on applying the same pressure (Bahadorimehr et al., 2010). Balaji and Bhat highlighted the distinctions among three diaphragm shapes (square, rectangular, and circular) with respect to linearity and burst strength. Finite Element Analysis (FEA) with Conventorware software has been used for the distribution of stress on the diaphragm. It was found that the circular diaphragm gives good linearity as well as higher burst stress acceptance characteristics (Balaji and Bhat, 2012). Komaragiri et al have found an optimal diaphragm shape that resulted in a reasonable output stimulus with minimal deflection and stress. They used Finite Element Method (FEM) with COMSOL. Among the three diaphragm shapes (square, rectangular, and circular) with relative dimensions, the square shaped diaphragm was found to perform better than others (Kattabooman et al., 2012). Krishnamurthy and Meena depicted the proper selection of the membrane geometry and piezo resistor position to enhance the sensitivity by the structural design and performance analysis of a silicon piezo resistive pressure sensor in terms of three diaphragm shapes viz. square, rectangular, and circular using Finite Element Analysis (FEA). Results depicted that from the design aspects, the circular diaphragm is most favored, and additionally, the square diaphragm is the preferential geometry because of high sensitivity, resulted from maximum induced stress (Krishnamurthy and Meena, 2013). Optimization of Piezo resistive MEMS Pressure Sensors in terms of an optimal diaphragm shape among the three diaphragm shapes viz. square, rectangular and circular by Finite Element Method (FEM) using ABAQUS showed that circular shaped diaphragms performed much more efficient than other shapes. The effect of holes in all the three shapes has also been found which resulted in better influence on the function of diaphragms by rectangular holes (Fathi and Moradi, 2014). 3. Deflection Analysis 3.1. Methods Used for Deflection Analysis To analyze silicon based piezoresistive pressure sensor diaphragm it is implicit that the diaphragm has a uniform thickness with completely clamped ends. Deflection of the diaphragm w(x, y) is governed via load-deflection equation as a function of pressure P (x, y), shown in Eq. (1) (Olszacki, 2009). 𝜕4 𝑤 𝜕𝑥 4

+

𝜕4 𝑤 𝜕𝑦 4

+2

𝜕4 𝑤 𝜕𝑥 2 𝑦 2

=

𝑃 𝐷

(1)

Where D defines the flexural rigidity and is specified by Eq. (2).

𝐷=

𝐸ℎ3 12(1−𝜐2 )

(2)

Where ν is the Poisson’s ratio, E is the Young’s modulus and h is the diaphragm thickness (Timoshenko and Krieger, 1987). The solution to Eq. 1 to determine the deflection of a diaphragm is complex and therefore, the Polynomial Approximation Method (Herrera-May et al., 2009) described in following sections is used. 3.1.1. Si Square Diaphragm With appropriate approximations and simplifications, the solution to Eq. (1) for maximum deflection (𝜔max) of a square diaphragm can be obtained using Eq. (3). 𝜔𝑚𝑎𝑥 = −

𝛼𝑃𝑎4 𝐸ℎ3

(3)

Where E is Young’s modulus in Gpa, P is the applied pressure in Mpa, h is diaphragm thickness in μm, a is diaphragm length in μm and ωmax is maximum diaphragm deflection in Mpa. 3.1.2. Si Rectangular Diaphragm With appropriate approximations and simplifications, the solution to Eq. (1) for maximum deflection (𝜔max) of a rectangular diaphragm can be obtained using Eq. (4). 𝜔𝑚𝑎𝑥 = −

𝛼𝑃𝑏 4

(4)

𝐸ℎ3

[287]

Banday and Qadri / COMMUNE-2015

Where E is Young’s modulus in Gpa, P is the applied pressure in Mpa, h is diaphragm thickness in μm, b is diaphragm length in μm and ωmax is maximum diaphragm deflection in Mpa. 3.1.3. Si Circular Diaphragm With appropriate approximations and simplifications, the solution to Eq. (1) for maximum deflection (𝜔max) of a circular diaphragm can be obtained using Eq. (5). 𝜔𝑚𝑎𝑥 = − Where 𝑊 = 𝜋𝑎2 𝑃 and 𝑚 =

3𝑊(𝑚2 −1)𝑎2

(5)

16𝜋𝐸𝑚2 ℎ3

1 𝜐

W is the total force acting on the plate, E is Young’s modulus in Gpa, υ is Poisson’s ratio, P is the applied pressure in Mpa, h is diaphragm thickness in μm, a is diaphragm radius in μm and 𝜔𝑚𝑎𝑥 is maximum diaphragm deflection in Mpa. 3.2. Experiments and Results Using MatLab Software, three diaphragms of approximately same area as given in table 1 and same thickness (3.002604μm) are analysed in this study. The three diaphragms are square diaphragm of side length 160μm, a rectangular diaphragm of dimensions 230μm×110μm and a circular diaphragm of radius 90 μm. The silicon diaphragm material properties used for the simulations are specified in table 2. Some of the design concepts used and fulfilled in this work to explore various parameters of the piezoresistive pressure sensors are (a) Wide operating range accessible is from fractions of psi to15, 000 psi; (b) Commercially, silicon diaphragms with less than 20 μm thickness are frequent; (c) The highest Stress that a silicon diaphragm can resist is 7Gpa which is equal to rupture stress; and (d) The thickness of the diaphragm should not surpass 10% of length of diaphragm (Krishnamurthy and Meena, 2013). Table 1: Description of Diaphragms in various Dimensions

Shape of Diaphragm (Square) (Rectangular)

Length

Width

Radius

Formula

Area

160µm (a)

160µm (a)

a2

25600μm2

230µm (a)

110µm (b)

axb

25300μm2

Not Applicable

Not Applicable

Not Applicable Not Applicable 90µm (a)

Πa2

25434µm2

(Circular)

Table 2: Material properties of silicon (Chang, 2012)

Material Property Young’s Modulus (Gpa) Poisson’s Ratio

Normal Range 130-187 0.25-0.36

Coefficients for α in Eq. (3) and Eq. (4) have been obtained from the aspect ratios shown in table 3. The maximum deflection produced in the diaphragms (square, rectangular and circular) are determined and compared for an applied pressure range of 0MPa to 1Mpa in the interval of 0.1Mpa. Results obtained from the MatLab analysis are shown in table 4. Table 3: Coefficients for α and β for different Aspect Ratios (a/b) (Madhavi et al., 2013)

a/b α β

1 0.0138 0.3078

1.2 0.0188 0.3834

1.4 0.0226 0.4356

1.6 0.0251 0.4680

1.8 0.0267 0.4872

2.0 0.0277 0.4974

∞ 0.0284 0.5000

On analysing recorded reading as given in table 3, it is clear that the circular diaphragm deflects more on the minimum values of Poisson’s ratio and Young’s modulus in comparison to other diaphragms under the same differential applied pressure.

[288]

Banday and Qadri / COMMUNE-2015 Table.4: Comparative deflection results for various silicon diaphragm shapes

Pressure (P) in Mpa 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Square Diaphragm

Rectangular Diaphragm

E=130Gpa

E=187Gpa

E=130Gpa

E=187Gpa

0.256993 0.513986 0.770979 1.027972 1.284966 1.541959 1.798952 2.055945 2.312938 2.569931

0.178658 0.357317 0.535975 0.714633 0.893292 1.07195 1.250608 1.429267 1.607925 1.786583

0.118155 0.23631 0.354465 0.472619 0.590774 0.708929 0.827084 0.945239 1.063394 1.181548

0.08214 0.164279 0.246419 0.328559 0.410699 0.492838 0.574978 0.657118 0.739258 0.821397

Circular Diaphragm E=130Gpa E=187Gpa υ=0.25 υ=0.36 υ=0.25 υ=0.36 0.327722 0.304265 0.227828 0.211521 0.655443 0.608531 0.455656 0.423043 0.983165 0.912796 0.683484 0.634564 1.310886 1.217062 0.911311 0.846086 1.638608 1.521327 1.139139 1.057607 1.96633 1.825593 1.366967 1.269129 2.294051 2.129858 1.594795 1.48065 2.621773 2.434124 1.822623 1.692172 2.949494 2.738389 2.050451 1.903693 3.277216 3.042655 2.278278 2.115214

Fig. 1: Linearity of Deflection for Square, Rectangular, and Circular Silicon Diaphragms Where Square Min. indicates Young’s Modulus (E) =130Gpa, Square Max. indicates Young’s Modulus (E) =187Gpa. Similarly, Rectangular Min. indicates Young’s Modulus (E) =130Gpa, Rectangular Max. indicates Young’s Modulus (E) =187Gpa and Circular Min. indicates both Poisson’s Ratio (υ) =0.25 and Poisson’s Ratio (υ) =0.36 for Young’s Modulus (E) =130Gpa, Circular Max. indicates both Poisson’s Ratio (υ) =0.25 and Poisson’s Ratio (υ) =0.36 for Young’s Modulus (E) =187Gpa.

With respect to the conditions of Load Deflection theory, linearity of deflection has been analysed on the linear applied pressure range from 0 to 1Mpa in the interval of 0.1Mpa with a uniform thickness of 3.002604μm for all the three diaphragm shapes (square, rectangular and circular). In figure 1 gives, maximum deflection recorded against pressure for the three studied diaphragm shapes. It is evident that the circular diaphragm shows maximum linearity of deflection on minimum values of the two geometrical parameters Viz. Poisson’s ratio (υ) and Young’s modulus (E) than that of rectangular and square diaphragms with the same applied pressure. 4. Stress Analysis 4.1. Methods used for Stress Analysis Stress is necessary for the design of diaphragm-based pressure sensors since two types of stress namely normal stress and shear stress emerge, when a diaphragm undergoes a strain by the application of uniform pressure (P) which is normal to its plane. According to Hooke’s law, the stress value for a square and a rectangular membrane as a function of the location on the diaphragm plane (x, y) may be expressed by using Eq. (6) and Eq. (7) respectively (Herrera-May et al, 2009). 𝜕 2 𝑤(𝑥,𝑦)

ℎ𝐸

𝜎𝑥 (𝑥, 𝑦) = 2(1−𝜐2 ) ( 𝜎𝑦 (𝑥, 𝑦) =

𝜕𝑥 2

𝜕 2 𝑤(𝑥,𝑦)

ℎ𝐸 2(1−𝜐2 )

(

𝜕𝑦 2

+ +

𝜕 2 𝑤(𝑥,𝑦) 𝜕𝑦 2

𝜕 2 𝑤(𝑥,𝑦) 𝜕𝑥 2

)

(6)

)

(7)

Formula for the circular membrane is only radius dependent as given by Eq. (8). 𝐸𝑧

𝜕2 𝑤(𝑟)

𝜎(𝑟) = − (1−𝜐2 ) (

[289]

𝜕𝑟 2

)

(8)

Banday and Qadri / COMMUNE-2015

For the analysis purpose, equations (6), (7) and (8) for stress in x and y directions are combined and simplified for various diaphragms as follows: 4.1.1. Square Diaphragm In a square diaphragm, utmost stress is at the centre of ends as given by Eq. (9).

𝜎𝑚𝑎𝑥 =

𝛽𝑃𝑎 2 ℎ2

(9)

Where P is the applied pressure in Mpa, a is diaphragm length in μm, h is diaphragm thickness in μm, and σmax is maximum diaphragm stress in Mpa. 4.1.2. Rectangular Diaphragm Likewise square diaphragm, for the rectangular diaphragm, utmost stress is at the centre of ends as given by Eq. (10).

𝜎𝑚𝑎𝑥 =

𝛽𝑃𝑏 2

(10)

ℎ2

Where P is the applied pressure in Mpa, b is diaphragm width in μm, h is diaphragm thickness in μm, and σ max is maximum diaphragm stress in Mpa. 4.1.3. Circular Diaphragm Maximum stress is at the edges in a circular diaphragm as given by Eq. (11).

𝜎𝑚𝑎𝑥 = Where 𝑊 = 𝜋𝑎2 𝑃 and 𝑚 =

3𝜐𝑊 8𝜋ℎ 2

(11)

1 𝜐

W is the total force acting on the plate, υ is poisson’s ratio, P is the applied pressure in Mpa, h is diaphragm thickness in μm, a is diaphragm length in μm and σ max is maximum diaphragm stress in Mpa. 4.2. Experiments and Results Using MatLab Software, three diaphragms of approximately same area as given in table 1 and same thickness (3.002604μm) have been analysed. The three diaphragms are square diaphragm of side length 160μm, a rectangular diaphragm of dimensions 230μm×110μm and a circular diaphragm of radius 90μm. The silicon diaphragm material properties used for the simulation are specified in table 2. The coefficients for β in Eq. (9) and Eq. (10) have been obtained from the aspect ratios given in table 3. The highest stress induced in the diaphragms (square, rectangular and circular) are determined and compared for an applied pressure range of 0MPa to 1Mpa in the interval of 0.1Mpa. Results obtained from MatLab analysis are shown in table 5. The stress distribution experienced by diaphragms under linear differential pressure applied from 0 to 1Mpa in the interval of 0.1Mpa with thickness 3.002604μm is shown in figure.2. Table.5: Comparative stress distribution results for various silicon diaphragm shapes

Pressure (P) in Mpa

Square Diaphragm

Rectangular Diaphragm

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

87.57 174.914 262.371 349.828 437.285 524.742 612.199 699.656 787.113 874.57

67.10568 134.2114 201.317 268.4227 335.5284 402.6341 469.7397 536.8454 603.9511 671.0568

[290]

Circular Diaphragm υ=0.25 υ=0.36 8.422872 12.12894 16.84574 24.25787 25.26861 36.38681 33.69149 48.51574 42.11436 60.64468 50.53723 72.77361 58.9601 84.90255 67.38297 97.03148 75.80584 109.1604 84.22872 121.2894

Banday and Qadri / COMMUNE-2015

Fig. 2: Stress distribution of square, rectangular and circular silicon diaphragms

Table 5 shows that the circular diaphragm has least stress at its ends when compared to other diaphragms. Further, stress points have for a square diaphragm is maximum while applying same pressure. From figure 2 it is clear that the circular diaphragm shows lesser stress distribution at its edges on minimum value of Poisson’s ratio (υ) while as the square diaphragm shows highest stress distribution in comparison to other two diaphragm shapes. Circular shape also fulfilled the condition that the stresses experienced by the diaphragms at the edges should not exceed the fracture stress, which is 7Gpa for silicon. 5. Conclusion Diaphragms of different geometries (square, rectangular and circular) have been analyzed for a pressure range from 0Mpa to 1Mpa. The experiments through simulations with various diaphragm geometries showed that the circular diaphragm shape yields more deflection but less stress at its edges. The square shaped diaphragm exhibits higher stress distribution in comparison to other two diaphragm shapes. Since the layout of circular shapes is not very common in VLSI manufacturing but can be useful where high stress is a vital constraint, the square shaped diaphragm is considered to be a good geometry for the design of MEMS pressure sensors for the reason that high stresses produced by applied pressure results in high sensitivity. References Bahadorimehr, A.R., Khakpour, R., Mansouri, Solmaz, R. M., 2010. Analytical Comparison for Square, Rectangular and Circular Diaphragms in MEMS Applications, IEEE, International Conference on Electronic Devices, Systems and Applications (ICEDSA). pp. 297-299. Bryzek Janusz, Shad Roundy, Brian Bircumshaw, Charles Chung, Kenneth Castellino, Joseph R., Stetter, Michael Vest, 2006. Marvelous MEMS, IEEE circuits & devices magazine, March/April 2006. Chang, Liu., 2012. Foundation of MEMS, second edition, Pearson Education Limited. Clark, Samuel, K., Wise, K D., 1979. Pressure sensitivity in anisotropically etched thin diaphragm pressure sensors, IEEE Tran. of Elec. Devices, vol. ED-26, no. 12, pp. 1887-1895. Fathi, Nabiollah, Abol., Moradi, Zohreh, Allah., 2014. Design and Optimization of Piezoresistive MEMS Pressure Sensors using ABAQUS, MiddleEast Journal of Scientific Research, Vol. 21, no.12, pp. 2299-2305, ISSN: 1990-9233. Herrera-May, A.L., Soto-Cruz, B.S., L´opez-Huerta, F., Aguilera Cort´es, L.A., 2009. Electromechanical analysis of a piezoresistive pressure MicroMexicana de fi´sica, vol. 55, no. 1, pp 14-24, Feb 2009. Kattabooman, N., Rama Komaragiri, Sarath S., 2012. VLSI Layout Based Design Optimization of a Piezoresistive MEMS Pressure Sensors using COMSOL, Excerpt from the Proceedings of the 2012 COMSOL Conference, Bangalore. Kung, J.T., Lee, H.S., 1992. An integrated air-gap-capacitor pressure sensor and digital readout with sub-100 attofarad resolution, IEEE J. Microelectromech.Syst., vol. 1, pp. 121–129. Lin, L., Yun, W., 1998. MEMS pressure Sensors for aerospace applications, Aerospace Conference, and Applications, IEEE, Vol.1, pp. 429-436. Madhavi, K.Y., Krishna, M., Chandrashekhara Murthy, C.S., 2013. Effect of Diaphragm Geometry and Piezoresistor Dimensions on the Sensitivity of a Piezoresistive Micropressure Sensor using Finite Element Analysis, International Journal of Emerging Science, and Engineering (IJESE), Vol.1, no. 9, ISSN: 2319–6378, July 2013. Najafi, K., Cho, S. T., Wise, K.D., 1990. Secondary Sensitivities and Stability of Ultrasensitive Silicon Pressure Sensors, Proceedings of IEEE conference, pp. 184-187. Olszacki, M., 2009. Modeling and optimization of piezoresistive pressure sensors, Ph.D.thesis, Université de Toulouse, France, July 2009. Shaikh, M.Z., Dr. Kodad, S.F., Dr. Jinaga, B.C., ©2005 - 2008. Performance Analysis of Piezoresistive MEMS for Pressure Measurement, Journal of Theoretical and Applied Information Technology (JATIT). Swathi Krishnamurthy, Meena, K.V., 2013. MEMS based Piezo resistive Pressure Sensor, International Journal of Research in Engineering & Advanced Technology (IJREAT), Vol. 1, no. 1, ISSN: 2320-8791, March 2013. Swathi Krishnamurthy, Meena, K.V., 2013. Performance Analysis of Piezoresistive Pressure Sensor, International Conference on Electronics and Communication Engineering (ECE), ISBN: 978-93-82208-84-67th April 2013, Bangalore. Timoshenko, S., Woinosky-Krieger, S., 1987. Theory of Plates and Shells. Vidhya Balaji, Bhat, K. N., 2012. A Comparison of Burst Strength and Linearity of Pressure Sensors having Thin Diaphragms of Different Shapes, INSTITUTE OF SMART STRUCTURES AND SYSTEMS, J. ISSS Vol. 2 ,no. 2, pp. 18-26, Sept 2012. Zhang, X., Mastrangelo, C. H., Tang, W. C., 1996.Surface-Micromachined capacitive differential pressure sensor with lithographically defined silicon diaphragm, IEEE J. Microelectromech. Syst., vol. 5, pp. 89–105.

[291]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Model Order Reduction of Large Scale Continuous Time Interval System Rajesh Bhatt*, Girish Parmar, Rajeev Gupta Department of Electronics Engineering, UCE, Technical University, Rajasthan, Kota, India

Abstract Two mixed methods for reducing the high order continuous time interval systems have been discussed. In these mixed methods, numerator and denominator are reduced separately. In first mixed method, both denominator and numerator are reduced by differentiation method. In second mixed method denominator is reduced by Mihailov criterion method, while numerator is reduced by Pade approximation method. Step responses of higher order interval system and reduced order interval system are taken for comparative analysis of the mixed methods. Integral square error has also been calculated for these two mixed methods. The second mixed method provides better stability of the reduced order interval system, if the original high-order system is stable. The methods are illustrated by one numerical example.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Mihailov Criterion; Pade Approximation; Reduced Order; Higher Order

1. Introduction The complexity of higher order linear system are large and order of matrices are higher and difficult to deal with. The lower order model helps to better understanding of physical system in view of, reduce computational and hardware complexity and provide better controller design. In continuous time interval system numerator and denominator coefficients are taken in the form of intervals as reported by Bandyopadhyay B. et al, 1997. This paper presents two mixed methods for order reduction of high order continuous time interval system. In first mixed method, both numerator and denominator are reduced by Differentiation method as given by Gutman improved by Kranthikumar D. et al, 2011. In second mixed method, denominator is reduced by Mihailov criterion method as reported by Kranthikumar D. et al, 2011 while numerator is reduced by Pade approximation method as reported by Shamash Y., 1975. 2. Problem Formulation The transfer function of a higher order interval systems is H(s) and reduced order interval system is R(s) as reported by Bandyopadhyay B. et al, 1997: H ( s) 

 ]  [c  , c  ]s  .............[c  , c  ]s n 1 [c-21, c21 22 22 2n 2n N ( s)  - , c  ]  [c  , c  ]s  .............[c   n D( s ) [c11 , c ] s 12 1, n  1 11 12 1, n 1

(1)

where n is the order of the higher order continuous time interval system and it should be reduced to k, applied that n>k always then reduced order interval system will be R( s ) 

    [d 21 , d 21 ]  [d 22 , d 22 ]s  ..........  [d 2k , d 2k ]s k 1 N ( s)  k     [d11, d11]  [d12 , d12 ]s  .........  [d1, k 1 , d1, k 1 ]s k Dk ( s)

(2)

*

Corresponding author. Tel.:+0-941-322-2142. E-mail address: [email protected] ISBN: 978-93-82288-63-3

Bhatt et al/COMMUNE – 2015

3. Mixed Methods 3.1

First mixed Method

In this method both numerator and denominator are reduced by Differentiation method. Following steps are taken during this method: Step 1: The denominator polynomial of the kth order reduced model as given in Eq (2) by using differentiation method: Dk ( s) 

1 1 D( ) s s

(3)

Step 2: Determination of the numerator coefficients of the kth order reduced model by using differentiation method: N k ( s) 

3.2

(4)

1 1 N( ) s s

Differentiation method

First of all higher order numerator or denominator is reciprocated and then differentiated. Number of time differentiation depends on the value of n-k, where n and k are orders of higher order and reduced order interval system. Then again, equation is reciprocated. High order transfer function

Reciprocal transformation

Differentiate the transfer function

The number of differentiation depends upon (n-k)

Second reciprocal transformation

Reduced order transfer function

Fig. 1: Flow chart of Differentiation-method

3.3

Second mixed method

In this mixed method, denominator is reduced by Mihailov criterion while numerator is reduced by Pade approximation method. Following steps are taken during this method: Step 1: Determination of the denominator polynomial of the kth order reduced model: Substituting s = jω in D(s) and separating the denominator into real and imaginary parts, D( j)  ([c11 , c11 ]  [c12 , c12 ]s  ...........  [c1,n1 , c1,n1 ]( j) n  [c11 , c11 ]  [c13 , c13 ] 2  .....)  j([c12 , c12 ]  [c14 , c14 ] 2  .......)

  ( )  j ( )

(5)

Where ω is the angular frequency in rad/sec. ξ (ω) = 0 and ɳ (ω) = 0, the frequencies which are intersecting 𝜔0 = 0, 

[1 , 1 ] ……..  [n1 , n1 ] are

obtained, where [1 , 1 ]  [2 , 2 ]  ............  [k1 , k1 ]

[293]

Bhatt et al/COMMUNE – 2015

Similarly substituting s = j𝜔 in 𝐷k(s), then obtain Dk ( j )   ( )  j ( )

(6)

Where      ( )  [d11 , d11 ]  [d13 , d13 ] 2  .........      ( )  [d12 , d12 ]  [d14 , d14 ] 2  .........

Put 𝜙 (𝜔) = 0 and 𝜓 (𝜔) = 0, then we get k number of roots and it must be positive and real and alternately distributed along the w axis. The first k numbers of frequencies are

0, [1 , 1 ], [2 , 2 ]............[k1 , k1 ] are kept unchanged and the roots of 𝜙 (𝜔) = 0 and 𝜓 (𝜔) = 0. Therefore,

 ( )  [1 , 1 ]( 2  [1 ,1 ]2 )( 2  [3 ,3 ]2 ).............

(7)

 ( )  [2 , 2 ]( 2  [2 , 2 ]2 )( 2  [4 ,4 ]2 ).............

(8)

For finding the coefficient values of [1 , 1 ] and [2 , 2 ] are calculated from ξ (0) = 𝜙 (0) and ɳ [1 Keeping these values of [1 , 1 ] and [2 , 2 ] in equations (7) and (8)



, 1 ] .

Respectively, 𝜙 (𝜔) and 𝜓 (𝜔) are obtained and 𝐷k(𝑗𝜔) is obtained as Dk ( j )   ( )  j ( )

(9)

Now replace j𝜔 by s, and then the 𝑘th order reduced denominator 𝐷k (𝑠) is obtained as     Dk ( j )  [d11 , d11 ]  [d12 , d12 ]s  .........  [d1,k 1 , d1,k 1 ]s k

(10)

Step 2: Determination of the numerator polynomial of the (k-1)th order reduced model by using Pade approximation method       [c 21 , c 21 ]  [c 22 , c 22 ][c 22 , c 22 ]s  .....  [c 2n , c 2n ]s n 1      [c 21 , c 21 ]  [c 22 , c 22 ]s  ........  [c1,n 1 , c1,n 1 ]s n       [d 21 , d 21 ]  [d 22 , d 22 ][d 22 , d 22 ]s  .....  [d 2k , c2k ]s k 1     [d 21, d 21]  [d 22 , d 22 ]s  ........  [d1,n 1 , d1,n 1 ]s k

(11)

Where,         [c21 , c21 ][d11 , d11 ]  [d 21 , d 21 ][c11 , c11 ]                 [c21 , c21 ][d12 , d12 ]  [c22 , c22 ][d11 , d11 ]  [d21 , d21 ][c12 , c12 ]  [d22 , d22 ][c11 , c11 ]

(12)

4. Integral Square Error(ISE) The integral square error (ISE) between the transient responses of higher order system (HOS) and reduced-order system (ROS) is determined to compare different approaches of model reduction, which is given by: 

ISE   [ y (t )  yr (t )]2 0

[294]

(13)

Bhatt et al/COMMUNE – 2015

Where, y (t ) and yr (t ) are the unit step responses of original system H(s) reduced order system R(s). 4.1

Numerical Example

Consider a third order system described by the transfer function as reported by Bandyopadhyay B. et al, 1997

G3 ( s) 

[15,16]  [17.5,18.5]s  [2,3]s 2 N ( s)  [20.5,21.5]  [35,36]s  [17,18]s 2  [2,3]s 3 D( s )

(14)

5.1 Reduction by first mixed method Step 1: For getting the second order model by using Differentiation method. Number of times to be differentiated will be n-k = 3-2 = 1.

D(s)  [20.5,21.5]  [35,36]s  [17,18]s 2  [2,3]s3

(15)

Now by doing reciprocal of D(s) we will get

[20.5,21.5]s 3  [35,36]s 2  [17,18]s  [2,3]

(16)

Differentiating above equation with respect to s

3[20.5,21.5]s 2  2[35,36]s  [17,18]  [61.5,64.5]s 2  [70,72]s  [17,18]

(17)

Now by doing reciprocal again we will get [61.5,64.5]  [70,72]s  [17,18]s 2

(18)

For gain settling according to higher order system this equation is divided by 4: [61.5,64.5]  [70,72]s  [17,18]s 2  [15.375,16.125]  [17.5,18]s  [4.25,4.5]s 2 4

D2 ( s)  [15.375,16.125]  [17.5,18]s  [4.25,4.5]s 2

(19)

(20)

Step 2: Reduction of Numerator N ( s)  [15,16]  [17.5,18.5]s  [2,3]s 2

(21)

[15,16]s 2  [17.5,18.5]s  [2,3]

(22)

Reciprocal of the N (s) is

Differentiating above equation with respect to s 2[15,16]s  [17.5,18.5]  [30,32]s  [17.5,18.5]

(23)

Doing reciprocal again [30,32]  [17.5,18.5]s

For gain settling according to higher order system this equation is divided by 3

[295]

(24)

Bhatt et al/COMMUNE – 2015

[30,32]  [17.5,18.5]s 3

 [10,10.6667]  [5.8333,6.16667]s

N 2 (s)  [10,10.6667]  [5.8333,6.16667]s

(25) (26)

So, the reduced order system by differentiation method is:

R2 ( s) 

[10,10.6667]  [5.8333,6.16667]s [15.375,16.125]  [17.5,18]s  [4.25,4.5]s 2

(27)

Figure 2 shows the step responses of higher order and reduced order interval system. GU, GL is corresponding higher and lower order limit of higher order model and RU, RL is corresponding higher and lower order limit of the reduced order interval system. It can be seen from figure 2 that response for higher order and reduced order interval system are very much similar, is very much accurate.

Fig. 2. Step responses of higher and lower order system by first mixed method

5.2 Reduction by second mixed method Step 1: Reduction of Denominator by Mihailov criterion method as reported by Kranthikumar D. et al, 2011: D( s)  [20.5,21.5]  [35,36]s  [17,18]s 2  [2,3]s 3

(28)

Put s = j𝜔 in the denominator D (s)

D( j )  ([20.5, 21.5]  [17,18] 2 )  j ([35,36]  [2,3] 2 )

(29)

Step 2: The intersecting frequencies are calculated as given in section 3.2 and step 1 [1 , 1 ]  0,[1.0929,1.0981],[3.46414, 4.1833]

(30)

Step 3: The denominator of the second order model is taken as given in equation 6 and coefficient [i , i ] are calculated as in equation 7, 8: , ] D2 ( j )  [1 , 1 ]( 2  [1.0929,1.0981]2 )  j[2 2

(31)

[1 , 1 ]  [17.0011,18.0007] [2 , 2 ]  [31.3826,33.61111]

[296]

(32)

Bhatt et al/COMMUNE – 2015

Step 4: Substitute the values of [1 , 1 ] and [2 , 2 ] in step 3 and also substitute j𝜔 = s. Hence, the denominator D(s) is given by D2 (s)  [17.0011,18.0007]s 2  [31.3826,33.6111]s  [20.3061,21.7052]

(33)

Step 5: Numerator is reduced by Pade approximation as reported by Shamash Y., 1975.  , c  ]  [c  , c  ]s [c0 1 1 [15,16]  [17.5,18.5]s  [2,3]s 2 0  2 3 2 [20.5,21.5]  [35,36]s  [17,18]s  [2,3]s [17.0011,18.0007]s  [31.3826, 33.6111]s  [20.3061, 21.7052]

(34)

[c0 , c0 ]  [14.1670,16.9406] [c1 , c1 ] 

 , c  ],[c  , c  ] are [c0 1 1 0

(35)

[216.2341,443.4788]  [10.0574,21.6331] [20.5,21.5]

calculated as given in equation 12.

So the reduced order system by Pade approximation and Mihailov criterion method is: R2 ( s) 

[10.0574,21.6331]s  [14.1670,16.9406] [17.0011,18.0007]s 2  [31.3826,33.6111]s  [20.3061,21.7052]

(36)

Fig.3. Step responses of HOS and ROS by second mixed method

Figure 3 shows the step responses of higher order and reduced order interval systems. GL, GU is corresponding higher and lower order limit of higher order model and RL, RU is corresponding higher and lower order limit of the reduced order interval system. And it is clearly seen from figure 3 that response for higher order and reduced order interval system is very much similar so this second mixed method is very much accurate then first mixed method. Integral square error is calculated by using equation 13 and above comments is very much proved by table 1. ISE for second mixed method is lower then first mixed method. Table 1. Comparison of ISE of Reduced Order Models

Mixed method of Order reduction First mixed method Second mixed method

ISE for lower limit 0.4858 0.3156

ISE for upper limit 0.5105 0.2247

6. Conclusions Two mixed methods have been presented for reducing the high order continuous time interval system. In these mixed methods numerator and denominator are reduced separately. In first mixed method, both denominator and numerator are reduced by differentiation method. In second mixed method denominator is reduced by Mihailov criterion method, while numerator is reduced by Pade approximation method. Step responses of higher order system and reduced order system are taken for comparative analysis of the mixed methods. Second mixed method is also compared with the first mixed method in terms of ISE which shows that second mixed method gives better ISE. [297]

Bhatt et al/COMMUNE – 2015

References Bandyopadhyay B., Upadhye Avinash, and Ismail Osman, August 1997 “    Routh Approximation for Interval Systems”, IEEE transactions on automatic control, Vol. 42, No. 8. Shamash, Y., 1975 “, Model Reduction Using Routh Stability Criterion and The Pade Approximation, International Journal of Control, 21, 475-484. Parmar, G., 2007 “A Mixed Method for Large-Scale Systems Modelling Using Eigen Spectrum Analysis and Cauer Second Form”, IETE Journal of Research, 53, 2, 89-93. Kharitonov, V. L., 1978, “Asymptotic Stability of an Equilibrium Position of a Family of Systems of Linear Differential Equations”, Differentsial’nyeUravneniya, 14, 2086–2088. Saraswathi,, G., 2007, “A Mixed Method for Order Reduction of Interval Systems”, International Conference on Intelligent and Advanced Systems, 1042-1046. Kranthikumar, D., Nagar, S.K., and Tiwari, J.P., August 2011, “Model Order Reduction of Interval Systems Using Mihailov Criterion and Factor Division Method”, International Journal of Computer Applications (IJCA), Vol. 28– No.11, Pp. 4-8. Kranthikumar, D., Nagar, S.K., and Tiwari, J.P., August 2011 “Model Order Reduction of Interval Systems Using Routh Approximations and Cauer Second Form”, International Journal of Advance Science and Technology (IJAST). Kranthikumar, D., Nagar S.K., and Tiwari, J.P., October 2011, “Model Order Reduction of Interval Systems Using Mihailov Criterion and Cauer Second Form”, International Journal of Computer Applications, 32(6):17-21, Bandyopadhyay B., Ismail O., and Gorez R., 1994, “Routh-Padè approximation for interval systems”, IEEE Trans. Auto. Cont., vol. 39, No.12, pp. 2454-2456. Bandyopadhyay B., Sreeram V., and Shingare P.,2008, “Stable γ-δ Routh approximation for interval systems using Kharitonov Polynomials”, International journal of Information and System sciences , Vol4 4, No.3: 348-361. Ismail O., and Bandyopadhyay B., 1995, “Model Order Reduction of Linear Interval Systems Using Pade Approximation”, IEEE International symposium on circuit and systems. Kranthikumar, D., Nagar S.K., and Tiwari, J.P., November 2012, “Mixed Methods for Order Reduction of Linear Continuous Time Interval Systems”, Control Instrumentation System Conference (CISCON - 2012).

[298]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Design of a Fractional Order Ramp Generator M. R. Dar*, F. A. Khanday Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India

Abstract In instrumentation and communication systems, ramp generators play an important and vital role. They form an important block in oscilloscopes, data converters, voltmeters, image sensors etc. Several design approaches have been reported in the open literature for the ramp generators using integer order circuit techniques. This paper provides hardware practical design of ramp generator using fractional order design technique which has property of domineering ramp generation with external control signal. The properties of the designed circuit which were investigated are, amplitude and noise immunity and these properties were found to get better using the proposed fractional order design. The experimental results were supported by the simulation results to get the idea about the various non-idealities of the practical design.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keyword: Fractional calculus; Fractional Order Circuits; Ramp Generator; Constant Phase Element (CPE); Fractance, Square Wave Generator; Continued Fraction Expansion (CFE)

1. Introduction Fractional calculus and its utility/applications in science and technology is becoming the interesting area of research from last decade and so. As the real world is generally fractional so we need to fractionalize the physical systems to understand their behaviour and enhance the performance of such systems. Fractional calculus and hence the fractional order models are used in fluid mechanics, quantum mechanics, medicine, bioimpedance modeling, etc. Fractional order (FO) models are more adequate for electronic circuit design (Maunday et al-2010) especially for biomedical circuits to handle noisy ECG as geven in (Tsirimokou, et al-2014), automatic control systems (Macahdo, 1997, Podlunby,1999), Neural Networks (Zhou, 2008), Electronic Oscillators (Radwan et al., 2008a 2008 b, Ahmad et al., 2001, Meilanov, 2002), analog Filters (Radwan et al., 2008) and fractional order circuit models used in biology and biomedicine as given in (Rigaud, et al., 1995, Ulgen and Sezdi, 1998, Morais et al., 2010, Ionescu et al., 2009). Various electronic circuits and systems have been designed as discussed above and one of the important circuits among those is ramp generator. Ramp generators have importance in electronic and electrical instruments; it is a circuit in which the output voltage increases linearly within the two limits. The important application of ramp generator is in oscilloscopes, function generators, voltmeters, image sensors, electrical motors and generators etc. In this paper, an externally controlled fractional order ramp generator is proposed which has enhanced performance interms of amplitude and noise performance. The external square wave signal used to control the ramp has been obtained from a square wave generator which has also been designed using fractional order modeling. 2. Fractional Order systems The physical systems which we design conventialy are based on the integer-order differential equations and are classified as first, second or n-th order systems but in fractional order systems, the order of differential equations is fractional typically ½ which was first introduced by a famous mathematician L. Hospital in 1695. So, as for as fractional order mathematics is concerned, differential equations are not essentially of integer-order (Oldham, Spainer, 1974, Samko, 1987). Various mathematicians have proposed, various mathematical definitions, theorems and formulas

*

Corresponding author. Tel.: +91 9797 993077. E-mail address: [email protected] ISBN: 978-93-82288-63-3

Dar and Khanday / COMMUNE – 2015

for differentials/integrals, but among those popular one is given by Riemann-Liouville known as Riemann-Liouville Definition of fractional derivative/integral (Miller and Ross B, 1993) and is given as 

D

a

t

1 d f (t )    (n   )  dt 

f ( )

n t

 a

(t  )

(  n 1)

d ;

(n  1)    n

(1)

Where n is integer, α is fractional order (  ) and a, t are limits. This is forward derivative and is also called semi-derivative.

aI

 t



t Ib

f (t )  f (t ) 

1  1 

t

 1

 (t u)

f (u )du ……………...........................…is forward integration

a b

 1

 (u t )

f (u )du ……...........................…………is backward integration

t

It is worth to note that kernel in forward integration is

t  u  1 and kernel in backward integration is u  t  1 .

A more powerful definition of fractional derivative is M. Caputo’s fractional derivative (1967) (Samko, S. Get-al- 1987) C a



D

t

1 f n ( ) f (t )  d ; (n   ) a (t  )( n1) t

(n  1)    n

(2)

Where n is integer, α is fractional order (  ) and a, t are limits. Fractional calculus plays vital role in enhancing the performance of physical systems and understanding the physical laws in better way especially in the field of technology. From last decade and so a significant number of researchers and prominent research institutes are working towards the development of the fractional order electronic systems/circuits as these have better performance over their integer order counterparts. These systems/circuits are developed by using fractional capacitor (Fractance) also known as constant phase element (CPE) instead of a normal capacitor.

3. Fractance or Constant Phase Element (CPE) As the fractional order systems/circuits are gaining more and more importance and is widely accepted among scientific and industrial community in analog signal processing, bioimpedance modeling, automatic control etc. but there is lack of fractance in its component form like capacitor, resistor and inductor, and also there is lack of software simulation tools/packages to support the simulation of fractional order circuits and systems. Hence for now it requires integer-order mathematical approximations to approximate the fractance; these approximations are typically realized using various combinations of R, L, C ladders as per various synthesis methods (Carlson, 1964, Valsa and Vlach, 2011). A fractance or constant phase element (CPE) is one whose impedance is proportional to z s  

1 , where α is s C 

arbitrary having typical values between 0 and 1 (Biswas, et al., 2006). In such a device, the phase difference between the voltage and the current is  . The various methods to approximate fractance are continued fraction expansion 2

(CFE) (Podlubny, et al., 2002), rational approximation methods and partial fraction expansion or self similar trees (Nakagawa et al., 1992) etc. The circuit to realize a fourth-order approximation of a CPE using the CFE method in (Krishna and Reddy, 2008) is shown in Fig. 1(a), the magnitude and phase response of CPE is shown in Fig. 1(b)

[300]

Dar and Khanday / COMMUNE – 2015

(a)

(b) .

Fig.1 Passive equivalent of CPE and the used symbol. (b) Magnitude and phase response of CPE centered over 1kHz.

4. Fractional order Ramp generator The circuit diagram of proposed externally controlled fractional order ramp generator is shown in Fig.2. Let

C be

1 where 0    1 and typically   0.5 . In the circuit shown s C in Fig.2, the transistor has been placed across the fractional order capacitor C . The base resistor Rb limits the current the fractional order capacitor whose impedance is



flowing through the base of the transistor. However,

R

b

is to be kept small to ensure that the transistor is driven into

saturation during the on-state of the clock, with a zero or negative control input voltage, the transistor is off. The capacitor charges through Rin from output of operational amplifier. The charging rate is given by V . The

R .C in

frequency of the ramp signal remains same as the square wave control signal input.

5. Fractional order Control signal/square wave generator The control signal fed to the ramp generator was generated using the fractional order square wave generator as shown in Fig. 2 which has the property of increased frequency generation (Maunday et al-2010) as compared to the integer-order square wave generator.

Fig. 2. Proposed Fractional order ramp generator.

[301]

Dar and Khanday / COMMUNE – 2015

Fig. 3. Fractional order control signal/square wave generator

6. Experimental and Simulation Results To verify the results for the circuit of Fig. 2, it was constructed on bread board and simulated in Multisim. A fractional capacitor of value 1uF was mathematically approximated using the partial fraction decomposition and is implemented in actual hardware to a fourth order, where the values of components used are, Rin=1.402KΩ, Ra=3.17KΩ, Rb=4.78KΩ, Rc=11.2KΩ, Rd=92.9KΩ, Ca=6.64µF, Cb=0.023µF, Cc=0.043µF, Cd=0.055µF. In order to design control signal generator, µA741 op-amp was used and the values of the passive components used were Cα=1µF , R1=1KΩ, R2=R3=10KΩ, Fig. 4(a) and Fig.4(b) shows the simulation and experimental results of fractional order square wave generator and here the frequency of oscillations was 45Hz for integer order and 4.36KHz for fractional order square wave generator (i.e. thousand times increase in frequency generation as in (Maunday et al-2010)) when the same value of capacitor and resistors were used that shows the usefulness of fractional order circuits (Maunday et al-2010). Using the same square wave generator parameters with   0.5 and Cα=1µF, the fractional order ramp generator was realized on bread board as shown in Fig. 5 and was simulated in Multisim, and the obtained simulation, experimental results of ramp generator is thereof shown in Fig. 6(a) and 6(b) respectively. Fig. 7(a) and 7(b) shows the simulation and experimental results of integer-order ramp generator and the results shows that fractional order ramp generators have better output amplitude and noise performance which is required in digital voltmeters, function generators, image sensors etc. 7. Conclusion The usefulness of the CPE in ramp generators is demonstrated in this paper. The amplitude and noise performance of the proposed circuit is shown to be much better for the same control input, amplitude and time constant in the integer order ramp generator. This opens up the possibility of greater control with much better amplitude, better noise performance in ramp generators as physical fractional capacitors soon become available which will make the linearity of the output better as well. Thus the proposed fractional ramp generator will be useful addition to the field of analog signal processing.

(a)

(b)

Fig. 4. (a) Simulation result of Fractional square wave generator (b) Expt. result of Fractional square wave generator.

[302]

Dar and Khanday / COMMUNE – 2015

Fig. 5. Bread board realization of Ramp generator.

(a)

(b)

Fig. 6 (a) Simulation results of proposed ramp generator, and (b) Experimental results of proposed ramp generator.

(a)

(b)

Fig. 7. (a) Simulation results of integer order ramp generator, and (b) Experimental results of integer order ramp generator.

References Maundy, B., Elwakil, A., Gift, S., 2010, On a multivibrator that employs a fractional capacitor, Analog Integrated Circuits and Signal Processing, 62(1), p. 99. Tsirimokou G., Psychalinos, C., 2014, Ultra-low voltage fractional-order differentiator and integrator topologies: an application for handling noisy ECGs, Analog Integrated Circuits and Signal Processing, 81(2), p. 393. Machado, J. A. T., 1997, Theory analysis and design of fractional-order digital control systems, Journal of System Analysis, Modelling and Simulation, 27(2–3), p. 107. Podlubny, I., 1999, Fractional order systems PIλDµ controller, IEEE Transactions on automatic control, 44, p. 208. Zhou S., Li, H., Zhu, Z., 2008, Chaos control and synchronization in a fractional neuron network system, Chaos, Solitons and Fractals, 36, p. 973. Radwan, A. G., Elwakil, A. S., Soliman, A. M., 2008, Fractional-Order Sinusoidal Oscillators: Design Procedure and Practical Examples, IEEE transactions on circuits and systems—I, 55 (7), p. . Radwan, A. G., Soliman, A. M., Elwakil, A. S., 2008, Design equations for fractional-order sinusoidal oscillators: Four practical design examples, International Journal of Circuit Theory and Applications, 36, p. 473. Ahmad, W., El Khazali, R., Elwakil, A. S., 2001, Fractional-order Wienbridge oscillator, Electronics Letters, 37(18), p. 1110. Meilanov, R., 2002, Features of the phase trajectory of a fractal oscillator, Technical Physics Letters, 28, p. 30,

[303]

Dar and Khanday / COMMUNE – 2015 Radwan, A. G., Soliman, A. M., Elwakil, A. S., 2008, First-order filters generalized to the fractional domain, J. Circuits Systems and Computers, 17, p. 55. Rigaud, B., Hamzaoui, L., Frikha, M. R., Chauveau, N., Morucci, J. P., 1995, In vitro tissue characterization and modelling using electrical impedance measurements in the 100 HZ—10 MHz frequencyrange, Physiological Measurement, 16 (3A), p. A15. Ulgen Y., Sezdi, M., 1998, Hematocrit dependence of the Cole-Cole parameters of human blood, Proceedings of the 1998 2nd International conference on Biomedical Engineering Days, p. 71, Morais, A.P., Pino A.V., Souza, M.N., 2010, A fractional electrical impedance model in detection of occlusal non-cavitated carious, 2010 Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), p. 6551. Ionescu, C. M., Keyser, R. D., 2009, Relations between fractional-order model parameters and lung pathology in chronic obstructive pulmonary disease, IEEE Transaction on Biomedecal Engineering, 56(4), p. 978. Oldham, K. B., Spainer, J., 1974, Fractional Calculus, New York: Academic. Samko, S. G., Kilbas, A. A., Marichev, O. I., 1987, Fractional Integrals and Derivatives: Theory and Application, New York: Gordon & Breach. Miller K. S., B. Ross, 1993, An Introduction to the Fractional Calculus and Fractional Differential Equations, New York: Wiley. Carlson, G. E., and Halijack, C. A., 1964, Approximation of fractional capacitors 1 / S 1 / n by regular Newton process, IEEE Transactions on Circuit Theory, 11(2), p. 210. Valsa, J., Vlach, J., 2013, RC models of a constant phase element, International Journal of Circuit Theory and Applications, 41, p. 59. Biswas, K., Sen, S., Dutta, P. K., 2006, Realization of a constant phase element and its performance study in a differentiator circuit, IEEE Transactions on Circuits and Systems II: Analog and Digital Signal Processing (USA), 53(9), p. 802 Podlubny, I., Petras, I., Vinagre, B., O’Leary, P., Dorcak, L., 2002, Analgoue realizations of fractional-order controllers, Nonlinear Dynamics, 29(1– 4), p. 281, Nakagawa, M., and Sorimachi, K., 1992, Basic characteristics of a fractance device, IEICE Transactions on Fundamentals of Electronics Communications and Computer Sciences (Japan), E75-A(12), p. 1814. B. Krishna, Reddy, K., 2008, Active and passive realization of fractance device of order 1/2, Active and Passive Electronic Components, Article ID 369421, 5 page.

[304]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Structural and Dielectric Studies of RFeO3(R=Pr, Eu and Ho) Khalid Sultana, Sajad Ahmad Mira, Zubida Habibb, M. Ikrama* a Department of Physics, National Institute of Technology, Srinagar, India. Department of Chemistry, National Institute of Technology, Srinagar, India

b

Abstract

The Rare-earth transition-metal oxide (RETMO) RFeO3(R=Pr, Eu and Ho) were synthesized by the solid-state reaction route. X-ray diffraction (XRD) was investigated to confirm chemical phase and the orthorhombic Pbnm structure. By varying rare earth ion, the unit cell volume and lattice parameters undergo non-monotonous changes. The slight shift in peaks towards higher 2θ for different R describes the lattice contraction, which is due to different ionic radii of rare earth ion in the samples.SEM images reveal that the average grain size of HoFeO3 is less than PrFeO3 and grain size of PrFeO3 is less than EuFeO3.Measurement of dielectric loss and dielectric constant were performed with respect to temperature and frequency. The dielectric constant and dielectric loss decreases as the ionic radii of rare earth ion decreases i.e., dielectric constant and loss of PrFeO3 is greater than EuFeO3 but it increases for HoFeO3. © 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Solid State Method; XRD; Dielectric Study; Small Polaron

1. Introduction Rare-earth transition-metal oxide (RETMO) compounds with perovskite ABO3 structure have been attracting renewed attention in connection with the discovery of high critical temperature ‘Tc’ superconductivity, colossal magnetoresistance and multiferroicity. They are promising eco-friendly material to replace toxic lead-based perovskite relaxers, sensors, capacitors and optical storage devices (Samara et al, 2003; Sultan et al , 2014). As one of the typical RETMO, Orthoferrite PrFeO3 belongs to perovskite family and has ABO3 distorted orthorhombic GdFeO 3 type perovskite structure with space group pbnm. In this structure, the distortion of the Fe octahedron is small and almost independent of ‘R’ (rare earth ion) (Marezio et al, 2008). However, the distortion of rare earth polyhedra is large and increases with decreasing ionic radius of R. As distortion increases, the 12 oxygen ions surrounding the ‘R’ separate into two types: R with eight first-nearest O ions and R with four second-nearest O ions. Such structural distortions influence the magnetic ordering and spin-state transitions (Marezio et al, 1971; Sultan et al , 2014; Sultan et al , 2015). 2. Experimental Bulk samples of RFeO3(R = Pr, Eu, Ho) were prepared by solid-state reaction route. Precursors of Pr6O11, Eu2O3, Ho2O3 and Fe2O3 were taken in the stoichiometry ratio and the compounds of the composition PrFeO3, EuFeO3and HoFeO3 were synthesized were ground to fine powder, pressed to the pellet form, and sintered at 1250 °C for 24 hours. Thecalcinated material was analysed by X-ray diffraction (XRD) and scanning electron microscope (SEM).The dielectric measurements as a function of frequency and temperature was made by using Agilent 4285A precision LCR meter.

* Corresponding author. Tel.: +91 8717 000375. E-mail address: [email protected].

ISBN: 978-93-82288-63-3

Sultan et al/ COMMUNE-2015

3. Results and Discussions 3.1.

Structural Analysis

3.1.1

XRD Analysis

The XRD patterns of the prepared samples are shown in Fig. 1. All the detectable peaks could be assigned as the RFeO3 orthorhombic structure with space group Pbnm. The shifts of peak positions were observed as the R-site rare earth ion was varied. The slight shift in peaks towards higher 2θ for different R describes the lattice contraction, which is also highlighted in inset of Fig. 1. This behaviour is due to different ionic radii of rare earth ion in the samples. Since the ionic radii of Ho is less as compared to Eu which in turn has less ionic radii than Pr so the XRD shows shift in peaks towards higher 2θ as shown in inset. Table 1 shows the calculated parameters and it is evident that there is an overall decrease in lattice parameters. The unit cell volume also shows the same pattern. This behaviour may be assigned to the different ionic radii of rare earth ions as explained above.

Fig. 1: The XRD patterns of the of RFeO3(R= Pr, Eu, Ho) Table I: The calculated structural parameters of the rare earth orthoferrites RFeO 3(R = Pr, Eu and Ho). Compound PrFeO3 EuFeO3 HoFeO3

3.1.2.

Crystal System (a) Space Group (b) Orthorhombic Pbnm Orthorhombic Pbnm Orthorhombic Pbnm

Lattice Parameters (c) 5.558 5.562 7.855 5.385 5.601 7.702 5.284 5.589 7.608

Cell volume (A3) 242.87 232.30 224.75

SEM Analysis

The in-depth external morphological studies of the samples were made using the Hitachi S-3000H scanning electron microscope (SEM). Typical SEM micrographs of RFeO 3(R = Pr, Eu and Ho) sintered at 12500C for 24 h are shown in Fig.2 (a-c).

Fig.2. (a-c) SEM micrograph of (a) PrFeO3 (b) EuFeO3 and (c) HoFeO3

The image resolution was set at 2,500X.The microstructure reveals that the particles are spherical in shape with distinguishable boundaries. From the Fig. 2, it is clear that the average grain size of HoFeO 3 is less than PrFeO3 and grain size of PrFeO3 is less than EuFeO3.But the grains and grain boundaries of HoFeO3 increases as compared to PrFeO3 and EuFeO3 which is the reason why dielectric constant and ac conductivity of HoFeO 3 is more than other two rare earth orthoferrites.

[306]

Sultan et al/COMMUNE – 2015

3.2

Dielectric studies

(a) Dependence of Dielectric Behaviour on Frequency The variation of dielectric constant as a function of frequency of RFeO3 (R = Pr, Eu and Ho) in an ac field ranging from 20 Hz-1 MHz is illustrated in Fig.3. From the Fig.3 it is observed that the value of dielectric constant decreases continuously with increase in frequency for all the three compositions. The decrease of dielectric constant (ε') and dielectric loss tangent (tan δ) [shown in Fig.4] with frequency is a general dielectric behaviour of ferrites.

Fig. 3.The variation of dielectric constant with frequency of RFeO3(R= Pr, Eu and Ho).

Fig.4.The variation of dielectric loss with frequency of RFeO3(R = Pr, Eu and Ho) at room temperature.

This type of behaviour has been reported by many investigators (lal et al, 2004; Ravinder et al, 2001; Hangloo et al, 2003; bhat et al , 1995) and can be explained on the basis of polarization mechanism. There are four primary mechanisms of polarization in materials i.e. ionic, electronic space charge and dipolar polarization. All the mechanisms of polarization contribute to the dielectric constant. At low frequencies, while the contributions from different polarizations starts decreasing at higher frequencies. For example, at very high frequencies of the order of 1015 Hz, only electronic polarization contributes to the dielectric constant. The steady behaviour in dielectric constant at higher frequencies indicates the inadequacy of electric dipoles to follow the variation in frequencies due to alternating applied electric field or it can be inferred that the electronic exchange between the ferric and ferrous ions i.e. , Fe2+ ↔ Fe3+ cannot follow the alternating field. From above figures, it is evident that dielectric loss and dielectric constant decreases as the ionic radii of rare earth ion decreases i.e., dielectric constant and loss of PrFeO 3 is greater than EuFeO3 but it increases for HoFeO3. From the literature (Berenov et al, 2008) it has been seen that the distortion in the FeO6 octahedron decreased from La to Gd and again increased in HoFeO3. This reason may be attributed to the anomaly observed. (b) Dependence of Dielectric Behaviour On Temperature The variation of the dielectric constant and dielectric loss with temperature at 400 KHz for RFeO3 (where R = Pr, Eu and Ho) is shown in Fig.5 and 6. Both dielectric loss and dielectric constant behaves independently at lower temperature while as it increases with temperature at higher values of temperature. This behaviour at higher temperature may be attributed to generation of extra thermal energy which enhances the mobility of charge carriers hence the rate of hopping increases. The thermal energy at low temperature does not contribute to mobility of charge carriers. This mechanism observed sets up the higher polarization at higher temperature which increases the dielectric constant. It is observed that both the dielectric constant and dielectric loss decreases as the ionic radii of rare earth ion decreases but in HoFeO3 it shows increase in these properties, which is again due to structural deformation.

[307]

Sultan et al/ COMMUNE-2015

Fig. 5. Shows the variation of dielectric constant with temperature at 400 KHz of RFeO3(R = Pr, Eu and Ho)

Fig.6. Shows the Variation of dielectric loss with temp.400 KHz of RFeO3(R = Pr, Eu and Ho)

4. Conclusion The rare earth orthoferrites RFeO3 (R=Pr, Eu and Ho) were successfully synthesized employing solid-state reaction method. These synthesized samples crystallize in an orthorhombically distorted perovskite phase with space group Pbnm. Phase formation was confirmed by X-ray diffraction. Varying rare earth ion results in significant changes in the physical properties of samples. The calculated structural parameters calculated from the XRD measurements change as the rare earth ion is varied. The slight shift in peaks towards higher 2θ for different R describes the lattice contraction, which is due to different ionic radii of rare earth ion in the samples. From SEM images, it is observed that the average grain size of HoFeO3 is less than PrFeO3 and grain size of PrFeO3 is less than EuFeO3. The dielectric constant and dielectric loss decreases as the ionic radii of rare earth ion decreases i.e., dielectric constant and loss of PrFeO 3 is greater than EuFeO3 but it increases for HoFeO3.The alteration in the ac conductivity identifies that the conduction mechanism follows the charge hopping between localized states and follow the small polaron conduction Acknowledgement Authors would like to thank IUAC, New Delhi for experimental facilities and Director NIT Srinagar for the encouragement provided during work. References G.A. Samara, J. Phys.: Condens. Matter. 15, 367 (2003). Khalid Sultan, Zubida Habib, Asima Jan, Sajad Ahmad Mir, M. Ikram, K. Asokan Adv. Mat. Lett. 2014, 5(1), 9-13 M. Marezio, J.P. Remeiko, P.D. Dernier, Acta Crystallogr. B 269, 2008 (1970). M. Marezio, P.D. Dernier, Mater. Res. Bull. 6, 23 (1971). K Sultan, M. Ikram, K. Asokan Vacuum 99 (2014) 251-258. Khalid Sultan, M. Ikram, Sanjeev Gautam, Han -Koo Lee, Keun Hwa Chae and K.Asokan. Journal of Alloys and Compounds 628 (2015) 151–157. B. lal, S.K. Khosa, R. Tickoo, K.K. Bamzai, P.N. Kotru, Mater. Chem.& Phys. 83, 158-168 (2004). D. Ravinder, K. Vijay Kumar, Bull. Mater.Sci. 24, 505-509 (2001). V. Hangloo, R. Tickoo, K.K. Bamzai, P.N. Kotru, Mater.Chem.Phys.81, 152-159 (2003). S. Bhat, S.K, Khosa, P.N. Kotru, R.P. Tandon, J. Mat. Sci. Lett. 14, 564-567 (1995). A. Berenov, E. Angeles, J. Rossiny, E. Raj, J. Kilner, A. Atkinson,Solid State Ionics 179, 1090–1093 (2008) .

[308]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16-18March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

WSN Based Secure Ambient Intelligent Hospitals A. H. Moon, Ummer Khan, Zaffar Kanth, Sheikh Junaid* National Institute of Electronics and Information Technology, Srinagar, India

Abstract Ambient Intelligence (AmI) is the technology appropriate for modern buildings, homes, and public infrastructure. The WSN is recognized as one of the technological corner stone of AmI. This paper proposes the design of a WSN based multi-Communication standard network for the intelligent monitoring of the fragile hospital environment. The proposed network includes sensor nodes to monitor air quality, water quality and crowd inflow inside various hospital buildings across the state of J&K. The outcome would be a system composed of ad-hoc sensor nodes providing the hospital administration with adequate non-intrusive and intelligent monitoring to improve the quality of hospital hygiene, cleanliness, and sanitation. The proposed design integrates WSN with packet switched networks/ web service communication into a knowledge and information service platform. Such a platform can support health administrators/ managers to improve the quality of ambient surroundings of hospitals. The design incorporates the custom developed authentication protocol based on hidden generator concept using Elliptic Curve Cryptography (ECC) suitable for low rate WSNs. © 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Ambient Intelligence; WSN; Base-Station;TinyOS; nesc; ECC; DSSS

1. Introduction Hospitals play a key role in supporting human survival and health standards. However, these critical pillars of life support are also prone to various dangers both intentional and unintentional like fire, pollution, contamination, noise etc. In order to make these temple’s of health fit for their purpose, effective mechanisms are needed to better monitor their ambient behavior. Revolution in sensor technology has made it possible to quantify any measureable characteristics of ambient phenomenon. Today we have line of sensors to measure air quality (Smoke, odor, suspended particles), water quality (pH, palatability etc), and other physical phenomenon. Sensors can be employed to count events (to count people, vehicles, or other events). These Sensors are available in various form factors and hence can be deployed across the range of surroundings. Due to the advances in sensor technology in recent years, wireless sensor networks (WSN) have extensively been employed for environmental monitoring, health monitoring and industrial monitoring (Lynch and Loh, 2006). WSNs can be used to monitor delicate environments and can prove vital to prevent environmental degradation by alerting the authorities about the possibility of a disaster. Although sensor networks have been utilized for numerous environmental monitoring tasks but their potential to monitor complex indoor habitats like hospitals has not been analyzed so far. By integrating WSN devices in the hospital environment, the level of protection against pollution, infections etc could be substantially raised, giving the hospital many new intelligent features. This paper proposes various design aspects related to implementation of WSN based hospital environs health monitoring system. The proposed system employs the use of the most modern sensors to monitor air quality, water quality, noise, crowd etc. in various hospitals across the state of J&K. The sensor data collected from hospitals would be securely transferred to a local base station system. The base station at each hospital is connected to a central database through the Internet cloud. The local base station would provide the hospital-health information to the local hospital administration. The central database can be interfaced with a web application to provide the status of the all hospitals.

2. Motivation Words like odour, waste, and pollution are almost synonymous with hospitals of India in general and J&K in particular. Waste materials speckled all around, crowded wards and waiting rooms are a grim reminder of the unhygienic condition of our health institutions. This is also a common scenario at J&Ks major health institutions, which has become a cause of worry for patients and * Corresponding author. Tel.+91 9622 773240. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Moon et al/COMMUNE – 2015

doctors. The motivation to choose this application for WSN and AmI integration is two folds. The biologists involved value the possibility to monitor the hospital environment inconspicuously, while for the technologists, they saw an opportunity to analyze the vision of WSN security in a real application. AmI and emergence of new computing devices, advances in sensor networking, smart sensors and other devices gives us the methods to come up with novel applications to better assist healthcare sector in general and its dependents in particular. We believe that people residing in hospitals are the ones who need the cleanest air to breath, palatable water to drink and a feasible environment that assists to their health recovery.

3. Recent Advances Environmental monitoring through sensor networks has evolved over time and improved in many studies (Mainwarring and Polystre, 2002) (Reis and Camara, 2008). As a large-scale sensor network deployment, Habitat monitoring system was developed using one hundred motes (R. Szewczyk, A. Mainwarring, 2004). The system was deployed to observe the effect of microclimatic changes on seabird nests. Citysee (X. Mao, X. Miao, Y. He, 2012), is a Co2 monitoring system designed for monitoring Co2 levels in an urban area in China. This paper mainly focuses on sensor node deployment problem. In (S. Choi, N. Kim, 2009), an air pollution monitoring system was implemented in outdoor setup. Various sensor devices were set up and sensor-monitored data were transmitted to the server. The server after doing some calculations based on the domain knowledge transmits the signal to an alarm system. The IEEE 1451 is a set of smart transducer interface standards developed to increase the interoperability and speed up the expansion of smart sensors and actuators usage within the area of wired and wireless network. ”Casattenta” (Elisabetta, Mirko 2009) is the demonstrator of a research project on ”Ambient Intelligence”, “Sensor Fusion” and “Wireless Sensor Networks”. The system which is based on Zigbee wireless protocol consists of fixed smart sensors distributed in the environment and wearable ones monitoring inhabitant’s health and activity. (M. A. Pillai, S. Veerasingham, 2010) presents a simplest air quality-monitoring module based on CAN (Controller Area Network) protocol. Sensor nodes consist of CAN transceiver and CAN controller. Each node is interfaced with VOC (Volatile Organic Compound) sensors, continuously monitoring environment and putting sensor data into CAN bus. Although the above discussed works all implement the environment monitoring, but most of them are outdoor systems with emphasis mainly on deployment and network feasibility in real time. However, for indoor environments like hospitals, various key issues like security, sensor technology and usability have seldom been analyzed. In this paper, we aim to propose a WSN based multi-communication monitoring system with emphasis on the aforementioned issues.

4. Design Approach The proposed secure ambient intelligent monitoring system based on WSN is illustrated in fig. 1. It is comprised of three tiers: monitoring sensor nodes, local base station node and remote monitoring centre. The proposed design encompasses the facilities of various hospitals across the state of J&K. At each hospital building, a large number of different types of sensor nodes (known as motes) silently deployed. A sensor node or a mote has sensing, processing and communicating capabilities. These minimally invasive motes are spread and placed at various locations across the hospital structure and are transparent to the public. The sensor position is to be chosen in a manner that leads to the accurate measurement of the concerned parameters. The heterogeneous set of sensors includes but is not limited to sensors like smoke detectors, palatability sensors, odour sensors, poisonous gas detectors, fire detectors, dissolved oxygen sensors, noise measurement sensors, counter sensors and many more. Fig. 2 depicts a sample sensor node layout. Some of the sensor nodes like smoke detectors, odour sensors, fire detectors etc need to be deployed all over the hospital including the patient wards, OPD’s, Corridors, intensive care units and elsewhere. The palatability sensors, dissolved oxygen sensors etc can be immersed in the water tanks that are the sources of the drinking water distribution system. Poisonous gas detectors and odour sensors can be deployed at locations like garbage collection points, laboratories, kitchens and many other locations where need is felt. Directional Count Sensors are deployed at all the hospital entrances to keep a count on number of hospital visitors present at a particular point of time. In the proposed design nodes will work independently and transmit the collected sensor parameters wirelessly over a WSN network towards the base station system. These nodes are programmed to work collectively to form a multi-hop network. The nodes do participate and exchange routing information to establish best possible route towards the base-station so that the sensor data can be routed to the base-station via multiple hops.

Fig 1: Proposed Architecture

[310]

Moon et al/COMMUNE – 2015

The base-station node is specially programmed (TOSBASE application) to act as Gateway between the WSN and the Packet switched network. The base Station node is connected to a base-station system that hosts a visualization and analysis application. This application can provide vital real time hospital health information to the concerned administration which can initiate a response to control the ambient parameters of the hospital. The base-station systems at every hospital facility further transfers the local hospital data to a global database via an Internet cloud setup. The Remote monitoring center accesses the global database to provide the real-time status of all the hospitals, which could be put to use by the general health administration. The proposed architecture would employ mote works platform. Moteworks nodes like micaz motes are 802.15.4 compliant and are based on CC2420 radio with upto 250 kbps data rate. CC2420 uses a digital direct sequence spread spectrum (DSSS) with OQPSK modulation and provides a spreading gain of 9 dB and an effective data rate of 250 kbps [ Jan Hinrich, Vlado, 2009] . The platform offers features like self-configuring, self healing, low power, XMESH based topology which requires minimal human intervention. The XMESH employs a power efficient variant of Adhoc Distance Vector Routing Protocol (AODV) to identify the best possible route from an end node to the base station node. These motes are tinyOS based and hence are custom programmable using embedded nesc applications. The nodes in the proposed network could be programmed with intelligent applications capable to dynamically adjust the sampling rates, manage the duty cycles to save energy and ensure secure transmission of messages to authenticated nodes. The base station system is configured with XSERVE a gateway application that logs the sensor data to local and global databases.

4.1.

Security

Since the deployed sensors are heterogeneous and the application is sensitive (related to health). Therefore, providing security with such critical application is essential. Because of the sensitive nature of application authentication mechanisms can be used to ensure that the data is coming from an authentic entity. Usage of ECC suits the requirement of building appropriate security protocols for resource constrained WSN.

Fig 2: Sensor Layout

A customized authentication protocol based on Zero Knowledge Protocol (ZKP) using hidden generator concept would be observed in the proposed architecture to prevent unauthorized capture and transport of the sensor data. For resource constrained WSN, ZKP is very useful as it requires less bandwidth, small computational power and less memory. The hidden generator concept helps to harden the protocol against an active attack like man-in-the-middle. The proposed security suit encompasses the following protocols developed:

a) Node- to -Node zero knowledge authentication protocol using split shares. b) Base -to -Node authentication Protocol. c) Node -to -Base authentication Protocol In the split share process the prover uses a secret r and divides it into two shares for achieving authentication. The usage of Private Key of prover and Shared Generator Gs makes the algorithm robust against attacks like impersonation, man-in-the-middle etc.

4.2

Sensor Implementation

A sensor is a transducer that detects a physical quantity like temperature etc and responds with an electric signal. The sensor selection should be based on several criteria such as low-power consumption, low cost and high precision. The low cost and low power sensors with additional temperature and humidity sensing capabilities are suitable for energy efficient ambient monitoring system that can measure pollutants and other environmental parameters. The specifications of the sensors to be used in the system are given in Table 1.

[311]

Moon et al/COMMUNE – 2015

Table 1: Identified Set of Sensors

Sensor Name

Range

Power Consumption

Response

Operating

LM35CZ Precision

-400C - 110 0C

1.7673 mW

2 sec Time

5.0 V Voltage

HIH-4000 Temp Sensor

0%-100%

1 mW

15 sec

5.0 V

COSensor Humidity

0.5ppm – 20ppm

9.093 mW

60 sec

7-9 V

CO2

50 ppm – 800ppm

1.1434 mW

5 min

3 V-3.3 V

NO2

0.01ppm-0.5ppm

1.767 mW

2 sec

5.0 V

SO2

0.04ppm-5ppm

12.722mW

60 sec

7 V-9 V

5. Conclusion The design process and implementation details of a sensor network based Ambient intelligent hospital were described. Since designing aWSN based system means working closely with hardware. Getting a design logically right does not guarntee that a function actually works when deployed in real time. As a continuation to this work, tests will begin on the physical implementation of the sensor network. A test bench will be setup where different implementation issues will be analysed in real time. The test bench would be employed to test the performance of the custom authentication protocol purposefully developed for the Low rate WSN based monitoring systems.

References Lynch J.P. and Loh. K.J. (2006). A Summary review of wireless sensors and sensor networks for structural health monitoring, Shock and Vibration Digest. Mainwaring, A.; Polastre, J.; Szewczyk, R.; Culler, D.; Anderson, J. Wireless Sensor Networks for Habitat Monitoring. In Proceedings of ACM International Workshop on Wireless Sensor Networks and Applications, Atlanta, GA, USA, 28 September 2002, pp. 88-97 Reis, I.A.; Câmara, G.; Assunção, R.; Monteiro, A.M.V. Data-Aware Clustering for Geosensor Networks Data Collection. In Proceedings of Anais XIII Simpósio Brasileiro de Sensoriamento Remoto, Florianópolis, Brasil, 21–26 April 2007, pp. 6059-6066 R. Szewczyk, A. Mainwaring, J. Polastre, J. Anderson, and D. Culler, “An analysis of a large scale habitat monitoring application, Proc. Int. Conf. Embedded Networked Sens. Syst. (SenSys 04), ACM Press, Nov. 2004, pp. 214-226. X. Mao, X. Miao, Y. He, X. Y. Li, and Y. Liu, “Citysee: Urban CO2 monitoring with sensors, Proc. IEEE INFOCOM, IEEE Press, Mar. 2012, pp. 1611-1619. S. Choi, N. Kim, H. Cha, and R. Ha, “Micro Sensor Node for Air Pollutant Monitoring: Hardware and Software Issues”, Sensors2009,Oct. 2009, pp. 7970-7987 Farella E, Falavigna M, Ricco B, “advances in sensors and interfaces, 2009. IEEE conference IWASI 2009, 3rd International workshop. Jan Hinrich Hauer, Vlado H, Adam Wolisz, Experimental Study of the impack of WLAN Interference on IEEE 802.15.4 BAN, 6th European Conf. On WSN feb2009

[312]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

QCA Full Adder Design and Noise Problems Shah JahanWania*, Zahoor Ahmad Peera, Fasel Qadira, K. A. Khanb a

Department of Computer Science, University of Kashmir, Srinagar, India b Govt. Degree College, Beerwah, Budgam, J&K, India

Abstract Nanotechnology based QCA basic logic layouts work well while simulating on QCA Designer software. The design of half adders has been reported by various researchers and most of the circuit combinations produce excellent results on simulation. Although design of QCA full adders has been reported by a large number of researchers but the simulation results on most of the circuits fail to produce claimed outputs as such the carry forward work is becoming difficult. There are certain difficulties in producing desired simulation results while joining two half adders to design full adder. The main cause for the problem seems to be the mismatch of input data, when earlier stage data is combined with the forward stage. We have tried to study some full adder designs on the basis of simulation results to move forward with our own designs for the purpose. With analysis of few reported adders, we have proposed two full adder circuit layout designs with analysis for efficient implementation and future improvements.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: QCA; Tunneling; Microelectronics; Boolean Algebra; Electrostatic Noise; QCA Designer; Full Adder; Simulation Result; Single Layer Layout Introduction

1. Introduction The system of device design based on electron positioning gave birth to concept of quantum-dot in early eighties. A quantum-dot thus can be defined as a nano-scale vessel in which an electron can be trapped. Dot as such can be termed as a potential well or ring in which a sufficiently low energy electron can be trapped. There are several ways of implementing quantum-dot and the tested and commonly used is Aluminum metal-dot developed with the help of electron lithography. The electrons in cell-dots can crossover through a technique known as Dolan bridge technique (Fulton, Dolan, 1987) that physically embeds a capacitor junction between the dots in a quantum cell. As the repulsive force is encountered by an electron from the adjacent electron it tunnels through Dolan bridge junction to the adjacent vacant dot. The dots produced with the help of electron beam lithography are not having a similar shape but varied shapes depending on the process and application. There are various processes of producing these devices and one of big challenges to achieve the objective is the precise location of quantum dots at the desired locations. Self organization is one of these processes and occurs when molecules of one crystal structure is deposited on the top of another. The lattice structure difference results in high stresses at the point of contact as such the material tends to clamp up at the point of contact in a manner of depositing oil on water. Although this process can produce dots of incredibly small size but one big problem is that the dots are not located at desired places. 1.1.

Quantum-dot Cell

A cell is a device used to store and transmit data using electrons and the Columbic interactions. The electrons change orientations from 0 (zero) to 1 (one) or vice-versa by changing positions through electron tunneling. A four dot quantum cell with two excess electrons is shown in figure 1 representing two binary states of 0 (zero) and 1 (one) and the arrangement will always place the electrons in the opposite corner or antipodal positions due to their repulsive force on each other. Lent, C. S., Almani and Porod, W. at Notre Dame University (Lent, 1993; Lent, Tougaw, 1997; Gregory, *

Corresponding author. Tel.: +91 9086 897920. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Wani et al/COMMUNE – 2015

et al, 2001; Amlani, et al., 1998) proposed a wireless two state quantum device cell of five dots as shown in figure 2. The modal similar to four dot cell modal yields two states of equal energy in the cell

Fig. 1 1.2.

Cell Working

The electrons in the cell always have the antipodal sites in both states of logic one (1) and logic zero (0) but the alignments are opposite as shown in figure 2 and figure 3.

Fig. 2 If two cells are brought close to each other they get aligned in the same direction due to inter columbic interaction as shown in Figure (4). The cells assume the order of lower energy in the system. In other words if a cell among two adjacent cells is brought to a state of ‘1’ or ‘0’ the adjacent cells will also get into same state. The carriage of state from one cell to its adjacent cell is said to have transmitted data and if a number of cells are placed adjacent to each other the data will travel from one end to other. Although no current flows but the conduction has taken place. This sort of conduction is the basic principle behind the working of quantum-dot cell devices. The shift of electron from one dot to other is electron from one dot to other is facilitated with the help of tunnel capacitor junction between the dots. This technique known as Dolan bridge technique physically embeds a capacitor junction between the dots in a quantum cell. As the repulsive force is encountered by an electron from the adjacent electron it tunnels through Dolan bridge junction to the adjacent vacant dot. The Dolan bridge junction between the dots in a quantum cell is demonstrated in Figure (4).

Fig. 3 and Fig. 4 2. Assembling QCA Devices On the basis of the basic properties of the quantum cells developed at Notre Dame University different algorithms have been designed to explore the possibilities of the devices with the help of software simulations. We have also studied and analyzed different logical device layouts with the help of QCA Designer software tool to arrive at some definite robust conclusions. As discussed above the quantum-dot cells arranged in a line produce a sort of conduction, this assembly of cells is termed as wire. A wire simulation is carried out by arranging a number of cells in a line and

[314]

Wani et al/COMMUNE – 2015

making the one end as input with respect to the other. Second important thing is to draw several output lines from a single input, this is termed as Fan-out. A fan-out is extension of the wire layout and only difference is to have similar more waveform appearing at different output lines. The logic NOT operation in QCA is performed in different ways and two ways to achieve this objective have been demonstrated in figure 5.

A Wire

Fan out

NOT Options

Fig. 5

Fundamental gate for the QCA exploration has been identified as the majority gate. It has three inputs and an output and its output is high when at least two of its inputs are high. The symbol and logic expression of the gate is given below:

a b c

M

Y= ab + bc + ca

Fig.6: Majority Gate Symbol

This gate is implemented by few quantum cells and has been demonstrated in the Notre Dame laboratory. The gate has a versatile property of acting as AND and OR gate with simply fixing one of its inputs. Let us assume the c input of the above to be at logic one i.e. c = +1 the Boolean expression for the above gate becomes 𝑌 = 𝑎𝑏 + 𝑎 + 𝑏 𝑂𝑟 𝑌 = 𝑎(𝑏 + 1) + 𝑏 𝑇ℎ𝑒𝑟𝑒𝑓𝑜𝑟𝑒

a b

𝑠𝑖𝑛𝑐𝑒 𝑏 + 1 = 1

𝑌 = 𝑎 + 𝑏

M

(𝑂𝑅 𝑔𝑎𝑡𝑒)

Y= a + b

+

In the similar way if the c input of1the majority gate is fixed to logic ‘0’ zero i. e. c = -1 the Boolean expression for the gate becomes 𝑌 = 𝑎𝑏 + 0 + 0 𝑂𝑟

a b

𝑌 = 𝑎𝑏

M

(𝐴𝑁𝐷 𝑔𝑎𝑡𝑒)

Y= ab

-1 Hence having NOT and the Majority gate to design OR and AND gates one has ample liberty to implement any kind of Boolean expression used for circuit design in digital electronics. Cell layout and simulation result for AND, OR and

[315]

Wani et al/COMMUNE – 2015

Majority gates has been reported by maximum researchers on five cell assembly. We have observed some noise distortion in the output as such an addition of one more cell at output stage was observed to produce robust results. The layout of the majority gate is presented in figure 7 below:

Fig. 7

Implementation of other logic gates like NAND and NOR can be simply designed by putting a NOT gate at the output of AND & OR gates respectively. The symbolic representation along with Boolean expression for these gates is given in Table 1. The main problem in the design arises when we start simulations for the Excusive-OR and ExclusiveNOR gates. These gates as we know are designed with the assembly of few gates, the layout imbalance and the clocking zone extensions produce interference in the output wave forms.

