TDMA-based Cooperative Sensing using SDR Platform for Cognitive Radio K. M. Khairul Rashid, S. K. Syed Yusof, N. M. Abdul Latiff, N. Fisal, M. Adib Sarijari, R. A. Rashid Telematics Research Group, Fakulti Kejuruteraan Elektrik, Universiti Teknologi Malaysia Johor Bahru, Johor, Malaysia [email protected], {kamilah; muazzah ; sheila; adib_sairi; rozeha}@fke.utm.my

Abstract—The current trend and demand of pervasive wireless services require more efficient spectrum usage. With policies in Cognitive Radio (CR) technology, spectral efficiency usage will be increased by exploiting unused licensed or primary users spectrum. CR with spectrum sensing mechanism is one of the main cores in CR system. To combat the problems of shadowing, multipath fading, and hidden nodes in CR network, cooperative spectrum sensing has been proposed. Testbed prototyping for sensing task is a challenging task. The demand for cooperative sensing to solve the above mentioned problems leads to proposals of physical and MAC layer functionalities. Therefore, in this paper, we demonstrate an experimental platform with USRP board that can facilitate the development of PHY and MAC layers functionality for cooperative spectrum sensing. In this paper, TDMA-based MAC protocol for cooperative sensing mechanism is proposed. The observation of the preliminary experimental measurement shows significant performance in term of feasibility of cooperative decision mechanism in wireless condition. Keywords- Software Defined Radio; Cooperative Spectrum Sensing; Time Division Multiple Access, Cognitive Radio

I.

INTRODUCTION

Cognitive radio (CR) technology, offers an attractive value to under-utilized licensed spectrum. Most of the primary or licensed spectrum is severely low usage in both time and spatial domain according to recent studies[1]. Cognitive radio introduced by J. Mitola [2] are now being studied extensively. Off the shell Software Defined Radio (SDR) platform such as Universal Software Radio Peripheral (USRP) and GNU Radio software tool package are great proponent to CR system. It is the key enabler for implementation of required cognitive system. However, this system encounters challenging issues in developing a reliable and realizable device as stated in previous work [3]. One of the main requirements of CR system is the ability to reliably detect or sense the presence of licensed primary user (PU) transmissions. From the sensing process, spectrum holes are identified for the cognitive or secondary users (SUs) usage. Previous works on the problem of sensing for cognitive radio have suggested the necessity of SUs cooperation in order to be able to detect PU transmission at really low signal-to-noise ratios or hidden node experienced by SUs in practical situations. This cooperation is known as cooperative spectrum sensing where a network of SUs share the sensed information they gained. This provides a better

picture of the spectrum usage over the area where the SUs are located. In CR networks, medium access control (MAC) protocol plays an important role. One of the objectives in developing MAC protocol for CR is to manage the secondary users (SUs) so that the spectrum hole is shared in a fairly and efficient manner. In CR, the spectrum hole is available when the licensed or primary users (PUs) vacate its spectrum. PU will encounter harmful interference if SUs doing a transmission with the same frequency and at the same time as PU. Therefore, mechanism of dynamic and opportunistic spectrum access by the SUs is expected in CR system. The development of cooperative sensing on a real platform has been a limited discussion in literature. It is expected that the MAC protocol developed for cooperative sensing has the capability to support the dynamic spectrum access policy in CR system. However, synchronization in frequency and time is a challenging issue to be addressed. Problems in cooperative sensing synchronization lead to devastating effects to the SUs. In this paper, preliminary TDMA (Time Division Multiple Access)-based cooperative sensing is developed using SDR platform. TDMA is chosen over CSMA (Carrier Sense Multiple Access) because it is well known that TDMA provides a more efficient solution in heavier traffic solution [4]. It should be noted that the initial groundwork of TDMA-based cooperative spectrum sensing hardware developed in this paper is for proof of concept purposes. The rest of this paper is organized as follows. In Section II, system design concept is introduced. Experimental implementation and results are described in section III. Finally, this paper is concluded in section IV. II.

