This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2007 proceedings.

An Energy-Aware Periodical Data Gathering Protocol Using Deterministic Clustering in Wireless Sensor Networks (WSN) Mohammad Rajiullah and Shigeru Shimamoto Graduate School of Global Information and Telecommunication Studies Waseda University 1011 Okuboyama, Nishi-Tomida, Honjo-shi, Saitama 367-0035, Japan [email protected], [email protected] Abstract - One major concern of energy constrained WSN technology is to design energy efficient communication protocol. Clustering the whole network is an efficient solution to increase its life time. In this paper, we propose a clustering based communication protocol for periodical data gathering application in wireless sensor network. It uses the scheduled rotation of cluster heads, which eliminates the problem with the variability of the number of clusters generated in the dynamic, distributed and randomized protocol. Proposed protocol shows more energy efficiency by only triggering the cluster formation when cluster heads energy level falls below a certain threshold. In the simulation, our protocol shows better performance than that of LEACH (Low Energy Adaptive Clustering Hierarchy) protocol in respect of energy dissipation and number of survival nodes. Furthermore this protocol performs higher number of rounds than that of LEACH in case of the first node and 60% nodes die in the network. Keywords- wireless sensor network (WSN); clustering; energy efficient design; network lifetime

I. INTRODUCTION WSNs are ad hoc networks where each node has the recent advancement of Micro-Electro-Mechanical (MEMS) technology [1] including sensors, actuators and RF communication components. Sensors nodes are randomly dispersed over the area of interest, capable of RF communication, contain signal processing engines to manage the communication protocols, and data processing tasks. Due to all these attractive characteristics, sensor networks are used in numerous areas of application e.g. large scale environmental monitoring, battle field surveillance, location tracking, industrial plant monitoring, medical monitoring and security management [2, 3, 4]. The goal is to enable the scattering of thousands of these nodes in areas that are difficult to access by using conventional method [2]. As sensor nodes are typically battery-powered, replacing and recharging batteries are often not possible [5]. So reducing energy consumption is an important design consideration for sensor networks. For these reasons, many end-to-end communications protocol that have been proposed for ad hoc networks in recent years are not suitable for WSNs.

Alternative approaches are essential. Also the sensed data collected by the sensor nodes are very time sensitive and should be delivered in timely manner [7]. Sensor network contains too much data for a base station to process [6]. Since all the sensor nodes are deployed in the dense manner, there are possibilities of significant amount of redundant data; so in network processing through data aggregation can eliminate data redundancy, reduce communication overload and thus save energy [6]. Traditional network performs wide variety of tasks but sensor network are designed for specific application. The networking protocol in WSN may vary according to the specific nature of the application [9]. A common sensor networking application is gathering sensed data at a distant base station (BS) for further analysis [6]. Here we consider periodical data collection from a sensor field to a remote BS in energy efficient manner. In this paper, we propose a deterministic clustering based periodical data gathering protocol for WSN. Our proposed protocol uses the non randomness in the cluster head selection process and reduces the problem with the inconsistent number of clusters selection in randomized protocol like LEACH [6, 7] (we will describe it more in section II). Similar to the algorithm described in LEACH, here also, cluster head collects sensed data from other members in the cluster and after applying aggregation over collected data; it sends the fused data to BS in single hop, as depicted in Fig.1. Cluster heads are rotated among the sensor nodes in system lifetime and every node gets their schedule of becoming cluster head at the beginning of the protocol. Moreover, before the beginning of data collection and transmission to BS, cluster head ensures energy sufficiency and releases its control otherwise to other sensor nodes. So, the cluster head rotation is based on the remaining energy level of the sensor nodes which simplifies the protocol implementation and decreases the overhead by reducing the unnecessary cluster head rotation. Remaining of the paper is organized as follows: Section II gives an overview of the related protocol. In section III we discuss network and radio model used in the proposed protocol. Detail description of the proposed protocol is given in the section IV. In section V we describe the simulation and

1525-3511/07/$25.00 ©2007 IEEE

3016

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2007 proceedings.