Fig.8(a)

Fig. 8(b) In digital electronics we usually design full adders by connecting two half adders to generate the output signal. While designing QCA full adders using same techniques number of problems arise in combining the signals of different stages due to mismatched clock. In the present study, we have tried different techniques to overcome this problem for the layout designs of full adder. 3. Study of Adder Layouts 3.1.

Designing Half Adders

The traditional way of designing adders is based on assembling half adders. In digital electronics the Boolean expression is derived for the required outputs and the circuit implementation is followed after minimization of the Boolean expressions. The binary sum of two bits can be simply got by an Exclusive-OR gate and the carry will be

[316]

Wani et al/COMMUNE – 2015

handled by a separate AND gate. The circuit assembly is known as Half Adder and its implementation using QCA is also simple because of the involvement of single stage of two gates. Here we move forward with the Exclusive-OR gates discussed in the previous section to convert them to half adders. Let us take the Exclusive-OR gate based on Boolean expression represented by equation (1) given below: 𝑺𝒖𝒎 = (𝑨𝑩). (𝑨 + 𝑩)

… … … … … … … . . (1)

𝑪𝒂𝒓𝒓𝒚 = 𝑨𝑩

… … … … … … … (𝟐)

Where A and B are the two input bits. The expression for the required carry can be simply generated by same assembly with a minor cell increase as the NOT form of AB is already present in the equation (1). The design layout of this Half Adder is given in Figure 9(a) and its simulation result is presented in figure 9(b). The results indicate perfect working of the layout with a minimum latency and a low count of cells for this implementation. Other way of designing the Half Adder is based on the Boolean expression represented by equation (3) given below: 𝑆𝑢𝑚 = (𝐴 + 𝐵) + (𝐴𝐵)

… … … … … … (3)

Fig. 9(a) and. 9(b) The generation of carry is again simple as the AB component is present in the assembly of the layout for this Exclusive-OR gate. The circuit layout is presented in figure 10a) along with its simulation results on QCA Designer in figure 10(b). The results in this case are almost same but this layout gives indication of more stability against electrostatic noise and has further advantage of occupying a bit lesser area in comparison to the above layout. These two layouts can form the basis for the design of Full Adders based on the traditional way of assembling two Half adders for the purpose. There are alternative ways too for designing Full Adders without using the traditional method. The methods arise due to Boolean alternative expression generation. The circuit algebra is used to convert the Boolean expressions of Full Adder to desired form for implementation. In QCA implementation a number of researchers have tried to look for the alternative of Full Adder layouts and before moving forward with our designs based on traditional method we will like to have survey of the proposed alternatives.

Fig. 10(a) and 10(b)

[317]

Wani et al/COMMUNE – 2015

Fig. 11 and Fig. 12(a) 3.2.

Zhang Full Adder Layout

A number of Full Adder layout designs have been reported using different techniques by various researchers. One of such circuit layouts is proposed by Zhang et al. (Firdous, Bhat, 2014) and is presented in Figure (11). The layout has been designed on the basis of reduced majority gates approach and appears to be a balanced compact assembly. We have analyzed the layout with the same algorithm and software (QCA Designer) and found the results inconsistent with the projected hypothesis. This multilayer layout was found to have lot of sneak noise paths as such an alternative single layer layout based on same logic was again simulated for crosscheck observations. The layout along with its simulation results is shown in figure 12(a) and figure 12(b) respectively. The analysis of the output yields correct behaviour for carry signal but failure in producing desired signal for the sum of this adder circuit on QCA Designer software. The implementation algorithm for this adder is based on the following Boolean expressions: Sum = MV(MV(A, B, Cin), Cin, MV(A, B, Cin)) ……………… (4) Carry = MV(A, B, Cin)

……………… (5)

These equations present a correct Boolean method of designing reduced majority gate based full adder, but the desired outputs are not produced even after changing orientation of layout from two layer to single layer

Fig. 12(b) and Fig. 13 3.3.

Cho Full Adder Layout

Another layout designed on the basis of reduced majority logic has been reported by (Zhang, et al, 2004) and is shown in figure 13. The design assembly was simulated on QCA designer software and produced results as shown in figure 13(a). Although we have not changed the orientation of the layout from multilayer to single layer but surely the hope of improvements in the results with QCA Designer software is very low.

Fig. 13(a) and Fig. 14(a) [318]

Wani et al/COMMUNE – 2015

3.4.

Simulation of Few More Layouts

Few more configurations based on reduced majority logic by (Walus, et al). (Heumpil, Earl, 2009) were also tested by us with the help of same software and found producing inconsistent results on simulation. Although the designs seem to be based on reduced majority gate equations but the result deviations suggest need of more study on both the robustness of software and the projected hypothesis. The layout assembles include designs proposed in (Wang, et al, 2003; Sara, et al, 2012; Mostafa, et al, 2007).

Fig.14 (b) and Fig. 15(a)

4. Proposed Full Adder Layouts Based on our previous work in (Amlani, et al, 1999; Wani, 2014) on the design half adders we have analyzed and simulated two layout configurations for the implementation of full adders and we will explain their analysis one by one in the following sub-sections 4.1.

Proposed Layout-First

This layout is based on the half adder design of figure 8(b). There is certainly a bit of complexity in generating the carry signal because it composes of two signal bits of different stages. The layout is given in figure 14(a) and its simulation results are given in figure 14(b). This layout is designed with the help of Boolean equations of traditional full adder circuits in microelectronics. Although the layout has a correct base of the Boolean algebra but the simulation results produce noisy result for the carry signal. Red circles in figure 14(b) have been drawn to show the signal variations at different places in the simulation output. The signal waveform for the sum signal is correct representation of the full adder layout with some propagation delay but the carry signal although incorrect in only one combination cannot be regarded as a valid result. This suggests a valid reason for looking into the robustness of the QCA Designer software for simulating the layout designs. We have tried a number of variations in this layout but could not generate the desired results on simulation for this case. 4.2.

Proposed Layout-Second

Projection from the above sections about the validity of QCA Designer software is only an option as the simulated results do not appear to be inconsistent. In maximum simulations the software produces convincing outputs. However some apprehensions regarding few outcomes need to be thoroughly investigated. An alternative layout based on same Boolean logic but using half adder of figure 8(a) was designed and simulated by us for desired output waveforms. This layout is given in figure 15(a) and its simulation results on QCA Designer are shown in figure 15(b).

[319]

Wani et al/COMMUNE – 2015

Fig.15 (b) The simulation results are completely consistent with the theory concept of full adders in Boolean algebra. Although circuit proposed layouts and some analyzed layouts of the previous section are perfectly according to Boolean or reduced majority logic but the results vary on simulation with QCA Designer software. Hence the evidence of sneak noisy paths projected in (Ismo, Jarmo, 2007; Kodam, Nanda, 2013) in the design of QCA layouts need to be taken seriously. 5. Conclusion and Discussion In this study we have not only proposed our version of QCA based full adder layout designs but also analyzed a good number of previously projected layouts to move ahead with concrete information regarding their working status on simulation. Our study suggests that the circuit layout comparison on the parameters of layout area, reduced majority logic, /Boolean logic, propagation delay, cell count or temperature are of secondary importance against considerations of working conformity with the help of some elegantly designed software. Although all the basic circuit layouts perform well on QCA Designer but to ascertain causes of varied results on simulation of full adders needs a through revisit in terms of both, the projected layouts as well as simulation software. We have already started our work regarding alternative options of layout designs with lesser noise crossover and simulation stability. The observations are projected for the fellow researchers to strive for the solution of the uncertainties regarding this promising field of technological development in future. References Amlani I. et al., 1998. Demonstration of a Functional Quantum-Dot Cellular Automata Cell, Journal of Vacuum Science and Technology B, 16, 3795 Amlani I., et al., 1999. Digital Logic Gates using Quantum-Dot Cellular Automata, Science, 284, , 289-291 Fasel Qadir, Ahmad, P. Z., Wani S. J., Peer M. A., December-2013. Quantum-Dot Cellular Automata: Theory and Applications, IEEE Conference on Machine Intelligence and Research Advancement (ICMIRA-2013), 540-544. Firdous, A,. Bhat, G. M., May 2014. Novel Code Converters Based on QCA, International Journal of Science and Research (IJSR), vol. 3, Issue 5, 364-371 Fulton, T. A. and Dolan, G. H., 1987. Observation of Single Electron Charging Effects, Physical Review Letters, 59, 189. Gregory, L. Snider, Alexci, O. Orlov, Vishwanath Joshi et. al., 2001. Quantum-Dot Cellular Automata: Introduction and Experimental Overview, IEEE-NANO, Heumpil Cho and Earl, E. swartzlander, June 2009. Adder and Multiplier Design in Quantum dot Cellular Automata, IEEE transactions on computers vol.58 No.6. Ismo Hanninan and Jarmo Takala, 2007. Robust Adders Based On Quantum dot Cellular Automata, Tampere Finland, Kodam Latha and Nanda Maharshi, M., September 2013. Design of Adders using QCA, IJAET, Vol. 6 No. 4, 1750-1759 Kyosum Kim, Kaijie Wu and Ramesh Karri, 2005. Towards Designing Robust QCA Architectures in the Presence of Sneak Noise Paths, Design, Automation and Test in Europe Conference and Exhibition, IEEE. Lent, C. S. and Tougaw, P. D., April 1997. A Device Architecture for Quantum Dots, Proceedings of the IEEE, vol. 85, No. 4, 541-557 Lent, C. S., and P. D., Tougaw, W., Porod and Bernstein, G. H., 1993. Quantum Cellular Automata, Nanotechnology, vol. 4, 49-57 Mostafa, R. A., kavehei, O., and navi, K., 2007. A Novel Design for Quantum-Dot Cellular Automata Cells and Full Adders, Journal of Applied Sciences, Vol. 7, No. 22. Navi, K, et al, 2010. A New Quantum Dot Cellular Automata Full Adder, Microelectronics Journal, vol.41, No. 12, pp. 820-826. Sara Hashemi, Mohammad Tehrani and Keivan Navi, January 2012. An Efficient Quantum-dot Cellular Automata Full Adder, Scientific Research Essays vol.7 (2) pp.177-189. Wang, W., Walus, K., and Jullien, G., 2003. Quantum-dot Cellular Automata Adders, In Proceedings of the 2003 3rd IEEE Conference on Nanotechnology, pages 461–464. Wani, S. J., Peer, Z., A,. Peer M. A., and Khan K. A., Mar-Apr 2014. Circuit Nanotechnology: QCA Adder Gate Layout Designs, IOSR Journal of Computer Engineering (IOSR-JCE), Vol. 16, Issue 2, 70-78. Zhang, R, Walus, K,. Wang, W,. and Jullien, G. A., 2004. A method of majority logic reduction for Quantum Cellular Automata.

[320]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

A Blind Watermarking Technique in Spatial Domain Using InterBlock Pixel Value Differencing Shabir A. Parah*, Javaid A. Sheikh, Nazir A. Loan, G. M. Bhat Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India

Abstract Robustness, Imperceptibility, Security, Payload and Computational complexity are main requirements of a watermarking system. It is a well-established fact that transform domain watermarking techniques are more robust than their spatial domain counterparts. However, high degree of robustness in frequency domain is achieved at the cost of high computational complexity. This paper presents a robust blind watermarking technique for gray scale images in spatial domain using interblock pixel value differencing that is robust to various image processing operations like rotation, cropping, histogram equalization, salt and pepper noise, Gaussian noise etc. An image is decomposed into 8×8 non overlapping blocks and difference between intensities of two pixels of adjacent blocks at the same position is calculated. Depending upon the watermark bit to be embedded, both the pixels, whose difference is calculated, are modified to bring the difference in a particular region. The comparison of experimental results obtained in our scheme with some existing techniques algorithm reveal that the proposed scheme is capable of providing high quality watermarked images in addition to being robust to image processing operations like JPEG compression, rotation, cropping, histogram equalization etc.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Blind Watermarking: Spatial Domain; Robust; Imperceptibility

_______________________________________________________________________________________ 1. Introduction The advent of internet and unprecedented growth of computer networks has enhanced the area of information technology to a greater extent. This has led to tremendous information being shared over the network infrastructure. Digital data being transferred over internet can however be easily copied, modified and distributed again. In such a scenario copyright protection, rational and material right protection for authors, proprietors, purchasers and suppliers and validity of content as are critical factors that need to be taken care of (Munesh and Shikha, 2010). Digital watermarking has grown-up as an effective solution for such a situation. Digital watermarking is a technique of embedding a secret message or information into a cover media message which can be perceived and extracted at a later stage for proof of copyright or usually against any illegal effort to either replicate or manipulate them in order to alter their identity. The hidden message embedded as watermark can virtually be anything, for example plain text, image or a serial number. (Mustafa and Rameshwar, 2012). In order to secure our watermark a key is needed to scramble the watermark before embedding it into cover media and also to extract the watermark the key is required. The most important requirements of any digital watermarking technique are Robustness, Security, Capacity and Imperceptivity (Shabir et al., 2014a). Depending on different applications and requirements, watermarking methods can be broadly classified into two categories, fragile and robust watermarking. Fragile watermarks are mainly used for content protection, authentication and tamper detection, while robust watermarks are used for ownership verification and copyright protection (Sami et al., 2010). A digital watermark is generally embedded into a cover media in spatial or transform domain (Namita and Jaspal, 2013 and Radhika and Kalpan, 2013). The spatial domain embedding methods operate directly on pixels, are easy to implement. The earliest and computationally efficient watermarking method in *Corresponding author. Tel.: +91 9596 529991 E-mail address: [email protected].

ISBN: 978-93-82288-63-3

Parah et al /COMMUNE – 2015

spatial domain is Least Significant Bit (LSB) substitution in which watermark bits are directly placed in the LSBs of the cover image (Shabir et al., 2014b). The major disadvantage of the LSB method is that the least significant bits may be easily destroyed (Samesh et al., 2013). Transform domain methods like Discrete Cosine Transform (DCT) and Discrete Wavelet Transform (DWT) modify the coefficients of transform domain of the cover image (Shabir et al., 2014c) and the watermarked image is obtained by the inverse transform. These methods are more complex than spatial domain methods but have much robustness against the image processing operations like median filtering, Gaussian noise, cropping etc. (Kalra et al., 2013), proposed a watermarking technique based on error correction codes and dual encryption along with DWT-DCT. This scheme is robust but, it is complex and has low capacity. (Jun et al., 2014) proposed a blind image watermarking scheme based on Fractional Fourier transform. This scheme is robust against JPEG compression and other image processing operations but, it is complex and has higher computational cost compared to spatial domain techniques. (Fang et al., 2012), proposed a blind watermarking technique in which Arnold scrambling and Quantification are used. By this scheme, the security of watermark is high but a watermarked image of low quality is obtained. Also for a cover image of size 512×512, only a 32×32 watermark can be embedded. (Chinmayee et al., 2014), proposed a novel blind watermarking algorithm in DCT domain, which is robust against JPEG compression but with this technique a 64×63 watermark can be embedded. This scheme is less robust to salt & pepper noise and Gaussian noise. (Lei and Bao-long, 2009), have proposed a localized image watermarking in spatial domain which is robust against both geometric attacks and traditional signal processing attacks but its capacity is very low. In this paper we propose a blind watermarking technique for gray scale images in spatial domain based. The proposed algorithm is easy to implement and has low computational cost as it is carried out in spatial domain. It is robust against the image processing operations like rotation, cropping, histogram equalization, salt & pepper noise etc. The rest of the paper is organised as follows: Section 2 explains the proposed algorithm. In Section 3 the experimental results are discussed. Section 4 concludes the paper. 2. Proposed Watermarking Algorithm The block diagram of the proposed system is shown in Fig. 1. A binary watermark W of size 64×64 is embedded into a cover image I of size 512×512. The original image I is divided into 8×8 non-overlapping blocks. In each block one bit of watermark is embedded, therefore the total number of watermark bits that can be embedded is equal to the total number of blocks i.e., a watermark of 64×64 can be embedded. Cover Image

Watermarked Image

Embedding

Watermark permutation

Watermark Fig.1 Block Diagram of Proposed Scheme

The binary watermark W shown in Fig. 2 is permuted with a key before embedding and is denoted by Wp.

Fig.2 Watermark

An arbitrary block of the cover image is denoted by B x. A pixel of xth block Bx at position (i, j) is denoted by PB x(i, j) where 1<=i<=8 is the row of block Bx and 1<=j<=8 is the column of block Bx. In order to embed the watermark, difference D between the two pixels of adjacent blocks at the same position is calculated and is given as: 𝐷 = 𝑃𝐵𝑥 (𝑖, 𝑗) − PBx+1 (i, j)

(1)

where 𝑃𝐵𝑥 (𝑖, 𝑗) is pixel of block Bx and 𝑃𝐵𝑥+1 (𝑖, 𝑗) is pixel of block Bx+1. When x is even, the pixels of blocks at position (3, 4) are taken for difference while the pixels of blocks at position (3, 3) are taken when x is odd as shown in Fig. 3. This is done to ensure proper watermark extraction. Depending upon the watermark bit to be embed both the pixels PBx(i,j) and PBx+1(i, j) are modified by a modification parameter m so that the difference D comes in a particular region, m is given as

[322]

Parah et al /COMMUNE – 2015

𝑚=𝑞×

𝑃𝐵𝑥(1,1)−𝑀𝑒𝑑(𝐵𝑥) 𝑃𝐵𝑥(1,1)

(2)

where PBx (1, 1) is the first pixel of block Bx, Med(Bx)is the median of 64 pixels of block Bx and q is a scaling variable ,we set q=0.1.

Fig. 3 Representation of blocks and pixels for calculating difference

The watermark embedding algorithm is summarised as under: Step1. Step2. Step3. Step4. Step5. Step6. Step7.

Read the original image I. Divide I into 8×8 non overlapping blocks, xth block of I is denoted by Bx Read the watermark W. Permute the watermark with a key, the permuted watermark is denoted by Wp. Compute the difference D using equation 1 according to the figure 2. Compute the modification parameter m using equation 2. For watermark bit Wp(x) =1; do the following: If {D >=60 then subtract m/2 from pixel PBx (i, j) and add m/2 to pixel PBx+1 (i, j) until the difference D becomes less than 60}. Else if {D<=20 and D>= - 40 then add m/2 to pixel PBx (i, j) and subtract m/2 from pixel PBx+1 (i, j) until difference D becomes greater than 20}. Else if {D< -40 then subtract m/2 from pixel PBx (i, j) and add m/2 to pixel PBx+1 (i, j) until the difference D becomes less than -100}. End of embedding of bit 1

For watermark bit Wp(x) =0 do the following: If {D>40 then add m/2 to pixel PBx (i, j) and subtract m/2 from pixel PBx+1 (i, j) until difference D becomes greater than 100}. Else if {D>= -20 and D<40 then subtract m/2from pixel PBx (i, j) and add m/2 to pixel PBx+1 (i, j) until the difference D becomes less than -20}. Else if {D>= -60 then add m/2 to pixel PBx (i, j) and subtract m/2 from pixel PBx+1 (i, j) until difference D becomes greater than -60}. End of embedding of bit 0

Step 8.

Repeat the steps from step5 to step7 until the whole watermark Wp is embedded into the image I.

The watermarked image denoted by WI is obtained by rearranging the modified blocks at their respective positions 2.2

Watermark extraction

For extraction of watermark from the watermarked image, the watermarked image WI is divided into 8×8 blocks and difference between the pixels PBx (i, j) PBx+1(i, j) of adjacent blocks is calculated using equation 1 according to Fig. 3. If the difference is less than -100 or falls between 20 and 60, a bit 1 is extracted and if the difference is greater than 100 or falls between -60 and -20, a bit 0 is extracted. The inverse permutation of bits is done to get the extracted watermark. 3. Results and Discussions The system robustness has been evaluated in terms of quality metrics Normalized correlation (NC) and Bit Error Rate (BER). NC gives the similarity between original watermark and extracted watermark while as the dissimilarity between the two is judged by BER. Lesser BER and higher NC values indicate that the algorithm is robust to the attacks. If W is the original watermark and EW is the extracted watermark then BER and NC are given as 𝐵𝐸𝑅=

[∑𝑅𝑖=1 ∑𝑠𝑗=1 𝑊(𝑖, 𝑗) ⊕ 𝐸𝑊(𝑖, 𝑗)]

1 𝑅𝑆

[323]

(3)

Parah et al /COMMUNE – 2015

𝑁𝐶 =

∑i ∑j W(i,j)EW(i,j)

(4)

∑i ∑j[W(i,j)]2

where W(i,j) is original watermark bit at position (i, j) and EW(i,j) is the extracted watermark bit at position (i ,j) and R×S is the size of watermark. Peak signal to noise ratio (PSNR) is used to measure the perceptual quality of watermarked image and is given as 𝑃𝑆𝑁𝑅 (𝑑𝐵) = 20𝑙𝑜𝑔10

255 1 √ ∑𝑀 ∑𝑁 [𝐼(𝑖,𝑗)−𝑊𝐼]2 𝑀𝑁 𝑖=1 𝑗=1

(5)

Where I is the original image and WI is the watermarked image of size M×N. Higher the PSNR value better the quality of watermarked image. 3.1.

Perceptual quality analysis

The proposed technique has been compared to those proposed by Fang et al., 2012 and by Chinmayee et al., 2014, for cover image 512×512 size and watermark size 64×64. The comparison of payload is depicted in Fig. 4. Further quality of the watermarked image, for the given watermark is depicted in Fig. 5. It is worthy to note that the PSNR of the watermarked image is of the order of 40.6dB.

Fig 4.Capacity comparison for 512×512 Lena image

3.2.

Fig. 5 Watermark insertion in Lena

Robustness Analysis

The scheme has been analysed for different attacks and the results obtained for different attacks are presented as under: 3.2.1

Watermark Extraction without Attack

Figure 5 shows the watermark insertion in Lena image when no image processing operation is done on the watermarked image. PSNR of watermarked image is 40.6 dB which means that a high quality watermarked image is obtained. Also BER and NC of the extracted watermark are 0 and 1 respectively which indicate that the watermark is extracted with no error. 3.2.2

Watermark Extraction after Rotation Attack

The watermarked image is rotated by 1 degree and 5 degree in counter clockwise direction. The watermark is extracted by re-rotating the image in clockwise direction. Figs. 5(a) and 5(b) respectively show the attacked image and extracted watermark after 1 degree and 5 degree rotation in counter clockwise direction. NC and BER of the extracted watermark after 1 degree rotation are 0.9934 and 0.0081 respectively. For 5 degree rotation NC and BER are 0.9480 and .0442 respectively. These results indicate that the watermark could resist rotation attack.

Figure5. Images obtained after rotation attack (a) 1 degree (b) 5 degree

[324]

Parah et al /COMMUNE – 2015

3.2.3

Watermark Extraction after Cropping

The watermarked Lena image is cropped 25% at top-left corner and 50% at bottom. A recognizable watermark from the cropped watermarked Lena image in both the cases is extracted. Fig. 6(a) shows 25% cropped watermarked Lena image and extracted watermark from it while as Fig. 6(b) shows 50% cropped watermarked Lena image and extracted watermark from it.

Fig. 6. Images obtained after cropping attack: (a) 25% cropping at left corner, NC=0.75and BER=0.181 (b) 50% cropping at bottom, NC= 0.4957 and BER=0.3501.

3.2.4

Watermark Extraction after JPEG Compression

Here we compress the watermarked Lena image using JPEG with a quality factor equal to 90. A recognizable watermark is extracted after the compression attack with NC=0.9524 and BER =.0435. For quality factor below 85, BER becomes too high and watermark extracted is not recognisable. Fig. 7 shows the compressed image with quality 90 and the extracted watermark.

Fig.7. Compressed image with PSNR=38.3 and extracted watermark

3.2.5 Watermark Extraction after Salt & Pepper Noise, Gaussian Noise, Histogram Equalization and Sharpening The watermarked media is attacked by salt and pepper noise, Gaussian noise, and Histogram equalization and sharpening. The parameters used for various mentioned attacks have been detailed as under. Fig. 8 shows watermarks extracted for different attacks. Salt and pepper: Salt and pepper noise with noise density 0.01 is added to the watermarked Lena image. PSNR of attacked image is 25.2dB. The extracted watermark is shown in Fig. 8(a). NC and BER of the extracted watermark are 0.9904 and 0.0107 respectively. Gaussian noise: Gaussian noise with mean zero and variance 0.001 is added to the watermarked Lena image. PSNR of attacked image is 29.61dB. The watermark extracted after this attack is shown Fig. 8(b). NC and BER of the extracted watermark are 0.9459 and 3.59 respectively. Histogram equalization: After performing histogram equalization, PSNR of attacked image is 19.9 dB and the watermark is extracted with NC and BER equal to 0.9483 and 0.0571 respectively. Fig. 8(c) shows the watermark extracted after histogram equalization.

Fig. 8. Extracted watermark after different attacks (a) Salt & pepper (b) Gaussian noise (c)Histogram equalization and (d) Sharpening

[325]

Parah et al /COMMUNE – 2015

Sharpening: Fig. 8(d) shows the watermark extracted from the sharpened image of PSNR 27.56 dB with NC=0.9766 and BER equals 0.0225. From Fig. 9 it is clear that proposed algorithm performs better than the algorithm proposed by( Chinmayee et al., 2014) ( i.e., watermark is extracted with less error as compared to (Chinmayee et al., 2014).

Fig. 9 BER comparison

4. Conclusions A blind and robust watermarking algorithm in spatial domain is proposed in this paper. The cover image has been divided in various 8×8 blocks and one bit of data has been embedded in each block. The relative difference between two predefined corresponding elements in a pair of adjacent blocks is modified depending upon the nature of data bit (whether 0 or 1) to be embedded. The scheme has been found robust to various image processing operations like cropping, rotation, salt & pepper noise, Gaussian noise. The proposed scheme has been compared to a state of art watermarking technique and the comparison results show that our technique is capable of providing high quality watermarked images besides being robust to various image processing attacks. Further, the scheme has been implemented in spatial domain which reduces the computational complexity involved. References Munesh, C., Shikha, P., 2010,"A DWT Domain Visible Watermarking Techniques for Digital Images." 2010 International Conference on Electronics and Information Engineering 2, p. 421-427. Mustafa, O., Rameshwar, R., 2012, "Digital Image Watermarking Basics, and Hardware Implementation". International Journal of Modeling and Optimization, 2, p.19-24. Shabir, A., Javaid. A., Abdul, M., Ghulam, M., 2014a," A secure and robust information hiding technique for covert communication" International Journal of Electronics, p. 1-14. Sami, E., Lala, Z., Thawar, A., Zyad, S., 2010," Watermarking of Digital Images in Frequency Domain''. International Journal of Automation and Computing 7, p. 17-22. Namita, C., Jaspal, B., 2013, "Performance Comparison of Digital Image Watermarking Technique: A Survey". International Journal of Computer Applications Technology and Research 2, p. 126-130. Radhika, V., Kalpan, B., 2013, "Comparative Analysis of Watermarking in Digital Image Using DCT &DWT". International Journal of Scientific and Research Publications 3, p. 1-4. Shabir, A., Javaid. A., Ghulam, M., 2014b, "A Secure and Efficient Spatial Domain Data Hiding Technique based on Pixel Adjustment". American Journal of Engineering and Technology Research 14, p. 33-39. Samesh, O., Adnane, C., Bassel, S., 2013, "Multiple Binary Images in Spatial and Frequency Domain". International Journal of Computer Theory and Engineering 5, p. 598-602. Shabir, A., Javaid. A., Ghulam, M., Abdul, M., 2014c, Data Hiding in Scrambled Images: A New Double Layer Security Data Hiding Technique, Computers and Electrical Engineering 40,p. 70-82. Kalra, G., Talwar, R., Sadawarti, H., 2013, Digital Image Watermarking in Frequency Domain Using ECC and Dual Encryption Technique". Research Journal of Applied Sciences, Engineering and Technology 6(18), p. 3365-3371. Jun, L., Zheng-guang, Z., 2014, Blind Digital Watermarking Method in the Fractional Fourier Transform Domain, Optics and Lasers in Engineering 53, p. 112-121. Fang , M., Jian, P., Wen, Z., 2012," A Blind Watermarking Technology Based on DCT Domain". 2012 International Journal Computer Science and Service System , p. 398-401. Chinmayee, D., Swetalina, P., Vijay, K., Kmaha, M., 2014, "A Novel Blind Robust Image Watermarking in DCT Domain Using Inter-block Coefficient Correlation". International Journal of Electronics and Communication (AEU)68, p. 244-253. Lei, D., Bao-long, G., 2009, "Localized Image Watermarking in Spatial Domain Resistant to Geometric Attacks". International Journal of Electronics and communication(AEU) 63, p. 123-131.

[326]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16-18March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

On the Study and Performance Evaluation of Multirate Filter Javaid A. Sheikh*, Jai Preet Kour Wazir, Shabir A. Parah, G. Mohiuddin Bhat Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India

Abstract Multirate digital signal systems in the processing of the digital signals have multiple sampling rates. Multirate systems have many applications in communication, compression, antennas etc. Multirate filter finds applications in multirate system. Cascaded Integrated Comb Filter is multiplier less realization digital filter. Cascaded Integrated Comb Filter acquaint with droop in pass band and attenuation in stop band. For attaining improved performance attenuation around the folding bands in stop-band must be high while that of pass- band droop must be reduced .This paper presents a review of some design and techniques used to decrease droop in pass band and improve attenuation in stop band various standard papers.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Multirate System; Multirate Filters; Cascaded Intergrated Comb Filters; Sample rate

1. Introduction

Multirate system finds many applications in software defined radios, communication, analog/digital conversion, efficient filtering, speech processing, audio, and video signals. Multirate systems are used in almost all area of digital signal processing. Multirate System’s main function is to alter the sample rate i.e., up sampling or down sampling. Down sampling means reduction in sample rate is called decimation. Up sampling means increase in sample rate is called interpolation. The simplest decimation filter is Comb Integrated Comb Filter. The Cascaded Integrated Comb Filter has low attenuation in stop band and high pass-band droop. This problem of low attenuation in stop-band and high pass-band droop can be eliminated using various techniques. This paper presents sharpening technique and compensation technique. 1.1

Comb Filter

Comb filter was first coined by Hogenauer two decades ago. Comb filter can be used for both interpolation and decimation. The Comb filter is the simplest decimation filter. The Comb decimation filter requires simple implementation of an addition and subtraction. They do not require any multiplier and requires only two building blocks. The design of Comb decimation filter, the stop-band attenuation should be high and pass-band droop low so that pass-band does not cause distortion in signal. Pass-band droop and attenuation in stop band depends upon the number of stages N and decimation factor R. So to low the pass-band droop and to rise the attenuation in stop band various techniques have been implemented and designed. 1.2

Cascaded Integrated Comb Filter

The Cascaded Integrator Comb filter has two main blocks integrators and combs. In multirate systems the Cascaded Integrated Comb filter are used for high sample rate conversion. Cascaded integrator comb filter has equal number of interpolators and decimators. The interpolators are used for up-sampling and decimators are employed in downsampling. The transfer function of Cascaded Integrated Comb filter is given below:

*

Corresponding author. Tel.:+91 9419 090554. E-mail address: [email protected] ISBN: 978-93-82288-63-3

Sheikh et al/COMMUNE – 2015 1

H (z) = [ (

1−𝑧 𝑅

𝑅 1−𝑧 −1

)]𝑁

R = Decimation factor N = Number of stages

I

I

I

R

C

C

C

Fig.1. Three stage Decimation Filter:

Comb Filter C

Down sampler R

Integrator Filter I

2. Techniques Implemented to increase the stop band attenuation and decrease in pass band droop in various papers some of them are discussed below Suverna Sengar: In this paper, the author has used CIC filter with compensation technique. In first stage with CIC filter followed by FIR filter. The paper presents the CIC filter with compensation technique to decrease the droop in pass band. The conversion ratio M. The CIC filter converts the sampling rate by a factor of N. FIR filter compensates the pass band characteristics and T(z) of FIR filter gives the decimation transition band. This paper uses the sharpening technique to improve the performances of pass band and stop band, high stop band attenuation and low droop in pass band. The author has considered the overall decimation factor M=10. The edge frequency in pass band is wp= 0.05π.The edge frequency of stop band is Ws π/N.

Figure.2. Gain Response of the Comb Filter and FIR Filter for K = 1 and K = 4

Figure.3. Gain Response of the Two-stage Decimator Implemented as a Factor-of-5 Comb Decimator and a Factor-of-Two FIR Decimator for K = 1 and K = 4

[328]

Sheikh et al/COMMUNE – 2015

Gordana Jovanovic Dolecek: In this paper, the author has used the compensation and sharpening technique in CIC filters. Simple equations are employed to increase the frequency response. R is decimation factor, K is number of stages, b is a parameter changes only when number of stages changes and does not depend on decimation factor. The magnitude response can be calculated as H (𝑒 −𝑗𝜃 ) =

𝜃𝑅

sin( 2 ) 𝑠𝑖𝑛

𝜃 2

The frequency response of compensation filter in this paper is 𝑅𝜃 G (𝑒 −𝑗𝜃𝑅 ) = 1+ε*( ) 2 By sharpening technique ε can be converted to increase compensation. The compensation filter provides delay to maintain filter’s linear phase.

Fig.4. Particulars of main lobe

Fig.5. Comparison of compensation CIC with sharpening technique

M. Laddomada, G. Jovanvic Dolecek: This paper has a multiplier less decimation filter and frequency characteristics can be improved by increasing attenuation around folding bands and adding more blocks of droop compensator to decrease the pass band droop. The filter is designed by addition and subtraction. The decimation filter proposed enhances the frequency response of CIC filter. The non-recursive architecture is used.

Fig.6. Magnitude Response

[329]

Sheikh et al/COMMUNE – 2015

Fig.7. Pass band and first folding band

Transfer function is: 𝐻(𝑧) = 𝐻𝑐 (𝑧)𝐻𝑠(𝑧)𝐺 (𝑧𝑅) G (zR) = Droop in compensator. H(z) = Increasing attenuation in stop band. Hc(z) = kthorder of comb filter. Magnitude response [G (ejwR) ]= (1 + 2 − 𝑏 𝑠𝑖𝑛2 (𝑊𝑅/2) Parameters considered while design N1, N2 and K R even = N1= 𝑅/2 − 1 N2 = 𝑅/2 + 1 R odd = N1 = R/2 N2 = 𝑁1 + 1 3. Conclusion This paper presents the comparison of various techniques for droop in pass-band and attenuation in stop band multirate signal processors. The compensated filter with sharpening technique show decrease in pass-band droop but has increased number of registers. The filter designed for even decimation values and rise in attenuation in odd folding bands. The comparison of droop compensator with sharpening technique shows better result near folding but has more complexity due to high rate foe filtering. References Phanendrababu H, Arvind Choubey, “Design of multirate Linear Phase Decimation Filters For Oversampling ADCs”, International Journal of Scientific & Technology Research, Vol 2, May 2013. Suverna Sengar, Partha Pratim Bhattacharya, “Performance Evaluation Of Multirate Filters, “International Journal of Science and Communication”,Vol.2,No.2, Feb 2012. Gordana Jovanovic Dolecek, Vlatko Dolecek, Isak Karabegovic, Javier Diaz-Carmona, “COMB-Based Method for narrow-band FIR filter Design”, Elsevier (September 2011). Ljiljana Milic, “Multirate Filtering for Digital Signal Processing: MATLAB Applications”, Hershey, PA: Information Science Reference, Jan 2009. G. Jovanovic Dolecek and Fred Harris, “On Design of Two-Stage CIC Compensation Filter”, IEEE International Symposium on Industrial Electronics, July 2009. Richard G.LYONS, “Understanding Digital Signal Processing”, Prentice Hall, 2011. Gordana Jovanovic Dolecek, M. Laddomada, “An improved class of multiplierless decimation filters: Analysis and design”, Elsevier (June 2013)1773-1782. Gordana Jovanovic Dolecek, Afono Fernandez-Vazquez, “Novel droop-compensated comb decimation filter with Improved alias rejections”, SciVerse Science Direct (October 2012)387-396.

[330]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Biomedical Sensor Interfacing Circuitry: A Watch Through M. Y. Kathjoo*, F. A. Khanday Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India

Abstract Biomedical engineering today is a highly emerging field of research and a growing industry. The health care today depends on the observations, which is in-turn dependent on devices used and procedures carried out. As the technology is advancing in terms of sensors and interfacing circuitry, we are having better systems for health care. This work has been carried out to present a review of the biomedical sensor interfacing circuitry, a classification of interfacing modules and their enhancements over time. This work access the downscaling of supply voltage and power consumed in circuits used for interfacing biomedical sensors. Trends in technology node, low noise design and spatial domain constraints have been discussed. This paper concludes with presenting the focus of current research being carried out and issue driving the future scope in this field.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Biomedical Engneering; Biomedical Sensor Interfacing Circuitry; Biomedical Amplifiers; Biomedical Rectifiers; Biomedical Situmilators; Biomedical A/D converters

1. Introduction Biomedical engineering has an important role in present day health care. Biomedical engineering as a rapid growing field is attaining heights of excellence yet there are provisions for lot of improvements in terms of faster diagnosis and adaptive technology, which is achieved in terms of going for low voltage, low power, higher integration capabilities, and better design node technologies. The field of biomedical engineering encapsulates various domains like biomechanics, biomaterials, physiological modelling, sensor design, sensor interfacing, image, and signal processing. The subject of sensor design and interfacing is quite important as for research carried in the field of biomedical engineering. Biomedical engineering found its importance as a research field with the arrival of first implantable device, a pacemaker in late 1950’s [Demosthenous, 2014, Ward et al, 2013, Vardas et al, 2014] and afterwards cochlear implants [Demosthenous, 2014, Zeng et al, 2008, Wilson and Dorman, 2008]. In 1960’s biomedical research became a point of attraction for many scientists, as many needful features need to be incorporated and implemented [Wang 2004]. Biomedical sensors broadly are classified based on working principle, in three categories. Physical sensors, Chemical Sensors, and Biological sensors (Bio Sensors). Physical sensors have a physical nature and effect such as metal resistance strain, piezoresistive, piezoelectric, and photoelectric sensors. Chemical sensors have a chemical nature and effect such as ion sensitive electrodes, ion sensitive tubes, humidity sensors, etc. In addition, bio-sensors are of biological nature such as enzyme sensors, microorganism sensors, immunity sensors, tissue sensors, DNA sensors and etc. [Wang 2004]. Biomedical measurements involve measuring physiological signals of electrical nature (ECGelectrocardiogram, EEG-electroencephalogram, EMG-electromyogram, EGG-electrogastrogram, VMG-vibromyogram, ENG-electroneurogram, VAG-vibroarthogram and CP-carotidartery pulse) and non-electrical (blood pressure, body temperature, pulse, breath, blood flow, etc). Since these signals are weak and random in nature having strong noise interference, additional circuitry is required to process them effectively. This circuitry termed as interfacing and conditioning circuitry forms an important part of the biomedical sensors and include amplifiers, filters, stimulators, A/D and D/A converters and rectifiers. In addition, a lot of work is being carried to make these effective and efficient in terms of voltage, power, space, portability and integration on chip [Demosthenous, 2014, Raikos et al, 2012, Carrillo et * Corresponding author. Tel.: +91 9622 772777 E-mail address: [email protected].

ISBN: 978-93-82288-63-3

Kathjoo and Khanday/COMMUNE – 2015

al, 2010]. By today, a lot of success has been achieved in regard to biomedical problems of monitoring health care as Pulse oximetry [Glaros and Drakakis, 2013, Wong et al, 2008, Jovanov et al, 2009, Wac et al., 2009, Yoo et al., 2010, Pandian et al., 2008, Haahr et al., 2012, Li et al, 2012], ECG [Roh et al, 2014, Tsai et al, 2012], EMG and neural spike detection [Weir et al, 2009, Gosselin, 2009, Laouudias, 2013, Li and Xu, 2011, Harrison et al, 2007], blood pressure and treating impairments as Visual prosthesis [Ong and Cruz 2012, Guenther, 2012], vestibular prosthesis [Wall et al, 2002-03, Fridman and Santina, 2012], Pacemakers [Ward et al, 2013, Vardas et al, 2014] deep brain stimulation (DBS) useful in providing therapeutic benefits for neurological disorders such as Parkinson’s disease, tremor, and dystonia [Demosthenous 2014, Miocinovic, 2013]. This success can be visualized in terms of developments in biomedical interfacing circuitry. In this paper, a classified overlook of trends and developments in biomedical circuitry has been provided. 2. Biomedical Interface Circuitry Building Blocks 2.1.

Amplifiers

Endogenous bioelectric signals-which are result of transient changes in transmembrane potential of living cells, like nerve acting potential and muscle acting potential, are very small in amplitude, of micro-volts and of very little time duration [Northop, 2004]. These signals for measurement need amplifiers with the distinct features of- rejection to dc input offset saturation, low leakage currents, very high adjustable gain, high CMRR, rapid calibration and a precise band operation of frequency as an EMG amplifier is generally coupled reactively with 3dB frequencies of 100-3k Hz and 0.5-100 HZ for ECG, which is not the case with regular amplifiers. These amplifiers are also preferably to have adhoc adjustments for bandwidth to give a high Signal to Noise Ratio (SNR). Fig 1 shows frequency and voltage range for various biopotentials. Tremendous work has been carried out on amplifier design for biomedical applications with visible achievements. Designs of differential amplifiers, instrumentation amplifiers [Denison et al, 2007], Chopper amplifiers [Enz et al, 1986, Masui et al, 2007], OTA based amplifiers [Khateb and Biolek, 2011, Khateeb et al, 2011], and auto-zero amplifiers [Enz and Temes, 1996, Chan et al, 2007] in terms of low noise, feedback topologies, and DC rejection ratio, have been so far proposed and experimentally demonstrated [Lim et al, 2007, Uranga et al. 2004, Harrison and Charles, 2007, Gosselin et al, 2007, Demosthenous and Triantis, 2005, Zhang et al, 2012, Khateb et al 2012, Tseng et al, 2012]. The recent design topologies have features like small dimensions (65nm CMOS design), low input capacitance (as there exists a trade off between chip area and input capacitance), interference neutralization (required for obtaining a specific neural signal), high gain and low power [Pude, 2013, Ng and Xu, 2013, Demosthenous et al, 2013, Kmon and Grybo’s 2013, Wu et al, 2013].

Fig. 1. Voltage and frequency range for various biopotentials

2.2.

Rectifiers

Biomedical devices used for continuous monitoring need to be portable, also implantable devices need a continuous power support. Low voltage devices can be operated by batteries yet this limits their duration of operation which has a solution of harvesting energy from bodies of bio-organisms to provide an extensive power support for these devices. This needs regulation of this harvested energy to a useable level, where in rectifiers find scope in biomedical devices. The common issue with rectifiers used for biomedical systems is presence of AC ripples which effect the operation of the devices. There are few other constraints like low voltage-to make it work on voltages available from harvested signals, Low power-to delivery maximum power to circuitry interfaced, tunablity- to be effective for different interfaces on chip , maximum output power which have been well addressed by researchers over time [Grace et al, 2011, Qingyun et al, 2011, , Cihun-Siyong , and Gong, 2011, Ivorra, Khateb et al, 2010, Koton et al, 2011, Dong and Yuanjin, 2009, Rodriguez-Villegas et al, 2009, Hyouk-Kyu, 2012]. A recent comparative study has been carried on various parameters

[332]

Kathjoo and Khanday/COMMUNE – 2015

of rectifier design with discussion on contemporary rectifier approaches in terms of output voltage, output power and the ripple factor [Haider et al, 2013]. A bulk driven CMOS technology rectifier has also been recently proposed with features of input power 2.14 µW and input voltage 600mV capable of rectifying frequencies from fraction of Hz to several KHz [Khateba 2013] 2.3.

Stimulators

Biomedical stimulators more precisely are called as neural stimulators, as electrical stimulus applied to nerves can cause action potential, which can drive a locomotive function in bio-organism or can overcome an imparity of vision/hearing. To produce an electric stimulus at least two electrodes are required and mode of stimulation is mainly voltage mode or current mode. Current mode is mostly used in implant devices. Charge balancing is an important parameter in stimulation as charge imbalance can be caused by many factors including leaking currents by cross-talk between adjacent stimulators and semiconductor/cable failure, which can be overcome by using a capacitor in series with each electrode. But since size is an important concern different alternate solutions have been proposed by researchers and most recent techniques for charge balancing are: Dynamic current balancing [Sit and Sarpeshkar, 2007], active charge balancer [Ortmanns et al, 2007], H-bridge with multiple current sinks [Williams and Constandinou, 2013], multiphase compensation approach [Jiang et al, 2011]. An important concern of research for biomedical stimulators is related to implants is integration on chip, incorporating more number of electrodes (focus on reducing size of implant SOC has limited the number of electrodes and applications like retinal implant need large number of stimulation sites), , power delivery(less power can result in improper stimulation and more power can cause cross-talk between two electrodes/stimulation sites), those have been well addressed by researchers over time [Das et al, 2013]. Success has been achieved upto integrating 4225 recording sites and 1024 stimulating sites using CMOS design [Bertotti et al, 2014]. 2.4.

A/D Converters

Biomedical Signals are analogue in nature and for processing these signals digitally or for storing these signals for further use/ remote analysis, these signals are to be digitized and thus require data (A/D and D/A) converters. The important parameters for data conversion are sampling rate and resolution. The various issues related to data conversions are aliasing-improper information as of having sampling frequency lower than nyquist rate, however increasing the sampling rate beyond a certain point does not significantly increase the fidelity with which the signal is rendered also increases the cost and sampling noise, which is caused by sampling a noise value adhered to signal during A/D conversion. These things are overcome by using filters with data converters. Successive Approximation register type ADC design is more suitable for biomedical applications. They are highly energy efficient for medium to high resolution applications with low speed requirements. A differential capacitive DAC is utilized to enhance the CMRR [Reda and El-Damak, 2012] 3. Conclusion Interfacing circuits forms the important building blocks of biomedical systems. The overall precision of biomedical systems in essence depends on the efficient design of interfacing circuits. In this paper, a review of biomedical interfacing circuits has been carried out. The classification of biomedical interfacing circuits were given and distinct parameters of the biomedical interfacing circuits which make them different from the interfacing circuits of the other domains were also presented. The parameters of interfacing circuits which are of concern as far as the present study of the biomedical systems is concerned were also covered. The work carried out in this paper shall therefore prove very useful for the furtherance of research in the area. References Demosthenous A., 2014, Review Article: Advances in Microelectronics for Implantable Medical Devices, Hindawi Publishing Corporation, Advances in Electronics, Volume 2014, Article ID 981295 Ward C., Henderson S.,and Metcalfe N. H., 2013, A short history on pacemakers, International Journal of Cardiology, vol.169,no.4, p.244–248,. Vardas P.E., Simantirakis E.N.,and Kanoupakis E.M., 2013, New developments in cardiac pacemakers, Circulation,vol.127, p. 2343–2350,. Zeng F.G.,Rebscher S., Harrison W.V., Sun X.,and Feng H., 2008, Cochlear implants: system design, integration, and evaluation, IEEE Reviews in Biomedical Engineering,vol.1, p.115–142. Wilson B. S. and Dorman M. F., 2008, Cochlear implants: a remarkable past and a brilliant future, Hearing Research,vol.242,no.18 Advances in Electronics 1-2, p. 3–21, Wang Ping, Liu Qingjun, 2004 Biomedical Sensors and Measurements, ISBN 978-3-642-19524-2 (Springer-Verlag GmbH) Raikos G., Vlassis S., Psychalinos C. 2012,0.5 V bulk-driven analog building blocks Int. J. Electron. Commun. (AEÜ) , p. 920–927 Carrillo J.M., Torelli G., Pérez-Aloe R., Valverde J.M., Duque-Carrillo J.F. 2010, Single-pair bulk-driven CMOS input stage: a compact low-voltage analog cell for scaled technologies Integr. VLSI J., p. 251–257 Glaros and Drakakis 2013 A sub-mw fully-integrated pulse oximeter front-end IEEE transactions on biomedical circuits and systems, vol. 7, no. 3, june

[333]

Kathjoo and Khanday/COMMUNE – 2015 Wong A.C.W. et al. 2008, A 1 V, micropower system-on-chip for vital-sign monitoring in wireless body sensor networks, in Proc. IEEE Int. SolidState Circuits Conf., Dig. Tech. Papers, pp. 138–602. Jovanov E., Poon C., Yang G., and ZhangY. 2009, Guest editorial body sensor networks: From theory to emerging applications, IEEE Trans. Inf. Technol. Biomed., vol. 13, no. 6, p. 859–863. Wac K. et al., 2009, Mobile patient monitoring: The mobihealth system, in Proc. Annu. Int. Conf. IEEE Eng. Med.-Biol. Soc., Sep. 2009, p. 1238– 1241. Yoo J., Yan L., Lee S., Kim Y.,and Yoo H. J., 2010, A5.2mWself-configured wearable body sensor network controller and a 12 wirelessly powered sensor for a continuous health monitoring system, IEEE J. Solid-State Circuits, vol. 45, no. 1, p. 178–188. Pandian P. et al., 2008, Smart vest: Wearable multi-parameter remote physiological monitoring system, Med. Eng. Phys., vol. 30, no. 4, p. 466–477. Haahr R. et al., 2012, An electronic patch for wearable health monitoring by reflectance pulse oximetry, IEEE Trans. Biomed. Circuits Syst., vol. 6, no. 1, p. 45–53. Li K., Warren S., and Natarajan B., 2012, Onboard tagging for real-time quality assessment of photoplethysmograms acquired by a wireless reflectance pulse oximeter, IEEE Trans. Biomed. Circuits Syst.,vol.6, no. 1, p. 54–63. Roh Taehwan, Hong Sunjoo, Yoo Hoi-Jun, 2014, Wearable Depression Monitoring System with Heart-rate Variability , IEEE 978-1-4244-7929-0/14 Tsai T.H., Hong J.-H., Wang L.-H., Lee S.-Y. 2012 Low power analog integrated circuits for wireless ECG acquisition systems IEEE Trans. Inf. Technol. Biomed., p. 907–917 Weir R.F., Troyk, P.R., DeMichele, G.A., Kerns, D.A., Schorsch, J.F. and Maas, H., 2009, Implantable Myoelectric Sensors (Imess) for Intramuscular Electromyogram Recording, IEEE Trans. Biomedical Engineering ,vol. 56, no. 1, p. 159–171. Gosselin B. and Sawan M., 2009, An ultra-low-power CMOS automatic action potential detector, IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 17, no. 4, p. 346–353. Laoudias C. and Psychalinos C., 2013, Ultra low-voltage low-power realization of non-linear energy operator for spike detection, Journal of Low Power Electronics, vol. 9, no. 1, p. 45–49. Li H. and Xu Q., 2011, Sub-threshold-based ultra-low-power neural spike detector, Electronics Letters, vol. 47, no. 6, p. 367–368. Harrison R.R., Watkins P.T., Kier R.J., Lovejoy R.O., Black D.J., Greger B., Solzbacher F, 2007, A low-power integrated circuit for a wireless 100electrode neural recording system IEEE J. Solid-State Circuits, 42 (1) , p. 123–133 Ong J. M. and Cruz L. da, 2012, The bionic eye: a review, Clinical & Experimental Ophthalmology,vol.40, p.6–17. Guenther T., Lovell N. H., and Suaning G. J., 2012, Bionic vision: system architectures—a review, Expert Review of Medical Devices, vol.9, no.1,p.33–48. Wall C. III, Merfeld D. M., Rauch S. D., and Black F. O., 2002-2003, Vestibular prostheses: the engineering and biomedical issues, Journal of Vestibular Research: Equilibrium and Orientation, vol.12, no. 2-3, p. 95–113. Fridman G. Y. and Della-Santina C. C., 2012, Progress toward development of a multichannel vestibular prosthesis for treatment of bilateral vestibular deficiency, Anatomical Record, vol.295,p. 2010–2029. Miocinovic S., Somayajula S., Chitnis S., and Vitek J. L., 2013, History, applications, and mechanisms of deep brain stimulation, JAMA Neurology,vol.70,no.2,p.163–171. Northrop Robert B., 2004, Analysis and Application of Analog Electronic Circuits to Biomedical Instrumentation, CRC Press Lim, Zou Xiaodan , Zheng Yuanjin , Tan Jun , 2007, Design of Low-Power Low-Voltage Biomedical Amplifier for Electrocardiogram Signal Recording, BIOCAS 2007 IEEE, Print ISBN:978-1-4244-1524-3 Enz C. C., Vittoz E. A., and Krummenacher F., 1986, A CMOS chopper amplifier, IEEE Journal of Solid-State Circuits,vol.22,no.3, p. 335–342. A. Uranga, X. Navarro, and N. Barniol, 2004, Integrated CMOS amplifier for ENG signal recording, IEEE Transactions on Biomedical Engineering,vol.51,no.12,p.2188–2194. Denison T., Consoer K., Santa W., Avestruz A.-T., Cooley J., and Kelly A., 2007, A2𝜇w 100 nV/rtHz chopper-stabilized instrumentation amplifier for chronic measurement of neural field potentials, IEEE Journal of Solid-State Circuits,vol.42,no.12, p.2934–2945. Enz C. C. and Temes G. C., 1996, Circuit techniques for reducing the effects of Op-Amp imperfections: autozeroing, correlated double sampling, and chopper stabilization, Proceedings of the IEEE, vol. 84, no. 11, p. 1584–1614. Chan C.-H., Wills J., LaCoss J., Granacki J.J.,and Choma J. Jr., 2007, A novel variable-gain micro-power band-pass auto-zeroing CMOS amplifier, in Proceedings of the IEEE International Symposium on Circuits and Systems (ISCAS ’07), p.337–340, May 2007. Masui Y., Yoshida T., Sasaki M., and Iwata A., 2007, 0.6 V supply complementary metal oxide semiconductor amplifier using noise reduction technique of autozeroing and chopper stabilization, Japanese Journal of Applied Physics,vol.46,no.4B, p.2252–2256. Harrison R. R. and Charles C., 2003, A low-power low-noise CMOS amplifier for neural recording applications, IEEE Journal of Solid-State Circuits,vol.38,no.6,p.958–965. Gosselin B., Sawan M., and Chapman C. A., 2007, a low-power integrated bioamplifier with active low-frequency suppression, IEEE Transactions on Biomedical Circuits and Systems, vol.1, no. 3, p. 184–192. Demosthenous A. and Triantis I. F., 2005, An adaptive ENG amplifier for tripolar cuff electrodes, IEEE Journal of Solid-State Circuits, vol.40,no.2,pp.412–420. Khateb F. and Biolek D. 2011 Bulk-driven current differencing transconductance amplifier Circuits, Syst. Signal Process, p. 1071–1089 Khateb F., Khatib N., Kubánek D., 2011 Novel ultra-low-power class AB CCII+ based on floating-gate folded cascode OTA Circuits, Syst. Signal Process, p. 447–464 Zhang Fan, Holleman Jeremy and Otis Brian P., 2012, Design of Ultra-Low Power Biopotential Amplifiers for Biosignal Acquisition Applications IEEE transactions on biomedical circuits and systems, vol. 6, no. 4. Khateb F., Kacar F., Khatib N., Kubánek D. 2012 High-precision differential-input buffered and external transconductance amplifier for low-voltage low-power applications Circuits, Syst. Signal Process. p. 453–476 Tseng Y., Ho Y., Kao S. and Su C.N.I.S., 2012, A0.09𝜇�W lowpower front-end biopotential amplifier for biosignal recording, IEEE Transactions on Biomedical Circuits and Systems,vol.6,no.5, p.508–516, Pude Mark, Mukund P. R., and Burleson Jeff, 2013, Positive feedback for gain enhancement in sub-100 nm multi-GHz CMOS amplifier design, Int. J. Circ. Theor. Appl. 43:111–124 Ng K. A. and Xu Y. P., 2013, A compact, low input capacitance neural recording amplifier, IEEE Transactionon Biomedical Circuits and Systems,vol.7,no.5, p.610–620. Demosthenous A., Pachnis I., Jiang D., and Donaldson N., 2013, An integrated amplifier with passive neutralization of myoelectric interference from neural recording tripoles, IEEE Sensors Journal,vol.13,no.9, p.3236–3248. Kmon P. and Grybo´s P., 2013, Energy efficient low-noise multichannel amplifier in submicron CMOS process, IEEE Transactions on Circuits and Systems I: Regular Papers,vol.60,no.7,pp.1764–1775, Wu C.Y., Chen W.M. and Kuo L.T., 2013, A CMOS power efficient low-noise current-mode front-end amplifier for neural signal recording, IEEE Transaction on Biomedical Circuits and Systems,vol.7,no.2, p.107–114 Grace, E.A., Rajan, S.E. and Asis, A.A.C., 2011, Performance Evaluation of Different Rectifiers for Piezoelectric Energy Harvesting Applications, in Proc. IEEE International Conference on Recent Advancements in Electrical, Electronics and Control Engineering, p. 248 –252. Qingyun Ma., Haider, M.R. and Massoud, Y., 2011, A Low-Loss Rectifier Unit for Inductive Powering of Biomedical Implants, in Proc. IEEE 19th International Conference on VLSI and System-on-Chip, p. 86-89.

[334]

Kathjoo and Khanday/COMMUNE – 2015 Cihun-Siyong and Gong Alex, 2011, An active-diode-based CMOS rectifier for biomedical power harvesting applications, International Journal of Circuit Theory and Applications Volume 39, Issue 5, p 439–449. Ivorra A., Remote electrical stimulation by means of implanted rectifiers. PloS one Journal Volume 6 ISSN 1932-6203 Khateb F., Vávra J., Biolek D., 2010 A Novel Current-Mode Full- wave rectifier based on one CDTA and two diodes Radio engineering, p. 437–445 Koton J., Lahiri A., Herencsar N., Vrba K., 2011, Current-mode dual-phase precision full-wave rectifier using current-mode two-cell winner-takes-all (WTA) Circuit Radioengineering, p. 428–432 Dong H., Yuanjin Z., 2009, A low-voltage low-power high precision digitally tunable transconductance rectifier Int. Symp. Integr. Circuits, p. 368– 371 ISIC ‘09 Rodriguez-Villegas E., Corbishley P., Lujan-Martinez C., Sanchez-Rodriguez T., 2009, An ultra-low-power precision rectifier for biomedical sensors interfacing Sens. Actuators, A, p. 222–229 Hyouk-Kyu C., 2012 A. CMOS Rectifier with a cross-coupled latched comparator for wireless power transfer in biomedical applications IEEE Trans. Circuits Syst. Express Briefs, p. 409–413 Haider Samnan, Mansor Hasmah, khan Sheroz, Arshad Atika, Shobaki Mohammed M. and Tasnim Rumana, 2013, A Comparative Study of Small Voltage Rectification Circuits for Implanted Devices, IOP Conf. Series: Materials Science and Engineering53 012024 Khateba Fabian, Vlassisb Spyridon, 2013, Low-voltage bulk-driven rectifier for biomedical applications Microelectronics Journal Volume 44, Issue 8 , p. 642–648 Sit J.-J. and Sarpeshkar R., 2007, A low-power blocking-capacitor free charge-balanced electrode-stimulator chip with lesst than 6 nA DC error for 1mA: full-scale stimulation, IEEE Transactions on Biomedical Circuits and Systems,vol.1,no.3, p.172–183. Ortmanns M., Rocke A., Gehrke M., and Tiedtke H.-J., 2007, A 232-channel epiretinal stimulator ASIC, IEEE Journal of Solid-State Circuits, vol. 42, no. 12, p. 2946–2959. Williams I. and Constandinou T. G., 2013, An energy-efficient, dynamic voltage scaling neural stimulator for a proprioceptive prosthesis, IEEE Transactions on Biomedical Circuits and Systems,vol.7,no.2, p.129–139. Jiang D., Demosthenous A., Perkins T., Liu X. and Donaldson N., 2011, A stimulator ASIC featuring versatile management for vestibular prostheses, IEEE Transactions on Biomedical Circuits and Systems,vol.5,no.2, p.147–159. Das Kushal and Lehmann Torsten, Rodrigues Cesar, 2013, Accurate Sensor Readout Circuitry for Reliability Measurement of Hermetically Sealed Chip-Scale Biomedical Implants, 978-1-4799-0066-4/13 IEEE Bertotti Gabriel et al, 2014, A CMOS-Based Sensor Array for In-Vitro Neural Tissue Interfacing with 4225 Recording Sites and 1024 Stimulation Sites, 978-1-4799-2346-5/14 IEEE Dina Reda and El-Damak , 2012, Design of Ultra Low Power Analog-to-Digital Converter for Ambulatory EEG Recording, PhD Thesis , MASSACHUSETTS INSTITUTE OF TECHNOLOGY, February-12

[335]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Privacy Preserving Data Mining using Fuzzy based Approach Majid Bashir Malika, M. Asgerb, Rashid Alic, Tasleem Arifd * a Department of Computer Sciences, BGSB University, Rajouri, India School of Mathematical Sciences and Engineering, BGSB University, Rajouri, India c Department of Computer Engineering, Aligarh Muslim University, Aligarh, India d Department of Information Technology, BGSB University, Rajouri, India

b

Abstract The process of data mining delivers valuable and previously unknown nuggets of information from vast volumes of data. The success of data mining is dependent on the quality of data submitted for data mining process. Most of the times, quality or accuracy of data collected for data mining purpose can be trusted but for some apprehensions like confidentiality, privacy or sensitiveness of the data, the real owners can submit false data. In addition, that, in turn, can affect the final results of the data mining. This is where privacy preserving data mining has to play a role, to develop confidentiality of the data, the data owners don’t want to reveal and at the same time the results of the data mining are not affected or are least affected. The aim of this paper is to propose a technique where results are not affected and at the same time, the privacy of data is preserved.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: PPDM;K-Means; Fuzzy; Fuzzy Membership Function;Clustering

1. Introduction With the advent of low cost storage devices, fast processors and extensive use of information technology large amounts of data is being collected every year in data repositories from almost every field that affects human life (Ann Cavoukian, 1977 and C. Li & G. Biswas, 2002). Such data in the data repositories can contribute in decision making and critical analysis provided sophisticated algorithms are available to extract the useful information (A. Ahmad and L. Dey, 2007). A lot of research has been done to achieve the same but till now it has proven to be a challenging task. Data mining is a specialized automated mechanism to achieve the goal of extracting useful information from enormously large data repositories (Zhihua, X, 1998). Modern computational and statistical techniques are being applied to uncover hidden and useful patterns in large data repositories (Tsantis L & Castellani J., 2001, Luan, J., 2002 and A. Ahmad and L. Dey, 2007). The data repositories may be distributed vertically or even horizontally. Data mining algorithms are being widely used to solve real world and real time problems associated extraction of unknown information from large data repositories. The need for data mining has been felt long ago in almost every field of life from banking to agriculture, medical diagnosis, telecommunication, intrusion detection, genetic engineering, education, marketing, investments, weather forecasting etc. (M. B. Malik et al, 2012). Conventional tools and techniques of data mining have to face certain challenges like high dimensionality, distributed databases, non-standardization of databases, missing values, changing data and even handling expired data (M. B. Malik et al, 2012). 2. Soft Computing Soft computing addresses many of such problems (Sushmita Mitra et al, 2002). Soft computing in itself is a consortium of synergistic mechanism to provide flexible and subtle processing of data related to the domain of real life ambiguous problems (L. A. Zadeh, 1994). The role model is human brain and it handles challenges like imprecision, uncertainty and partial truth successfully (M. B. Malik et al, 2012). A number of solutions have been developed using

* Corresponding author .Tel.: +91 94191 74250. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Arif et al /COMMUNE – 2015

soft computing techniques where conventional data mining were not able to deliver properly. The solutions so delivered possess robustness, tractability and at the same time having low cost in terms of time and space complexity. Privacy preserving data mining Data mining is flourishing as a valuable technique but at the same time privacy of the individuals has emerged as a serious concern. Identity in case revealed is threat to the privacy of individual because it is vulnerable to misuse. If this is the scenario, the individuals may be reluctant to share such information or even if they share it may not be true. This may affect the overall results of the data mining. Therefore, privacy needs to be preserved. The problem is not with the data mining but with the process of data mining. The mechanism needs to be modified to incorporate privacy preserving during data mining. Consequently, lots of work is being done on the privacy issue where privacy that is at stake is being tried to be preserved but yet a complete solution is missing (M. B. Malik et al, 2012). The data can be divided into four categories (Benjamin C M Fung et al, 2010): Explicit Identifiers, Quasi Identifiers, Sensitive attributes and non-sensitive attributes. Normally explicit identifiers and sensitive attributes need to be preserved as far as privacy preserving in data mining is concerned because these attributes when combined with publically available data can reveal the identity of the individual with good amount of confidence (Sweeney L, 2002). Such attacks are categorized under linkage attacks. Many a time background knowledge can also be cause of privacy attack (Gayatri Nayak and Swagatika Devi, 2011). This can be explained with the help of following example: Suppose a pathology lab maintains records of patients for research purpose and shares it with other labs having attributes like Name, Parentage, Address, Age, Contact number, name of the test, results (few attributes related to results). Although identifier attributes may be like Name, Parentage may be removed for the purpose. Quasi attributes like contact number, age have to be like that as adding noise to it can affect the results of the data mining. Same is the case with the results also. Now in case such attributes if shared with third parties that too in case of semi honest or malicious model can be combined with publically available data like Voter records. Voter record contains attributes like Voter ID, Name, Parentage, Address, DOB. DOB from Voter record can be combined with the Age attribute of the patient’s records from lab and with a good amount of confidence, the identity of the individual can be predicted. As such for the purpose sophisticated techniques are also available. So the privacy is under threat. Various solutions for addressing the problem have been developed. The solutions include noise addition through suppression, generalization, perturbation, data swapping etc but it affects the quality of data. Various other techniques have also been proposed like anonymization, perturbation, Randomized response, condensation, cryptography etc. But most of them suffer from one problem or other like information loss, high computation cost or even failure against attacks on privacy (Gayatri Nayak & Swagatika Devi, 2011; Y. Lindell & B. Pinkas, 2000; Aggarwal C & Philip S Yu, 2004; Aggarwal C & Philip S Yu, 2008; Clifton C, Kantarcioglu M et al 2002). 3. Proposed solution In this paper, we have proposed a solution of changing the values using fuzzification, where the age attribute has been fuzzified using S-shaped fuzzy membership function. K means Clustering has been performed first with normal values of age attribute and then with fuzzified values of the Age attribute. The clusters so formed in both the cases are exactly similar. It means that if age of person falls under cluster X, after fuzzification of the age of the same person, it still belongs to the same cluster. The data that has been used for the experimental purpose is related to the probable diabetic patients to whom the doctors had recommended for Sugar test. The data is a set of 400 records, collected from Pathology labs and has already been used in (Abid Sarvar and Vinod Sharma, 2014). Permission for the same has already been taken. The values of the ‘Age’ attribute have been randomly chosen from the whole dataset for this experiment. 4. Experimental results: K means clustering algorithm has been applied to the age attribute of nine patients. The code has been developed in Java. Age attribute values are: 67, 60, 83, 25, 60, 28, 5, 45 and 21. The same values of age have been fuzzified using S-shaped fuzzy membership function using the equation depicted as below: 0, 𝑥≤𝑎 𝑥−𝑎 2 𝑎+𝑏 2( ) , 𝑎≤𝑥≤ 𝑏−𝑎 2 (1) 𝑓(𝑥; 𝑎, 𝑏) = 𝑥−𝑏 2 𝑎+𝑏 1 −2( ) , ≤𝑥≤𝑏 𝑏−𝑎 2 { } 1, 𝑥≥𝑏 The code for it has been developed in Java. The fuzzified vales so generated are: 0.954588, 0.878401, 1, 0.284799, 0.878401, 0.423200, 0, 0.591293 and 0.157587.

[337]

Arif et al / COMMUNE-2015

In the next step, K-means clustering has been performed on original values of the age and there after it is again performed on the final fuzzified values of the corresponding original age values. The clusters assignments after 5 iterations so generated are the same. Table 1: Experimental results: Cluster assignments of original and fuzzified values

S.No

Original Age value

Fuzzified Age value

1 2 3 4 5 6 7 8 9

67 60 83 25 60 28 5 45 21

0.954588 0.878401 1.000000 0.284799 0.878401 0.423200 0.000000 0.591293 0.157587

Cluster assignments 1 1 1 0 1 0 0 0 0

Second column in the table i.e ‘Orignal Age value’ is age in years of the probable diabetes patient. Third column ‘Fuzzified Age value’ is the fuzzified value of the age in years corresponding to the second column age values. Fourth column ‘Cluster assignments’ shows to which cluster the original and fuzzified age values belong. It is clear from table 1 that, after fuzzification of the ‘Age’ attribute, the assignment of the clusters is exactly the same as compared to the cluster assignments of the original values of Age attribute. 5. Conclusion In this paper, we have proposed a potential approach to preserve privacy of an individual against linkage attacks and background knowledge attacks by transforming original data into fuzzified data using S-shaped fuzzy membership function. The above experimental results as shown in table 1 prove that the relativity of the data is preserved even after fuzzification of the data. So it can be used for preserving privacy in data mining that has evolved as a major concern along with the success of the data mining. Here using this solution, the privacy has been preserved and at the same time relativity of the data is same, thereby not affecting the results of the data mining. In future the work can be extended for data classification, association rule mining and machine learning along with preserving privacy in data mining. References Ann Cavoukian (1977), Information and Privacy Commissioner, Ontario, “Data Mining Staking a Claim on Your Privacy”, www.ipc.on.ca C. Li, and G. Biswas (2002), “Unsupervised learning with mixed numeric and nominal data,” IEEE Transactions on Knowledge and Data Engineering, vol. 14, no. 4, pp. 673-690. A. Ahmad, and L. Dey (2007), “A k-mean clustering algorithm for mixed numeric and categorical data,” Data & Knowledge Engineering, vol. 63, no. 2, pp. 503-527. Zhihua, X (1998), “Statistics and Data Mining”, Department of Information System and Computer Science, National University of Singapore. Tsantis L & Castellani J. (2001), “Enhancing Learning Environment Solution-based knowledge Discovery Tools: Forecasting for Self-perpetuating Systematic Reform”, JSET Journal. Luan, J. (2002), “Data Mining Application in Higher education”, SPSS Executive Report. Retrieved from http://www.crisp-dm.org/CRISPWP.pdf A. Ahmad, and L. Dey (2007), “A k-mean clustering algorithm for mixed numeric and categorical data,” Data & Knowledge Engineering, vol. 63, no. 2, pp. 503-527. M. B. Malik, M. A. Ghazi, R. Ali (2012), “Privacy preserving data mining techniques: Current Scenario and future prospects”, Third International Conference on Computer and Communication Technology. Sushmita Mitra, Sankar K. Pal and Pabitra Mitra (2002), “Data Mining in Soft Computing Framework: A Survey”, IEEE Transactions on Neural Networks, Vol 13, No. 1. L. A. Zadeh (1994), “Fuzzy Logic, Neural Networks, and Soft Computing.” Communications of the ACM, vol. 37, no. 3, pp: 77-84, March. Benjamin C M Fung, Ke Wang, Rui Chen, Philip S Yu (2010), "Privacy Preserving Data Publishing: A Survey of recent developments", ACM Computing Surveys, Vol. 42, No. 4, Article 14. B. Karthikeyan, G Manikandan, V. Vathiyanathan (2011), “A fuzzy based approach for privacy preserving clustering”, Journal of Theoretical and Applied Information technology”, Vol. 32, No. 2. Sweeney L (2002), "Achieving k-Anonymity privacy protection using generalization and suppression" International journal of Uncertainty, Fuzziness and Knowledge based systems, 10(5), 571-588. Gayatri Nayak, Swagatika Devi (2011), "A survey on Privacy Preserving Data Mining: Approaches and Techniques", International Journal of Engineering Science and Technology, Vol. 3 No. 3, 2127-2133. Y. Lindell and B. Pinka (2000), “Privacy Preserving Data Mining”, Journal of Cryptology, 15(3), pp.36-54. Aggarwal C, Philip S Yu (2004), "A condensation approach to privacy preserving data mining", EDBT, 183-199. Aggarwal C, Philip S Yu (2008), "A General Survey of Privacy-Preserving Data Mining Models and Algorithms", Springer Magazine, XXII, 11-52. Clifton C, Kantarcioglu M, Vaidya J, Xiaodong L, Michael Y (2002), "Tools for Privacy Preserving Distributed Data mining", SIGKDD Explorations letters Vol. 4, Issue 2. Abid Sarvar, Vinod Sharma (2014), “Comparative analysis of machine learning techniques in prognosis of type II diabetes”, AI and Society, Journal of Knowledge, Culture and Communication, Vol. 29, No. I.

[338]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

High Capacity Data Hiding using Random Plane Indicator Technique for Color Images Shabir A. Parah*, Javaid A. Sheikh, Jahangir A. Akhoon, G. M. Bhat Post Graduate Department Of Electronics and Instrumentation Technology, University Of Kashmir, Srinagar,India

Abstract The exponential increase in internet traffic demands more security for the data being transmitted over insecure channels. Cryptography has been used as one of the methods for ensuring data security during transit; however, camouflaged appearance of the scrambled data alerts the adversary about some critical information being shared. In such a scenario, image steganography has been used as an alternate solution to secure the secret information. This paper presents a high capacity data hiding technique for color images based on the concept of Random Plane Indicator Technique. The proposed technique uses Most Significant Bits (MSBs) of two channels of a color image as indicators to embed the data in the third channel. Since MSB of pixels are used as indicators in a cover image, any attempt to distort the indicator channels could be easily recognized. Further, unlike basic Pixel indicator method, the proposed technique uses all the three channels of cover image for embedding the secret message, hence increasing the hiding capacity of the system.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Pixel Indicator Technique; Pixel Value Modification; Stego-image; Cover image; Pixel Value Diffrence; Data Hiding

1. Introduction Nowadays the use of internet for transmitting digital information is more than as it was in the past, so algorithms for defending and securing important message needs to be discovered and implement them. This can be achieved with the help of Steganography and Cryptography .The purpose of cryptography and Steganography is to provide secret communication. However, Steganography is not the same as cryptography. Cryptography hides the contents of a secret message from a malicious people, whereas Steganography even conceals the existence of the message [Shabir et al, 2014]. Image steganography is the art and science of hiding information into hidden channels, so as to obscure the information and avert the uncovering of the hidden information [Frank, 2008]. The basic aim of Image Steganography is to achieve high capacity but there is always a tradeoff between Capacity and Quality [Shabir et al., 2012]. Steganography can be accomplished in spatial domain as well as in transform domain. In spatial domain secret information is embedded in cover image by modifying pixel values directly. The oldest and simplest spatial domain embedding technique is Least Significant Bit (LSB) substitution. Spatial domain techniques are computationally efficient. [Adnan, 2010] proposed a Pixel Indicator Technique for color image Steganography. This technique has low capacity. [Mandal and Das, 2012] proposed a blind steganographic techniques based on pixel value difference. Both the capacity and quality of stego image in this technique are very low. [Shamim and Kattamanchi, 2012] have proposed a steganographic technique, where the secret message is first encrypted and then embedded into LSBs of cover image. A high quality stego image is obtained with this technique but having low capacity. [Vijaya et al., 2014] proposed image steganography by enhanced pixel indicator method using MSB compare. This technique is complex and has low capacity. [Vaithilingam et al., 2013] have proposed image steganography scheme based on Pixel Value Modification (PVM) by modulus function. This technique does not produce a stego-image of much better quality and also its capacity is low.

* Corresponding author: Tel:+91 9596 529991. Email address: shabireltr@gmai;.com. ISBN: 978-93-82288-63-3

Parah et al/COMMUNE – 2015

The proposed paper presents a data hiding technique based on Random Plane Indicator Technique. The cover image is separated into three channels i.e. Red (R), Green (G) and Blue (B). The two MSB’s of the two channels out of three (RG, GB or BR) are used as indicators to embed the data in the remaining channel. Rest of the paper is organized as follows. Section 2 gives detailed description of proposed scheme. The results have been presented and discussed in section 3. The paper concludes in section 4. 2. Proposed Random plane Indicator Technique The flow chart of the proposed scheme has been shown in Fig. 1. In the proposed scheme the host image is partioned into three M×N planes i.e. Red(R), Green (G) and Blue (B) channel (plane). The MSB bits of two pixels of planes R and G, G and B or B and R are used as indicator to embed the secret message in plane B, R and G respectively. In order to embed message bit in pixel of B channel the MSB bits of R and G channel pixels are used as indicators. The indicators give us the bit position of pixel where message bit is to embed. For example if RG= ‘00’ or ‘01’, the secret bit is embedded in LSB of B plane pixel and if RG= ‘10’ or ‘11’, the secret bit is embedded in ISB of B plane pixel. Similarly, for embedding message bit in pixel of G plane the MSB bits of R and B channel pixels are used as indicators and so on. This makes embedding of message bits random in nature which in turn gives security to this scheme. The channel and pixel selection for embedding is depicted by Table 1. By using this technique three bits of secret data are embedded into each pixel and the secret bit is placed either in Least Significant Bit (LSB) or first Intermediate Significant Bit (ISB). Table1. Channel and pixel selection for embedding

2.1

Embedding algorithm

Steps of embedding are given below:

Fig. 1. Embedding algorithm

[340]

Parah et al/COMMUNE – 2015

1. 2. 3. 4.

5. 6. 2.2

Read the cover image and secret message. Partition the cover image into three channels i.e. Red, Green and Blue each of size 512×512. Get MSB bits of two pixels of RG, GB or RB channels. For embedding a message bit in a particular channel, the other two channels are used as indicators and accordingly message bit is placed in the pixel at the bit position obtained from indicators as depicted by Fig. 1. The steps 3 and 4 are repeated until secret data is embedded in all the pixels of the cover image. The stego-image is obtained by recombining the three planes.

Extraction algorithm

The steps of extraction of secret data from stego image are given below: 1. Read stego image 2. Decompose the stego image into three planes (i.e. R, G, and B). 3. Get MSB bits of two pixels of RG, GB or RB channels. 4. MSB bits of pixels of stego image are used to extract the secret message bit according to fig. 2. 5. Steps 3 and 4 are repeated until data bits are extracted from all the pixels of stego image.

