Deterministic Clustering Based Communication Scheme for Energy Constrained Wireless Sensor Network Mohammad Rajiullah, Mohammad Abu Nawar Siddique and Md. Akhtaruzzaman

Abstract— Since nodes in unattended autonomous wireless sensor network have limited energy, prolonging the network lifetime becomes the unique challenge. Clustering the whole network is an efficient solution to increase its life time where sensors communicate information only to cluster heads and then the cluster heads communicate the aggregated information to the base station may save energy. In this paper, we propose a Deterministic Clustering based Communication Scheme (DCCS) for periodical data gathering application in wireless sensor network. It uses 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. DCCS shows more energy efficiency by only triggering the cluster formation when cluster heads energy level fall below a certain threshold. In the simulation, we consider both homogeneous and heterogeneous form of network (in term of nodes initial energy level )and in both cases, DCCS shows better performance than the existing randomized scheme such as legacy LEACH (Low Energy Adaptive Clustering Hierarchy) in respect of energy dissipation, number of survival nodes and number of data transmission to the base station. DCCS can send an order of magnitude more data to the base station with minimum energy dissipation than LEACH and its variation. Furthermore, DCCS performs higher number of rounds than that of LEACH in case of the first node and 60% nodes die in the both homogeneous and heterogeneous network. Index Terms— wireless sensor network (WSN), clustering, energy efficient design, network lifetime.

W

I. INTRODUCTION

ireless sensor network is the revolutionary technology that can observe and control the physical world. Wireless sensor network is the special case of ad hoc networks where each node has the recent advancement of MicroElectro-Mechanical (MEMS) [1] technology including sensors Mohammad Rajiullah was with the Computer Science and Engineering Department, Asian University of Bangladesh, Uttara, Dhaka. He is now with the School of Engineering and Computer Science (SECS), Independent University, Bangladesh, Baridhara, Dhaka (e-mail: rajiullah@ secs.iub.edu.bd). Mohammad Abu Nawar Siddique was with the Computer Science and Engineering Department, Asian University of Bangladesh. Now he is doing his Master of Telecommunications Engineering in University of Melbourne, Australia. (e-mail: [email protected]). Md. Akhtaruzzaman is with Computer Science and Engineering Department, Asian University of Bangladesh, Uttara, Dhaka. (e-mail: [email protected]).

actuators and RF communication components. Sensor 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]–[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]. Unlike traditional networks, wireless sensor network imposes a set of new limitation for the protocols designed for this type of network. 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 protocols those have been proposed for ad hoc networks in recent years are not suitable for wireless sensor network with only incremental modifications. 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]. Highly unpredictable nature of wireless sensor network necessitates high redundancy. This is one of the positive characteristics that wireless sensor network has and must be utilized and exploited in the protocol designs. Since 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 communication scheme in wireless sensor network may vary according to the specific nature of the application [11]. A common sensor networking application is periodical data gathering where sensed data from remote sensor field are periodically collected at a distant base station (BS) for further analysis [6], here data aggregation and hierarchical routing mechanism has been the most effective due to the applications severe energy constraints and time criticality [16]. Hierarchical (clustering) mechanisms can increase the energy efficiency, network scalability and reduce data latency.

Cluster Head (CH) Cluster member

Base station

Fig. 1. Single hop clustering based system model for the proposed DCCS.

In this paper, we propose and evaluate a deterministic clustering based periodical data gathering algorithm, DCCS (Deterministic Clustering based Communication Scheme) for wireless sensor network. 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 scheme 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 its schedule of becoming cluster head at the beginning of the protocol operation. 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. We assume two kinds of network where every node starts with same energy level (i.e. homogeneous) in the first case and with different energy level (i.e. Heterogeneous) in later. In the simulation result, DCCS has been proved more energy effective than other clustering schemes regardless of their initial energy level.