SYSTEM DESIGN CONCEPT

In this research, cooperative mechanism TDMA-based MAC protocol is developed for multichannel functionality. GNU Radio software package [5] incorporates with USRP (Universal Software Radio Peripheral) [6-7] are chosen as the main entities of SDR platform for the development of the CR network system. Cognitive features such as spectrum sensing [8] and spectrum decision are included which complies with the overlay opportunistic spectrum access.

The enhancement of the work of spectrum sensing mechanism is made from single sensing decision to cooperative sensing decision. Quite similar project of cooperative test bed using different approach is presented in [9] which two cooperative sensing algorithms evaluation were performed. In our design, hard decision fusion of cooperative sensing is accomplished upon receiving TDMA distributed sensing information gained by each SU for the respective channel. The elements of the cooperative mechanism will be explained in detailed in the following section. A. Time frame structure Fig. 1 shows a timing structure diagram of the TDMAbased protocol frames. As shown in Fig. 1, there are three different frames or states, which are Initial state, Wait and Discovery (WnD) state, and Active state. The descriptions of the states are as follows:

The period of initial frame, PI , depends on the node's activity in which from it starts running until it receives the first synchronization packet from the synchronizer. The frames in both WnD state and active state are further divided into N timeslots with the same size correspond to N of SU’s nodes, SUn where n = 1, 2, ..., N . Initial frame and WnD frame are only at earlier phase for node’s timeslot selection purpose. The rest of the frames are active frames. As it has been described before, SU’s timeslot selection is made within WnD frame. Meanwhile during active frame, local sensing, transmission of hard sensing information, data fusion and regional decision are performed. Note that regional decision only took place in the Nth timeslot of each SU in active frame.

Initial - By assuming the network is provided with a packet of pre-determined synchronizer which contains its unique ID, each node will listen to and triggered by synchronizer’s packet. The main purpose of synchronizer’s packet is for the start-up of synchronized timeslot. The packet is firstly broadcasted by synchronizer periodically. Upon completion of this procedure, initial time synchronization will be achieved. Once SU has been synchronized with the timeslot, it will enter the wait and discovery state. Wait and Discovery (WnD) – After the initial synchronization stage, the SU node listens and captures incoming packet which contains data of timeslot occupancy. The timeslot occupancy will be analyzed for the node to choose different timeslot after waiting for several frames to make sure that the decision of each time slot is different from others. Once the SU has chooses the timeslot, the active state is triggered. The SU will self-choose the timeslot based on the timeslot occupancy. Each timeslot will be occupied simply by one SU. When timeslot nth is used (SU receiving packet at timeslot n th) , where n = 2,3,…N; N is total number of SU, the timeslot decision will be different. The probability of choosing the same timeslot is minimized due to random selection method of unoccupied timeslot and a frame delay. The random selection will pick only in range of n = 2,3…N excluding occupied timeslot. The unique value of timeslot occupancy will be assigned to the value of a node ID. For instance, if a node chooses the second timeslot, the node ID will be two (2) and so on. After all nodes have their own timeslot and ID number, cooperative spectrum sensing will take place. Active– In this state, spectrum sensing will occur and the node can start broadcasting sensing information in a beacon packet allocated to it. The sensing information will be sent in SU’s signaling channel. Finally, the regional cooperative decision which is area covered by the cooperating SUs are made at each node. For future works, the node who wants to communicate can then use the channel decided by the mechanism.

Fig. 1. Generic timing structure scheme

Fig. 2 shows the data flow diagram of the design concept during active frame for three SU nodes where Fig. 2(a), Fig. 2(b) and Fig. 2(c) illustrate the data flow at sequential timeslots. There is only one node broadcasting in each timeslot while the other nodes are listening for the sensing information. At the end of the timeslot for each active’s frames, data fusion will take place at each node where the sensing information bits of all SU nodes will be fused. After that, the decision on the available channel to be accessed by the SU will be made. The details of active frame are discussed in the following sections.

a)

b) J bits J bits SU1

c)

SUN SU2

d)

Fig. 2. Data flow diagram of proposed design. a) First timeslot b) Second timeslot c) Nth timeslot d) Decision fusion.