Fig. 2. Extraction algorithm

3. Experimental Results In the proposed scheme we have hidden the secret data in the RGB channels of the cover image of size 512×512×3. The proposed technique produces a stego-image of cover image (Lena) with Peak signal to noise ratio (PSNR) 47.19 dB which indicates a high quality stego-image as shown in fig. 3 and also the secret data is extracted with no error. The histogram analysis is also performed on Lena image. The histogram of Lena image and stego-image is shown in fig.4. The least changes in the histogram of stego images will not able the steganalyst to detect the secret message. This makes our scheme secure. The proposed algorithm is also applied on other images like Baboon, Peppers and Plane, in each a PSNR above 46 dB is obtained, also data is extracted from each stego-image with no error. The cover images

[341]

Parah et al/COMMUNE – 2015

and stego-images of Baboon, Airplane and Pepper are shown in Fig.5, Fig. 6, and Fig 7. The results of proposed technique are compared with Vaithilingamet al. From Table 2 it is clear that the quality of stego- image obtained by our technique is better than the one proposed by Vaithilingam et al., and also the capacity of our technique is higher than the capacity of Vaithilingam et al.

Fig. 3. Lena

Fig. 4. Histograms of cover Lena and Stego Lena image

Table 2. Comparison between proposed method and PVM method

[342]

Parah et al/COMMUNE – 2015

Fig.5:Baboon

Fig. 6: Plane

Fig. 7: Pepper

5. Conclusion A high capacity steganographic technique is proposed in this paper which takes the advantage of channels of color images. Three bits of data are embedded into each pixel and bit of a pixel of a channel is replaced by a data bit depending on the MSB bits of two pixels of other two channels. A comparison of results is made on the basis of quality of stego image and capacity of the technique .The comparison showed that the proposed scheme has high capacity and also provides a high quality stego image. References Shabir, A., P, Javied, A., S, Abdul M., H, G., M., Bhat, 2014. “A secure and robust information hiding technique for covert communication”.International Journal of Electronics. DOI:10.1080/00207217.2014.954635. Shabir A. P, Javaid A. S and G. M. Bhat, 2012, “On the realization of a secure, high capacity data embedding technique using joint top-down and down- top embedding approach”. Computer Science and Engineering. Vol. 4, pp. 10141-10146. Frank, Y., S, 2012. Watermarkin and Steganography. Introduction to Digital Steganography Digital, CRC Press, New York pp. 137 Adnan, A., A., G., 2010. “Pixel Indicator Technique for RGB Image Steganography”. Journal of emerging technologies in web intelligence. Vol: 2, NO. 1. Mandal, J., K., and Das, D., 2012. "Colour Image Steganography Based on Pixel Value Differencing in Spatial Domain" , International Journal of Information Sciences and Techniques (IJIST). Vol: 2, No. 4. Shamim, A., L. and Kattamanchi, H., 2012. “High Capacity data hiding using LSB Steganography and Encryption”. International Journal of Database Management Systems. Vol.4, No. 6. Vijaya R. K., Dr.B. Tarakeswara R., Mr.B.Satyanarayana R., 2014. “Image Steganography by Enhanced Pixel Indicator Method Using Most Significant Bit (MSB) Compare, International Journal of Computer Trends and Technology (IJCTT). Vol. 15, No. 3. V.Nagaraj, Dr. V. Vijayalakshmi and Dr. G. Zayaraz, 2013 "Color Image Steganography based on Pixel Value Modification Method Using Modulus Function”. International Conference on Electronic Engineering and Computer Science. Vol.4. pp. 17-24.

[343]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Control of IP Address Spoofing - A Comparative Study of IPv4 and IPv6 Networks M. Tariq Banday, Reyaz Ahmad Mathangi* Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India

Abstract In addition to other fields, the IP address of source node is carried in each packet received by hosts in the network. This address is used by the receiving host to reply to the source in an IPv4 network. The source IP address is not verified for its correctness and therefore, may have been intentionally spoofed for some malicious intent. IPv6 incorporates a set of mandatory security protocols such as IPsec besides other mechanisms to prevent IP address spoofing. These are efficient anti-spoofing measures; however, spoofing is still possible because IPv6 networks have not completely replaced IPv4 networks as they support IPv4 through tunneling, transition, and translation. This paper discusses IP address spoofing in IPv4 networks and their control through various host based, router based, and combined spoofing control measures. The security control protocols such as IPsec and other features included in IPv6 for anti-spoofing are discussed and compared to security features of IPv4 networks. It also discusses their efficiency in dual stack, tunneling, and transition modes where both IPv6 and IPv4 protocols are used.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: IPv4; IPv6; IP Spoofing; IPSec; Tunneling; IP Address; Internet Protocol

1. Introduction The IP address spoofing is a serious threat to the legitimate use of Internet because it is an IP level threat that can affects other higher layer of TCP/IP protocol suite. IPv4 initially developed for research purpose possessed limited features such as addressing through a 32-bit address space, complex header format, manual configuration and little security control mechanism. With the Internet’s progression and its exponential growth by the development of application protocols such as www, e-mail, video sharing, VOIP, and teleconferencing several issues have surfaced. These include address exhaustion and security threats. To resolve the issue of limited address space, some techniques such as IP address classes, sub-netting, Classless Inter Domain Routing, Network Address Translation, etc. were implemented in the IPv4 networks. These methods complicated the addressing mechanism and further increased the security concerns. Many defense prevention mechanisms are thwarted by the ability of attackers to change, or spoof, the source addresses in IP packets. Attackers can evade detection and put a substantial burden on the destination network for policing attacked packets. Detection and prevention of IP address spoofing can be classified as host based, router based and combinational methods. Host-based methods are implemented on hosts, which allow them to recognize spoofed packets. They offer an advantage of easy deployment on the existing infrastructure without any need to change, however, their response is slow, as the spoofed packets must reach the host before being detected. Router based methods are implemented on routers and spoofed packets can be detected before they reach the destination. In comparison to host based methods, router based methods are more efficient but the efficiency of the system depends on the overall participation of routers.

* Corresponding author. Tel.: +91 9419900421. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Banday and Mathangi/ COMMUNE – 2015

However, it is difficult to ensure that every router on the Internet deploys these methods. Combinational methods engage both hosts and routers in a coordinated manner to identify spoofed packets. Introduction of IPv6 has resolved several issues of IPv4 and has mitigated some of its security threats by offering features such as large address space, hierarchal addressing, simpler and efficient header format, auto-address configurations, quality of services (QoS), and built in security (IPsec). It has replaced many IPv4 procedures to make IP spoofing difficult; however, they have not been able to eliminate it. Further, IPv6 has not so far replaced IPv4 in Toto and both of them are used in dual stack, tunneling, and transition modes. The complete transition from IPv4 to IPv6 requires replacement of entire IPv4 based hardware, skilled persons, and more importantly, it would incur heavy costs. Therefore, a strong need persists to develop new anti-spoofing methods and to improve existing ones that which can work effectively on both IPv4 and IPv6 networks. The remaining paper is organized as follows: section II briefly discusses security loopholes of IPv4. This section discusses the limitations of IPv4 due to which IP spoofing is possible. Section III evaluates the performance of various spoofing control mechanisms i.e. router based, host based and combinational methods, particularly designed for IPv4 and their advantages and disadvantages. Section IV describes IPv6 features that reduce chances of IP spoofing. Section V describes co-existence techniques and their effect on IP spoofing followed by conclusion. 2. Security Concerns in IPv4 IPv4 was initially designed for research purposes (RFC791, 1981), but due to its easy implementation and introduction of application protocols such as www, e-mail, VOIP, video sharing, teleconferencing, etc., it became widespread. It gained huge success in business, education, entertainment, banking, etc. IP made world a global village. However, it could not meet the growing demand for secure and vast connectivity of devices owing to one or more of the following reasons. 2.1. Address Exhaustion Once considered very huge, the address space of 232 hosts in IPv4 networks became insufficient to accommodate growing demand for connectivity. This led to the development of methods such as address classes (Class A, Class B and Class C) (Reynolds, Postel, 1983), Classless Inter Domain Routing (CIDR) (Fuller, 2006), Network Address Translator (NAT) (Srisuresh, Egevang, 2001), and Port Address Translation (PAT). These techniques also create some difficulties and introduce a few limitations such as slow mapping, which can be exploited by attackers to spoof packets. 2.2. Inefficient Addressing Schemes IPv4 uses three types of addressing schemes namely unicast, multicast, and broadcast (Hedrick, 2003). In unicast addressing, IP packets are send from one host to another host in a network. Multicast addressing allows a single host to communicate with a specific group of hosts in a network. While broadcast addressing allows a host to send packets to all devices of a network or broadcast addressing could communicate all hosts on a subnet. Broadcast addressing is prone to bandwidth attacks and therefore can open door for IP address spoofing. 2.3. Increased Routing Table Information The progressing rate of Internet forced Internet governing organizations to allocate IPv4 address prefixes. Presently more than 85,000 routes are in the routing tables of the Internet’s backbone (Hedrick, 2003). The use of increasing number of routers and entries within them has increased the chances of inconsistency in routing entries. As such, threats at network level including IP address spoofing have increased. 2.4. Complex Address Configuration The address configuration of IPv4 based networks is either manual or by using state full address auto-configuration using Dynamic Host Configuration Protocol (DHCP) (Templeton, Levitt, 2003). The growing rate of connections to the network and their manual configurations has further increased the possibility of threats at IP level. 2.5. Fragmentation at Router and Host IPv4 is inefficient protocol for data transmission, because it performs the fragmentation of large packets either at host or at router. This process is the integral part of the IPv4 protocol (DARPA, 1981) but degrades the efficiency and increases the possibility of DoS based attacks like ping of death, Tiny Fragment attack, Teardrop attack, Overlapping Fragment attack, and unnamed attack. Thus, fragmentation at the intermediate router level increases the possible IP

[345]

Banday and Mathangi/ COMMUNE – 2015

address spoofing. 2.6. Inefficient QoS IPv4 does not supports guaranteed and reliable delivery of packets, however, use of fields namely type of service and identification in IPv4 permit minimum QoS, but type of service field has limited functions in IPv4 and identification field using either TCP or UDP does not function when the payload of packets is encrypted. Therefore, the design of IPv4 is inefficient and vulnerable to spoofing attacks due to the non-prioritization of packet. 2.7. Needs Security at IP Level Due to the lack of security features in IPv4, IETF has developed a collection of security protocols such as IPSec, which protect communication at IP level. IETF designed IPSec with an aim to add confidentiality, integrity, and authenticity. However, it is very hard to deploy IPSec in IPv4. Implementation of IPSec can be undertaken in integrated form, bump in the wire form or bump in the stack form. However, all of them are difficult to implement and face deployment problems. 3. Spoofing control Methods in IPv4 Different mechanisms have been devised to control IP address spoofing at various levels of network. This section, describes spoofing control mechanisms in IPv4 at host level, router level and at both levels. 3.1. Host Based Methods Host based methods can be categorized as active and passive. Active methods includes those methods which uses cryptographic techniques such as IPSec, active probes and IP puzzling while passive methods gather information locally at end hosts without probing the source packet. IPSec is an active cryptographic method, which requires encryption and handshaking operation to setup the connection between hosts by using the secret key. Due to the use of secret key, attackers would be unable to spoof IP packets, and thus will not receive necessary replies, which are essential for handshaking operation (Kent, Seo, 2005). In addition, attacker would be unable to interfere in the existing communication, because of the use of secret key. The main objective of the design of IPSec was to achieve confidentiality, integrity, and privacy and to prevent date from spoofing. However, it is unrealistic to use IPSec in some conditions. Its use is optional in IPv4 and faces deployment problems on incompatible OS’s. Further, it is very difficult to make various active connections due to high computational costs involved for encryption of packets. Active probing methods include OS fingerprinting, IP identification field, and TCP specific probing. These methods though not specifically designed for anti-spoofing possess some sort of ability to defend IP spoofing. Probing tools such as Nemesis, NMAP, etc. can be used to determine the OS and as well as they can be used to detect the spoofed packets by sending specially designed packets to end hosts and observe their response. However, most of OS follow specification of standard protocols and their implementations may differ. These differences in implementations act as OS fingerprinting. Suppose a host actively fingerprints a particular host and find it running OS ‘Y’ while as passive OS fingerprinting on the original received packet determines it to be running OS ‘Z’. This indicates spoofing of packets. However, OS fingerprinting has a high amount of overheads to function correctly (Taleck, 2003; Zalewski, 2009c). When a sending host finds a suspicious IP packet, it can send packets to a supported source to observe response of its identification field (Zalewski, 2009a; Zalewski, 2009b). Further, different packets may have sent different IP identification field e.g. some hosts may choose random identification number while some hosts may use ascending or descending order of each packet. Suppose some hosts must increment the identification field of each packet then identification field in these probing responses must approach the identification field number of suspicious packets. Otherwise, the suspicious packet is spoofed one. However, it is a very complicated procedure and has many complexities involved in its implementation. As packet numbering may be easily guesses by attackers through TCP sequence numbers, therefor, simple TCP handshaking is not a secure procedure. Some OS’s uses the random sequence number. However, the pseudo-random number generator used may not be random in nature (Chang, et al, 2005). On the other hand, TCP specific probes add another layer for protection by adding acknowledge message. In TCP specific probes an attacker who sends specified packets is unable to see the replies and in response a receiving host sends acknowledgements that must change the TCP window size or causes packet retransmission and observes the supposed source responds correctly or not. If the supposed source does not change window size or does not retransmit packets, then recipient host may treats the packets as spoofed.

[346]

Banday and Mathangi/ COMMUNE – 2015

Another method used to defend spoofing is SYN cookies by which TCP handshaking becomes stateless. In SYN cookies, sending host does not allow resources to connect until 3-way TCP handshake is completed. First sending host sends SYN+ACK with packets with encoded initial sequence number (cookies) which includes hash of TCP header received from the receiving hosts SYN packet, a timestamp and clients maximum size. A SYN cookies using secure hash by encoding initial sequence number in 3-way TCP handshaking and when it receives the receiving hosts response, the sending host checks the sequence number and creates the necessary state if the receiving hosts sequence number is cookies value plus one (Bernstein, 2010). As such attacker are not able to guess the cookies values. Due to performance concerns and incompatibilities with TCP extension, SYN cookies are no longer used. The use of IP puzzles can prevent attackers from sending spoofed packets. In IP puzzling, sending host sends puzzles to receiving host and receiving host has to solve these puzzles at some computational cost (Chang, et al, 2005;). When the sending host receives the solved puzzles, it then allows receiving host to make connection otherwise not. IP puzzles can defend the IP spoofing. One of the passive host based method is hop count filter, which observes the hop counts of arriving packets at host by counting its normal hops and creates a mapping IP addresses hop counts. Thus, a spoofed packet may be detected by the hop count filter. The legitimate hop counts may vary due to routing changes, filtering all packets do not match would leads to false positive. To minimize false positive, HCF filters traffic only if some amount of threshold of packets do not match the expected hop counts. This threshold protects against mistakenly filtered packets but also makes HFC ineffective against slow amount of spoofed packets that do not reach the threshold (Jin, et al, 2003, Farguson, 2000). 3.2. Router Based Methods One of the earliest and simplest basic router based method is Martin filtering. To filter spoofed source address of packet, it simply cheeks IP field and looks for invalid address like non-unicast, loopback or any other spoofed address (Martin, 2006). However, from its design, it can detect and defend simple and common type of attacks only otherwise it is ineffective. Ingress and egress filtering methods are considered to be among the best solutions as they run on boarder router protocol. It can filter almost any type of spoofed packets if deployed on all routers of the network (Mircovik, et al, 2006). It limits the attackers to their local networks only. However, it requires full deployment. Reverse path forwarding (RPF) functions similar to ingress and egress but the only difference is, RPF needs forwarding table information at routers (Brenler, Barr, 2005) and it is based on routing symmetry of the internet, which is not always possible. The second type of router-based methods are distributed defence methods wherein routers co-operate using a key, which only valid packets can carry. Spoofing prevention method (SPM) which validates packets based by checking their secret key, which is embedded into the packets are deployed on the border routers. The packets containing the secret key are passed and others are treated as spoofed (Lin, Li, 2008). However, packets, which come from ASes, cannot be checked because ASes do not deploy SPM. Passport running routers define a new system of header, which is considered a passport containing many visas, and each visa indicating one ASes on the path that packets pass through. Visa is a keyed message authenticated code (MAC) generated by a passport system and is embedded in the sending packet. Each pair of ASes uses Deffe-Hallman exchange based on BGP update message (Park, Lee, 2001). If the packet is large, router fragments it due to which the passport becomes invalid because key exchange on BGP message is no more valid. Deterministic packet filtering (DPF) assumes all routers to maintain incoming knowledge of packets on each incoming interface. When spoofed packet arrive at the interface, the router can detect and filter them. DPF faces problems in learning the incoming direction knowledge (Daun, et al, 2008). Inter Domain Packet Filtering (IDPF) attempts to provide an implementation of DPF, the incoming direction comes from BGP update messages and the BGP protocol guarantees that attributes of a BGP update message are correct (Li, et al, 2002). In this way IDPF become invalid. SAVE and DPF operate in a similar way in which packets are filtered on the basis on their incoming detection. SAVE helps routers to learn the incoming direction knowledge (Li., et al 2008; Lee, et al, 2007). However, SAVE creates an incoming tree, which keeps track of topological relationship of source address. Whenever any router change happens, the incoming direction is affected and other participating routers are updated on the incoming tree. However, SAVE assumes its full deployment, which is infeasible and shows inefficiency. BASE completely relies on BGP and functions similar to SAVE, which sends updates to the routers to learn the correct incoming direction of packets. BASE is path based filtering mechanism and updates must travel the same path as the BGP update. However, BGP updates do not always travel the same path (Shen, et al, 2008). BASE also uses control message means BASE routers are only able to respond to the spoofed IP packets after receiving packets. 3.3. Combination of Host and Router Based Methods The combinational defense methods for IP Spoofing use both routers as well as hosts based methods, these are generally packet-marking solutions. In these methods, packets are marked first by routers in a network, and when these

[347]

Banday and Mathangi/ COMMUNE – 2015

IP packet reach at destination host, then the hosts take the action by using the marking, such as tracing the origin of an IP packet regardless of its source address field (Lv, Sun, 2007). The disadvantage of combinational methods is that no action can be taken upon the attacking packets until they reach their destination or victim at the edge of the network. In combinational methods, attacking packets are close to their destination. Therefore, attacks cannot be mitigated, even if the system can identify the packets. The path-identification or Pi designed against DoS attacks can also provide solution for IP Spoofing (Takuruu, et al, 2007; Wang, et al, 2007; Savage, et al, 2000; Belenky, Ansari, 2003). Pi uses IP packets fragmentation field to identify the path that it travelled. To identify the attacking packets is another problem in the Pi system. The improved version of Pi is the StackPi system, which stores the legitimate IP packet marking as a mechanism against spoofing packets (Yaar, et al, 2003; Yaar, et al, 2006). The host checks the stored marking, and then filters out packet, which do not have correct marking. StackPi assumes its full deployment, which is not possible. Furthermore, it cannot fully identify all spoofed packets. Table 1 shown below lists various characteristics of different spoofing control mechanisms. Table 1: Spoofing Control Mechanism in IPv4 C1 IPsec OS Fingerprinting IP ID Field Probing TCP Probing SYN Cookies IP Puzzles Hop-count Filtering Martian Address Filtering Ingress/Egress Filtering Reverse Path Forwarding SPM Passport DPF IDPF SAVE BASE Pi StackPi

C2 Active cryptography Active Probing Active Probing Active Probing Active others Active others Passive Basic Basic Basic Distributed Distributed Distributed Distributed Distributed Distributed Combined Combined

C3

C4

C5

C6

C7

A C C C A A B D D D C B A A A A A A

D A A A A D A C C C C B A A D C C C

D D D D D D D A C A C A A A A A D D

X X X X X X X X X X D D A B A A C C

(Kent, Seo, 2005) (Taleck,2003;Zalewski,2009c) (Zalewski, 2009a) (Zalewski, 2002b) (Bernstein, 2010) (Chang, 2005) (Jin, et al, 2003) (Martin, 2006) (Farguson, 2000; Baker,et al, 2004) (Mircovik, 2006) (Brenler, Levy, 2005) (Lin, 2001 ) (Park, Lee, 2001) (Daun, 2008) (Li, et al, 2002; Li, et al, 2008) (Lee, et al, 2007) (Yaar, 2003; Shen, 2008) (Yaar, et al, 2006)

Column Legends: C1-Spoofing Control Mechanism, C2-Category, C3-Spoofed Packet Detection, C4-Deployment, C5- Attack Mitigation, C6- Locating Attackers, C7- Reference(s), (A-Excellent, B- Better, C- Good D- Poor, X- None)

4. Anti-Spoofing Features in IPv6 As discussed in section 3 above spoofing control mechanisms of IPv4 have many limitations. IPv6 has been designed by IETF to overcome these limitations of IPv4 that surfaced during its use. It has added new features some of which provide more adequate spoofing control mechanisms than those in IPv4. These include: 4.1. Large Address Space Internet address space has been extended in IPv6 to 128 bits in comparison to 32 bits in IPv4. This address space provides 2128 (i.e. 3.4X1034) unique addresses in comparison to 232 (i.e. 4.3X109) address space in IPv4 (Hinden, et al, 2006). Therefore, address translation techniques such as NAT, PAT, address classes, and CIDR used in IPv4 networks are no longer required. Extremely larger address space makes address scanning difficult in IPv6 networks which considerably reduces the chanced of spoofing. 4.2. Streamlined Header Format IPv6 header structure is much simpler and streamlined than IPv4 header (Deering, et al, 1996). It comprises of eight fields in comparison to thirteen in IPv4. Three header fields namely version, source address, and destination address have not been changed while as five header field namely IHL, identification, flags, fragment offset, and header checksum have been removed. The names of four header fields namely type of service, total length, protocol and time to live have been respectively renamed to traffic class, payload length, next header and hop limit. The position of these fields have also changed. Further, flow label field has been added. IP packets in IPv6 are processed fast, efficiently and securely which requires less time that considerably reduces the chance of IP spoofing extremely. 4.3. Extension Headers All optional fields of IPv4 have been moved to extension field in IPv6. This change allows fast and efficient processing of packets as processing of these optional fields is done in a sequential order only if required. The extension

[348]

Banday and Mathangi/ COMMUNE – 2015

field includes field names Hop-by-Hop, Routing Header, Fragmentation Header, Destination Options Header, Authentication Header (AH), and Encapsulating Security Payload (ESP) Header, each of which is defined by the Next Header field of IPv6 packet. Chances of IP address spoofing can be minimized as Extension Header processing has become fast, efficient and secure using IPSec (AH & ESP) (Deering, et al, 1996). 4.4. Efficient Addressing IPv6 has efficient addressing schemes like unicast, multicast, and anycast. A packet is delivered to the interface it identifies in a network or a sub network through unicast addressing. A packet is delivered to the group hosts for which it is intended to and others do not receive it in multicast addressing. In the anycast addressing mechanism, a packet is delivered to the closest member of the group in a network (Blanchet, 2003). Unlike IPv4, IPv6 does not have broadcast addressing wherein a packet is delivered to all the nodes of the network, which considerably reduces the chances of IP address spoofing. 4.5. Stateless and Stateful Address Configuration The rapid growth of Internet demands some mechanism for automatic address configuration. Improved Dynamic Host Control Protocol DHCPv6 makes two types of automatic address configurations namely stateless address autoconfiguration and stateful autoaddress configuration available in IPv6 (Gont, 2013). As manual address configuration is not permitted in IPv6, IP related threats are therefore minimized. 4.6. Better QoS Internet protocol does not guarantee delivery of packets at destinations and is thus unreliable but it makes best efforts not only to make the packets delivered fast at destinations. IPv6 and IPv4 both include features for quality of service but in IPv6 two fields namely traffic class and flow label have been set aside to prioritize the delivery of IP packets. These fields prioritize time sensitive packets such as video, voice over IP (VOIP), and teleconferencing packets. Therefore, IPv6 has better quality of service. Due to prioritizing of particular packet flow, IP level threats like IP spoofing can therefore be minimized. 4.7. Fragmentation at Nodes In IPv6, the fragmentation of packets occurs only at the source host, which are reassembled at the destination host (Thomson, et al, 2007). This not only avoids resending of entire packet in case some fragmented packet gets corrupt or lost but also minimizes fragmentation attacks in comparison to that in IPv4 networks. 4.8. Mobile IPv6 (MIPv6) MIPv6 has another enhanced feature of IPv6 as it supports rooming for mobile nodes on the move. IPv6 has enormous address space and can support growing number of mobile devices. Using Neighbor Discovery (ND) protocol, IPv6, solves the handover issue at IP level without disconnection and minimizes the security issues of mobile devices (Johnson et al, 2004; Perkins, Ed, 2002). 4.9. Inbuilt Security One of the main features of the IPv6 is its mandatory security protocol namely IPSec, which secures communication at IP and other higher layers. IPSec is a set of security protocols and various other security components. It provides confidentiality, integrity, and authenticity to IP packets thereby securing IP packets from IP spoofing. Data protection is achieved by the use of authentication header (AH) and Security Encapsulation Payload (ESP) from the next header in the sequence (Kent, Seo, 2006; Kent, 2005a; Kent, 2005b). IPSec operates in two modes namely transport mode and tunneling mode. Transport mode encrypts the payload of packet while as tunneling mode encrypts whole packet. 4.10. Extensibility Features offered by IPv6 are extensible by addition of new fields to its extension header.

[349]

Banday and Mathangi/ COMMUNE – 2015

5. Co-Existence Mechanisms Replacing IPv4 networks with IPv6 networks rapidly has not been possible owing to multidimensional requirements such as requirement for upgrading hardware infrastructure such as hosts, routers, switches, hubs, etc., deployment of supportive operating system, and training human resource. Therefore, a few co-existence mechanisms through which smooth transition from IPv4 to IPv6 protocol is supported without disturbing Internet and its processes have been devised. These include Dual Stack, Tunneling, and Transition modes. Dual stack mode is the first co-existing mechanism wherein end hosts and routers implement both IPv6 and IPv4 protocols. In this mode, no abrupt change in ISP’s existing infrastructure is required but to support both IPv4 and IPv6 protocols in parallel, devices such as nodes, routers, etc. have to be configured. Dual stack mode is not a secure transition mechanism, as it is vulnerable to attacks including that from IP spoofing. IPv6 packets are tunneled or bridged over IPv4 networks in tunneling mode. Packets are encapsulated by the boarder router at the transmitting end and sent through the IPv4 network, which are expanded at the receiving end by outer router. Various types of tunneling modes are GRE, 6RD, 6to4 and ISATAO modes. Tunneling mode transition mechanism mostly uses IPv4 supported network at intermediate level. During the transmission, IPv6 packets are treated as IPv4 packets and therefore, are prone to IP spoofing attacks. Transition mode allows IPv4 networks and IPv6 networks to communicate with each other by translating packets from one version to the other. The transition mechanism allows deployment of new devices as IPv6 devices while older devices remain IPv4 systems. This method creates bottleneck when packets are translating from IPv6 to IPv4 and vice versa. 6. Conclusion The limitations of IPv4, which include its limited address space, its broadcast addressing scheme, complex header format, manual address configuration, fragmentation at routers, and inefficient QoS threaten to degrade the performance of the global Internet. These limitations permit IP spoofing and other IP related attacks. In IPv4 domain, the attackers can easily spoof the IP address of the victim host without need of complicated hacking tools or skilled human resource. As such, various defense mechanisms at hosts, routers or at both hosts and routers were developed. The host-based mechanisms are easy to deploy without changing the existing infrastructure and are low cost mechanisms but their response is very poor. The active host based methods show quick response while passive host based methods act slowly. On the hand router based mechanisms, have quick response but their deployment and cost is challenging. Their performance is variable in terms of spoof detection, spoof mitigation, locating attacks, etc. The basic filtering methods perform filtering at routers while distributed methods mark each IP packet on every router. The methods employing both hosts and routers have higher efficiency but are costly and their deployment is difficult. None of the methods has completely controlled spoofing. Recognizing the limitations of IPv4, IETF has designed its successor, the IPv6. IPv6 has included a range of new features such as very large address space, hieratical-addressing scheme for unicast, multicast, and anycast, simple and streamlined header format, extension headers for rarely used headers, statefull auto-address configuration, and stateless address configuration (DHCPv6) and efficient transmission of IP packets by fragmenting large packets at the source node reducing fragmentation attack. These features make IP addressing spoofing a complicated task for attackers though not impossible. The attackers have to change their attacking tools and needs skilled human resource. However, IPv6 has not completely replaced the existing protocol. Currently both protocols exist at the same time in the form of co-existence i.e. dual stack mode, tunneling mode and translation mode. In these co-existing mechanisms, the possibility of IP spoofing still exists. In dual stack mode, both the protocols functions simultaneously and the threats related to IPv4 persist. IPv4 IP spoofing control mechanisms face new challenges in the co-existence schemes because they are not applicable to both IPv4 and IPv6 packets in their current forms. Therefore, during the co-existence period, it is required to improve already deployed IPv4 based spoofing control mechanisms or design new and efficient methods. IPv6 is very efficient against spoofing attacks and thus useful to minimize them with its full deployment, however, prevention and detection of spoofing is still an ongoing battle between developers and attackers.

References Baker, F., 1995. Requirements for IP Version 4 routers. Request for comments RFC 1812, IETF. [Online]. Available: http://www.ietf.org/RFC/RFC1812.txt Baker, F., Savola, P., 2004. Ingress Filtering for Multihued Networks. Request for comments RFC 3704. IETF. Belenky, A., Ansari, N., 2003. IP Traceback with deterministic packet marking. IEEE Communication Letters, vol. 7, pp. 162-162. Bernstein, D. J., 2010. SYN Cookies. http/www.cr.yp.to/syncookies.html. Blanchet, M., April 2003. A Flexible Method for Managing the Assignment of Bits of an IPv6 Address Block. Request for Comments RFC3531. Brenler-Barr, A., Levy, H., 2005. Spoofing Prevention Method. IEEE INFOCOM. IEEE, pp. 536-547.

[350]

Banday and Mathangi/ COMMUNE – 2015 Chang, F., Kaiser, W., Chi, F., Liu, A, 2005. Design and implementation of network puzzles. The annual joint conference of IEEE, computer, and communications societies (InfoCom). IEEE, pp. 2372-2382. Daun, Z., Yuan, X., Chandarshekar, J., 2008. Controlling IP Spoofing through Inter Domain Packet Filters. IEEE Transactions on Dependable and Secure Computing, vol. 5, pp. 22-36. Deering, S., Hinden, R., December 1998. Internet Protocol, Version 6 (IPv6). Request for Comments RFC2460. Farguson, P, 2000. Network Ingress Filtering: defeating Denial of service Attacks which employ IP Source Address. Request for Comments RFC2267, IETF. Fuller, V., Li, T., August 2006. Classless Inter-domain Routing (CIDR). The Internet Address Assignment and Aggregation Plan. Request for Comments RFC4632 Gont, F., August 2013. Security Implications of IPv6 Fragmentation with IPv6 Neighbor Discovery, Request for Comments RFC6980. Hedrick, C., June 1988. Routing Information Protocol. Request for Comments RFC1058, IETF. Hinden, R., Deering, S., February 2006. IP Version 6 Addressing Architecture. Request for Comments RFC4291. Jin, C., Wang, H., Shin, K. G. 2003. Hop-count filtering: An effective defense against spoofed DDoS traffic. In Proceedings of the Conference on Computer and Communications Security, pp. 30–41. Johnson, C. D., Perkins, Arkko, J., June 2004. Mobility Support in IPv6. Request for Comments RFC3775. Kent, S., 2005. IP Encapsulating Security Payload (ESP). RFC4303. Kent, S., December 2005a. IP Authentication Header. Request for Comments RFC4302, IETF. Kent, S., Seo, K., 2005b. Security Architecture for the Internet Protocol. Request for Comments RFC4301, IETF. Lee, H., Kwon, M., Hasker, G., Perring, A., 2007. BASE: an incrementally deployable mechanism for IP Spoofing prevention, 2nd ACM Symposium on information, computer, and Communication Security (ASIACCS’07). Singapore, pp. 20-31. Li, J., Mirkovic, J., Ehrenkranz, T., Wang, M., Zhang, L., 2008. Learning the valid incommoding direction of IP pockets. Computer Networks, vol. 52, pp. 399-417. Li, J., Mirkovic, J., Mengqui Wang, P.R., Zhang, L., 2002. SAVE: Source Address validity enforcement protocol. Annual Joint Conference of the IEEE Computer and Communications Societies (InfoCom). IEEEpp. 1557-1566. Lin, X., Li, A., Yang, X., Wetharall, d., 2008. Passport: Secure and adoptable source authentication. USENIX Symposium on Network Systems Design and Implementation, USENIX, pp. 365-378. Lv, G. F., Sun, Z. G., 2007. Towards Spoofing Prevention Based On hierarchal coordination model, IEEE Workshop on high performance switching and Routing (HPSR). IEEE, pp. 446-239. Martin, April 2006. Stop the bots, Security Focus, Available at: http://www.securityfocus.com/columnists/398 Mircovik, J., Jectic, N., Reibher, P, 2006. A practical IP Spoofing Defense Through Route-Based Systems, cis-tr-2006 – 332, University of Daleware CIS Department, Park, k., Lee, H., 2001. On the effectiveness of rout based packer filtering for Distributed Denial of Services attack prevention in power-law internets, computer communication, pp. 15-26. Perkins, C., August 2002. IP Mobility Support for IPv4. Request for Comments RFC3344. Reynolds, J., Postel, J., October 1983. ASSIGNED NUMBERS. Request for Comments RFC870. RFC791, September 198. Defense Advanced Research Projects Agency. Request for Comments RFC0791. Savage, S., Wetherall, Karlin, A. D., Anderson, T., September 2000. Practical network support for IP Traceback. Computer communication review, vol. 30, pp. 295-306. Shen, Y., Bi, J., Wu, J., Liu, Q., 2008. A two-level source address prevention based on automatic signature and verification mechanism. The passive and active conference. pp. 392-397. Srisuresh, P., Egevang, K., January 2001. Traditional IP Network Address Translator (Traditional NAT). Request for Comments RFC3022. Takuruu, H., Matsuura, K., Imai, H., 2007. IP Traceback by pocket Marking Methods with Bloom Filters, 41st Annual IEEE international Carnaha Conference. IEEE, pp. 255-263 Taleck, G., 2003. Ambiguity Resolution via Passive OS Fingerprinting. The symposium on recent advance in intrusion detection, pp. 192-206. Templeton, S. J., Levitt, K. E., 2003. Detecting Spoofed Packets, DARPA information survivability conference and Exposition (DISCEX’03). Thomson, S., Narten, T., Jinmei, T. September 2007. IPv6 Stateless Address Autoconfiguration. Request for Comments RFC4862. Wang, H., Jin, c., Shin, K. G., 2007. Defense against spoofed IP traffic using hop-count filtering. IEEE-ACM Transactions on Networking, Vol. 15, no. 1, pp. 40-53. Yaar, A., Perring, A. and song, D, “StackPi, 2006. New packet marking and filtering mechanism to defend against DDoS attacks and IP Spoofing defense. IEEE journal of selected areas in Communications, vol. 24, pp. 1853-1863. Yaar, A., Perring, A., song, D., 2003. A path identification mechanism to defend against DDoS attacks. IEEE Computer Society Symposium on Research on Security and Privacy. IEEE, pp. 93-107. Zalewski, M, 2009a. Strong Attractors and TCP/IP Sequence No Analysis. http/www.lcamtuf.coredumo.cx/oldtcp/ 2001. Zalewski, M, 2009b. Strong Attractors and TCP/IP Sequence No Analysis. http/www.lcamtuf.coredumo.cx/newtcp/,2002. Zalewski, M, 2009c. Passive OS Fingerprinting Tool. http/www.lcamtuf.coredumo.cx/p0f.shtml, 2006.

[351]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Performance Evaluation and Comparison of Speech Compression using Linear Predictive Coding and Discrete Wavelet Transform Javaid A. Sheikh*, Shabir A. Parah, Sakeena Akhtar, G. Mohiuddin Bhat Post Graduate Department of Electronics and IT, University of Kashmir, Srinagar, India

Abstract This paper presents a tutorial overview of speech coding techniques with emphasis on Linear Predictive Coding and Discrete Wavelet Transform that are part of speech processing standards for wireless communication for compression of speech signal to improve the data rates. We attempt to provide a comparative study between the two compression techniques based upon the survey of relevant literature about the speech compression. We think that this approach will not only mention key references but will also provide a valuable background to the beginner. We start with a brief introduction to speech compression and its types and consequently Linear Predictive Coding and Discrete Wavelet Transform have been discussed. Finally, we present concluding remarks on these two techniques followed by a discussion of opportunities for future research.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Speech Compression; LPC; DWT; Peak Signal to Noise Ratio; Compression Ratio; Mean Square Error

1. Speech Compression The purpose of speech compression is to reduce the number of bits required to represent speech signals (by reducing redundancy) in order to minimize the requirement for transmission bandwidth (e.g., for voice transmission over mobile channels with limited capacity) or to reduce the storage costs (e.g., for speech recording) (Rabiner L.R and Schafer R.W.,2012). Speech compression, especially at low bit rate explores the nature of human speech production mechanism. When we speak, the air from lungs push through the vocal tract and out of the mouth to produce a sound. For some sounds for example, a voiced sound, or vowel sounds of ‘a’, ‘i’ and ‘μ’, the vocal cords vibrate (open and close) at a rate (fundamental frequency or pitch frequency) and the produced speech samples show a quasi-periodic pattern. For other sounds (e.g., certain fricatives as ‘s’ and ‘f’, and plosives as ‘p’, ‘t’ and ‘k’ , named as unvoiced sound, the vocal cords do not vibrate and remain open during the sound production. The waveform of unvoiced sound is more like noise. The change of the shape of the vocal tract (in combination of the shape of nose and mouth cavities and the position of the tongue) produces different sound and the change of the shape is relatively slow (e.g., 10–100 ms). This forms the basis for the short-term stationary feature of speech signal used for all frame-based speech coding techniques. 2. Speech Compression Techniques Speech compression aims to remove redundancy in speech representation to reduce transmission bandwidth and storage space (and further to reduce cost). There are in general three basic speech compression techniques, which are waveform-based, parametric-based, and transform coding techniques. 2.1.

Waveform Based Speech Compression

As the name implied, waveform based speech compression is mainly to remove redundancy in the speech waveform *

Corresponding author Tel.: +91-9419090554. E-mail address: [email protected].

ISBN: 978-93-82288-63-3

Sheikh et al/COMMUNE – 2015

and to reconstruct the speech waveform at the decoder side as closely as possible to the original speech waveform. Waveform-based speech compression techniques are simple and normally low in implementation complexity, whereas their compression ratios are also low. The typical bit rate range for waveform-based speech compression coding is from 64 kb/s to 16 kb/s. At bit rate lower than 16 kb/s, the quantization error for waveform-based speech compression coding is too high, and this results in lower speech quality. Typical waveform-based speech compression codecs are PCM and ADPCM (Adaptive Differential PCM). 2.2.

Parametric Based Speech Compression

Parametric-based speech coding is based on the principles of how speech is produced. It is based on the features that speech signal is stationary or the shape of the vocal tract is stable in short period of time (e.g., 20 ms). The spectral characteristics of the vocal tract can be represented by a time-varying digital filter. For each speech segment, the vocal tract filter parameters, voiced/unvoiced decision, pitch period and gain (signal energy) parameters are obtained via speech analysis at the encoder. These parameters are then coded into binary bit stream and sent to transmission channel. The decoder at the receiver side will reconstruct the speech (carry out speech synthesis) based on the received parameters. Compared to waveform-based codecs, parametric-based codecs are higher in implementation complexity, but can achieve better compression ratio. The quality of parametric based speech codecs is low, with mechanic sound, but with reasonable intelligibility. A typical parametric codec is Linear Prediction Coding (LPC) vocoder (discussed in detail below) which has a bit rate from 1.2 to 4.8 kb/s and is normally used in secure wireless communications systems when transmission bandwidth is very limited. 2.3.

Transform Coding Techniques

As parametric-based codecs cannot achieve high speech quality because of the use of simple classification of speech segments into either voiced or unvoiced speech and simple representation of voiced speech with impulse period train, transform coding techniques were proposed (Satish Kumar et al., 2012). In Transform coding, the data to be compressed can be natural data like audio/speech signal or photographic images. In Transform coding technique some of the information having least importance about the data is discarded and hence lowering the bandwidth. The remaining data of valuable information is then compressed with the help of various types of Transform Coding Techniques such as Fast Fourier Transform (FFT), Discrete Cosine Transform (DCT), Discrete Wavelet Transform (DWT) and Discrete Packet Wavelet Transform (DPWT). In this paper we will be considering DWT in detail used for data transform into another mathematical domain for suitable compression. 3. Linear Prediction Coding The fundamental idea is that for a given segment of speech (e.g., 20 ms of speech, which corresponds to 160 samples at 8 kHz sampling rate), if we can detect whether it is voiced or unvoiced and estimate its LPC filter parameters, pitch period (for voiced signal) and its gain (power) via speech signal analysis, we can then just encode and send these parameters to the channel/network and then synthesize the speech based on the received parameters at the decoder. For a continuous speech signal which is segmented for 20 ms speech frames, this process is repeated for each speech frame. At the encoder, the key components are pitch estimation (to estimate the pitch period of the speech segment), voicing decision (to decide whether it is a voiced or unvoiced frame), gain calculation (to calculate the power of the speech segment) and LPC filter analysis (to predict the LPC filter coefficients for this segment of speech). These parameters/coefficients are quantized, coded and packetized appropriately (in the right order) before they are sent to the channel and are then used to synthesize the speech based on the received parameters at the decoder (Amol. R. Madane et al., 2009). The parameters and coded bits from the LPC encoder are listed below.  Pitch period (T): for example, coded in 7 bits as in LPC-10 (together with voicing decision).  Voiced/unvoiced decision: to indicate whether it is voiced or unvoiced segment. For hard-decision, a binary bit is enough.  Gain (G) or signal power: coded in 5 bits as in LPC-10.  Vocal tract model coefficients: or LPC filter coefficients, normally in 10-order, i.e. a1, a2…, a10, coded in 41 bits in LPC-10. 4. Discrete Wavelet Transform Wavelet is a new technique for examining and comparing a speech signal, it is more advantageous technique because it holds both time and frequency aspect of a signal. Wavelet breaks speech signal into different coefficients. Some of the coefficients having small value are treated as insignificant during data compression and are hence discarded (Amol. R. Madane et al., 2014). Wavelets are obtained by a single Mother Wavelet by delay and shifting.

[353]

Sheikh et al/COMMUNE – 2015

(𝑡 − 𝑏) 𝜓 𝑎 𝑎 √ Where ‘a’ is the scaling parameter and ‘b’ is the shifting parameter. Choosing a Mother Wavelet function used in designing high quality speech coders is of prime importance. In DWT, a signal to be examined is passed through an analysis filter bank followed by a particular decomposition level. At each decomposition level, the analysis filter bank consists of a high pass and low pass filter. The signal is passed through a series of such high pass and low pass filters. The output of high pass filter is known as detail coefficients and contains the valuable information of the signal while as the output of the low pass filter is known as approximation filter and contain least information of the signal. Detail coefficients are low scaled high frequency components while approximation coefficients are high scaled low frequency components. The frequency components that are not very prominent in the original signal will have very low amplitude and this part of the DWT signal can be discarded without loss of any valuable information, allowing data compression at higher data rates. Wavelets decompose a signal into different resolutions or frequency bands. Signal compression is based on the concept that selecting small number of approximation coefficients and some of the detail coefficients can represent the signal components accurately. Choosing a decomposition level for the DWT depends on the type of signal being used or parameters like entropy. ψa,b=

1

5. Performance Evaluation To evaluate the performance of Linear Predictive Coding and Discrete Wavelet Transform various performance parameters related to speech signals are calculated. These include mean square error, peak signal to noise and compression ratio. (Dr. Tarik Zayad et al., 2005). The above quantities are calculated using the following formats: MSE = {∑ 𝑒𝑟𝑟^2}/N Where err is the error signal ∑ (𝑥(𝑛))−𝑦(𝑛))2

NRMSE=√∑𝑛

2 𝑛(𝑥(𝑛))−𝑢(𝑛))

Where x(n) is the original signal, y(n) is the reconstructed signal and u(n) is the mean of speech signal. N is the size of original signal. PSNR = 10log10{max(A)/MSE} Where A is the original signal CR = x(n)/y(n) Where x(n) is the length of original signal, y(n) is the length of compressed signal. A brief comparative study between the two techniques can be viewed from the tables 5.1 -5.2 Table 5.1 (Shiyo. M. Joseph et al., 2011) Speech Signal S1 S2 S3 S4 S5 S6 S7 S8 S9 S10

MSE 0.0022 0.0590 0.0006 0.0040 0.0997 0.0009 0.0008 0.0045 0.0144 0.0298

DWT PSNR 86.018 69.556 82.302 82.899 66.937 80.008 80.835 82.419 76.602 72.971

NRMSE 1.621 1.562 1.577 1.568 1.669 1.536 1.553 1.592 1.524 1.654

MSE 0.0321 0.0205 0.0621 0.0412 0.0355 0.0645 0.0318 0.0486 0.1344 0.0883

LPC PSNR 80.557 86.814 63.556 66.519 58.004 80.823 74.750 68.321 72.711 70.625

NRMSE 0.043 0.023 0.226 0.121 0.290 0.075 0.050 0.119 0.049 0.075

Table 5.2 (D. Ambika et al., 2012) Speech signal S1 S2 S3 S4 S5 S6 S7

PSNR 15.18606 14.84219 14.26461 15.19349 14.64838 15.35273 14.54993

DWT NRMS 0.00255 0.00311 0.00434 0.00254 0.00348 0.00232 0.00368

CR 0.9684 0.3684 0.9820 0.7807 0.6150 0.7748 0.6280

[354]

PSNR 14.57136 14.21918 13.66169 14.58647 14.04234 14.74981 13.92235

LPC NRMSE 0.00364 0.00446 0.00614 0.00361 0.00493 0.00328 0.00529

CR 1.0146 1.0244 1.0010 1.0057 1.0046 1.0010 1.0298

Sheikh et al/COMMUNE – 2015

Speech signal S8 S9

PSNR 14.93618 15.57102

DWT NRMS 0.00295 0.00204

CR 0.8096 0.6147

PSNR 14.32655 14.96110

LPC NRMSE 0.00419 0.00289

CR 1.0087 1.0010

Table 5.3 (Manvendr et al., 2012) Man/Women Man speech(clean) Man speech(noisy) Women speech(clear) Women speech(noisy)

Max error using LPC 0.497 0.0480 0.380 0.620

Fig 5.4: Original and reconstructed signal using LPC

Max error using DWT 0.061 0.063 0.070 0.100

Fig 5.5: Original and reconstructed signal using DWT

6. Conclusion and Future Work This paper presents a comparative study of Linear Predictive Coding and Discrete Wavelet Transform. Linear Predictive Technique causes time delay and some loss of quality. But they are negligible in terms of cost when compared with the advantages of storage space saving, bandwidth requirement etc. In general a good reconstructed signal is one having low MSE and high PSNR and from the above given results it is clear that the signal compressed with DWT shows better results than LPC. Another advantage of DWT over LPC is that the compression factor is not constant and can be varied. Thus we conclude that the performance of DWT technique is better than LPC. The performance of each of the techniques has been presented in the form of various tables and plots. In future both these techniques can be combined together to get a hybrid technique for speech coding so that much better results can be obtained. References Rabiner L.R, and Schafer R.W. Digital Processing Of Speech Signal, Prentice Hall. 2012 Satish Kumar, O. P. Singh, G. R. Mishra, Sourabh Kumar Mishra, Akanksha Trivedi. Speech Compression and Enhancement using Wavelet Coders, International Journal of Electronics and Communication and Computer Engineering, Vol.3, 2012. Amol. R. Madane, Zalak Shah, Raina Shah, Sanket Thakur. Speech Compression Using Linear predictive coding, Proceedings of the international Workshop on Machine Intelligence Research, 2009. Hemant Amhia, Ratish Kumar. A new approach of speech compression by using DWT and DCT, International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering, Vol.3, July 2014. Dr. Tarik Zayad, Ahlam Hanoon. Speech Signal Compression using Wavelet and Linear Predictive Coding, Al-Khwarizmi Engineering Journal, Vol.1,pp 52-60,2005. Shiyo. M. Joseph, Firoz Shah A., Babu Auto. Comparing Speech Compression using Waveform coding and Parametric coding, International Journal of Electronics Engineering, 3(1), 2011. D. Ambika, V. Radha. A comparative study between Discrete Wavelet Transform and Linear Predictive Coding”, World Congress on Information and Communication Technologies, 978-1-4673-4873-4804-1, IEEE, 2012. Manvendr, A. K. Jaiswal, Mukesh Kumar, Alok Singh, “Voice synthesis using wavelet transform”, International Journal of scientific and research Publications, volume 2, May 2012.

[355]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

0.5V Design of Signal Conditioning Circuit for ECG Signal Retrieval I. N. Beigh* and F. A. Khanday Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India

Abstract Today biomedical telemetry is an important field of research. Biomedical telemetry makes it possible to decentralize the available facilities and expert human resource and therefore giving comfort to the patients across the globe. This field also saw the boom due the fact that patient need to be continuously monitored and the patient data is later analysed at his place or at far distance using telemetry. For these cases, data logging is a very important step in the cure. Before the corresponding signals from patients are digitized and stored on a digital media, signals need to be amplified and filtered from noise etc. This circuitry at the first step is called signal conditioning circuitry and is of great importance for biomedical systems as the overall measurement in essence will depend on the efficiency of this circuitry. In this paper an ultra low-voltage signal conditioning circuit for ECG signal retrieval is presented using sinh-domain technique. The proposed circuit design offers attractive features of low-voltage operation, resistorless realizations, electronic tunability and space optimization essential for biomedical signal analysis. In addition, the inherent class-AB operation of sinh-domain filters offers the capability for handling signals greater than the bias current, leading to a power saving. The aforementioned benefits have been evaluated through simulation results using the HSPICE simulation tool and 0. 35µm CMOS process.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Biomedical; Engineering; Biomedical Systems; ECG Analysis; Signal Conditioning; Data Logging; Sinh-Domain Technique

1. Introduction Biomedical telemetry is growing very fast as a field of research as the number of patients is getting benefited due to technology across the globe. Data loggers constituent one of the most important hardware section of biomedical telemetry. Data loggers also play an important role in the continuous monitoring of the patients. Data logging involves collecting, analyzing and storing data. Data logging systems typically monitors an event or process over a period of time. It is an electronic instrument, which has to be portable; battery operated and should interface with computer or as a stand-alone instrument. Data loggers find applications in a number of fields. The significant areas are Aeronautics, meteorology, automotive engineering, biomedical engineering etc. In biomedical systems data logger is used to monitor health of patient by analyzing different bioelectrical signals such as EEG, ECG, EMG etc. Data loggers consists essentially two parts: signal conditioning circuitry and data storage. The data stored is analysed for diagnosis/treatment purposes. Therefore, the efficient diagnosis/treatment of the patients depends on the accuracy with which the data is stored which in essence depends on the efficiency with which the biomedical signals are retrieved by the signal conditioning circuitry. Bioelectrical signals convey the useful information regarding the health of a patient on which diagnosis is made and the patient is treated accordingly. Extraction of these signals is a complicated task because apart from being corrupted by external noise, every bioelectrical signal acts as noise for every other signal making it difficult to filter out the desired one. There are different kinds of biological signals like ECG, EEG, EGG etc and each one has its own significance for example ECG gives the information about the state and behaviour of heart, So it is very important to get these signals efficiently, effectively and noise free. As it is desired to have a portable device which the patient can carry with him/her in order to have real-time monitoring of biomedical signals which makes it essential that the circuit should consume less power so that battery life will be long.

*

Corresponding author. Tel.: +91 9797730217. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Beigh and Khanday/COMMUNE-2015

As far as circuit design techniques are concerned, the companding filtering technique has emerged a choice for designing portable circuits. This is because a number of advantages required for portable systems such as low-voltage low-power operation, electronic tunability, high dynamic range, resistorless realizations etc. are offered by it. SinhDomain filtering is an important subclass of companding filters realized using non-linear transconductors with a hyperbolic-sine/cosine relationship between their output current and input voltages. Besides, the features offered by the other subclasses of companding filters, it offer an additional feature of class-AB operation and, consequently, they are capable for handling currents with an amplitude greater than that of the bias current and therefore leading to power saving. [Serdijn et al., 1999, Poort et al., 1999, Katsiamis et al., 2008, Sawigun and Serdjin, 2009, Kasimis and Psychalinos, 2011, Kasimis and Psychalinos, 2012, Khanday et al., 2013, Kant et al., 2013, Kafe et al., 2014, Skotis et al., 2014, Tsirimokou et al., 2015, Khanday et al., 2013, Panagopoulou et al., 2013, Roumelioti et al., 2015, Khanday et al., 2014]. In this paper, an ultra-low-voltage and ultra-low-power signal conditioning for the retrieval of ECG signal realized using the Sinh-Domain technique, is introduced. The proposed circuit design offers attractive features, as mentioned above, essential for biomedical signal analysis. The paper is organized as follows: the proposed signal conditioning circuit is given in section 2, the simulation results are given in section 3 and the paper is concluded in section 4. 2. Proposed Sinh-omain Signal Conditioning Circuit The block diagram of signal conditioning system is given in Fig. 1. From the block diagram, it is clear that it needs two sub-blocks, namely, an instrumentation amplifier and the bandpass filter. These blocks have been implemented in sinh-domain technique. The main building blocks of sinh-domain technique are non-linear transconductor cell, summer/subtractor, integrator, and two-quadrant divider. The non-linear transconductor cell is depicted in Fig. 2 and it realizes the expression given in (1) in the case of the hyperbolic sine output (Kafe et al., 2014)

 vˆ  vˆ IN  i  2 I o sinh IN   nVT

  

(1)

(2) in the case for hyperbolic cosine output

 vˆ  vˆ IN  i  2 I o cosh IN   nVT

  

(2)

(3) in the case for inverted hyperbolic sine output

 vˆ  vˆ IN    i  2 I o sinh IN   nVT  and (4) in the case for weighted hyperbolic sine output  vˆ  vˆ IN    i  2 KI o sinh IN   nVT 

(3)

(4)

where Io is a dc current, VT is the thermal voltage (26mV @ 27oC), n is the subthreshold slope factor (1
v

v

and IN  , IN  are the voltages at the non-inverting and inverting inputs, respectively. The sinh-domain realization of three input summer/subtractor is given in Fig. 3 (a). Defining the SINH operator by equation (5)

 ˆ  VDC i  SINH (ˆ )  2 I o sinh  nVT

  

(5)

and after some routine sinh-domain algebraic manipulations, it can be deduced that the transfer function of the SinhDomain summer/subtractor is given by equation (6)

SINH vˆOUT   K1 SINH vˆ1   K 2 SINH vˆ 2   K 3 SINH vˆ3 

(6)

The sinh-domain realization of lossless integrator is given in Fig. 3(b). After some routine sinh-domain algebraic manipulations, it can be deduced that the transfer function of the Sinh-Domain lossless integrator is given by equation (7)

H ( s) 

SINH vˆout  1  SINH vˆin  ˆs

[357]

(7)

Beigh and Khanday/COMMUNE-2015

Where the time constant ˆ 

Cˆ nVT 2I o

Also, the two-quadrant divider employed in the lossless integrator is constructed from appropriately configured S cells as shown in Fig. 3(c) in order to realize the relationship:

iout  Io.

i1 [Kasimis and Psychalinos, 2012]. Using i2

the blocks given in Figs. 2, 3(a), 3(b) and 3(c), the instrumentation amplifier and bandpass filter were designed as given in Figs. 4 and 5 respectively. After some routine sinh-domain algebraic manipulations, the transfer functions for instrumentation amplifier and bandpass filter can be respectively written as:

SINH vˆout   2K1K 2 SINH vˆ1   SINH vˆ2 

(8)

Where K1, K2 are the constant multipliers.

SINH vˆout  b0  2 SINH vˆin  b2 s  b1s  b0 Where, the time-constant of the integrators is given by the relationship ˆi 

(9)

ˆ nV bj C i T , j  1,2 2 I oi b j 1

Employing the blocks in Fig. 4 and 5, the optimized proposed sinh-domain design of signal conditioning circuit is given in Fig. 6, where it was considered that S and/or C cells with the same input voltages could be combined into one resulting into multiple output non-linear transconductance cell.

Fig. 1. Block diagram of signal conditioning circuit.

Fig. 2. Non-linear transconductor cell (Multiple output).

[358]

Beigh and Khanday/COMMUNE-2015

(a)

(b)

(c) Fig. 3. (a)Sinh-domain three-input summer/subtractor, (b) Sinh-domain lossless integrator, and (c) Sinh-domain realization of Two-quadrant multiplier.

Fig. 4.Sinh-domain realization of instrumentation amplifier.

Fig. 5. Sinh-domain realization of bandpass filter.

Fig. 6. Optimized proposed sinh-domain realization of signal conditioning circuit.

3. Simulation Results A bias scheme with dc supply voltages VDD=0.5V and VDC=0.1V, and dc current Io=100pA was employed. Using

[359]

Beigh and Khanday/COMMUNE-2015

MOS transistors models provided by the AMS 0.35µm CMOS process, the aspect ratios of MOS transistors in order to be biased in the weak inversion region, are given in Table I. The obtained transient response of the differential amplifier with gain variations (g = 2, 4, 8) is shown in Fig. 7. The frequency response of the band pass filter centred around 1Hz and the demonstration of its electronic tunability are given in Figs. 8(a) and 8(b) respectively. In order to demonstrate the electronic tunability of the differential amplifier and bandpass filter, the values given in Table 2 were used with value of capacitor as 100pF in all the cases. The obtained values of the gain and center frequencies are given in Table II. The small deviations in the parameters are due to the imperfections of the transistors and current mirrors. The overall signal conditioning circuit was tested for noisy ECG signal which was obtained by adding the applied ECG signal with the noise signal (50 kHz sinusoidal signal with peak amplitude of 0.2pA). The overall response of the designed circuit is given in Fig. 9 where the applied ECG signal, the noisy ECG and the obtained results are shown. The obtained response from the circuit shows that the effect of the noise is significantly reduced and the ECG signal is retrieved with very small noise. 4. Conclusion Signal conditioning circuit to retrieve the ECG signal from its noisy counterpart is introduced in this paper. The design can also be extended to retrieve other biomedical signals as well. Besides, the design can be moderated for multi-electrode biomedical systems. The design offers the benefits of low-voltage operation, resistorless realizations, electronic tunability and space optimization. Therefore, the proposed design shall prove a valuable addition to the area of biomedical signal processing.

Fig. 7. Demonstration of gain change of differential amplifier via Transient response.

(a)

(b)

Fig. 8. (a) Response of bandpass Filter, and (b) Demonstration of Electronic tunability of bandpass filter.

Fig. 9. Demonstration of ECG Signal, Noisy ECG signal and over-all response of the conditioning circuit.

[360]

Beigh and Khanday/COMMUNE-2015

Table I. Aspect Ratio of MOS Transistors in Weak inversion. MOS Transistor Mp1 – Mp4

Aspect Ratio ( µm) (W/L) 35/0.55

Mp5 – Mp12

58/0.6

Mn1 – Mn8

21/1

Table II. Electronic tunability parameters for differential amplifier and Bandpass filter.

K0

Differential Amplifier Gain K2 Theoretical Simulated

I0(pA)

Bandpass filter f0 Theoretical

Simulated

1

1

2

1.885

50

4.7Hz

3.98Hz

3

1

4

3.83

75

7.05Hz

8.12Hz

2

2

8

7.57

100

9.41Hz

12Hz

References Serdijn, W., Kouwenhoven, M., Mulder, J., van Roermund, A., 1999 Design of High Dynamic Range Fully Integratable Translinear Filters, Analog Integrated Circuits and Signal Processing, 19, p. 223. Poort, P., Serdijn, W., Mulder, J., van der Woerd, A., 1999, A 1-V class AB translinea integrator for filters applications Integrated Circuits and Signal Processing, 21, p. 79. Katsiamis, A., Glaros, K. K., Drakakis, E., 2008, Insights and Advances on the Design of CMOS Sinh Companding Filters, IEEE Transactions on Circuits and Systems-I, 55, p. 2539. Sawigun, C., Serdijn, W., 2009, Ultra-low power, class-AB, CMOS four quadrant current multiplier, Electronics Letters, 45, p. 483. Kasimis, C., Psychalinos, C., 2011, 0.65 V class-AB current-mode four-quadrant multiplier with reduced power dissipation, International Journal of Electronics and Communications (AEU), 65, p. 673. Kasimis, C., Psychalinos, C., 2012, Design of Sinh-Domain Filters Using Complementary Operators, International Journal of Circuit Theory and Applications, 40, p. 1019. Kasimis, C., Psychalinos, C., 2012, 1.2V BiCMOS Sinh-Domain Filters, Circuits Systems and Signal Processing, 31, p. 1257. Khanday, F. A., Shah, N. A., 2013, A Low Voltage and Low Power Sinh-Domain Universal Biquadratic Filter for low frequency Applications, Turkish Journal of Electrical Engineering and Computer Sciences, 21, p. 2205. Kant, N. A., Khanday, F. A., Psychalinos, C., Shah. N. A., 2013, 0.5V Sinh-Domain Realization of Activation Functions And There Employment in Neuron Network Design for Logic Gate Implementations, ASP Journal of Low Power Electronics, 9, p. 1. Kafe, F., Khanday, F. A., Psychalinos, C., 2014, A 50mHz Sinh-Domain highpass filter and its application for removing the baseline wander in an ECG signal acquisition system, Circuits, Systems and Signal Processing, 33, p. 3673. Skotis, G. D., Khanday, F. A., Psychalinos, C., 2014, Sinh-Domain Complex Integrators, International Journal of Electronics, DOI:10.1080/00207217.2014.963891 Tsirimokou, G., Psychalinos, C., Khanday, F. A., Shah, N. A., 2015, 0.5V Sinh-Domain Differentiator, International Journal of Electronics, 3, p. 34. Khanday, F. A., Pilavaki, E., Psychalinos, C., 2013, Ultra Low-Voltage, Ultra Low-Power Sinh-Domain Wavelet filter for ECG analysis, ASP Journal of Low Power Electronics, 9, p. 1. Panagopoulou, M., Psychalinos, C., Khanday F. A., Shah, N. A., 2013, Sinh-Domain Multiphase Sinusoidal Oscillator, Microelectronics Journal, 44, p. 834. Roumelioti, K., Psychalinos, C., Khanday, F. A., Shah, N. A., 2015, 1.2V Sinh-Domain Allpass Filter, International Journal of Circuit Theory and Applications, 43, p. 22. Khanday, F. A., Kasimis, C., Psychalinos, C., Shah, N. A., 2014, Sinh-Domain Linear Transformation Filters, International Journal of Electronics, 101, pp. 241.

[361]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

A Survey of Spell Checkers Available for Hindi and Punjabi Kamal Deep Garg*a, Ajit Kumarb *a

Department. of Computer Science, Lovely Professional University, Phagwara, India b Department of Computer Science, Multan Mal Modi College, Patiala, India

Abstract Spell checker is a program used for examining the spelling errors and removing the faults in the text or a document. Spell checker is a platform, which is used to identify the words, which are misspelled and update the user about those misspelled words. The components of spell checker either replaces the misspelled word or ask the user to select the right word from the recommendations, which are existing for that specific misspelled word. Spell checker may be an operation which has ability of functioning on a huge part of text or as an element of larger application such as text editor, email, blog writing, keyword searching. Developing a spell checker for Indian languages such as Punjabi, Hindi, and Oria etc. step-up several new problems that are not faced in English, which makes the design of spell checker very challenging. The very first constraint for developing any spell checker is to have dictionary of different words of that language which will work as word list. This paper survey the different available spell checker for Indian language (Hindi and Punjabi) and their approach used to detect and correct spell errors.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Spell checker; Error detection; Error Correction; Natural Language Processing

1. Introduction Spell-checkers are the basic tools needed for word processing and document preparation. It is one of the most Studied Natural Language Processing (NLP) task. Many NLP applications like Machine Translation Systems, Text to Speech Systems, and Information Retrieval Systems require automated spell checking of text. Many different techniques for detection and correction of spelling errors are based on English. Designing a spell checker for Indian languages such as Hindi and Punjabi has many new research challenges that are not found in English, which complicates the design of the spell checker 2. Error Correction Spelling and typing errors are well-to-do in human produced electronic text. Error in word belongs to one of the two types: non-word error and real word error. Non-word error are those errors normally emerge due to an incorrect key press or shortfall of understanding of spelling of the right word. There are different types of non-word error, which are explained as follows: a) Typographic Errors: Typographic errors normally happen when person knows the correct spelling but the error arises because of the typing fault or some kind of mechanical downfall. Example: ਕੀਮਤ--ਕੀਮਰ, कीमत—कीमर. b) Cognitive Errors: Cognitive errors arise because of lack of knowledge on the part of user or there may be some misinterpretation present in the mind of the user. Example: ਪ੍ਰਸਾਦ--ਪ੍ਰਸ਼ਾਦ, प्रसाद –प्रशाद.

* Corresponding author. Tel.:+91 9988 493359. E-mail address: [email protected].

ISBN: 978-93-82288-63-3

Garg and Kumar/COMMUNE – 2015

c) Phonetic Errors: Phonetic errors arise when the user writes a word which is phonetically correct but there may be incorrect order of the characters for the required word. This type of error is a special type of cognitive errors. Example: ਮਮਹਨਤ--ਮੇਹਨਤ, मेहनत --ममहनत Real word error: These are those when word does not fit into the context of the sentence. There are different types of real word error which are explained as follows: a) Transposition Error: This type of occurs when two adjacent letters are written in swapped way. For Example: -ਕਮਲ ਪ੍ਾਣੀ ਮ ਿੱ ਚ ਉਗਦਾ ਹੈਕਲਮ ਪ੍ਾਣੀ ਮ ਿੱ ਚ ਉਗਦਾ राम, पंकज के पास गया है राम,पंकज के

पास गाय है . b) Substitution error: This type of error occurs when one or more letters are substituted by some another letter. Example: -ਉਹ ਸ਼ੀਸ਼ਾ ਦੇਖਦਾ ਹੈਉਹ ਸ਼ੀਸਾ ਦੇਖਦਾ ਹ राम अच्छा श्रोता है राम अच्छा सोता है c)

Deletion Error: This type of error occurs when one or more letter is removed from the required word. For Example: -ਉਹ ਯੋਗਾ ਕਰਦਾ ਹੈਉਹ ਯੋਗ ਕਰਦਾ ਹੈ मोहन कला प्रस्तुत करे गामोहन कल प्रस्तुत करे गा

d) Insertion error: This type of error occurs when one or more extra letters are inserted in the required word. Example:-ਉਸਦੇ ਕਮਰ ਮ ਿੱ ਚ ਦਰਦ ਹੈਉਸਦੇ ਕਮਰਾ ਮ ਿੱ ਚ ਦਰਦ ਹੈचत े क नाम के घोड़े मे बहुत बल है चेतक नाम के e)

घोड़े मे बहुत बाल है Run- on error: This type of error occurs when two or more valid words are erroneously typed side by side without a space in the middle.

Example: - ਸਬ ਦਾ ਆਦਰ ਮਾਣ ਕਰੋਸਬ ਦਾ ਆਦਰ ਮਾਣ ਕਰੋ घर मे परम – आत्मा का ननवास है घर मे परमात्मा का ननवास है

3. Techniques of Error Detection and Correction in Spell Checker 3.1 Error Detection 3.1.1.

N-Gram Analysis

A method to find spelled words in text that are incorrect and for non-word errors is the n-gram analysis method. Ngrams are controlled which leaves no comparing of each word in a text to a dictionary. An n-dimensional matrix having actual n-gram frequencies are acquired. The word is considered flagged if a missing or odd n-gram is encountered, else not. A collection of successive characters from a string is what n-gram is. It is classifies as uni-gram if n is 1, bi-gram if n is 2 and tri-gram if n is 3. Bi-gram array of 2-D array of size 41*41 where element depict all doable 2-D letter blend of alphabet is the simplest. It needs no awareness of the lingo and is frequently termed as lingo self-sufficient or an unbiased string matching algorithm. The resemblance amid two strings is obtained by finding the amount of exclusive n-grams that are shared and then scheming a resemblance coefficient, that gives the number of common n-grams (intersection), separated by the union of n-grams in the two words. 3.1.2.

Dictionary Lookup

This technique simply lookup every word in the dictionary if the word is not there then it is said to be an error. A dictionary having most frequently used words along with extra dictionaries for specific areas as computer science or economy is often considered to be a large dictionary. They use much space and longer access time. Checking every word with reference to a dictionary can find the non-word errors. Keeping the dictionary up to date and also provide better system response time are the problems faced. Very small dictionary will provide the user with much fake denial of legitimate words. Hash table is the most common technique applied to gain quick access. Computing the hash address and retrieving the word available at that address in the pre constructed hash table is done to look up a string. If the word available at the hash address is dissimilar from the Input string, a misspelling is depicted. The advantage of hash tables is their random-access character that removes the bulky amount of comparisons essential to look for the dictionary. The core shortcoming is the want to develop a quick hash function that averts collisions. Hash function is calculated for storing the words in the dictionary and the vector entries are set equivalent to the evaluated values to true. To uncover out whether a word belongs to the dictionary, one can evaluate the hash values used for that word and search in the vector. If all entries equivalent to the values are factual, then the word is in the dictionary, else it is not.

[363]

Garg and Kumar/COMMUNE – 2015

3.2 Error Correction Techniques 3.2.1.

Edit Distance

Edit Distance Edit distance is a simple technique. First edit distance spelling error correction algorithm was implemented by Damerau Simplest method is based on the assumption that the person usually makes few errors if ones, i.e. errors from keyboard input therefore for each dictionary word .The minimal number of the basic editing operations (insertion, deletions, and substitutions) necessary to covert a dictionary word into the non-word As edit distance is useful for correcting errors resulting from keyboard input, since these are often of the same kind as the allowed edit operations. It is not quite as good for correcting phonetic spelling errors. 3.2.2.

N-Gram Technique

N-grams can be used in two ways, either without a dictionary or together with a dictionary. Letter N-gram including tri-gram, bi-gram and uni- gram have been used in variety of ways in text recognition and spelling correction techniques. Used without a dictionary, n-grams are employed to find in which position in the misspelled word the error occurs. If there is a unique way to change the misspelled word so that it contains only valid n-grams, this is taken as the correction. The performance of this method is limited. It is that it is simple and does not require any dictionary. Together with a dictionary, n-grams are used to define the distance between words, but the words are always checked against the dictionary. 3.2.3.

Similarity Keys

In this technique we map every string into a key such that the similarly spelled strings will have similar key. It is known as SOUNDEX system. In this it is not necessary to directly compare misspelled string to each word in dictionary. For example suppose a customer comes in a bank and said his name is zayheijendn. So in this case you cannot ask him to speak his name as his English is poor and others customers are waiting. So we want a key which sounds like his name and find a name resembles with it. 3.2.4.

Rule Based Technique

It attempts to represents knowledge commonly spelled errors i.e. mistyping by mistake in the form of rules for converting it into valid words. Each word which is correct can be taken as a suggestion. It consist the process consist of applying all applicable rules to a misspelled string. 3.2.5.

Neural Network

This method works on small dictionaries. Back propagation algorithm is used in neural network. It consist of 3 layers input, hidden and output layer. It has potential to adapt specific error pattern. In this input information is represented by on- off pattern. A=1 indicates that node is turned on and A=0 means node is turned off. For e.g. in spell checking applications misspellings represented as binary n-gram vector may be taken as input and output pattern might be vector of m elements means number of words in lexicon. 3.2.6.

Probabilistic Techniques

This is based on some statistical features of the language-gram technique led to probabilistic technique in spell correction and text recognition. Two methods are used in this. Transition probabilities which is similar to n-grams .It is language independent. Confusion probabilities estimates of how. 4. Spell Checker Available for Hindi and Punjabi Jaspreet Kaur et.al (2014) has developed hybrid approach for spell and grammar check for Punjabi language. The main steps of spell checker are Dictionary Creation, Pre-processing, Error Detection, displaying the suggestions and then replacing the error with most suitable suggestion. While implementing the spell checker, the erroneous word is firstly checked in the standard dictionary and the dictionary or lexicon created by the user. If the word is not found in the lexicon, it will be treated as an error. After that, suggestion list will be generated and then list will be sorted displaying the suggestions. Then final output will be displayed after correcting the errors. The average accuracy of the resulting system is 83.5 %. (Jaspreet Kaur, 2014) Ritika Mishra et.al (2013) have developed Raftaar online Punjabi spell checker. This system is based on dynamic programming. Dynamic programming is technique for solving complex problems by breaking them down into simpler sub-problems. The spell checker process consisted of three mechanism: an error detector that detects misspelled words,

[364]

Garg and Kumar/COMMUNE – 2015

a candidate spelling initiator that gives spelling suggestions for the detected misspelled words and an error corrector that choose the finest correct spellings out of the list of candidate spelling. . The main features of this spell checker are big database, online application, easy to operate, email and printing option. This system gives the result accuracy as 80% according to the research work for Punjabi words. (Ritika Mishra, 2013) Rupinderdeep Kaur et.al (2010) have developed spell checker for Gurumukhi and named it as SUDHAAR. The main features of this system are dictionary creation, error detection, error correction, and replacement. It is a standalone application capable of taking input from any source. There are two different applications in the system. First creates the dictionary which is already been developed and executed once to create the dictionary. Second is designed for spelling check of Punjabi text where user gives the input Punjabi text and system detects the error by looking up for that word in the dictionary and provide suggestions for those errors. The system is divided into three modules: first the Punjabi dictionary is created followed by next module of Error detection and error correction, and the last phase replacement. This system detects and corrects only non-real word errors. The system is efficient as it detects approximately 80% of the errors and provides 85% of the correct suggestions. (Rupinderdeep Kaur, 2010) Gurpreet Singh Lehal (2007) has developed a Punjabi spell checker. The major components of this spell checker include the Tokenization and normalization module, Lexicon Lookup/Error Detection Module, and Detection Module. The first step to develop a spell checker is to create a lexicon of correctly spelled words. There are two main issues involved in lexicon development- the size of the lexicon and format of the words. This spell checker only deals with the non-real word errors. The primary suggestion list has been generated using the Reverse minimum edit distance. The suggestion list has been generated using five knowledge sources: KS1 (Character Substitutor), KS2 (Character Inserter), KS3 (Character Remover), KS4 (Subsequent Character Switcher), KS5 (Phonetic Similar Tester. In 81.14% cases of wrongly spelled words, this system provided the correct word on top of the suggestion list. (Lehal, 2007) Amit Sharma et.al (2013) has developed a Hindi spell checker. In this system, mainly real word errors are detected and corrected. They have used a dictionary with word, frequency pairs as their language models. A lookup into the dictionary categorizes a word as correct or erroneous. To produce candidate corrections, the strings at edit distance one and two from the erroneous string is calculated and further the strings that are not present in the dictionary are filtered out. Damerau-Levenshtein is the edit distance that has been used. Words possessing the same edit distance are stored in order of their frequencies. Along with the frequencies of occurrence, they have created 2-grams and 3-grams so as to deal with real word errors. For real word errors checking, every 2-gram, 3-gram and 4-gram of the sentence is checked in the created set. An error occurs in the case if the frequency of the gram is low. This system gives the accuracy of 69.1% of correction and 90.3% of detection (Amit Sharma, 2013). 5. Conclusion The intent of the paper is to discuss different type of error detection and error correction technique in spell checker. This paper also tells about available spell checker for Hindi and Punjabi Language. In future we will design spell checker that can be used to detect and correct error in both Hindi and Punjabi. We will use Dictionary Lookup and Rule based technique to design such a system. References Sumreet Kaur Randhawa, C. S. (2014). Study of Spell Checking Techniques and Available Spell Checkers in Regional Languages: A Survey. International Journal For Technological Research In Engineering , 148-151. Jaspreet Kaur, Kamaldeep Garg (2014). Hybrid Approach for Spell Checker and Grammar Checker for Punjabi. International Journal of Advanced Research in Computer Science and Software Engineering , 62-67. Ritika Mishra, N. (2013). Design and Implementation of Online Punjabi Spell Checker Based on Dynamic Programming. International Journal of Advanced Research in Computer Science and Software Engineering , 987993. Neha Gupta, P. M. (2012). Spell Checking Techniques in NLP: A Survey. International Journal of Advanced Research in Computer Science and Software Engineering , 217-221. Amit Sharma, P. J. (2013). Hindi Spell Checker. Indian Institute of Technology Kanpur . Rupinderdeep Kaur, P. B. (2010). Design and Implementation of SUDHAAR-Punjabi Spell Checker. International Journal of Information and Telecommunication Technology , 10-15. Lehal, G. S. (2007). Design and Implementation of Punjabi Spell Checker. International Journal of Systematic cybernetics and informatics , 70-75.