II. RELATED WORK 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 local head or 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) [12], 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 [13] 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 [14]. All of these algorithms have a complexity of O(n) which makes them suitable only for a small wireless sensor network [15]. LEACH [6], [7] is a probability based clustering protocol for periodical data gathering applications in wireless sensor network. 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 in Eq. 1, without negotiating with other nodes and then broadcast an advertise message to the rest of the nodes. k ⎧ ⎪ pi (t ) = ⎨ N − k (r mod N / k ) ⎪0 ⎩

if

n∈G

(1)

otherwise

Where k is the desired number of CH, N is the total number of the nodes, G is the set of nodes that have not been CH in the last r mod N/k rounds. The non CH node examines the received signal strength of the advertise message from different CH nodes and decide of which CH node it will be cluster member for that round. In steady state phase, CH aggregates data from the cluster members and sends the composite data to the BS 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. However it has some limitations. Authors at [17], [18] have shown that due to the random CH selecting strategy, the number of CH resulted by LEACH and other scheme 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 [19]. 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. Subsequently, the clustering techniques in wireless sensor network have been extensively researched. We review some of the energy efficient clustering mechanism so far have been proposed. In [7], to cope with the heterogeneous energy

circumstances where each node starts with different amount of initial energy, the sensor node with higher energy has the probability to become the CH. This has been achieved by setting the probability, pi (t) of becoming a CH as a function of a node’s energy level relative to the aggregated energy remaining in the network: p i (t ) =

E i (t ) k E total (t )

(2)

Where, Ei(t) is the current energy of the node i in the network and in Eq. 3 N

E total = ∑ E i (t )

(3)

i =1

This energy aware LEACH (we call it EA_LEACH) is more energy efficient in the heterogeneous scenario. But to calculate the probability in Eq. 2, each node must have an estimation of the total network energy. This requires a routing protocol that allows each node to determine the total energy, whereas the probability in Eq. 1 enables each node to make completely autonomous decision. PEGASIS (Power-Efficient Gathering in Sensor information System) [20] 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) [21] provides responses in the sudden change in the network and does not suit perfectly with the periodical data collection application.

III. NETWORK AND RADIO ENERGY DISSIPATION MODEL To achieve DCCS, we consider a sensor network model similar to those used in [6], [7] with the following reasonable assumptions: • The fixed BS is located far from the sensor field; • All sensor nodes and BS are immobile after deployment; • All sensor nodes are homogeneous (i.e. they have the same capacity) and energy constrained; • The nodes are equipped with the power control abilities to vary their transmitter power; • Each sensor node also possesses the ability to transmit to any other node or directly to BS; • Nodes located closed to each other have correlated data; • All sensor nodes are location-unaware; • Every node is provided with an identification number.

These assumptions are quite reasonable due to the technological advancement in the integration of low power communication, sensing, energy storage, and computing in a tiny sensor node. In all the analysis, we use the same radio model for energy calculation used in [7]. We refer the readers to [7] for more details. The total transmission cost for sending an l bit message to a distance d meter is given by the Eq. 4 ⎧⎪lE elec + lε fs d 2 , d < d 0 E tx (l , d ) = ⎨ 4 ⎪⎩lE elec + lε mp d , d ≥ d 0

(4)

Where, Eelec presents energy consumption to run the transmitter or receiver circuitry and the εfs and εmp 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 [22]. And to receive this message, radio expends E rx (l , d ) = lE elec

(5)

Communication channel 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. Moreover, the energy cost for data aggregation is considered as EDA. IV. DCCS: DETERMINISTIC CLUSTERING BASED COMMUNICATION SCHEME DCCS is a clustering based protocol, where the CHs are selected deterministically. DCCS starts with the primary cluster formation phase. In this phase, several clusters are formed. Then the protocol operations are divided into rounds. Where firstly, in each round new set of CHs are selected based on the energy level of present CHs. Secondly, there is data collection and transmission phase that includes the node level data collection, data aggregation and the final transmission of information to BS. Each phase of the scheme has been described into details in the followings. A. Primary Cluster Formation Phase In this phase, several CHs are selected and then the network is partitioned into number of clusters. At the beginning, all the nodes broadcast their identification number within a finite time in a pre-specified cluster range R (we will explain more about it in section V). Any node does not receive any identification number lower than its own from this R, elects itself as a CH. On the contrary, any node receiving lower identification number than its own does not make any decision to elect itself as a CH or not until all neighboring nodes in R with lower identification number than its own have already decided. So, at the end there are number of CHs selected within their own R. Then all the elected CHs broadcast an

10 3

1

8

6 9 11

5

7 2 4

Cluster Head (CH) Cluster member

,

Fig. 2. Scheduling by CH in a single primary cluster (color blue in the picture) gives a serial number to all the cluster member (Number 1, 2, 3 etc. along with nodes stand for their own serial number).

