Sunteți pe pagina 1din 5

The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'06)

CLUSTER HEAD SELECTION USING ANALYTICAL HIERARCHY PROCESS FOR WIRELESS SENSOR NETWORKS
Yaoyao Yin1, Juwei Shi1, Yinong Li2, Ping Zhang2 Beijing University of Posts and Telecommunications Beijing, China ABSTRACT The cluster-based wireless sensor network (WSN) can enhance the whole network lifetime. In each cluster, the cluster head (CH) plays an important role in aggregating and forwarding data sensed by other common nodes. A major challenge in the WSN is the appropriate cluster head selection approach. In this paper, we propose a cen-tralized cluster head selection approach using Analytical Hierarchy Process (AHP). Three factors contributing to the network lifetime are considered and they are energy, mobility and the distance to the involved cluster centroid respectively. Simulation results demonstrate that the pro-posed approach is effective in prolonging the network lifetime. I. INTRODUCTION complicated WSN environments. In this case, the node with high energy but close to the edge of the cluster may be selected as CH. The other nodes have to spend more energy in delivering the data to CH, which in turn shortens the network lifetime. Proposals in the second class take multiple factors into account [6, 7, 8, 9]. Whereas the over-restricted assumptions, such as homogeneous nodes and nodes with a little or no mobility, reduce the feasibility of the system models to a large extent. In order to increase the feasibility of the system model, some unnecessary constrains on the assumptions should be removed and more factors should be considered. The cluster head selection can be modelled as a multiple factors decision-making process. However, the different measurements units and complex interrelation between multiple factors complicate the cluster head selection process. It is difficult to find a generic metric to simplify the multiple-factors problem. Naturally, a question arises: how to incorporate multiple factors to choose a suitable CH? In this paper, we propose a novel cluster head selection approach using AHP [10] algorithm. Three factors which directly influence the network lifetime are considered and they are energy, mobility, and the distance to the involved cluster centroid, respectively. BS performs the proposed selection approach in a centralized fashion due to its abundant power, huge storage and high processing speed. Our exhaustive simulation results exhibit the multiple criteria decision-making based AHP algorithm can effectively prolong the network lifetime especially in the case of heterogeneous nodes with different initial batteries. The rest of this paper is organized as follows. Section II is devoted to the system model. Section III elaborates the cluster head selection using AHP algorithm. Section IV describes the operation of WSN and some necessary exchanging information. Section V contains some simulation results conducted in a simple topology. Finally, Section VI presents the conclusions of this paper. II. SYSTEM MODEL Consider the single-hop cluster-based WSN with hundreds or thousands sensor nodes dispersed in a field as shown in Fig. 1. BS, an observer, is located outside the field remotely. The observed field is composed of several clusters. Each cluster has one CH which acts as a local control centre to coordinate the data transmissions. All of these components are based on the following assumptions and the radio model. A. Assumptions In our research, we remove some unnecessary assumptions in [5, 7, 8, 9] to increase the feasibility of the system model. We only keep the following key assumptions: -- BS: far away from the sensor nodes and stationary

Recently WSN has become one of the most interesting networking technologies since it can be deployed without communication infrastructures. Various and useful applications, such as detection of chemical activity in military field, healthcare monitoring, and wild life sensing etc., exploit the strengths of WSNs. In general, WSN consists of hundreds or even thousands of wireless sensor nodes. They are deployed in remote or dangerous environment, allowing users to extract information in ways that would not have been possible otherwise. Hence, it is inconvenient or even impossible to recharge nodes batteries once the deployment of nodes is finished. Therefore a major challenge in the design of WSNs is to lessen the battery consumption in order to enhance the network lifetime. Network lifetime can be defined as the time elapsed until the first node (or the last node) in the network depletes its energy [1]. One solution is to reduce the data transmission between sensor nodes and the Base Station (BS) by adopting hierarchical network architecture [2, 3]. In the hierarchical network architecture, WSN is divided into several clusters. In each cluster, one special node acts as cluster head (CH) which collects and compresses the data sent by common nodes within that cluster, and then transmits the processed data to BS. Once CH is destroyed by accident or drained for the above heavy communication load, it is no longer operational and all common nodes belonging to that cluster lose communication ability. The increasing invalid CHs become the bottleneck of the whole WSN and shorten the network lifetime. Apparently, CH plays an important role in WSN. Thus an appropriate cluster head selection approach is urgently needed. Many proposals have been proposed for the cluster head selection. These proposals can be classified into two classes according to the factors of selection. Proposals in the first class merely consider distributing the energy load to select CH [2, 4, 5]. Such single factor based case is not adequate in
1-4244-0330-8/06/$20.002006 IEEE