comparative analysis. Finally we give some concluding remarks and some indications of future work in section VI. II. RELATED WORK Here as we are networking with a lot of low power nodes together, conventional technique like direct transmission must be avoided. In this case the distant nodes from the BS will loss more energy than the node closer to the BS [6]. On the other hand multi hop transmission like Minimum Transmission Energy (MTE) routing causes equally undesirable effect. In this case nodes closest to the BS will incur more energy loss as these nodes are engaged for the routing of large data message to the BS [6, 7]. Different communication protocol has been proposed to solve such kind of problem. Since periodic data collection application perquisites energy awareness and time criticality, hierarchical routing protocols perform better than other solutions [14]. Cluster based communication protocol is an effective use of hierarchical routing. A cluster based network can be partitioned in to disjoint clusters. Then in each cluster, one particular node is assigned as the local head and other member nodes of the cluster work as the followers of that cluster head (CH); thus implementing a virtual backbone of the network. Then the CH collects all the sensed data from the follower and sends the same to the BS. Different clustering techniques have already been proposed. Among them one class of protocol uses node identity to choose CH. In Linked Cluster algorithm (LCA) [10], one node can elect itself as the CH if it has the highest identity among all nodes within its one hope or among all nodes within one hope of one of its neighbors. LCA2 [11] elects the node having the lowest identity among all the nodes, those are neither CH nor are within the range of one hope of already chosen CH. It is an improvement over LCA by generating small number of CH. The Weighted Clustering algorithm elects a node as a CH based on their weights. The weight of an individual node signifies the remaining battery energy, degree, mobility, and distances to the neighbor or their combination. Distributed clustering algorithm (DCA) uses application based weights of a node to select it as the CH among its neighbor [12]. All of these algorithms have a complexity of O(n) which makes them suitable only for a small WSN [13].

Cluster Cluster

III. NETWORK AND RADIO MODEL To achieve the proposed protocol, we consider a sensor network model similar to those used in [6, 18] with the following reasonable assumptions:

Base Station Cluster Head

LEACH [6, 7] is a probability based clustering protocol for periodical data gathering applications in WSN. In LEACH the whole operation is divided into rounds. Each round consists of a set up phase followed by steady state phase. During the set up phase, nodes decide on their own to be a CH or not by evaluating a certain probability without negotiating with other nodes and then broadcast an advertise message to the rest of the nodes. The non CH node examines the received advertise signal strength and decide of which CH node it will be cluster member for that round. Any node can be selected as the CH once in a 1/ P rounds, where P is the desired percentage of the CH in any round. In steady state phase CH aggregates data from the cluster members and sends the composite data to the base station by single hope communication. Since in each round a new set of CHs are chosen randomly, LEACH evenly distributes the energy consumption of the CHs. Among the class of distributed, dynamic and randomized protocol, LEACH has the highest dominance due to its effectiveness in providing load balancing, scalability and energy efficient solution [16]. However it has some limitation. Authors at [15, 16] have shown that due to the random CH selecting strategy, the number of CH resulted by LEACH and other protocol in its class is not guaranteed to be equal to the expected optimal value and the probability that there is only one CH or there is no CH is high when the desired value of CH is small. So when no CH is elected, clustering structure is broken; every node must communicate with the BS directly [17]. Moreover, when few numbers, for example only one CH is elected, its energy will be drained rapidly. Therefore, the variability in the number of CHs adversely affects the energy efficiency, fairness among nodes and the system life time. Also it uses cluster formation in every round which increases the overhead of the protocol. Lastly in the CH selection procedure any node select itself as the CH does not ensure that it’s residual energy sustain for the whole round. PEGASIS (Power-Efficient Gathering in Sensor Information System) [18] is a chain based near optimal routing protocol. Here, in each round instead of forming clusters, each node communicates only with the closest neighbor to form a chain. Though it reduces the expensive cluster formation cost at every round, this causes an extensive delay of data transmission for a distant node in the single chain. Also this delay is unaffordable when the network size increases. Threshold sensitive Energy Efficient sensor network protocol (TEEN) [19] provides responses in the sudden change in the network and does not suit perfectly with the periodical data collection application.