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 message (primary cluster join message) back to the chosen CH. Then several clusters are formed. We call these clusters, the primary cluster of the network. After the cluster formation is finished, CH nodes create a schedule containing both the primary CH id and the serial information in which order the member nodes 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. Fig. 2 describes the scheduling in a single primary cluster. So in the protocol lifetime all the cluster members of the respective primary cluster follow their own serial number in the schedule message to take the control as CH. According to the scheme, in Fig. 2 when the node with serial no 2 finishes its service time as a CH than the node with serial no 3 can elect itself as the CH and so on. This schedule works in a round fashion. B. CH Selection Phase After the primary cluster setup phase, DCCS is divided into rounds. Here one round or round of activity consists of both cluster formation (or not based on the residual energy level of the present set of CHs) and the data collection. DCCS starts its first round 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 CHs schedule follows round fashion, any node can be CH for multiple times as long as it has sufficient energy. Every working CH checks its energy sufficiency in the beginning of every new round according to the following: E _ residual > En _ diff × E _ dissipate

(6)

Where E_residual is the CH’s residual energy and E_dissipate is the CH’s total dissipated energy in the last round. So, if the En_ diff value increases then the number of cluster formation are 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 Eq. 6, 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 next 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 declare as the next CH if it has more E_residual than the E_dissipate (the dissipated energy in the last round it was CH). 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. C. Message Transmission Phase After the cluster formation, sensor node 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 their sensed 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 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.

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 DCCS, 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 when only one CH broadcasts one extra control message. And in the worst case, it is |CH| + 2 |CH| + (N − |CH|) = 2 |CH| + N. So the overhead of DCCS is also near optimal. But it should be mentioned that in the proposed scheme, 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 scheme. A. Simulation Environment We evaluate the performances of DCCS through the simulation program implemented by C++. We use the same simulation parameter values used in LEACH. The parameters of simulation are listed in Table I. We consider a network of randomly distributed 100 sensor nodes in a place of 100 m × 100 m with the BS at location (x =50 m, y =175 m). We adapt the same MAC protocol in DCCS as used in the LEACH. We assume the probability of signal collision and interference in the wireless channel is ignorable. In the homogeneous network, all the sensor nodes start with 1 Joule of energy. In the heterogeneous network, all the sensor nodes start with random amount of energy between 1 and 2 Joule. 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 in every round. 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 assume any energy dissipation when nodes are idle or in carrier sensing operations. Unless otherwise specified, every simulation result presented below is the average of 50 independent experiments where each experiment uses a different randomly generated uniform topology of wireless sensor nodes. The primary idea of DCCS and its efficiency evaluation have been done previously in [8], [9]. In this paper, we have extended the idea in details. Also we have done more experiments to evaluate different performance metrics for DCCS. Moreover, here we have shown the efficiency of DCCS in both homogeneous and heterogeneous environment in terms of initial energy level of the sensor nodes. B. Experiment on DCCS parameters We first run simulation to explain the relation between the system lifetime and different parameters in DCCS in both homogeneous and heterogeneous scenario. For these simulations only, we measure the lifetime in terms of round when the

TABLE I SIMULATION PARAMETERS Parameter Area (A) Location of BS Message Size (l) Header Size Eelec

εfs εmp

d0 Data aggregation cost, EDA

Value 100 m X 100 m (50 m, 175 m) 500 bytes 25 bytes 50 nJ/bit 10 pJ/bit/m2 0.0013 pJ/bit/m4 87 m 5 nJ/bit/message

first node dies. In the followings, we examine the impact of parameter, En_diff (used in cluster head selection phase for validating energy sufficiency for any present set of cluster head in the beginning of a new round) and R (used as clustering radius in primary cluster formation phase) on the sensor network lifetime. We first examine the impact of the value of En_diff on the network life time of the sensor nodes. According to the DCCS protocol, small value of En_diff minimizes the frequency of CH rotation but after a certain range, it shortens the lifetime of the CH in a certain cluster. Also the cluster formation frequency increases as the value of En_diff increases. Fig. 3(a) and Fig. 3(b) show the relation between En_diff and the network life time in case of both homogeneous network and heterogeneous network. We get the optimal range of the En_diff value is 67 to 68 in case of homogeneous network and 100 to 103 in heterogeneous network for the given simulation scenario. The optimal values give the highest number of rounds of operation before the first node dies and in data gathering application certain area may not be monitored once a node dies. 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 k is the number of optimal cluster, R is cluster radius and A is the total area of the network, we get, k = A/πR2, where clusters are assumed to be circular. Then we can derive that R = A kπ