[365]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Effect of Semiconductor Thickness on Al/p-CuIn0.81Al0.19Se2 Schottky Diodes Usha Parihara*, R. Sachdevaa, C. J. Panchalb, N. Padhaa a Department of Physics & Electronics, University of Jammu, Jammu, J&K, India Applied Physics Department, The M.S. University of Baroda, Vadodara, Gujrat, India

b

Abstract Copper Indium Aluminium Diselenide (CIAS) Al/p-CuIn0.81Al0.19Se2 Schottky diodes of the areas of 6x10-3 cm2 were fabricated by depositing CIAS layer of different thicknesses i.e. 200 nm, 500 nm and 700 nm over Ag (~100nm) coated glass substrates by using flash evaporation technique. Prepared diodes were undertaken for current-voltage (I-V) as well as capacitance-voltage (C-V) characterization at room temperature. The diode parameters such as ideality factor (η), barrier height (bo) and series resistance (Rs) were determined from the downward curvature of current-voltage characteristics using Cheung and Cheung method. However, it has been seen that the Schottky barrier height deduced from the room temperature I-V analysis is less than that obtained from the C-V characteristics when measured at room temperature. This difference in the values of IV and CV may be due to the presence of a native oxide layer at the metal/ CuIn0.81Al0.19Se2 interface.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar Keywords: Schottky Diodes ; Ideality Factor; Barrier Height ; Series Resistance; Current-Voltage; Capacitance-Voltage

1. Introduction Metal-semiconductor (MS) contacts are of great importance because these are present in every semiconductor device. Further, most of the electronic devices that make an integrated circuit are formed of metal-semiconductor contacts (Sze et al, 2007 Rhoderick E.H. and Williams R.H., 1978). M–S structures are important research tools in the characterization of new semiconductor materials, further; the fabrication of these structures plays a vital role in constructing some useful devices (Tung R. T., 2001; Sx et al, 2003; Chattopadhyay P; Daw AN., 1986). The electrical behaviour of the Schottky diodes depends on the density of interface states, which play a crucial role in the determination of diode parameters. Therefore, it is necessary to have a better understanding of the contacts made between metals to the semiconductor (Dogn et al, 2006; J P et al 1991). Depending on the work functions of metals as well as semiconductors and their types (n or p), the contact may be either non-rectifying (ohmic contact) or rectifying (Schottky barrier). Literature reveals that large barriers heights and good thermal stability are critical for the use of Schottky diodes in electronic devices. However, the development of Schottky diodes of CuIn0.81Al0.19Se2 with high barrier heights and low leakage current is still a challenge. The present investigation has therefore, been focused to study the current transport behaviour (I-V as well as C-V) of Al\p-CuIn0.81Al0.19Se2 Schottky diodes of varying thickness of CIAS layers. 2. Experimental Details Al/p-CuIn0.81Al0.19Se2 Schottky diodes of the areas of 6x10-3 cm2 were fabricated by depositing CIAS layer of different thicknesses i.e. 200 nm, 500 nm and 700 nm over Ag (~100nm) coated glass substrates at 473K by flash evaporation technique which was subsequently followed by the deposition of Aluminium thin films (~150 nm) by using

*

Corresponding author. Tel.: not available. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Parihar et al/COMMUNE – 2015 (2015)

mica sheet masks. The overall design of Al/p-CuIn0.81Al0.19Se2 Schottky Diodes is shown in the Fig.1 where CIAS layer is a P-type semiconductor and Aluminium forms a Schottky contact while Silver (Ag) makes ohmic contact (back contact) to the CIAS film. Al/p-CuIn1-xAlxSe2 Schottky diodes were undertaken for the current-voltage (I-V) as well as capacitance-voltage (C-V) characterization at room temperature by using probe station fitted with computer interfaced setup comprising a programmable Keithley Source Meter (model-2400) and Precision programmable LCR meter (Aglient make 4284A). Interfacing of I-V and C-V measurement equipments were achieved by using LabVIEW software by National Instruments (U.S.A).The data of current-voltage measurements were recorded to a personal computer using GPIB data transfer card.

(a)

(b) Fig.1 (a) The animated elevated view of the structure of Al/p-CuIn0.81Al0.19Se2 Schottky diode & (b) Pictorial view of the fabricated Al/p CuIn0.81Al0.19Se2 Schottky diodes.

3. Results and Discussion 3.1.

Current –Voltage (I-V) Characterization

Current-voltage (I-V) characteristics of Al/p-CuIn0.81Al0.19Se2 Schottky diodes, fabricated over CIAS layer with varying thickness of the values of 200 nm, 500 nm and 700 nm have been measured at room temperature as shown in Fig. 2. The current-voltage (I-V) characteristics of Al/p-CuIn0.81Al0.19Se2 Schottky diodes revealed that the current increased with the thickness in forward bias. However, the same has been found decreased with increase in thickness in reverse bias condition. This increase in the current value with film thickness has been explained by the increase in grain size and better enforcement of crystal growth (Cova P and Singh A., 1990). Thus, the main cause of the forward current enhancement has been found to be the grain size effect (Kang et al, 2008). Thus, it is concluded that with the increase in film thickness, the grain size increases and electrical resistivity of thin films systematically decreases which ultimately results in increase in conductivity of the thin films. Thus, the electrical properties of the Al/p-CuIn0.81Al0.19Se2 Schottky diodes are grain size dependent.

log(I)Amp

1E-4

200nm(as-deposited) 500nm(as-deposited) 700nm(as-deposited)

1E-5

1E-6

1E-7 -1.0

-0.5

0.0

0.5

1.0

Voltage(V)

Fig. 2. Forward as well as reverse log(I) versus voltage plot of Al/p-CuIn0.81Al0.19Se2 Schottky diodes with varying CuIn0.81Al0.19Se2 layer thicknesses.

However, the reverse leakage current of Al/p-CuIn0.81Al0.19Se2 Schottky diodes has been found to be decrease from 3.57E-6 A to 9.2E-6 A at -1V for its layer thickness changing from 200nm to 700nm thick layer. Other diode parameters such as ideality factor (), barrier height (bo) and series resistance (Rs) extracted by using Cheung’s method have been presented in Table 1. The values of electrical parameters determined by I–V measurements indicate that the ideality factor () and series resistance (Rs), was high for thin CIAS layer (~200nm) in comparison with thicker ones (700 nm). Conversely, a reverse effect was observed on the barrier height (bo) i.e. smaller values were obtained for thin CIAS [367]

Parihar et al/COMMUNE – 2015 (2015)

layer (~200nm). Therefore, it is observed that 700nm thick layer showed better performance than the thinner layer Schottky diodes. Table 1. Various diode parameters ideality factor (), barrier height (bo) and series resistance (Rs) experimentally obtained by using Cheung’s Method Barrier Height (bo) (eV)

Ideality Factor (η)

CIAS Layer (nm) 200

0.57

2.06

12

500

0.58

1.90

10.5

700

0.59

1.76

8.12

Thickness of

3.2

Series Resistance (Rs) (K-)

Capacitance-Voltage (C-V) Characterization

The depletion region of a Schottky barrier behaves in some respect like a parallel-plate capacitor. It is important to know about the factors used to determine its capacitance, not only because reverse-biased diodes are used in practice as variable capacitors (varactors), but also because measurements of the capacitance under reverse bias can be used to give information about the barrier parameters (Shih I.,and C.X. Qiu, 1988). Therefore, for the comparative study, an attempt has been made to access the doping concentration and barrier height from the C-2–V measurement. The C–V relationship applicable to M-S Schottky barriers can be written as [9]

2Vo  VR  1  2 C q s N A A2

(1) where εs is the permittivity of the semiconductor, VR ; the reverse bias voltage, NA ; the acceptor concentration, q ; the electronic charge, A ; the area of the diode (= 9x 10 -2 cm2). The slope of the above plot gives the value of the acceptor concentration (NA) while the x-axis intercept of the plot of (1/C2) versus VR gives Vo and is related to the builtin potential or diffusion potential (Vbi) by the equation,

Vbi  Vo 

KT q

(2)

where T is the absolute temperature in Kelvin. The zero-bias barrier height from the C–V measurement is defined by

 bo Vbi  Vn

 kT   N v  ln  Vn    q   NA

(3)

  

(4)

Where Vn is the voltage axis intercept of the above plot and represents the energy difference between the Fermi level and the bottom of the conduction band edge in CuIn0.81Al0.19Se2 and

 2m p kT   N v  2 2 h  

3/ 2

(5)

Nv is the effective density of states in the conduction band of CuIn 0.81Al0.19Se2, where mp is the effective mass of CuInAlSe2 = 0.15 m0 eV

 2 N A   2  q o  s A

 dv   2   dc 

(6)

is the acceptor density of CuIn0.81Al0.19Se2, s = 8.1 is the dielectric constant of CuInSe2, εo = 8.85 × 10−12 F m–1 is the permittivity of the free space, A = 9 × 10−2 cm2 is the area of the Schottky diode. Since the value of dielectric constant of CuInAlSe2 is not known, therefore, the value of its close derivative CuInSe2 was taken. The value of carrier concentration (NA) is calculated from the slope of reverse bias C–2-V characteristics is 4.23x1019 cm–3 in close agreement with that obtained from electrical analysis(1019cm–3). A plot of 1/C2 as a function of bias voltage measured at 1MHz frequency for Al\p-CuIn0.81Al0.19Se2 Schottky diode with different CIAS layer thicknesses is shown in Fig. 2. Various electrical parameters deduced from the 1/C2 –V measurement such as bo and NA are presented in Table 2. The barrier height for the Schottky contact obtained by capacitance-voltage measurements were in agreement with those given by I–V characteristics. However, it has been seen that the Schottky barrier height deduced from the room temperature I-V analysis is less than that obtained from the C-V characteristics when measured at room temperature.

[368]

Parihar et al/COMMUNE – 2015 (2015)

22

1x10

200nm 500nm 700nm

21

9x10

21

8x10

21

21

6x10

2

-2

1/C (F )

7x10

21

5x10

21

4x10

21

3x10

21

2x10

21

1x10

0 -2.0

-1.5

-1.0

-0.5

0.0

0.5

1.0

Voltage(V)

Fig. 3. Plot of 1/C2versus V of Al/CuIn0.81Al0.19Se2 Schottky Diode with different CIAS layer thickness at room temperature. Table 2. Various electrical parameters deduced by the 1/C2 –V measurement Thickness of CIAS Layer (nm)

Acceptor Density of States x 1019 NA (cm-3) 1.11

Barrier Height (CV) (eV)

700

Effective Density of States x 1024 Nv (cm-3) 1.26

500

1.38

1.15

0.78

200

1.40

1.21

0.76

0.80

This difference in the values of IV and CV may be due to the presence of a native oxide layer at the metal/ CuIn0.81Al0.19Se2 interface. Any damage at the interface affects the I–V behavior because defects may act as recombination centers or as intermediate states for trap assisted tunnel currents. C–V measurements are less prone to such defects. Since interfacial capacitance and capacitance due to the depletion layer are in series, the total capacitance decreases and as a result C-2 increases. This increase the intercept of a C-2 versus V plot resulting in the increasing of barrier height. Since the I–V method involves the flow of electrons from semiconductor to metal, the barrier height obtained from this method will yield lower barrier heights than from C–V measurements. Another possibility, lowering the barrier height by an image force due to current flow across the barrier, can have some effect. According to Werner and Guttler (Werner J.H. and Guttler H.H., 1991), spatial inhomogeneities at the metal/semiconductor interface of Schottky contact can also cause such differences in the barrier height determined from I–V and C–V measurements. 4. Conclusion In this paper, Al/p-CuIn0.81Al0.19Se2 Schottky diodes of different semiconductor layer thicknesses were fabricated and its effect over the characteristics of Al/p-CuIn0.81Al0.19Se2 Schottky diode has been analyzed on the basis of current–voltage (I-V) as well as capacitance-voltage (C-V) measurements. The results so obtained showed good junction parameters with a value of ideality factor close to unity, a high barrier height and a small value of series resistance for Schottky diodes with a thick CIAS layer (i.e. 700 nm). Thus, depicting reasonably good quality Schottky diodes were formed with 700 nm CIAS layer. References Sze S.M., Kwok. K Ng, 2007. “Physics of semiconductor devices”, Hoboken, New Jersey: John Wiley & Sons, Inc. Rhoderick E.H. and Williams R.H., 1978. “Metal-Semiconductor contacts”, 2nd Edition, Clarendon Press, Oxford. Tung R.T., 2001. Materials Science and Engineering: R, 35 1. Altındal Sx, Karadeniz S, Tugluog lu N, Tatarog lu A, 2003. Solid State Electronics, 47 1847. Chattopadhyay P, Daw AN., 1986. Solid State Electronics, 29(5) 555. Dogan H., Yilirim N.,and Nuhoglu C., 2006. Semicond. Sci. Techonol., 21 822. Sullivan J P, Tung R T, Pinto M R and Graham W R, 1991. J. Appl. Phys. 70 7403. Cova P., Singh A., 1990. Solid-State Electronics, 33 1. Shih I., C.X. Qiu, 1988. J.Appl. Phys., 63(2). Kang Y.J., Park H.J., Choi G.M., 2008. Solid State Ionics, 179 1602. Werner J.H. and Guttler H.H., 1991. J. Appl. Phys., 69 1522.

[369]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Question Classification using Knowledge based Semantic Kernel Mudasir Mohda*, Zahid Maqboolb a Department of Computer Science, University of Kashmir, Srinagar, India Department of Computer Engineering, Islamic University, Awantipora, Kashmir, India

b

Abstract Question classification is an important component of any question answering system as it plays a vital role in the overall accuracy of the QA system, and the key to the accuracy of a question classifier depends on the set of features which we extract. Generally for all text classification problems the data is represented in a vector space model with bag-of-words (BOW) approach. But despite the simplicity of BOW approach it suffers some serious drawbacks like it can’t handle synonymy or polysemy and does not take into account the semantic relatedness between the words. In this paper we propose knowledge based semantic kernel that uses WordNet as its knowledgebase and a semantic relatedness measure SR. We experimented with five machine-learning algorithms viz. Nearest Neighbors (NN), Naive Bayes (NB), Decision Tree (DT), and Support Vector Machines (SVM) to compare the results. For SVM we experimented with linear kernel and the proposed semantic kernel represented by SVMSR.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Ambiguity problem; Semantic Relatedness; SCM; SPE; WordNet

1. Introduction The search engines or the current information retrieval systems available at present return whole document in response to a user query when the user may not be all the time interested in the whole document. This problem got solved with the advances in the research of Question answering technology that identifies and extracts exact answers to the user queries posed in natural languages, from a large collection of text documents. Since the user is interested in exact answers rather than whole document that contains it thus question answering systems are more focused for example for a query like “ Who was the first women prime minister of India” should return “Indira Gandhi” and not documents containing “Women”, “Prime Minister” and “India”. Question Classification is the defining phase of the question answering system firstly because the overall accuracy of the QA system depends heavily on the accuracy of the underlying question classifier. It has been proved by the results obtained from the error analysis of the question answering systems that 36.4% of the errors are due to miss classification of questions. And secondly because of its ability to eliminate candidate answers that are irrelevant to the question thus reducing the search space. There are two common approaches to question classification: (i) Surface pattern identification based question classification. (ii) Semantic categorization based question classification. Surface pattern identification based approach classifies the questions the set of word based patterns and answers are retrieved based on these patterns. Such type of question classification strategy suffers from limited capability to extract answers that are in irrelevant classes. While as semantic categorization based question classifiers use external knowledge base like WordNet to classify the questions taking care of Synonymy and polysemy. While text categorization has been researched better than question classification, which is relatively new field. With there being very small difference that is “what, when, is, of, the” etc. are neglected in text classification while these * Corresponding author. Tel.: +91 8715083633 E-mail address: [email protected].

ISBN: 978-93-82288-63-3

Mudasir and Zahid/COMMUNE – 2015

words are important in question classification. 2. Related Work The major research conferences in the field of question answering have been TREC. The body of Question answering research most related to our work focuses on the automatic question classification. Ravichandran and Hovy [2002] proposed a question classification method that does not rely on external knowledge base but classifies questions on the different sets of surface patterns. Li and Roth [2002] used a diverse feature set consisting of both syntactic features like parts of speech tags and semantic features like named entities to achieve a performance of 84.2%. D. Zehang et al. in 2003 proposed support vector based question classification with a tree kernel used for finding the syntactic information about the question. Krishnan et al. in 2005 contributed the concept of informer a short two to three word phrase present in the question that can be used to accurately classify a particular question. They used a Meta learning model to classify informers and then combine the features of the predicted informer with more general features into a single large feature vector and used linear support vector machines to classify. This approach achieved an accuracy of 86.2%. Haung etal in 2008 derived features from head words of the principal noun phrases in the question (such as WordNet) hypernyms achieving an accuracy of about 89.2%. The machine learning approach used by R C Balabantaray et al. [2013] showed good improvement in the baseline accuracy. 3. Question Classification This component of the question answering system is concerned with mapping the questions into several semantic categories with the intension that this classification, potentially with other constraints on the answer, will be used by a downstream process which selects a correct answer from among several candidates. We have used a two layered question hierarchy which contains 6 coarse grain classes (ABBREVIATION, ENTITY, DESCRIPTION, HUMAN, LOCATION and NUMERIC VALUE) and fifty fine grained categories Table 1 shows the details. Each coarse grained category contains a non-overlapping set of fine grained categories. Table-1 Coarse and fine grained question categories Coarse ABBR DESC ENTY

HUM LOC NUM

Fine abbreviation, expansion Definition, description, manner, reason animal, body, color, creation, currency, disease/medical event, food , instrument, language, letter, other, plant product, religion, sport, substance, symbol, technique Term, vehicle, word. description, group, individual, title city, country, mountain, other, state code, count, date, distance, money, order, other

4. The Ambiguity Problem This is the biggest problem associated to question classification because there is there is no completely clear boundary between classes. Therefore, the classification of a specific question can be quite ambiguous. Consider a below given situation. Who do you buy groceries for? The answer to such a question can be a pet animal or a human family member thus it is hard to categorize a question of this type into one single class and it is likely that there will be miss classification involved in process if we do so. To avoid this problem, we allow our classifiers need to assign multiple class labels for a single question. This method is better than only allowing one label because we can apply all the classes in the later processing steps without any loss. But to simplify the experimentation process, we assume that one question resides in only one category. That is to say, an ambiguous question is labeled with its most probable category. We use machine learning to methods for question classification because it has got advantages over manual classification. The construction of manual methods is a cumbersome task that requires the analysis of a large number of questions. Moreover, mapping questions into fine grain classes requires the use of specific words (syntactic and semantic information) and therefore an explicit representation of the mapping can be very large. On the contrary , in our learning (using machine learning ) approach we can define only a small number of “types” of features, which can then be expanded in a data-driven way to a potentially large number of features (Cumby and Roth, 2000), relying on the ability of the learning process to handle it. It is hard to imagine writing explicitly a classifier that depends on thousands or more features. Finally, a learned classifier is more flexible because it can be trained on a new taxonomy in a small amount of time.

[371]

Mudasir and Zahid/COMMUNE – 2015

5. Dataset and Features We used the UIUC questions data set that is publicly available. This data set contains approximately 5500 training questions from TREC 8, 9 QA track and 500 testing questions from TREC 10 QA track. The dataset has been manually labeled by UIUC according to the coarse and fine grain categories in the Table 1. We take two types of feature simple bag-of-words (BOW) and bag-of -words with semantic relatedness (BOW with SR). Since there are some major short comings of simple bag-of –words approach .Despite its ease of use it cannot handle word Synonymy and polysemy and thus does not take into account semantic relatedness. In this paper we overcome these short comings of the bag-of –words by adding a WordNet based semantic relatedness measure called SR for a pair of words into our semantic kernel. The proposed measure uses the TF-IDF weighing scheme, thus we create a semantic kernel that combines both statistical as well as semantic information from the text. Thus we present a semantic smoothing matrix and kernel for text classification, based on a semantic relatedness measure that takes into account all of the available semantic relations in WordNet by using the semantic relatedness measure SR. From our experimental results we prove that by embedding a semantic relatedness measure through a semantic kernel results in increased accuracy in question classification. 6. Semantic Relatedness Measure Semantic relatedness semantic relatedness using WordNet considers two component measures path length and path depth. The two measures are combined to represent the semantic relatedness (SR) between two terms. The Path length is represented by compactness (SCM) and the path depth is represented by semantic path elaboration (SPE). The semantic relatedness between any pair of terms in the WordNet is given by: Definition 1: Given a word thesaurus W, a weighting scheme for edges that assigns a weight e {0,1} for each edge, a pair of senses S=(s1 ,s2) and a path of length l connecting the two senses, then the semantic compactness of S is given by: SCM(S,W)=∏li =1 ei Where e1, e2,…,el are the path’s edges. If s1= s2 SCM (S, W) =1. If there is no path between s1 and s2 SCM(S, W) =0. Semantic compactness considers the path length and has values in the set [0, 1]. Higher compactness between senses declares higher semantic relatedness and larger weight are assigned to stronger edges. The basic idea behind edge weighting is that some edges provide stronger semantic connections than others. All the paths are weighted and the path with maximum weight is selected. Definition 2: Given a word thesaurus W, and a pair of senses S= (s1, s2) where s1, s2 length l between the two senses, then the semantic path elaboration is given by

W and s1

s2 and a path of

2di.di+1

SCM(S,W)=∏li =1 (𝑑𝑖+𝑑𝑖+1)𝑑𝑚𝑎𝑥 Where di is the depth of sense si in W and dmax is the maximum depth of W. d If s1 = s2, and d1 = d2 = d, SPE(S, W) = 𝑑𝑚𝑎𝑥 If there is no path between s1 and s2 then SPE(S, W)=0. Semantic path elaboration is the second parameter that affects the semantic relatedness between the terms. The standard method of measuring the depth of a sense node in WordNet is by checking for hypernym/hyponym hierarchical relation for the noun and adjective parts of speech and hypernym/troponym for the verb parts of speech.

Definition 3: Given a word thesaurus W, a pair of terms T = (t1, t2), and all pairs of senses S = (s1, s2), where s1, s2 senses of t1, t2 respectively. The semantic relatedness of T is defined as SR (T, S, W) = max {SCM(S, W) ・SPE(S, W)}. SR between two terms ti, tj where ti ≡ tj ≡ t and t ∉ W is defined as 1. If ti ∈ W but tj ∉ W, or ti ∉ W but t∈ W, SR is defined as 0. 7. Semantic Relatedness Based Semantic Kernel Given a question q, we know it is represented in simple bag-of-words representation as:

[372]

Mudasir and Zahid/COMMUNE – 2015

(q) = [tf-idf(t1,q), tf-idf(t2,q),…,tf-idf(tN,q)]T

RN

Where tf-idf (ti, q) is the TF-IDF weight of term ti in a particular question, and N is the total number of terms (e.g. words) in the dictionary (the superscript T denotes the transpose operator). In the above expression, the function (q) represents the question q as a TF-IDF vector. The tf–idf representation implements a down weighting of irrelevant terms as well as highlighting potentially discriminative ones, but nonetheless is still not capable of recognizing when two terms are semantically related. It is therefore not able to establish a connection between two questions that share no terms, even when they address the same topic through the use of synonyms. The only way that this can be achieved is through the introduction of semantic similarity between terms. So in order to do that we have to enrich the bag-of-words representation with some semantic information or in other words we have to transform the tf-idf matrix into a semantic relatedness matrix (let us call this matrix as M) using the semantic relatedness measure SR defined above. The ij th element of M is given by SR (T, S, W) which quantifies the semantic relatedness between terms T (t i, tj). Thus M is an NxN matrix with 1’s in the principle diagonal. Thus this smoothing matrix can be used to transform the questions vectors in such a way that semantically related questions are brought closer together in the transformed (feature) space. Mathematical representation of this semantically enriched bag of words as: ^ (q) = [ (q) TM] T This new feature space can be as such used for many classification purposes but since we want to use this feature space inside a kernel function. The general representation of a kernel function is as: K(X, Z) = (X) T (Z) Where X and Z are vectors. The kernel function simply computes the inner product of these vectors. In our case the questions are represented as vectors so our kernel function will be K (qi, qj) = ^ (qi) T (qj) = (qi) TMMT (qj) Now in order to prove that this is a valid kernel the Gram matrix G formed from the above kernel function should hold the Mercer’s conditions. Mercer’s conditions are satisfied only when the Gram matrix is a positive semi- definite. The Proof that the matrix G formed by the kernel function is indeed a positive semi-definite with the outer matrix product MMT has been proved already. 8. Experimental Results We performed our experiments on the UIUC dataset for question classification. The data is preprocessed using tokenization and TF-IDF matrix construction it is noteworthy to say that we did not use the stop word removal technique as our questions are small documents comprising of only few terms and we assume all the terms are important for correct classification. We employed four machine learning algorithms namely Nearest Neighbors (NN), Naïve Bayes (NB), Decision Tree (DT) and Support Vector Machines (SVM) with bag-of-words approach and compared them with Support vector machine with semantic smoothing o semantic relatedness approach. We performed our experiments using WEKA (for NN, NB, DT, SVM) and SVM Light package (for SVM with semantic smoothing). The results of our experiments are given in Table 2 and Table 3. We used the default parameters for the different machine learning algorithms where the subscripts with SVM algorithm viz Linear and SR mean linear kernel and semantic relatedness kernel respectively. The table shows that by using the semantic kernel for the support vector machines the accuracy of the classification gets increased and in fact is better than SVM with linear kernel. We have tested the SVM on poly kernel and RBF l kernel as well bur we found that there is no significant difference from linear kernel do we decided to include only linear kernel. In summery the experimental results showed that by using a semantic kernel in SVM out performs the other method including SVM with linear kernel. Table 2: Question classification accuracy using different machine learning algorithms under the coarse grain category definition Classifier NN NB DT SVMlinear SVMSR

Accuracy 75.6% 77.4% 84.2% 85.8% 91.9%

Table 3: Question classification accuracy using different machine learning algorithms under the fine grain Category definition Classifier NN NB DT SVMlinear SVMSR

Accuracy 68.4% 58.4% 77.0% 80.2% 86.41%

[373]

Mudasir and Zahid/COMMUNE – 2015

9. Conclusion In this paper we presented a semantic kernel approach to question classification by enriching the BOW with semantic information. We evaluated the impact on question classification using SVM with semantic kernel on the commonly used question classification dataset. We found that by using SVM with semantic kernel there is a significant improvement in the performance. References Budanitsky, A. and Hirst, G., 2006. Evaluating wordnet-based measures of lexical semantic relatedness. Computational Linguistics, 32(1):13–47. Gabrilovich, E. and Markovitch, S., 2007. Computing semantic relatedness using Wikipedia-based explicit semantic analysis. In Proc. of the 20th IJCAI, pages 1606–1611. Hyderabad, India. Hovy, E., Gerber, L., Hermjakob, U., Lin, C. and Ravichandran, D., 1999. Towards Semantics-based Answer Pinpointing. In proceedings of the DARPA Human language Technology Conference (HLT), San Diego, CA. Balabantaray, R. C., Mudasir Mohd and Nibha Sharma, 2012. Multi-class emotion classification: A new approach. International Journal of Applied Information Systems International Journal of Applied Information Systems (IJAIS) – ISSN: 2249-0868 Foundation of Computer Science FCS, New York, USA Volume 4–No.1, September. Sinha, R., and Mihalcea, R., 2007. Unsupervised graphbased word sense disambiguation using measures of word semantic similarity. In Proc. of the IEEE International Conference on Semantic Computing. Li, X., and Roth, D., 2002. Learning Question Classifiers. In Proceedings of the 19th International Conference Computational Linguistics (COLING'02).

[374]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering March 16-18, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India.

Recognition of Offline Handwritten Devanagari Numerals using Statistical Techniques Shraddha Aryaa*, Indu Chhabrab, G.S. Lehalc a

Sri Guru Gobind Singh College, Sector 26, Chandigarh, India, [email protected] b Department of Computer Science and Application, Panjab University, Chandigarh, India c Computer Science Department, Punjabi University, Patiala, Punjab, India

Abstract In this paper we present an Offline Devnagari Handwritten Numeral Recognition System and have experimented with statistical feature extraction techniques such as Zoning, Directional Distance Distribution, Hu Moments, Zernike Moments, Pseudo Zernike moments and Discrete Cosine Transform. The classifiers used are Nearest Neighbor and Support Vector Machine. The Standard Benchmark Handwritten Devnagari Numeral Database provided by ISI, Kolkata is utilized for the purpose. The maximum accuracy achieved is 99.73% using Directional Distance Distribution and 99.33% using Discrete Cosine Transform.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Handwritten Numeral Recognition; Devnagari Characters; Zoning; Directional Distance Distribution; Zernike Moments; Discrete Cosine Transform

1. Introduction Handwriting recognition is a challenging area as infinite variations of a shape exists for even a single handwritten character. Each shape rendered is influenced by the physical and mental state of the writer with respect to the time and the situation. Moreover for offline recognition no real time information is available for support. Devnagari is a two dimensional script as compared to one dimensional Roman script. The second dimension is added as the vowels are written as diacritic marks or modifiers when they occur after the consonant in a word. There are 14 vowels and 33 consonants. Each vowel has its own way of attaching with the consonant and the attachment position can be before or after or above or below or in combination of these. The existence of half characters, new character shape for different combination of half characters, the variation of the position and attachment point of diacritic mark based on character being attached to along with additional special indicator symbols add to the complexity of the script . 2. Related Work Character recognition has been an active research area and numerous feature extraction and recognition techniques are found in literature. A survey on feature extraction methods for character recognition is presented by Trier et al. (1996). Oh and Suen (2002) introduced the Directional Distance Distribution based feature and used the modular neural network classifier for CENPARMI handwritten numeral databases character recognition. The use of moments as invariant binary shape representation was first proposed by Hu (1961). Upneja and Singh (2014) gave performance and error analysis on Orthogonal Rotational Invariant Moments using grayscale images. An OCR system to read printed Bangla and Devnagari scripts was given by Pal and Chaudhuri (1997). Extensive work has been done by Bansal and Sinha (2001a, 2001b) on printed Devnagari character recognition. The pioneer development of standard benchmark Handwritten databases for Devnagari and Bangla scripts is done by Bhattacharya and Chaudhuri (2009). The following work is reported on handwritten devanagari numerals. Ramteke and Mehrotra (2008) applied Moment Invariants, Principle Component Axis, Correlation coefficient and Perturbed moments for feature extraction and Gaussian distribution function classifier obtaining 92.28% accuracy. Hanmandlu and Murthy (2007) used box approach and *

Corresponding author. Tel.: +91 9815 655540. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Arya et al/ COMMUNE – 2015

Bacterial Foraging Fuzzy model giving 96% results. Srivastava and Gharde (2010) used Moment and Affine Moment invariants with SVM classifier giving 99.48% results. The work reported using ISI, Kolkata Devnagari Numeral Standard Database is as follows. Pal et al. (2007) used 64 and 400 dimensional features with MQDF classifier and attained 99.56% accuracy. Bhattacharya and Chaudhuri (2009) applied Wavelet based technique and Chain code histogram using MLP classifier achieving 99.27% accuracy. Singh et al. (2011) performed recursive subdivision of character image with SVM, K-NN and Quadratic classifier reporting 98.49% accuracy. Aggarwal et al. (2012) applied Gradient feature using Sobel operator with SVM classifier achieving accuracy of 99.6%. 3. Techniques Used for Feature Extraction 3.1

Zoning

In this technique, the extracted character image (raw or re-scaled), is divided into windows of equal size. Density values (Number_of_foreground_pixels / Total_number_of_pixels) are computed for each zone, which are used as the feature values. We have used the zone size 5x5 and hence the feature size is 25. 3.2

Directional Distance Distribution (DDD)

Directional Distance Distribution (DDD) is a distance based feature proposed by Oh and Suen (2002). For every pixel in the input binary array, two sets of 8 bytes which are called W (White) set and B (Black) set are allocated. For a white pixel, the set W is used to encode the distances to the nearest black pixels in 8 directions  0 , 45 , 90 , 135 , 180, 225 , 270 , 315  The set B is simply filled with value zero. Likewise, for a black pixel, the set B is used to encode the distances to the nearest white pixels in 8 directions. The set W is filled with value zero. The eight direction codes are 0(E), 1(NE), 2(N), 3(NW), 4(W), 5(SW), 6(S), 7(SE). We divide the image into 5x5 zones and extract 16 DDD features for each zone thus total feature size is 400. 3.3

Hu Moments

Hu derived the following set of seven rotational invariant moment functions which form the suitable shape representation. The feature size is 7. Φ1 = η20 + η02

Φ2 = (η20 _ η02)2+4 η112 Φ3 = (η30 _ 3η12)2 + (3 η21 – η03)2 Φ4 = (η30 + η12)2 + (η21 + η03)2 Φ5 = (η30 _ 3 η12) (η30 + η12)[ (η30 + η12)2 _3(η21 + η03)2 + (3 η21 – η03)( η21 + η03) [3(η30 + η12)2 _ (η21 + η03)2] Φ6 = (η20 _ η02) [(η30 + η12)2 _ (η21 + η03)2 ] + 4 η11 ( η30 + η12) ( η21 + η03) Φ7 = (3η21 _ η03) (η30 + η12) [(η30 + η12)2 _3(η21 + η03)2 ] _ (η30 _3 η12) (η21 + η03) [3(η30 + η12)2 _ (η21 + η03)2] 3.4

(1) (2) (3) (4) (5) (6) (7)

ORIMs- Orthogonal Rotational Invariant Moments

ORIMs are widely used for Image Processing applications. Being orthogonal and complete, they possess the minimum information redundancy. The magnitude of moment is invariant to rotation and reflection and with some geometric transformation they can be made translation and scale invariant. They are robust to image noise, have reconstruction capability with minimum error and high discriminatory power. The ORIMs of order n and repetition m of a continuous signal f(x, y) over a unit disk is defined by

A

nm



n 1



 f (x, y) V

* nm ( x, y ) dx dy

(8)

x2  y2  1

where n is a non-negative integer and m is an integer ( negative or positive) and V*nm (x, y) is the complex conjugate of the Moment Basis function Vnm (x, y). The function Vnm(x, y) has the following invariant form. Vnm( x, y)  Vnm(  , )  Rnm(  ) eim

(9)

where ρ is the length of vector from origin to pixel (x,y) i.e.   x 2  y 2 , θ is the angle between vector ρ and x axis R

( )

in counter clockwise direction i.e. θ = tan-1(y/x), i   1 and nm is the Radial Polynomial. The Basis function Vnm(x, y) is orthogonal to each other. Owing to the property of orthogonality and completeness of the basis function, the image function f(x,y) can be represented as

[376]

Arya et al/ COMMUNE – 2015

f ( x, y) 

 A

(10)

nmVnm ( x, y)

n

m

The Zernike Moment and Pseudo Zernike moment differ only in the Radial function and the constraints imposed on n and m. 3.4.1.

Zernike Moment

Zernike first introduced the set of complex polynomials which form a complete orthogonal set over a unit disk of x2 + y2 ≤ 1 in polar co-ordinates. The Zernike Polynomials are defined by (

(1) s (n  s)! (n  2s ) n | m | n | m |  s)! (  s)! s  0 s !( 2 2



Z Rnm ( ) 

3.4.2.

n |m| ) 2

with constraints (n ≥ 0) and n-|m| is even, |m| ≤ n.

(11)

Pseudo Zernike Moment

The Radial function of Pseudo Zernike polynomials is defined by ( n|m|) P Rnm ( ) 

 s 0

3.5

(1) s (2n  1  s)! (n  s) s !(n  | m | 1  s )! (n | m | s)!

with constraint |m| ≤ n.

(12)

Discrete Cosine Transform(DCT)

Discrete Cosine Transform feature is a technique to convert data of the image into its elementary frequency components. DCT of a 2D array A[M,N] is a matrix say B[M,N] of same size as A whose each element f(u,v) is calculated by the given formula for corresponding element f (m, n) in A where M and N are the width and height of the image respectively. DCT coefficients f(u, v) for f(m, n) are computed by M 1 N 1

f (u, v)  a(u ) a(v)

  f (m, n) cos  2m2M1 u  cos  2n 2N1 v 

(13)

m 0 n 0

where  1   M a(u )    2   M

   a (v )     

, u0

(14)

and;

, 1 u  M 1

1

, v0

N 2 N

(15) , 1 v  N 1

The resultant matrix B[M,N] is actually the matrix with DCT coefficients f(u,v) corresponding to each pixel f(m,n) and the feature size is MxN. DCT cluster high value coefficients in the upper left corner and low value coefficients in the lower right corner of matrix B[ M,N]. These high value coefficients are obtained in zigzag pattern from top left corner of matrix and used as features. 4. Techniques Used for Classification Classification is generally performed by comparing the feature vectors corresponding to the input character with the representative(s) of each character class, using a distance metric. 4.1

Nearest Neighbor Classifier

It compares the input feature vector with a library of reference vectors and the pattern is identified to be of the class of the library feature vector to which it has the closest distance. The Euclidean distance between an input feature vector X N

and a library feature vector C is given by D 

 C  X  i

i

2

where Ci is the ith library feature, Xi is the ith input feature, and N

i 1

is the number of features used for classification. The input character is labelled with the class of the library feature vector giving the minimum Euclidean distance.

4.2

Support Vector Machine

SVM are a set of related supervised learning methods used for classification and regression. It is a discriminative classifier formally defined by a separating hyperplane. Given labelled training data, the algorithm outputs an optimal hyperplane which categorizes the unknown data. The SVM in its elementary form can be used for binary classification.

[377]

Arya et al/COMMUNE – 2015

For multiclass problems like classifying 10 classes of Devnagari numerals as in our case, it is extended using one against the rest approach. We have used LIBSVM 2.8.2 classifier tool as in Chang and Lin (2011) with Linear and Radial Basis Function (Gaussian) kernel for the purpose. These kernels are opted based on their frequent use and good results in literature survey. The polynomial kernel is yet to be experimented with. 5. Experimentation Database used: The Standard Benchmark Handwritten Devnagari Numeral Database (Bhattacharya and Chaudhuri, 2003) provided by CVPR Unit, ISI, Kolkata is used for experimentation purpose. Figure 1 shows few samples of the database and Table 1 gives the database size. The database provides the grayscale tiff images of individual Handwritten Devnagari Numerals. We use Otsu method for the Binarization purpose. No other preprocessing technique is applied. The Normalization to 32x32 size is done for Hu moment Feature computation only and for the other Feature extraction techniques of Zoning, Direction Distance Distribution, Zernike moment, Pseudo Zernike moment and Discrete Cosine Transform the original size of the binarized image is used for feature extraction. Table 1. The size of database

Feature size: To select the optimum feature size for Zernike, Pseudo Zernike and DCT features, a Sample Set was made taking first 200 characters of ISI database training data of each class for training purpose and next 40 characters from ISI database training data only for testing purpose. Zernike and Pseudo Zernike were tested with different orders. The maximum recognition was obtained at order n=10 for Zernike and n=5 for Pseudo Zernike and hence these orders are opted. Also the best reconstruction obtained using Zernike was at order n=12 and Pseudo Zernike was at order n=5. Similarly, various sizes for DCT were tested and size 6x6 gave the best results. Hence we opted for Zernike order n=10, Pseudo Zernike order n=5 and DCT size 6x6. 6. Results and Discussion Table 2 shows the accuracy obtained with individual features. SVM values are computed as per the directions in Practical Guide (Hsu et al. 2003). The recognition accuracy for Zoning (98.27%), DDD (99.73%) and DCT (99.34%) were better than other features used. The SVM classifier with RBF kernel has given better results than Nearest Neighbor overall. Our results 99.73% are better than earlier reported results by Pal et al. (2007) (99.65%), Bhattacharya et al. (2009) (99.27%), Singh et al. (2011) (98.49%) and Aggarwal et al. (2012) (99.6%)

Fig.1.Sample Devnagari Numerals from ISI Database

[378]

Arya et al/COMMUNE – 2015 Table2. Recognition Accuracy achieved for Feature Set used S#

Feature

1-NN

Linear

SVM

gamma

Cost

1

Zoning

97.71%

RBF

2

DDD

97.39%

3

3072

98.27%

99.73%

0.25

256

3

Hu

4

Zernike

56.09%

26.42%

47.94%

32

32768

88.20%

70.42%

92.11%

4

4096

5 6

Pseudo Zernike

84.24%

64.07%

90.06%

6.7272

6888.62

DCT

98.94%

93.73%

99.34%

0.7071

724.077

7. Conclusion In this paper the experiments are conducted with different statistical features and classifiers for the recognition of handwritten Devnagari numerals. The results are comparable with the peer researchers as Standard Benchmark database has been used.

Acknowledgement We are grateful to CVPR unit, ISI, Kolkata for giving us the Standard Benchmark Handwritten Devnagari Numeral Database free of cost for research purpose and saving our time and effort to develop the same.

References Aggarwal, A., Rani, R., & Dhir, R., 2012. Handwritten Devanagari character recognition using Gradient features. International Journal of Advanced Research in Computer Science and Software Engineering, 2(5), p. 85-90. Bansal, V., & Sinha, M. K., 2001, September. A complete OCR for printed Hindi text in Devanagari script. In 2013 12th International Conference on Document Analysis and Recognition (pp. 0800-0800). IEEE Computer Society. Bansal, V., & Sinha, R., 2001, A Devanagari OCR and a brief overview of OCR research for Indian scripts. Proceedings of STRANS01, IIT Kanpur. Bhattacharya, U., & Chaudhuri, B. B., 2005, August. Databases for research on recognition of handwritten characters of Indian scripts. In Document Analysis and Recognition, 2005. Proceedings. Eighth International Conference on (pp. 789-793). IEEE. Bhattacharya, U., & Chaudhuri, B. B., 2009. Handwritten numeral databases of Indian scripts and multistage recognition of mixed numerals. Pattern Analysis and Machine Intelligence, IEEE Transactions on, 31(3), p. 444-457. Chang, C. C., & Lin, C. J., 2011. LIBSVM: a library for support vector machines. ACM Transactions on Intelligent Systems and Technology (TIST), 2(3), p. 27. Chaudhuri, B. B., & Pal, U. 1997, August. An OCR system to read two Indian language scripts: Bangla and Devnagari (Hindi). In Document Analysis and Recognition, 1997. Proceedings of the Fourth International Conference on (Vol. 2, pp. 1011-1015). IEEE. Due Trier, Ø. Jain, A. K., & Taxt, T., 1996. Feature extraction methods for character recognition-a survey. Pattern recognition, 29(4), p. 641-662. Hanmandlu, M., & Murthy, O. V., 2007. Fuzzy model based recognition of handwritten numerals. Pattern Recognition, 40(6), p.1840-1854. Hsu, C. W., Chang, C. C., & Lin, C. J., 2003. A practical guide to support vector classification. Hu, M. K. 1962, Visual pattern recognition by moment invariants. Information Theory, IRE Transactions on, 8(2), p. 179-187. Oh, I. S., & Suen, C. Y. 2002. A class-modular feedforward neural network for handwriting recognition. Pattern Recognition, 35(1), p.229-244. Pal, U., Sharma, N., Wakabayashi, T., & Kimura, F., 2007, September. Handwritten numeral recognition of six popular Indian scripts. In Document Analysis and Recognition, 2007. ICDAR 2007. Ninth International Conference on (Vol. 2, pp. 749-753). IEEE. Ramteke, R. J., & Mehrotra, S. C., 2008. Recognition of Handwritten Devnagari Numerals. International journal of Computer processing of Oriental languages. Shrivastava, S. K., & Gharde, S. S., 2010. Support vector machine for handwritten devanagari numeral recognition. International Journal of Computer Applications, 7(11), p. 9-14. Singh, C., & Upneja, R., 2014. Error analysis in the computation of orthogonal rotation invariant moments. Journal of mathematical imaging and vision, 49(1), p.251-271. Singh, M. J. K., Dhir, R., & Rani, R., 2011. Performance Comparison of Devanagari Handwritten Numerals Recognition. International Journel of Computer Application (0975-8887) volume-22 No.-1.

[379]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Sentiment Analysis of Views Written in Gurmukhi Script Deepalia*, Vishal Goyalb, Ajit Kumarc a

AP,CA Department, GZS-PTU Campus, Bathinda Punjab, India b AP,DCS Department, Punjabi University, Patiala, India c AP,DCS Department, Multani Mal Modi College, Patial, India

Abstract With the flourishing of internet, digital media like web pages, websites, social media, blogs has fuelled the growing market in personal opinions like reviews, ratings and other forms of online expressions. Digital review mining and summarization has become a blister research area. Reviews can be positive and negative based on sentiment orientation of the opinion it contains. In this paper, we focus on various digital media where sentiment analysis is flourishing.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar . Keywords: Sentiment Analysis; Punjabi Language; Naïve Bayes; N-Gram;Online Reviews

1. Introduction In this technical era, internet provides immense amount of information. Social media plays major role in the conveying information. These sites offer valuable information imminent into the sentiment analysis of particular product or a movie. Most social groups, e-commerce websites, blogs identify these reviews and ratings as an important part of decision making, these reviews provide a baseline of information that indicates levels and supports business intelligence. Traditionally, manufacturers conduct consumer feedback and surveys manually for this purpose. So an emerging field known as sentiment analysis is taking shape around digital world. Sentiment analysis (opinion mining) is an application of natural language processing (NLP) and data miming task that aims to obtain feelings of the writer expressed in positive or negative. Sentiment analysis is a form of text classification that classifies texts based on the sentimental orientation (SO) of opinions. Reviews are classified as positive and negative based on the SO of opinions. Opinion miming is performed at two different levels (Hu and Liu, 2004; Leung et al.) i.e. Document level and Sentence level. 1. At Document level it determines the polarity of whole document ਪ ਿੰ ਡ ਦੇ ਮਜਾਪਿਆ ਨੌਜਵਾਨ ਦਾ ਰੋਲ ਅਿਸ਼ੈ ਿੁਮਾਰ ਨੇ ਬਹੁਤ ਵਧੀਆ ਢਿੰ ਗ ਨਾਲ ਪਨਭਾਇਆ| ਇਸਦੇ ਇਲਾਵਾ ਿੈਟਰੀਨਾ ਿੈਫ ਨੇ ਵੀ ਆ ਣੇ ਅਪਭਨੈ ਨਾਲ ਪਿਸੇ ਤਰਹਾਂ ਦੀ ਬੇਈਮਾਨੀ ਨਹੀਂ ਿੀਤੀ| ਇਸਦੇ ਇਲਾਵਾ ਜੇਿਰ ਪਫਲਮ ਦਾ ਪਮਊਪਿਿ ਵੇਪਿਆ ਜਾਵੇ ਤਾਂ ਉਹ ਵੀ ਗਿਬ ਦਾ ਹੈ. ਹਰ ਗੀਤ ਲੋ ਿਾਂ ਦੇ ਪਦਲਾਂ ਨਿੰ ਟੁਿੰ ਬਦਾ ਹੈ| ਿੁੁੱ ਲ ਪਮਲਾਿੇ ਪਫਲਮ ਆ ਣੇ ਨਾਮ ਅਤੇ ਦਰਸ਼ਿਾਂ ਦੀ ਉਮੀਦ ਉੱਤੇ ਖ਼ਰੀ ਉਤਰਦੀ ਹੈ| Piḍa dē majāki'ā naujavāna dā rōla akaśai kumāra nē bahuta vadhī'ā ḍhaga nāla nibhā'i'ā| isadē ilāvā kaiṭarīnā kaipha nē vī āpaṇē abhinai nāla kisē tar'hāṁ dī bē'īmānī nahīṁ kītī| isadē ilāvā jēkara philama dā mi'ūzika vēkhi'ā jāvē tāṁ uha vī gazaba dā hai. Hara gīta lōkāṁ dē dilāṁ nū ṭubadā hai| kula milākē philama āpaṇē nāma atē daraśakāṁ dī umīda utē ḵẖarī utaradī hai| This document has positive opinion.

*

Corresponding author. Tel.: +91 7837 242200. E-mail address: [email protected] ISBN: 978-93-82288-63-3

Deepali et al/COMMUNE – 2015

2.

At Sentence level it determines polarity of sentences.

Table1. Sentence Level Polarity Punjabi sentence ਸੰ ਸੰ ਗ ਇਸ ਗੂੜ ਕੋਮ੍ਪਾਨੀ| ś'āmśung iś good company| ਕੈਮਰਾ ਚੰ ਗਾ ਨਹੀ ਹੈ| camera change nahi hai |

Reviews Positive Review

Negative Review

Punjabi is the most widely spoken language, mostly popular in Punjab (INDIA) and Pakistan. Till today most research work on sentiment analysis is carried out in English, Hindi, Bengali Amitava Das and Sivaji Bandyopadhyay, AFNLP-2010 language but little work is done in Punjabi language because there are limited resources available in Punjabi. The rest of the paper is organized as follows: Section 2 describes the related work performed in sentiment analysis. Section 3 explains the system description .Experimental results are presented in Section 4. The last we concludes the study and future scope. 2. Related Work Most prominent work on has been done by Amitava Das and Bandopadhya,they developed Sentiwordnet for Bengali language. They created the subjectivity detection system which was evaluated on Multi Perspective Question Answering (MPQA) corpus and on Bengali corpus,Amitava Das and Sivaji Bandyopadhyay, IEEE-09. Sentiment analysis algorithms were designed for binary classification by (Turney, 2002;Pang et al 2002:dave et al 2002).Some recently proposed algorithms extend binary sentiment classification to classify reviews with respect to multi-point scales(Pang and Lee, 2005). Sentiment analysis is closely related to subjectivity analysis (Wiebe et al, 2001; Esuli and Sebastiani, 2005).Subjectivity analysis specifies whether the given text is subjective or objective in nature. The first method considers the subjectivity analysis a binary classification problem. Pang and Lee (2005) adopted this method to identify the subjectivity of sentences on domain (movie reviews). The second method is part-of speech tagging (POS) (Turney, 2002; Hu and Liu, 2004). With the emergence in E-commerce, online shopping is booming, so the summarization of customer’s review about the product was carried out by Minqing Hu and Bing Lu, KDD’04. This task was performed in three steps. First, the Product Features are mined. Secondly, opinions are identified as positive or negative. Social media (twitter) is widely used platform to express our views during election campaign (Khurshid Ahmad,Nicholas daly;2011) to monitor political sentiments and based upon public opinions election results are predicted(Adam Bermingham and Alan F. Smeaton,2011). N-Gram approach along with machine learning techniques was used to determine the polarity of words(Vasudeva Verma) and 82.9% accuracy was obtained using SVM approach (Pang et al.), ,78.32% on movie dataset and 70.06% on multi-category dataset. Score = x * Count _Trigram + y * Count _Bigram + z * Count _Unigram Count N-Gram = Number of N-Grams matched (N=Uni/Bi/Tri). 3. System Description The Sentiment classification techniques are applied on different data sets such as reviews, News articles, Web Discourse, blogs. A review includes movie reviews, product reviews. Web discourse includes social media (Facebook, Twitter), News groups etc. Different approaches are used to extract information are Naive Bayes, Support Vector Machine (SVM), Rule Based Classifier and Hybrid Classifier. This paper describes Naïve Bayes approach. Naïve Bayes Classifier: Naïve Bayes Classifier: Naïve Bayes is an effective machine learning technique based on Bayes Rule. It is simple probabilistic classifier which performs well on problems related to linearly and non-linearly separable.. Bayes rule: 𝑃(𝑐|𝑑) =

𝑃(𝑐)𝑃(𝑑|𝑐) 𝑃(𝑑)

(1)

Where P(d) plays no role in selecting c*. To estimate the term P (d | c), Naive Bayes decomposes it by assuming the fi's are conditionally independent given d's class:

[381]

Deepali et al/COMMUNE-2015

𝑃𝑁𝐵 (𝑐|𝑑) ≔

𝑚 𝑃(𝑐)(𝜋𝑖−1 𝑃(𝑓𝑖 |𝑐)𝑛 𝑖(𝑑)) 𝑃(𝑑)

(2)

The system works on two phases i.e. training phase and testing phase. In the training phase the system is trained to analyse the data. There are no resources available for Punjabi language. The corpus (database) is collected from various Punjabi websites. In the Testing phase the data collected is first normalized (punctuation marks, spaces etc are removed), followed by tokenization (each paragraph is tokenized using delimiter “the dandi” (|), further we split the sentence into word using the delimiter space (“ “). After that the probability score is calculated (positive and negative) and words like country name, pronoun etc. are ignored from probability calculations.

Fig1. Testing Phase

To train the system we find the frequency of words in both the polarity data using following equation:

PFi  WordiPositive _ CorpusPFi  1

(3)

NFi  WordiNegative_ Corpus NFi  1

(4)

After finding the frequency of the words in both the corpus we need to normalize the value between 0 and 1. Because there were huge variation in the frequency of the word and if we use direct frequency value in finding the polarity of the words it gave us wrong results. Some of the word frequency is either low or high. To normalized the polarity frequency score we used following equations: POS _ SCOREi 

NEG _ SCOREi 

PFi PFi  NFi

(5)

NFi PFi  NFi

(6)

We find the both negative as well as positive polarity score using the formula

n

POL _ POS   POS _ SCOREi

(7)

i 0

[382]

Deepali et al/COMMUNE – 2015

n

POL _ NEG   NEG _ SCOREi

(8)

i 0

Where POL_POS is the positive polarity score POL_NEG is the negative polarity score n is the total number of words POS_SCOREi is the positive probability score of the word currently processed NEG_SOCREi is the negative probability cscore of the word currently processed. 4. Experiments and Results Experiment is conducted on domain movie reviews collected from different Punjabi websites (www.jagbani.com, www.punjabitribune.com, www.ajitjalandhar.com). Three evaluation measures are used on the basis of which system performance is computed, these are   

Precision Recall F-Measure

Table 2:

Machine says yes tp fp

Human says yes Human says no 𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛(𝑃) =

𝑅𝑒𝑐𝑎𝑙𝑙(𝑅) =

𝐹=

Machine says no fn tn

𝑡(𝑝) 𝑡(𝑝) + 𝑓(𝑝)

(9)

𝑡(𝑝) 𝑡(𝑝) + 𝑓(𝑛)

(10)

2 ∗ 𝑅𝑒𝑐𝑎𝑙𝑙 ∗ 𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 𝑅𝑒𝑐𝑎𝑙𝑙 + 𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛

(11)

Following results are calculated: N-Gram Unigram Bigram Trigram (Unigram+ Bigram+ Trigram)

Recall (%) 91 97 73 36

Precision (%) 68 53 58 59

F-measure (%) 77 68 64 45

Recall is the number of correct answers given by the system, so we conclude that our system has 97% recall which is highest in case of Bigram. Precision is number of actually correct answers, i.e. Unigram is best in this case with 68% precision-measure is the accuracy of the system, which is highest in case of Unigram. The size of training and testing datasets experimented is 80thousand reviews. So, from the results we conclude that due to less size of training dataset (corpus) the combination of (Unigram+Bigram+Trigram) and Trigram provides less accuracy. 5. Conclusion and Future Work People have shortage of time and to filter out specific polarity of text is very time consuming. Sentiment analysis tool processes a set of reviews based on a given domain, item or product and then provides output as the overall polarity. However the process of determining appropriate polarity of any paragraph (review) is full of uncertainty about subjectivity, imprecision in rating etc. Therefore building such system is a very complicated task. Finally we can conclude that this work is an attempt to explore new area of opinion mining in Punjabi language. Our work can be considered as base of development of Punjabi or any other Indian Language sentiment analyzer. In future we are experimenting on different approaches in which we take a list of positive and negative corpus to improve the accuracy of the system. [383]

Deepali et al/COMMUNE-2015

References Adam Bermingham and Alan F. Smeaton(2011)” On Using Twitter to Monitor Political Sentiment and Predict Election Results” IJCNLP,Chiang Mai, Thailand, pp. 2–10. AkshatBakliwal,AnkitPatil,PiyushArora,VasudevaVarma,(2011)“Towards Enhanced Opinion Classification using NLP Techniques”, Proceedings of the Workshop on Sentiment Analysis where AI meets Psychology (SAAIP),Chiang Mai, Thailand, IJCNLP,pp.101-107. Amitava Das and SivajiBandyopadhyay (2010).“SentiWordNet for Indian Languages”, AFNLP, pp. 56-63. Amitava Das and SivajiBandyopadhyay.(2010).“Opinion-Polarity Identification in Bengali”, ICCPOL, pp.169-182. Andrea Esuli and Fabrizio Sebastiani,(2006)” SentiWordNet: A High-Coverage Lexical Resource for Opinion Mining” Kluwer Academic Publishers,Netherlands,pp.1-26. Bo Pang, Lillian Lee, and ShivakumarVaithyanathan(2002)Thumbs up? Sentiment classification using machine learning techniques. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), pp 79–86. Faraaz Ahmed, Barath Ashok, SaswatiMukherjee,MeenakshiSundaram, Murugeshan, Ajay Sampath(2008).“Effect of Modifiers for Sentiment Classification of Reviews”, ICON. Hemalatha,G.P.S.Varma(2012)” Preprocessing the Informal Text for efficient Sentiment Analysis”IJETTCS,vol 1,/Issue 2,pp. 58-61. Janyce Wiebe, Theresa Wilson(2005)” Annotating Expressions of Opinions and Emotions in Language” Kluwer Academic Publishers,Netherlands,pp. 1-50. Kushal Dave, Steve Lawrence, and David M. Pennock(2003)”Mining the peanut gallery: Opinion extraction and semantic classification of product reviews”,pp. 519–528. Khurshid Ahmad, Nicholas Daly(2011)” What is new? News media, General Elections, Sentiment, and named entities”, IJCNLP 2011,Chiang Mai, Thailand, pp.80–88 . Lun-Wei Ku, Yu Thing and Liang Hsin-His Chen. (2006). “Opinion Extraction, Summarization and Tracking”, AAAI, pp. 100-107. Minqing Hu and Bing Lu (2004), “Mining and Summarizing Customer Reviews”. KDD,ACM New York, pp.168-177. Minqing Hu and Bing Liu, (2005) “opinion Extraction and summarization on web”, AAAI,pp. 1621-1624. Peter Turney(2002)”Thumbs up or thumbs down? Semantic orientation applied to unsupervised classification of reviews”pp. 417–424. Proceedings of the Workshop on Sentiment Analysis where AI meets Psychology (SAAIP), IJCNLP 2011,Chiang Mai, Thailand, pp.1. RudyPrabowo and Mike Thelwall (2009)”Sentiment Analysis: A Combined Approach”,ScienceDirect pp.143-157. Richa Sharma,Shweta nigam,rekha Jain(2014)”Polarity detection of Movie Reviews in Hindi Language”IJCSA, Vol.4,pp.49-57. Shailendra Kumar Singh,Sachita Paul,Dhananjay Kumar(2014)”Sentiment Analysis Approaches on different Dataset Domains:Survey”IJDTA,vol 7,pp.39-50. Vishal Goyal, AnkurRana ,Vimal K. Soni,(2011) “Renaissance of Opinion Mining”,Springer,pp. 60-67. www.jagbani.in accessed on November 03, 2011. www.punjabitribuneonline.com accessed on November 05, 2011. www.navapanga.com accessed on November 05, 2011. www.ajitjalandhar.com accessed on November 07, 2011. N-gram, http://en.wikipedia.org/wiki/N-gram accessed on December 30, 2011. WordNet, http://wordnet.princeton.edu/. accessed on November 12, 2011. www.24dunia.com accessed on November 05, 2011. Naïve Bayes Classifier, http://en.wikipedia.org/wiki/Naive_Bayes_classifier/accessed on February 15, 2012. Precision and Recall, http://en.wikipedia.org/wiki/Precision_and_recall accessed on February 28, 2012. Support Vector Machine" Internet Source,www.statsoft.com/textbook/support-vector-machines/ accessed on March 15, 2012.

[384]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Rectangular Patch Antenna using Metamaterial for Multi Band Applications Sunita*, Gaurav Bharadwaj, M. M. Sharma Govt. Women Engineering College, Ajmer, India

Abstract Design of multiband application antenna using metamaterial for enhancement in the return loss. This design is operated at multi band frequency as comparison to RMPA alone. An antenna whose substrate is of FR4 of 80mm × 80mm and patch of 32.908mm × 25.43mm is taken as reference and by adding, another substrate on which patch is formed by which metamaterial is created. This metamaterial enhance the return loss of the antenna and the structure become multiband. For verifying the negative value of permittivity and permeability, author has used the MS-Excel. The structure is simulated in CST MICROWAVE Studio.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Enhanced Return Loss, Negative Refractive Index, Omega Structure, Multiband, Metamaterial.Introduction

1. Introduction Recently, the microstrip patch antenna have gained much attention because of their many advantages including ease of installation, mechanical reliability with respect to radiation property, they are versatile in polarization and resonant frequency. Many researchers have done research to improve these drawbacks. In this field of research, the theoretical concept of metamaterials was introduced by (Victor Veselago,1968), (Engheta and Ziolkowski, 2006). According to the theory of Vesalago, these materials are generally providing properties which are not found in readily available materials in nature (J.B. Pendry, 2000, Bimal Garg, et al 2011) They are also called as the Negative Refractive Index materials or Left handed materials LHM because they have Negative Refractive Index (NRI) , (Smith, 2000 et al, 2001) he first structure of LHM is made by Smith (Bimal Garg, et al, 2013) Metamaterial is a material which has negative permeability and permittivity, so called as double negative metamaterial. In this paper, microstrip patch antenna using metamaterial for multi band operation has been proposed. The proposed antenna is designed on FR4 substrate. The metamaterial structure is composed of three omega ring connected each other with strip lines and split ring resonator present in each omega ring. This metamaterial structure improves the efficiency of RMPA alone. It also increases the return loss and improves the structure of RMPA alone. 2. Antenna and Metamaterial Structure A rectangular shaped antenna whose substrate is of FR4 of 80mm × 80mm with dielectric constant 4.3 and loss tangent 0.02 and patch of 32.908mm × 25.43mm with a slot of 8mm × 4mm, is taken as reference antenna as shown in Fig.1 (a). This structure resonate at 2.673 GHz with return loss of -11.318 dB as shown in Fig.1 (b). As shown in Fig.1 (b), the return loss is poor and bandwidth is very small. By adding another substrate with patch on it with the reference antenna, metamaterial is created. This metamaterial enhance the return loss of the antenna and the structure become multiband. In this metamaterial design, three omega structure connected with each other through strip-line and each omega structure consist of split ring resonator. This design gives the better improvement in return loss. This proposed antenna has been used for multi band applications. It cannot be proved on CST MICROWAVE STUDIO, that the structure is metamaterial. So the negative permittivity * Corresponding author. Tel.: +91 95303 66515. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Sunita et al/ COMMUNE-2015

(R. A. Shelby et al, 2001) and permeability of the metamaterial is shown in MS-Excel, as shown in Fig.3.(a) and Fig.3.(b). To calculate the permittivity and permeability of metamaterial structure, given below relation is used respectively by equation (1) and equation (2).

(a)

(b)

Fig.1.(a) Dimensional view of reference antenna;

(b) S-parameter of reference antenna.

(a)

(b)

Fig.2.(a) Meta covered between two waveguide port;

µ𝑟 = 𝜀𝑟 =

(b) Dimensional view of metamaterial (All dimensions are in mm)

2.𝑐.(1−𝑉1)

(1)

𝑤.𝑑.𝑖(1+𝑉2) 2.𝑠11.𝑐.𝑖 𝑤.𝑑

+ µ𝑟

(2)

Where µr= relative permittivity of metamaterial, εr= relative permeability of metamaterial, c = Speed of Light/ vacuum, ω = Frequency in Radian, d = Thickness of the Substrate, V1= Voltage Maxima, V2= Voltage Minima As shown in Fig. 3(a) and (b) these values are negative. So this structure follows the metamaterial characteristics. 3. Result and Discussion After the simulation of antenna and metamaterial structure using simulator CST microwave studio and using dielectric substrate FR-4 by conventional microstrip integrated circuit technology, respectively. We have obtained the different results of antenna. The results of antenna are simulated with and without proposed metamaterial structure. The comparison of different characteristics, with and without metamaterial, is also shown in this section. S-parameter with metamaterial is shown in Fig.4. S-parameter with and without metamaterial is shown in Fig. 6. It shows that return loss is enhanced on the resonant frequency of the reference antenna and it become multiband structure also.

[386]

Sharma et al/COMMUNE – 2015

(a)

(b)

Fig.3.(a) Graph between permittivity and frequency; (b) Graph between permeability and frequency.

Fig.4. S-parameter of structure with metamaterial (simulated).

Fig.5. Antenaa structure with metamaterial.

Fig.6. S-parameter of antenna with and without metamaterial.

4. Conclusion A rectangular shaped antenna is turned into metamaterial structure to enhance the return loss and bandwidth of the structure. The merit of the structure is that its multiband characteristic can be used in deep space exploration. In future return loss, directivity, and bandwidth can be further enhanced and single band can used in application such as Bluetooth. References J.B. Pendry, 2000, Negative refraction males a prefect lens, Phys Rev Lett, 85, pp.3966–3969. Bimal Garg, Rahul Tiwari, Ashish Kumar and Tilak Chitransh, 2011 , Design of factored ‘X’ shaped metamaterial structure for enhancement of patch antenna gain”, International Conference on Communication Systems and Network Technologies. D. R. Smith, W. J. Padilla, D. C. Vier, S. C. Nemat-Nasser, and S. Schultz. 2000. , Composite medium with simultaneously negative permeability and permittivity, Phys. Rev. Lett., vol. 84, no. 18, pp. 4184–4187. R. A. Shelby, D. R. Smith, and S. Schultz. 2001 , Experimental verification of a negative index of refraction,” Science, vol. 292, pp. 77–79, April. R. W. Ziolkowski and E. Heyman. 2001., Wave propagation in media having negative permittivity and permeability” Phys. Rev. E, vol. 64, pp. 056625:1–15, Bimal Garg, Anupam Das, Pankaj Chand, Bitty Singh 2013, Quadruplet Circular Hollow Ring Structure to Ameliorate RMPA parameters for WLAN” Journal of Network Security Volume 1, Issue 1, ,pp.1-5. Silvio Hrabar, Juraj Bartolic, 2003, Backward Wave Propagation in Waveguide Filled with Negative Permeability Meta Material”, Antennas and Propagation Society International Symposium, vol. 1 pp. 110 – 113. Silvio Hrabar, Gordan Jankovic, Berislav Zivkovic, Zvonimir Sipus, 2005, “Numerical and Experimental Investigation of Field Distribution in Waveguide Filled with Anisotropic Single Negative Metamaterial”, Applied electromagnetics and communications (ICEcom), pp. 1- 4.

[387]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Identification of Clause Boundary in Punjabi Language Sanjeev Kumar Sharmaa*, Gurpreet Singh Lehalb a

DAV University, Jalandhar, India Punjabi University, Patiala, India

b

Abstract With the growing research in the field of Natural language processing, need to process the larger complex sentence has become an essential part in most of the natural language processing applications. Some of these applications include machine translation, information extraction, grammar checking and text to speech system etc. large complex sentences can be processed by splitting them in their smaller unit i.e. clauses. In our research, we have proposed a morphological based technique for identification of clause boundary i.e. to mark starting and end position of a clause in a sentence. We have tested our system on simple, compound and complex sentences and the results have been manually evaluated.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Clauses; Dependent; Independent; Clause Boundary

1. Introduction The performance of natural language processing application decreases as the complexity of sentences increases. A complex sentence cannot be processed until its clauses are not properly identified (Leffa, 1998). Clause boundary identification is one of the shallow semantic parsing tasks consisting of marking the starting and end position of a clause in the sentence. Larger sentences (compound and complex) are composed of more than one clause. The clauses present in a sentence can be of same type (independent clauses) or of different types (dependent and independent). The identification of clause boundaryis important in various NLP applications such as Machine Translation, parallel corpora alignment, Information extraction and speech applications. Grammatically a clause is a group of words having a subject and a predicate of its own, but forming part of a sentence.Clause identification is a special kind of shallow parsing, like text chunking nevertheless, it is more difficult than text chunking, since clauses can have embedded clauses. Clause boundary identification of natural language sentences poses considerable difficulties due to the ambiguous nature of natural languages. 2. Introduction to Punjabi Language Punjabi language belongs to the Indo-Aryan family of languages. It is 11th most spoken language in India. Besides an official language of Punjab state Punjabi is also spoken Haryana and Himachal Pradesh.It is mostly spoken by the peoples living in Punjab region of India and Pakistan. It is the 10 th most spoken language in the world. Other than India, it is spoken in USA, England, Canada, and Pakistan. It is spoken by nearly 102 million people in the world. 3. Related Work Work on clause identification has been done on a few Indian languages and some foreign language. In Indian language a clause boundary, identification system has been proposed by D. Parveen et al. (2009) for Urdu language. They used Conditional Random Field as the classification method and the clause markers. They used the word related feature like current word, next word, previous word and Part-Of-Speech (POS) tag related features like POS tag of *

Corresponding author. Tel.: +91 94650-05780 E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Sharma and Lehal/ COMMUNE-2015

current, next and previous words. Nine types of clause marker for marking nine different types of subordinate clauses were employed. Vinh Van Nguyen et al. (2007) have proposed Conditional Random Fields (CRFs) framework for clause splitting problem in English language. Abney (1990) used a clause filter as a part of his CASS parser. It consists of two parts: first part recognizes the basic clauses and the second part repair difficult cases like clauses without subjects and clauses with additional VPs. Ejerhed (1996) showed that how a parser could benefit from automatically identified clause boundaries in discourse. Papageorgiou (1997) used a set of handcrafted rules for identification of clause boundaries in the text. Leffa (1998) proposed a set of clause identification rules. Orasan (2000) used a memorybased learner with post-processing rules for predicting clause boundaries in Susanne corpus. Georgiana Puscasu (2004) proposed a multilingual method for detecting clause boundaries in unrestricted texts. ErifK.Tjong et al. proposed a memory based learner to CONLL-2001 shared task. Erik F. Tjong et al. (2001) proposed a machine learning system for identification of clauses. EraldoFrenandes et al. (2009) have proposed Entropy guided transformation learning (ETL) method for clause identification. Ani Thomas et al. (2011) used dependency relationship for identification and separation of clauses in a sentence. NaushadUzZaman et al. (2011) proposed a rule based system for the simplification of the sentences. Xavier Carreras et al. (2001) used Ada Boost Learning Algorithm to solve the simple decision for clause splitting problem. Aniruddha Ghosh et al. (2010) proposed a hybrid approach i.e. rule based and statistics based approach for clause identification and separation system in Bengali language. 4. Clause Boundary Identification Problem A clause is the largest unit of the sentence that contains a predicate and explicit or implied subject and the problem is to split a sentence in to clauses by identifying the starting and end position of the clause. The problem of clause boundary identification is not only detecting a non-recursive phrase of the sentence rather it is a three-step process: identifying clause starts, identifying clause ends, and finding complete clauses (Sang and Dejean, 2001).Consider the following example: ਮੀਂਹਪੈਰਿਹਾਸੀ,ਤੇਲੋਕਰ ਿੱ ਜਿਹੇਸਨ। ( It was raining and people were on spree ) In above sentence there are two clauses; one is ਮੀਂਹਪੈਰਿਹਾਸੀ(It was raining)and second is ਲੋ ਕਰ ਿੱ ਜਿਹੇਸਨ (people

were on spree). Both these clauses are separated by comma followed by conjunction. The system should be able to identify both the clauses as shown below: ਮੀਂਹਪੈਰਿਹਾਸੀ,ਤੇ⏟ ⏟ ਲੋਕਰ ਿੱ ਜਿਹੇਸਨ 𝑐𝑙𝑎𝑢𝑠𝑒 1

𝑐𝑙𝑎𝑢𝑠𝑒 2

5. Introduction to Punjabi Clauses In Punjabi language, there are two types of clauses. One is dependent clause and second is independent clause. Independent clauses are simple sentences that contains subject and predicate. On the other hand dependent clauses can be further classified in to two categories: one is predicate bound clause and the other is non-predicate bound clauses. The further classification of these dependent clauses has been given in fig 1.