Cluster

Cluster member node

• • •

Figure 1. Cluster based system model of the proposed protocol

The fixed BS is located far from the sensor field; All sensor nodes are homogeneous, energy constrained and started with uniform energy; The nodes are equipped with the power control abilities to vary their transmitter power;

3017

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2007 proceedings.

• • •

Each sensor also possesses the ability to transmit to any other node or directly to BS; Every node is provided with an identification (id) number and there is no mobility; Sensors are sensing the environment at a fixed rate and thus always have data to send to the BS.

In all the analysis, we use the same radio model used in [7]. We refer the readers to [7] for more details. The total transmission cost for sending a k bit message to a distance d meter is given by the (1) 2 k × E elec + k × e fs × d , d < d 0  E tx (k , d ) =  4 k × E elec + k × e mp × d , d ≥ d 0

E _ residual > En _ diff × E _ dissipate

(1)

Where, Eelec presents energy consumption to run the transmitter or receiver circuitry and the efs and emp stands for the radio energy dissipation, which is consumed in the transmitter amplifier. For the radio transmission, if the distance is greater than threshold d0, multi path (mp) fading channel model is used; otherwise, free space (fs) channel model is used [20]. And to receive this message, radio expends E rx (k ) = k × E elec

After the primary cluster setup phase the proposed protocol is divided into rounds. And the protocol starts with the same CH nodes and respected clusters selected in the primary cluster formation phase. As new set of CHs are not elected in the beginning of a new round, all the present CHs continue until they meet the threshold energy. Since the CH schedule follows round fashion, any node can be CH for multiple times as long as it has sufficient energy. Any node working as CH for the first time checks its energy sufficiency in the beginning of every new round according to the following equation:

(2)

Communication is symmetric so that the same energy is required to transmit a message in both direction between node A and node B at a given SNR. We assume the same set of communication energy parameters used in [7]: Eelec = 50 nJ/bit, efs = 10 pJ/bit/m2, emp = 0.0013pJ/bit/m4, d0 = 87.0m. Radio speed is set as 1 mbps. Moreover the energy cost for data fusion is considered as EDA = 5nJ/bit/message. IV. DETERMINISTIC CLUSTERING BASED PROTOCOL Proposed protocol is a clustering based protocol, where the CHs are selected deterministically. This protocol starts with the primary cluster formation phase. In this phase, several CHs are selected. At the beginning, all the nodes broadcast their id number within a finite time in a pre-specified cluster range R (we will explain more about it later). Any node receiving lower id number than its own id gives up the competition without receiving the subsequent broadcast. Otherwise it will be elected as CH at the end. Then all the elected CHs broadcast an advertise message across the network. All the nodes receive these broadcasts and depending on the received signal strength, nodes decide to which cluster they want to belong and eventually choose the nearest CH. All the nodes inform their decision back to the chosen CH. Then several clusters are formed. We call these clusters, the primary cluster of the network. CH nodes then create a schedule containing both the primary CH id and the serial information in which order the member nodes will elect themselves as CH in the network lifetime. This serial is determined according to the order, each CH receives decision message (primary cluster join message) from its cluster member. CH nodes broadcast this message to all the cluster member nodes. This schedule works in a round fashion through the lifetime of the sensor nodes in the network.

(3)