B. Bandwidth of interest In active frame, sensing information that is transmitted by SUn in every timeslot consists of J bits (J bits/SUn). The j-th decision bit represents the condition of each channel, Ch j where j = 1, 2, …, J. Bandwidth that is allocated for Chj is signified by BWj. Hence, the total sensing bandwidth is, J

(1)

BWT   BW j j 1

where the unit of bandwidth is in Hz. C. Cooperative Mechanism Description. The MAC protocol used in this cooperative mechanism is inspired by eL-MAC [10] and the dynamic spectrum access mechanism in [11]. The main feature of this newly developed MAC protocol is it provides distributed-like decision fusion cooperative sensing. In this mechanism, SU can be in initial, WnD and active state as shown in Fig. 3 where the descriptions of the states have been discussed before.

the primary users, the final decision declares a primary user is present. The PD of the final decision is [12], k

PD,coop ( AND )   PD,i

(3)

i 1

where PD,k, is the probability of detection for the i-th secondary user and k number of cooperating nodes. The work in [13] states that the OR fusion mechanism can usually achieves higher PD than the alternative mentioned scheme. From the cooperative fusion decision, hidden node problems can be solved and each SU node is able to identify and select reliably the available channel to use for communication purposes. In spectrum sensing, energy detection is widely used due to it simple mechanism. Energy detection is usually based on hypothesis testing (Hypothesis H 1, is the primary user is active while hypothesis H 0, is the primary user is inactive) at time instant, tn and its random variables as follows [14]. H 0 : Y (tn )  I (tn )  N A (tn )  (4)  H 1 : Y (tn )  S (tn )  I (tn )  N A (t n ) where Y means the observation result at each SU; S implies signal from PU; I for the interference from co-existing multi-radio wireless network; meanwhile NA is Additive White Gaussian Noise (AWGN).

Fig. 3. Flow of the node's state

D. Cooperative Spectrum Sensing Decision One of the methods to improve spectrum sensing reliability is through cooperative sensing [12]. There are two types of techniques namely soft fusion decision and hard fusion decision. The first type requires all nodes to send their whole sensing data and it consumes high network overhead. Meanwhile the second type requires only 1-bit decision from a node and it is merely low network overhead. Next, the focus of discussion is on hard decision fusion.

Fig. 4 shows the cooperative decision activity during active state. Initially, each SU senses the spectrum and then broadcasts its corresponding local spectrum sensing information during its respective timeslot’s period. The decision is based on a pre-determined sensing threshold, γ which depends on the false alarm and detection probabilities as in these two equations [8]: PFA  P(Y   | H 0 )

(5)

PD  P(Y   | H1 )

(6)

The hard decision fusion that is considered in cooperative sensing are whether OR-rule or AND-rule. The probability of detection for OR-rule hard decision fusion can be written as in the following [12], k

PD,coop (OR )  1   (1  PD,i )

(2)

i 0

where PD,k, is the probability of detection for the i-th secondary user and k number of cooperating nodes. This OR-rule hard decision fusion mechanism can give protection to PU from harmful interference by the SUs. There must be at least one out of k SUs that detects the PU, then the final decision states that a primary user is present. Meanwhile, in AND fusion rule, when all the k SUs detect

Fig. 4. Cooperative decision activity in active state.