(7)

According to k in [7], in the present simulation environment, we get 1 < k < 6 and using this k in Eq. 7, 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. Also too many primary clusters increases the number of long distance transmission of data to BS per round of activity results more energy consumption which ultimately decrease the system lifetime. On the other way, small number of primary cluster decreases the number of long distance transmission but increases energy burden on few CHs and consequently decreases the nodes

1800 Number of rounds until the first node dies

Number of rounds until the first node dies

1200

1100

1000

900

800

700

30

40

50

60

70

80

90

100

En_diff

1600 1400 1200 1000 Homogeneous network

800

Heterogeneous network 600 30

35

40

45

50

55

R

Fig. 4. Simulation results of the impact of R on the span of network lifetime (in terms of number of rounds when the first node dies).

(a) Number of rounds until the first node dies

1800

1600

1400

1200

1000

800

50

100

150

180

En_diff

(b) Fig. 3. Simulation results of the impact of En_diff on network lifetime (in terms of number of rounds when the first node dies), in case of (a) homogeneous network (b) heterogeneous network

lifetime. Fig. 4 suggests that the optimal value of R lies in 40.0 m in homogeneous network and in also 40.0 m for heterogeneous network when we get maximum number of rounds before the first node dies. C. Performance Comparison Finally, we make the performance comparison in both homogeneous and heterogeneous network scenario. Performance is measured by quantitative metrics of energy consumption, network lifetime and number of data transmission in case of DCCS, LEACH, using Eq. 1 and energy aware LEACH or EA_LEACH, using Eq. 2. The simulation result shows the significant performance improvement of DCCS against the randomized scheme. In all our experiment for performance comparison, in the homogeneous network we set, En_diff= 68, R=40.0 m and in the heterogeneous network, we set, En_diff=102, R=40.0 m. Fig. 5(a) and Fig. 5(b) show the comparison of energy consumption of the proposed protocol, DCCS with both LEACH (in case when all sensor nodes start with the same energy level or homogeneous condition) and EA_LEACH (in case when all nodes start with the different energy level or heterogeneous condition) over the rounds of operation. These plots clearly

demonstrate that DCCS has the much more desirable energy dissipation curve than those of its counterparts. This is because DCCS can minimize the clustering overhead by eliminating unnecessary CH rotation in every round through residual energy based CH rotation. Also as we already described in section II that the stochastic CH selection in both LEACH and EA_LEACH may lead to an undesirable number of CH production in different rounds and decrease the energy efficiency. The improvement gained through DCCS is further exemplified by the network lifetime graph in Fig. 6(a) and Fig. 6(b) for both homogeneous and heterogeneous network respectively. In both graph, we present the three types of communication schemes together for better understanding of the performance comparison of EA_LEACH and LEACH in two different types of network under consideration. These plots show the number of nodes that remain alive over the number of rounds of activity for the simulation scenario. In the homogeneous network with DCCS, all the nodes alive for 1160 rounds, while the respective numbers for LEACH and EA_LEACH are 946 and 899. Since EA_LEACH is proposed to cope with heterogeneous scenario, it performs no better (even worse) than LEACH in homogeneous network. Because of the relevant overhead to get the overall energy information of the network somewhat affects the system efficiency. Furthermore, if network lifetime is defined as the number of rounds for which 60% of the nodes remain alive, DCCS outperforms the network lifetime of both LEACH and EA_LEACH. In the heterogeneous scenario, the initial energy varies from 1J to 2J. With DCCS the first node dies in 1687 rounds, where the corresponding numbers for LEACH and EA_LEACH are 978 and 1080. Here the EA_LEACH shows improvement over LEACH. But DCCS performs the best and prolongs the network lifetime significantly over both LEACH and EA_LEACH. Next we analyze the number of data message transmitted to BS for the routing protocols under consideration in the same simulation environment with both homogeneous and heterogeneous network. The more data the BS receives, the more accurate its view of the remote environment will be. Both