Where E_residual is the CHs residual energy and E_dissipate is the CHs total dissipated energy in the last round. So, if the En_diff value is large then the number of cluster formation is increased in the network life time and the minimum En_diff value causes the quick die of the CH nodes for longer continuation as high energy dissipating nodes. We conclude that the value of En_diff must be set as it lengthens the time at which the first node dies yet minimizes the number of cluster formation process. Once all the CHs ensure that they have sufficient energy to continue the present round as validated in (3), they use the same cluster for present round. On the other hand if any of the CH does not find sufficient energy for serving the present round, it broadcasts a cluster_head_release message with its serial number and primary CH id. So, the next scheduled node checks the serial number in the broadcasted message and if the CH id matches with the primary CH id it belongs to, it declares itself as the CH and broadcasts an advertise message to all the nodes. In this stage of the protocol, all the CH in the sensor field again performs cluster formation process even if any of them has sufficient energy for the present round. It is required for the scalability of all the clusters in the network. Here it should be mentioned that if any node wants to declare itself as the CH for the first time, E_dissipate is always a constant for that round only. Moreover, after the first time, a node is only allowed to act as CH if it has more E_residual than the E_dissipate (the dissipated energy in the last round it was CH) and it is validated at the beginning of every new round. Again if any scheduled node does not have much energy or if the node is dead then after certain timeout period, the next schedule node examines its energy for becoming CH and so on. After the cluster formation, sensor starts data collection and transmission. Although new clusters are not formed in each round, data transmission phase is performed in every round. This phase is quite similar to that of the LEACH. So, when the new clusters are formed, each CH makes one schedule for the data transmission with the number of nodes included in the respective cluster. This is a TDMA schedule describes when each of the cluster member node can transmit data to the CH. This schedule is broadcasted to all the nodes in a cluster and then nodes start data transmission. Also like LEACH, all the nodes keep their radio off while they are not transmitting data. It minimizes energy dissipation. When all the data is received, CH uses data aggregation and then sends the composite data to the BS. After a certain time, the next round begins where every CH node again determines whether it should continue the next

3018

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2007 proceedings.

(4)

According to n in [7], in the present simulation environment, we get 1< n <6 and using this n in (4), we get 23 < R <56. Then we observe the relation between R and the network lifetime. The value of R is directly related with the number of primary clusters and there sizes which consequently affects the system life time. If the primary cluster is too small then CH is rotated among those nodes in that cluster very frequently which causes quick die of those nodes. This situation is also true when the sensor nodes are not uniformly deployed. And this can be improved by using node degree instead of only node id in the primary cluster formation phase. Fig. 3 suggests that the optimal value of R lies in 40 and 41 m. Fig. 4 shows the comparison of energy consumption of the proposed protocol with LEACH where, En_diff = 45 and R = 40m. Here the proposed protocol shows significant energy saving against LEACH which results a longer system life time. Fig. 5 shows the system lifetime of the proposed protocol in the same simulation environment. And by using the proposed protocol, the network can operate higher number of rounds before the first and 60% nodes die. We repeat similar experiments with the different initial energy given to the sensor nodes. But in every case our proposed protocol outperforms LEACH. The data from the experiment are shown in Table I. In every different simulation environment, we vary the value of En_diff and R for the best performance. Our ultimate improvement of the proposed protocol against LEACH is that, it reduces energy consumption by nonrandom CH selection process and eliminating cluster formation process in every round. 1200 # rounds until the first node dies

V. COMPARATIVE ANALYSIS AND SIMULATION RESULTS In LEACH, CHs are selected without any iteration and the process consists of only two messages, CH advertise message and schedule message and for the other nodes it is only cluster join message. So the overhead is near optimal, which is 2NP+N (1-P) = NP+N, where N is the total number of nodes and P is the probability of any node to be CH. In the proposed protocol, when any CH reaches the threshold energy, it broadcasts cluster_head_release message which is followed by one CH advertise message, cluster join message from the nodes and at last TDMA schedule message from the CH as LEACH. So it is CH+2|CH|+ (N-|CH|) = CH+ |CH|+N, in the best case while only one CH broadcasts cluster_head_release message. And in the worst case it is |CH|+2|CH|+ (N-|CH|) = 2|CH|+N. So the overhead of the proposed protocol is also near optimal. But it should be mentioned that in the proposed protocol, this overhead is not repeated in every round and rather cluster formation overhead depends on the residual energy of the CH, which reduces the complexity of the protocol. In the simulation program, we mainly compare our proposed protocol with LEACH. We use the same simulation parameter values used in LEACH. We consider a network of randomly distributed 100 sensor nodes in a place of 100 m X 100 m with the BS at location (x=50m, y=175 m). We adapt the same MAC protocol in the proposed protocol as used in the LEACH. All the sensor nodes start with 1 Joule of energy. The size of the data message and the packet header size for each type of packet are set as 500 bytes and 25 bytes respectively. We run the simulation until 60% of the total nodes die. We consider that the network is unstable when 60% nodes are died. For these simulations, energy is removed whenever a node transmits, receives or performs data aggregation. We do not assumed any energy dissipation when nodes are idle or in carrier sensing operations. We first examine the impact of the value of En_diff on the network life time of the sensor nodes. According to the protocol, small value of En_diff minimizes the frequency of CH rotation but after a certain range, it shortens the time after which the first sensor node dies. Also the cluster formation frequency increases as the value of En_diff increases. Fig. 2 shows the relation between En_diff and the network life time using R=40.00 m. The optimal range of the En_diff value is about 44-46 in the given simulation scenario. In [7] authors have shown how to find the optimal number of clusters and we can use this induction to find the optimal R which controls the number of primary clusters. If n is the number of optimal cluster, R is cluster radius and A is the total area of the network, we get, n=A/πR2, where clusters are assumed to be circular. Then we can derive that