Energy detection sensing that is based on NeymanPearson criterion, is a good candidate for implementation of spectrum sensing. The equation of spectrum sensing is [13]: Y

1 N

N

 ( X [n])

2

(7)

communicate. Additionally, a drifted timeslot due to hardware delay or earlier timeslot’s problems will lead to a mess in the system. Hence, by implementing the synchronizer, the time synchronization issues are mostly overcome.

n 1

where, X[n] is a sampled received signal at the receiver; Y is the output of the energy detector which serves as the test statistic; N is the number of samples with n = 1, …, N. Each SU node uses j bits to represent the availability of active channels. Bit ‘1’ means channel is busy meanwhile bit ‘0’ denotes the channel is vacant. Fundamentally, each SU’s sensing information is shared or distributed with other SUs within the regional cooperative members. Next, each SU node makes a regional cooperative sensing decision based on cooperative fusion mechanism. III. EXPERIMENTAL IMPLEMENTATION AND RESULTS In the experimental setup, the network consists of four nodes which are positioned next to each other in a row. Full architecture of each node is shown in Fig. 5 which is composed of GNU Radio host (PC/laptop) and USRP node. Three nodes represent SUs with cooperative CR nodes and they are able to communicate with each other. The remaining one node represents PU node with controlled signal transmission activity manually. Each of SU performs spectrum sensing as well as transmits and receives data, meanwhile PU node can just transmits and it acts as complement to the cognitive network. Performance measurement is conducted to measure the reliability of the system design. The RF (radio frequency) implementation is using unlicensed band 2.4GHz.

A. Physical layer model A total bandwidth of 1MHz provided by the USRP for sensing is divided into 4 active channels and one common/beacon signaling channel. Each channel’s bandwidth with 250 kHz bandwidth is shown in Fig. 6. Meanwhile Table I shows the characteristic of each transceiver’s physical parameter. In order to allow the cooperative sensing to be conducted experimentally within the limitation of the hardware features, sensing is performed by SU at each respective active channel of interested band. Cooperative spectrum sensing is accomplished upon receiving TDMA distributed sensing information gained by each SU at the respective channel. For future works, the SU will be able to opportunistically access the spectrum hole within the four active channels.

Fig. 6. Channel fragmentation

Fig. 5. Architecture of each SU and PU platform.

The assumption made in this work is that the spectrum hole and channel state information (CSI) are in stationary state throughout an active frame. In addition, it is assumed that the exchanging of SU’s sensing information is conducted in SU’s signaling channel and the channel is always available for SU network. A synchronizer is also provided in the network to synchronize timeslot between nodes. In this preliminary work, each frame of WnD state and active state consists of three timeslots which are able to serve up to three SUs. The size of WnD’s period frame, PWnD, and active frame, PA are the same (3.3 seconds). The long period of 3.3 seconds was chosen due to the unpredictable hardware delay. Moreover, the time chosen is secure enough to make the system runs using the minimal requirement. The timing is critical for synchronization between each node’s timeslot as it is crucial for them to

TABLE I.

CHARACTERISTIC O F THE T RANSCEIVER

Parameter

SU and PU

Modulation

GMSK

Bitrate

125kb/s

Band

2.4GHz ISM

B. Synchronization between nodes Preliminary node is pre-assigned to act as a time synchronizer. When a new SU enters the network, second timeslot will be assigned to that particular node randomly. Consequently, when third node arrives, it chooses the third timeslot consecutively due to limited timeslot allocated. Using this procedure, all the timeslots will be synchronized to the synchronizer. It is crucial that the synchronization of all nodes in the cluster or network in terms of time and frequency is maintained. In time synchronization, all nodes

need to dwell in the dedicated timeslot but with diverse tasks or activities. As in conventional TDMA, the SU transmits during its own timeslot meanwhile other SU or nodes listen. In frequency synchronization, transmitter and receiver are required to communicate in the same signaling frequency for the sensing information exchange. Synchronization issues occur during communication of nodes in which they need to use the same frequency and their own timeslots. Hence, there is possibility of time delay during communication that leads to performance degradation of cooperative sensing mechanism. Consequently, in the proposed design as mentioned before, reset timer is performed to solve the synchronization problem. C. Single spectrum sensing decision Spectrum sensing using energy detection is implemented on SDR platform for proof of concept purposes. By introducing PU in a controlled manner, spectrum sensing is performed. Decision of each channel can be obtained from equation (8) and equation (9) as follows: FFTBinAvrg M  Bin  

1 M

M

 FFT

bin

(i )

that the channel is busy gives decision bit ‘1’. Here the threshold level is determined to be -30dB. The values observed show that the adjacent channel is also slightly affected in term of energy measurement but the result still proves that the spectrum decision is reliable and robust. TABLE II.