Fig. 1. Classification of dependent clauses in Punjabi language

[389]

Sharma and Lehal/ COMMUNE-2015

6. Methodology Used We have used the syntactic cue of the clauses for their identification. The syntactic cues used are suffix information of non-finite verb, presence of conjunction or comma and even part of speech tag at some places. We have performed our research work mainly for identification of clause boundaries in complex sentences, as they contains different types of clauses and in some cases these dependent clauses are embedded in independent clause of the sentence. Different syntactic cues have been used for different type of dependent clauses. 6.1. Predicate Bound Clause Predicate Bound clauses are those in which non-finite verb phrases bound the dependent clause (predicate) with independent clause e.g. Sentence 1:ਰੱ ਸੀਟੱ ਪਦਿਆਂਮੇਰੇਪੈਰਨੂੰ ਮੋਚਆਗਈ। (I sprained my foot while skipping the rope) In above example ਰੱ ਸੀਟੱ ਪਦਿਆਂ‘skipping the rope‘ is the dependent clause (predicate) and the word ਟੱ ਪਦਿਆਂ‘skipping’ is non-finite verbal phrase and this non-finite verbal phrase bound this clause with in-dependent clause i.e. ਮੇਰੇਪੈਰਨੂੰ ਮੋਚਆਗਈ ‘I sprained my foot’. As Clause in Punjabi language follow SOV (Subject-Object-Verb) order, so the non-finite verb will indicate the end of dependent clause. Non-finite verbs can be identified by presence of one of the suffix i.e. (ਰਿਆਂ, IAA, ਕੇ, NE, NON, N).In above examples sentence 1 contains ਟਿੱ ਪਰਿਆਂ‘skipping’ as nonfinite verb with suffix ਰਿਆਂ. In predicate bound clauses subordinate verbal phrase of predicate bound types of sentences are positioned just before the predicate clause of the sentence. Therefore, the starting point of the dependent clauses in participial type of sentence will be the start of the sentence and the end point will be the non-finite verb of the subordinate verbal phrase or dependent clause. The clause boundary mark of dependent and independent clauses in above sentence can be represented in the following way: ਿਿੱ ਸੀਟਿੱ ਪਰਿਆਂ ਮੇ ⏟ ⏟ਿੇਪੈਿਨੂੰ ਮੋਚਆਗਈ

𝐷𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑐𝑙𝑎𝑢𝑠𝑒 𝐼𝑛𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑐𝑙𝑎𝑢𝑠𝑒

6.2. Non Predicate Bound Type of Clauses: Non-predicate bound sentences are those in which the bounding particle is not the predicate rather it may be some other element of the sentence like subject, adjective etc. For example: ਧੂੰ ਿਤਾਂਸਿਜਨੂੰ ਇੂੰ ਜਨਿੱਪੀਬੈਠੀਸੀਰਜਵੇਂਕੋਈਤਕੜਾ ਲਵਾਨਮਾੜੇ ਲਵਾਨਿੀਰਗਚੀਤੇਗੋਡਾਿਿੱ ਖੀਬੈਠਾਹੋਵੇ । (The fog has subdued the sun like a strong wrestler subdues a weak one with his knee on his throat) In above example the predicate of dependent clause ਧੂੰ ਿਤਾਂਸਿਜਨੂੰ ਇੂੰ ਜਨਿੱਪੀਬੈਠੀਸੀ‘The fog has subdued the sun

‘does not bound it with the independent clause ਕੋਈਤਕੜਾ ਲਵਾਨਮਾੜੇ ਲਵਾਨਿੀਰਗਚੀਤੇਗੋਡਾਿਿੱ ਖੀਬੈਠਾਹੋਵੇ ‘a strong

wrestler subdues a weak one with his knee on his throat’’, rather it is the conjunction ਰਜਵੇਂ which bound the dependent clause with independent clause. This type of clauses are further of two types: 6.1.1 Sequential This type of dependent clauses contains ਰਕ as a conjunction for separating dependent and independent clauses. So

the clause boundary can be easily identified by scanning the sentence from left to right. The starting point of the dependent clause is the start of the sentence and the end point of the dependent clause will be ਰਕ conjunction. The word next to ਰਕ will be the starting point of independent clause and the end of sentence will be the end of independent clause. 6.1.2 Non-Sequential Clauses Some other conjunction except ਰਕ is used. E.g. ਜਿੋਂ, ਜੇ, ਾਵੇਂ, ਰਕਉਂਰਕ etc. these are further classified in following two types:

[390]

Sharma and Lehal/ COMMUNE-2015

6.1.3 Restrictive This clause provides the crucial information about the preceding subject and limits the meaning of the subject. No comma is used to separate the clauses.Conjunctions starting with ਜ character (ਰਜੂੰ ਨਹਾਂ,ਜੋ etc.) are used to provide relevant information of the clause.The position of this clause in the sentence is in between the subject and the predicate of the independent clause.So there will be three pairs of clause boundary markers in sentence:first pair marks the start and end of the subject of independent clause, second pair marks the start and end of restrictive clause and the third pair will mark the start and end of the predicate of independent clause. The first pair can be easily identified as the subject starts from the beginning of the sentence and end just before the restrictive clause and the restrictive clause starts with a conjunction. So conjunction serves the purpose of the end of subject and beginning of the restrictive clause. Now it is a challenge to identify the end of restrictive clause and beginning of the predicate of independent clause as no comma or conjunction is used in between the restrictive clause and predicate of independent clause. Consider the following example: ਉਨਹਾਂਰਵਰਿਆਿਥੀਆਂਨੂੰ ਰਜੂੰ ਨਹਾਂਨੇਰਮਹਨਤਨਹੀਂਕੀਤੀਪਿੀਰਖਆਿੇਣੀਹੀਨਹੀਂਚਾਰਹਿੀ (Those students who have not worked hard, should not give the exam) In above sentence the restrictive clause “ਰਜੂੰ ਨਹਾਂਨੇਰਮਹਨਤਨਹੀਂਕੀਤੀ”‘who have not worked hard’is positioned in between subject“ਉਨਹਾਂਰਿਆਿਥੀਆਂਨੂੰ ”’‘those students’ and predicate “ਪਿੀਰਖਆਿੇਣੀਹੀਨਹੀਂਚਾਰਹਿੀ”‘should not give the exam ‘of independent clause. So three separate portion of the sentence will be:

𝒓𝒆𝒔𝒕𝒓𝒊𝒄𝒕𝒊𝒗𝒆 𝒅𝒆𝒑𝒆𝒏𝒅𝒆𝒏𝒕 𝒄𝒍𝒂𝒖𝒔𝒆

ਉਨਹਾਂਰਵਰਿਆਿਥੀਆਂਨੂੰ ⏟

⏞ ਰਜੂੰ ਨਹਾਂਨੇਰਮਹਨਤਨਹੀਂਕੀਤੀ

𝒔𝒖𝒃𝒋𝒆𝒄𝒕 𝒐𝒇 𝒊𝒏𝒅𝒆𝒑𝒆𝒏𝒅𝒆𝒏𝒕 𝒄𝒍𝒂𝒖𝒔𝒆

ਪਿੀਰਖਆਿੇਣੀਹੀਨਹੀਂਚਾਰਹਿੀ ⏟

𝒑𝒓𝒆𝒅𝒊𝒄𝒂𝒕𝒆 𝒐𝒇 𝒊𝒏𝒅𝒆𝒑𝒆𝒏𝒅𝒆𝒏𝒕 𝒄𝒍𝒂𝒖𝒔𝒆

As clear from above diagram, there is no identification mark between the restrictive dependent clause and predicate of independent clause. Here we used the part of speech tagging information to identify the end of restrictive dependent clause. As all the dependent clause contains verb phrase as one of the essential element and this verb phrase will contain the helping verb or auxiliary verb as the last element, so we scan the restrictive clause and the word with verb part-ofspeech tag and followed by a word having non-verb part-of-speech tag will be the end of the restrictive clause. Now after assigning part-of-speech tag to all the words of above sentence we obtained the sentence in the following form: (ਉਨਹਾਂ_PNDBPO ਰਵਰਿਆਿਥੀਆਂ_NNMPO ਨੂੰ _PPUNU ਰਜੂੰ ਨਹਾਂ_AJIMSDਨੇ_VBAXBPS1 ਰਮਹਨਤ_NNFSD ਨਹੀਂ_PTUN ਕੀਤੀ_VBMAFSXXPTNIA ਪਿੀਰਖਆ_NNFSD ਿੇਣੀ_VBMAFSXXXTNNA ਹੀ_PTUE ਨਹੀਂ_PTUN ਚਾਰਹਿੀ_Unknown ।_Sentence) As shown in above sentence the restrictive clause is: ਰਜੂੰ ਨਹਾਂ_AJIMSD ਨੇ_VBAXBPS1 ਰਮਹਨਤ_NNFSD ਨਹੀਂ_PTUN ਕੀਤੀ_VBMAFSXXPTNIA This clause ends with the word ਕੀਤੀ having part-of-speech tag “VBMAFSXXPTNIA” which shows that it is a main verb and the next word to this main verb in the sentence is ਪਿੀਰਖਆ‘exam ‘having part-of-speech tag “NNFSD”, that shows it is a noun. Therefore, ਕੀਤੀ is the end of the restrictive clause and ਪਿੀਰਖਆ is the beginning of the predicate of the independent clause. 6.1.4 Nonrestrictive In contrast to restrictive clauses, Nonrestrictive clauses tell us something about a preceding subject and they do not limit, or restrict, the meaning of that subject. Conjunctions starting with ‘ਜ’ character are used to provide relevant information of the clause. Even if we remove this clause from the sentence then the meaning of sentence is not much affected. Comma is used to separate different clauses. Like restrictive clause this clause is also positioned in between the subject and predicate of the independent clause. Therefore the sentence containing nonrestrictive clause can have three pairs of clause boundaries; one pair containing start and end of the subject of independent clause, second pair containing the start and end of the nonrestrictive clause and the third pair containing the start and end of the predicate of independent clause. Consider the following example: ਥੋੜਹੀਆਂਰਜਹੀਆਂਚੀਜ਼ਾਂ,ਜੋਅਿੱਗੇਰਪਛੇਪਈਆਂਹਨ,ਬੜੀਆਂਹੀਸੋਹਣੀਆਂਤੇਮੂੰ ਹੋਂਬੋਲਿੀਆਂਸਨ।(A few objects, which has been placed together, looks very beautiful and lively) In this sentence the nonrestrictive clause “ਜੋਅਿੱਗੇਰਪਛੇਪਈਆਂਹਨ” ‘which has been placed together‘ is positioned between the subject “ਥੋੜਹੀਆਂਰਜਹੀਆਂਚੀਜ਼ਾਂ” ‘A few objects‘and predicate “ਬੜੀਆਂਹੀਸੋਹਣੀਆਂਤੇਮੂੰ ਹੋਂਬੋਲਿੀਆਂਸਨ”‘looks very beautiful and lively‘ of the independent clause. So three separate portion of the sentence will be:

[391]

Sharma and Lehal/ COMMUNE-2015

𝒏𝒐𝒏𝒓𝒆𝒔𝒕𝒓𝒊𝒄𝒕𝒊𝒗𝒆 𝒅𝒆𝒑𝒆𝒏𝒅𝒆𝒏𝒕 𝒄𝒍𝒂𝒖𝒔𝒆

ਥੋ ⏟ੜਹੀਆਂਰਜਹੀਆਂਚੀਜ਼ਾਂ ,

𝒔𝒖𝒃𝒋𝒆𝒄𝒕 𝒐𝒇 𝒊𝒏𝒅𝒆𝒑𝒆𝒏𝒅𝒆𝒏𝒕 𝒄𝒍𝒂𝒖𝒔𝒆

⏞ ਜੋਅਿੱਗੇਰਪਛੇਪਈਆਂਹਨ

, ਬੜੀਆਂ ਹੀਸੋਹਣੀਆਂਤੇਮੂੰ ਹੋਂਬੋਲਿੀਆਂਸਨ ⏟ 𝒑𝒓𝒆𝒅𝒊𝒄𝒂𝒕𝒆 𝒐𝒇 𝒊𝒏𝒅𝒆𝒑𝒆𝒏𝒅𝒆𝒏𝒕 𝒄𝒍𝒂𝒖𝒔𝒆

The boundary of these clauses can be detected with comma as the comma is used to separate the subject of independent clause with nonrestrictive clause and nonrestrictive clause with predicate of dependent clause. The first pair of clause boundary marker starts with the first word of the sentence i.e.ਥੋੜਹੀਆਂ’a few’ and ends with the first comma that is with the word ਚੀਜ਼ਾਂ‘objects’. Similarly second pair of clause boundary starts with the conjunction that appears just after the first comma and end with the word just before the second comma i.e. ਹਨ, and third pair of clause boundary marker starts with a word just after the second comma i.e. ਬੜੀਆਂ and end with the last word of the sentence. 6.1.5 Conditional Conditional sentences are used to describe the consequences of a specific action, or the dependency between events or conditions. There are two clauses in the conditional sentence; independent clause and a dependent clause. The dependent clause generally start with a correlated conjunctions (conjunctions occurred in pair). Consider the example: ਜੇਉਹਰਜਉਂਿਾਿਰਹੂੰ ਿਾਤਾਂਇਕਰਿਨਅਵਿੱ ਸ਼ਵਿੱ ਡਾਆਿਮੀਂਬਣਜਾਂਿਾ।(If he had been were alive, he would definitely have become a big man). In above sentence a correlated conjunction ਜੇ-ਤਾਂ has been used. In this sentence the dependent clause is “ਜੇਉਹਰਜਉਂਿਾਿਰਹੂੰ ਿਾਤਾਂ” ‘If he had been were alive‘and the independent clause is “ਇਕਰਿਨਅਵਿੱ ਸ਼ਵਿੱ ਡਾਆਿਮੀਂਬਣਜਾਂਿਾ” ‘he would definitely have become a big man’. This can be diagrammatically represented in the following diagram: ਜੇਉਹਰਜਉਂਿਾਿਰਹੂੰ ਿਾ ⏟ ਤਾਂਇਕਰਿਨਅਵਿੱ ਸ਼ਵਿੱ ਡਾਆਿਮੀਂਬਣਜਾਂਿਾ ⏟ 𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑐𝑙𝑎𝑢𝑠𝑒

𝑖𝑛𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑐𝑙𝑎𝑢𝑠𝑒

There are two clause in the sentence containing conditional clause; one is conditional clause and other is independent clause. In Punjabi language most of the time a conditional clause is followed by an independent clause. A clause boundary is marked in between the conditional and independent clause. The conditional clause starts with the first word of the sentence and ends with the second part of the correlated conjunctions. So the clause boundary mark can be set at the end of the dependent clause. 6.1.6 Non-Conditional These type of clauses starts with ਜਿੋਂ,

ਾਵੇਂ and ਰਕਉਂਰਕ conjunctions. These clauses are positioned either before the

independent clause or after the independent clause. If this occurs before the independent clause then comma is used to separate the clause and if these occurs after the independent clause then no comma is used. Consider the following example: ਜਿੋਂਉਸਿੀਜਾਗਖਿੱ ਲਹੀ,ਚੋਿੀਹੋਚਿੱ ਕੀਸੀ। (When she woke up, the robbery had already happened) This sentence contains two clauses; one is non-conditional clause “ਜਿੋਂਉਸਿੀਜਾਗਖਿੱ ਲਹੀ” ‘When she woke up‘and the second is conditional clause “ਚੋਿੀਹੋਚਿੱ ਕੀਸੀ” ‘the robbery had already happened’. In this sentence, the non-conditional clause occurs before the independent clause and therefore comma has been used. As this sentence contains two clauses, so a clause boundary can be marked after the non-conditional clause: ਜਿੋਂਉਸਿੀਜਾਗਖਿੱ ਲਹੀ ⏟

ਚੋਿੀਹੋਚਿੱ ਕੀਸੀ ⏟

𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑐𝑙𝑎𝑢𝑠𝑒 𝑖𝑛𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑡 𝑐𝑙𝑎𝑢𝑠𝑒

7. Results We tested our system on a sample of 250 randomly selected complex sentences containing predicate bound and nonpredicate bound clauses. Our system was able to correctly identify the boundary of clauses in 239 sentences. The failure of system in 11 sentences was due to presence of unknown word (word not present in the morph). Since currently, our system does not contain any component to deal with the unknown words, so this can be further improved by adding a component to handle the unknown words.

[392]

Sharma and Lehal/ COMMUNE-2015

8. Future Scope This work can be further extended for simplification of Punjabi sentence i.e. converting the large complex sentences in to small simple sentences. This will be helpful in various NLP applications like machine translation, summarization etc. also the same technique can be implemented in other Indian languages having same morphological features. References Vilson J Leffa., 1998. Clause processing in complex sentences. Volume 1, p. 937–943. DarakshaParveen, SanyalRatna, and Ansari Afreen, 2011. Clause Boundary Identification using Classifier and Clause Markers in Urdu Language. Vinh Van Nguyen, Nguyen Minh Le and Shimazu Akira, 2007.Using Conditional Random Fields for Clause Splitting,Proceedings of the 10th Conference of the Pacific Association for Computational Linguistics, p. 58–65. Steven Abney, 1990. Rapid incremental parsing with repair. p. 1–9. Eva I Ejerhed, 1988. Finding clauses in unrestricted text by finitary and stochastic methods. Association for Computational Linguistics. . p. 219–227 Harris V Papageorgiou, 1997. Clause recognition in the framework of alignment. p. 417–426. Georgiana Pu¸sca¸su, 2004. A Multilingual Method for Clause Splitting. Orasan,C, 2000. A hybrid method for clause splitting. Proceedings of ACIDCA 2000 Corpora Processing, Monastir, Tunisia, p. 129 – 134. ErifK.Tjong and Sang Kim, 2001. Introduction to the CoNLL-2001 Shared Task: Clause Identification. Erik F. Tjong, Sang Kim, D´ejeanHerv´e, 2001. Introduction to the CoNLL-2001 Shared Task: Clause Identification, Proceedings of the 5th Conference on Natural Language Learning (CoNLL-2001) at the 39th Annual Meeting of the Association for Computational Linguistics (ACL 2001)’, Toulouse, France, p. 52–57. Eraldo R. Fernandes, A. Pires Bernardo, N. dos Santos C´ıcero, Milidi´uRuy L, 2009. Clause Identification using Entropy Guided Transformation Learning. Zhemin Zhu, Bernhard Delphine and GurevychIryna, 2010. A Monolingual Tree-based Translation Model for Sentence Simplification,Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), p. 1353–1361, Ani Thomas, Kowar M K, Sharma Sanjay and Sharma H R, 2011. Extracting Noun Phrases in Subject and Object Roles for Exploring Text Semantics, International Journal on Computer Science and Engineering (IJCSE) vol-3 NaushadUzZaman , Jeffrey P. Bigham and James F. Allen, 2011. Multimodal Summarization of Complex Sentences, IUI 2011, February p. 13-16. Xavier Carreras and Lluis Marquez, 2001. Boosting Trees for Clause Splitting. Aniruddha Ghosh, Das Amitava, BandyopadhyaySivaji, 2009. Clause Identification and Classification in Bengali,Department of Computer Science and Engineering JadavpurUniversity .

[393]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Parameter Optimization of Ballistic Carbon Nanotube Field Effect Transistor Devi Dass, Rakesh Prasher, Rakesh Vaid* Department of Physics and Electronics, University of Jammu, Jammu, India

Abstract Carbon nanotubes can be considered as the most promising building blocks of a future nanoelectronic technology. For further extension of Moore’s Law, it is necessary to combine novel nanotechnologies with Silicon onto the same Silicon platform. Carbon nanotube field effect transistor (CNTFET) is one of the novel nanoelectronics devices that have been investigated as a potential alternative to CMOS devices in future. The present paper investigates the performance of a cylindrical shaped ballistic n-CNTFET and the results have been compared with the 15nm n-MOSFET and 60nm Tri gate n-MOSFET. It has been concluded that the ballistic CNTFET has high drive current (Ion), low leakage current (Ioff), high transconductance, high Ion/Ioff ratio (fast switching speed), and improved subthreshold slope (SS) as well as drain induced barrier lowering (DIBL).

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Carbon Nanotube; CNTFET; Insulator Thickness; Dielectric constant; Threshold voltage

1. Introduction Moore Law has been the guiding principle for the semiconductor industry in the last four decades (Chau et al., 2003; Lundstrom, 2003; Theis et al., 2010). The semiconductor industry and academia have pushed the downscaling of MOSFET for achieving its requirements such as low power consumption, high performance and for extending Moore’s Law (Hasegawa et al., 2004; Hashim et al., 2008). ITRS predicts that MOSFETs will reach sub-10 nm dimensions by 2016, however, to realize devices beyond the 45 nm technology node, novel device architectures along with high mobility materials are required for enhanced performances to improve the on-current and to reduce the power absorption (Prasher et al., 2013). As the MOSFET size reaches its limiting value, various short channel effects appear to degrade its performance. Multigate transistors are a solution of reducing short-channel effects and also playing a crucial role for the continuation of Moore’s Law well through the middle of the next decade (Ferain et al, 2011; Dass et al., 2013). In order to extend Moore’s Law further into the next decade, it is necessary that the other novel nanotechnologies may be combined with silicon onto the same silicon platform. For further nanoelectronics applications much progress has been made in the research of non-Si nanotechnology, therefore, several emerging nanoelectronics devices such as carbon-nanotube field effect transistor (CNTFET), semiconductor-nanowire FET, and planar III-V compound semiconductor FET to be integrated onto the silicon platform for enhancing circuit functionality and for further extending Moore’s Law (Chau et al., 2003). CNTFET were first fabricated in 1998 by Martels (Tans et al, 1998) and Dekkers group (Martel et al., 1998). Later on, Ali Javey et al. have reported the room-temperature ballistic transistors with ohmic metal contacts and high-k gate dielectric in 2003. CNTFETs exhibit near-ballistic transport characteristics, resulting in high-speed devices (Javey et al., 2003). In this research paper we have studied the performance of a cylindrical shaped ballistic n-channel carbon nanotube field effect transistor. Various results of the device shall be explained such as IDS vs. VGS characteristics, IDS vs. VDS characteristics, and mobile charge. We have also compared the results of ballistic n- CNTFET with the 15nm nMOSFET and 60nm Tri gate n-MOSFET.

* Corresponding author. Tel.: +91 9419 106794. E-mail address: [email protected] ISBN: 978-93-82288-63-3

Das et al/COMMUNE-2015

2. Device Description The structure of a cylindrical shaped ballistic n-CNTFET as shown in Fig. 1. The cylindrical gate geometry offers the best electrostatic gate control. In this structure, the intrinsic CNT channel is separated from the source/drain metal contact by the heavily n-doped CNT source/drain extension to minimize the Miller capacitance between gate and source/drain electrode. The source/drain region could also be realized by using weakly coupled metal–nanotube contacts with an appropriate metal work function. We assume that the metal–nanotube contact resistance, RC = 0, and carrier transport through nanotube is ballistic.

Fig. 1. Schematic diagrams of the cylindrical gated ballistic n-type CNTFET (a) 3D view (b) 2D view.

3. Results and Discussion

In this section, we have discussed the various simulation results of cylindrical shaped ballistic n–CNTFET such as IDS vs. VGS characteristics, IDS vs. VDS characteristics, and mobile charge by optimized the parameters like CNT diameter d = 1 nm, threshold voltage Vth = 0.3 V, temperature T = 300 K, insulator thickness tins = 1.5 nm, gate insulator dielectric constant k = 3.9 (SiO2), gate control parameter αG = 1, drain control parameter αD = 0 while the gate and drain voltages are varied. All results are carried out using the nanohub simulator FETtoy (Rahman et al., 2006). 3.1.

IDS vs. VGS Characteristics

Fig. 2 (a) shows how the drain-source current (IDS) changes with the change in gate-source voltage (VGS) for a positive drain-source voltage (VDS) of 0.1 V and 1 V. The drain current has been plotted using a linear scale and a logarithmic scale. In a linear scale, no current flows for below threshold voltage because energy barrier between source and drain is high since no electrons travels from source to drain. Above the threshold voltage, the current increases linearly with the applied gate bias due to the reason that energy barrier between the source and drain is reduced with the increase in gate-source voltage (VGS) and more electrons flows from source to drain which further indicates that more current flows. It is also evident from the figure that the transconductance of device improves with the increase in drainsource voltage (VDS) due to more controllability of gate on the channel.

(a)

(b)

[395]

Das et al/COMMUNE-2015 Fig. 2. (a) IDS vs. VGS characteristics and (b) IDS vs. VDS characteristics

Further, for low drain-source voltage (VDS = 0.1 V), faster saturation occurs in the curve which indicate that the device has fast switching speed when the voltage scales down. In a logarithmic scale, it is clear that the drain current varies exponentially with the gate voltage below the threshold value. The slope of this exponential increase on a logarithmic scale is called the subthreshold slope (SS) and is expressed in millivolts per decade of current. The value for the subthreshold slope is 60 mV/decade means that a 60 mV increase in the gate voltage brings about a tenfold increase in the drain current. Also, in logarithmic scale, for below threshold voltage, the I DS curve overlap which means that the DIBL effect is zero. Hence we can say that the cylindrical shaped ballistic n-CNTFET controls the short-channel effects. Because of similar behavior of CNTFET is observed as in conventional MOSFET hence the named as MOSFET-like CNTFET. 3.2.

IDS vs. VDS Characteristics

Fig. 2(b) shows how the drain-source current (IDS) changes with the change in drain-source voltage (VDS) for a given change in the values of gate-source voltage (VGS). It has been observed that the saturation current increases with the gate-source voltage (VGS) and degree of this positive effect increases as we go for higher gate-source voltage (VGS). This figure further indicates that drain current increases with increase in drain voltage up to pinch-off voltage and beyond this point there is no effect of drain voltage over the drain current since it remains constant as happens in conventional MOSFETs. Further, it can be seen from the figure that the output conductance increases with the increase in gate-source voltage (VGS). 3.3.

Variation of Mobile Charge/q vs. VDS

Fig. 3 (a) illustrates how the mobile charge changes with the change in drain- source voltage (VDS) for a given change in the values of gate-source voltage (VGS). It has been observed that the variation of mobile charge/q w. r. t. drain -source voltage (VDS) show inverse relationship with the variation of drain current w. r. t. drain-source voltage. It is observed that increasing the drain voltage beyond a pinch off voltage has no longer an effect on the shape of the curves since the mobile charge remains constant. It is also observed that low drain voltage produces higher mobile charge and high drain voltage produces lower mobile charge. Further, it has been noticed that the mobile charge increases with the increase in gate-source voltage (VGS).

(a)

(b)

Fig. 3. (a) Variation of mobile charge/q vs. VDS and (b) Variation of mobile charge/q vs. VGS

3.4

Variation of Mobile Charge/q vs. VGS

Fig. 3(b) illustrates how the mobile charge changes with the change in gate-source voltage (VGS) for a given drainsource voltage (VDS) of 0.1V and 1V. It has been observed that the variation of mobile charge/q w. r. t. gate-source voltage (VGS) show direct relationship with the variation of drain current w. r. t. gate-source voltage i.e. for below threshold voltage (VGS < Vth), the mobile charge remains constant since the device is off and for above threshold voltage (VGS > Vth), the mobile charge increases with the increase in gate-source voltage. The results further show that the mobile charge increases due to the reduction in drain- source voltage.

[396]

Das et al/COMMUNE-2015

3.5

Comparison of n-CNTFET with 15nm Single Gate n-MOSFET and 60nm Trigate n-MOSFET

The comparison of ballistic n-CNTFET with 15nm n-MOSFET and 60nm Trigate n-MOSFET as shown in Table 1. It has been observed in table 1 that the ballistic n-CNTFET has very high drive current (Ion) which is 7295.55 % maximum than single gate MOSFET and 2819.29 % maximum than tri gate MOSFET. The ballistic n-CNTFET has low leakage current (Ioff) which is 79.44% lower than single gate MOSFET and 47.14 % lower than tri gate MOSFET. The ballistic n-CNTFET has high Ion/Ioff ratio (~106) i.e. fast switching speed which is maximum as compared to 15nm n-MOSFET (103) and 60nm Trigate n-MOSFET (104). The short-channel effects such as Subthreshold Slope (SS) and Drain Induced Barrier Lowering (DIBL) are ideal in case of ballistic n-CNTFET. Hence, we can say that the controllability of gate over the channel will be more in case of ballistic n-CNTFET. It is further observed that the CNTFET is an excellent way for further extension of Moore’s Law. Table 1. Comparison of ballistic CNTFET with single gate MOSFET and tri gate MOSFET.

Ion

15nm MOSFET (Chau et al., 2003) 450 µA/µm

60nm Tri gate MOSFET (Chau et al., 2003) 1140 µA/µm

Ioff

180 nA/µm

70 nA/µm

37 nA/µm

Ion/Ioff

2500

16285.71

898730

SS

95 mV/decade

68 mV/decade

60 mV/decade

DIBL

100 mV/V

41 mV/V

0.60 mV/V

Device Metrics

Ballistic CNTFET with CNT diameter = 1nm 33, 280 µA/µm

4. Conclusion This paper investigates the performance of cylindrical shaped ballistic n-CNTFET. It can be concluded that the CNTFET perform better when the gate or drain voltage is increased and it has very high drive current (Ion), low leakage current (Ioff), high Ion/Ioff ratio (~106) i.e. fast switching speed and improves short channel effects in comparison with single-gate and tri gate MOSFET. 5. Acknowledgements One of the authors Mr. Devi Dass gratefully acknowledges the University Grants Commission (U.G.C.), for the award of ‘Rajiv Gandhi National Fellowship’ under the scheme funded by Ministry of Social Justice & Empowerment, Govt. of India. References Chau, R., Boyanov, B., Doyle, B., Doczy, M., Datta, S., Hareland, S., Jin, B., Kavalieros, J., Metz, M., 2003. Silicon Nano-transistors for Logic Applications, Physica E 19, p. 1. Lundstrom, M., 2003. Moore's Law Forever?, Science 299, p. 210. Theis, T. N., Solomon, P. M., 2010. It's Time to Reinvent the Transistor, Science 327, p. 1600. Hasegawa, H., Kasai, S., Sato, T., 2004. Hexagonal Binary Decision Diagram Quantum Circuit Approach for Ultra-Low Power III-V Quantum LSls, IEICE Transaction on Electron, E87-C, 1757 (2004). Hashim, A.M., Pung, H.H., Pin, C.Y., 2008. Characterization of MOSFET-like Carbon Nanotube Field Effect Transistor, Jurnal Teknologi 49 (D), p. 129. http://www.public.itrs.net Prasher, R., Dass, D., Vaid, R., 2013. Study of Novel Channel Materials Using III-V Compounds with Various Gate Dielectrics, International Journal on Organic Electronics (IJOE) 2, p. 11. Prasher, R., Dass, D., Vaid, R., 2013. Performance of a Double Gate Nanoscale MOSFET (DG-MOSFET) Based on Novel Channel Materials, Journal of Nano and Electronic Physics 5, p. 010171. Ferain, I., Colinge, C. A., Colinge, J.P., 2011. Multigate Transistors as the Future of Classical Metal-Oxide-Semiconductor Field-Effect Transistors, Nature 479, p. 310. Dass, D., Prasher, R., Vaid, R., 2013. Impact of Scaling Gate Insulator Thickness on the Performance of Carbon Nanotube Field Effect Transistors (CNTFETs), Journal of Nano and Electronic Physics, 5, p. 020141. Tans, S. J., Verschueren, A.R.M., Dekker, C.,1998. Room-Temperature Transistor Based on a Single Carbon Nanotube, Nature 393, p. 49. Martel, R., Schmidt, T., Shea, H. R., Hertel, T., Avouris, P., 1998. Single-and Multi-Wall Carbon Nanotube Field-Effect Transistors, Applied Physics Letters 73, p. 2447. Javey, A., Guo, J., Wang, Q., Lundstrom, M., Dai, H., 2003. Ballistic Carbon Nanotube Field-Effect Transistors, Nature 424, p. 654. Rahman, A., Wang, J., Guo, J., Hasan, M. S., Liu, Y., Matsudaira, A., Ahmed, S. S., Datta, S., Lundstrom, M., 2006. FETToy.

[397]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Morphological Analysis of Proper Nouns in Punjabi Umrinderpal Singh*, Vishal Goyal, Gurpreet Singh Lehal Department of Computer Science Punjabi University Patiala, India

Abstract The Morphological is the branch of linguistics. This field is related to the study of structure of words. The Morphological Analysis becomes a very popular branch of research in all languages, especially for the morphological rich languages like Indo-Aryan languages. The Morphological Analysis and Generator play essential roles in many Natural Language Processing (NLP) applications like Part of Speech (POS) Tagger, Named Entity Recognitions (NER) and many other applications. This paper presents an approach to Morphological Analysis of Proper Nouns in Punjabi. A thorough analysis has been done for Punjabi and found many suffix patterns that can be used to classify Proper Nouns. Two thousand thirty unique suffixes has been found during analysis. Analysis has been done on a large Punjabi Corpus having 278098 unique words. Based on these suffixes proposed algorithm was able to get 97.42% accuracy to identify proper nouns.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: NLP: Part of Speech Tagger; POS; Punjabi; Rule Based; HMM

1.

Introduction

The Morphological analysis and generation is an essential part of any NLP applications. It has a vital role in morphological rich languages such as Punjabi, Hindi and other Indo-Aryan languages. In morphological analysis, study of the structure of the words based on its root and affixes. The morphological analysis has become essential in IndoAryan languages, where words can be inflected in many forms and yield different meaning. The main concern of the morphological analysis to get the grammatical information by studying the word's structure like gender, number person etc.[Goyal and Lehal. 2008]. A word can be of two types; simple and compound. Where simple words combination of root words and its suffixes on the other hand compound words can be broken into two independent words and there independent words has their own meaning[ Goyal and Lehal 2008]. Inflection morphology [Bansal et.al 2011] gives us different form of a word by adding or removing affixes. Changes in the word meaning are minimal for exp: Cat Faster

---

Cats Fast

In derivational morphology, derives new forms of words from existing words and word class is changed on deriving for example: Modern (adj) Drink (v) --

-Modernize (v) Drinkable (adj)

2. Related Work Punjabi is one of the rich Morphological languages in Indo-Aryan language's family. However, very few morphological analyzer and generator has been developed for Punjabi. [Gill 2007] had been developed rule based morphological analyzer for Punjabi. This system was developed for Punjabi grammar checker. The system was based on database lookup for root words. If word not found in database, it marks it as unknown [Bansal et.al 2011]. [Singh *

Corresponding author. Tel. +91 9914996719 E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Singh et al/ COMMUNE-2015

and Kumar 2013] has been developed a Punjabi Morphological analyzer for Nouns, Pronouns and Adjectives. This Morphological analyzer was the part of UNL (Universal Network Language) based Machine Translation system (MT) system. [Bansal et.al 2011] suggests way to increase the accuracy of Morphological analyzer system developed by [Gill et.al 2007],[Mohd. Hunayoun et. Al] has been described as an implementation of morphological and development of corpus for Punjabi in Shamukhi script, which is spoken in Pakistan Punjab. [Goyal and Lehal 2008] had been developed morphological analyzer for Hindi as part of the MT system Hindi to Punjabi. [Snigdh paul et.al 2013] has been proposed a method for over stemming word in Hindi by various rules to make the word proper stem. Reader may fallow [Shweta Vikram 2013; Antony PJ et. al. 2012] paper for more details of morphological analysis for Indian languages and available techniques. Related word shows that very limited work has been done for Indian languages, especially for Punjabi. One should need to explore all morphological rich languages in detail. Efforts were made to explore Proper Nouns in Indian Languages and shows that how rich are these languages even for Proper Nouns. To best of our knowledge, we did not find any work that particular based on morphological analysis of proper nouns in Indo-Aryan languages. 3. Morphology of Proper Nouns in Indian Languages Indian languages have very rich morphological features as compared to European languages. Words can be inflected in many forms and may yield different meanings. Many researcher has been proposed various algorithm and methods only for open class words like nouns, verb and adjectives etc. [Rohit kansal 2012; Harinder Singh 2013; Mayuri Rostagi at.al 2014]. However, none of the researcher works on morphological analysis of Proper Nouns. By analysis of proper nouns in Indian Languages (IL) we can found various hidden features or information which can help us to classify the word as proper noun. The Morphological analysis of proper nouns is essential for resource poor languages. It is not possible to collect all the proper nouns for a particular language and proper nouns may came in any form. Most of the proper nouns are ambiguous for example word ਲਾਲ (Lal)(Red) is a valid noun word and it also valid person name. In this paper, morphological analysis has been done for proper nouns in Punjabi language and find out various suffixes, which were part of the person and location names in Punjabi. By doing the analysis of these suffix patterns, one can identify the word as proper noun. This kind of analysis can be very useful in POS tagging, where algorithm can check, out of vocabulary (OOV) words and classify them into proper nouns. 4 Patterns in Proper Nouns Morphological rich languages like Punjabi; Hindi Urdu has some unique features related to proper nouns. Analysis has been done to find out these unique suffixes in proper nouns. Most of the proper nouns in Indo-Aryan languages consist of two words, root word and its suffix for example. person name 'Ramdas' consist of two independent words 'Ram' and 'Das' where 'Ram' means 'lord Rama ' and 'das' means 'Servant' combined meaning of both words servant of load Rama. Based on these suffixes we can able to find out all proper names that ends with das. Following table shows proper nouns suffixes. Table 1. Compound word for Proper Nouns S.No. 1

Name ਜੀਵਨਦਾਸ(Jīvanadāsa)

Suffix ਦਾਸ(Dāsa)

2

ਹਰਨਾਮਦਾਸ(Haranāmadāsa)

ਦਾਸ(Dāsa)

3

ਹਰਰਦਾਸ(Haridāsa)

ਦਾਸ(Dāsa)

4

ਜੀਵਨਦਾਸ(Jīvanadāsa)

ਦਾਸ(Dāsa)

5

ਰਰਵਦਾਸ(Ravidāsa)

ਦਾਸ(Dāsa)

6

ਮੁਰਨਦਾਸ(Munidāsa)

ਦਾਸ(Dāsa)

Table 1 shows compound words consists of two independent words. There are many proper nouns, which have simple word forms, consist of root and their suffix but have unique features to identify them as proper noun. Following table 2 shows simple form of words, which are valid candidate words for proper nouns. Table 2.Simple words form for Proper Noun S.No. 1

Name ਤਾਰਰਕਾ(Tārikā)

Suffix ਾਾਰਰਕਾ(̔Ārikā)

2

ਸਾਰਰਕਾ(Sārikā)

ਾਾਰਰਕਾ(̔Ārikā)

3

ਹਜਾਰਰਕਾ(Hajārikā)

ਾਾਰਰਕਾ(̔Ārikā)

4

ਦਵਾਰਰਕਾ(Davārikā)

ਾਾਰਰਕਾ(̔Ārikā)

[399]

Singh et al/ COMMUNE-2015

S.No. 5

Name ਸਾਰਰਕਾ(Śārikā)

Suffix ਾਾਰਰਕਾ(̔Ārikā)

6

ਰਨਹਾਰਰਕਾ(Nihārikā)

ਾਾਰਰਕਾ(̔Ārikā)

By performing an analysis on Punjabi language and find out 230 categories, which were unique to Punjabi language and no evidence where found where these suffixes conflict with other words of Punjabi language. Most categories of suffixes have unique patterns but few categories were also found those suffixes might conflict with other Punjabi words, which were not person names for example; word ਉਦਾਸ has valid suffix ਦਾਸ but may or may not be person name. These words were ambiguous for names category. All these words treated as special cases and tried to resolve their ambiguity by checking previous and next word of current position. The person designation list like ਸਰੀ, ਸਰੀਮਾਨ, ਸਾਰਹਬ etc. and last name ਕੁਮਾਰ, ਰਸਿੰ ਘ etc. has collected to resolve ambiguity of these words. If ambiguous words start with some designation or end with some last name based on this localized information we can resolve their ambiguity. A proper name may have two valid suffixes; we have used longest suffixes stripping approach to identify the word. Following table shows some names having two suffixes. Table 3. Two or more valid suffixes for Person Names S.No. 1

Name ਅਜਮੁੁੱ ਦੀਨ(Ajamudīna)

Suffix1 ਾੁੱਦੀਨ

Suffix2 ਦੀਨ

2

ਅਨਵਰੁੁੱ ਦੀਨ(Anavarudīna)

ਾੁੱਦੀਨ

ਦੀਨ

3

ਉਮਾਵਤੀ(Umāvatī)

ਾਾਵਤੀ

ਵਤੀ

4

ਅਮਰਾਵਤੀ(Amarāvatī)

ਾਾਵਤੀ

ਵਤੀ

5

ਅੋਮਾਨਿੰਦ(A̔ōmānada)

ਾਾਨਿੰਦ

ਨਿੰਦ

6

ਆਤਮਾਨਿੰਦ(Ātamānada)

ਾਾਨਿੰਦ

ਨਿੰਦ

7.

ਕਮਲੇ ਸ਼ਵਰੀ(Kamalēśavarī)

ਾੇਸ਼ਵਰੀ

ਸ਼ਵਰੀ

8.

ਸੁਰੇਸ਼ਵਰੀ(Surēśavarī)

ਾੇਸ਼ਵਰੀ

ਸ਼ਵਰੀ

9.

ਰਾਮਪ੍ਰਸਾਦ(Rāmaprasāda)

ਪ੍ਰਸਾਦ

ਸਾਦ

10.

ਪ੍ੀਰਪ੍ਰਸਾਦ(Pīraprasāda)

ਪ੍ਰਸਾਦ

ਸਾਦ

Stripping the largest suffix in to identify the proper nouns always is better clue as compared to smallest suffix stripping. Smallest suffix striping may yield ambiguous information to identify proper nouns, for example. ਰਾਮਪ੍ਰਸਾਦ name has two valid suffixes ਪ੍ਰਸਾਦ and ਸਾਦ has length 5 and 3 respectable. Suffix ਪ੍ਰਸਾਦ always helps to identify word as proper name and there is no chance to classify word wrongly, but second suffix ਸਾਦ is also a valid name to identify word ਪ੍ਰਸਾਦ but has high chances to identify word wrongly for example, ਫ਼ਸਾਦ word also ends with suffix ਸਾਦ but ਫ਼ਸਾਦ is not a person name. Therefore, always better to check longest suffix first to identify proper nouns. Like person names, location names also have morphological features to identify them. Following table 4 shows location names that end up some particular suffixes. Table 4. Location names with suffixes S.No 1

Location Names ਸੁਲਤਾਨਪ੍ੁਰ(Sulatānapura)

Suffix ਪ੍ੁਰ(Pura)

2

ਹੈਦਰਾਬਾਦ(Haidarābāda)

ਾਾਬਾਦ(Ābāda)

3

ਰਕਸਨਪ੍ੁਰਾ(Kisanapurā)

ਪ੍ੁਰਾ(Purā)

4

ਉਜਬੇਰਕਸਤਾਨ(Ujabēkisatāna)

ਸਤਾਨ(Satāna)

5

ਮਨੀਮਾਜਰਾ(Manīmājarā)

ਮਾਜਰਾ(Mājarā)

5. Methodology used to Extract Proper Nouns In-depth analysis has been done on the Punjabi corpus to extract all valid suffixes for Proper Nouns, 80% of the total corpus has been used for this analysis. Following diagram 1 shows process to extract suffixes for proper noun. Initially we have collected around 30000 proper nouns related to person names and locations. All person names belong to different communities of Indo-Aryan cultures.

[400]

Singh et al/ COMMUNE-2015 Table 5.Training Corpus Detail Domain News Data Health Data Tourism Data Agriculture Data

Total Words 400200 200541 230853 208422

Unique Words 150040 60087 63082 72421

Along with this proper noun lexicon, a huge raw corpus of Punjabi was used to find out suffix patterns related to names. Table 5 shows the details of corpus. A huge raw corpus was needed to check proper noun suffixes, to extract only those suffix patterns, which were unique in all domains and words, should not conflict with any domain's data. We have developed the algorithm to extract all unique suffix patterns along with their frequency and suffix length. All unique words have been extracted from corpus of 1040016 words. We have found 278098 total unique words. All unique suffixes having length greater than 3 applied on these unique words, to extract all those words that end with given suffixes. We have chosen the length greater than 3 because length less than this yield many ambiguous words, which were not candidate words for proper nouns. After applying all suffixes, we got 47501 words, which were 17% of all the unique words. All proper nouns extracted from 47501 words, which already part of the proper names lexicon, along with their suffixes, and we left with two lists. The List having new words, which may be the candidates for proper nouns and list of words, which we already have in our proper name lexicon. We have divided the list of all words end with valid suffix into two lists because it made us easier to find out all ambiguous words. Proper Nouns

Raw Corpus

Suffixes length >=3 Extract Unique Suffixes

Prop er

Extract Unique Words

Find Valid

Words Ends with Suffixes

Valid Names with suffixes

No

New Names/ Words with suffixes

Yes

If Suffixes Common in both list

Ambiguous suffixes for Names

Valid suffixes for Names

Tag and remove all Verbs

List of Verbs

Manually check all names and suffixes

Valid list of suffixes for Proper Nouns

Fig. 1: Methodology to extract valid suffixes

Compassion of both list's suffixes yield all ambiguous words having common suffix in both lists. If suffix is only part of the proper name list, we can choose that suffix as a valid suffix to identify proper noun without any ambiguity. On the other hand, we had a list of ambiguous words and their suffixes. To refine ambiguous word list, we had used the list of verbs. The verb tag helps us to tag and remove all those words, which are verbs. The verb list consists of around

[401]

Singh et al/ COMMUNE-2015

23000 unique words. All forms of the verb have been used to tag verbs in list. Following table 6 shows the list of Verb tags used for tagging the verbs in ambiguous list. Table 6.All Verb POS tags Verb Types Main Verb Non-Finite Infinite Gerund Auxiliary

Label V_VM V_VM_VNF V_VM_VINF V_VM_VNG V_VAUX

Proper nouns mainly confessed with Adjectives and Nouns. Therefore, we need to check all these words manually. Whether they were valid words for proper nouns or not. Manually analysis has been done to extract all those suffixes, which can be used to classify words as proper nouns. Ambiguous word list also extracted that ends with valid suffixes. 6. System Architecture to Find Proper Nouns In out proposed system to find proper noun based on suffixes, four different lists were used to identify names correctly. Initially, the systems take Punjabi input text. The tokenization and normalization process token the text into words and remove unwanted symbols. The system find and remove all ambiguous words from the input text like word ਰਜਿੰ ਦਰਾ end with valid suffix ਾਿੰਦਰਾ but never used as person name. Suffix ਾਿੰਦਰਾ used to identify many other names like ਵਰਰਿੰ ਦਰਾ, ਰਰਜਿੰ ਦਰਾ, ਭੁਰਪ੍ਿੰ ਦਰਾ. Therefore, we have developed a list of words, which were mostly nouns to identify ambiguous words. This list contains 203 ambiguous words. Following table 7 shows some of the ambiguous words having valid suffixes. Table 7. Ambiguous words SNo. 1

Names ਮਨੋਰਿੰ ਜਨ(Manōrajana)

Suffix ਰਿੰ ਜਨ(Rajana)

2

ਹਸਤਪ੍ਾਲ(Hasatapāla)

ਤਪ੍ਾਲ(Tapāla)

3

ਪ੍ੁਸਰਤਕਾ(Pusatikā)

ਰਤਕਾ(Tikā)

4.

ਜੀਵਨ(Jīvana)

ਜੀਵਨ(Jīvana)

5.

ਏਕਾਂਤ(Ēkānta)

ਕਾਂਤ(Kānta)

The list of valid suffixes applied on all tokens to identify proper nouns and mark them for further processing. We have used 230 suffixes to identify proper names. Appendix A shows some of the most frequently used suffixes for proper nouns. Input Text

Tokenization/Normali zation

List of Ambiguous words which are not names

Find and Remove ambiguous words

Compare words with valid suffix list

List of Ambiguous names

Find and Tag ambiguous names

Rules to resolve ambiguity Fig 2: System Architecture Output Text

[402]

List of valid suffixes

List of Designation s middle names and last names

Singh et al/ COMMUNE-2015

The list of ambiguous proper nouns has been used to identify and mark them. Ambiguous names like word ਸਮੁਿੰ ਦਰ, ਜੀਵਨ etc. can be used as proper nouns or they can be used can adjective and noun. The algorithm marks them as ambiguous names and various rules have been used to resolve their ambiguities. The rules used localization information and list of designation, middle names and last names to resolve their ambiguities for example word ਸਮੁਿੰ ਦਰ is an ambiguous word, which can be used as proper names or can be used as adjective. ਫੁੁੱ ਲ ਬਹੁਤ ਹੀ ਸੁਿੰ ਦਰ ਹੈ.( Flower is very beautiful) ਸਰੀ ਸੁਿੰ ਦਰ ਕੁਮਾਰ (Shree Sunder Kumar) Algorithm check previous and next words to disambiguate it like if it has some designation ਸਰੀ or ਸਰੀ ਮਾਨ etc or may have surname ਕੁਮਾਰ, ਰਸਿੰ ਘ etc. then it marked as proper noun by algorithm. The system output is all valid proper names based on their suffix information. 7 Evaluation valuation of the all the suffixes has been done on 20% of the corpus which was not used for suffix pattern extraction. Testing data consists of four domains namely political news data, Heath data, Tourism data and agriculture data. Table 8 shows the detail of testing data. Table 8. Testing Data Details Domain News Data Health Data Tourism Data Agriculture Data

Total Words 30300 8998 5859 9893

Total unique words in test data was 2502. Testing data contains 453 words ends with valid suffixes used for proper noun classification. Out of 453 words, 54 words were ambiguous. These words ends with valid suffix but words can be classified as noun, adjectives or proper nouns. Evaluation has been done using standard metrics. Recall(R) =

𝐂𝐨𝐫𝐫𝐞𝐜𝐭 𝐚𝐧𝐬𝐰𝐞𝐫𝐬 𝐠𝐢𝐯𝐞𝐧 𝐛𝐲 𝐬𝐲𝐬𝐭𝐞𝐦 𝐓𝐨𝐭𝐚𝐥 𝐩𝐨𝐬𝐬𝐢𝐛𝐥𝐞 𝐜𝐨𝐫𝐫𝐞𝐜𝐭 𝐚𝐧𝐬𝐰𝐞𝐫𝐬

Precision (P) = F1-Measure =

𝐂𝐨𝐫𝐫𝐞𝐜𝐭 𝐚𝐧𝐬𝐰𝐞𝐫𝐬 𝐀𝐧𝐬𝐰𝐞𝐫𝐬 𝐩𝐫𝐨𝐝𝐮𝐜𝐞𝐝

𝟐∗𝐑∗𝐏 𝐑+𝐏

Table 9. System evaluation of system

Recall Precision F1-Measure

97.13 97.72 97.42

Algorithm yield 97.42% accuracy to find proper nouns based on their suffixes. Sometime the system failed to find some proper nouns because of their ambiguities. Some person names were ambiguous and have no information around them in sentence to resolve their ambiguity like designation and surnames. The system also marks some words, which were not person names like word ਅਰਨਿੰਦਰਾ having valid suffix ਾਿੰਦਰਾ. 8 Conclusion The morphological analyzer has been used by many researchers in different NLP applications like POS tagger and NER systems. It’s role becomes very essential in morphological rich languages. In this paper, an approach has been discussed to find proper based on their suffix information. Analysis shows that how rich Indo-Aryan languages are for

[403]

Singh et al/ COMMUNE-2015

even proper nouns. To the best of our knowledge, there has been no work done for proper nouns, analysis based on their morphological information. Various morphological analyzers and generators have been proposed by the researcher for Indo-Aryan languages. These analyzers were based on verbs, nouns and other part of speech. Algorithm successfully extracted many suffix patterns for Punjabi and used them to extract proper nouns with the accuracy of 97.42%. Most of the suffixes can be directly used in other Indian languages without any conflicts with other words of that language. In future work, effort will made to explore Hindi and Urdu languages for Proper Name's suffixes. References Antony P J and Dr K P Soman. 2012. Computational Morphology and Natural Language Parsing for Indian Languages: A Literature Survey, Volume 3 No. 4: 136-146 Deepak Kumar Manjeet Singh and Seema Shukla. 2012. FST Based Morphological Analyzer for Hindi, volume 9, 349-353 Gagan Bansal, Satinder Pal Ahuja, Sanjeev Kumar Sharma. 2011. Improving Existing Punjabi Morphological Analyzer, Volume 5: 221-229 Gill Mandeep Singh, Lehal Gurpreet Singh, Joshi S.S. 2007. A full form lexicon based Morphological Analysis and generation tool for Punjabi, International Journal of Cybernatics and Informatics, Hyderabad, India,October 2007, pp 38-47 Harinder Singh and Parteek Kumar. 2013. Analysis of Noun, Pronoun and Adjective Morphology for NLization of Punjabi with EUGENE, volume 2: 436-442 Mohd. Shahid Husain. 2012. An Unsupervised Approach to Develop Stemmer, International Journal on Natural Language Computing, 15-23 Manish Shrivastava and Pushpak Bhattacharyya. 2008. Hindi POS Tagger Using Naive Stemming : Harnessing Morphological Information Without Extensive Linguistic Knowledg, International Conference on NLP (ICON08), Pune, India, December, 2008 Also accessible from http://ltrc.iiit.ac.in/proceedings/ICON-2008 Mayuri Rastogi and Pooja Khanna. 2014. Development of Morphological Analyzer for Hindi, International Journal of Computer Applications Volume 95-No-17: 1-5 Muhammad Humayoun and Aarne Ranta. 2010. Developing Punjabi Morphology, Corpus and Lexicon, n. The 24th Pacific Asia conference on Language, Information and Computation, : 163-172 Snigdha Paul, Mini Tandon, Nisheeth Joshi And Iti Mathur. 2013. Design Of A Rule Based Hindi Lemmatizer: 67-74 Vaishali Gupta, Nisheeth Joshi and Iti Mathur. 2013. Rule Based Stemmer in Urdu, Computer and Communication Technology (ICCCT), 129 - 132 Vishal Goyal, Gurpreet Singh Lehal. 2008. Hindi Morphological Analyzer and Generator, First International Conference on Emerging Trends in Engineering and Technology: 1156-1159. Appendix A. List of frequently used suffixes S.No. 1.

Suffixes

Examples

ਨਪ੍ਰੀਤ(Naprīta)

ਹਰਮਨਪ੍ਰੀਤ,ਮਨਪ੍ਰੀਤ(Haramanaprīta,manaprīta)

2.

ਨਾਰਾਇਣ(nārā'iṇa)

ਜੈਨਾਰਾਇਣ,ਰਸਵਨਾਰਾਇਣ(ainārā'iṇa,śivanārā'iṇa)

3.

ਪ੍ਰਸਾਦ(prasāda)

ਰਸ਼ਵਪ੍ਰਸਾਦ, ਰਾਮਪ੍ਰਸਾਦ(śivaprasāda, rāmaprasāda)

4.

ਪ੍ਰਕਾਸ਼(prakāśa)

ਜੈਪ੍ਰਕਾਸ਼, ਓਮਪ੍ਰਕਾਸ਼(jaiprakāśa, ōmaprakāśa)

5.

ਰਕਸ਼ੋਰ(kiśōra)

ਨਿੰਦਰਕਸ਼ੋਰ,ਰਾਜਰਕਸ਼ੋਰ(nadakiśōra,rājakiśōra)

6.

ਕੁਮਾਰ(kumāra)

ਰਸ਼ਵਕੁਮਾਰ,ਰਾਜਕੁਮਾਰ(śivakumāra,rājakumāra)

7.

ਗੋਪ੍ਾਲ(gōpāla)

ਰਕਰਸ਼ਣਗੋਪ੍ਾਲ,ਰਾਮਗੋਪ੍ਾਲ(kriśaṇagōpāla,rāmagōpāla)

8.

ਚਿੰ ਦਰ(cadra)

ਰਾਮਚਿੰ ਦਰ,ਖੇਮਚਿੰ ਦਰ(rāmacadra,khēmacadra)

9.

ਰਜਿੰ ਦਰ(jidara)

ਰਾਰਜਿੰ ਦਰ,ਰਰਜਿੰ ਦਰ(rājidara,rajidara)

10.

ਜੇਂਦਰ(jēndara)

ਰਾਜੇਂਦਰ,ਬਰਜੇਂਦਰ(rājēndara,brajēndara)

11.

ਾਿੰਰਤਕਾ(tikā)

ਅਿੰ ਰਤਕਾ,ਅਵਿੰ ਰਤਕਾ(atikā,avatikā)

12.

ਪ੍ਰੀਤ(prīta)

ਰਦਲਪ੍ਰੀਤ,ਹਰਪ੍ਰੀਤ(dilaprīta,haraprīta)

13.

ਪ੍ਾਲ(pāla)

ਸੁਖਪ੍ਾਲ,ਸੁਰਰਿੰ ਦਰਪ੍ਾਲ(sukhapāla,suridarapāla)

14.

ਰਪ੍ਿੰ ਦਰ(pidara)

ਨਰਪ੍ਿੰ ਦਰ,ਹਰਰਪ੍ਿੰ ਦਰ(napidara,harapidara)

15.

ਪ੍ੁਰ(pura)

ਸਰਹਬਾਜ਼ਪ੍ੁਰ,ਫੁੱ ਤੇਪ੍ੁਰ(sahibāzapura,phatēpura)

20.

ਮੀਤ(mīta)

ਗੁਰਮੀਤ,ਪ੍ਰਮੀਤ(guramīta,paramīta)

21.

ਮੇਸ਼(mēśa)

ਰਮੇਸ਼,ਸੋਮੇਸ਼(ramēśa,sōmēśa)

22.

ਮੇਂਦਰ(mēndara)

ਧਰਮੇਂਦਰ,ਉਮੇਂਦਰ(dharamēndara,umēndara)

23.

ਮੋਹਨ(mōhana)

ਮਨਮੋਹਨ,ਰਾਜਮੋਹਨ(manamōhana,rājamōhana)

24.

ਰਿੰ ਜਨ(rajana)

ਨਰਿੰ ਜਨ,ਰਸਵਰਿੰ ਜਨ(narajana,śivarajana)

25.

ਰਜੀਤ(rajīta)

ਸੂਰਜੀਤ,ਇਿੰ ਦਰਜੀਤ(sūrajīta,idarajīta)

26.

ਾੇਾਦਰ(̔ēndara)

ਜੀਤੇਂਦਰ,ਸ਼ੈਲੇਂਦਰ(jītēndara,śailēndara)

27.

ਜੋਤ(jōta)

ਮਨਜੋਤ,ਨਵਜੋਤ(manajōta,navajōta)

28.

ਤਪ੍ਾਲ(tapāla)

ਪ੍ਰਨੀਤਪ੍ਾਲ,ਸੁਖਜੀਤਪ੍ਾਲ(pranītapāla,sukhajītapāla)

29.

ਤੇਜ(tēja)

ਗੁਰੂਤੇਜ,ਕੁਲਤੇਜ(gurūtēja,kulatēja)

30.

ਾਿੰਦਰ(dara)

ਸੁਰਜਿੰ ਦਰ,ਤੇਜਰਵਿੰ ਦਰ(sujidara,tējavidara)

[404]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Issues in Word Segmentation of Handwritten Text in Devanagari Script Rohit Sachdevaa, Dharam Veer Sharmab a

Department of Computer Science, M M Modi College, Patiala, India, Department of Computer Science, Punjabi University, Patiala, India

b

Abstract Word segmentation means dividing the word into sub-parts to extract identifiable units. In handwritten word recognition system, segmentation is a significant pre-processing step. The Indic scripts (such as Bangla, Devanagari, Gujrati, Gurumukhi, etc.) are cursive in nature, making it a challenging task to segment the word. As character set of Devanagari scripts consists of large number of characters further subdivided into consonants, half consonants compound characters, modifiers conjunct consonants and vowels etc. that makes it difficult to segment word Devanagari script. As every person has different method of writing, so it is challenging task to fragment the word into three zones- upper, middle, lower as compare to printed version. The variance in handwritten word is due to: different size of letters, spacing between characters, writing at different angle etc. At times, while writing, the characters tend to be merged, overlapped which adds to complexity of segmentation of these words. This is a general problem, which also occurs very often in other Indic Scripts. Due to the curved nature and inclination to write word at an angle, it is painstaking task to the segment the words of Devanagari script. This paper discusses the issues and problems related to segmentation of handwritten words written in Devanagari Scripts.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: OCR; Handwritten Word Segmentation; Devanagari Script

1. Introduction Text recognition is heavily dependent on appropriate segmentation of text into lines, words and then individual characters or sub-characters as well as feature extraction and classification of the individual characters. Correct segmentation is highly important because incorrect recognition happens if the segmentation has not been done properly. Cursive nature of a script makes the process more assiduous. The structural characteristics of Devanagari character set, curve nature and different ways of writing words makes handwritten word segmentation in Devanagari script onerous job. The core objective of this paper is to identify the complexities involved in the segmentation of hand written words of Devanagari script, to devise suitable solutions for the same in near future. The rest of the paper is organized as follows: section 2 discusses the work done in this area, section 3 describes the properties of Devanagari script, section 4 briefly discusses the method of segmenting hand written words, section 5 deals with the issues faced in segmentation of handwritten word, the papers findings have been summarized in the section 6 and references are given at the end. 2. Previous Work and Related Research Researchers have applied various different approaches for the segmentation. A lot of research has taken place in the field of segmentation of handwritten text in Arabic, Chinese, Japanese and Roman scripts and various methodologies have been postulated by various researchers in handwritten text recognition (Wang et al, 2000; Verma, 2003; Ariki,1995; Plamondon, Srihari, 2000; Chin, Han, 2000; Kim, Bang, 2000).

___________________________ * Corresponding author. Tel.: +91 8146 700347. E-mail address: [email protected].

ISBN: 978-93-82288-63-3

Sachdeva and Sharma/ COMMUNE-2015

During the recent years number of researchers has tried to recognize the off-line handwritten text of Indian scripts. A functional system exists for recognition of handwritten numerals and characters of off-line Bangla script Roy et al, 2005). For Indian postal automation a system has also been developed for unconstrained Bangla handwritten word recognition (Pal et al, 2006). Some research work has been reported for other Indic scripts including Gurmukhi Sharma, (Lehal, 2006), Oriya (Tripathy, Pal, 2006) and Devanagari (Gargl et al, 2010; Ramteke, Rane, 2012; Palakollu et al, 2012). N. K. Garg et al. (Gargl et al, 2010) suggested a technique for line segmentation of handwritten Hindi text. It is mutated form of their earlier proposed method with some presumptions associated with the height of consonant, lower modifiers and maximum height of consonant and skew between the two lines in a text. Their system was depends on detection of head line, base line and contour following method. A. S. Ramteke et al. (Ramteke, Rane, 2012) proposed a method which is implemented in Matlab for segmentation of offline handwritten Devanagari Script. The accuracy of segmentation for this method depends upon the proper writing i.e. proper association of characters with Shirorekha, decent space between words and characters, non-overlapping or characters. The accuracy of segmentation for character, word and numerical achieved by system is 97%, 98% and 100% respectively. Results for broken characters are not so good. S. Palakollu et al. (Palakollu et al, 2012) proposed new technique for segmentation of line and overlapping characters of Handwritten Hindi text. Initially the text is segmented into lines then lines are segmented into words. Later from these words, headline detected and transformed as straight line. Headline and base line is detected by reckoning the average line height and based on it. Before segmentation, skew detection and correction is needed for straightening the image for proper segmentation and feature extractions. N. K. Garg et al. (Gargl et al, 2010) introduced procedure for segmentation of handwritten Hindi text which is based on structural approach. The accuracy of segmentation for proposed procedure gives good results if headline and base line meticulously determined. The accuracy of segmentation for line, word, consonants, ascenders, descenders achieved by procedure is 91.5%, 98.1%, 79.12%, 95.5% and 82.6% respectively. Results for large skewed and touching lines are not great. Munish Kumar et al. (Kumar et al, 2014) illustrated a method for to identify and segment of touching characters in Gurumukhi script by using water reservoir method. By using reservoir base area point, the accuracy for character segmentation achieved by method is 93.51%. Results for broken and overlapped characters are not so virtuous. Preeminent mechanisms for skew detection encompass Correlation Method, Hough Transform, Fourier method, Projection profile, Historate Method, Nearest Neighbor (Postl, 1986; Hashizume et al,1986; Yan, 1993; Baird, 1987; Srihari, Govindaraju, 1989; Hinds et al, 1990; Sharma et al, 2009). Sharma et al. (Sharma, Lehal, 2009) gave the sturdy method, which detect the skew and corrected the isolated words of machine printed Gurumukhi papers. According to authors, if isolated words have straight headline then it is not considered as skewed but when length of headline is less than a threshold value then it is considered as skewed word and need to be corrected. Scripts in which word contain headline used to connect characters, this method is equally adequate for machine printed documents. This method is adequately practiced on Bangla, Devanagari, and Gujarati script words as these have the same structural characteristics as Gurmukhi Script. 3. Characteristics of Devanagari Script Devanagari is part of Brahmi family of scripts of India, which are used in Nepal, Tibet, and south East Asia. Devanagari script has 34 consonants, 12 vowels, 14 modifiers of vowels. It also has compound characters, which are, compose by combining two or more basic characters. As Devanagari script is phonetic and syllabic script so words are written literally as they are speaks. Another typical feature of Devanagari is the existence of a horizontal line on the top of all characters. This line is known as headline or Shirorekha. This line divides the word into two parts. The upper part consists, mostly, of vowels (matras) and modifiers. 3.1

Data Collection:

To analyse the complexities of the handwritten text a form is designed which contains 25 words. Databases constructed by taking data from 200 users it contain a set of 5000 isolated words. These forms are scanned at 300 dpi resolution. In the database, words contained different sizes of characters, touching characters, overlapped characters and slanted characters. In pre-processing step normalization is performed on the data set. Fig. 1 shows some samples of the handwritten words collected from different users.

Fig. 1 Example of handwritten words in Devanagari script

4. Handwritten Word Segmentation in Devanagari Script Handwritten words can be classified into following categories:

[406]

Sachdeva and Sharma/ COMMUNE-2015

a. b. c. d. e.

Words with headline (Fig. 2.a) Words with slanted headline (Fig. 2.b) Words with broken headline (Fig. 2.c) Words without headline (Fig. 2.d) Words with uneven headline (Fig. 2.e)

For segmenting words to individual characters, headline detection is crucial step. In handwritten text, due to skew in the word of writing style of the user, headline may not be a straight line. By using Horizontal Projection Profiles (HPPs) existence or absence of headline can be identified. If the headline is not determined because of its absence then vertical Projection Profiles (VPPs) are constructed to determine gaps for segregation. Figure 3 shows some example of handwritten word’s headline detection.

Fig 3 shows the headline detection using HPPs

If headline is determined, then the whole image is prorated horizontally into upper zone and middle-lower zone. Upper zone part is over the headline and the middle-lower zone under the headline. The determined headline is stripped, to detach the characters in upper and middle-lower zone. This makes gaps in the characters of the word, which are segmented by constructed Vertical Projection Profiles (VPPs). For upper zone characters, deal with the area just over the headline as the bottom of upper zone and creates VPPs. Each gap in the VPPs shows the cut point for segmentation of upper zone characters. Upper zone contains all the vowels modifiers (matras) above the headline. Middle-lower zone is further divided into two sub zones: middle zone and lower zone. Middle zone characters are determined using Vertical Projection Profiles (VPPs), constructed by considering the area just under the headline as the top starting point. Words may contain some broken characters to avert over-segmentation a distance threshold value is used. This step gives us the number of columns constructed in the word, which helps in relating characters in the upper and lower zone with characters in the middle zone. Middle zone characters segmentation is accomplished in three stages. Under first stage the word is segmented, words which are under-segmented are more segmented to the maximum possible limit in the second stage and in the third stage over segmentation is managed resulting from the first stage, which mainly occurs because of broken characters in handwritten text. From the lower part, just under the headline, again constructed Horizontal Projection Profiles and from the bottom move upwards to locate each gap in the HPPs. Existence of any gap in the HPPs shows existence of characters in the lower zone. If no gaps are located to a threshold then either the lower zone characters are overlapped or connected with the middle zone character or not present in lower zone. Figure 4 shows word divided into three zones.

Fig 4 Word divided into three zones

[407]

Sachdeva and Sharma/ COMMUNE-2015

5. Word Segmentation of Handwritten Text in Devanagari Scripts- Issues and Problems 5.1. Problem of Skewed Words: For proper segmentation of a word, the image should be free of any skewness. Some examples of skewed words written in Devanagari script are shown in Figure 5.

Fig 5: Shows skewed words written in Devanagari script

There are several proposed approaches as options for skew angle detection of document images. All the approaches need a rich text area to be present in order to work accurately. Rich text areas carry a well-known characteristic structure, one or more (separate) lines of printed or handwritten words, sharing a common direction. The method proposed in (Hinds et al, 1990) has been used for skew angle detection and correction. Results of applying this algorithm on skewed images are shown in table 1. Table 1: Comparison of word before and after skew correction

Skewed Word

After Skew Correction

5.2. Headline Detection and Removal for segmentation For segmenting words to extract individual characters, headline detection is necessary. Existence or absence of headline can be identified by using Horizontal Projection Profiles (HPPs) as discussed in section 4. This can also be done by locating the rows which have maximum number of black pixels in a word. After the detecting headline we stripped it so as to divide the word vertically into two parts. Table 2: Comparison of word before and after skew correction

Skewed Word Headline detection

After Skew Correction Headline detection

5.3. Overlapping, Connected, Merged or Fused and Broken Characters Segmentation Middle zone contains the overlapped, connected, merged or fused and broken characters. Segmentation of these types of words is a challenging job. Figure 6 shows some examples of these types of words:

(a) Fig 6: (a) Words with overlapping characters

[408]

Sachdeva and Sharma/ COMMUNE-2015

(b)

(c)

(d) Fig 6: (b) Words with connected characters (c) Words with fused or merged characters (d) Words with broken characters

Presence of these types of complexities makes the segmentation task more challenging as shown in Figure 7.

Fig 7.1 Overlapped characters returned two characters as single character (under-segmentation)

Fig 7.2 Connected characters returned two characters as single character (under-segmentation)

Fig 7.3 Fused characters return fused characters as a single character (under-segmentation)

Fig 7.4 Broken Characters returns single character in two characters (over-segmentation)

6. Conclusion If the words are skewed then with the help of skew detection and correction algorithm, the headline detection becomes easy. Then the word is segmented into three zones using Horizontal Projection Profiles (HPPs). Zone wise characters are recognized by using Vertical Projection Profiles (VPPs). Upper zone and lower zone character identification is uncomplicated as compare the characters in middle zone. Middle zone characters segmentation becomes more complex if it contains the overlapped, connected, merged or fused and broken characters words. This eliminates the vertical gaps between the characters and further segmentation is done using cut-classification or recognition based segmentation References Wang, X., Govindaraju, V. and Srihari, S. N., 2000. Holistic Recognition of Handwritten Character Pairs, Pattern Recognition, Vol. 33, pp. 19671973. Verma, B., 2003. A Contour Code Feature Based Segmentation for Handwriting Recognition, in the proceedings of 7th International Conference on Document Analysis and Recognition, pp. 1203 – 1207. Ariki, Y. and Mot, Y., 1995. Segmentation and Recognition of Handwritten Characters using Subspace Method, in the proceedings of 3rd International Conference on Document Analysis and Recognition, Vol. 1, pp. 120-123. Plamondon, R. and Srihari, S. N., 2000. On-Line and Off-line Handwritten Recognition: A Comprehensive Survey, IEEE Transaction on Pattern Analysis and Machine Intelligence, Vol.22, pp.62-84. Chin, Z. and Yan, H., 2000. A Handwritten Character Recognition using Self-organizing Maps and Fuzzy Rules”, Pattern Recognition, Vol.22, pp. 923-937.

[409]

Sachdeva and Sharma/ COMMUNE-2015 Kim, K. and Bang, S.Y., 2000. A Handwritten Character Classification using Tolerant Rough Set, IEEE Transaction on Pattern Analysis and Machine Intelligence, Vol.22, pp.923-937. Roy, K., Pal, U. and Kimura, F., 2005. Recognition of Handwritten Bangla Characters, in the proceedings of 2nd International Conference on Machine Intelligence, pp.480-485. Pal, U., Roy, K. and Kimura, F., 2006. A Lexicon Driven Method for Unconstrained Bangla Handwritten Word Recognition, in the proceedings of 10th International Workshop on Frontiers in Handwriting Recognition, pp.601-606. Sharma, D. V. and Lehal, G. S. 2006. An Iterative Algorithm for Segmentation of Isolated Handwritten Words in Gurmukhi Script, in the proceedings of 18th International Conference on Pattern Recognition, Vol. 2, 1022-1025. Tripathy, N. and Pal, U. 2006. Handwriting Segmentation of Unconstrained Oriya Text, Sadhan, Vol. 31, pp. 755–769. Garg, N. K., Kaur, L. and Jindal, M. K., 2010. A New Method for Line Segmentation of Handwritten Hindi Text, in the proceedings of 7th International IEEE Conference on Human Technology: New Generations (ITNG), pp. 392- 397. Ramteke, A. S. and Rane, M. E., 2012. Offline Handwritten Devanagari Script Segmentation, International Journal of Scientific and Technology Research, Vol. 1, Issue 4, pp. 142-145. Palakollu, S., Dhir R. and Rani, R., 2012. Handwritten Hindi Text Segmentation Techniques for Lines and Characters, in the proceedings of the World Congress on Engineering and Computer Science, Vol. 1, pp. 640-644. Garg, N.K., Kaur, L., and Jindal, M. K., 2010. Segmentation of Handwritten Hindi Text, International Journal of Computer Applications (0975 – 8887), Volume 1 – No. 4 pp. 19-23. Kumar, M., Jindal, M. K. and Sharma, R. K., 2014. Segmentation of Isolated and Touching Characters in Offline Handwritten Gurmukhi Script Recognition, International Journal of Information Technology and Computer Science, Vol. 02, pp 58-63. Postl, W. 1986. Detection of Linear Oblique Structures and Skew Scan in Digitized Documents, in the proceedings of International Conference on Pattern Recognition, pp. 687-689. Hashizume, A., Yeh, P. S. and Rosenfeld, A., 1986. A Method of Detecting the Orientation of Aligned Components, Pattern Recognition Letters, Vol. 4, pp. 125-132. Yan, H., 1993. Skew Correction of Document Images Using Interline Cross-correlation, Computer Vision Graphics and Image Processing: Graphical Models and Image Processing, Vol. 55, Issue 6, pp.538-543. Baird, H. S., 1987. The Skew Angle of Printed Documents, in the proceedings of Conference on Photographic Scientists and Engineers, pp. 14-21. Srihari, S. N. and Govindaraju, V., 1989. Analysis of Textual Images using the Hough Transform, Machine Vision and Applications, Vol. 2, pp. 141153. Hinds, S. C., Fisher, J. L. and Amato, D. P. 1990. A Document Skew Detection Method using Run-length Encoding and the Hough Transformh transform, in the proceedings of International Conference on Pattern Recognition, pp. 464-468. Sharma, D. V., Gupta, S. and Beri, P., 2005. Skew Angle Detection and Correction of Hand Written Gurmukhi Words using Historate Method, in the proceedings of the International Conference on Cognition and Recognition, pp. 22-24. Sharma, D. V. and Lehal, G. S., 2009. A Fast Skew Detection and Correction Algorithm for Machine Printed Words in Gurmukhi Script, presented in an International Workshop on Multilingual OCR, published by ACM, article no 15.

[410]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Remote Monitoring of Water Pollution in Dal Lake using Wireless Sensor Networks in Realtime Sofi Shabir, Roohie Naaz National Institute of Technology, Srinagar, 190006

Abstract Pollution monitoring in water is an important part of environment monitoring. The population boom, high urbanization levels and the birth of new industries are producing adverse effects on water bodies and leading to pollution of water and on environment in general. Due to the constraints of the natural conditions and time and temporal factors, the traditional monitoring methods have some limitations. In recent years remote sensing technologies were widely applied in the investigation and monitoring of various pollutions in aquatic environments. Field investigation is reliable but too expensive and time consuming. Satellite remote sensing have the advantages of region scale coverage and moderate timeliness but restricted by the factors of heavy mist, low image resolution and prohibitive cost. For better accuracy and spatial coverage Wireless Sensor Networks can be promising approach with real time monitoring of water pollution and aquatic life. In this paper we analyse the different techniques used in water pollution monitoring and propose a cost effective new technique for the monitoring of water bodies based on Wireless Sensor Networks with transferring of the data collected in real time and viewing of the same via a network. The GSM model of the same has been tested in the Lab and is working satisfactorily.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Wireless Sensor Network ; Dissolved Oxygen ; Wi-Max Technology;Gateway; GSM;Pollution, Environment

1. Introduction Water is the essential resources for all the living organics of the earth to survive. The quality of life on earth is linked inextricably to the overall quality of the environment. Nevertheless, water is suffering from various pollutions due to thousands of reasons. The major reasons are population boom; high urbanization levels and the birth of new industries producing adverse effects on water bodies and leading to pollution of water and on environment in general. Natural phenomenon such as volcanoes, algal blooms, storms and earthquakes also cause major change in water quality and ecological status of the water. Since water environment monitoring is an important part of environment monitoring, therefore it is of great importance to develop an effective water pollution monitoring system. Due to the constraints of the natural conditions and time and temporal factors, the traditional monitoring methods have some limitations. Field investigation is reliable but too expensive and time consuming. In recent years remote sensing technologies were widely applied in the investigation and monitoring of various pollutions in aquatic environments. But they are also not accurate and limited by many factors especially they fail to provide the realtime monitoring. A new approach is the need of the hour and one of the promising technologies is the wireless sensor networks. 2. Traditional approaches The traditional water inspection method is called offline inspection approach (Qingbo and Jianzhong, 2006). The inspectors collect sample water from the monitored area and bring back to the laboratory for analysis. With this method, 

Corresponding author Tel.: +91 9419 009971. E-mail address: [email protected]

ISBN: 978-93-82288-63-3

Sofi and Naaz/COMMUNE – 2015

the inspection circle is long sometimes it takes a few days to get the result, moreover the analysis result is limited to sampling area and it is unable to carry out realtime monitoring in a big range (Jung and Lee, 2008). The different chemical parameters which are essential indicators of pollution level of the water body are given in table below. Each parameter is having its threshold value in water which should not be allowed beyond a certain limit. For this purpose samples are taken to lab and the analysis is done. Chemical Parameters

Reason for the Analysis

Temperature

Temperature can exert great control over aquatic communities. If the overall water body temperature of a system is altered, an aquatic community shift can be expected.

pH value

pH is an indicator of the existence of biological life as most of them thrive in a quite narrow and critical pH range. DO is essential for aquatic life. A low DO (less than 2mg/l) would indicate poor water quality and thus would have difficulty in sustaining many sensitive aquatic life. Conductivity indicates the presence of ions within the water, usually due to in majority, saline water and in part, leaching. It can also indicate industrial discharges.

Dissolved Oxygen (DO)

Conductivity

3. Remote Sensing The development and maturation of remote sensing technology has made a great breakthrough on testing for the environmental pollutants and has solved dozens of practical problems. The remote sensing technology has wide monitoring range; high speed, low cost and being convenient to make long term dynamic monitoring as compared to traditional approach (Sun et al, 2010). 4. Polarized Remote Sensing Using multi-band, multi-date and hyper-spectral remote sensing data, the ability to identify surface features increases. It is found that the angle information has made great influence and contribution on the distinguish and classifying of remote images (Fenghui et al, 2009) which is called three dimensional spectral characteristics of surface features in 2π space. The polarized remote sensing information can show richer content in expressing dark objects, it provides a new effectual measure for water remote sensing (Yu et al, 2010). Polarization makes great sense of expressing the nature of water and the nature of material in water. Thus polarized remote sensing technology can be used as a kind of supplementary means so that it could play its role in water quality monitoring. Satellite remote sensing have the advantages of region scale coverage and moderate timeliness but restricted by the factors of heavy mist, low image resolution and prohibitive cost and accuracy. For better accuracy and spatial coverage Wireless Sensor Networks can be promising approach with realtime monitoring of water pollution and aquatic life. 5. Wireless Sensor Networks (WSN) Advances in electronics and wireless communications have enabled a new evolution in wireless sensor networks. Over the last 5 years WSN have attracted a great deal of interest due to their cost-effectiveness, ability to perform multiple functions simultaneously and make decision based on information gathered from various sensing elements placed at different locations. A wireless sensor network consists of a number of nodes which are low costs, with wireless communication, sensing, and data processing capabilities which can be distributed across a geographical area. A sensor node consists of information collection module, responsible for monitoring the area of information collection and data conversion; information processing module, responsible for controlling the operation of the sensor nodes, storage and processing their own data collection and data sent by the other nodes; information exchange module, with other sensor nodes for wireless communication, exchange control messages and send and receive data collection; system power module for the sensor nodes to provide the energy required to run, usually in a miniature battery. The sensor part of the node is responsible for the physical sensing of the parameter such as temperature, ph value, dissolved Oxygen etc. The parameter list as given in the table above can be more to depending upon the availability of the sensors. Only the sensor part is changed for the WSN node rest of the components remains same as shown in fig 1. remains same.

[412]

Sofi and Naaz/COMMUNE – 2015

Fig 1: Sensor Node Architecture

A wireless sensor network is kind of ad hoc network which is easily configured, without infrastructure, such as cables or structures. Since a sensor node is low priced and can be placed in a large area of water. The cost of deployment is less. Also many sensor nodes could monitor the same area at the same time through compressed configuration so that the inspection result will be more accurate through redundancy data analysis, the inspection period is short and the inspection is in realtime. Proposed Wireless Sensor Network for water monitoring Architecture:

Fig 2: Proposed Architecture

The architecture of the proposed water monitoring system is shown in fig. 2. Wireless sensor nodes are deployed at various locations of the water body depending on the need and parameters to be monitored. The nodes can have ability of monitoring one or many parameters at a time. The topology for the network is cluster based. Each group of the nodes [413]

Sofi and Naaz/COMMUNE – 2015

is having a cluster head. All the nodes in the cluster are communicating with the cluster head for onward transmission to the gateway. The cluster head can be static or dynamic. In static case the cluster head is predefined and it knows the complete topology of the network beforehand. In case of the dynamic cluster head each node may get a chance to become cluster head depending upon the energy level of each node. The node with higher energy level will become the cluster head. In the later case the network life will be improved (shabir and Naaz, 2010). In our case we are using a topology based on static cluster head. The cluster head will be without the sensing part and having a higher energy level than the individual nodes [dummy node] and having higher transmitting range with the capability for the communicating with the nodes using Zigbee technology and with the gateway using Wimax technology (Silva et al). The gateway in turn can communicate with the nearest base station and the base station in turn is communicating with WSN Data Server. The server is having the necessary software to analyse the data received and then the processed data is stored in a database. The database is then connected to the internet and the users can access the data in realtime irrespective of the place. Since the monitoring is not only required at one place of the water body but we need to establish many such clusters with a separate gateway for each cluster. The base station can be one or many depending on the area of the water body and the range of the base station. The nodes may directly communicate with the base station or the data may be routed through the other nearby nodes in case the sending node is out of range. Again the gateway may communicate the data to be sent to the base station via another nearby gateway in case it is out of range. 6. Conclusion Due to the constraints of the natural conditions and time and temporal factors, the traditional monitoring methods have some limitations. Satellite remote sensing have the advantages of region scale coverage and moderate timeliness but restricted by the factors of heavy mist, low image resolution. For better accuracy and spatial coverage, Wireless Sensor Networks can be promising approach with realtime monitoring of water pollution and aquatic life. The proposed architecture can use the available technologies of communication with the power of physical monitoring of the environment around us by the tiny WS nodes. The architecture has the limitation that it can be useful to medium sized water bodies like lakes etc. only. There are many challenges which include: deployment, topology control, battery recharge, device design etc. All these can be taken for future work. References Jiang Peng, Huang Qingbo and Wang Jianzhong Research on Wireless Sensor Networks Routing Protocol for Water Environment Monitoring 07695-2616-0/06 2006 IEEE. JY Jung and JW Lee, “ZigBee Device Access Control and Reliable Data Transmission in ZigBee Based Health Monitoring System,” ICACT 2008. 10th International Conference on Advanced Communication Technology, vol. 1, pp. 795-797, February 2008. Sofi shabir and Roohie Naaz “ Dummy Node based Design of Energy Efficient Wireless Sensor Network” IET 2010 Steven Silva , Hoang N ghia Nguyen , Valentina Tiporlini and Kamal Alameh Electron Science Research Institute Web Based Water Quality Monitoring with Sensor Network: Employing ZigBee and WiMax Technologies Yang Yu, Xinhua Li and Liang Chen “The Application of Wireless Sensor Networking in Environmental Monitoring Based on LEACH Protocol”IEEE 2010 Zhang Fenghui, Zhou Huiling and Zhou Xiaoguang, “A Routing Algorithm for ZigBee Network Based on Dynamic Energy Consumption Decisive Path,” CINC’09. International Conference on Computational Intelligence and Natural Computing, vol. 1, pp. 429-432, June 2009. Zhongqui Sun, Yusheng Zhao and Shaoping Li “Research on polarized remote sensing of monitoring of water pollution “ IEEE 2010

[414]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Machine to Machine (M2M) Control & Communication for Internet of Things (IoT) using DTMF G. Mohiuddin Bhat, Rouf-ul-Alam Bhat, Naazira Badar, Malik Rabaie Mushtaq, Afzan Hussain Hamdani Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar, India

Abstract Dual Tone Multiple Frequency (DTMF) is a low cost, global wireless control technology used in diverse application areas. Although a lot of work has been done on the applications based on DTMF, major portion of the research and development has been carried out to develop systems that involve a manual data entry at the transmitter side and an automatic response at the receiver side e.g. the Interactive Voice Response (IVR) System that creates a lot of scope for human to machine (H2M) interaction. With the advent of Cloud centric Internet of Things (IoT) technology, there is a need to design and develop a low cost, globally accessible machine to machine (M2M) interaction technology. In this paper, a system is proposed and implemented that makes the DTMF technology two way interactive and autonomous so that it can be used for M2M IoT market. Although a class of data sources can be used, the experimentation has been performed through a test bed where an image processing application is automatically generating control signals to be relayed over Global System for Mobile Communications (GSM) to a remote system thus emulating an M2M system

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: IOT; M2M; GSM; DTMF; H2M

1. Introduction Kevin Ashton was the first person to coin the term Internet of Things in 1999. He introduced this term in reference to the concept of supply chain (K. Ashton, 2009) management. However, this definition has been more comprehensive in the past decade covering broader range of applications such as health care, utilities and transport (H. Sundmaeker et al., 2010). Although with the evolution of technology, the definition of ‘Things’ has changed, the goal of prime importance still remains to make a computer sense information without the aid of human intervention. Internet of Things (IoT) is expected to be a huge leap radically evolving the current Internet into a network of interconnected objects that not only takes information from the environment via the sensing module and interacts with the physical world via the actuation, command and control modules, but also uses the status quo Internet standards to provide various services like information transfer, communications, analytics and applications. Strongly propelled by the prevalent open wireless technologies like Bluetooth, radio frequency identification (RFID) and Wi-Fi as well as embedded sensor and actuator nodes, IoT has stepped out of its initial stages and is on the verge of transforming the current static Internet into a fully integrated Future Internet (J.Buckley, 2006).The propagation of mobile Internet has resulted in ubiquitous mobility and nationwide coverage. The cost of broadband data service provided by today's advanced wireless networks is significantly less than in the past due to extensive standardization (Surobhi et al., 2014). Machine to machine (M2M) broadly refers to any technology that helps networked devices in exchanging information and makes these capable of performing tasks without the manual human assistance. M2M is an indispensable part of the IoT and is very beneficial for the industry and business in general.