100

100 LEACH DCCS Number of alive sensor node

Energy dissipation (J)

80

60

40

20

0

90 80 70 60 EA_LEACH LEACH DCCS

50

300

500

700 900 1100 Number of round

1300

40

1500

0

500

(a)

2000

100 EA_LEACH DCCS Number of alive sensor node

120 Energy dissipation (J)

1500

(a)

150

90

60

30

0

1000 Number of round

90 80 70 60 LEACH EA_LEACH DCCS

50

500

900

1300 1700 Number of round

2100

2500

(b) Fig. 5. Simulation results for the total amount of energy dissipation in system over rounds of activity, in case of (a)homogeneous network (b)heterogeneous network.

Fig. 7(a) and Fig. 7(b) show the number of data transmitted to BS over the number of rounds of activity. The plot clearly describes the improvement of DCCS in delivering significantly more data message than both LEACH and EA_LEACH in different scenarios. Our ultimate improvement of the proposed scheme against LEACH is that, it reduces energy consumption by nonrandom CH selection process and eliminating cluster formation process in every round. VI. CONCLUSION AND FUTURE WORK In this paper, we propose a fully decentralized and energy efficient communication scheme, DCCS for periodical data gathering in the energy constrained wireless sensor network. A novel clustering approach has been introduced to ensure low overhead cluster formation. DCCS overcomes the problem with the already existing randomized scheme in terms of the variability of the number of clusters generated in each round and ultimately extends the network’s lifetime. The efficiency of DCCS goes further by introducing the residual energy based CH rotation which simplifies the overhead and saves node energy. So, DCCS not only solves the problem with the

40

0

500

1000 1500 Number of round

2000

2500

(b) Fig. 6. Simulation results of the network lifetime, in case of (a) homogeneous network (b) heterogeneous network.

varying number of clusters formation in different round but also reduces the overhead of cluster election in every round. We assume both homogeneous and heterogeneous form of network in terms of energy level of the nodes. In the both network, simulation result shows the significant performance improvement of the proposed scheme against LEACH (and its variation) in terms of energy dissipation, network lifetime and number of data transmission to BS. Our future work will focus on the more energy efficient CH scheduling in the primary cluster formation phase. And we believe there are still many scopes to improve the performance of data transmission phase. Beside this, we plan to address the other issues like node mobility, time synchronization in DCCS as part of our future work.

[9]

Number of data transmission to BS

8000 7000

LEACH DCCS

6000

[10]

5000 4000

[11]

3000 2000

[12]

1000 0

[13] 300

500

700 900 1100 Number of round

1300

1500

[14]

(a)

Number of data transmission to BS

14000 12000

EA_LEACH DCCS

[15]

10000

[16]

8000 6000

[17]

4000 2000 0

[18] 500

1000 1500 Number of round

2000

2500

(b) Fig. 7. Simulation results for the number of data transmission to BS, in case of (a)homogeneous network (b) heterogeneous network.

REFERENCES [1] [2]

[3] [4] [5] [6]

[7]

[8]

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. D. Estrin, L. Girod, G. Pottie, M. Srivastava, “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. R. Szewczyk, J. Polastre, A. Mainwaring and D. Culler, “Lessons from a sensor network expedition,” Proc. of the 1st European Workshop on Wireless Sensor Networks (EWS N’04), Berlin, , pp. 1-16,Jan. 2004. A. Schmidt, and K.V. Laerhoven, “How to build smart applications?,” IEEE Personal Communications, pp. 66-71, Aug. 2001. V. Mhatre, C. Rosenberg, “Homogeneous vs. Heterogeneous Clustered Sensor Networks: A comparative study,” Proc. of IEEE international conference on communication, ICC, June 2004. W. R. Heinzelman, A. Chandrakasan, and H. 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. W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan, “An Application Specific Protocol Architecture for Wireless Microsensor Networks,” IEEE Trans. On Wireless Communication, vol. 1, no. 4, pp. 660-670, Oct. 2002. Mohammad Rajiullah and Shigeru Shimamoto, “Low Overhead Deterministic Cluster Heads Selection in Wireless Sensor Net-works,” Proc. of Asian Simulation Conference (ASC) 2006, Springer, ISBN: 4431-49021-3, pp. 450-454, Tokyo, Japan, October 30-November 1, 2006.