LOOK-UP TABLE FOR SENSING DECISION Bit Decision

PU transmission GMSK, 125kHz Bit[3]

Bit[2]

Bit[1]

Bit[0]

2.401624512 GHz (ch0)

0

0

0

1

2.401874512 GHz (ch1)

0

0

1

0

2.402124512 GHz (ch2)

0

1

0

0

2.402374512 GHz (ch3)

1

0

0

0

No Transmission

0

0

0

0

Decision bit Received Energy level (dB)

(8)

i 1

1, FFTBinAvrg M  Bin    FFTBinStatusM (bin )   0, FFTBinAvrg M  Bin   

(9)

Regarding the FFT(Fast Fourier Transform) bin, the results of the energy detector is analyzed per subcarrier basis. In equation (8), value of each subcarrier, i is averaged over the sample size, M. The average of i value is compared with the pre-determined threshold, γ to imply the status of spectrum availability indicated in equation (9). Table II shows the look-up table for sensing decision with center frequency for each active channel. The PU ‘s transmission coexists with any of the active channel in order to emulate the cognitive radio background. On the other hand, Fig. 8 shows the channel activity with the presence and absence of PU that is obtained from the spectrum sensing action. The left side shows sensing decision information (in bits) on four active channels which are named from the left, as ch3 (Bit[3]), ch2 (Bit[2]), ch1 (Bit[1]) and ch0 (Bit[0]). Energy level of noise floor within the range of -56dB to -53dB can be observed without the presence of PU. Meanwhile with the presence of PU, the energy level of noise changes within the range of -51dB to 38dB. Fig. 8 also shows the changes of channel from free channel to busy channel and vice versa. Free channel means that PU is inactive while busy channel indicates that PU is active. During the experimental measurement, PU transmission is at 2.401874512 GHz with bit rate of 125 kilobit/s, and using Gaussian Minimum Shift Keying (GMSK) modulation. The highlighted space shows that the presence of PU’s transmission which consequently means

Fig. 8. Spectrum Sensing Information Using Energy Detection

D. OR-rule Hard Decision Fusion For the hard decision fusion, the decision involves four channels. By employing four active channels, four bits of sensing information are broadcasted by each SU node at the sensing period of each timeslot. In this experiment, three SU nodes are used to reduce the complexity of the proposed protocol. Nevertheless, the algorithm supports any numbers of SUs in the network. The results of experimental arrangement for cooperative activity for three nodes operating in TDMA manner are depicted in Fig. 9. The process of spectrum decision is done in the third timeslot (in rectangle) of each active frame as shown in Fig. 9(a). The OR-rule hard decision fusion is implemented by each node to decide on the channel or spectrum holes availability. When the nodes operate in active state, where all synchronization and timeslot management procedures are carried out. Fig. 9(b), 9(c) and 9(d) represent the SU node activities on respective timeslot for sensing information exchange. The node transmits its node ID and local sensing decision as shown in circle’s mark. Other nodes listen to the information of the first node and transmit the same parameter as the first node during their timeslot. Based on the experiment, the observed results in Fig. 9(b) is (0,0,0,1) by node 1, in Fig. 9(c) is (0,1,0,0) by node 2 and in

Fig. 9(d) is (0,0,1,0) by node 3. After OR-rule hard decision fusion process, each node has the final decision of available spectrum which is (0,1,1,1) that corresponds to channel 3 labeled as free channel (underlined). Consequently, the performance of spectrum sensing can be significantly improved by utilizing the cooperative sensing approach where the final detection results is based on the fusion of local sensing decisions collected by multiple SUs . In such a way, sensing time can be decreased and the hidden node problem can be kept away.