_________________________________ * Corresponding author. Tel.: +91 9596 088846 E-mail address: [email protected].

ISBN: 978-93-82288-63-3

Bhat et al/ COMMUNE-2015

The applications of M2M spread over a large extent in areas like industrial automation, logistics, Smart Grid, Smart Cities, health and defence mostly for monitoring but also for control purpose. Technology is one of the major driving forces of M2M.The growing semiconductor industry provides better miniaturization along with improved yield, thereby continuing to reduce cost and power consumption per chip. The coverage of wireless networks can be further extended employing the use of peer-to-peer communication, relays and small cells while drastically reducing the cost per transmitted bit. The economic reasons are also fervent motivations for the wireless industry to adopt M2M. The deteriorating voice revenue has put operators are under massive pressure to introduce new services to fill this revenue gap. Cloud computing, M2M and application stores lead the list of potential revenue-generating services (Geng Wu et al.,2011).Based on the definition given by the European Telecommunications Standards Institute (ETSI), M2M architecture can technically refer to a variable number of communicating machines. However, it is generally accepted that M2M principles are valid particularly well for networks where a large number of machines is used, even up to the estimated 1.5 billion wireless devices of the future. This means that when an M2M application is discussed, it is presented on a global scale with a profusion of centrally coordinated sensors (M.J. Booysen et al., 2009). 2. Background DTMF is a very reliable means of signaling and is being used largely for global communication and control. Although a number of protocols have been developed for wireless communication, these are limited by the short range (Nasim uz zaman Chowdhury and Md. Khaled, 2013). Employing the use of DTMF, this limitation can be overcome. DTMF can be transmitted over telephone lines as well as over the Internet. Whenever a key is pressed on the DTMF encoder, a DTMF tone is generated and transmitted. This DTMF tone is decoded at the receiving end and used for practical applications related to communication and control. In case of mobile phones, DTMF tones can be generated only after the connection is established. Most of the applications employing DTMF for control use a GSM phone connected to a cellular network for transmitting signals. The controls are provided by manual pressing of the keypad by a human. These DTMF-based systems have the following drawbacks: ● The speed of control is limited by the rate at which a human can press the buttons of the DTMF keypad. ● These designs are costly. ● These designs require more power. In this research paper, we have focused on the proposed system to make both the encoder and decoder modules of our DTMF - controlled robot fully automated and low cost employing the M2M technology unlike the H2M based control architectures of other robots. The devices that constitute the Internet of Things (IoT) have a variety of low power microcontrollers, sensor networks and communication protocols at the core. These underlying technologies operate in a fast and dynamic environment providing services like tele-monitoring, M2M communications and motion control. Microcontrollers are monitoring and controlling power close to the load point, providing better device management. When low power design architectures combine with manufacturing designed to be power sensitive, the total power budget and costs are dramatically reduced. Moreover, with the use of a GSM based technology, we have achieved a global control. 3. Problem Definition A lot of work has been done on DTMF technology and its efficient implementation, but most of the advancement in the DTMF has been done at the decoder end. Human intervention is still required at the encoder end which restrains the system from being a fully automatic M2M system. Moreover, the conventional systems are costly and have large power requirements. 4. Proposed Solution – Design And Implementation In this paper, a system is proposed and implemented that makes the DTMF technology two way interactive and autonomous so that it can be used for M2M IoT market. The block diagram in Fig. 1 describes the overall functionality of our proposed design. To emulate an M2M device, an image processing application has been designed that senses color information using the image acquisition device and converts it into a set of codes. The acquired data is fed to a computing device with MATLAB installed on it. MATLAB has been programed to generate serial data corresponding to the input. The serial data are interfaced with the processing module consisting of an Atmega328 microcontroller based Arduino UNO board using an RS-232 cable. The Atmega328 has been programed to do serial to parallel conversion the parallel data act as control signals for the DTMF encoder module corresponding to which different DTMF tones are generated which are fed to the communication network (GSM) through a communication module. The communication module acts as an interface between the encoder module and a GSM mobile handset that has been used as a relaying element such that these tones reach the receiving station over the GSM link. The backbone of encoder module is the DTMF encoder IC-UM91214B and a CMOS analog switching ICCD4066BE. Two CD4066BE ICs have been used for producing a total number of eight DTMF tones. The parallel data [416]

Bhat et al/ COMMUNE-2015

generated by the processing module are the controlling inputs for the switches of these ICs. The outputs of these switches are given to DTMF encoder IC which generates a particular tone for a particular input combination of low and high group frequency which is the basis of DTMF operation. The frequency combination table is shown in Table 1. Table 1. Frequency combination table High group frequencies

1209

1336

1477

1633

697

1

2

3

A

770

4

5

6

B

852

7

8

9

C

941

*

0

#

D

Low group frequencies

Fig. 1. Block diagram of the proposed architecture

The communication module consists of a microphone jack that acts as a tone-handset interface and has been primarily designed for taking the input from condenser or capacitive microphones. The following two design approaches have been considered for the same: ● Use of acoustic coupling: In this approach, the output from the encoder is directly fed to a speaker which is acoustically coupled with a microphone. This approach resulted in an advantage of providing a better isolation but had lesser speed. The maximum number of tones that could be transmitted per second was two.

Fig. 2. Acoustic Coupling Device

● Use of pi attenuator: In this approach, the output of encoder is fed to microphone input through an attenuator. The attenuator is needed because the incoming signal from encoder is of the order of hundreds of millivolts but the maximum input that can be fed to the microphone jack is of the order of few hundred microvolts.This approach although lacked in proper isolation but resulted in better speed and performance than acoustic coupling. Hence a pi type attenuator has been used in the prototype with a low pass filter for filtering out the high frequency noise. The prototype M2M transmitter for our proposed design is shown in Fig. 4. While as The prototype M2M receiver for our proposed design is shown in Fig. 5.

[417]

Bhat et al/ COMMUNE-2015

5. Results Out of all the possible frequency combinations shown in Table 1, only eight have been used for our prototype. For each of the eight combinations, a DTMF tone was generated. The resultant tones were viewed on the Digital Storage Oscilloscope as shown in Fig. 6.

Fig. 3. Attenuator

Fig. 4. Prototype M2M transmitter

Fig. 5. Prototype M2M receiver

(a)

(b)

[418]

Bhat et al/ COMMUNE-2015

(c)

(d)

(e)

(f)

(g)

(h)

Fig. 6. (a) tone produced corresponding to Frequency Combination 1209 Hz with 697 Hz, (b 1336 Hz with 697 Hz, (c) 1477 Hz with 697 Hz, (d) 1209 Hz with 770 Hz, (e) 1336 Hz with 770 Hz, (f) 1477 Hz with 770 Hz, (g) 1209 Hz with 852 Hz

For each of the tone shown in Fig. 6, the resultant tone amplitude is 500-800mV. 6. Conclusion The paper presents the proposed architecture for the M2M based DTMF encoder design. The results clearly reveal that the DTMF tones can be faithfully generated without manually pressing the keypad of mobile phone eliminating the human intervention. Two design approaches have been considered for interfacing the tone with the microphone. The piattenuator approach resulted in better performance and has been implemented in the final design. The proposed architecture should help in shifting the DTMF encoder design from H2M technology to M2M technology for the IoT. References K. Ashton, 2009. That ‘‘Internet of Things’’ thing, RFiD Journal. H. Sundmaeker, P. Guillemin, P. Friess, S. Woelfflé,March 2010. Vision and challenges for realising the Internet of Things, Cluster of European Research Projects on the Internet of Things—CERP IoT. J. Buckley, 2006. The Internet of Things: From RFID to the Next-Generation Pervasive Networked Systems, Auerbach Publications, New York. N.A. Surobhi, A. Jamalipour, 2014.A Context-Aware M2M-Based Middleware for Service Selection in Mobile Ad-Hoc Networks, IEEE Transactions on Parallel and Distributed Systems, p. 3056-3065. Geng Wu,S. Talwar, K.Johnsson,N. Himayat , K.D. Johnson,April 2011. M2M: From mobile to embedded internet, Communications Magazine, IEEE, p.36-43. M.J. Booysen, J.S. Gilmore1, S. Zeadally, G.J.van Rooyen,Feb. 2012. Machine-to-Machine (M2M) Communications in Vehicular Networks/ KSII Transactions on Internet and Information Systems, p.529-546. Nasimuzzaman Chowdhury, Md. Khaled Hussain, 2013. M2M: GSM Network for Robots using DTMF, The Global Journal of Researches in Engineering, p.23-29.

[419]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Framework for Web Security Using Multimedia Password Authentication Manzoor Ahmad Chachoo, Farah Fayaz Quraishi, Summera Ashraf * Department of Computer Sciences , University of Kashmir , Srinagar , Kashmir, India

Abstract Various studies over a long period give sufficient proof that humans are better at remembering images compared to text. By taking into the account the above fact , the team members strived to bring such a system (Multimedia Password Authentication) that is more secure, supports memorability and has a user friendly interface and uses the memory trigger connected to previously seen image. For enhanced security and efficiency, various sensitivity levels viz. Easy, Medium, Hard were introduced. The relation between security and usability is an anomaly that is not grasped in the current systems(mostly alphanumeric).The research is focused on whether the multimedia passwords(images along with audio track and video clips) deal with important issues like memorability and the ever important security factor at the same time while being practical in usage. To aid in remembrance of passwords (if forgot), images are used in place of traditional security questions for enhanced security. In addition, user is allowed to provide for Password Hint, in which text is converted to speech. The converted text which can be referred to as the sound at this point of time provides the user with the convenience of recovering forgotten passwords.Upon usage of the multimedia password authentication system, a significant increase was seen in the performance of the system regarding important benchmarks such as accuracy and speed. This significant increase in various important aspects of the system was seen independent of the time consumption of the user, which usually took longer than alphanumeric ones.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Information Security, Multimedia passwords, Image Click points, Audio Tracks, Video Clips, Sensitivity Levels

1. Introduction Patrick and Dourish (2003-04 respectively)stated that security is fundamentally a human-computer interaction(HCI)problem. The human-computer interaction is extremely important in two ways 1) Usability of the security mechanisms. 2) Interaction of security mechanisms with the user. The usability or efficiency of any given password depends on the authentication system software and how the information regarding that particular password is stored. Alphanumerical passwords are used to meet two quite opposite aspirations of the user. First of all, they must be easily recalled by a user while at the same time they have to be difficult for a hacker to crack. The general perception regarding the alphanumerical passwords is the usage of words and numbers that can be recalled by the user quite effortlessly (S. Chiasson et al.2007, 2008, 2009, E. Stobert et al, 2010 G.Agarwal, 2010) The dichotomy of using a particularly hard to remember password is that user has to write them down somewhere making them susceptible to hackers rather easily (Adams & Sasse, 1999). (G.Agarwal et al, 2010, Pinks, B. and T. Sander. 200,2). Biometric systems which are supposed to relieve the user of the above dilemma have their own disadvantages. (L. Jones et al , 2007, L. O‟Gorman, 2003, A. Jain, et al, , 2006). As compared to the biometric systems, graphical passwords offer or present a more practical alternative and are the focus of this paper. Blonder defined the graphical passwords in 1996.The Image Based Authentication rests on cognitive ability of association based memorization of humans compared with traditional textual passwords(Zhi Li et al, 2005). In addition to images, audio tracks and video clips have been used in the project to increase the remembrance of passwords. *

Corresponding author. Tel.: +91 9797 847507. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Chachoo et al/COMMUNE – 2015

Multimedia Password Authentication System (MPAS) is an alternative to solve the hardships that come to the fore regarding text based or graphical password system. Multimedia passwords are comparatively tougher to crack either by plain guessing or by using search of any kind. The potential click points in the images, audio clips and videos is rather large, therefore the presumable password area of MPAS scheme tend to overtake text based schemes and hence offer a more durable resistance to attacks. This paper tries to show a fresh, more adaptable and a relatively higher secure password system that we have ventured to design and consequently implement. In addition, we compare the traditional password systems to the multimedia authenticated system. The principal research questions that lead to the development of MPAS are stated below: RQ1: What kind of passwords is required by the user for multiple accounts RQ2: How does a user remember passwords for either a single account or multiple accounts? RQ3: What kind of limitations a user faces during password recovery process? 2. Current Authentication Methods The general classification of the current authentication methods can be expressed under following three headings: a) Token- based Authentication(e.g., Credit cards ) b) Biometric based Authentication(e.g., Iris scan, Facial recognition) The most disappointing aspects of the above two authentication methods is that they are quite expensive and demand special instruments. c) Knowledge based Authentication.(Two types—Text based and Picture based) Text Based: Due to their ease of use textual or text passwords are the go-to choice for humans. As the human capacity to remember is rather limited, the passwords chosen are rather easy to remember. The conventional passwords patterns and schemes are susceptible to various attacks. Keeping this fact in mind it is highly essential that tougher passwords are used for important accounts and the habit of changing the passwords regularly should be encouraged. Picture Based:The inherent problem of using an image as a password is that the image can be easily predicted and hence susceptible to be used by unwanted sources. Assumption is that the location of the user’s three gestures is all chosen independently. In practice, this is not realistic; there will often be some pattern. For instance, in the example the user chose to tap on three windows in sequence .(Fig 1)

Fig. 1 If you’ve guessed that the first tap location is over a window, then it would be natural to guess that maybe the next two are on some other windows, too.

2. Smudge Attacks Past researches have shown if someone gets hold of your laptop/phone, they could guess your picture password by looking at the pattern of smudge marks on the touchpad left by your finger oils (Adam J et al, 2010) (Amazingly, they found that the smudge marks remained clearly visible even if the user put the phone in their pocket -you might expect this would wipe the fingerprints off, but nope, they remained visible! An attacker can guess the picture password and significantly reduce the entropy in the password. For instance,

[421]

Chachoo et al / COMMUNE-2015

suppose the attacker gets lucky and there are only 3 smudge marks on the screen. Then there are 3! = 3*2*1 = 6 possible re-orderings of these. The attacker gets 5 tries to guess the picture password. So, in this scenario, the attacker has a 5/6 chance of guessing the picture password correctly before being locked out. That said, this is almost the best possible case for the attacker. 3. Design of Authentication System based on Multimedia Passwords (MPAS) The main objective behind the development of an authentication system based on Multimedia passwords was to providing a security method for any software application offline or online as an alternative to traditional text passwords, whereby a user must remember an image (or parts of an image) in place of a word. They are motivated in part by the well-known fact that people are better at remembering images than words. In MPAS, a supportive Audio and Video signature is provided to increase in remembrance of passwords, thus enhancing efficiency.(Fig 2) Further the security of this system is increased by providing the user with number of levels for password sensibility which are: a)

Level EASY: In this level, users are allowed to keep click points on Image or combination of click points on same or multiple Images, as their password. b) Level MEDIUM: In this level, users in addition to image click points, are allowed to keep Audio too , as their passwords i.e., users can keep combination of Images and Audio as their passwords. This ensures better security as well as better memorability. c) Level HARD: In this level, users in addition to image click points and audio, are allowed to keep click points on video, as their passwords. A combination of Images, Audio and Video ensures a Highly Secure authentication system.

Fig 2: Levels in MPAS

User choice is influenced by persuasion, encouraging users to select more random, and thus more difficult to guess click-points. MPAS is Highly Reliable and Highly Flexible. It provides lot of benefits: a) Authentication, Authorization and Access Control. b) The system is provided with easy-to-use GUI(graphical user interface) that guides the user in a simple manner to follow the necessary steps without much hindrance. c) Also, To aid in remembrance of passwords(if the password is forgotten), images are used in place of traditional security questions for enhanced security. d) In addition user is allowed to provide for Password Hint, in which text is converted to speech. This sound signature will be used to help the user to recover password.(Fig. 3)

[422]

Chachoo et al/COMMUNE – 2015

Fig .3: Password Recovery 4. Empirical Results and Analysis In the learning Phase, the participants entered the password repeatedly until they achieved correct input submission. The users were allowed to use this system for 10 days, after which they were asked to fill a questionnaire based on the memorability, usability and security of the system. The participants gave the score out of 5 to each of the available password systems.

Fig. 4: 2-D Line Chart depicting the feedback of participants

3. Security An authentication system should give required safety or security for its potential environment otherwise it cannot meet its prime objective. Password brute forcing attack: A system will be susceptible to a brute forcing attack if it does not possess a proper protocol to make sure that passwords are strong and follow the adequate password policy. a)

Dictionary based password attack: As the name suggests, this attack represents the usage of every word in a dictionary as password by the hacker to enter the system through some other user’s account. If the user had used a word present in the dictionary then the attacker is bound to be successful. (Van Oorschot, P.C., S. Stubblebine, 2006) b) Guessing attack :This is a hit or miss type of attack with least success rate among various types of hacks or attacks. In this particular attack the hacker tries to enter the system by presuming that the user password must be based on the personal information of the user e.g. nicknames, etc.

[423]

Chachoo et al / COMMUNE-2015

c)

Spyware attacks: Spyware is a malware that is covertly installed in a system with the intention of scanning information about users by employing key logger or even a key listener. Since the data collected through this procedure is reported to an outside source that needs information about a particular user. While graphical password authentication is going on the hacker tries to obtain the confidential information like a particular username or even a particular image. d) Social engineering attack: This type of attack involves one or more than one interaction of the attacker to obtain sensitive and highly confidential information to infiltrate an organization or computer systems. The hacker asks many relevant questions to achieve his/her goal 4. Conclusion The model proposed above named as MPAS promises to deliver as an easier to use and a memorable authentication mechanism. The MPAS model takes advantage of user’s ability to recognize images and hence has a distinct advantage in terms of usability. MPAS has an edge over other mechanisms as it is definitely more secure. MPAS multiplies the essential workload for hackers by compelling them to first get hold of a particular image selected by the user. This step represents an EASY LEVEL of intrusion for the attackers. Also MPAS makes the attackers acquire image and relative audio tracks chosen by the user(MEDIUM LEVEL).The most HARD LEVEL for an attacker to reach or cross is the one where he/she has to obtain image, audio tracks and video clips and subsequently conduct hotspot analysis (Thorpe, J et al, 2007)on each of the image. This approach has been proven to be very effective at lowering the formation of hotspots, consequently avoid shoulder surfing and at the same time provide the ever needed high security success without compromising usability.

References S. Chiasson, R. Biddle, and P. van Oorschot, 2007 A Second Look at the Usability of Click-Based Graphical Passwords,”Proc. ACM Symp. Usable Privacyand Security (SOUPS). S. Chiasson, A. Forget, R. Biddle, and P. van Oorschot, 2008.Influencing Users towards Better Passwords: Persuasive Cued Click- Points,”Proc. British HCI Group Ann. Conf. People and Computers: Culture, Creativity, Interaction, S. Chiasson, A. Forget, E. Stobert, P. van Oorschot, and R. Bddle, 2009., Multiple Password Interference in Text and Click-Based Graphical Passwords,Proc.ACM Conf. Computer and Comm. Security CCS). E. Stobert, A. Forget, S. Chiasson, P. van Oorschot, and R.Biddle, 2010, Exploring Usability Effects of Increasing Security in Click-Based Graphical Passwords,Proc. Ann. Computer Security Applications Conf. (ACSAC). G.Agarwal,S.Singh and R.S.Shukla, ., 2010, Security Analysis of Graphical Passwords over the Alphanumeric Passwords. Pinks, B. and T. Sander. 200,2, Securing Passwords Against Dictionary Attacks. ACM, CCS, L. Jones, A. Anton, and J. Earp, 2007, Towards Understanding User Perceptions of Authentication Technologies, Proc. ACM Workshop Privacy in Electronic Soc. L. O‟Gorman, 2003, Comparing Passwords, Tokens, and Biometrics for User Authentication, Proc. IEEE, vol. 91, no. 12, pp. 2019-2020, Dec.. A. Jain, A. Ross, and S. Pankanti, 2006, Biometrics: A Tool for Information Security,” IEEE Trans. Information Forensics and Security (TIFS), vol. 1, no. 2, pp.125-143. Zhi Li, QibinSun ,YongLian, and D.D.Giusto,2005, An Association Based Graphical Password Design Resistant to Shoulder Surfing Attack, IEEE International Conference on Multimedia and Expo(ICME) 0-7803-9331-7 ,pp245-248 Adam J. Aviv, Katherine Gibson, Evan Mossop, Matt Blaze, and Jonathan M. Smith, 2010, Smudge Attacks on Smartphone Touchscreen- WOOT. Van Oorschot, P.C., S. Stubblebine, 2006, On Countering Online Dictionary Attacks with Login Histories and Humans-in-the-Loop.ACM Trans. Information and System Security 9(3), 235-258, Thorpe, J. and P.C. van Oorschot, 2007. Human-Seeded Attacks and Exploiting Hots-Spots in Graphical Passwords.16th USENIX Security Symposium, R. N. Shepard, 1967, Recognition memory for words, sentences, and pictures, Journal of Verbal Learning and Verbal Behavior, vol. 6, pp. 156163, Blonder, G.E. 1996. Graphical Passwords.United States Patent 5559961

[424]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Computational Approaches for Emotion Detection and Classification in Textual Data Abid Hussain Wania*, Rana Hashmyb a

Department of Computer Science, South Campus, University of Kashmir, Srinagar, India b Department of Computer Science, University of Kashmir, Srinagar, India

Abstract Men are born with a “human heart” and express their feelings, behavior, experience, physiology, conceptualization, and cognitions through emotions. Internet provides a global platform for people to communicate and express themselves, their feelings, their emotions even when they are not staying at the same place. In the absence of face-to-face contact to detect facial expressions and intonations in voice, people decipher emotions using text on online social networking sites, which have thereby emerged as a vast repository of human communication. Emotion recognition and its deep classification has recently gained much attention due to its potential applications ranging from customer satisfaction in business decision making to cognition understanding in HumanComputer-Interaction. In this paper, we present a study of different emotion detection & recognition approaches, the potential applications and the major challenges faced in this area.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Emotion Detection; Emotion Classification Models;Deep Classification; Sentiment Analysis; Text Analysis

1. Introduction Due to the profound advances in computing and communication technologies, there has been a steep growth in popularity of online social networks. Communication over online social platforms has considerably affected the way people interact with friends and acquaintances nowadays. Indeed, interacting through online social networks and online messaging systems has become ubiquitous, and a major component of a person’s life. The exponential growth of electronic communication, particularly text-based, associated with the Internet has led to a huge increase in the amount of social data repository which is not currently warehoused or mined in the ways relevant to various areas of study like Human Social Cultural Behavior analysis (Liu et al., 2012). There has been a recent swell of interest in the automatic identification and extraction of emotions, attitudes, opinions, and sentiments in textual data. Strong motivation for this task comes from the aspiration to provide tools and support for information analysts in government, commercial, and political domains, who want to be able to automatically track attitudes and feelings in the news and on-line blogs and forums. How do people feel about the recently launched Direct Benefit Transfer of LPG (DBTL) scheme in India? Is there a change in the customer satisfaction for AirAsia since its plane crashed in the Java Sea? Trying to answer these types of questions could be profoundly made easier if we have a system that could automatically detect and extract opinions, sentiments and emotions rather than sifting through the vast repository of news and data. Researchers from AI (especially Natural Language Processing and Affective Computing) have been working on the automatic detection of emotions and their classification, though it is in fact an interdisciplinary problem involving researchers from computer science, biology, psychology, cognitive science and so on.

* Corresponding author. Tel.: +91 9797 078253.

E-mail address: [email protected].

ISBN: 978-93-82288-63-3

Wani and Hashmy/COMMUNE – 2015

2. Emotion Classification Models An emotion is a certain human feeling arising out of certain mental and physiological processes in response to internal or external events that characterizes a state of mind, such as joy, anger, love, fear and so on (Agarwal et al., 2012). Classification of emotions from data first requires the choice of an emotion model for which the major consideration is the accuracy with which models represent emotions. Although the number of emotions encountered in real life can be quite large yet many theorists in this field have suggested a set of basic or fundamental emotions and are of the opinion that all other emotional states are a mixed manifestation of theses fundamental emotions. There are two considerably different models for representing emotions: the categorical model and the dimensional model. Each type of model helps to express a unique aspect of human emotion and both of them can provide an insight into how emotions are represented and interpreted within the human mind. The categorical model and dimensional models employ two different methods for estimating the actual emotional states of a person. In the categorical model, a subject is usually required to choose one emotion out of a set of emotions that best represents the feeling conveyed whileas in dimensional model employs the rating scales for each dimension by using tool like Feeltrace (Cowie et al., 2000). A brief description of different emotion models which have proved to be of particular interest for the purpose of emotion classification is presented as under. 2.1

Categorical Emotion Models

Categorical model encompasses identifying emotions by simply making the use of emotion-denoting words, or category labels. This model either assumes that there are fundamental and discrete emotional categories such as six basic emotion categories namely anger, disgust, fear, joy, sadness and surprise (Ekman, 1992), or uses expressive categories that are domain-specific. There are both primary and unrelated emotions in the model. Each emotion is uniquely distinguished by a specific set of features, expressing eliciting conditions or responses/reactions. Most work, till date, in affective computing has focused on the six basic emotions as proposed by (Ekman, 1992). However, many researchers have argued that different sets of emotions are required for different domains, for instance, in the field of teaching and education. For instance, D’Mello et al in 2007 proposed five categories (boredom, confusion, delight, flow, and frustration) for describing affect states in the student-system dialogue. Learners rarely feel sadness, fear, or disgust, whereas they typically experience boredom or delight, which is an argument for the need for domain-specific categories. The principle advantage of categorical representation model is that it represents human emotions intuitively with easy to understand emotion labels. However, it has several weaknesses due to the limited number of labels. For example, the emotional categories consist of discrete elements, and a great variety of emotions within each discrete category can be frequently observed. The categories do not cover all emotions adequately because numerous emotions are grouped together under one category. Furthermore, the same affective states can be expressed by means of dissimilar emotional categories owing to linguistic, environmental, cultural or personality differences, which leads to meager and poor agreement among emotional categories. These findings indicate that representative set of emotional categories may not characterize distinct affective states although the set of emotion categories is defined. In addition, this incomplete and hence problematic conceptualization may result in non-optimal or inefficient affect-detection. Firstly, it can potentially lead to a forced-choice identification problem, wherein the subjects are likely to discriminate among presented categories rather than to identify an emotion label themselves. This can force the subjects to choose an inappropriate and irrelevant category. The second problem is more serious and related to the first one. It is occasionally not possible for subjects to select an appropriate category since it does not exist in the label set. Therefore, a categorical model has the limitations of an identification task in attempting to identify the precise emotional states perceived by people. For instance, subjects cannot help selecting one of six basic emotions (e.g. anger, disgust, fear, joy, sadness, and surprise) even though they feel neutral and want to choose that category. Nevertheless, the categorical model with its many variations has emerged as a dominant one due to its simplicity and familiarity. 2.2

Dimensional Emotion Models

Dimensional model represents affects in a dimensional form. Emotional states in this model are related to each other by a common set of dimensions and are generally defined in a two or three-dimensional space with each emotion occupying a location in this space. A variety of dimensional models have been proposed and studied till date (Russell, 1980) (Mehrabian, 1996), (Plutchik, 1980). In Russell’s model (Figure 2.1) of affect is introduced as a reference circumflex through a figure with a setting of points representing the emotions. Emotion-related terms in Russell’s model are organized in a circumplex shape which enables a subject to choose a position anywhere between two discrete emotion-related terms. Numerical data are obtained from the relative position of the points in the two-dimensional bipolar space (valence-arousal). The valence dimension depicts positive and negative emotions on different ends of the scale while as the arousal dimension differentiates excited vs. calm states. The proximity of two emotion categories in the circumplex represents conceptual similarity of the two categories The representation and description of emotional states by means of emotion dimensions has certain advantages. A major advantage of dimensional models is that they are not correlated to a certain emotional state (e.g. angry or happy).

[426]

Wani and Hashmy/COMMUNE – 2015

Two or three dimensions of emotional meaning are commonly identified by means of rating. Due to their gradual nature, emotion dimensions are able to capture subtle emotion concepts that differ only slightly in comparison with broad emotion categories. Emotion dimensions can depict very specific identification and a large range of people’s emotion concepts. In particular, a dimensional description is well suited for the task of measuring the full-defined emotional states. Furthermore, emotional states are related to each other on a dimensional space, which is a significantly different approach from the categorical model. A dimensional model provides a means for measuring the degree of similarity between emotion categories; adjacent categories on the space are very similar, while as the opposite, categories are distinctly different from each other. Conclusively, a dimensional emotion model is a useful representation capturing all relevant potential emotions and offers a means for measuring similarity between affective states.

Fig. 2.1 Russell’s circumplex model of emotion (Russell, 1980)

Fig. 2.2 Plutchik’s emotion wheel (Plutchik)

3. Emotion Detection methods and their challenges Over the past several years, much research has been done utilizing linguistics, machine learning, information retrieval, and other theories to detect emotions. Experiments conducted so far have shown that, computers although in a coarse way, can detect and recognize emotions from texts like humans. However, all methods have certain limitations, and they lack context analysis to refine emotion categories with existing emotion models, where much work has been done to put them computationalized in the domain of believable agents. (Kao et al., 2009). Currently, three approaches dominate the emotion recognition and detection task; keyword based, learning based and hybrid based approach. These make use of features mainly selected from syntactic (e.g. n-grams, pos tags, phrase patterns) and semantic (e.g. synonym sets) data to detect emotions (Binali et al., 2010). The hybrid approach attempts to make up for the weaknesses in other two approaches (Table 3.1). Hybrid model takes advantage of the fact that the syntactic and semantic information can be highly helpful for emotion detection. Contemporary methods are lacking in in-depth semantic analysis for detecting hidden phrase patterns and more research needs to be done to identify, build, and incorporate knowledge rich linguistic resources that have a focus on recognizing and detecting emotions. The main benefit of this approach is that it can yield higher accuracy results from training a combination of emotion classifiers and supplemented by knowledge-rich linguistic information from dictionaries and thesauri. An important and direct consequence of this is that it will reduce the high cost involved in using human indexers for information retrieval tasks and minimize complexities encountered while integrating different lexical resources. Table 3. Emotion Detection Methods; Strengths and Challenges Method

Strength(s) of the method

Keyword -Based



Most intuitive, easy to implement

Learning -Based



Facilitates easy implementation of classifiers by novices who can then apply the learned model to new instances.

 Hybrid



Offset the high cost involved in using human indexers for information retrieval tasks Minimize complexities encountered while integrating different lexical resources

[427]

      

Challenges for the method Ambiguity in Keyword definitions Keyword-less sentences Conflicting keywords Determination of features (emotion indicators) The number of emotion categories that can be recognized Acquisition and processing of multi-domain corpus in supervised techniques Acquisition and processing of training data corpus when supervised learning techniques are used

Wani and Hashmy/COMMUNE – 2015

4. Conclusion In this paper, we briefly surveyed the various emotion detection models and detection and classification approaches, which are popular in the research community in this field. A comparison of different emotion detection approaches was presented. We outlined the fact that the syntactic and semantic information can be beneficial for emotion detection and classification task. References Agarwal, A., An, A., 2012. Unsupervised Emotion Detection from Text using Semantic and Syntactic Relations, In proceedings of the 2012 2012 IEEE/WIC/ACM International Conference on Web Intelligence and Intelligent Agent Technology (WI-IAT). Washington, DC. Binali, H., Wu, C., & Potdar, V., 2010. Computational approaches for emotion detection in text, 4th International Conference on Digital Ecosystems and Technologies (DEST). Dubai, UAE. doi:10.1109/DEST.2010.5610650 Cowie, Roddy, Douglas-Cowie, E., Savvidou, S., McMahon, E., Sawey, M., & Schröder, M. "FEELTRACE': An instrument for recording perceived emotion in real time. In ISCA Tutorial and Research Workshop (ITRW) on Speech and Emotion. D’Mello, S., Picard, R., & Graesser, A., 2007. Towards an effect sensitive auto tutor, IEEE Intelligent systems 22, p. 53-61. Ekman, P., 1992. An argument for basic emotions, Cognition & Emotion, 6, p. 169-200. Kao, E. C., Liu, C. C., Yang, T. H., Hsieh, C. T., Soo, V. W., 2009. Towards Text-based Emotion Detection a Survey and Possible Improvements, In International Conference on Information Management and Engineering. Kaula lumpur, Malaysia. p. 70-74 doi: 10.1109/ICIME.2009.113 Liu, X., Tang, K., Hancock, J., Han, J., Song, M., Xu, R., & Pokorny, B., 2012. SocialCube: A Text Cube Framework for Analyzing Social Media Data. In proceedings of ASE International Conference on Social Informatics. Washington, DC. Mehrabian, A., 1996. Pleasure-arousal-dominance: A general framework for describing and measuring individual differences in temperament, Current Psychology 14, p. 261-292 Plutchik, R., 1980. Emotion: A psychoevolutionary synthesis, Harper & Row, New York, p. 440 Russell, J. A., 1980. A circumplex model of affect, Journal of personality and social psychology 39, p. 1161

[428]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16-18March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Achievements and Limitation of the First Machine Translation System to convert Hindi into Dogri Preeti*, Devanand Department of CS&IT, Central University of Jammu, Jammu, India

Abstract The Hindi-Dogri machine translation system has been developed to facilitate translation of Hindi text into Dogri text. A user can get the translations done on the click of a button. The development of this system is an effort to bring the Dogri language on the map of machine translation. This paper summarizes the achievements and limitations of this research work.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Dogri; Machine translation, Achievements,Limitations

1.

Introduction

The machine translation system for Hindi-Dogrihas been developed using ASP.Net and the databases are in MSAccess with Hindi and Dogri text in Unicode format. It is based on the direct approach of machine translation. It takes Hindi text as input and provides Dogri text as output in Unicode. This is the first machine translation system developed to convert Hindi text into Dogri. Both Hindi and Dogri use the Devanāgarī Script. These are closely related languages and share not only the script but also the lexicon. The major difference lies in the inflections of the words in these languages. The phases involved in Hindi-Dogri MTS are: pre-processing of the source text (Hindi), word to word translation of the text to the target language (Dogri), followed by inflectional analysis and ambiguity resolution then handling peculiarities of Dogri followed by output generation. The pre-processing phase involves text normalization, proper noun recognition and handling of collocations.(Dubey, 2014) 1.1 Pre-processing Activities 





Text Normalization: Text normalization refers to keeping a standard spelling for spelling variations of the same word. Therefore, all the variants of the same word need not be stored in the dictionary database. A database of such words has been created and has a collection of 400 words. It contains a standard word for its variants. The standard word has its meaning in the dictionary database. Some examples of such words are:गई/gī, गयी/ gayī;ह द िं ी/ hiṃdī, ह न्दी / hindīetc, Identifying Collocations:Words that cannotbe translated word to word are called collocations.They need to be handled properly for accuracy of the output. If these words are translated word to word, the word sense is changed. e.g. आपके(āp ke) should be translated into Dogri as तिं’दे / tuṃ’de. If not handled properly, it will be translated word to word asतसदे / tusa de. Proper Noun Identification: After checking for collocations in the source text, the system extracts proper nouns such as names of months, countries, days of a week, universities, banks etc. These words are usually transliterated to preserve their meaning. In our system both the source and target language use the same script, therefore these words are not to be translated; word to word so as to preserve the word sense. In this sub phase,

* Corresponding Author. Tel:+919419180084. Email address: [email protected].

ISBN: 978-93-82288-63-3

Preeti and Devanand/COMMUNE – 2015

the proper nouns are identified by referring the proper noun database. These words are not sent for further translation. The architecture of the system is presented in Fig I. 1.2 Tokenizer It segments the input stream of characters into single meaningful units called tokens. The output of the preprocessing phase goes to the tokenizer, which extracts individual words from the sentence for further processing for translation into the target language. The tokens are extracted from the text using space, a punctuation mark, as delimiter. 1.3 TargetText Generation The translation engine consists of the following sub-phases: a) Lexicon look up b) Ambiguity Resolution c) Inflectional Analysis d) Handling special cases pertaining to Dogri. i. Handling kar at the end of words ii. Handling words before raha iii. Handling words before laga The final output i.e. Dogri text is generated after going through all the above mentioned phases.The system has been tested using the human evaluation method. Both Quantitative (include intelligibility testing and accuracy testing) and Qualitative tests (include WER and SER) have been performed on the system. The accuracy of the system on the basis of Intelligibility test has been calculated as 98.54% and the accuracy test reports 98.71% accuracy. In the quantitative tests the Word Error Rate is found out to be 2.011% whereas Sentence Error Rate is 20.40% (Dubey, 2014).

Fig I: Architecture of Hindi-Dogri MTS

2.

Achievements in the Course of Development of Hindi-Dogri MTS

The achievements during the course of development of Machine Translation System for Hindi-Dogri language pair are discussed the following section:  The first survey to study the Digital Divide factors has been conducted in Jammu, to The efforts made by the Government to bridge the digital divide have also been presented (Dubey et al, 2011a).  The first Grammatical and inflectional analysis of Hindi and Dogri has been done to study the closeness between these languages (Dubey et al, 2011b ).

[430]

Preeti and Devanand/COMMUNE – 2015

 

3.

The first digital dictionary for Hindi- Dogri has been developed(Dubey, Devanand, 2013). There was no dictionary available for the language pair for Machine Translation purpose. Rules for morphological analysis of Hindi-Dogri have been devised, which can be used in future for other research work on translation of Dogri. Modules for handling Proper Nouns and collocations have been developed successfully. These modules facilitate increased efficiency of the system. Sometimes, during translation proper nouns and collocations are translated leading to inaccurate results. The module for handling proper nouns identifies proper nouns and does not send them for further translation. The module for handling collocations does the same when collocations are seen in source text. 3. Limitations of the Hindi-Dogri MTS

The system shows good accuracy of translation, but it still has some limitations. Some common errors are explained with examples: 3.1 Proper noun Recognition Failure The names that are not present in the Dogrinames database, but have their translations in the dictionary database will be translated. For identifying a proper noun, it must either be preceded with a title or be succeeded with a surname. In the absence of both (surname and title) is not identified; and hence translated. For example: Input sentence:

विजयआयाथा (vijay āyā thā)

Output Sentence:

जजत्तआएदा ा (jitt āedā hā)

3.2 Ambiguity Resolutions Words with multiple meanings are not resolved by the system. For example: Input sentence:

कैसेख्यालोंमें खोर ीथी।

Incorrect Output:

कक ’यािंख्यालें चखोआदी ी।( ki ’yāṃ khyāleṃ ca khoā dī hī )

Correct Output:

कने े ख्यालें चखोआदी ी।(kanehekhyāleṃ ca khoā dī hī)

(kaise khyāloṃ meṃ kho rahī thī)

33 Disagreement of दा/da postposition before Verb phrase: In Dogri, all the Verb phrases in the sentence must agree with the postposition दा.However, in some cases, it fails as shown in following example: Input sentence: ब नकीशादी ोगई(bahan kī śādī ho gī) Incorrect Output: भैनदीब्याह् ोईगेई(bhain dī byāh hoī geī) Correct Output: भैनदाब्याह् ोईगेआ(bhain dī byāh hoīgeā) 3.4 Gender Disagreement: The output of the system sometimes does not reflect the correct gender of a word and therefore causes gender disagreement with verb/postposition in the target language. For example: Input sentence: उन् ें रोडआइलैंडमें एकशिममला ै / (unheṃ roḍ āilaiṃḍa meṃ ek śav milā hai) Incorrect Output:उ’नेंगीरोडआइलैंडचइकलाशममलेआऐ/ (u’neṃ gī roḍ āilaiṃḍa ca ik lāśmileāai) Correct Output:उ’नेंगीरोडआइलैंडचइकलाशममलीऐ/(u’neṃ gī roḍ āilaiṃḍa ca ik lāśmilīai) 3.5 ी Inflection Missing In Dogri, most of the words end with ी . Therefore, there is a need to add this inflection in many words. Some failure cases are discussed below: Input sentence: इसमें बािंधलो। (isameṃ bāṃdh lo) Output Sentence:एह्दे चब’न्नलैओ। (ehde ca b’nn laio) [431]

Preeti and Devanand/COMMUNE – 2015

Correct output:एह्दे चब’न्नीलैओ।(ehde ca b’nnīlaio ) 3.6 Wrong Sentence Structure In Dogri, if र ा,र ी, र े are preceded with न ीिं, the structure of the sentence changes in the following manner. Input sentence: गठरीबिंधन ीिंर ीथी। (gaṭharī baṃdha nahīṃ rahī thī) Incorrect Output: गिंढब’न्ननेईंरे ी ी।(gaṃḍha b’nna neīṃ rehī hī ) Correct Output: गिंढब’न्नोआदीनेईं ी। (gaṃḍha b’nnoā dī neīṃ hī) 3.7 Disagreement of Subject Noun Phrase with Verb Phrase Some failure cases due to disagreementof all the verb phrases in the sentence with the subject noun phrases are seen in some translations. The following example illustrates the disagreement of subject noun phrase with verb phrase Input sentence: उसेक्यासजादीजानीचाह ए।(use kyā sazā dī jānī cāhie) Incorrect Output: उस्सीकेह् स’जाहदत्तीजानीचाह दा।(ussī kehs’jā dittī jānī cāhidā) Correct Output: उस्सीकेह् स’जाहदत्तीजानीचाह दी। ( ussī kehs’jā dittī jānī cāhidī) 3.8 Limited Dictionary The Hindi-Dogri dictionary used in this system consists of 18510 words, which is not very large. The efficiency of the system can be improved by increasing the size of the dictionary. 4.

Future Directions

Although this system is showing good results, using the direct translation approach but still there is lot of scope for improvement. Following are some of the future directions:  



 

5.

Increase in Database size: The present databases used by the system such as dictionary, proper noun database, surnames, titles, bigrams and trigrams for WSD can be extended to improve the accuracy of the system. Use of Newer Methods of Translation:Statistical Machine Translation approach is the latest approach of MT and is believed to give highly accurate results. The development of SMT system requires a high quality parallel corpus. Thus, with the use of the present system, parallel corpus for Hindi-Dogri Language pair can be developed for use in future researches. Use of Automatic Evaluation Techniques: Automatictechniques of evaluation such as BLUE, METEOR etc are used worldwide for evaluation of translation software.These techniques also require parallel corpus for the language pair undertaken. These techniques can be adopted in future for Hindi-Dogri Language pair, once the parallel corpus has been developed. Word Sense Disambiguation: WSD should be extended to include polysemous words also; this will increase the accuracy of the system. Though not many polysemous words were encountered during testing; but adding this module will disambiguate such words also leading to increase in translation accuracy. Website Translation:The Machine translation System can further be extended for translating websites and emails. Conclusion

This paper in brief introduces the Hindi-Dogri MTS, the achievements in the course of its development and the limitations of the system, some of which can be handled in future. Future directions of this work have also been presented. References Dubey,P., 2014. Study and development of a machine translation system from Hindi language to Dogri language: an important tool to bridge the digital divide, PhD Thesis, Department of Computer Science &IT, University of Jammu. Dubey, P., Devanand, 2013. Machine Translation System for Hindi-Dogri Language Pair, in proceedings of IEEE Conference (ICMIRA), held at SMVDU, in Dec 2013, ISBN: 978-0-7695-5013-8, pages: 422-425, DOI 10.1109/icmira.2013.89 Dubey, P., Jyoti,J., Devanand,2011a. A Study to Examine digital Divide Factors: Jammu and Kashmir Perspective, BVICAM’s International Journal of Information Technology (BIJIT) , Volume 3 ,No.2, Issue 6, July-December, 2011, ISSN: 0973-565 Dubey, P., Pathania, S.,.Devanand, 2011b. Comparative Study of Hindi and Dogri Languages with regard to Machine Translation , Language In India, Volume 11:10 October 2011,ISSN 1930-2940.

[432]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Computational Aspect of different Neuroimaging Modalities to Study the Structural and Functional Brain Networks: Analysis and Measure Based on Advanced Statistical and Classical Graph Theory Approach Vinay Shukla*, Shrawan Kumar, Dharmendra Singh Department of Computer Science & Engineering, Institute of Technology & Management Chehari, Maharajganj UP, India

Abstract Advancement in brain imaging technology gives a good insight of inner brain circuits and its functional neuronal network. Brain is a nonlinear dynamical system, considering the dynamical aspect one can also think brain as hub of neuronal networks. In recent years there has been huge study on network modelling of brain connectivity. One of the challenge to establish the network connections between heterogeneous brain regions. The purpose of this study is to explore the structural and functional brain networks using task based and non-task based data of MRI, Diffusion MRI and EEG. The temporal resolution is very promising in EEG while MRI is only give spatial resolution but combined these two imaging modality. Using EEG for functional network connectivity, has some issues and severely limited by volume conduction and its accuracy is entirely depends on source modelling. Considering the above lacuna we are trying to propose a novel approach based on interacranial cerebral EEG (SEEG) recordings in human brain. We are also delineating theoretical graph analysis of human brain networks based on various neuroimaging modalities such as structural MRI, diffusion MRI and EEG. To investigate the salient characteristics of brain regions of preferred network pathway and its connectivity hub location. We are also proposing ML (maximum likelihood) estimation between exact functional connectivity and expectation with minimization ratio between brain functional regions. Using this study one can make a possible assumption of topological organization of human brain networks and its connected network hubs. In later stage of proposed research one can also deal with various neuropsychiatric diseases and one can also see brain development and change in network properties during different stage of age.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords : EEG; Brain Network; MRI; MLE; Graph Theory; Diffusion MRI

1. Introduction Neural function, and by extension neural codes, are constrained by connectivity. Brain connectivity is really crucial to elucidating how neurons and neural networks process information (Brandes and Erlebach, 2005). The functional connectivity in human brain is spontaneous, whole process is quite debatable. Advancement in computational neuroimaging technique (CNT) gives more uncertainty and overlook brain activity. Till now most of the studies is based on external stimuli based and which is challenge for higher cognitive challenge. All these assumptions based on average statistical random analysis. In imaging technique the task based paradigm activation is kind of manipulation which clearly result of cerebral activation of circuits, and its important for performing the task. All these studies are totally based on some short of ideal base lines (Gustavo, 2011). The BOLD activity is entirely based on the signal change during performing some task and totally depends on task using this one can see the functional network during activation which leads towards default mode network (DMN) (Raichle et al, 2001). Imaging structural and functional brain connectivity has revealed the complex brain organization into large-scale networks. Such an organization not only permits the complex information segregation and integration during high cognitive processes but also determines the clinical consequences of alterations encountered in development, ageing, or neurological diseases. Recently, it has also been demonstrated that human brain networks (Sporns et al., 2004) shared topological properties with the so-called 'small-world' mathematical model, allowing a maximal efficiency with a minimal energy and wiring cost (Guye et al., *

Corresponding author. Tel.: +91 9554 776314 E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Shukla et al /COMMUNE – 2015

2008). To understand the complexity behind the human brain networks one can only understand using various kinds of neuroimaging modalities such as Structural MRI, Diffusion MRI and EEG/MEG (Horwitz, 2003). Structural MRI deal with anatomical network of different neuronal regions and its morphology. Diffusion MRI to understand and delineate the bunch of fibre connectivity in different regions in white matters tracts. Diffusion tractography and functional/effective connectivity MRI provide a better understanding of the structural and functional human brain connectivity (Guye et al., 2008). 2. State of the Art In this section we present the most prominent work relevant to our research. The aim is to understand the structural brain network followed by functional brain network using structural MRI combined with diffusion MRI followed by EEG. There are already enough research is ongoing and also significant result are also achieved. But if we combined MRI EEG and DTI we will be able to understand better brain networks from Structural to functional. Diffusion tensor imaging is very novel neuroimaging technique, using DTI one can see the white matter track and also explore the connectivity from other brain regions. One can also do fibre tractography to understand the fibre connection by each voxel. Using Probabilistic tractography one can compute the connectivity probabilities without touching the actual white matter pathways (Rajapakse et al., 2008). EEG has better temporal resolution and time dependent signal, one can easily calculate the time dependent signal. Based on time varying signal one can easily map the scalp on brain surface. One can also combined Diffusion Tensor Imaging (DTI) with EEG to understand the connectivity voxel by voxel. We are proposing novel method to understand the brain network and its connectivity through graph theory approach and measure of these networks is going to be done by advanced statistical technique. We are proposing complex graph analysis based on binary graph which is highly usable in complex networks. The Graph theoretic analysis of functional brain connectivity has helped to identify functional hubs, which are highly connected and central to information flow and integration (McIntosh and Gonzalez, 1994). Functional connectivity studies in the frequency domain have provided evidence for a fractal organization of functional brain networks (Friston, 2005). 3. Problem Statement Advancement in brain imaging modality give us very good understanding of inner brain networks. In this paper, we are going to use multimodal imaging data sets, which is purely based on particular imaging machine. Because different imaging machines has different imaging protocol. Our basic aim of study is to understand the brain in structural to functional and its connectivity using DTI. In structural brain network we would like to understand the resting state network to understand the coordinates of the brain regions. For functional network, we would like to understand the task based activity in particular brain regions. Using above we would be able to judge the development of brain networks in different age period. We would also be able to predict the networks changes in different kinds of neurological diseases. We would like to investigate some challenging research questions like: ●Brain network changes in neurological diseases and also we would like to answer different kind of connectivity structural, functional and effective connectivity or causal connectivity and its measure in certain kind of brain disorders. ●We will also explore the functional connectivity how it maps to structural connectivity. ●The parallel analysis of structural connectivity maps of the human brain and patterns of functional and effective connectivity recorded in various conditions of cognitive activation and cognitive maps. 4. Proposed Solution: General Idea To start the preliminary work we will get all the data sets provided by our collaborators. After extracting the data sets using dicom manipulation or visualization software like Mricro. Mricron will convert all the data sets in to nifti format for visualization and analysis process. For EEG data sets we will use EEG lab software and Matlab function to extract the EEG data sets using Fourier or Wavelets transform. To extracting the EEG feature extraction for time varying signal we will use wavelet transform. Wavelets transform is better to extract the signal and modularize the EEG signals. For structural MRI we used anatomical brain template like MNI brain template to match our brain data and superimpose on it. One can use (SPM, n.d) or (FSL, n.d) software for Diffusion MRI data sets. To extract the brain network point by point. For mapping the brain scalp in to network one can use theoretical graph theory approach (Bullmore et al., 2009). As we know in graph has good features to establish the complex networks. For EEG signal we will use causality modelling over time varying signals. One can also think of EEG signal is good for time and space dependent signals to map the causal signals for each node using advanced modern algebra methods like Lie Theory and its operators specially Zeeman's causality. One can also use some short of Granger causality (Sato et al., 2010) method for EEG Signals to map the entire data sets over time dependent signal into space and time. For structural MRI we will use ICA (Independent component Analysis) to localize the brain data sets into point of interest brain regions (Rowe et al., 2006). For diffusion MRI its good to use Probabilistic method for tracking the fibres and compute probabilities of [434]

Shukla et al /COMMUNE – 2015

theses fibers without touching the pathways of these fibres. Finally one can develop tools based on Matlab functions using C++ or any other programming language to make all our queries automated for possible predictions based on given query. C and C++ is good because one can easily import or convert these functions into Matlab or vice-versa using MEX compiler. 5. Conclusion The topology of structural and functional human brain networks offers a common framework to merge structural and functional imaging as well as dynamical data from electro physiology that might allow a comprehensive definition of the brain organization to understand the differences between different modality of brain networks in multimodal neuroimaging techniques. But one can only understand the differences between underline said phenomenons if and only if to dig between structural connectivity and its coherence networks. References Gustavo, D., 2011, Emerging concepts for the dynamical organization of resting-state activity in the brain. Nature Reviews.Vol. 12. Raichle, M. E. et al., 2001 A default mode of brain function. Proc. Natl Acad. Sci. USA 98, 676–682. SPM (n. d) - Statistical Parametric Mapping url: www.fil.ion.ucl.ac.uk/spm/ FSL(n.d)-url: www.fmrib.ox.ac.uk/fsl/ Sato JR et al., 2010, Analyzing the connectivity between regions of interest: an approach based on cluster Granger causality for fMRI data analysis. Neuroimage. 52(4):1444-55 Rajapakse JC et al., 2008, Probabilistic framework for brain connectivity from functional MR images. IEEE Trans Med Imaging. 27(6):825-33. Horwitz B, 2003, The elusive concept of brain connectivity. Neuroimage 19, 466-470. Sporns, O, Chialvo, D, Kaiser, M, Hilgetag, CC, 2004, Organization, development and function of complex brain networks. Trends Cogn Sci 8, 418425. Brandes, U, Erlebach, T, 2005, Network Analysis. Springer, Berlin. McIntosh, AR, Gonzalez-Lima, F., 1994 Structural equation modeling and its application to network analysis in functional brain imaging. Hum Brain Mapping 2, 2- 22. 15. Friston, KJ. 2005, Models of brain function in neuroimaging. Annu Rev Psychol 56, 57Guye M. et al., 2008, maging structural and functional connectivity: towards a unified definition of human brain organization? Curr Opin Neurol.21 (4):393-403. Bullmore E. et al., 2009, Complex brain networks: graph theoretical analysis of structural and functional systems. Nat Rev Neurosci. 10(3):186-98. Rowe DB. et al., 2006, Multivariate statistical analysis in FM.

[435]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Study of the Emergence of Sky Computing Vivek Chalotra* Department of Physics & Electronics, Baba Saheb Ambedkar Road, University of Jammu, Jammu, India.

Abstract Sky computing is truly changing the method for computing. Numerous machine resources, for example, hardware and programming modules are gathered into the resource pool which can be used by the clients by means of the web through web programs, desktops, or cell phones. It is not a new idea; it is identified with cluster computing, grid computing standard and utility computing. All these computing paradigms have really helped in the improvement of cloud computing. With the expansive scale utilization of web everywhere throughout the world, everything can be conveyed over web utilizing the idea of sky computing as a service like electrical energy, cooking gas, water etc. In this paper, we will analyze all the advances in detail, which leads to the emergence of Sky computing.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: grid computing; sky computing; cloud computing; resource balancing; cluster computing; parallel processing.

1. Introduction We have encountered an enormous change in computing from more seasoned times till today. Long ago, large machines were kept behind the glass dividers and just the experts are permitted to work on them (Lucky, 2009). Later the idea of grid computing which permits the clients to have computing on interest as per need come in existence. After that, we got such computing, which makes resource provisioning more straightforward and on need of customer. By then, finally we got the thought of distributed computing which concentrates on the provisioning and deprovisioning of calculation, storage, information services to and from the user without user being not mindful of the way that from where these resources are coming to him/her (Singh, Kirit, 2012). With the substantial scale utilization of web everywhere throughout the world, everything can be delivered over web utilizing the idea of sky computing as a service like water, cooking gas and power and so forth. The remaining part of the paper is composed as follows: Part 2 depicts the cluster-computing paradigm including its pros and cons. Part 3 depicts grid-computing paradigm including its pros and cons. Part 4 depicts sky-computing paradigm including its pros and cons. Part 5 shows the key attributes of cluster, grid, and sky computing. Finally, conclusion is displayed. 2. Cluster Computing Paradigm Cluster computing is a kind of computing in which a couple of nodes are made to run as a one large computer. The different nodes included in cluster are ordinarily associated with one another utilizing some fast LAN's (Indu et al, 2011). There are principally two reasons of deploying a cluster rather than a standalone machine, which are execution and fault resistance. An application requires high processing regarding response time, memory, and throughput particularly when we discuss real time applications. Cluster computing gives high computation by utilizing parallel programming, which is utilization of numerous processors all the while for various or a single task. An alternate reason is fault resistance which is really the capacity of a framework to work smoothly even near any flaw. As the clusters are

* Corresponding author. Tel.: +91 9419 304382. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Chalotra/COMMUNE – 2015

the copies of comparative parts, the flaw in one part just influences the cluster's energy however not its accessibility. Along these lines, clients dependably have a few components to work with even in the vicinity of fault.

Fig. 1: Cluster computing (Indu et al, 2011) Here Fig. 1 demonstrates the general idea of cluster computing as per which a few nodes merge together and are introduced as a single interface/node to the client.

2.1.

Pros and cons of Cluster Computing

Table 1.Pros and cons of Cluster Computing S.No.

1.

2.

3.

Pros Manageability: It requires a great deal of effort, expense, and cash to deal with a substantial number of segments. In any case, with cluster, vast quantities of segments are consolidated to act as a single element. In this way, administration gets to be simple. Single system Image: Again, with cluster, client simply gets the vibe that he is working with a solitary framework, but in real practice, he is working with an expansive number of parts. He requires not stressing over that segments, he just needs to deal with a single framework image. High Availability: As all the parts are imitations of one another, so if one part goes down in view of any specialized reason, then some other segment can take its place and client can keep on working with the framework.

Cons Programmability Issues: This may be the situation if the parts are distinctive regarding programming from one another, and afterward there may be issues when joining every one of them together as a single substance.

Issue in Finding Fault: Because we are managing a single element, so issue may emerge when discovering fault that which of the segment has some issue connected with it. Hard to handle by a Layman: As cluster, processing includes combining diverse or same segments together with distinctive programmability, so a non-proficient individual may think that it hard to manage.

3. Grid Computing Paradigm Grid Computing is the isolation of resources from various destinations to take care of an issue that cannot be comprehended by utilizing the processing of a single machine. It utilizes utilization of different clusters that are loosely coupled, heterogeneous and are geographically scattered. Here individual client gets access to the resources (like processors, storage and data and so forth.) on interest with practically little or no idea of the concept that where those resources are physically placed. For instance, we utilize power for running cooling systems; computers and so on through wall sockets without worried about the reality that from where that power is coming and how it is being created. It is all the more prevalently known as a collection of servers that are bound together to assault a single task. Grid computing is mainly concerned about offering, gathering, facilitating and giving services to different consumers. Despite the fact that the Grid relies upon the computer systems and communication networks of the basic web, novel programming permits clients to get to machines distributed over the network. This product is called "middleware" in light of the fact that it sits between the operating systems of the machines and the applications software that can tackle a client's specific problem. As an occurrence, the Worldwide LHC Computing Grid (WLCG) – a distributed computing framework masterminded in tiers – gives a group of in excess of 8000 physicists close continuous access to LHC information (CERN, n.d). The Grid builds on the innovation of the World Wide Web, which was invented at CERN in 1989.

[437]

Chalotra/COMMUNE – 2015

Fig. 2: Grid computing (CERN, n.d) Here Fig. 2 demonstrates the tier structure of Worldwide LHC Computing Grid (WLCG)

3.1.

Pros and cons of Grid Computing

Table 2.Pros and cons of Grid Computing

S.No. 1.

2.

3.

Pros Access to Additional Resources: Grid computing can provide access to CPU, storage and many other resources as well. Resource Balancing: Grid computing framework joins different frameworks into a single system image. Grid enabled applications needs resource balancing which is done by network by scheduling jobs on nodes that are demonstrating low usage. Reliability: Grid computing sites are topographically scattered. In case, for example, there is power or cooling bafflement at one site, then that won't impact the other site, hence high reliability will be there uncommonly.

Cons Not Stable: Grid programming and benchmarks are not steady in correlation to other computing models. Its standards are yet advancing. Fast Internet Connection Required: Gathering and amassing different resources from geologically scattered destinations require fast internet connection, which brings about high financial expense. Distinctive Administrator Domains: Sometimes political issues emerge when offering assets among diverse domains. Some extra devices are needed for having legitimate adjusting and managing among diverse environments like cfengine, opsware and so forth.

4. Sky Computing Paradigm Sky computing or Cloud computing is the new computing model, which gives huge pool of dynamical versatile and virtual resources as a service on demand. The main principle behind sky computing model is to offer computing, storage, and software as a service or as a utility. We just require web to utilize these utilities. Sky is a parallel and distributed computing framework comprising of a collection of interconnected and virtualized machines that are dynamically provisioned and exhibited as one or more bound together computing resources focused around servicelevel agreements (SLA) built through negotiation between the service provider and purchasers. Sky computing cuts the operational and capital expenses and permit the IT divisions to concentrate on vital activities as opposed to keeping the server farm running (Velte et al, 2010). It gives the service on Infrastructure level, Platform level, and Software level. It gives numerous features, for example, speed, versatility of resources, parallel processing, simply pay the utilized resources, pick an alternate technology whenever to further work, every minute of every day accessibility of services, independent location, gives reliability and security and so on. Some of the areas of cloud computing are Banking sector, Insurance sector, Space exploration and Weather forecasting. Infrastructure as a service means that hardware, software and other equipments can scale up and down dynamically. Platform as a service offers high level integrated environment to build, test and deploy custom apps. Software as a service delivers special purpose software that is remotely accessible. Sky computing research addresses the difficulties of meeting the necessities of cutting edge private, public and hybrid sky computing architectures, additionally the difficulties of permitting applications and improvement stages to exploit the profits of sky computing. The research on sky computing is still at an early stage (Reddy et al, 2011). Numerous existing issues have not been completely tended to, while new difficulties continue rising up out of industry applications. Some of the challenging research issues in sky computing are Service Level Agreements (SLA’s), Cloud Data Management & Security, Data Encryption, Migration of virtual Machines, Interoperability, Access Controls, Energy Management, Multi-tenancy, Server Consolidation, Reliability & Availability of Service, Common Cloud Standards and Platform Management.

[438]

Chalotra/COMMUNE – 2015

In the Sky deployment model, networking, platform, storage, and software infrastructure are provided as services that scale up or down depending on the demand as shown in figure 3.

Fig. 3: Sky deployment model (Reddy et al, 2011) Here Fig. 3 demonstrates sky deployment model.

Pros and cons of Sky Computing

4.1.

Table 2. Pros and cons of Sky Computing

S.No.

1.

2.

3.

Pros Resource sharing: In sky computing user can scale up and scale down the resources on need basis. It offers resources, which give services to various users. Pay-As-You-Go: Clients just need to pay only for those resources which are used by them. They can demand for more resources in the event that they require later on and they can moreover release their resources after usage. Better Hardware Management: It is simple for cloud administration supplier to deal with the hardware effectively on the grounds that all machines run the same equipment [2].

Cons Less Reliability: Cloud Computing is less solid in light of the fact that it used to impart the resources to different clients. So there is possibility to take the information of a client or information of one association may blend with the information of an alternate association. Web: The principle prerequisite for clients to utilize the administrations of distributed computing is web. Clients require high speed web connection. Inaccessibility of web would result in inaccessibility of information. Non-Interoperability: On the off chance that customer put away data in one cloud then later on he/she can't move it to an alternate cloud service supplier in light of the fact that there is non-interoperability between sky based frameworks.

5. Key attributes of cluster, grid and sky computing S.no.

Attributes

4.

Installation Place

Same physical location

5.

Possession

6.

Network and bandwidth

Single owner Low latency and high bandwidth dedicated network

7.

Privacy

Password based login

8.

Discovery

Membership services

9. 10.

Service Negotiation Client Management Resource Management Resource Allocation

Limited Centralized

Grid computing High-end Servers (rack/tower/blade) 1000 computers Mainly Unix/Linux, Middleware Different physical locations Multiple owners High latency and low bandwidth dedicated network Public/Private key pair based authentication Centralized indexing and decentralized data services Yes, SLA based VO based, Decentralized

1.

Composition

Standard PCs

2.

Magnitude Operating System and software

100 computers

Centralized

Distributed

/Distributed/ Centralized

Centralized

Decentralized

Centralized/Decentralized

3.

11. 12.

Cluster computing

Windows/Linux/Unix

[439]

Sky computing Standard PCs, servers and NAS 100 to 1000 computers Multiple system software’s running on Virtual Machines Different physical location Single owner Low latency and high bandwidth dedicated network Each Application/ User is provided with a Virtual machine Membership services Yes, SLA based Centralized or third-party

Chalotra/COMMUNE – 2015

S.no.

Attributes

Cluster computing

13.

Standards

VIA based

14. 15.

Single System Image Capability

16.

Failover Management

Yes Stable Failed tasks are generally restarted

Grid computing Open Grid forum standards No Varies but high Failed tasks are generally restarted

17.

Pricing

Non-open market

Mainly internal pricing

18.

Potential

Limited

19.

Internetworking

20.

Applications

Multiple clusters within an Organization Business, Science, and Data centers

Limited, mainly scientific computing oriented Multiple grid sites composed of clusters Collaborative Scientific and HPC Applications

Sky computing Web services (SOAP/ REST) Yes, but optional On demand provision Strong support for failover and Content replication. Pricing according to service provider and client Highly potential, offer individual or combined cloud services to users Loose coupling of services from different clouds Web applications and Content delivery

Table 2. Key attributes of cluster, grid and Sky Computing [2]

6. Conclusion Sky computing is the latest innovation of computer network system, giving the web services at lower expense contrasting with typical strategy. It helps enhance the services in other related technologies, for example, Grid, cluster and utility computing. In this paper, we highlighted the pros, cons and looked at the various attributes of cluster computing, grid computing, and sky computing.

References Lucky, R. W., 2009. Reflections Cloud computing, May 2009, IEEE Spectrum. Indu, G., Pawanesh, A., Pooja, G., Rohit, U. and Sandeep, S., 2011. Cloud Computing Over Cluster, Grid Computing: a Comparative Analysis, Journal of Grid and Distributed Computing, pp-01-04. CERN, Conseil Européen pour la Recherche Nucléaire, url:http://home.web.cern.ch/about/computing Velte, A.T., Velte, T.J. and Elscnpeter, R., 2010. Cloud Computing- A Practical Approach, The McGraw-Hill Companies, New York. Singh J. Y., and Kirit, M., 2012. Cloud Computing Concepts, Architecture and Challenges, International Conference on Computing, Electronics and Electrical Technologies [ICCEET], IEEE. Reddy, V. K., Rao, B. T., Reddy, L. S. S. and Saikiran, P., 2011. Research Issues in Cloud Computing, Global Journal of Computer Science and Technology, Volume 11, Issue 11.

[440]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Digital Identity Called Selfie-Means of Narcissism, Self-Exploration or Entertainment? A Review Aadil Masood Wani, Benish Ali Bhat* MERC, University of Kashmir, Srinagar, India

Abstract Digital technology has converted Selfie into a phenomenon. This paper seeks to trace the dual nature of a Selfie which simultaneously can be a social discourse and a profound personal communication tool. A Selfie has a unique nature as it can defy time and space by being narrowcast and broadcast at the same time which gives an individual a power to share and portray the ‘self’. This paper firstly looks at the identity created by a Selfie and whether it has the power to bring out egotism in personalities. Secondly, many people tend to look at selfies as a medium of documenting and exploring individuality and the paper gazes into that aspect. Finally, the paper touches the entertainment aspect of the Selfie where celebrities use it as a tool to casually connect with their audiences and they in turn are offered an opportunity by the Selfie to grab their slice of fame.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Selfie; Narcissism; Self Exploration; Online Social Networks

1. Introduction Certain tilt in the front facing cell phone camera, usually about 45 degrees above the eyeline, (Day, 2013) a light source, a good pose or may be a pout, a click, a flattering filter and a new Selfie is ready to go on your social network profiles. Facebook likes and the Instagram hearts are the aim of this new Selfie. According to Berit Skog, 2013, receiving many likes on Facebook can be interpreted as the amount of appreciation gotten from Facebook friends. It subsequently develops an identity and appreciation of the self. Communication solely through the written word or texts is long gone and ‘photographs are dominating all dialogues’ now. (Fuerst, 2013) "Selfie" was named as word of the year in 2013 by Oxford Dictionaries. "Language research conducted by Oxford Dictionaries editors reveals that the frequency of the word Selfie in the English language has increased by 17,000% since this time last year," Oxford wrote in justifying its choice. The choice was definitive and raised a dialogue throughout the world about the importance and the consequent formation of the digital identity by the Selfie. The same dictionary defines a Selfie as, “A photograph that one has taken of oneself, typically with a smartphone or webcam and uploaded to a social media website”. The definition clearly indicates to the technologies that have made Selfies rampant i.e. smartphone or a webcam and social media website. Self-portraits are not a new phenomenon but owing to the growing technology camera id no more a luxury. Robert Cornelius, a ‘Ductch born US chemist’ made the ‘first self-portrait in 1839’ using the ‘daguerreian light process’ (Cahill, 2013). Then came the Polaroid cameras that ‘freed individuals from the photographic dark rooms’ adding ease but the boom of ‘self-portraits came with the invention of the compact digital cameras’ and then came the Selfie revolution when ‘Apple launched their iphone4 with a front facing camera’ (Day, 2013) Smartphones grew and so grew the amalgamation of different technologies. The camera came together in these devices with the vast opportunities of sharing the photographs across social networks. In short people got an access to a platform where they could turn there their daily activities and their looks into viewable commodities and create their own digital identities. R. Swaminathan puts the process precisely as:

* Corresponding author. Tel.: +91 9796 722980. E-mail address:[email protected]. ISBN: 978-93-82288-63-3

Wani and Bhat/ COMMUNE-2015

“What animates a selfie from a mere picture to a complex life world of multi-vocal meanings is its unique ecosystem of production, distribution, and consumption. The camera-embedded smartphone is its means of production, mobile Internet-enabled social media is its distribution platform and its consumption is through a variety of networked smart devices. This unique digital DNA of the selfie, analogous to a human genetic code, allows it to exist in multiple forms and spaces simultaneously” 2. Selfie and Narcissism The lookout for self-worth and admiration is a part of human nature but as it slips out of hand, narcissism comes into play. ‘Narcissus a beautiful Greek youth’ fell in love with his own reflection and pined away ‘yearning for a mere image’ and so the word narcissism has its roots in there and it is ‘recognized as a psychiatric personality disorder’ by the American Psychiatric Association (Race, 2002). Given the rise of social networking sites and the smartphones with cameras that make them accessible at all times and all places, ‘Narcissists need not wait’ until others are available to ‘engage in self-aggrandizement, but can instead curate, manage, and promote an online “self” throughout the day’ by posting selfies (Panek, Konrath, 2013). According to Mark R. Leary, Professor of Psychology and Neuroscience at Duke University and author of The Curse of the Self: Self-Awareness, Egotism, and the Quality of Human Life i “By posting selfies, people can keep themselves in other people’s minds. In addition, like all photographs that are posted on line, selfies are used to convey a particular impression of oneself. Through the clothes one wears, one’s expression, staging of the physical setting and the style of the photo, people can convey a particular public image of themselves, presumably one that they think will garner social rewards.” The realisation of self-worth through the number of likes or comments you get is common among the youth today. And across the various platforms like Facebook, Instagram and Snapchat etc. the kind of selfies posted differ. According to José van Dijckii , Professor of Comparative Media Studies at the University of Amsterdam, Facebook features mostly normal and ‘ordinary self-portraitures’ while as Instagram is for “stylish selfies or stylies” and Snapchat selfies are more like ‘funny postcards’. Taking ‘excessive selfies and posting’ them has also been seen as a ‘psychiatric disorder’ (Graham, 2014). In an Article written by Graham for Mail Online, Dr David Veale, a consultant psychiatrist at the South London and Maudsley NHS Trust and The Priory Hospital, points out that people taking excessive selfies suffer from Body Dysmorphic Disorder (BDD)iii and says, “Two out of three of all the patients who come to see me with BDD since the rise of camera phones have a compulsion to repeatedly take and post selfies on social media sites.’ In the same article Graham refers to a British teenager Danny Bowman who tried to ‘commit suicide because he was unsatisfied with his appearance in the selfies he took’. According to recent findings from the Pew Research Centre, ‘teenagers in America are sharing more information than ever’ about themselves on social media (BBC News, 2013) and of those studied, 91% post photos of themselves online - up from 79% in 2006. Since narcissistic disturbance involves an ‘intense need to gain recognition and admiration through some form of exhibiting one’s self’, Selfie coupled with social media allows for an endless opportunity to gain both the ‘superficial attention that a narcissistic person may crave’, as well as an ‘easy avenue for manipulating one’s image’ (Beck, 2014). Seeking ‘validation is normal’ but with social media it can easily ‘spiral out of control’ and become a narcissistic addiction says Dr Jill Weber a psychologistiv . Getting liked on Facebook or Instagram seems to fill the individual with ‘reassurance and approbation’ which is very ‘addictive’ and the ‘cycle is repeated’ (Day, 2013). 3. Selfie as Means of Self-Exploration Not all find Selfies useless means of show-off but certain scholars find in them an ability to give expression and to provide a better perspective to study the human race. For the future generations the selfies might be ‘important historical documents’ to understand the current era (Cahill, 2013). The view that Selfie is nothing more than an ‘outlet for self-expression’, which just happens to be ‘shared more publicly via the communication modes of our times’, is catching up (Sifferlin, 2013). “Self-captured images allow i Oxford University Press blog features the scholarly reflections about their choice of the word of the year. Retrieved from: http://blog.oup.com/2013/11/scholarly-reflections-on-the-selfie-woty-2013/ ii ibid iii BDD is characterised by a preoccupation with one or more perceived flaws in appearance, which are unnoticeable to others, according to the BDD Foundation. As well as the excessive self-consciousness, individuals with BDD often feel defined by their flaw. They often experience an image of their perceived defect associated with memories, emotions and bodily sensations – as if seeing the flaw through the eyes of an onlooker, even though what they ‘see’ may be very different to their appearance observed by others. Sufferers tend repeatedly to check on how bad their flaw is - for example in mirrors and reflective surfaces - attempt to camouflage or alter the perceived defect and avoid public or social situations or triggers that increase distress. iv Taken from an article by Melissa Walker, ‘The Good, the Bad, and the Unexpected Consequences of Selfie Obsession’ written for ‘Teen Vogue’ in August 2013

[442]

Wani and Bhat/COMMUNE – 2015

young adults and teens to express their mood states and share important experiences,” says Dr. Andrea Letamendi, a clinical psychologist and research fellow at UCLA. She adds that developmentally, selfies make sense for children and teens. And for the most part, they are ‘simply reflections of their self-exploration and nothing more’ (Sifferlin, 2013). According to Dr. Rutledge, 2013: “One of the most effective ways to know yourself is to see yourself as others see you. Selfies offer the opportunities to show facets of yourself, such as the arty side, the silly side, or the glamorous side. We learn about people by accumulating information over time. Our understanding of everything, include other people, is a synthesis of all the things we know about them. By offering different aspects through images, we are sharing more of ourselves, becoming more authentic and transparent—things that digital connectivity encourages.” Experts also find that selfies can be used to look into the psychological state of individuals especially ‘adolescents’ and provide a ‘deeper window into their issues’ (Sifferlin, 2013). Selfies are not always about self-absorption, they can easily be ‘artistic expressions’ or ‘fashion statements’ (Rutledge, 2013). By sharing selfies individuals can also be expressing the desire to be from a ‘certain community or a group’ (Tifentale, 2014). Fink, 2014 says that the selfie is a form of documentation, a modern diary and there is nothing inherently wrong or negative about documenting one’s life. This way a selfie can be a catalyst for introspection. A more appropriate reaction to this new form of documentation according to Fink is to attempt to maximize it and use every selfie as an opportunity for self-reflection. 4. Selfie and Entertainment: In Edelman'sv eighth annual study on how and why people consume and share entertainment, it was found consumers in the U.S., UK and China want their entertainment "selfie-style"-- centered on the individual, immediately gratifying, engaging and sharable across social networks (Becker, 2014). Celebrities have always taken to social media to attract their audiences and they are equally fond of Selfies as a medium to reach out. Celebrity Twitter and Instagram profiles are common and filled with their selfies. International celebrities like Rihanna, Justin Bieber, Lady Gaga and Madonna are all ‘serial uploaders’ of selfies (BBC News, 2013). In their ‘unique style’ and ‘casual rawness’, selfies feel more ‘immediate’, ‘intimate and personal’, enhancing the ‘celebrity’s connection’ to their fans (Rutledge, 2013). The famous TV celebrity Kim Kardashian is releasing a 352-page book of curated selfies called Selfishvi . According to R. Swaminathan, Kardashian may not have intended it as such, but the title of the book is richly infused with interpretative possibilities for the fields of social and philosophical anthropology and the complex notions of manufactured self and selfhood. As a relational social construction the selfie is a product of popular culture: the song #selfie by Chainsmokers7 reached the Billboard Top 10vii . It’s also a material, non-material and ‘epistemological foundation’ for creating ‘a sense of meaning’ of daily life and human interactions (Swaminathan, n.d.). As Becker, 2014 puts it in her summary of the Edelmans report, Today's global consumers expect unprecedented control over what they watch and when and where they watch it. They want content that is instantaneous, self-revolving, engaging in the moment and engaging others at their choosing. In short, they want entertainment "selfie-style." 5. Conclusion The selfie is increasingly becoming a symbol of a slightly ‘shifting sense of self’, one that is more aware of how we always function in at least two modes at once, the private and the public, the internal and the external (Alang, 2013). Looking at all the three aspects of the Selfie it can be concluded that its pervasive nature is because of the availability of smartphones and social media and it has changed the way we conceived communication. Whether as narcissism or expression or entertainment Selfie has become an important part of the digital footprint and has given rise to alternate digital identities. The people living in the two worlds- the real and the virtual are maintaining the two images constantly; the selfimage and the selfie-image. Technology aspires to improve human life but there is a thin line between the good it brings and its negatives. An autobiography of the ‘self’ through images and not just words is a powerful tool that technology offers. However, the constant dependence on how others value ‘you’ through their likes and comments destroy the very ‘self’ that one wants to explore. This is where the thin line between self-exploration and narcissism blurs. Albert Einstein quotes, “I fear the day technology will surpass our human interaction. The world will have a generation of idiots.” v

Edelman is an international public relations firm founded and named after Daniel Edelman and currently run by his son Richard Edelman. www.latimes.com/entertainment/gossip/la-et-mg-kim-kardashian-selfie-book-20140808-story.html vii Please watch: http://www.youtube.com/watch?v=kdemFfbS5H0 Accessed on January 05, 2015 vi

[443]

Wani and Bhat/ COMMUNE-2015

Is the Selfie entertainment and self-exploration required or are we actually breeding generations of narcissistic Selfie idiots? The link between Selfie and growing self-love leading to narcissism is an area that needs a proper thought and research. The fact that Selfie was word of the year in 2013 shows the power of an image and subsequently leads us to the understanding that human nature seeks validation and pleasure. The heights of it give it a narcissistic form, moderate use make its self-exploration and the fact that it is just pleasure makes it pure entertainment. References Self-portraits and social media: The rise of the 'selfie'. (June 7 2013). Retrieved January 14, 2014, from BBC News: http://www.bbc.co.uk/news/magazine-22511650 How to Spot a Narcissist Online. (January 16 2014). Retrieved January 17, 2014, from The Atlantic: http://www.theatlantic.com/health/archive/2014/01/how-to-spot-a-narcissist-online/283099/ ALANG, N. (November 26 2013). You are wrong about 'selfies,' they are not proof of narcissism. Retrieved January 17, 2014, from theglobeandmail: http://www.theglobeandmail.com/technology/digital-culture/you-are-wrong-about-selfies-they-are-not-proof-of-narcissism/article15600483/ Beck, J. (January 16 2014). How to Spot a Narcissist Online. Retrieved January 17, 2014, from The Atlantic: http://www.theatlantic.com/health/archive/2014/01/how-to-spot-a-narcissist-online/283099/ Becker, G. (June 19 2014). Entertainment in the Era of the Selfie. The Huffington Post. Blaine, L. (August 14 2013). How Selfies Are Ruining Your Relationships. Retrieved January 17, 2014, from Time NewsFeed: http://newsfeed.time.com/2013/08/14/how-selfies-are-ruining-your-relationships/ Cahill, D. (October 17 2013). No filter needed: The origin of the Selfie. New York Post. Retrieved january 3, 2015, from http://nypost.com/2013/10/17/the-art-of-taking-selfies-is-nothing-new/ Connell, J. O. (December 11 2013). Selfie, word of 2013, sums up our age of narcissism. Retrieved January 17, 2014, from Irishtimes: http://www.irishtimes.com/life-and-style/selfie-word-of-2013-sums-up-our-age-of-narcissism-1.1623385 Cross, T. (November 8 2013). The Culture of Now – The rise of imagery in social media. Retrieved January 15, 2014, from The Wall: http://wallblog.co.uk/2013/11/08/the-culture-of-now-the-rise-of-imagery-in-social-media/ Day, E. (July 14 2013). How selfies became a global phenomenon. The Guardian. Fink, R. E. (January 5 2014). Is the selfie narcissism at its finest? Retrieved Janauary 17, 2014, from HAARETZ: http://www.haaretz.com/jewishworld/rabbis-round-table/.premium-1.567136 Fuerst, E. (2013). Cultural Hybridity: Remix and Dialogic Culture. Retrieved January 12, 2014, from blogs.commons.georgetown.edu: https://blogs.commons.georgetown.edu/cctp-725-fall2013/2013/11/04/social-medias-impact-on-photography/ Gervais, S. J. (January 22 2013). Does Instagram Promote Positive Body Image. Retrieved January 17, 2014, from Psychology Today: http://www.psychologytoday.com/blog/power-and-prejudice/201301/does-instagram-promote-positive-body-image Graham, S. (April 10 2014). Take a lot of selfies? Then you may be MENTALLY ILL: Two thirds of patients with body image disorders obsessively take photos of themselves. Retrieved from MailOnline: http://www.dailymail.co.uk/sciencetech/article-2601606/Take-lot-selfies-ThenMENTALLY-ILL-Two-thirds-patients-body-image-disorders-obsessively-photos-themselves.html Kaufman, M. T. (February 24 2003). Robert K. Merton, Versatile Sociologist and Father of the Focus Group, Dies at 92. The New York Times. Kleinman, Z. (August 16 2010). How the internet is changing language. BBC News. Knox, S. (October 30 2013). Enough with the selfies already. Retrieved January 18, 2014, from live4: http://www.live4.com.au/enough-with-theselfies-already/ LOSSE, K. (June 5 2013). THE RETURN OF THE SELFIE. Retrieved January 5, 2014, from New Yorker: http://www.newyorker.com/online/blogs/elements/2013/06/the-return-of-the-selfie.html McIntosh, A. (September 15 2013). NARCISSISTS LIKE SOCIAL MEDIA. Retrieved January 17, 2014, from murdochindependent: http://www.murdochindependent.com.au/narcissists-like-social-media/ Studies in world Chritianity. (n.d.). Retrieved from Project MUSE:https://muse.jhu.edu/journals/studies_in_world_christianity/summary/v014/14.2.malhone.html NEWS, B. (June 7 2013). Self-portraits and social media: The rise of the 'selfie'. BBC News Magazine. Nicola Bruno, M. B. (Febuary 6 2013). Self-Portraits: Smartphones Reveal a Side Bias in Non-Artists. Retrieved January 18, 2014, from plosone: http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0055141 Panek, N. K. (2013). HOW NARCISSISM DIFFERS ON FACEBOOK AND TWITTER. Retrieved January 17, 2014, from sarakonrath: http://www.sarakonrath.com/media/publications/narcissism_SNS_sites.pdf Panek, N., & Konrath, S. H. (2013). HOW NARCISSISM DIFFERS ON FACEBOOK AND TWITTER. Retrieved January 15, 2104, from Sarakonrath: http://www.sarakonrath.com/media/publications/narcissism_SNS_sites.pdf Pearse, D. (March 17 2012). Facebook's 'dark side': study finds link to socially aggressive narcissism. Retrieved January 17, 2014, from Guardian: http://www.theguardian.com/technology/2012/mar/17/facebook-dark-side-study-aggressive-narcissism Race, T. (July 29 2002). New Economy; Like Narcissus, executives are smitten, and undone, by their own images. Retrieved january 15, 2014, from New Yor Times: http://www.nytimes.com/2002/07/29/business/new-economy-like-narcissus-executives-are-smitten-and-undone-by-their-ownimages.html?src=pm Rawlings, K. (Novemnber 21 2013). Selfies and the history of self-portrait photography. Retrieved January 14, 2014, from OUP blog: http://blog.oup.com/2013/11/selfies-history-self-portrait-photography/ Rutledge, P. (April 18 2013). #Selfies: Narcissism or Self-Exploration? Retrieved January 17, 2014, from Psychology Today: Here to Help: http://www.psychologytoday.com/blog/positively-media/201304/selfies-narcissism-or-self-exploration Sifferlin, A. (September 6 2013). Why Selfies Matter. Retrieved from http://healthland.time.com/2013/09/06/why-selfies-matter/ Skog, B. (February 27 2013). What's the thing about "Like!" on Facebook. Retrieved from Popularsocialscience.com: http://www.popularsocialscience.com/2013/02/27/whats-the-thing-about-like-on-facebook/ Slavin, L. (January 17 2014). The Evolution of Selfie Culture: Self-Expression, Narcissism, or Objectification? Retrieved January 18, 2014, from feminspire: http://feminspire.com/the-evolution-of-selfie-culture-self-expression-narcissism-or-objectification/ Swaminathan, R. (n.d.). Self, Selfhood and a Selfie: The Anatomy of a Virtual Body and Digital Identity. Tifentale, A. (February 2014). The Selfie: Making sense of the “Masturbation of Self-Image” and the "Virtual Mini-Me". Retrieved from Selfiecity.net: http://d25rsf93iwlmgu.cloudfront.net/downloads/Tifentale_Alise_Selfiecity.pdf Titlow, J. P. (January 31 2013). #Me: Instagram Narcissism and The Scourge Of The Selfie. Retrieved January 17, 2014, from REadWrite: http://readwrite.com/2013/01/31/instagram-selfies-narcissism#awesm=~otc2NlnjHXOpwC

[444]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Experimental Study of Different Wavelets for Real Time Environmental Monitoring in Wireless Visual Sensor Networks Umar Farooq, Shabir Ahmad Sofi*, Roohie Naaz Department of Information Technology, NIT Srinagar, India

Abstract Wireless Sensor Networking is a modern information gathering technology that has a wide range of applications in environmental monitoring. However due to the fact that sensor nodes have limited battery and are deployed in remote and harsh environments, energy efficient and real time transmission of the information are still open challenges. Since in a wireless sensor network data transmission is the most power consuming task (80% approx.), so far most useful techniques for the purpose of energy efficient and real time transmissions are based on data compression, the majority of them are based on wavelet transform. In this paper we describe the robust use of wavelet transform and make the Performance analysis of different wavelets for energy efficient and real time image data transmission in environment monitoring using wireless visual sensor networks. Our performance evaluation shows that compared to other wavelets, Haar wavelet is better in terms of energy efficiency and transmission time.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Wireless Visual Sensor Network ; Discrete Wavelet Transform ; Skipped High-pass Sub-band ; Image Compression.