R = A / nπ

1000 800 600 400 200 10

20

30

40 50 En_diff

60

70

80

Figure 2. Impact of En_diff on the span of network life time

1200 # rounds until the first node dies

round or not as described previously. However as the cluster formation is not done in every round, TDMA schedule is not given to the cluster members in every round. So to ensure the scalability of the TDMA scheduling, CH observes whether any node misses to send data in any round. So if there is no cluster formation in the subsequent round and any node dies and misses to send data then CH only changes and rebroadcasts the TDMA schedule at the beginning of that round.

1000 800 600 400 200 0 24

30

36

42

48

54

R

Figure 3. The impact of R on the span of the network life time.

3019

Energy dissipation (J)

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2007 proceedings.

REFERENCES

10 9.6 8.8 8 7.2 6.4 5.6 4.8 4 3.2 2.4 1.6 0.8 0

proposed protocol

100

LEACH

200 300 400 Number of rounds

500

Figure 4. Energy consumption comparison between the proposed protocol and LEACH 110

Number of alive sensor nodes

100 90 80

LEACH

70 proposed protocol

60 50 40

0

300

600

900

1200

1500

1800 2000

Number of rounds

Figure 5. Network lifetime of the proposed protocol and LEACH TABLE I. Energy (J/node) 0.50

0.75 1.00

NETWORK LIFETIME USING DIFFERENT AMOUNTS OF INITIAL ENERGY GIVEN TO THE SENSOR NODES Protocol

Proposed protocol LEACH Proposed protocol LEACH Proposed protocol LEACH

# of rounds when 1st node dies 557 467 854 693 1175 970

# of rounds when 60% nodes die 915 684 1225 1019 1754 1293

VI. CONCLUSION AND FUTURE WORK In this paper, we propose a protocol for periodical data gathering in the energy constrained WSNs. Our proposed protocol overcomes the problem with the already existing randomized protocol in terms of the variability of the number of clusters generated in each round and ultimately extends the network’s lifetime. The efficiency of the proposed protocol goes further by introducing the residual energy based CH rotation which simplifies the overhead and saves node energy. Simulation result shows the significant performance improvement of the proposed protocol against LEACH in terms of energy dissipation and network lifetime. Our future work will focus on the more energy efficient CH scheduling in the primary cluster formation phase. And we believe that there are still many scopes to improve the performance of data transmission phase.