ACKNOWLEDGMENT The authors wish to express their gratitude to Research Management Center (RMC), Universiti Teknologi Malaysia and Ministry of Higher Education Malaysia (MOHE) for the financial and technical support of this project. REFERENCES [1]

[2]

[3]

listen [4]

listen (a)

(b)

[5] [6] [7]

listen

listen [8]

listen

listen

[9]

(c)

(d)

Fig. 9. Experimental arrangement (a) Scenario of all nodes communicating to each other. (b) First node activity (c) Second node activity (d) Third node activity.

IV.

[10]

CONCLUSION

In this paper, implementation of energy detection cooperative sensing with distributed-alike TDMA-based for SDR platform is developed. Cooperative fusion decision contributes to the reliability of spectrum decision at each local sensing node in spatial spectrum activity when coexistence with PU is presence. The results of the works have been encouraging as the timeslots are synchronized and available channels can be identified from the OR-rule hard decision fusion. This preliminary work can be further improved and enhanced for the next CR system future works generally. In specific, this work may include spectrum mobility management for the enhancement. Beside that, timing can be improved for faster and optimized sensing time for durable in data transmission.

[11]

[12]

[13]

[14]

I. F. Akyildiz, et al., "NeXt generation/dynamic spectrum access/cognitive radio wireless networks: a survey," Comput. Netw., vol. 50, pp. 2127-2159, 2006. J. Mitola, III and G. Q. Maguire, Jr., "Cognitive radio: making software radios more personal," Personal Communications, IEEE, vol. 6, pp. 13-18, 1999. J. C. d. F. O'Sullivan, P.; Anyanwu, U.K.; DaSilva, L.A.; MacKenzie, A.B., "Multi-hop MAC implementations for affordable SDR hardware," in 2011 IEEE Symposium on Dynamic Spectrum Access Networks (DySPAN), Aachen, Germany, 2011. H. Kuo-Chun, et al., "MAC Protocol Adaptation in Cognitive Radio Networks: An Experimental Study," in Computer Communications and Networks, 2009. ICCCN 2009. Proceedings of 18th Internatonal Conference on, 2009, pp. 1-6. E. Blossom, "GNU Radio: Tools for Exploring the RF Spectrum. ," Linux Journal, 2004. M. Ettus, "USRP User’s and Developer’s Guide," Ettus Research LLC. F. A. Hamza, "The USRP under 1.5X Magnifying Lens!," 2008. M. A. S. R. A. Rashid, N. Fisal, A. Lo, S.K.S. Yusof, N. Mahalin, "Spectrum Sensing Measurement using GNU Radio and USRP Software Radio Platform," presented at the ICWMC 2011, The Seventh International Conference on Wireless and Mobile Communications, Luxembourg City, Luxembourg, 2011. Q. C. Raghavendra Rao, Aditya Kelkar, Dhaval Chaudhari. (2011, June 23, 2011 ) Cooperative Cognitive Radio Network Testbed. ICST’s global community magazine. L. A. Latiff, et al., "Implementation of enhanced lightweight Medium Access (eL-MAC) protocol for wireless sensor network," in Communications (APCC), 2010 16th Asia-Pacific Conference on, 2010, pp. 267272. R. A. Rashid, et al., "Enabling dynamic spectrum access for cognitive radio using software defined radio platform," in Wireless Technology and Applications (ISWTA), 2011 IEEE Symposium on, 2011, pp. 180-185. E. Peh and L. Ying-Chang, "Optimization for Cooperative Sensing in Cognitive Radio Networks," in Wireless Communications and Networking Conference, 2007.WCNC 2007. IEEE, 2007, pp. 27-32. A. Haniz, et al., "Spectrum sensing on emergency radio spectrum management system," in Communications and Information Technologies (ISCIT), 2010 International Symposium on, 2010, pp. 985-990. M. A. Sarijari, et al., "Energy detection sensing based on GNU radio and USRP: An analysis study," in Communications (MICC), 2009 IEEE 9th Malaysia International Conference on, 2009, pp. 338-342.