Introduction Wireless Visual sensor Network (WVSN) is a self-organized intelligent system and consists of a number of visual sensor nodes (VSNs) deployed over large environmental area that can be used to monitor environment ( Akyildiz et al 2002)(Guang-yu et al, 1999). These nodes are capable of sensing, processing, and transmitting the gathered information. Fig.1 shows the Wireless Sensor Network architecture that can be used for environmental monitoring. Sensor nodes will communicate with each other and transmit the processed data to sink node over a wireless communication link. Sink node collects data from all the nodes, and transmits the analyzed data to user via Internet (Mhatre and .Rosenberg, 2004). WVSNs are unique and more challenging as compared to other sensor networks as they produce two-dimensional data whereas in other sensor networks, the scalar data is produced. Due to large amount of data, WVSNs place more stringent requirements on power bandwidth and processing capabilities. Under the context of environmental monitoring, WVSNs are to be deployed in remote areas therefore changing battery becomes impractical and assiduous task. Hence, the energy consumption of the sensor nodes is proving to be a critical constraint in employing WVSNs. Energy consumption operation of a wireless sensor node comprises of sensing, processing, and transmission. Among these operations, data transmission is the most power consuming task (80% approx.) and it is widely accepted that the energy consumed in one bit of data transfer can be used to perform a large number of arithmetic operations in the sensor processor (Pottie and. Kaiser, 2000)( Anastasi, et al, 2009). Moreover, in a densely deployed sensor network, the physical environment would produce very similar data in near-by sensor nodes and transmitting such data is more or less redundant. Under these energy constraint conditions, it is useful to transmit only a limited amount of data bits by compressing the data using the different compression techniques like Discrete Cosine Transform, Wavelet Transform in order to reduce the energy consumption( Farooq, 2014; VijendraBabu, et al , 2008).In this context, image (2-D signal) transmission over WSN has been done mainly after using the compression algorithm based on Wavelet transform in order to reduce number of bits used to represent the image by eliminating the various redundancies, thereby reducing the energy consumption in image transfer. Adaptive compression based congestion control technique (JH and IB 2010) is one of the approaches utilizing the wavelet transform to reduce the *

Corresponding author. Tel.: +91 9419 009971. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Farooq et al /COMMUNE – 2015

number of packets so as to provide efficient energy consumption mechanism. In work (Lecuire et al, 2007) open loop and closed loop image transmission schemes based on Discrete Wavelet transform (DWT) were proposed which provide graceful trade-off between image quality and energy utilisation to transmit the image data. Skipped High-pass Sub-band (SHPS) technique (Nasri, et al, 2010) is one of the effective approaches of image transmission utilising DWT. A distributed algorithm based on lifting scheme (Ciancio and Ortega, 2004) to decorrelate the collected data at nodes by the exchange of data among other sensor nodes in network path is also an efficient approach of reducing overall energy consumption using the right trade off among local processing and transmission operations.

Fig.1 Architecture of Wireless Visual sensor Network

2. Design of Wireless Visual Sensor Network 2.1

System Overview

The system overview of the WVSN for monitoring the environment is shown in Fig.2.The architecture consists of three parts: (i) Data monitoring nodes in the sub region. (ii) Base station and (iii) Data monitoring centre. The sensor nodes are deployed in remote environment area where they collect the information of that environment in the form of image data. This image data after suitable processing techniques is sent to the monitoring centre. In each sub region, wireless network based on ZigBee technology ( Willig , 2005)is developed. ZigBee technology is preferred because it consumes significantly less power and is intended to be simpler and less expensive which makes it more suitable for wireless sensor network applications. The working frequency bands of ZigBee include 868 MHz (for Europe), 915 MHz (for USA), and 2.4 GHz (Global) ISM frequency bands.

Fig. 2 System Overview of WVSN for environment monitoring.

[446]

Farooq et al /COMMUNE – 2015

In each sub region, base stations based on the ZigBee gateway and CDMA transmission channel are deployed to perform the data collection and condition monitoring of sensor nodes in ZigBee network and transmit the parameters of the sub region to monitoring centre by CDMA wireless network. 2.2

Design of Data Monitoring Node

The system design of a data-monitoring node is shown in Fig. 3. The design consists of several units/modules such as data sensing unit, processing unit, radio frequency unit and power supply unit. The sensing unit consists of the sensors and an Analog to Digital Convertor (ADC) for data acquisition. Type of the sensors to be used depends on the application of the sensor network e.g. in case of environment monitoring the sensors required are temperature sensor, humidity sensor, Pressure sensor etc. Processing unit includes a microcontroller and memory for processing and storing of data. Power module consists of tiny battery to provide the necessary energy for other units of the monitoring node. Each node is connected with and controlled by the ZigBee communication protocol (Willig, 2005)

Fig.3 A typical data monitoring node

3. Wavelet Analysis Wavelet analysis has emerged as an important milestone in the field of spectral analysis because of its multiresolution and localisation capabilities in time as well as frequency domain. In contrast to a wave, wavelets are localized waves and have their energy concentrated in time or space and are suited to analysis of transient signals (Mallat, 1999 )( Mathieu and Daubechies,1992). The wavelet decomposition of the signal at various frequencies reveals the low and high frequency components thereby helps in localising these features in time domain. There are number of algorithms that are designed to perform such decomposition and the selection of particular algorithm depends on the application at hand. For example in case of processing of medical images or seismic signals, Continuous Wavelet Transform (CWT) is used which calculates the wavelet transform as given by the equation 3.1, where x(t) is the signal to be analyzed. Ψ(t) is the mother wavelet or the basis function. XWT (τ,s) =

𝟏 √|𝒔|

∫ 𝒙(𝒕) · Ψ*(

𝒕− 𝝉 𝒔

)dt

(3.1)

All the wavelet functions used in the transformation are derived from the mother wavelet through translation (shifting) and scaling (dilation or compression).The mother wavelet used to generate all the basis functions is designed based on some desired characteristics associated with that function. The translation parameter τ relates to the location of the wavelet function as it is shifted through the signal and hence corresponds to the time information in the Wavelet Transform. The scale parameter ‘s’ corresponds to frequency information of the transform. Discrete Wavelet Transform (DWT) is an alternate to the CWT for certain applications like image and signal compression and is usually preferred over other transforms. Although it is not the only possible choice, Discrete Wavelet transform (DWT) coefficients are usually sampled from the Continuous Wavelet Transform (CWT). There are a number of basis functions that can act as the mother wavelet for Wavelet Transformation. Since the mother wavelet produces all wavelet functions used in the transformation, it determines the overall characteristics of the resulting Wavelet Transform. Based on the number of coefficients and vanishing moments each family of the wavelets are having wavelet subclasses. Haar wavelet is the oldest and simplest wavelet. The number of vanishing moments in this wavelet is one(1). Daubechies wavelets are also called Maxflat wavelets as their frequency responses have maximum flatness at

[447]

Farooq et al /COMMUNE – 2015

frequencies 0 and π. These wavelets are completely supported with extreme phase. Sym wavelet (Symlet) is both orthogonal and biorthogonal and have compact support and has highest number of vanishing moments for given support width. Table I illustrates the characteristics of some of the wavelets which are investigated in this paper. Each of the wavelet is having a different filter length which is an important property in determining the computational cost. Obviously the wavelet filter length is directly proportional to the computational cost (Abbate et al, 1995). Table I Characteristics of wavelet functions where N is the order

3.1

Wavelet

Support width

Filter length

Orthoganality

Haar

1

2

Yes

Daubechies

2N-1

2N

Yes

Coiflet

6N-1

6N

Yes

Bior

2Nd+1 for decomp. 2Nr+1 for reconst.

Max(2Nr,2Nd) +2

No

Algorithm implementation using Wavelet Transform.

The algorithm of data transmission consists of two modules viz. encoder and decoder module as shown in Fig. 4

(a)

(b)

Fig. 4 (a) Encoder module (b) Decoder module

Image is transformed into matrix form so as to make image suitable for compression before applying the Discrete Wavelet Transform (DWT) which separates the data signal into fine-scale information known as detail coefficients and rough-scale information known as approximate coefficients ( Abbate, 1995). Since image is typically a two dimensional signal, a 2-D equivalent of the DWT is performed. This is achieved by first applying the Low Pass (L) and High Pass (H) filters to the lines of samples, row-by-row and then re-filtering the columns of the output by the same filters. As a result, the image is divided into 4 sub-bands, LL, LH, HL and HH. The LL sub-band contains the low-pass information and the others contain high-pass information of horizontal, vertical and diagonal orientation. The LL sub-band provides a half-sized version of the input image which can be transformed again to have more levels of resolution. After applying DWT, next step is to perform SHPS technique. In SHPS, attempts to conserve energy are made by skipping the least significant sub band Hi in each transform level. The low pass sub bands are further decomposed leading to LLi and LHi sub bands. By skipping two out of every four sub bands, SHPS technique reduces computational loads and data to be transmitted because only LLi and LHi sub bands are computed. The next step is quantification of the sub-bands which is used to reduce the number of bits needed to represent the image. The next step is Coding of the quantised coefficients. We use here Huffman Coding which is one of the simplest compression techniques .The main idea behind the use of this coding technique is that on one hand it gives better results in terms of compression ratio and the average bits per character than run length Encoding, Shanon-Fano coding and dictionary based techniques like Lempel Ziv Scheme (LZ77 and LZ78) (Shanmugasundaram and Lourdusamy, 2011). On other hand experimental results have revealed that although the compression ratio of the arithmetic coding for different image sizes is higher than the Huffman

[448]

Farooq et al /COMMUNE – 2015

coding but comparatively less execution time is taken by Huffman coding ( Shahbahrami et al , 2011). Finally data packets are transmitted over the Wireless Sensor network. At the decoder end, after receiving the bit stream, the first step is Inverse Huffman decoding and Finally Inverse Discrete Wavelet Transform (IDWT) is applied to these decoded coefficients to recover the image. 4. Performance Analysis In this section we implement and test the encoder and decoder module for the sensor node based on Imote2 platform for different gray scale images in MATLAB. We evaluate the energy consumption and transmission time for different wavelets using the same transmission scenario (Fig.5). Encoding and decoding is performed respectively at the source and destination node.

(a)

(b)

Fig.5(a) Original images (b) Transmission Scenario

4.1

Energy Analysis

For the energy analysis of the transmission system for different wavelets, we consider gray scale Lena image of size 128×128 with the distance between the source and destination node as 10m. The results are shown in the Table II which shows that Haar wavelet is the best option for energy efficient data transmission as the minimum energy is consumed in that case. Table II Comparative analysis of energy consumption for different wavelets Wavelet

Compression ratio 4.21 4.0 3.65 3.07 3.76 2.91

Haar db1 db4 db10 coif1 coif4

4.2

Energy(mJ)

Wavelet

Compression ratio

Energy(mJ)

19.54 20.57 22.55 26.76 21.88 28.25

Sym1 Sym8 bior1.1 bior2.2 bior2.4 dmey

4.0 3.25 4.0 3.76 3.54 1.26

20.57 25.32 20.57 21.88 23.22 65.27

Transmission Time Analysis

Table III shows the transmission time analysis of the system for different wavelet functions employed in the encoding and decoding with same transmission scenario as used in energy analysis. Table III Comparative analysis of transmission time of 128×128 image at 230 kbps for different wavelets Wavelet Haar db1 db4 db10 coif1 coif4

Transmission time (Second) 0.41 0.43 0.46 0.56 0.46 0.59

Wavelet Sym1 Sym8 bior1.1 bior2.2 bior2.4 dmey

Transmission time (Second) 0.43 0.53 0.43 0.46 0.48 1.36

From the table it is quite clear that least amount of transmission time is achieved in case of employing Haar wavelet functions in the encoding and decoding modules of the sensor node

[449]

Farooq et al /COMMUNE – 2015

5. Conclusion In this work encoder and decoder module of a sensor node employing wavelet transform technique was implemented and investigated for different wavelets in MATLAB. Based on the experimental study of image data transfer from source node to destination node, the most suitable mother wavelet was found and it was Haar wavelet, which proved to be the best option as compared to other wavelets, which were investigated in this work. The study revealed that minimum energy consumption and least amount of transmission time for data transfer between the two nodes were achieved when Haar wavelet was employed in the encoding and decoding module. In future, this work could be extended for multimedia wireless sensor networks (MWSNs) and similar type of analysis could be made for packet losses, memory usage and execution time. References A., C.M. Decusatis and P.K. Das, Wavelets and Subbands: fundamentals and application, 1995. Graps, “An Introduction to Wavelets”, IEEE Computational Science & Engineering Magazine, Vol. 2, June 1995, pp. 50 –61. Alexandre Ciancio and Antonio Ortega, “A Distributed Wavelet Compression Algorithm for Wireless Sensor Networks using Lifting,” Proceedings of the International Conference on Acoustics, Speech and Signal Processing, Canada, May 2004. Antonini, Barlaud, Mathieu and Daubechies, “Image Coding using Wavelet Transform,” IEEE Transactions on Image Processing, April 1992, pp. 205-220. Asadollah Shahbahrami, Ramin Bahrampour, Mobin Sabbaghi, Mostafa Ayoubi, “Evaluation of Huffman and Arithmetic Algorithms for Multimedia Compression Standards,” International Journal of Computer Science, Engineering and Applications, 2011, pp. 34–47. D VijendraBabu, N R Alamelu, P Subramanian, N Ravikannan, “EBCOT using Energy Efficient Wavelet Transform,” International Conference on Computing, Communication and Networking (lCCCN), 2008. G. Anastasi, M. Conti, M. Francesco, A. Passarella, “Energy Conservation in Wireless Sensor Networks: a Survey,” Ad-Hoc Networks, Vo1.7, Issue 3, 2009, pp. 537-568. G.J. Pottie and W.J. Kaiser, “Wireless Integrated Network Sensors”, Communications of the ACM, 2000, pp. 551–558. Ian F. Akyildiz, Weilian Su, Yogesh Sankarasubramaniam and Erdal Cayirci, “A Survey on Sensor Networks,” IEEE Communications Magazine, August 2002, pp. 102-114. Karl H, Willig A, “Protocols and Architectures for Wireless Sensor Networks,” John Wiley publishing, 2005. Lee JH and Jung IB, “Adaptive Compression Based Congestion Control Technique for Wireless Sensor Networks,” Sensors (Basel), 2010, pp. 2919– 2945. Li Guang-yu, Ye Si-yuan, Zhang Zheng-xian and Gao Zong-jun, “Study and Protection of Wetland,” Marine Geology Letters, 2005, pp. 8-11. Mallat, “A Wavelet Tour of Signal Processing,” 2nd ed. Academic Press, 1999. Mohsen Nasri, AbdelHamid Helali, Halim Sghaier and Hassen Maaref, “Energy Efficient Wavelet Image Compression in Wireless Sensor Network,” International Conference on Wireless and Ubiquitous Systems, Tunisia, 8-10 October 2010. Senthil Shanmugasundaram and Robert Lourdusamy, “A Comparative Study Of Text Compression Algorithms,” International Journal of Wisdom Based Computing, Vol. 1, December 2011, pp. 68-76. Umar Farooq, Jyoti Saxena and Shabir ahmad Sofi, “Wavelet Transform based Effective Energy Utilisation Approaches of Data Transfer in Wireless Sensor Networks: A Survey,” Proceedings of ICAET, Roorke, India, May 2014, pp. 599-604. V. Mhatre and C. Rosenberg, “Design Guidelines for Wireless Sensor Networks: Communication, Clustering and Aggregation,” Ad Hoc Networks, vol. 2, No. 1, Jan. 2004, pp. 45-63. Vincent Lecuire, Cristian Duran Faundez and Nicolas Krommenacker, “Energy Efficient Transmission of Wavelet Based Images in Wireless Sensor Networks,” Eurasip journal on Image and Video Processing, 2007.

[450]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Search Interfaces of Select Online Public Access Catalogues: An Assessment Huma Shafiq*, Sheikh Mohammad Shafi, Shazia Rasool, Tariq Shafi, Zahid Ashraf Wani Department of Library & Information Science, University of Kashmir, Hazratbal, Srinagar, India

Abstract Purpose –This study is an attempt to explore and assess the search interfaces of select online public access catalogs viz WorldCat, Library of Congress Online Catalog, and NLM’s LocatorPlus. Design/methodology/approach – Due to the adoption of various information & communication, technologies by libraries in the late 80's and early 90's world over, especially in developed countries like US, the search and retrieval paradigm of library resources by users have changed considerably. keeping this as the background, the authors have undertook this study to assess and explore the new possibilities opened up by the revolutionary breakthroughs in ICT which have provided the users with new vistas for searching, retrieving and using the resources offered by the libraries to its users. To assess the changing trends in locating, searching, accessing and retrieving information, the three prominent Online Public Access Catalogs (OPACs) from Libraries in United States of America are selected for the study (WorldCat, Library of Congress Online Catalog and National Library of Medicine's (NLM’s) LocatorPlus) because of the technologically advanced nature of the users served by these famous libraries. Search Interfaces of the select OPACs are explored and assessed based on six parameters: Query Formulation, Search Limits, Help Mechanism, Alerting Services, Result Manipulation, and Additional Search Features. Findings – After assessing the search interfaces on the basis of six parameters, it is found that among the three OPACs, only Library of Congress Online Catalog has at least proved better than the other two (WorldCat and LocatorPlus) in terms of search interface features. It is apparent from the study that there is a great need to improve the search interfaces of these select OPACs as the results are not too promising due to the lack of some basic and important information retrieval features. Originality/value – The paper makes an endeavor to explore and assess the search interfaces of select Online Public Access Catalogs used worldwide.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Information Retrieval; Search Interface; Online Public Access Catalogs; WorldCat; NLM’s; LocatorPlus; Congress Online Catalog

1. Introduction An enormous amount of information is being produced and published every day. This information, which is as difficult to retrieve as is to manage, is what is called as information explosion. Whenever a piece of information is to be retrieved, a large number of results are obtained. Some of this information is pertinent to the needs of the user while some of it is a mere noise (Ryder, 2011). In order to retrieve the useful information effectively and efficiently, there must be a proper knowledge of how to retrieve the relevant information (Kules, Wilson & Shneiderman, n.d). The role of the search user interface is to aid in the searcher’s understanding and expression of their information needs, and to help users formulate their queries, select among available information sources, understand search results, and keep track of the progress of their search (Hearst, 2011). Each search interface varies in its content, presentation style and query proficiencies (He, Meng, Yu & Wu, n.d). With the intention of creating more effective search interfaces, human computer interaction experts and web designers have been developing novel interactions and features that enable users to conveniently visualize, parse, manipulate, and organize their Web search results (Wilson, Kules, Schraefel &

* Corresponding author. Tel.: +91 9622 909266. E-mail address: [email protected].

ISBN: 978-93-82288-63-3

Shafiq et al/COMMUNE – 2015

Shneiderman, 2010). An efficient information retrieval is based on the knowledge of search interfaces of the information resources and the proper use of these search interfaces. Search interfaces can be either simple or complex. For a past decade now, there has been many considerable changes in the ways and methods adopted by users to acquire the information they need. Due to the obvious outburst of information and the omnipresence of computer access, it has become much easier for the academic community to have access to this significant amount of information at their homes without personally going to the library. A vast number of online resources become available on a single click (Hiller, 2002). It is obvious that every online resource has different search interface. The present study is based on the exploration and evaluation of these search interfaces of select Online Public Access Catalogs. An Online Public Access Catalog (OPAC) is an online storehouse holding the list of all the documents available in the library or even group of libraries. OPAC helps users to search all types of documents (like Journals, Digital Materials, Books, etc.). Users primarily search OPACs to locate the exact position of documents in the library (Online Public Access Catalog, 2014). 2. Review of Literature Internet primarily started as a communication tool and now has advanced and changed significantly over the years thus, turns out to be a very vital information resource (Brinkley & Burke, 1995). Various studies, including (Berazi, 1981), (Stonier, 1991), (Aina, 2004) and (Bello and Ajala, 2004) provide proofs of the emergence of information as the most important factor in modern industrial systems (Nnadozie, 2008). New knowledge is most often produced by combining information from different sources. To produce knowledge, therefore, requires information retrieval skills. An information retrieval skill is defined as the ability to find information in such a way that non-relevant data (noise) are excluded while relevant information is found (Wien, 2000). The information that is accessible and the array of topics covered have grown to massive proportions (Brinkley & Burke, 1995; Vickery & Vickery, 1993). It has become a growing concern for users at their individual levels as well as at the organizational levels to retrieve and represent related data from enormous amount of information reservoirs. Manual retrieval systems have become outdated and unproductive to a large extent due to effective information access and sharing all over the globe with no limitation to organizational or even geographical boundaries. Thus, there is an immense necessity for effective IT-based retrieval system support (Hu, Ma & Chau, 1999). OPAC’s interface and its searching and retrieving abilities together act as a gateway to library resources and determines the user’s success in retrieving the required information. Users recognize, select and obtain the library resources with the help of bibliographic information provided by OPACs. Thus, the effectiveness in OPAC’s bibliographic display is essential for the users in order to retrieve the precise information (Mi & Weng, 2008). These OPACs started appearing in the later stage of 1990s. Since then, various libraries have implemented these and many others are considering their implementation. These online catalogues prove to be more advanced than the traditional ones due to innumerable features particularly its potential to assimilate a number of document types and sources through a single interface and in terms of providing remote access to the users (Babu & O’Brien, 2000). The way traditional catalogues used to be accessed has been totally changed over the years with the advent of a new search tool called as OPAC. It provides a number of different techniques for searching the same data, thus adding a layer of functionality. It acts as an inherently rich tool offering vast number of search features as compared to traditional catalogues (like card catalogue). It is capable of providing fast, easy and enhanced access to the users even from the remote areas hence saving their time, besides incorporating all the information of library including circulation or new arrivals (Sridhar, 2004). A study discussing library OPACs divulge that OPACs can be practically more useful by reducing the potential complexity of information to a certain manageable degree of simplicity (Wells, 2007). Many studies have been done which illustrate that OPAC’s default search options can affect users while retrieving information, thus there is a need to provide customizable search facilities in OPACs. Another study on OPACs such as Ohio State University’s Gateway and Okapi reveals that the improvements in OPACs are very promising as these systems facilitate users to their effective use without having any knowledge of resources recorded in the catalog or even without any support from experts (Thompson, Pask, Peterson & Haynes, 1994). 3. Data Analysis and Interpretation The three Online Public Access Catalogs (OPACs) are selected from United States of America (USA) and are prominently used worldwide. The OPACs are assessed on the basis of six parameters namely; (1) Query Formulation, (2) Search Limits, (3) Help Mechanism, (4) Alerting Services, (5) Result Manipulation, and (6) Additional Features. 3.1 Query Formulation Query formulation involves checking of search levels (Simple and Advanced Search) and search techniques (Keyword, Phrase, Boolean, Truncation, etc.) supported by the interfaces of the select OPAC's. All the OPACs are analyzed on the basis of these levels and techniques of query formulation. It is evident from Table 1 that under search levels both simple and advanced search facilities are available in all catalogs. While analyzing search techniques, six

[452]

Shafiq et al/COMMUNE – 2015

different search techniques are recognized. Among the six techniques, WorldCat and Library of Congress Online Catalogue support five of them (Keyword, Phrase, Truncation, Boolean and Parentheses) while LocatorPlus supports only three techniques (Keyword, Phrase and Boolean Search). ‘Proximity Operators’ technique is not supported by any of the three catalogues. Table 1: Query Formulation

Levels

Search Techniques

Simple Search Advanced Search Keyword Search Phrase Search Truncation Search Boolean Operators Parentheses Proximity Operators

WorldCat

LocatorPlus (NLM)

1 1 1 1 1 1 1 0

1 1 1 1 0 1 0 0

Library of Congress Online Catalog 1 1 1 1 1 1 1 0

*1 stands for AVAILABLE, 0 stands for NOT AVAILABLE

3.2 Search Limits While accessing the select OPAC's for retrieving the desired information, a large number of results are obtained pertaining to a particular query. In order to retrieve the most relevant information, the results need to be limited and these results can be refined by using various search limiting techniques provided by many information retrieval tools. Table 2 presents an assessment of OPACs on the basis of such search limiters. The OPACs were analyzed through ten different search limiters as listed in Table 2. A maximum of seven (Year, Material Type, Format, Location, Language, Publication Status, and Place of Publication) are supported by LocatorPlus while WorldCat and Library of Congress Online Catalogue support only five of them, (Year, Format, Language, Audience, and Content) and (Year, Material Type, Location, Language, and Place of Publication) respectively. ‘Publication Date’ search limit is offered by none of the catalogues. Table 2: Search Limits Search Limiters

WorldCat

Year Material Type Format Location Language Publication Status Place of Publication Publication Date Audience Content

LocatorPlus (NLM)

1 0 1 0 1 0 0 0 1 1

1 1 1 1 1 1 1 0 0 0

Library of Congress Online Catalogue 1 1 0 1 1 0 1 0 0 0

*1 stands for AVAILABLE, 0 stands for NOT AVAILABLE

3.3 Help Mechanism Help section in any information retrieval tool aids the user in searching the relevant information efficiently and effectively. Almost all information retrieval tools offer this facility. The three OPACs are assessed on the basis of various parameters of help mechanism. It is evident from Table 3 that six different help options exist in the OPACs under study. Library of Congress Online Catalog supports five help options (Search Assistance/Guide, F.A.Q., Feedback Facility, Contact Facility, and Instant Chat). WorldCat supports three of them (Search Assistance/Guide, Feedback Facility, and Online Tutorials) and LocatorPlus supports the least number of help options i.e. only two (Search Assistance/Guide, and Online Tutorials). Table 3: Help Mechanism WorldCat Search Assistance/Guide F.A.Q. Feedback Facility Contact Facility Instant Chat Online Tutorials

LocatorPlus (NLM)

1 0 1 0 0 1

1 0 0 0 0 1

*1 stands for AVAILABLE, 0 stands for NOT AVAILABLE

[453]

Library of Congress Online Catalog 1 1 1 1 1 0

Shafiq et al/COMMUNE – 2015

3.4 Alerting Services Alerting service keeps the user up-to-date about the latest information of his/her interest that has been added to the resource. Table 4 compares these select OPACs on the basis of various alerting services provided by them. Out of the three alerting services, Library of Congress Online Catalog supports two types (RSS Feed, and Email), and WorldCat supports only one (Email) while LocatorPlus does not support any of the three alerting services. One of the alerting services ‘Atoms Feed’ is not supported by any of the OPACs. Table 4: Alerting Service WorldCat

LocatorPlus (NLM)

0 1 0

0 0 0

RSS Feed Email Atoms Feed

Library of Congress Online Catalog 1 1 0

*1 stands for AVAILABLE, 0 stands for NOT AVAILABLE

3.5 Result Manipulation Result manipulation involves controlling the results as per the convenience of the user like reducing the results displayed per page, or sorting the results as required. The selected OPACs are evaluated on the basis of result manipulation options listed in Table 5. It is evident from Table 5 that LocatorPlus supports both result manipulation options while the other two catalogs only support one of them, WorldCat supports ‘Sorting Options’ facility and Library of Congress Online Catalog supports ‘Result Display per page’ feature. Table 5: Result Manipulation

Results Display per page

WorldCat 0

Sorting Options

LocatorPlus (NLM) 1

1

1

Library of Congress Online Catalog 1 0

*1 stands for AVAILABLE, 0 stands for NOT AVAILABLE

3.6 Additional Search Features Additional search features include all other features (e.g. Quick Search, Allied Links) offered by the OPACs that are not included in any of the five parameters of search interfaces analyzed above. Table 6 compares OPACs on the basis of these additional search features. It is depicted in Table 6 that two additional search features (Quick Search and Allied Links) have been analyzed. All the three catalogs support both these features. Table 6: Additional Search Features WorldCat

LocatorPlus (NLM)

Quick Search 1 Allied Links 1 *1 stands for AVAILABLE, 0 stands for NOT AVAILABLE

1 1

Library of Congress Online Catalog 1 1

4. Findings The major findings are enumerated as follows:  All the three OPACs provide the facility of both the levels of searching and most of the search techniques.  Most of the search limiting options can be seen in all the three Online Public Access Catalogs, thus providing the users with the facility of customizing their search according to their needs.  It is important that the users should be provided with information feedback, support and guidance in order to understand and use an information retrieval system efficiently (Park & Lim, 1999). While analyzing the help mechanism, it is not too convincing except for the Library of Congress Online Catalog.  Alerting services are also not too promising except for the Library of Congress Online Catalog again.  Result Manipulation options can be seen in all the three catalogs.  All the three OPACs fully support both the identified additional search facilities (Quick Search and Allied Links). Thus, assessing each resource on various parameters of search, it is found that Library of Congress Online Catalog provides many of the search features as compared to the rest. The other two OPACs do not show substantial results. It is clear from the above assessment that all the resources under study are in need of improving their interfaces. In order to enhance the effective, efficient and satisfactory use of application, it is important to design interface signs for the users

[454]

Shafiq et al/COMMUNE – 2015

(Islam, 2013). It can thus be concluded that online resources have to be designed in such a way that the users can make the most from their use. 5. Conclusion Gone are the olden days when a user had to pass through a cumbersome task of searching the various catalogue cabinets housed at a separate place in the library for locating a desired book. OPAC's have now revolutionized the way information is being located and searched in the libraries. Developing the more user friendly OPACs is the first step in ensuring that the users of a particular library are satisfied upto the optimum level and it can go a long way in saving the time and reducing the frustration of the users in locating their desired documents in the library. It is the need of the hour and foremost responsibility of the librarians to develop the OPACs as per the requirements of the users they serve, as the complex nature of the information sources acquired by the libraries of the twenty first century are changing considerably. References Babu, B.R., O’Brien, A. 2000. Web OPAC interfaces: An overview. The Electronic Library, 18(5), 316-330. DOI: 10.1108/02640470010354572 Brinkley, M., Burke, M. 1995. Information retrieval from the internet: An evaluation of the tools. Internet Research: Electronic Networking Applications and Policy, 5(3), 3-10. DOI: 10.1108/10662249510104595 He, H., Meng, W., Yu, C., Wu, Z. n.d. Construction Interface schemas for Search Interfaces of Web Databases. Retrieved from: http://cs.binghamton.edu/~meng/pub.d/He_p136.pdf Hearst, M.A. 2011. User Interfaces for Search. Modern Information Retrieval: The Concepts and Technology behind Search Engines (2nd ed.). Boston, USA: Addison Wesley Professional. Retrieved from: http://people.ischool.berkeley.edu/~hearst/papers/mir2e_chapter2_hearst_uis_references.pdf Hiller, S. 2002. The impact of information technology and online library resources on research, teaching and library use at the University Of Washington. Performance Measurement and Metrics, 3(3), 134-139. DOI: 10.1108/14678040210454923 Hu, P.J.H., Ma, P.C., Chau, P.Y.K. 1999. Evaluation of user interface designs for information retrieval systems: A computer-based experiment. Decision Support Systems, 27(1-2), 125-143. DOI: 10.1016/S0167-9236(99)00040-8 Islam, M.N. 2013. A systematic literature review of semiotics perception in user interfaces. Journal of Systems and Information Technology, 15(1), 45-77. DOI: 10.1108/13287261311322585 Kules, B., Wilson, M.L., Shneiderman, B. n.d. From Keyword Search to Exploration: How Result Visualization Aids Discovery on the Web. Retrieved from: http://hcil2.cs.umd.edu/trs/2008-06/2008-06.pdf Mi, J., Weng, C. 2008. Revitalizing the library OPAC: Interface, Searching, and Display Challenges. Information Technology and Libraries, 27(1), 522. DOI: 10.6017/ital.v27i1.3259 Nnadozie, C.O. 2008. Use and Non-Use of Information Resources by Low-Ranking Industrial Employees in Nigeria. Trends in Information Management, 4(2), 140-157. Retrieved from: http://ojs.uok.edu.in/ojs/index.php/crdr/article/view/130/118 Online Public Access Catalog, 2014. Retrieved from: http://en.wikipedia.org/wiki/Online_public_access_catalog, Jun. 19, 2014 [Oct. 5, 2014] Park, K.S., Lim, C.H. 1999. A structured methodology for comparative evaluation of user interface designs using usability criteria and measures. International Journal of Industrial Ergonomics, 23(5-6), 379-389.Retrieved from: http://ac.els-cdn.com/S0169814197000590/1-s2.0-S0169814197000590-main.pdf?_tid=622575ba-b482-11e2-8366 00000aab0f01&acdnat=1367648466_02ab7aebb51b8350b86caa7a82e9cb40 Ryder, B. 2011, January 30. Too much information. The Economist. Retrieved from: http://www.economist.com/node/18895468#footnote2 Sridhar, M.S. 2004. OPAC vs card catalogue: A comparative study of user behaviour. The Electronic Library, 22(2), 175-183. DOI: 10.1108/02640470410533443 Thompson, D.M., Pask, J., Peterson, B., Haynes, E. 1994. Online public access catalogs and user instruction. Reference & User Services Quarterly, 34(2), 191-202. Retrieved from: http://www.jstor.org/stable/pdfplus/20862644.pdf Vickery, B., Vickery, A. 1993. Online search interface design. Journal of Documentation, 49(2). Retrieved from: http://www.emeraldinsight.com/search.htm?st1=Online+search+interface+design&ct=all&ec=1&bf=1 Wells, D. 2007. What is a library OPAC?. The Electronic Library, 25(4), 386-394. DOI: 10.1108/02640470710779790 Wien, C. 2000. Teaching online information retrieval to students of journalism. Aslib Proceedings, 52(1), 39-47. DOI: 10.1108/EUM0000000006999 Wilson, M.L., Kules, B., Schraefel, M.C., Shneiderman, B. 2010. From Keyword Search to Exploration: Designing Future Search Interfaces for the Web. Foundations and Trends in Web Science, 2(1), 1-97. DOI: 10.1561/1800000003

[455]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Big Data: A Growing Tide not Hype Samiah Jan Nastia*, M. Asgarb , Muheet Ahmed Buttc, Majid Zaman Babad a Department of Computer Sciences, BGSB University, Rajouri, India Department of Mathematical Sciences and Engineering, BGSB University, Rajouri, India c Department of Computer Sciences, University of Kashmir, Srinagar, India d Directorate of IT and SS, University of Kashmir, Srinagar, India

b

Abstract Data revolution is just at its infancy; everyone is talking about Big Data. Big Data is an explosion of data and as such traditional systems are not scalable enough to handle this enormous data. The explosion of the Big Data is a very recent phenomena it is quite recently, companies have started to realize that they should capture all this data that is being producing and not only capture they should try to analyse it and try to get some value of it. This paper explores the Sources of Big Data, Architecture of Big Data, challenges and issues produced by it and Hadoop, a Big Data tool

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Big Data: Architecture: Hadoop

1. Introduction Data is a new class of economic asset, like currency and gold (World Economic Forum 2012). Data is growing at enormous rate, so it is very difficult to manage and handle this huge and gigantic volume of data. It is very difficult to handle this enormous data because it is growing so rapidly in comparison to the computing resources. The term Big Data is very confusing as it gives us a feeling that after a certain size the data is big and below a certain size the data is small (Dong, X.L and .Srivastava, D). The Big Data could start from any point. There is no definitive definition for Big Data. However it is mostly defined this way that “Big Data is a data that becomes difficult to be processed because of its size using traditional system”. Traditional systems including relational databases are not capable of handling the Big Data and challenges spring up at multiple levels including capturing, storing, analysing, searching, sharing, transforming the data and even visualizing the data. The Big Data becomes a challenge for traditional systems not merely because of its size that could be a challenging point but challenge may also arise because of its speed at which the Big Data is coming in and also because it is unstructured and it could contain data items of various formats. So Big Data is usually measured by three attributes, velocity, volume and variety. The velocity refers to the speed at which the data is coming in e.g. the Scientific Experiments that they do at atomic reactors where they do the collision of sub-atomic particles, 40TB of data could come in within one sec, so that is a very high speed. Volume is of course a problem, the data keeps on accumulating and the file becomes too large to be handled by traditional system. The Facebook is generating 25TB of data daily so just imagine the size of the files that are there since the beginning of time. In traditional systems data is structured and is stored well in planned tables, each table has specific columns and each column could accept values of specific data types. However in case of Big Data, the third V creates problem sometimes i.e. variety. When the Big Data comes in it may include items of variety of formats. It could have audio files, video files, and unstructured data like text messages so that becomes challenging sometimes for a traditional system to handle. The explosion of the big data is a very recent phenomenon and it is quite recently, companies have started to realize that they should capture all this data that is being produced and not only capture they should try to analyse it and try to get some value out of it. These days the decision making is solely performed on structured data which is mostly stored in applications like ERP’s and other related applications that are running in an Enterprise. So, the most of this unstructured data gets wasted, it is not captured and *

Corresponding author. Tel.: +91 8491 027772. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Samiah et al/COMMUNE – 2015

even if it is captured it is not analysed and even if an attempt is made to analyse this data, the real value is not expected because of the limitations. This is the area (90% area) that represents the focus of the companies in coming years and these companies will try to analyse this unstructured data and extract meaningful information out of it. 2. Related Work Kaisler and his Co-workers analyse the issues and challenges in Big Data and begin a collaborative research work program into methodologies for Big Data analysis and design. They conclude that the traditional databases do not solve the various aspects of the Big Data problem and thus advocates for machine learning algorithms which should be more robust and easier for unsophisticated users to apply. Bakshi. K discusses architecture of Big Data and ends up with the conclusion that despite the various architectures and design decisions the analytics system aim to scale out elasticity and high availability. The concepts of Big Data and the available market solutions used to manage and explore the unstructured data are discussed. The results and observations thereof concluded that the analytics is a pivotal part for adding value for the social business. Demchenko , Zhiming and Wibison introduces the universal model known as the Scientific Data Infrastructure (SDI).Authors show that with the help of Cloud based Infrastructure such a model can be implemented easily. Courtney, M investigates the difference in Big data applications and how they are different from the traditional methods of analytics existing from a long time. Smith and his co-workers analyze social media sites such as Flickr, Locr, Facebook and Google+. Based on this analysis they have discussed the privacy implications and also geo-tagged social media; an emerging trend in social media sites. They presented a concept with which users can stay informed about which parts of the social media deluge relevant to them. 3. Sources of Big Data A growing number of users, applications, systems, sensors, mobile devices, social media etc, are producing large and large files. These files are not only large, they are being produced at a very high speed and sometimes these file contain variety of data items which are not even structured such as video files, audio files, images, photos, logs, click trails, text messages, emails documents, books, transactions, public records etc. So all these attributes are creating challenges for traditional systems and hence the term Big Data. Here are some examples of data generation points: (Katal, A. et al) 3.1

Data from Enterprises

Now –a- days the profitability of business department is mainly influenced by the IT and digital data. It has been estimated that the data used and produced by the different business companies may double every 14 months. Amazon terminal operations are processed in millions and third party sellers’ queries are more than 500K per day .Also1 million customer trades are processed by Wal-Mart per hour. Big Data is also coming from activities like Trading. New York Stock Exchange (NYSE) produces 1TB per trading day; in 2020 the total data estimated is 35 ZB. 3.2

Internet of Things

IoT is a vital source of Big Data. The cities which have been designed on the basis of IoT are referred to as Smart Cities and in these cities the data comes from various fields such as medical care, Agriculture, Traffic, Industry etc. The data generated by the IoT has mainly the following features: i. Large Scale data, ii. Heterogeneity of Data , iii. Time and Space Co-relation. 3.3

Biomedical Data

The human genome project (HGP) and the sequencing technology lead to a generation of huge amount of data. A data of about 100-600 GB is generated by one sequencing of human gene. By 2015 the traceable biological samples will reach to 30 million. The development of gene sequencing and bio-medical technologies will contribute to the continuous growth of Big Data of Bio-medicine beyond all doubts. 3.4

Other Sources

Airbus generates 10 TB every 30 minutes. About 640 TB is generated in one flight. This is a lot of data usually one warehouse would be of this size .Smart meters read the usage every 15 minutes and record 350 billion transactions in a year. By the end of 2014, there were 200 million smart meters. Camera phones are the world wide and most of them are

[457]

Samiah et al/COMMUNE – 2015

location aware meaning when we take a photo the location information could go in the photo as well as because of the GPS capabilities and 22% of these phones are smart phones. By the end of 2013, it was estimated that the number of smart phones exceed the number of PC’s. The cell phones and smart phones are major players in creating large volume of data. According to CISCO it was estimated that by 2014, internet traffic was reached 4.8 Zeta Bytes. Sloan digital sky survey (SDSS) and High energy physics also are the major sources of Big Data 4. Architecture of Big Data Big Data resembles to the Cloud Architectures and is having four layers as shown below in figure 1:

Fig 1: Cloud Architectures

Infra-structure as a service: - It consists of network as a base, storage, servers and inexpensive commodities of Big Data stack which can be bare metal or virtual (cloud). Distributed file system are part of this layer. Platform as a service: - Distributed caches and NoSQL stores, form the platform layer of the Big Data. The logical model is provided by this layer for the unstructured and raw data stored in the files. Data as a service layer: - In this layer the integration of all the tools available is done with the PaaS Layer using integration adapters, search engines, Batch programs etc. Big Data function as a service: Packaged applications can be built by the industries like e-commerce, health, retail and banking to serve a specific business need and the DaaS layer is leveraged for the cross cutting data functions. 5. Challenges and Issues in Big Data The challenges and issues in big data include (Madden, S and Wand W): 5.1

Information Sharing and data access

In order to make timely and accurate decisions it is mandatory that the data should be timely available and besides its timely availability it should be also complete and accurate. This necessity of making data open and available to government agencies in standardized manner leads to decision making , Business intelligence and productivity improvements but this process of making data open makes the management and governance process bit complex and challenging. In order to get an edge in business the sharing of data by the companies about their operations and clients, threatens the intrinsic of secrecy and competitiveness. So it is very much akward to expect the sharing of data between the companies. 5.2

Storage Issues

The large amount of data generated by almost everything such as social media sites, sensor devices etc. needs to be stored and the storage available is not enough for such amount of data. Uploading such large amount of data in Cloud seem an option but it does not solve the problem in actual. The data which is in Tera Bytes make take a large amount of time for uploading in Cloud and it is hard to upload this rapidly changing data in real time. Besides the above the Cloud distributed nature is a hurdle of analysis of Big Data. This outsourcing this Big Data to cloud leads to Capacity and Performance Issues. 5.3

Processing Issues

It takes large amount of time for processing such large amount of data. Whole of data set is required to be scanned to find suitable elements somewhat which is not possible and building up of indexes right at the start while storing and collecting the data is a good option as it reduces time of processing. It considerably reduces analytical challenges.

[458]

Samiah et al/COMMUNE – 2015

5.5

Analytical Challenge

This huge and massive amount of data which can be structured, semi-structured and structured requires skills for its analysis. Moreover the analysis to be done on the data depends highly on the results to be obtained i.e. decision making, and this can be done either by incorporating massive data volumes in analysis or determine upfront which Big Data is relevant. The main challenges can be summarized as:      5.6

What, When data volume gets varied and large and how to deal with as it is not known? How data can be used to best advantage? Is it necessary to store all data? Is it necessary to analyse all the data? How to search or find out data points which are really important?

Skill Requirement

Big Data is a new and emerging technology and needs to attract the organizations and people with diverse new skill sets, which should not only include technical but also should extent to other areas such as research analytical and creative ones. 6. Hadoop: A Big Data Tool Hadoop – an open source framework was developed by Google and later on adopted by Yahoo and handed over to Apache. Hadoop supports the processing of large data sets in a distributed environment. It breaks the data into smaller pieces and thus breaks the computation into smaller pieces as well as each smaller piece of computation is sent to the smaller piece of data so that instead of performing on Big computation, numerous smaller computations are performed obviously much faster and finally the result is aggregated and the aggregated result is sent back to the application. There are two core components of Hadoop viz. Hadoop File Distributed File System, HDFS (Storage) and Map Reduce (Processing) as shown below in Figure 2

Fig. 2: Hadoop File Distributed File System

7. Conclusion Everything we do generates data. We swim in such a sea of data whose level is increasing. Millions and Billions of people and sensors and trillions of transactions are rapidly working to generate unimaginable amounts of data without any doubt. Technology evolution and placement guarantee that in few years more data will be available in year than has been created since the dawn of the man. References Bakshi, k., 2012. Considerations for Big Data: Architecture and Approach, IEEE , Aerospace Conference, p.1 Courtney, M., 2012. The Larging-up of Big Data, IEEE, Engineering &Technology,p.72 Demchenko, Y, Zhiming Zhao ; Grosso, P., Wibisono, A., de Laat, C. , 2012. Addressing Big Data Challenges for Scientific Data Infrastructure, IEEE , 4th International Conference on Cloud Computing Technology and Science,p.614 Dong, X.L. , Srivastava, D. , 2013. Big Data Integration, IEEE 29th International Conference on Data Engineering (ICDE), p. 1245 Katal, A., Wazid, M., Goudar, R.H. , 2013. Big data: Issues, challenges, tools and Good practices , Sixth International Conference on Contemporary Computing (IC3) , p. 6 Kaisler, S., Armour, F., Espinosa, J.A. ,Money, W. , 2013. Big Data : Issues and Challenges Moving Forward, IEEE, 46th Hawaii International Conference on System Sciences, p.995 Madden, S ., 2012 . From Databases to Big Data, IEEE, Internet Computing,p.4 Singh, S. ; Singh, N., 2012. Big Data Analytics, IEEE, International Conference on Communication, Information & Computing Technology (ICCICT),p.19 Smith, M., Szongott, C., Henne, B., von Voigt, G. , 2012. Big Data Privacy Issues in Public Social Media, IEEE, 6th International Conference on Digital Ecosystems Technologies (DEST), p.1 Wang, W , 2014 .Big Data , Big Challenges ,, IEEE International Conference on Semantic Computing (ICSC) ,p. 6

[459]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

English-Kashmiri Machine Translation: Issues and Challenges Mir Aadila*, Mohammad Asgerb, Vishal Goyalc a Department of Computer Sciences, BGSBU, Rajouri, India. School of Engineering & Mathematics, BGSBU, Rajouri, India c Department of Computer Sciences, Panjabi University, Patiala, India b

Abstract Machine Translation is now considered as a challenging task for research by the academicians. It is an interesting and promising study of research, even though a flawless and correct translation by an intelligent computer is yet a dream to be realized due to the complexity and challenges that slowly came to notice. Most of these problems are independent of the methodology or tools used to achieve overall translation but still vary with each language pair. Every language pair puts forth a different level of challenges and issues which latter on becomes the reason of undesirable translation quality and fluency. This work tries to bring forth few of the main challenges that are faced at the very initial stages of the process of machine translation for English-Kashmiri machine translation.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Machine Translation; English-Kashmiri Translation; Challenges with Machine Translation; Divergence in English-Kashmiri

1. Introduction Machine translation is not merely an automatic linguistic word by word translation, rather a translation of one natural language to another and preserving the meaning just like a human translator. And just like a human translator, machine translators also face multiple divergences in any language pair. The languages differ in their lexicon, syntax, semantics, pragmatics, culture and background so these need to be taken in account for a reliable translation. Kashmiri culture, its beauty, its Sufi saints and its literature is diverse and unique and so is its language. However, the language (like other around 3000 languages) is facing a treat of extinction. That is why globalization of its language is the need of the hour. Machine translation is a promising solution. But Kashmiri language has a really scarce corpus available online or in digital form. And the development of a huge database of parallel corpora is one of the biggest challenges. Also like any language pair, English-Kashmiri translation also exhibits a general and an idiosyncratic difference in realization of their syntax and word order and has a lexical and morphological divergence. These divergences put forth some issues and challenges, most of which creep in usually the initial stages of machine translation and continuously damage the overall efficiency of the translation. The main challenges and issues that arise are a result of the ambiguity in source language, divergence across the languages and finally the variations in the target language. A study of these root causes is necessary as these are to be kept in view while devising algorithms for machine translation. 2. Source Language Ambiguity Nagamani and Ananth proposed an image compression technique for high resolution, grayscale Satellite urban images. The proposed technique used discrete wavelet transform together with EZW (Embedded Zero tree wavelet) and SPIHT (Set Partitioning in Hierarchical Trees) coding techniques in order to achieve high compression ratio and better image quality. The compression ratio and peak signal to noise ratio determined using EZW and SPIHT codings have been compared to each other for same set of images. The results obtained showed possibility to achieve higher * Corresponding author. Tel.: +91 9086 750369. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Aadil et al/ COMMUNE-2015

compression ratio and PSNR (approximately CR of 8 and PSNR of 29.20) for SPIHT coding compared to EZW coding (approximately CR of 1.07 and PSNR of 13.07) for applications related to satellite urban imagery (Nagamani, Ananth, 2011). Ambiguity in source language are not itself a big issue as these are usually solved in context, but what damages the efficiency of translation is that these ambiguities multiply. A 2-gram sentence(two words) with each word having only two different meanings can be interpreted in four different ways and a 3-gram with same degree of ambiguity shall have eight different interpretations all correct as per grammar. 2.1

Lexical Ambiguity.

When there are multiple meanings of same word or phrase in the source language. e.g. The word “bank” can be used for river edge as well as for a financial institution. 2.2

Syntactic Ambiguity.

When there are multiple interpretations for a sentence because of unclear modifying expression in its structure- its syntax. e.g. “We need apple or banana and sugar” may mean either apple or both banana and sugar or it may mean apple and sugar or banana and sugar. Similarly, for the sentence “It is a little pretty girl's doll” may mean the doll is little and girl is pretty or the girl is little and pretty or the doll is little and pretty. 3. Cross Lingual Divergences Divergence is the most common observation between any two languages to be translated. When two sentences of the source language that are structurally similar show divergence in their structure in the target language, such languages show cross-lingual divergence. [Dorr, 1993]. For our study we used some of the divergence types based on the Dorr Classification [1993]. Dorr categorizes translation divergence broadly in  Syntactic Divergences like Constituent order divergence, Adjunction divergence, Preposition Stranding divergence, Movement divergence, Null Subject divergence, Dative Divergence and Pleonastic Divergence.  Lexical-Semantic Divergences like Thematic Divergence, Promotional Divergence, Demotional Divergence, Conflactional/Inflational Divergence, Categorial Divergence and Lexical Divergence. Some of these divergences are universal and some exist only for particular language pairs. Some particular of these divergences in English-Kashmiri language pair are as described below. 3.1. Categorial Divergence: It is the most common type of divergence found in any language pair. Categorial divergence results when parts of speech in the source language are translated into the target language using different parts of speech, e.g. if Noun in source language is translated into adjective in the translated language or vice-versa, or if verb in source language is translated into noun in the target language or vice-versa. So, any change in the lexical category of a word in the source language during translation process leads to categorial divergence, e.g. Noun Adjective or Adjective Noun, Verb Noun or Noun Verb Some of the examples showing categorial divergences in English-Kashmiri Translation are. 

Verb to Noun: English: My mom loves me. Kashmiri: Transliteration: Maineh Majeh Che Main Maiye. Or Kashmiri: Transliteration: Maineh Majeh Che Maiye Main. In English sentence the word “loves” is a verb while in Kashmiri language it is realized as noun.



Noun to Verb: English: His residence is near the river. Kashmir: Transliteration: Su chu daryavas nazdeek roozan.

[461]

Aadil et al/ COMMUNE-2015

In English the word “residence” is a noun while in Kashmiri it is realized by verb. 

Noun to Adjective: English: She is a beauty. Kashmir: Transliteration: Sou che Khoobsurat. The word “beauty” in English is a noun while in Kashmiri it is adjective.



Adjective to Noun: English: Her behavior is harsh. Kashmir: Transliteration: Tinhindis wartravas manz chu troushar.

3.2. Conflational and Inflational Divergence: Conflational Divergence occurs when two or more words in the source language are translated in a single word by combining their meaning in the target language. Such divergence is also known as Lexical Gap. e.g. English: He slipped away. Kashmiri: Transliteration: (su) Tsul English: You may leave now Kashmir: Transliteration: Neeriv Inflational Divergence is just the opposite of the Conflational Divergence. It is when a single word in source language is translated in multiple words in the target language. English: Please leave. Kashmiri: Transliteration: Meharbani kareth neiriv toye. English: It suffices Kashmiri: Transliteration: ye chu hajtas mutabik 3.3. Structural Divergence: Structural divergence is the difference in the realization of the incorporated arguments of the source language and in the target language. This is the difference in Phrasal Categories, e.g. when an ad-position-phrase (PP) category in one language is realized by a noun-phrase (NP). English: I have to go to Punjab. Kashmiri: Transliteration: Meh chu gasun Punjab. English: I have to write a letter. Kashmiri: Transliteration: Meh che chithe lyekan 3.4. Head Swapping Divergence: Head Swapping arises when the role of the main verb in source language is diminished and the role of the modifier verb is promoted in the target language. The first phenomenon is known as demotional divergence and the second is known as promotional divergence respectively. Since in almost all cases demotional and promotional divergences work together, these are together entitled as Head Swapping Divergence. English: The Music is on. Kashmiri:

[462]

Aadil et al/ COMMUNE-2015

Transliteration: saaz chu bahjaan ya wazaan English: The Jhelum is flowing. Kashmiri: Transliteration: Wyeth che pakaan 3.5. Thematic Divergence: Thematic divergence arises due to switching of role of arguments in the Source Language to Target Language, e.g. the difference in the argument structure of verb in the two languages. English: Where from are you? Kashmiri: Transliteration: Toye kateh chiv roozan. (OR) Kashmiri: Transliteration: Toye kateh peth chiv. English: Why are you late. Kashmiri: Transliteration: Toye kyazi gov tsheer. 3.6. Lexical Divergence: Lexical divergence is not very uncommon. This is relatively simple type of divergence where exact match for translation of a word or phrase in the Source language is not available in the Target Language. The lack of exact translation map for a certain construction between two languages gives rise to such divergence. English: Good Luck Kashmiri: Transliteration: Khudaiye sund fazal aesney English: Please Sit Kashmiri: Transliteration: Meharbani Kareth thayev tashreef. 3.7. Honorific Divergence: As other South Asian languages, Kashmiri Language also exhibits some of honorific features. Honorific divergence occurs when we use plural inflectional elements (verb and the genitive noun) for some nouns known as Honorific nouns just for the purpose of showing respect or honor. Such features are not present in English language and other European languages. English: He is my friend. Kashmiri: Transliteration: Suh chu meh doost English: He is my teacher. Kashmiri: Transliteration: Tem cheh mein ustaad English: He has come. Kashmiri: Transliteration: Suh aaw (or) Timav aun tashreef.

4. Target Language Variation For a single sentence, each word can be translated into multiple options resulting in a large number of translated sentences in target language. This may result due to ambiguous structure, but the main cause is word sense disambiguation. Although for corpora rich languages like English it can be solved by anaphora resolution, coherence, and inference and mostly by improving relevance of search engines. However for a corpora deficient language like

[463]

Aadil et al/ COMMUNE-2015

Kashmiri, it is a bigger problem. Statistical Machine translation approach uses the probability and frequency of each sentence out of the multiple candidate sentences to decide the most appropriate translation. This may not always work as a sentence is usually a reflection of thought of mind and the most frequent or probable translated sentence may not be always correct. However, Kashmiri language is still not abundantly available on Internet, so resolving word sense disambiguation for Kashmiri language is typically tough. 5. Conclusion There are multiple issues and challenges in English-Kashmiri machine translation that need to addressed to ensure fluent, efficient, and desirable output. The scarcity of parallel corpora, the divergences and variations at different levels make English-Kashmiri translation difficult. These divergences and variations need to be studied properly to devise proper algorithms and tuning of the translation mechanisms for a better output. There are various techniques and methods that can resolve the issues of these divergences. However, not each one of these challenges can be solved yet. References Aasim Ali and Malik, M. K., 2010. Development of parallel corpus and english to urdu statistical machine translation. Int. J. of Engineering & Technology IJET-IJENS, 10:31–33. Jawaid, B. and Zeman, D., 2011. Word-order issues in english-to-urdu statistical machine translation. Number 95, pages 87–106, Praha, Czechia. Dave, S. and Parikh, J. and Bhattacharya, P., 2002. “Translation Technical Report", LAMP 88. Dorr, B., 1994. “Classification of Machine Translation Divergences and a Proposed Solution Computational Linguistics”. 20 (4) 597–633. Dorr. Bonnie, J., 1994. “Machine Translation Divergences: A Formal Description and Proposed Solution”, Computational Linguistics, 20:4, pp. 597-633. Dorr, Bonnie, J. and Nizar Habash,, 2002. “Interlingua Approximation: A Generation-Heavy Approach”, In Proceedings of Workshop on Interlingua Reliability, Fifth Conference of the Association for Machine Translation in the Americas, AMTA-2002,Tiburon, CA, pp. 1—6. Dorr, Bonnie J., Clare R. Voss, Eric Peterson, and Michael Kiker,. 1994. “Concept Based Lexical Selection”, Proceedings of the AAAI-94 fall symposium on Knowledge Representation for Natural Language Processing in Implemented Systems, New Orleans, LA, pp. 21—30. Dorr, Bonnie J., Lisa Pearl, Rebecca Hwa, and Nizar Habash,. 2002. “DUSTer: A Method for Unraveling Cross-Language Divergences for Statistical Word-Level Alignment," Proceedings of the Fifth Conference of the Association for Machine Translation in the Americas, AMTA-2002, Tiburon, CA, pp. 31—43. Dorr, Bonnie, J., and Nizar Habash, 2002. “Handling Translation Divergences: Combining Statistical and Symbolic Techniques in Generation-Heavy Machine Translation”, In Proceedings of the Fifth Conference of the Association for Machine Translation in the Americas, AMTA-2002, Tiburon, CA, pp. 84—93. Goyal, P., and Sinha. R.M.K., 2008. “A Study towards English to Sanskrit Machine Translation system”. SISSCL. Haspelmath, Martin. 2002. “Understanding Morphology”, Oxford University Press. Jawaid, B., Zeman, D., Bojar, O., 2010. “Statistical Machine Translation between Languages with Significant Word Order Difference”. PBML. Kameyama, Megumi and Ochitani, Stanley Peters. 1991. “Resolving Translation Mismatches With Information Flow” Annual Meeting of the Assocation of Computational Linguistics. Koehn, P., 2010. “Statistical Machine Translation”: Cambridge University Press. Levin, B., 1997. “English Verb Classes and Alterations: A Preliminary Investigation”, the MIT Press. Lewis, Paul, M., Simons, G.F., Fennig. C.D., 2013. “Ethnologue: Language of the World”. Seventeenth edition. Dallas, Texas: SILI. Sinha, RMK and Thakur, A., 2005. “Translation Divergence in English-Hindi MT EAMT”, 10th Annual Conference, Budapest, Hungary

[464]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16-18March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

A Comparative Analysis of Full Adder Cells in Nano-Scale for Cascaded Applications Afshan Amin Khan*, Shivendra Pandey, Jyotirmoy Pathak Lovely Professional University,Punjab,Jalandhar,India

Abstract This paper focuses on the different designs of an Adder cell with an aim of finding an Adder cell from the literature, which can be used for the future VLSI application. To ensure this all the designs have been implemented in Cadence Virtuoso 90nm Technology with least possible size of transistors available. A critical parameter in identifying circuits suitable for VLSI applications is that the implementation area and power dissipation must be as small as possible, thus in this paper we are trying to compare different possible design of an adder cell ranging from the most stable 28 transistor(28T) adder cell to a low area 8 transistor(8T) adder cell. All the designs have been compared for their threshold voltage loss, power dissipation, delay, and PDP, to choose the more reliable adder cell for cascaded VLSI applications.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Low Arear Full Adder; Nano Scale Adder; Lower Trasistor Count; Threshold Loss; Cascaded Logic

1. Introduction In recent years, a good number of researchers have focused their work for the optimization of system architectures such as that of a Filter, multipliers, MAC units etc. This methodology of optimization includes a huge amount of theoretical calculations and improvisations in the logic used to implement the architecture. However, a simpler method to optimize any logic circuit can be to identify a base cell that can be repeated to form the complete architectural design of a system. For systems like multipliers and others the base cell is clearly a Full Adder Cell. Thus optimization of a Full Adder Cell will lead to the optimization of the system as a whole while all the efforts are focused on optimizing a single unit out of the whole architecture. Realizing the significance of a Full Adder Cell we have gone through the literature and identified a wide range of versatility in the design and also in the number of transistors used to implement each design. Some of the promising designs of Full Adder Cells have been implemented and compared in this paper. However a disadvantage of reducing the number of transistors being used to implement the Adder cell is that the driving capability is also expected to vary with the number of transistors used to implement the cell. Thus the cells given in this paper have been compared on the basis of amount of the degradation attained in the output logic level for respective input combinations, power consumption and delay. As we tend to reduce the number of the transistors used to implement an adder cell starting from the fully symmetrical design of a 28T Adder cell R. Zimmermann et al., 1997 and going down to least possible 8T Adder cell Fayed et al., 2001 there is a variation in the amount of the degradation levels attained in either Sum or Carry output or both, which ultimately restricts the driving capability of respective Full Adder Cell. Each adder cell has been somewhat modified by varying the W/L ratio of the transistors used in the design of that particular adder cell so as to attain some acceptable logic levels of Sum and Carry outputs, which help in proper driving of the next stages of the circuit.

2. Implementation and Design Analysis A detailed study of the literature provides different possible designs of a Full Adder Cell that are optimized in one or the other respect. However, an optimization is reliable only if it fits in some real time application such as multi-bit addition, Multiplication and so on, in order to find an application suitable adder cell following designs where implemented and analyzed for the features given below.

* Corresponding author. Tel.: +91 9796 570562. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Khan et al/COMMUNE – 2015

2.1.

28T Adder Cell.

The 28T CMOS adder cell uses 28 transistors for its implementation, an efficient design of this 28T full adder R. Zimmermann et al., 1997 provides a symmetric design of 28T full adder as in Figure 1. The basic advantage of this adder is that it has no threshold loss problem and more over power dissipation is of acceptable values. As this adder is being designed using CMOS design, it stands as one of the adder that has a versatile application and will be as a stand out performer in most of the cases, whether it be the large systems or an adder cell itself, 28T adder is a promising design, hence becomes a major challenge to the researcher to optimized such a stable design. The two major concerns for the implementation of this cell is the amount of area consumed and the amount of delay of this adder cell thus has becomes an area of research for the designers. As a result of this research designers have developed different circuit level implementations which take care of reducing the implementation area and delay with some or little tradeoff of the circuit performance. However some designs not only provide an area reduction but also show a significant improvement in performance characteristics.

Fig 1. 28T CMOS Full Adder cell

2.2.

24T Adder Cell.

The 24T adder cell S. Goel et al., 2006 given in Figure 2 is also one among the promising adder cell in terms of less threshold loss problem as well as reduces the number of transistors by four, however the amount of the power dissipation in this circuit increases as compared to the most stable 28T adder cell and is nearly double than the power of the 28T adder cell, whereas the amount of area saved is very less. Hence it is a requirement for the designer to optimize this design for power consideration. Hence can’t be regarded as efficient trade-off of power and area.

Fig 2. 24T Full Adder cell

2.3.

20T Adder Cell.

This adder cell uses 20 transistors for implementing both sum and carry functions. A possible design of Adder cell using 20T N. Weste and K. Eshraghian et al., 1993 is given in Fig 3. This adder cell was also observed to have very little or no threshold loss problem, more over the number of transistors is reduced by 8. Thus area is reduced to a great deal with significant improvement in power consumption. Hence it assures to be better performer in terms of both power and area.

[466]

Khan et al/COMMUNE – 2015

Fig. 3. 20T Full Adder cell

2.4.

16T Full Adder Cell.

The 16T adder cell N. Zhuang and H. Hu., 1992 given in Fig. 4 proves to be a better design than the above designs in terms of power dissipation as well as delay is almost similar to that of 28 and 24T adder cells, whereas there is a slight degradation in output logic levels. However the driving capability is still expected to be fairly good. The design is a mixture of the pass transistor logic, GDI and pass transistor based MUX cells as in Figure 4.

Fig. 4. 16T Full Adder cell

2.5.

14T Adder Cell.

Two different designs of 14T adder cells have been taken from Chip-Hong Chang et al., 2003 and T.Sharma et al., 2010. Among these 14T adder of Chip-Hong Chang et al., 2003 proves to be a better adder and gives astonishing results in terms of voltage threshold loss problem, however the amount of power consumed increases. The one in T.Sharma et al., 2010 gives some serious threshold loss problems and even supplies some very weak logic values for the w/l ratios used to implement this design, hence is a further topic of research. In this work we have considered the 14T adder cell in Chip-Hong Chang et al., 2003 shown in Figure 5 only for the rest of analysis and comparison with other designs.

[467]

Khan et al/COMMUNE – 2015

Fig. 5. 14T Full Adder cell

2.6.

12T Full Adder Cell

The 12T adder cell Yingtao Jiang et al., 2004 given in Fig. 7 proves to be a promising design in terms of least amount of power dissipation among all the circuits implemented in this work but on the other hand the output logic voltage levels for respective logic are not satisfactory and needs some level restoration circuitry or improved sizing parameters to be used for obtaining proper logic voltage swings for both Sum and Carry outputs of a Full Adder cell.

Fig6. 12T Full Adder cell

2.7.

10T Adder Cell.

The 10T adder cell Fayed et al., 2001 given in Figure 8 proves to be very useful adder cell in terms of reducing the overall power of the circuit but as expected the more we go away from the conventional design possible chances of threshold loss problem are more as a result the output obtained is degraded in nature. Thus this adder can’t be used for a circuit where the cascading of the adder cell structures is very huge. However comparing the adder for the amount of area saved and amount of power dissipation obtained , the amount of degradation in the output is somewhat acceptable and may be able to drive some limited number of stages in a system were cascading is a choice of design.

Fig. 7. 10T Full Adder cell

[468]

Khan et al/COMMUNE – 2015

2.8.

8T Adder Cell.

The design of 8T adder cell Shiwani Singh et al., 2012 is observed one of the reliable designs among all other 8T adder cell analyzed so far in Afshan Amin et al., 2014. This cell gives the advantage of reducing the number of transistors to a great deal however its huge amount of power dissipation and significant threshold loss problem make this design less feasible for general operations, especially for the circuits where amount of cascading involved is large. The cascading logic is expected not only to consumes huge amount of power but also due to degraded output logic swings for both sum and carry for a single stage 8T full adder cell, thus the driving capability of this cell is expected to be limited. The methodology that can be used to eradicate this threshold loss problem can be use of a logic implementation technique that itself restores the voltage swings at respective stages or else logic level restoration circuits are used at required places.

Fig 8. 8T Full Adder cell In cases of circuits using this adder cell and a logic level restoration circuit as well, we may be able to eradicate the problem of weak voltage swings but the advantage of using only 8 transistors for implementation of a full adder cell will be lost as the logic level restoring circuit will again increase the over-all area of the circuit.

3. Simulation Results and Analysis We have analyzed all implemented full adder cells for their power , delay and PDP performance using Cadence Virtuoso 90nm technology file and the results are as below with their respective implemented circuits given in the figures from figure 1 to figure 8 . The size of the MOS transistors used is kept as least as possible to support the low area implementation and for any modifications same is reflected in the figures. Table

1: Carry output voltages obtained for respective input combinations (abc)CY

28T

24T

20T

16T

14T

12T

10T

8T

(000)0

.01473E-3

.01468E-3

.008511E-3

.01097E-3

.009866E-3

316.861E-3

262.700E-3

.000156E-3

(001)0

.01673E-3

.015209E-3

.74897E-3

.75534E-3

35.260E-3

318.420E-3

480.800E-3

492.450E-3

(010)0

.01385E-3

.023990E-3

.73686E-3

.008906E-3

.008820E-3

330.760E-3

285.700E-3

.002237E-3

(011)1

1.798E+0

1.7980E+0

1.8000E+0

1.7980E+0

1.7980E+0

1.643E+0

1.79E+0

1.66E+0

(100)0

.01523E-3

.011296E-3

.008081E-3

.74399E-3

.75619E-3

275.170E-3

289.100E-3

.7256E-3

(101)1

1.798E+0

1.7980E+0

1.7980E+0

1.7990E+0

1.7990E+0

1.594E+0

1.8E+0

1.872E+0

(110)1

1.798E+0

1.7980E+0

1.7980E+0

1.7980E+0

1.7980E+0

1.537E+0

1.79E+0

1.799E+0

(111)1

1.798E+0

1.7980E+0

1.8000E+0

1.7990E+0

1.8000E+0

1.594E+0

1.8E+0

1.799E+0

Table 2: Sum output voltages obtained for respective input combinations (abc)SM

28T

24T

20T

16T

14T

12T

10T

8T

000)0

0.0166E-3

0.0175E-3

0.7442E-3

0.8612E-3

0.8097E-3

317.7E-3

277.7E-3

0.0295E-3

(001)1

1.798E+0

1.798E+0

1.798E+0

1.799E+0

1.799E+0

1.64E+0

1.798E+0

1.615E+0

(010)1

1.798E+0

1.789E+0

1.794E+0

1.789E+0

1.791E+0

1.53E+0

1.794E+0

1.421E+0

(011)0

0.0147E-3

0.0147E-3

0.0014E-3

1.525E-3

1.456E-3

0.028E-3

787.2E-3

341.6E-3

(100)1

1.798E+0

1.798E+0

1.794E+0

1.793E+0

1.792E+0

1.798E+0

1.794E+0

1.564E+0

(101)0

0.0201E-3

0.0152E-3

1.468E-3

1.299E-3

1.291E-3

316.6E-3

0.786E-3

267.8E-3

(110)0

0.0170E-3

0.0161E-3

0.7552E-3

0.7454E-3

0.6096E-3

285.2E-3

279.0E-3

136.9E-3

(111)1

1.798E+0

1.798E+0

1.798E+0

1.799E+0

1.799E+0

1.595E+0

1.799E+0

1.609E+0

[469]

Khan et al/COMMUNE – 2015

The inputs a,b and c are provided by the user and CY and SM denoted the expected carry and sum logic levels respectively, where as each row contains the logic voltage levels generated by the adder cell with respective input combination in each row. The ideal voltage generated by an adder cell for logic 0 must be 0V and for logic 1 it should be V dd of the respective circuit or else the maximum input voltage value. Table 1 and 2 show the logic voltages generated for each combinations of the inputs possible starting from (abc) = (000) and analyses it upto all inputs being one i.e (abc) = (111). The best case being that of a 28T adder cell which show very less deviation from the expected voltage levels whereas other cells trade off this deviation of logic voltage levels but at the same time save some power or else some area. Moreover the Bar graphs give a clear idea about the amount of power and respective reduction in number of transistors.