[19]

[20] [21] [22]

Mohammad Rajiullah and Shigeru Shimamoto, “An Energy-Aware Periodical Data Gathering Protocol Using Deterministic Clustering in Wireless Sensor Networks (WSN),” Proc. of IEEE Wireless Communications and Networking Conference (WCNC), 2007, IEEE, ISBN: 1-4244-0659- 5, pp. 3014-3018, Hong Kong, China, 11-15 March 2007. B. Krishnamachari, D. Estrin, S. Wicker, “The impact of data aggregation in wireless sensor networks,” International Workshop on Distributed Event-Based Systems, (DEBS ’02), July 2002. J. N. Al-karaki and A. E. Kamal, “Routing Technique in Wireless sensor networks: A Survey, ” IEEE Wireless Communications, vol. 11, no. 6, pp. 6-28, Dec. 2004. 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. 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. S. Basagni, “Distributed Clustering for Ad Hoc Networks,” Proc. of the IEEE International Symposium on Parallel Architectures, Algorithms, and Networks (I-SPAN), Perth, Western Australia, pp. 310-315,June 1999. S. Bandhyopadhyay, E.J. Coyle, “An energy efficient hierarchical algorithm for wireless sensor networks, ” Proc. of 22nd IEEE INFOCOM 2003, Vol. 3, pp. 1713-1723,30th March- 3rd April 2003. 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, pp. 8 -13, Mar. 2005. Y. Wang and M. Xiong, “Monte Carlo Simulation of LEACH Protocol for Wireless sensor networks,” Proc. of sixth International Conference on Parallel and Distributed Computing Applications and Technologies, pp. 85-88,PDCAT 2005. 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, 2004. L. Zhao, X. Hong, Q. Liang, “Energy-Efficient Self-organization for Wireless sensor networks: A fully distributed approach,” Proc. of Global Telecommunication Conference, 2004 GLOBECOM ’04, IEEE Vol. 5, pp. 2728-2732, 29 Nov.- 3 Dec. 2004. S. Lindsay and C. Raghavendra, “PEGASIS: Power- Efficient Gathering in Sensor Information Systems,” Proc. of 2002 IEEE Aerospace Conference, pp. 1-6, March 2002. 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. T. Rappaport, Wireless Communication: Principles and Practice. Prentice-Hall,Inc.,New Jersy 1996.

Mohammad Rajiullah completed his B.Sc. in Computer Science and Information technology from Islamic University of Technology (IUT), Gazipur, Dhaka in 2002. Then he finished his M.Sc. (research based) in Global Information and Telecommunication Studies from Waseda Uni-versity, Tokyo, Japan under JDS (Japanese Grant Aid for Human Resource Development Scholarship) program in 2007. Mr. Rajiullah is now working as a lecturer in the school of Engineering and Computer Science (SECS) of Independent University, Bangladesh. His research interest includes Mobile Adhoc Network, Mobile Cellular network, Wireless Sensor Network, Body Area Network. He has publication in Springer and major IEEE comsoc conference like WCNC

(Wireless Communication and Networking Conference). He is a member of IEEE. Mohammad Abu Nawar Siddique completed his B.Sc (hons) in Computer Science & Engineering from University of Dhaka, Bangladesh in 2005. Now he is pursuing his Master of Telecommunications Engineering in University of Melbourne, Australia which is due to be completed by the end of 2008. He worked as a Senior Lecturer in the Department of Computer Science & Engineering, Asian University of Bangladesh from March 2005 to February 2008. He has done a fair bit of research throughout his study and career. He has presented his papers in various conferences in home and abroad. He started his research during the final year of his Bachelors study in 2004 and went to Gifu, Japan in Virtual Systems & Multimedia (VSMM 2004) conference to present his papers. Afterwards he continued his research onwards. His research interests include Wireless network architecture, Distributed computing and Network traffic control in WCDMA & OFDM.

Md. Akhtaruzzaman completed his M.Sc in Computer Science from University of Dhaka, Bangladesh in 1999. He has been working as an Assistant Professor in the Department of Computer Science & Engineering, Asian University of Bangladesh, Uttara, Dhaka-1230 since April 2006. He had been working as a Lecturer in the same since March 2005. Mr. Zaman has done some researches throughout his study and career. His research interests include Artificial Intelligence, Wireless network architecture, Distributed computing and Network Traffic Control.