[1] I. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, “Wireless Sensor Networks: A. Survey,” Computer Networks, vol. 38, no. 4, pp. 393-422, March 2002. [2] Estrin, D. Girod, L. Pottie, G. Srivastava, M, “Instrumenting the world with wireless sensor networks,” Proc. of the International Conference on Acoustics, Speech and Signal Processing (ICASSP 2001), Salt Lake City, Utah, May 2001. [3] R. Szewczyk, J. Polastre, A. Mainwaring and D. Culler, “Lessons from a sensor network expedition,” Proc. of the 1st European Workshop on Wirless Sensor Networks (EWSN’04), Berlin, Jan. 2004, pp. 1-16. [4] A. Schmidt, and K.V. Laerhoven, “How to build smart applications?,” IEEE Personal Communications, Aug. 2001, pp. 66-71. [5] V. Mhatre, C. Rosenberg, “Homogeneous vs Heterogeneous Clustered Sensor Networks: A comparative study,” Proc. of IEEE international conference on communication, ICC, June 2004. [6] Wendi Rabinder Heinzelman, Anantha Chandrakasan, and Hari Balakrishnan, “Energy-efficient communication protocol for wireless microsensor networks,” Proc. of the Hawaii International Conference on System Sciences (HICSS), pp. 3005-3014, Maui, Hawaii, January 2000. [7] Wendi Rabinder Heinzelman, Anantha Chandrakasan, and Hari Balakrishnan, “An Application Specific Protocol Architecture for Wireless Microsensor Networks,” IEEE Trans. On Wireless Communication, vol. 1, no. 4, Oct. 2002, pp. 660-670. [8] B. Krishnamachari, Deborah Estrin, Stephen Wicker, “ The impact of data aggregation in wireless sensor networks,” International Workshop on Distributed Event-Based Systems, (DEBS '02), July 2002. [9] Jamal N. Al-karaki and Ahmed E. Kamal, “Routing Technique in Wireless Sensor Networks: A Survey, ” IEEE Wireless Communications, vol. 11, no. 6, pp. 6-28, Dec. 2004. [10] D. J. Baker and A. Ephremides, “The Architectural Organization of a Mobile Radio Network via a Distributed Algorithm,” IEEE Transactions on Communications, Vol. 29, No. 11, pp. 1694-1701, Nov. 1981. [11] A. Ephremides, J.E. Wieselthier and D. J. Baker, “A Design Concept for “Reliable Mobile Radio Networks with Frequency Hopping ,” Proc. of IEEE, vol. 75, No. 1, pp. 56-73, 1987. [12] S. Basagni, “Distributed Clustering for Ad Hoc Neteorks,” Proc. of the IEEE International Symposium on Parallel Architectures, Algorithms, and Networks (I-SPAN), Perth, Western Australia, June 1999, pp. 310315. [13] Seema Bandhyopadhyay; Coyle, E.J, “An energy efficient hierarchical algorithm for wireless sensor networks, ” Proc. of 22nd IEEE INFOCOM 2003, Vol. 3, 30th March- 3rd April 2003, pp. 1713-1723. [14] S.D. Muruganathan, D.C.F. Ma, R.I. Bhasin, and A.O. Fapojuwo, “A centralized energy- efficient routing protocol for wireless sensor networks,” IEEE Radio Communications Magazine, Mar. 2005, pp. 8 -13. [15] Ying Wang and Mudi Xiong, “Monte Carlo Simulation of LEACH Protocol for Wireless Sensor Networks,” Proc. of sixth International Conference on Parallel and Distributed Computing Applications and Technologies, PDCAT 2005, pp. 85-88. [16] Q. Wang, G. Takahara, and H. Hassanein, “Stochastic Modeling of Distributed, Dynamic, randomized Clustering Protocols in Wireless Sensor Networks,” Proc. of the 2004 International Conference on Parallel Processing Workshop (ICPPW’04), pp. 456-463. [17] Liang Zhao, Xiang Hong, Qilian Liang, “Energy-Efficient Selforganization for Wireless Sensor Networks: A fully distributed approach,” Proc. of Global Telecommunication Conference, 2004 GLOBECOM ’04, IEEE Vol. 5, 29 Nov.- 3 Dec. 2004, pp. 2728-2732. [18] S. Lindsay and C. Raghavendra, “PEGASIS: Power- Efficient Gathering in Sensor Information Systems,” Proc. of 2002 IEEE Aerospace Conference, March 2002, pp. 1-6. [19] A. Manjeshwar and D.P. Agrawal, “TEEN: A Routing Protocol for Enhanced Efficiency in Wireless Sensor Networks,” Proc. of 1st Intl. Workshop on Parallel and Distributed Computing, Apr. 2001. [20] T. Rapport, Wireless Communication: Principles & Practice. PrenticeHall, 1996.

3020