Fig. 9: Delay Analysis

Fig. 10: Power Analysis

Fig. 11: PDP Analysis

4. Conclusion and Future Scope Among various other parameters important for the realization of a VLSI circuit, two critical parameters include power and area. In this work, we have focused on optimization of a base unit of a large system to improve the over-all system without modifying its architecture. The base cell chosen here is the Full Adder Cell. However an important concern is to ensure that the logic voltages

[470]

Khan et al/COMMUNE – 2015

generated at the Sum and the Carry outputs must be such that they are able to drive the subsequent stages attached with these terminals, as a result driving capability becomes a check note for any Full Adder Cell. Hence, this work is an effort towards finding an Adder Cell that can be marked as a reliable adder cell among the implemented cell designs. All the designs have been compared for their power dissipation, delay, and PDP, to choose the more reliable adder cell especially for cascaded VLSI applications. More over tables give the idea about the voltage value associated with both sum and carry outputs for each input combination. After detailed analysis it is observed that 8T transistor Adder cell uses minimum number of transistors in its implementation but has a serious problem of consuming huge amount of power. The huge amount of power consumption of this cell can be accounted to poor Vdd and Gnd isolation and also to use of high power consuming 3-T Xor cell. It was observed that while analyzing the individual power of each MOS cell used to implement the 8-T adder cell, the MOS cells comprising the Xor operation consume maximum power among all other MOS cells. Thus the next best choice for a designer is the 10T adder cell which not only gives the lower amount of the power consumption but also reduces the number of transistors to a great deal. A very close contender to this cell is the 12T adder cell as it gives the lowest amount of power dissipation among all other designs in this work;however, it fails at the check point of having poor driving capabilities, as the amount of degradation in this adder cell is expected to be very large. Thus keeping in view the future VLSI application the preferred choice of a designer must be the 10T adder cell as it not only promises to be low power and low area cell but is also expected to have some significant driving capabilities. This feature of 10T adder cell can prove very useful in systems where the primary job of an adder cell would be to drive another cell. The future scope of this work can be use of this 10T adder cell for some real time VLSI applications such as multipliers, multi-bits adders etc.

5. Acknowledgements We are thankful to the staff of LPU both teaching and non-teaching who stood by our side in terms of providing the required facilities to accomplish our goals.

References R. Zimmermann and W. Fichtner, “Low-power Logic Styles: CMOS versus Pass-Transistor Logic,” IEEE J. Solid-State Circuits, vol. 32, pp.1079-90, July1997. S. Goel, A. Kumar, M. A.Bayoumi, “Design of robust, energy-efficient full adders for deep submicrometer design using hybrid-cmos logic style,” IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol. 14, pp. 1309–1321, Issue 12, Dec. 2006. N. Weste and K. Eshraghian Principles of CMOS VLSI design, a system perspective, Addison-Wesley, 1993. N. Zhuang and H. Hu, "A new design of the CMOS full adder," IEEE J. of Solid-State Circuits, vol. 21, no. 5, pp. 840-844, May 1992. Chip-Hong Chang, Mingyan Zhang and JiangminGU,”A Novel Low Power Low Voltage Full Adder Cell”, Proceedings of the 3rd International Symposium on Image and Signal Processing and Analysis,Proc. ISPAO3, pp.454-458, 2003. T.Sharma, K.G.Sharma, B.P.Singh, N.Arora, “High Speed, Low Power 8T Full Adder Cell with 45% Improvement in Threshold Loss Problem, “Proceedings of the 12th International Conference on Networking, VLSI and Signal Processing, p. 272, Coimbatore and University of Cambridge, UK, Feb. 2010. Yingtao Jiang, Abdulkarim Al-Sheraidah, Yuke Wang,Edwin Sha and Jin-Gyun Chung, “A Novel Multiplexer –Based Low Power Full Adder”, IEEE Transaction On Circuits And System-II: Express Briefs, Vol. 51 ,No. 7, July 2004. Fayed, AA.; Bayoumi, M.A., "A Low Power 10-Transistor Full Adder Cell for Embedded Architectures," in Proc. IEEE In/. Symp. Circuits and Systems, vol. 4, pp. 226-229, Sydney, Australia, May 2001. Shiwani Singh,Tripti Sharma, K. G. Sharma and Prof. B. P. Singh, “New Design Of Low Power 3t Xor Cell” International Journal Of Computer Engineering &Technology. Vol. 3, Issue 1, pp. 76-80, 2012. Afshan Amin,Shivendra Pandey and Jyotirmoy Pathak,“A Review Paper On 3-T Xor Cells And 8-T Adder Design in Cadence 180nm,”IEEE I2CT,Pune,978-1-4799-3759-2/14,2014.

[471]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Synthesis and Characterization of Chemical Bath Deposited CuZnSnS Nano/Microstructures Suresh Kumar*, Virender Kundu, Mamta, Nikhil Chauhan Department of Electronic Science, Kurukshetra University, Kurukshetra, India.

Abstract In this paper CuZnSnS (CZTS) nano/ microstructure grown on commercial glass slide by chemical bath deposition technique has been presented. The as-deposited CuZnSnS nano/microstructures were characterized by X-ray diffractometer, scanning electron microscope and UV-Vis spectrophotometer. The SEM micrograph confirmed the formation of CuZnSnS nano/microstructures and have ball like structures. The XRD studies confirmed the polycrystalline form consisting of kesterite crystal structures nature of Cu2ZnSnS4 nano/microstructure. The optical energy band gap for CuZnSnS Microstructure was found to be 2.36eV. The CZTS nanostructures have potential application in CZTS nanostructured solar cell.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Chemical Bath Deposition, CZTS, Nano/Microstructures, Photovoltaics

1. Introduction Cu2ZnSnS4 (CZTS) is an important and promising photovoltaic material and has been studied widely for solar cell and other optoelectronic device applications. It is considered an excellent absorber layer in thin film solar cell because of low cost, high absorption coefficient in visible light region, optical band gap of 1.45eV and environmental friendly material (Wang et al, 2010; Katagiri et al, 2008). Many researchers studied and described the preparation of CZTS thin film for their potential application in thin film photovoltaic cells (Katagiri et al, 2008; Barkhouse et al, 2012; Bag et al, 2012; Schurr et al, 2012; Schbert et al, 2011; Weber et al, 2009). (Lin et al, 2012) fabricated nanoparticles based solar cell by using CZTS thin film deposited solution-based chemical route as an absorber layer. (Xosrovashvili, Gorji, 2013) analysed the nanostructured hetro-junction solar cell using TiO2 nanoparticles and CZTS thin film. (Jimbo et al, 2007), (Guo et al, 2009), (Moholkar et al, 2011) and (Dhakal et al, 2014) fabricated CZTS thin film solar cells using costly methods such as electron beam evaporation, hot infusion method, pulsed laser deposition and sputtering respectively. In this paper, CZTS nanostructures have been synthesized by low cost chemical bath deposition technique which can be used for fabricating low cost solar cell.

2. Experimental Details CZTS nanostructures have been deposited by chemical bath deposition technique on commercially available glass slides. The cleaned glass substrate is immersed in dilute solutions containing metals, hydroxide, sulfide ions. The bath was prepared by using 0.2M copper sulfate (CuSO4) in 10ml, 0.2M zinc sulfate (ZnSO4) in 10ml, 0.1M tin chloride(SnCl2)) in 10ml, 0.5M sodium thiosulfate (Na2S2O3.2H2O) in 10ml, 0.1M tri-sodium citrate(C6H5Na3O7.2H2O) in 10ml, 2 ml tri-ethanolamine (TEA) and 5ml ammonia in a beaker. In a beaker, a total volume of 57 ml bath solution was prepared by the sequential addition of all the chemicals. First, the chemicals are added and a solution is prepared prior to the addition of TEA and NH3. The color of the solution prepared was light white color (cream color). Now, TEA was added to this solution which turns the color into dark color. Lastly, NH3 was * Corresponding author. Tel.: +91 9416 377282. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Kumar et al/COMMUNE – 2015

added drop wise and the color of the solution was turned to dark brown and continuously controlled the pH of the final solution to 9.5. Continuous stirring was applied during the whole steps involved in preparing final solution. Then, the cleaned glass slide was immersed vertically in the final solution for 12 hours to grow the CZTS nanostructures at room temperature. The glass slides were taken out from the beaker and dried. The dried sample was then characterized to study its structural, morphological and optical properties. 3. Results and Discussion 3.1

SEM Analysis

The morphological studies of chemical bath deposited CZTS nano/microstructures have been studied by scanning electron microscope. The as-prepared samples were viewed under scanning electron microscope (JEOL JSM-6100) at an accelerating voltage of 15kV under high vacuum. The figures 1a and 1b shows the SEM micrographs of as prepared CuZnSnS (CZTS) nano/microstructures at low and high magnifications. The SEM micrographs show the ball like morphologies of CuZnSnS nano/microstructures. The structures are composite of nano/micro crystals in the size range of nano and micrometers.

Fig. 1 CZTS nano/microstructure (a) low magnification (b) high magnification

3.2

Structural Analysis

XRD is the most essential tool used to characterize crystalline nature of a material. The XRD studies of CuZnSnS nano/microstructures were performed using X’PERT-PRO Phillips X-ray diffractometer using Cu-Kα radiation at 45mA, 45keV. Fig.2 shows the XRD patterns of as-deposited CuZnSnS nano/microstructure. The XRD peaks of (110), (112), (200), (105), (220), (312) and (008) which belong to kesterite CZTS (CZTS JCPDS 26-0575) as observed in the as-prepared sample which indicated the formation of kesterite phase of CZTS structures. The crystalline size of the CZTS was calculated from the Debye-Scherrer’s relation (Alexander, Klug, 1950) and was calculated in the range from 80nm to 1μm.

Fig. 2. XRD pattern of CZTS nano/microstructure

3.3

Optical Analysis

The optical analysis plays an important role in studying the optical properties of a material. The optical absorption properties of CZTS nano/microstructures were studied by Shimadzu 2550 UV-Vis spectrometer in the UV/VIS regions of the electromagnetic spectrum. Fig.3a shows the absorbance spectra of as-deposited CuZnSnS nano/microstructure [473]

Kumar et al/COMMUNE – 2015

 h  vs h ).The optical band gap energy of CZTS nano/microstructures is and Fig.3b shows the Tauc plot of calculated from the plot of the absorption coefficient as a function of wavelength by using Tauc relation (Davis, Mott, 1970) given by 1/ 2

 h 

 h 

1/ 2

Fig. 3 (a) Absorption spectra of CZTS (b) Plot of

vs

n

h

 B  h  Eg 

(1)

for CZTS nano/microstructures

where ‘n’ is the index having values 2, 3, 1/2 and 1/3 corresponding to indirect, indirect forbidden, direct and direct forbidden type of band to band transitions respectively, α is the absorption coefficient, B is a constant called band tailing parameter, h is the Planck’s constant, ν is the frequency of incident radiation photon energy and Eg is the optical energy band gap. The extrapolating of the linear portion of Tauc Plot for n=2 in the equation (1), gives the best linear fit for direct transition. From the fig.3b, it was observed that the optical band gap energy of CZTS nano/microstructures is 2.36eV. The optical band gap energy of CZTS thin film was observed between 1.4-1.6eV as reported by (Lin et al, 2012). It is concluded that the energy band gap of CZTS nano/microstructures is increased as compared to CZTS thin films. 4. Conclusions CZTS nano/microstructures have been prepared on the glass substrates via chemical bath deposition approach at room temperature. The X-ray diffraction study revealed that CuZnSnS nano/microstructures were polycrystalline in nature with kesterite phase. The SEM studies revealed that morphologies of CuZnSnS have ball like nano/microstructures. The optical band gap energy of CuZnSnS nano/microstructure is found to be about 2.36eV. The optical studies revealed that CuZnSnS quartnary materials have wide optical bandgap and thus can be used in optoelectronics and future nano-photovolatics devices. It is concluded that chemical bath deposition method is a simple and useful method for the deposition of CuZnSnS thin film and nano/microstructures. 5. Acknowledgements This work is supported by Science and Engineering Research Board, Department of Science & Technology (DST), Govt. of India (Grant No. SERB/F/2139/2013-14). References Wang K., Gunawan O., Todorov T., Shin B., Chey S.J., Bojarczuk N.A., Mitzi D. and Guha S., 2010, Appl. Phys. Lett. 97, p143508. Katagiri H., Jimbo K., Yamada S., Kamimura T., Maw W.S., Fukano T., Ito T. and Motohiro T., 2008, Appl. Phys. Express 1, p41201. Barkhouse D.A.R., Gunawan O., Gokmen T., Todorov T.K. and D. B. Mitzi,2012, Progr. Photovolt.: Res Appl 20, p6. Bag S., Gunawan O., Gokmen T., Zhu Y., Todorov T.K. and Mitzi D.B., 2012, Energy Environ. Sci. 5, p 7060. Schurr R., Holzing A., Jost S., Hock R., Vo T., Schulze J., Ennaoui A., Lux M., Ahmed S., Reuter K.B., Gunawan O., Guo L., Romankiw L.T. and Deligianni H., 2012, Adv. Energy Mater 2, p253. Schubert B.A., Marsen B., Cinque S., Unold T., Klenk R., Schorr R. and Schock H.W., 2011, Progr. Photovolt.: Res. Appl 19, p93. Weber A., Krauth H., Perlt S., Schubert B., Kotschau I., Schorr S. and Schock H.W., 2009, Thin Solid Films 517, p2524. Lin X., Kavalakkatt J., Kornhuber K., Levcenko S., Lux-Steiner M.C. and Ennaoui A., 2012, Thin Solid Films, http://dx.doi.org/10.1016/j.tsf.2012.10.034. Xosrovashvili G. and Gorji N.E., 2013, Journal of Modern Optics, 60 (11), p936. Jimbo K., Kimura R. and Kamimura T., 2007, Thin Solid Films, 515 (15), p 5997. Guo Q., Hillhouse H.W., and Agrawal R., 2009, Journal of the American Chemical Society, 131(33), p. 11672. Moholkar A.V., Shinde S.S., A. R. Babar et al., 2011, Solar Energy, 85(7), p. 1354. Dhakal T.P., Peng C.Y., Tobias R.R., Dasharathy R. and Westgate C.R., 2014, Solar Energy,100, p23. Alexander L. and Klug H.P., 1950, Journal of Applied Physics 21, p137. Davis E.A. and Mott N.F., 1970, Phil. Magn. 22, p903.

[474]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16-18March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Verification using Multimodal Biometric Fusion Saba Mushtaq*, Shoaib Amin Banday, Ajaz Hussain Mir Department of Electronics and Communication Engineering, National Institute of Technoloy Srinagar, India

Abstract Verification using biometrics has offered a wide range of advantages over conventional possession and knowledge based methods. Almost all the biometric modalities have been tested by now but there are various factors that limit their accuracy. This paper presents a multimodal biometric system for verification. We have fused the matching scores for features extracted from iris and handwritten signatures. GLCM features for iris and GLRLM features for signatures have been used. The verification results so obtained are exceptionally good in comparison to both unimodal iris verification and handwritten signature verification systems.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Multimodal Biometrics; Iris Recognition; Handwritten Signature Verification; Texture Recognition.

1. Introduction A large variety of application requires confirming the identity of the individual before providing them access to the application. Some of these applications need to be highly secured from illegitimate access like bank transaction, security systems, surveillance databases, entrance to high security zone areas etc. Even a small amount of inaccuracy can compromise the access. Biometrics has become a prominent and popular technique to provide personal verification. A biometric system simply matches a pre-stored sample of an individual with the current input sample and matches certain feature to ascertain the identity of the individual. A biometric system operates to either identify or verify an individual. In identification a comparison is made between the submitted sample and all N stored samples in the database (1: N) while as in verification in addition to a submitted sample some pin or password are also entered and a 1:1 comparison is made. A single biometric system may fail to extract enough information for verifying an individual so multimodal biometric that is, biometrics that involve more than one biometric modality to obtain improved performance are used. The most important feature of multimodal systems is to collect information from multiple biometric modalities to reduce the error introduced in monomodal systems (Ross et al 2006). Multimodal systems make it difficult for an intruder to copy more than one biometric traits. The main aim of multimodal systems remains to fuse information obtained from biometric samples at different fusion levels (Rattani et al. 2006). This fusion can be performed at four different levels sensor level, feature level, matching level and decision level. The first two levels i.e. sensor and the feature level are referred to as a pre-mapping fusion while as if the matching is performed at matching score level and the decision level then it is referred to as a post-mapping fusion (Sanderson.C, K. K. Paliwal 2003). In this paper, we fuse the information at matching score level. We have made use of GLCM (Grey level co-occurrence matrix) to calculate features of right iris and GLRLM (Grey level rum length matrix) to calculate features of handwritten signature images. Iris recognition is the most promising for high security environments (J. Daugman 1993). Iris based biometric recognition systems have achieved a very high accuracy as high as 97% (C. Sanchez-Avila et al. 2001). A brief description of the two texture based techniques viz GLCM and GLRLM are given in next section.

*

Corresponding author. Tel.: +91 9906 118357. E-mail address: [email protected]. ISBN: 978-93-82288-63-3

Mushtaq et al./ COMMUNE-2015

2. Generalized Description of Feature Extraction schemes Texture is one of the important characteristics used in identifying an object in an image and to discriminate the images. The texture coarseness or fineness of an image can be interpreted as the distribution of the elements in the matrix (Harlic 1973). The gray tone spatial dependence was first used by ( Julesz 1962) for texture classification. 2.1. GLCM GLCM is a second order statistics method, which describes the spatial interrelationships of the gray tones in an image (R.W. Conners and C. A. Harlow, 1980). It contains elements that are counts of the number of pixel pairs, which are separated by certain distance and at some angular direction. Typically, GLCM is calculated in a small window, which scans the whole image. Bachoo and Tapamo (2005) have used GLCM for pattern analysis of iris however in this method the selection of window size remains a problem. In the proposed scheme we have normalize, the GLCM and assumed GLCM represent probabilities instead of counts. The co-occurrence matrix is constructed by the joint probability density function between the gray level tones, which gives the spatial relationship between any two points in the image. It is denoted by P(i,j,d,θ), where i and j give ith line and jth column of co-occurrence matrix respectively, d is the distance between any two points and θ is the direction. Normalization involves dividing by the total number of counted pixel pairs. There are eight texture features based on GLCM as studied by (Haralick 1973). The mathematical expressions for these features are given below:

Entropy

=

(1)

Correlation

=

(2)

Contrast

=

(3)

Dissimilarity

=

(4)

Homogeneity

=

(5)

Angular Second Moment

=

(6)

Mean

=

i(µ)i

(7)

Variance

=

(σ 2 )

(8)

Herei GLCM features are computed based on two parameters, which are the distance between the pixel pair ‘d’ and their angular relation θ. The angular relation is quantized at four angles i.e., 00, 450, 900 and 1800. 2.2. GLRLM The technique used to calculate features of handwritten signatures is GLRLM. The GLRLM is based on computing the number of gray-level runs of various lengths. A gray level run is a set of consecutive and collinear pixel points having the same gray level value. The length of the run is the number of pixel points in the run. The Gray Level Run Length matrix is constructed as follows:

R(θ) = ( g (i,j) | θ ), 0 ≤ I ≤ Ng , 0 ≤ I ≤ Rmax;

(9)

Where Ng is the maximum gray level and Rmax is the maximum length. Let p (i, j) be the number of times there is a run of length j having gray level i. There are five Run Length Matrix based features computed for 4 directions of run (0°, 45°, 90°, 135°). For each matrix in a particular direction following seven GLRLM features viz SRE, LRE, GLN, RLN, RP, LGLRE, HGLRE are obtained. These features were suggested by Gallow 1975

[476]

Mushtaq et al./COMMUNE – 2015 Table 1 : GLRLM features

S.NO.

Features

1

Short Run Emphasis(SRE)

2

Long Run Emphasis(LRE)

3

Grey Level Nonuniformity(GLN)

4

Run Length NonUniformity

5

Run Percentage(RP)

6

Low Grey level Run Emphasis(LGLRE)

7

High Grey Level Run Emphasis(HGLRE)

Formulae

3. Proposed Scheme This section presents the proposed scheme for verification using multimodal biometric fusion. 3.1. Database We have used two databases one for signatures and one for iris images. For iris experiments are carried on CASIAiris-V4 thousand database and for hand written signatures a signature database collected at NIT Srinagar is used. 3.2. Algorithm The above discussed textural feature extraction method are used to extract features from right iris and hand written signatures. GLCM textural features are calculated using equations shown in section 2. The distance d between pixel pairs is first selected as 1 and gray level co-occurrence matrix features are calculated. The features are calculated at an angles of 00,450,900 and 1800. The experiment is continued with calculation of GLCM features for value of d=2 and 4. The features are again measured at the pre-defined angles of rotation over the whole iris image. To make the GLCM invariant to the rotation of the images, GLCM obtained at d = 1,2 and 4 is averaged through four angular relations (00,450,900 and 1800 ). Once the textural features of image (iris) is calculated, the feature set is stored in the database as trainer. The test image (iris) is similarly processed to obtain a textural feature vector. This feature vector of the test image is processed by the matching unit of the multi unit biometric verification system and compared against the templates stored in the database. The matching unit because of taking Euclidean distance as the classifier outputs a matching score corresponding to each template in the database. The matching score is fed to the decision unit which because of some predefined threshold classifies test image as Genuine or Imposter based on the score obtained at the matching unit. Similarly, the signature image is pre- processed separately as shown in the figure 1. The GLRLM features are extracted for the signature image in the similar way followed by matching. In case Once the scores are obtained from both the matching units ( i-e., from right iris and handwritten signatures), they are fused using SUM method of fusion.

[477]

Mushtaq et al./ COMMUNE-2015

Fig 1: Block Diagram of proposed system

4. Experimental Results We evaluated the proposed system on a data set of 525 signature images collected at NIT Srinagar and CASIA-irisV4 thousand database. The training set contains 200 each, a signature and an iris image assumed to be belonging to same individual. The testing set is a set of 50 each, signatures and iris images for verification. 4.1

Training

For an individual one image for each trait i.e iris and signature is enrolled in the database for which features are extracted using GLCM and GLRLM respectively. Which are also used for score level fusion and saved in database for verification. 4.2

Testing

Pair of iris and signature are used for testing.. Fused feature vector is generated from the pair and is compared with the database score value.We have calculated FAR and Accuracy for the individual systems and then for the multimodal proposed system. The results thus obtained are given in table below. We also calculated the Genuine and imposter scores for the two systems that is for iris using GLCM and for Signatures using GLRLM and for the proposed system which are given below in fig2, fig 3 and fig4 respectively . It can be clearly seen that the overlap of genuine and imposter scores is more in unimodal systems and for every verification systems the aim is to reduce this overlap as much as possible. Lesser the overlap more accurate is the system.

[478]

Mushtaq et al./COMMUNE – 2015

Fig 2 : Score distribution for iris using GLCM

Fig 3 : Score distribution for Signature using GLRLM

Fig 4 Score density for proposed system. Table 2: Accuracy and FAR S. No.

Modality

FAR

Accuracy

1 2 3

Signatures Iris Proposed Multimodal system

13.33 0.8 0.17

85.15% 93.2% 97.6%

5. Conclusion The field of multimodal biometrics is a challenging and novel area of research aiming at reliable and accurate personal verification and identification. This paper presents a score level fusion technique for human verification. The proposed technique uses texture features viz GLCM and GLRLM for feature extraction. The features thus obtained are normalized. In addition, stored in database. Matching scores are calculated for iris and signatures separately, then fused using sum rule of fusion, and results are compared to unimodal systems based on accuracy and FAR. The experimental results establish the effectiveness of fusion of the individual matching scores and accuracy of 97.6% is abtained in comparision to individual signature and iris biometric systems that provide accuracy of 85.15% and 93.2 % respectively. References Asheer Kasar Bachoo and Jules-Raymond Tapamo : Texture detection for segmentation of iris images. In Annual Research Conference of the South African Institute of Computer Information Technologists, pages 236–243, 2005. C. Sanchez-Avila, R. Sanchez-Reil, D. de Martin-Roche : Iris-Based Biometric Recognition using Dyadic Wavelet Transform , IEEE AESS Systems Magazine, October 2002 Ross. A., Nandakumar, K., Jain, A.K.: Handbook of Multibiometrics. Springer Verlag (2006) Rattani, A., Kisku, D.R., Bicego, M., Tistarelli, M.: Robust Feature-Level Multibiometrics Classification. IEEE Biometric Consortium Conference, Biometrics Symposium, pp. 1—6 (2006) R.W. Conners and C. A. Harlow. A theoretical comparison of texture algorithms. IEEE Trans. on Pattern Analysis and Machine Intelligence, 2(3):204–222, 1980. Sanderson.C, K. K. Paliwal. Information Fusion and Person Verification Using Speech and Face Information. IDIAP-RR, pp.02-33, 2003. Haralick, R.M. , Shanmugan, K.. and Dinstein, I.( 1973) ‘Textural Features for Image Classification’, IEEE Tr. on Systems, Man, and Cybernetics, Vol SMC-3, No. 6, pp. 610-621 J. Daugman, November 1993, High confidence visual recognition by test of statistical independence, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 15, pp. 1148-1 161. Julesz B, “Visual pattern discrimination,” IRE T M In~form, Theory, vol. 8, no. 2, pp. 84-92,Feb. 1962. M. M. Galloway,“Texture analysis using gray level run lengths”, Computer Graphics Image Process., Vol. 4, pp. 172–179, June 1975.

[479]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Extension to the K-Means Algorithm for Automatic Generation of Clusters for Mixed Datasets Anupama Chadhaa*, Suresh Kumarb a Faculty of Computer Applications, MRIU, Faridabad, India Faculty of Engineering and Technology, MRIU, Faridabad, India

*

Abstract A lot of work has been done and is still in progress on the famous partition based K-Means clustering algorithm. Various forms of KMeans have been proposed depending on the type of data sets being handled. Most popular ones are K-Modes for categorical data and K-Prototype for mixed numerical and categorical data. In all these forms of K-Means, one major limitation is dependency on prior input of number of clusters K, and sometimes it becomes practically impossible to correctly estimate the optimum number of clusters in advance. Various ways have been suggested in literature to overcome this limitation for numerical data. But for categorical and mixed data work is still in progress. In this paper, we have proposed a method for clustering mixed data, which is based on K-Means, but has advanced features for automatic generation of appropriate number of clusters.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Clustering; K-Means; Mixed Dataset; Automatic Generation of Clusters

1. Introduction Clustering is the method of segregating the objects into groups such that the objects in a certain cluster have high degree of similarity with each other than the objects in the other clusters. The clusters formed are then analyzed for making decisions . For example, in Educational Institutes, clustering students according to their academic performance can help in identifying the weaker ones, who can be provided with more tuitions. In Health sector, data mining and clustering may help in identifying disease symptoms, ethnicities, water quality etc. and establishing links between them. In Banking sector it can be used to group customers with overdue credit card payments. In Market Research, Data Mining and Clustering can be used to identify customers having certain buying patterns. Number of clustering methods have been suggested in the literature (Guojun et al, 2007; RuiXu et al, 2005). Out of these, the partition based method is known for its speed in clustering large data sets. One of the famous algorithm based on this method is K-Means. K-Means is a simple algorithm known for its speed. However, there exist some limitations in this algorithm. One major limitation is the requirement to input the number of clusters at the very beginning based on anticipation.. This is domain specific, and if the person using the algorithm is not domain expert, then an incorrect number of clusters may be input, leading to inefficient grouping. Work has been done for auto generation of clusters with numerical data. However, for categorical and mixed type of data, this limitation still exists. In this paper we have proposed an algorithm based on K-Protoype, which is an extension of K-Means to deal with mixed type of data to automatically generate suitable number of clusters. 2. Literature Survey 2.1. This part of the literature survey discusses the works of the authors done to remove the limitation of inputting the number of clusters required with numerical data set in K-Means.

*

Corresponding Author. Tel.: +91 9818 856542. Email-Address: [email protected]. ISBN: 978-93-82288-63-3

Chadha and Kumar/COMMUNE – 2015

Pelleg et al. (2000) suggested XMeans algorithm as an extension of K-Means which required the user to input a range representing the lower and upper instead of a particular value of K. The algorithm initially takes lower bound of the given range as K and continues to add centroids until the upper bound is reached. The algorithm terminates as soon as it gets the centroid set that scores the best. The drawback lies in the fact that it requires the user to input a range suggesting the lower and upper bound of K. Tibshirani et al. (2000) used the technique of Gap Statistic. In this technique output generated by any clustering algorithm was used to compare the change in within cluster dispersion to that expected under an appropriate reference null distribution. The algorithm works well with well separated clusters. Wagstaff et al. (2001) suggested utilizing information about the problem domain in order to put some constraints on the data set. During the clustering process it is ensured that none of the constraint is violated. This algorithm requires some domain specific information, which sometimes becomes difficult to attain. Cheung (2003) proposed a new extension of K-Means clustering technique named STep-wise Automatic Rival penalized (STAR) K-Means algorithm overcoming two of its major limitations of dependency on initial centroids and inputting K. In the first step of the algorithm cluster centres are provided and in the second step the units are adjusted adaptively by a learning rule. The limitation of this algorithm is the complex computation involved in it. Shafeeq et al. (2012) proposed an algorithm in which the optimal number of clusters was found on the run. The main drawback of the proposed approach is that its computational time is more than the K-Means for larger data sets. Also the user has to input the number of clusters (K) as 2 in the first run. Leela et al. (2013) proposed Y-means algorithm. Initially, clusters are found using K-Means algorithm on the data set. A sequence of splitting, deleting and merging the clusters is then followed to find the optimal number of clusters. The limitation of this algorithm is that it depends on K-Means algorithm to find the clusters initially. Abubaker et al. (2013) presented a new approach to based on the K-Nearest Neighbour method. The only input parameter taken by the algorithm is kn (the number of nearest neighbour). The drawback of this algorithm is that we have to input the number of nearest neighbours kn. 2.2. In this part, the work done in the field of automatic generation of clusters with categorical data has been discussed. Not much of the work has been done in this field. In this section we will be discussing two of the research papers dealing with this limitation of inputting the value of K. Liao et al. (2009) proposed the new algorithm which extends the K-Modes clustering algorithm by introducing a regularization parameter to control the number of clusters in a clustering process. A suitable value of regularization parameter is chosen to generate the most stable clustering result. The major limitation of the above proposed algorithm is that the computation involved is much more than the original K-Modes algorithm. Also this algorithm requires an input parameter representing the initial cluster centres. Cheung et al. (2013) proposed a similarity metric that can be applied to categorical, numerical, and mixed attributes. Based on this similarity metric an iterative clustering algorithm is developed to overcome the limitation of inputting K. This algorithm requires some initial value of K which should not be less than the original value of K. The cluster accuracy is more as compared to the original K-Modes and K-Modes with Ng’s dissimilarity metric (Ng, 2007). But as in the clustering algorithm proposed by (Liao, 2009), this algorithm too has much computation involved in it. 2.3. Very little work has been done to automatically generate clusters in a dataset containing mixed attributes. Two research papers providing solution to this problem are discussed below: Liang et al. (2012) extended K-Prototype algorithm by proposing a new dissimilarity measure for mixed data set. A mechanism was developed to make use of within-cluster entropy and between-cluster entropy to identify the worst cluster in a mixed dataset. The major limitation of the above proposed algorithm is that this algorithm requires input parameters representing the minimum and maximum number of clusters that can be generated from the data set. Ahmad et al. (2007) proposed a new cost function for mixed data set. The authors extended K-Means algorithm that worked well for data with mixed numeric and categorical features by introducing a new distance measure and a new way of finding the centroids of the clusters. The algorithm used the concept of the significance of an attribute in the clustering process. In this algorithm computation times increases with the increase in the dimensions of the data. 3. Proposed Algorithm As discussed in section 2.3, the algorithms proposed to overcome the limitation of inputting the value of K for mixed data require either some input parameter or some initial value of K to achieve good clustering results. In the proposed algorithm we have extended K-Prototype algorithm to overcome this limitation. In the proposed algorithm we have utilized the methods proposed by (Ahmad et al, 2007) to find the centroids of the clusters and the distance of the records from the centroid. Also we have utilized the concept of the most significant attribute suggested by (Ahmad et al, 2007) to create initial clusters. To compute the significance of numeric attributes, they have been discretized using equal

[481]

Chadha and Kumar/COMMUNE – 2015

width interval method (Ahmad et al, 2007). Input: dataset of n objects with m mixed attributes. Output: clusters or groups distributing the objects in the given dataset. a) b) c) d)

e)

f) g) h) i) j) k)

Discretize all numerical attributes and find the most significant attribute using the way proposed by (Ahmad et al, 2007) Create initial clusters with attribute values with maximum distance in most significant attribute in different clusters. Normalize all the numerical attributes and find the centroids of all the clusters using the way proposed by (Ahmad et al, 2007). Find the distance of every tuple from the respective centroids using the way proposed by (Ahmad et al, 2007). The minimum of all the average distance of every cluster of all the distance values is taken as d. Find the outliers in the initial two clusters according to the following objective function: i. Distance(x, q) <=d not an outlier. ii. where Distance(x,q) is the distance between object x and centroid q. Calculate the new centroids of the clusters. Calculate the distance of every outlier from the new cluster centroids and find the outliers not satisfying the objective function in step 5. Let B={Y1,Y2,…..Yp) be the set of outliers obtained in step 7 (value of p depends on number of outliers). Repeat until (B==Φ). assume this set B as a new data set. perform steps 1 to 8.

4. Experimental Run of Proposed Algorithm The proposed algorithm is explained using a small dataset with categorical and numerical attributes as shown in Table 1. Table 1. Dataset with mixed attributes Alpha A A A B B B A

Beta C C D D C D D

Gamma 1.2 3.0 3.2 5.0 4.2 3.5 4.8

Discretize attribute Gamma and represent the values as ‘a’ and ‘b’. After discretizing the table 1 will appear like this: Table 2. Dataset after discretizing Table 1 Alpha A A A B B B A

Beta C C D D C D D

Gamma A A B B B B B

In order to find the most significant attribute calculate the conditional probabilities and the distance between various values of the attributes as given in Table 3 and Table 4.

Table 3. Probability table for attribute Alpha Conditional Probability with respect to attribute Beta P(C|A)=1/2 P(D|A)=1/2 P(C|B)=1/3 P(D|B)=2/3

[482]

Conditional Probability with respect to attribute Gamma P(a|A)=1/2 P(b|A)=1/2 P(a|B)=0 P(b|B)=1

Chadha and Kumar/COMMUNE – 2015

Similarly calculate the conditional probabilities of Beta and Gamma. Calculate the distance between various values of the attributes of Table 2. Table 4 shows all the distance values. Table 4. Distance between various values of the attributes of Table 2 Alpha δ(A,B)=1/3

Beta δ(C,D)=5/12

Gamma δ(a,b)=7/10

As the distance between the values ‘a’ and ’b’ of attribute Gamma is the highest so it comes out to be most significant attribute. a) Initial clusters are formed by keeping the values ‘a’ and ‘b’ of attribute Gamma into separate clusters. Cluster 1 {(A,C,1.2), (A,C,3.0)}

Cluster 2 {(A,D,3.2),(B,D,5.0),(B,C,4.2),(B,D,3.5),(A,D,4.8)}

b) After normalizing the attribute gamma the records in cluster1 and cluster2 appear as shown in Table 5. According to the normalization scheme used by (Ahmad et al, 2007), for ith attribute and kth value, the normalized value of xik is obtained as dik, as shown in Equation 1. 𝑑𝑖𝑘 = (𝑥𝑖𝑘 − 𝑥𝑖,𝑚𝑖𝑛 )/(𝑥𝑖,𝑚𝑎𝑥 − 𝑥𝑖,𝑚𝑖𝑛 )-- (1) Table 5. Records in cluster 1 and cluster2 after normalizing attrinute Gamma Cluster1 Alpha A A

Beta C C

Cluster2 Gamma 0 1

Alpha A B B B A

Beta D D C D D

Gamma 0 1 0.56 0.17 0.89

Centroid of cluster1 is: 1/2(2A), 1/2(2C), 0.5, Centroid of Cluster2: 1/5(2A,3B), 1/5(C,4D), 0.52 Table 6 shows the distance of the records from their respective centroids for cluster1 and cluster2. a) The total distance between an object and a cluster center as proposed by (Ahmad et al, 2007) for a mixed data set is defined in Equation 2. 𝑚𝑟 𝑚𝑐 𝑟 𝑐 𝜗(𝑑𝑖 , 𝐶𝑗 ) = ∑𝑡=1 (𝑤𝑡 (𝑑𝑖𝑡 − 𝐶𝑗𝑡𝑟 ))2 + ∑𝑡=1 (Ω(𝑑𝑖𝑡 , 𝐶𝑗𝑡𝑐 ) )2 − −(2)

Where mr and mc represent the number of numeric and categorical attributes, respectively. b) The various components of distance of record (A,C,0) in clsuter1 is calculated as : 1 𝑐 𝑐 ) 𝛿(𝑑11 , 𝐶11 = ( (2𝛿(𝐴, 𝐴)))2 = 0 𝑎𝑠 𝛿(𝐴, 𝐴) = 0 2 1 𝑐 𝑐 𝛿(𝑑12, 𝐶12 ) = ( (2𝛿(𝐶, 𝐶)))2 = 0 𝑎𝑠 𝛿(𝐶, 𝐶) = 0 2 𝑟 𝑟 (𝑤1 (𝑑11 − 𝐶11 ))2 = (0(0.5 − 0))2 = 0 Table 6. Distance of the records from their respective centroids

Distances in Cluster1 δ(A,C,0)=0 δ(A,C,1)=0

Distances in Cluster2 δ(A,D,0)=0.02 δ(B,D,1)=0.01 δ(B,C,0.56)=0.16 δ(B,D,0.17)=0.01 δ(A,D,0.89)=0.02

c) The average of all the distances in cluster1 is 0 and cluster2 is 0.04. So d=0.04. d) There is only one outlier in cluster2: (B,C,4.2) New cluster2 {(A,D,3.2), (B,D,5.0),(B,D,3.5),(A,D,4.8)}

[483]

Chadha and Kumar/COMMUNE – 2015

e) f)

New centroid of cluster2 is 1/4(2A, 2B), 1/4(4D), 0.52. After calculating the distance of the outlier from centroid of cluster1 and new centroid of Cluster2, it was found that they did not fit either in cluster1 or cluster2. g) Since there is only one outlier, so create a new cluster with this outlier as its member. (otherwise steps 1 to 8, would have been repeated) Finally three clusters are formed: Cluster1: {(A,C,1.2), (A,C,3.0)} Cluster2: {(A,D,3.2),(B,D,5.0),(B,D,3.5),(A,D,4.8)} Cluster3: {(B,C,4.2)} The clusters obtained are compared with the clusters obtained using K-Prototype algorithm using software RapidMiner as shown in Table 7. Table 7. Results of K-Modes and Proposed algorithm for the data set in Table 1. Cluster1

K-Prototype {(A,C,1.2),(A,C,3.0)}

Cluster2

{(B,D,5.0), (B,C,4.2)}

Cluster3

{(A,D,3.2), (B,D,3.5),(A,D,4.8)}

Proposed Algorithm {(A,C,1.2),(A,C,3.0)} {(A,D,3.2),(B,D,5.0),(B,D,3.5),(A, D,4.8)} {(B,C,4.2)}

As can be seen from Table 7, both the algorithms have tried to categorize the data set with objects having similarity in two of the attributes out of three, though the results are slightly different. The algorithm is further explained using a data set with two numerical and one categorical attribute. Table 8. Dataset with mixed attributes Subject1 A B C B B C

a)

Subject2 80 60 40 65 62 35

Subject3 90 68 38 67 65 36

Discretize attributes Subject2 and Subject3. After discretizing the Table 8 will appear like this:

Table 9:. Dataset after discretizing Table 8 Subject1 A B C B B C

Subject2 A B C A B C

Subject3 A B C B B C

b) The signifance value of Subject1 and Subject3 come out to same, so choose any of the two as the most significant attribute. Choosing Subject1 as the most significant attribute initial clusters are: Cluster 1 {(A,A,A), (B,B,B), (B,A,B), (B,B,B)}

c)

Cluster 2 {(C,C,C), (C,C,C)}

After normalizing the attributes Subject2 and Subject3 the records in cluster1 and cluster2 appear as shown in Table 10.

Table 10. Records in cluster 1 and cluster2 after normalizing attrinutes Cluster1

Cluster2

Subject1

Subject2

Subject3

Subject1

Subject2

Subject3

A B B B

1 0 0.25 0.1

1 0.12 0.08 0

C C

1 0

1 0

Centroid of Cluster1: 1/4(A,3B), 0.34, 0.3, Centroid of Cluster2: 1/2(2C), 0.5, 0.5 [484]

Chadha and Kumar/COMMUNE – 2015

d)

Find the outliers by calculating the distance of the records from their respective centroids.

Table 11. Distance of the records from their respective centroids Distances in Cluster1 δ(A,1,1)=0.51 δ(B,0,0.12)=0.057 δ(B,0.25, 0.08)=0.039 δ(A,0.1,0)=0.018

Distances in Cluster2 δ(C,1,1)=0 δ(C,0,0)=0

The average of all the distances in Cluster1 is 0.16 and cluster2 is 0. So d=0.16 So outlier in cluster1 is (A,80,90). New Clsuter1: {(B,60,68), (B,65,67), (B,62,65)} Cluster2: {(C,40,38), (C,35,36)} e) After calculating the distance of the outlier from new centroids of cluster1 and centroid of Cluster2, it was found that it did not fit either in cluster1 or cluster2. Since only one outlier is there so a new cluster3 is formed containing this outlier. Finally, 3 clusters are formed: Cluster1: {(B,60,68), (B,65,67), (B,62,65)} Cluster2: {(C,40,38), (C,35,36)} Cluster3: {(A,80,90)} Table 12. Results of K-Prototype and Proposed algorithm for the data set in Table 8. K-Prototype Cluster1

{(B,60,68),(B,65,67), (B,62,65)}

Cluster2 Cluster3

{(C,40,38), (C,35,36)} {(A,80,90)}

Proposed Algorithm {(B,60,68), (B,65,67), (B,62,65)} {(C,40,38), (C,35,36)} {(A,80,90)}

As can be seen from Table 12, the results of both the algorithms are same. 5. Conclusion and Future Work In this paper we have proposed an algorithm based on K-Prototype algorithm which creates appropriate number of clusters without need of prior input of number of clusters, K. The proposed algorithm is explained using two small synthetic data sets. The results obtained found to be better than the original K-Prototype. In Future Work, the proposed algorithm will be implemented on many real data sets and the accuracy of the clusters produce will be compared to that of original K-Prototype algorithm. References Abubaker, Mohamed, Ashour, Wesam, 2013. Efficient Data Clustering Algorithms: Improvements over Kmeans. International Journal of Intelligent Systems and Applications, Vol.5, no.3, p. 37-49. Ahmad, Amir, Dey, Lipika, 2007. A K-Mean Clustering Algorithm for Mixed Numeric and Categorical Data. Data & Knowledge Engineering, 63, p. 503–527. Cheung, Yiu-Ming, 2003. k*-Means: A new generalized k-means clustering algorithm. Pattern Recognition Letters, 24, p. 2883–2893. Cheung, Yiu-ming, Jia, Hong, 2013. Categorical-and-numerical-attribute data clustering based on a unified similarity metric without knowing cluster number. Pattern Recognition, 46, p. 2228–2238. Gan, Guojun , Ma Chaoqun, Wu, Jianhong , 2007. Data clustering: theory, algorithms, and applications. SIAM: Society for Industrial and Applied Mathematics. Leela, V., Sakthipriya, K., Manikandan, R., 2013. A comparative analysis between k-mean and y-means Algorithms in Fisher’s Iris data sets. International Journal of Engineering and Technology, Vol.5, no.1, p. 245-249. Liang, Jiye, Zhao, Xingwang, Li, Deyu, Cao, Fuyuan, Dang, Chuangyin, 2012. Determining the number of clusters using information entropy for mixed data. Pattern Recognition, 45, p. 2251–2265. Liao, H., Ng, M., K., 2009. Categorical Data Clustering with Automatic Selection of Cluster Number. Fuzzy Information and Engineering, Vol.1, no.1, p. 5–25. Ng, K., Michael, Li, Junjie, Mark, Huang, Zhexue, Joshua, He, Zengyou, 2007. On the Impact of Dissimilarity Measure in k- Modes Clustering Algorithm. IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.29, no.3, p. 503-507. Pelleg, Dan, Mooreg, W., Andrew, 2000. X-means: Extending K-means with Efficient Estimation of the Number of Clusters. Proceeding ICML '00 Proceedings of the Seventeenth International Conference on Machine Learning, p. 727-734. Shafeeq, Ahamed, B., M., Hareesha, K., S., 2012. Dynamic Clustering of Data with Modified K-Means Algorithm. International Conference on Information and Computer Networks (ICICN 2012), IPCSIT, 27, p. 221-225. Tibshirani, R., Walther, G., Hastie, T., 2000. Estimating the number of clusters in a dataset via the gap statistic. Technical Report 208, Department of Statistics, Stanford University, California. Wagstaff, Kiris, Cardie, Claire, Rogers, Seth, Schroedl, Stefan, 2001. Constrained K-means Clustering with Background Knowledge. Proceedings of the Eighteenth International Conference on Machine Learning, p. 577-584. Xu, Rui, Wunsch, Donald, 2005. Survey of Clustering Algorithm. IEEE Transactions on Neural Networks, Vol.16, no.3, p. 645- 678.

[485]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Implementation of an Embedded Device to Prevent Friendly Fire in Battle Field Padma Prasad, Sathisha* Department of Electronics and Communication Engineering, Mangalore Institute of Technology and Engineering, Moodabidri, India

Abstract Friendly fire is the situation where military forces attacks on friendly forces while attempting to attack the enemy. Causative reasons may be either misidentifying the target as hostile or due to errors. Such attacks results in unwanted casualties in the battlefield. Utilization of improved technology to assist in identifying friendly forces is an ongoing process to prevent any fratricides. In this proposed paper, one such attempt is made to develop an embedded device to identify friendly forces. In this approach, encrypted data is sent through laser beam for an authentication. Implementation method considers power down strategies, efficient encryption and decryption methodologies, and theft security for the module. Software programming algorithm and their implementation is also discussed. PIC16F877A from Microchip Corporation is used as heart of this design. It features 256 bytes of EEPROM, selfprogramming, 8 channels of 10 – bit analog to digital converter, UART. using mickroC PRO IDE, entire software programming done in embedded C language.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Encrypted Laser Beam; Fratricides; Friendly Fire; Pulse Detector; PIC Microcontroller

1. Introduction This paper aims in developing an embedded system, which identifies friendly forces to assist in prevention of fratricides at battlefield. In the confusion of battle, it is easy to shoot friendly forces, accidently. The effects of friendly fire are not just unwanted casualties, but being hit by their own forces causes a huge negative impact on morale. Forces start doubting the ability of their command, and existence of friendly fire makes commanders more cautious in the battlefield. To reduce this effect by military leaders generally starts with identifying the causes of friendly fire and overcoming repetition of such incidents through supplying adequate training, tactics and making use of technology. In this proposed system, encrypted laser beam is being transmitted from the initiative unit to the opponent responder unit. The responder unit is expected to decrypt the received encrypted data and sends back a message indicating friendly force is identified. A soldier equipped with the proposed system has a responder unit on their body armour, and an initiator unit mounted on their rifle. The rifle module transmits an encrypted laser beam. If the rifle points towards a friendly force, phototransistors equipped on the target body armour detects an incident of laser beam. If decryption is successful, the master control unit identifies which friendly force is currently targeted. A pulse detector is used to ensure that the system to be deactivated soon after separation from the soldier body; enemy forces cannot recover system from dead soldier and use it to masquerade as friendly force. The rest of this paper is organized as follows: section II has discussion on case study of friendly fire crisis in past. Section III exposes system configuration and architecture. Section IV discusses hardware design and interface techniques. In section V, Implementation and workflow of the system is portrayed. In Section VI, validation and experimental results are discussed. Section VII concludes this paper.

* Corresponding author. Tel.: +91 9538 651528. E-mail address: [email protected].

ISBN: 978-93-82288-63-3

Padma and Sathisha/COMMUNE – 2015

2. Friendly Fire: Case Study Most militaries use extended training to ensure troop safety as part of normal coordination and planning, but are not always exposed to possible friendly fire situations to ensure they are aware of the situations. Some tactics make friendly fires are inevitable. History gives many such examples of friendly fire causing death of royalist commanders. Few cases are discussed in this section. Six Philippine soldiers were killed in one of the bloodiest clashes against al – Qaida – linked militants on July 11, 2014 were hit by friendly fire. A former football player, American Ranger Pat Tillman were shot and killed by American forces in 2004. In 2009, a British Military Police Officer was shot and killed by British sniper while on patrol. In Afghanistan Five American soldiers were killed by friendly fire incident on June 10, 2014. On October 2012, Indian army witness friendly fire led to death of one soldier and injuries to two. Army spokesperson says that ‘these are the realities of combat. Factors like combat stress, the fog of war affect decision making’. Proper use of advanced technologies may found helpful in solving friendly fire problems. In response to such crisis, in this proposed paper an attempt is made to develop an embedded device to identify friendly forces in battlefield and avoid unwanted casualties. 3. System Configuration and Architecture The proposed system consists of two major working units, initiator unit and a responder unit. Initiator unit is mounted on rifles of soldier and responder unit equipped with identification of friend/foe system is placed in the soldiers body armour. Its functional diagram is depicted in Fig 1. Initiator unit mounted on the soldier’s rifle generates a key encrypted laser beam; identification of friend or foe query. If the rifle points on friendly soldier, phototransistor mounted on soldiers body armour; responder unit, will detect an incident of laser beam. The microcontroller unit in the responder unit performs the decryption of encoded laser beam. If decryption is successful, MCU unit identifies which friendly soldier is currently in target. This information is broadcasted through RF packet generation.

Fig 1: Block Diagram of Proposed System.

4. Hardware Design and Interfacing Techniques As discussed earlier the said system consist of two major units; an initiator unit and a responder unit. This section focuses mainly on hardware design techniques of these two units. 4.1.

Initiator Unit

Initiator module is responsible for generation of encrypted laser beam to identify friendly soldier to prevent possible fratricides. Hardware design of initiator unit is shown in Fig. 2. The PIC microcontroller generates a PWM signal that is used to switch the MOSFET; regulates the laser output. This enables a laser beam appears to be steady to the soldier, but a very rapid series of pulse with an interval unique to each soldier. This unit also periodically checks for the friend foe identification RF signal from the responder unit.

[487]

Padma and Sathisha/COMMUNE – 2015

Fig 2: Initiator Unit Hardware Design.

4.2.

Responder Unit

Responder unit is to be mounted on soldier’s body armour. Its hardware design is depicted in Fig. 3. As discussed, phototransistor placed in the module detects an incident of laser beam from the initiator unit. Upon receiving encrypted laser beam, responder unit performs decryption with the preset key to identify whether friend or foe. If the decryption is successful, soldier is identified as friend and responder unit passes this message to initiator unit through RF message signal.

Fig 3: Responder Unit Hardware Design.

Pulse Detector is used to detect soldier is dead or alive, in order to avoid misuse of the proposed system by enemy forces to masquerade as friendly force. To reactivate the system it demands to enter password for theft security of the system. 5. Implementation and Work Flow The proposed system can be implemented in two steps, first designing initiator unit, wherein performing encryption with preset key and generating laser encrypted beam. Simultaneously detecting RF signal to identify friendly soldier. Secondly, Implementation of responder unit that receives encrypted laser beam and initiates decryption with preset key. The flow diagram of the entire implementation is illustrated in Fig. 4. Pulse Detector is being used in this system in order to ensure the deactivation of the system soon after separation from the body of soldier. Each pulse from the pulse detector generates an interrupt, which is used to reset software based watchdog timer. If the timer count exceeds predetermined timing intervals, it indicates that the system has been detached from the soldier body and system will be deactivated. To start the system again it is required to enter valid password. This feature helps to ensure that the system cannot be recovered from dead soldier and used by enemy forces to masquerade as friendly force. The flow diagram for the said feature is given in Fig. 4.

[488]

Padma and Sathisha/COMMUNE – 2015

The use of infra-red lights and thermal tape that are invisible to observers without night-goggles, or fibres and dyes that reflect only specific wavelengths are still in their infancy, but may prove to be key identifiers for friendly infantry units at night. The proposed system is working effectively at night as well.

Fig 4: (a) Flow Diagram of entire System, (b) Flow Diagram of Pulse Detector.

6. Validation and Experimental Results Final circuit is implemented on bread board as per the requirements discussed in section IV. Initiator unit consists of RF receiver; to receive message from friendly soldier, a laser beam transmitter; to transmit encrypted laser beam and a central processing unit, PIC microcontroller. Phototransistors placed on soldier’s armor detects incident of laser beam. After successful decryption of the received signal, responder unit generates RF message signal to be transmitted to initiator unit to acknowledge that targeted soldier belongs to the same troop. Pulse detector is also implemented in this module to provide theft security; avoid false use of the system. The initiator unit placed on soldier’s rifle transmits laser encrypted beam and it periodically checks for the friendly fire signal from the responder unit. Responder unit is responsible for registering laser beam that is sensed by phototransistors placed on body of soldier. If the sender is friend, then it sends pass signal to the initiator unit. In case soldier is dead, the proposed system cannot be recovered by the enemy forces. This is successfully implemented with the help of pulse detector. Fig. 5 gives the snapshot of the responder module incase of soldier is dead. It wipes out the program.

Fig 5: Responder Unit Displaying Soldier Is Dead.

To reactivate the responder unit, it demands authorized password. Fig 6 gives the snapshot of the responder unit’s action. Hence, anti – theft security is ensured for the proposed system effectively

[489]

Padma and Sathisha/COMMUNE – 2015

Fig 6: Responder Unit Asking Authorized password.

The system works well and it’s responsive. It takes 1s to leave the area to be registered. This is an acceptable level of lag in detection, since it amounts to extending the no – fire time and ensures no friendly fire crisis. A range of 10m is being obtained for the designed system. By using sophisticated amplification stage for the phototransistor outputs and use of decoupling capacitor to remove DC bias, it can be viewed to obtain much higher sensitivity and extend the range of the system. 7. Conclusion In this paper, we propose an adoption of embedded technology to overcome the crisis of friendly fire happens in the battlefield. The proposed architecture is well designed and implemented using PIC microcontroller and tested for its operating conditions. Within the specified range of 10m, the system works as specified and produces no false signals. However, environments with high lighting contrast may exacerbate the problem. In addition, there is a possibility of radio interference of other group’s using the same radio modules and operate on the same channel. This system is designed to be usable by all personnel in the field, since it requires no special instruments. It can be mounted on the existing equipments. In the future, we intend to refine this proposed architecture to get higher range of the system. Also by implementing more number of phototransistors in the responder unit; can cover larger part of the body and increases the detection rate of laser beam. Infra – red laser can also be utilized to avoid detection in the visible spectrum. 8. Acknowledgements The assistance of Mr. Mohammed Nayeem & Mr. Sashanka S Naik, Dept. of E&C Engg, MITE in the implementation of proposed architecture and preparation of this manuscript is gratefully acknowledged. References John B. Roes, Deepak Varshneya, “Secure covert combat identification friend-or-foe (IFF) system for the dismounted soldier” US 7308202 B2, Cubic Corporation. Daniel A. Britton, “Identification friend or foe system including short range UV shield”, The United States of America as represented by the Secretary of the Navy. John R. Wootton, Gary Waldman, Gregory L. Hobson, David Holder, “Laser vibrometer identification friend-or-foe (IFF) system”, Electronics & Space Corp. Sharef Neemat & Michael Inggs, “Design and Implementation of a Digital Real-Time Target Emulator for Secondary Surveillance Radar / Identification Friend or Foe” University of Cape Town, IEEE A&E SYSTEMS MAGAZINE, JUNE 2012 US Congress, Office of Technology Assessment, “Who Goes There: Friend or Foe”, (Washington, DC: US Government Printing Office, June 1993). LCDR William Ayers III, “Fratricide: Can It Be Stopped”, Marine Corps University Command and Staff College, 1993. Fratricide: Avoiding the Silver Bullet, Marine Corps University Command and Staff College Report, 1995

[490]

2015 International Conference on Advances in

Computers, Communication and Electronic Engineering 16 -18 March, 2015

PG Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, India

Improving the Network Capacity and Coverage of 4G-LTE-Advanced Systems using Relaying Technology Javaid A. Sheikh, Mehboob ul Amin*, Shabir A. Parah, G. M. Bhat Department of Electronics and Instrumentation technology, University of Kashmir, Srinagar

Abstract In this paper we have proposed a new technique in Long-Term-Evolution-Advanced (LTE-A) by incorporating the relaying technology to improve the coverage and capacity of the 4G cellular networks, especially at the cell borders where users experience low Signal-to- interference- noise- ratio(SINR) and low throughput in terms of capacity. The main focus is laid upon the selection of the location of Relay node (RN) in the network, so that the access link attains much improved performance. The main aim of the proposed technique is to attain maximum throughput capacity and improved SINR.

© 2015 Published by University of Kashmir, Srinagar. Selection and/or peer-review under responsibility of Department of Electronics and Instrumentation Technology, University of Kashmir, Srinagar. Keywords: Long-Term-Evolution-Advanced ; Generation Partnership Projects ; 4th Generation ; Signal-to-Interference-Noise –Ratio ; Relay Mode

1.

Introduction

LTE-A has been incorporated in the market to meet the requirements set by ITU-R for 4th generation wireless communication. These include a peak bit rate of 1Gbit/sec on the downlink and 500Mbit/sec on the uplink (Dahlman et al; 2011). LTE uses pre-coded version of OFDM in the downlink transmission and SC-FDMA in uplink transmission because SC-FDMA offers lesser peak-to average power ratio (PAPR), that is regarded as one of the bottleneck problem in multi-carrier communication, because it can degrade the system performance due to various non-linearites like spectral spreading and intermodulation noise that exist because of inherit non-linear nature of power amplifiers. The SC-FDMA involves both FFT and IFFT operations that are applied on the transmitter as well as receiver side. The SCFDMA multiple access scheme employed in LTE involves the use of frequency band in which the overall bandwidth ranges from 1.4 MHz to 20 MHz with the sub-carrier spacing of 15KHz. Transmitting 12 consecutive sub-carriers (180KHz) with 0.5ms gives us one physical Resource Block (RB) (D.H.te Hennepe j et al; 2012). The time period of one RB is called a slot and two slots form a frame. LTE uses certain scheduling algorithms to assign these RBs to the users. The three main scheduling algorithms available in the market are: Fair Fixed Assignment (FFA), Fair Work Conserving (FWC) and Maximum Added value (MAV). Out of these three algorithms FWC has better performance over other because there is the efficient utilization of the RBs. In FWC scheduler algorithms resources are equally distributed within each sub-frame. The through-put is thus maximized because resources are shared as fair as possible within each sub-frame. The main drawback associated with LTE-A networks is severe radio propagation losses at 2GHz frequency, especially at cell edges as a result of low SINR resulting in small coverage area and capacity of cell borders. In order to ensure the capacity enhancement and improvement in SINR of the cellular networks, the implementation of Relaying technology is being investigated by the team of 3GPP.A Relay Station (RS) is actually the intermediate station that can be incorporated between eNode B and Mobile Station (MS). The RS can operate either in half duplex mode or full duplex mode. There are several ways of incorporating the relay technology in the LTE networks. One of the simplest ways is Amplify and Forward. These types of relays amplify the signal along with noise, so they serve best in noise limited environment. Another type of relaying is Decode and Forward (DF) relays. These types of relays use certain error detecting algorithms to detect the signal and thus forward only the desired signal and thus can give the improved performance over conventional relaying. Another important reason of implementing the relaying technology * Corresponding Author, Tel. +919596127081 Email address: [email protected] ISBN: 978-93-82288-63-3

Sheikh et al/COMMUNE – 2015

is that RS helps in minimizing the path-loss, as the user can catch the signal directly from RS instead of BS, thus reducing the distance between MS and BS, resulting in lesser path-losses. In this paper we have laid focus on the selection for the best location for Relay Node so as to maximize throughput capacity of the cell and to attain much significant improvement in SINR, so that the overall performance of LTE-A system improves. 2. Related Work The deployment of relays in LTE networks have main objectives such as maximizing system throughput capacity, increasing Signal to interference noise ratio (SINR) especially at cell edges, enhancing the overall system performance ,attenuating the path-loss. Attaining these objectives eventually leads to improvement in system capacity and coverage area. However there are many challenges associated for the deployment of relays in LTE networks e.g. CCI mitigation, inter cell interference, wastage of some resource blocks. The following literature contains diverse techniques to attain these objectives and resolve these challenges. I. (Xiaoxiazhang et al; 2014) proposed “Joint Sub-carrier and Power Allocation for Cooperative Communications in LTE-Advanced Networks”. In this paper each user is allocated the sub-carrier with best channel quality for maximising the throughput and power is distributed in a water filling manner to improve the downlink transmission efficiency in LTE-Advanced relay systems. Decode and Forward Relay strategy is use for the joint allocation of resources. Numerical results show that proposed scheme improves the fairness as well as overall throughput of system. II. (Mattia Minelli et al; 2014) proposed “Optimal Relay placement in cellular Networks”. In this paper the authors focus on the placement of the relay nodes in the cellular networks in order to maximize the cell capacity and to obtain the SINR values that are very close to the specifications set by 3GPP-LTE-A.The impact of backhaul link (BS-RS) is also analysed for inband relays. It can be seen that when out of band relays are placed on the cell edges, the overall cell capacity increases. III. (Omer Bulacki et al; 2013) proposed “Performance of Coarse Relay Site Planning in composite Fading/Shadowing Environments”. In this paper the authors presented the performance of Coarse Relay Site Planning (RSP). The RSP selects a RN deployment location from the set of alternatives considering SINR on the relay link. The relay link is modelled by Nakagami-lognormal distribution while as access link is modelled by Rician-lognormal distribution. The results show that Coarse RSP can yield high system performance even in the shadowing or fading environments. IV. (Tommaso Beniero et al; 2009) proposed “Effect of Relaying on coverage in 3GPP LTE-Advanced”. In this paper Decode and Forward relays have been used to improve the performance of the LTE systems. The authors have proposed a methodology that aims at incorporating relay extension in the LTE networks in such a way so that there is an enhancement in system performance in terms of coverage. Simulation results show improvement in the throughput with the deployment of relays. V.( Jolly Parikh et-al ; 2013) have investigated “Effect of ISD on Path-loss and SINR in 4G Systems,” The paper investigates the performance of the LTE system in terms of path-loss and signal to interference and noise ratio (SINR) parameters in two different scenarios –rural and urban environments. Simulations results show the effect of inter site distance (ISD) on these parameters. The importance of deploying relay nodes for extending the coverage, reducing the path-loss and thereby improving the SINR at the cell edges, has been focused upon. 3. Proposed System Model In the proposed work we have considered the LTE-A system that has only two hops because this would reduce the system complexity and would make the system more practical. As per this approach three different types of links are possible: Direct link: The link between Base station (BS) and User equipment (MS). Backhaul link: The link between Base station (BS) and Relay station (RS). Access link: The link between Relay station (RS) and Mobile station (MS). The relay link and the access link should be placed in different time slots in order to mitigate the effects of interference. The use of relays reduces the interference of users situated near cell edges and hence enhances the capacity of cell edge users. As per the conventional model Relay node (RN) is incorporated at a predefined location and is placed usually at closest base station of cell. There are various possible sets of locations for the deployment of RN in the cell taking into consideration various parameters like shadowing effects and relay link quality. In the proposed model we assume there are N set of locations for the deployment of RN in a cell and best location is selected on the basis of SINR parameter and through-put capacity of access link. The SINR of the Access link should of the form: 𝛾𝑁,𝑗 = max⁡{𝛾𝑛,𝑗𝑖 : 𝑛 = 1,2 … … … … … 𝑘}

(1)

Where 𝛾𝑁,𝑖 is the SINR of nth location in the j-th cell.

[492]

Sheikh et al/COMMUNE – 2015

4. Mathematical Modelling

4.1

Maximizing Throughput Capacity of access link

In this section we derive the end-to-end throughput for the MS served by RS, we consider k number of users connected through k connection to RS. Let R be the total number of resource blocks available and each kth user uses a fraction of 𝑅𝑘 of total available resources R then 𝑅 = ∑𝑘𝑘=1 𝑅𝑘

(2)

The end-to end throughput for k-th link is given by 𝐺 = min{𝑅1 𝐺1 , 𝑅2 𝐺2 , − − − − 𝑅𝑘 𝐺𝑘 } /𝑅

(3)

Where k=1, 2, …………, k In the optimal case, when equal amount of information is transferred over each link so that throughput is maximized, we have 𝑅. 𝐺 𝑜𝑝𝑡 = 𝑅𝑛 𝐺𝑛 , n =1, 2…, k

(4)

𝐺 𝑜𝑝𝑡 = 𝑅𝑛 𝐺𝑛 /𝑅

(5)

Solving equations (5) and (2), we can get 𝐺𝑒2𝑒= 𝑅𝑛 𝐺𝑛 / ∑𝑘𝑘=1 𝑅𝑘 =

𝐺𝑛 / ∑𝑘𝑘=1(𝑅 𝑘 ) = 𝐺𝑛 / ∑𝑘𝑘=1(𝐺 𝑛 )

= 𝐺𝑒2𝑒 = (∑k𝑘=1

(7)

𝐺𝑘

𝑅𝑛

4.2.

(6)

1 −1 𝐺𝑘

)

(8)

Formulation of Access link SINR

The CDF of the access link SINR is given by (Bou Saleh et al; 2012) 𝐹(𝛾) = 1 − where,

𝑣=

𝛾𝑘 𝛾𝑙

=

𝑣 𝑣+𝛾

𝑒 −𝛾/𝛾𝑘

(9)

𝐸{𝑃𝑇𝑥,𝑘 /𝐿𝑘 } 𝐸{𝑃𝑇𝑥,𝑙 /𝐿𝑙 }

⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡𝛾𝑘 = 𝐸{

𝑃𝑇𝑥,𝑘

𝑃 𝑁 𝐿𝐾

} ,⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡𝛾𝑙 = 𝐸{

𝑃𝑇𝑥,𝑙

𝑃 𝑁 𝐿𝑙

}

Where 𝑃𝑇𝑥,𝑘 and⁡𝑃𝑇𝑥,𝑙 are the transmit powers of the serving and interfering RNs respectively. 𝐿𝑘 and 𝐿𝑙 are the corresponding path-losses, 𝑃𝑁 is the thermal noise power 𝛾𝑘 and 𝛾𝑙 are mean SINRs dependent on user distance and shadowing to the serving and interfering nodes and 𝑣 defines the mean SINR of access link. When this SINR of access link is plotted against Cumulative Distribution Function (CDF)⁡𝐹(𝛾) it shows much improved performance as compared to the SINR of the direct link. Furthermore relay link can be positioned at different sectors of the cell to examine at which location it attains maximum SINR. This is called optimal location of relay. 5. LTE-Advanced Performance The deployment of RNs in LTE-A improves the overall performance of the access link in terms of throughput and SINR values. The use of various error correcting codes like Turbo coding and convolutional coding further improves the performance of LTE-A networks. But the main area of interest lies in the best location for the relay node in the cell. The position of the access link is of the main importance because its performance determines the fate of cell edge users, which experience low SINR and coverage area of cell borders, where there is relatively less connectivity. The various parameters that are required for the enhancement of system performance of LTE-A networks are listed in table 1.

[493]

Sheikh et al/COMMUNE – 2015

Table 1. Parameters of Simulation Parameters Network Inter Site Distance (ISD) Bandwidth Carrier Frequency No of Sub-carriers Cell radius No of relays per cell Path-loss model Path-loss exponent, α,lfix Path-loss direct link H(l) Path-loss Access link H1(l)

Assumption Tri-sector cell/Six –sector cell 500m/1723m 10MHZ 2GHZ 1024 1000 meters 3/6 Cost Hata 231 3.53,141.6 10𝜎𝑓𝑖𝑥/10 ∗ (𝑙/2)𝛼 (linear) l is the distance between Tx and Rx 0.087*H(l)

6. Conclusion and Future Work This paper discusses the importance of the deployment of relay nodes in LTE-A networks for improving the coverage and capacity for the 4th generation wireless communication. The paper focuses on the selection for the best location for RN so that the access link achieves maximum throughput capacity per user and the considerable improvement in SINR. Sectoring can be done in a cell to improve the system capacity, since it reduces the inter-cell interference, SINR will improve. The RN can be placed in each sector to allow the capacity and coverage comparisons for the different positions of the access link and the positions with the improved performance can be selected. References Dahlman, E., Parkvall, S. and Sköld, J. 2011 4G LTE/LTE-Advanced for Mobile Broadband, Elsevier. D.H.te Hennepe, J.L van DenBerg et al; 2012 “Impact of Relay Station positioning on LTE Uplink performance at flow level” IEE Global communication Conference (GLOBECOM). Xiaoxiazhang, Xiamen (Sherman) Shen et al; 2014 “Joint Sub-carrier and Power Allocation for Cooperative Communications in LTE-Advanced Networks”. IEEE transactions on wireless communications, vol. 13, no. 2, February. Mattia Minelli Maode Ma et al; 2014 “Optimal Relay placement in cellular Networks” IEEE transactions on wireless communications, Vol. 13, No. 2. Omer Bulacki,Hamalainen et al; 2013 “Performance of Coarse Relay Site Planning in composite Fading/Shadowing Environments.” IEEE 24th International symposium on Personal Indoor and Mobile Radio Communications (PIMRC). Tommaso Beniero, Simone Redana et al; 2009 “Effect of Relaying on coverage in 3GPP LTE-Advanced”. IEEE 69th Vehicular Technology Conference. Jolly Parikh et-al; 2013 “Effect of ISD on Path-loss and SINR in 4G Systems”, Proceedings of IEEE International Conference on Multimedia Signal Processing and Communication Technologies - IMPACT. Bou Saleh et al 2012 “Analysis of Site Planning on the performance of Relay Deployments”, IEEE Transactions on Vehicular Technology Vol. 61, No 7. A.Bou saleh, S.Redana et al 2009; “On the Coverage extension and Capacity Enhancement of Inband Relay Deployments in LTE-Advanced”, IEEE VTC 2009-Fall, Anchorage, USA. Report ITU-R M.2134, “Guidelines for evaluation of radio interface technologies for IMT-Advanced.”[Online]. Available: http://www.itu.int/dms pub/itu-r/opb/rep/R-REP-M.2135-1-2009-PDF-E.pdf 3GPP, TR 36.806 Relay architectures for E-UTRA (LTE-Advanced), Tech. Rep., April 2010 [Online]. Available: http://www.3gpp.org/ftp/specs/html-info/36806.htm.

[494]

2015 Author Index Author Name

Page No. (s)

A. Ahmed A.H. Moon Aadil Amin Kak Aadil Masood Wani Abid Baba Abid Hussain Wani Adil H. Khan Adil Rashid Afshan Amin Khan Afzan Hussain Hamadani Aijaz Aiman Ajaz Hussain Mir Ajit Kumar Amit Mishra Anupama Chadha Ashaq Hussain Dar Atul Kumar Babu Lal Sharma Baljeet Singh Sinwar Benish Ali Bhat Burhan Khurshid C. J. Panchal Chetan Chouhan Deepali Deepika Jamwal Deepti Sharma Devanand Devi Dass Dharam Veer Sharma Dharmendar Singh Ekta Khimani F. A. Khanday Faizan Kitab Farah Farah Fayaz Quraishi Farhana Ahad Farooq Aadil Faroze Ahmad Fasel Qadir Fayiqa Naqshbandi Fozia G. Mohiuddin Bhat

Gaurav Bharadwaj Girish Parmar

200 309 110, 115 441 97 425 119 103 465 415 161 161 260, 475 362, 380 255 480 92 34 236 236 441 82 366 280 380 214, 250 195 429 78, 394 155, 405 433 185 45, 66, 143, 299, 331, 356 149 161 420 270 71 226 313 97 97 97, 161, 270, 321, 327, 339, 352, 415, 491 385 236, 292

Author Name Gurpreet Singh Lehal Harish Patidar Harjeet Kaur Huma Shafiq I. N. Beigh Indu Chhabra Jagannatham V. V Jahangir A. Akhoon Jai Preet Kour Wazir Javaid A. Sheikh Javeed Reshi Javid Ahmad Rather Jayesh C. Prajapti Joytirmoy Pathak K. A. Khan Kaisar Ahmad Kamal Deep Garg Kapil Dev Goyal Khalid Sultan Kshitij Pathak, Liyaqat Nazir M. A. Peer M. Asger Ghazi M. Ikram M. L. Singh M. M. Sharma M. Mustafa M. R. Beigh M. R. Dar M. Tariq Banday M. Y. Kathjoo Majid Bashir Malik Majid Zaman Baba Malik Rabaie Mushtaq Mamta Mansoor Farooq Manzoor A. Chachoo Manzoor Ahmad Mehboob ul Amin Mir Aadil Mithilesh Kumar Mohammad Ahsan Chishti Mudasir Mohd Mudasir Wani Muheet Ahmed Butt Mursal Ayub

Page No. (s) 375, 388, 398 255 218 451 356 375 265 339 327 161, 270, 275, 321, 327, 339, 352, 491 143 52 185 465 131, 313 200 362 34 305 280 165 131 29, 336, 456, 460 305 223 385 245 245 299 52, 61, 71, 143, 149, 286, 344 331 29, 336 456 415 472 110, 115 180, 420 87 491 460 236 40 370 87 456 265

2015 International Conference on Computers, Communication and Electronic Engineering, 16-18 March, 2015

495

2015 Author Name Musavir Ahmed Muzamil Ahmad N. Padha N.A. Kant Naazira Badar Nadiya Mehraj Naheeda Reshi Navdeep Lata Nazir A. Loan Nikhil Chauhan Nusrat Parveen Padma Prasad Prashant Bansod Preeti R. Sachdeva Rajandeep Singh Rajeev Gupta Rajesh Bhatt Rajshekhar Rakesh Prasher Rakesh Vaid Rana Hashmy Rashid Ali Renu Reyaz Ahmad Mathangi Richa Gupta Rockey Gupta Rohit Agnihotri Rohit Sachdeva Roohie Naaz Mir Roshani Gupta Rouf Ul Alam Bhat S. P. Ahmad S. Umira R. Qadri Saba Mushtaq Sajaad Ahmed Lone Sajad Ahmad Mir Sajjad Ahmed Sakeena Akhtar Samiah Jan Nasti Sanjeev Kumar Sharma Sanna Sathisha Shabir A. Parah Shabir Ahmad Sofi Shafiya Afzal Sheikh Shah Jahan Wani Shameem Yousf Shazia Rasool

496

Page No. (s) 103 175 240, 366 45 415 149 97 189 321 472 138 486 280 429 240, 366 223 292 292 265 78, 394 78, 195, 214, 232, 250, 394 425 29, 336 232 344 214, 250 126, 218 280 405 82, 165, 411, 445 126 97, 415 200 286 475 226 305 40 352 456 388 161 486 161, 270, 275, 321, 327, 339, 352, 491 445 52 131, 313 175 451

Author Name

Page No. (s)

Sheikh Junaid Sheikh Mohammad Shafi Sheikh Nasrullah Shifaa Basharat Shiv Kumar Shivani Raval Shivendra Pandey Shoaib Amin Banday Shraddha Arya Shrawan Kumar Simpel Rani Jindal Sofi Shabir Sukhdev Singh Sumaira Nabi Sumaya Jehangir Sumia Tariq Summera Ashraf Sunita Suresh Kumar Suriya Jabeen Susheel Sharma Syed Mohsin Syed Zaffer Iqbal T. R. Jan

309 451 175 180 255 185 465 260, 475 375 433 189 411 155 110, 115 110 115 420 385 472, 480 207 126, 218 275 138 103, 119, 207

Tariq Shafi Tasleem Arif Tawheed Jan Shah Tawseef Tazeem Zainab U. Parihar Uferah Maqbool Umar Farooq Ummer Khan Umrinderpal Singh Usha Parihar Uzma Vinay Shukla Virender Kundu Vishal Goyal Vivek Chalotra Z.A.Bangi Zaffar Kanth Zahid Ashraf Wani Zahid Hussain Zahid Maqbool Zahoor Ahmad Peer Zia Malik Zubair Manzoor Shah Zubida Habib

451 29, 336 61 161 170 240 97 445 309 398 366 161 433 472 380, 398, 460 436 66 309 170, 451 275 370 313 149 92 305

Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar

तमसो मा ज्योित

UN

IVE

RSIT

‫ﻟﻨﻮر‬ ‫ا‬

‫ﻣﻦ اﻟﻈﻠ ٰﻤﺖ‬ ‫اﻟﯽ‬

य गम

Y OF KAS

HM

IR

Department of Electronics and Instrumentation Technology University of Kashmir, Srinagar, J & K