The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'06)

Figure 1: Cluster-based WSN architecture -- Sensor nodes: energy-constrained and location-aware -- Cluster: single-hop -- Propagation channel: symmetric. It is important to note that compared with existing models, no assumptions in our model are made about: -- homogeneity of node category -- little mobility of node -- homogeneity of node dispersion radius. B. Radio Model We use the same radio model as stated in [1] with amp = 100 pJ / bit / m 2 as the energy dissipation of the transmission amplifier, E elec = 50 nJ / bit as the energy being dissipated to run the transmitter or receiver circuitry, and E DA = 5 nJ / bit / signal as the energy for data aggregation. The energy cost of transmission for common sensor nodes is calculated as: (1) ETx ( k , d ) = E elec k + amp * k * d The energy cost of reception is: (2) E Rx ( k ) = E elec k with k as the length of the message in bits, d as the distance between transmitter and receiver node and as the path-loss exponent. The energy needed by CH is: (3) E CH = n[ E Rx ( k )] + n[ R DA ] + ETx ( c n k , d ) with c as the compression coefficient and n as the number of nodes. C. Objective To prolong the WSN lifetime, we choose one appropriate CH for every cluster. Thus the procedure of selection can be expressed as: (4) CH = ni * = arg max LT ( ni ) where ni denotes the ith alternative sensor node; LT is the lifetime of the WSN. In real WSNs, three factors influence the network lifetime directly, that is, energy, mobility and distance to the involved cluster centroid: -- Energy ( ): Remaining battery of the sensor node. The initial energy is predefined and can be set to different degrees. And the energy consumption is calculated using (1)-(3) by BS. -- Mobility ( ): Degree of senor node mobility. It can be divided into three degrees: high, medium and low. The high

mobility node results in the continual change of the cluster topology, causing frequent re-selection of CH. -- Distance to the Involved Cluster Centroid ( ): Distance of a node to the cluster centroid. It can be calculated by the GPS receiver of BS. The node with nearest distance to the cluster centroid is preferred to become CH. With the introduced three factors, (4) can be re-expressed as (5) CH = ni * = arg max LT [ ni ( , , )] The optimized cluster head selection process is a multiple factors optimization problem and can be achieved using the AHP approach, which will be introduced in the next section. III. CLUSTER HEAD SELECTION USING AHP AHP is a multiple criteria decision-making method. It decomposes a complex problem into a hierarchy of simple subproblems (or factors), synthesizes their importance to the problem, and finds the best solution. In this paper, AHP is used to deal with cluster head selection and is carried out in three steps: Step 1: Collect information and structure the cluster head selection problem as a decision hierarchy of independent factors. Step 2: Calculate the relative local weights of decision factors or alternatives of each level. Step 3: Synthesize the above results to achieve the overall weight of each alternative sensor node and choose the node with largest weight as CH. A. Structuring Hierarchy The goal of the decision choosing an appropriate CH is at the top level of the hierarchy as shown in Fig. 2. The next level consists of the decision factors which are called criteria for this goal. And at the bottom level are the alternative sensor nodes to be evaluated. B. Calculating Local Weights Local weight consists of two parts: the weight of each decision factor to the goal and the weight of each nominee to each factor. Both of them are calculated with the same procedure. Taking the former as an example, we describe the procedure as the following three steps.

Figure 2: Structuring AHP hierarchy for cluster head selection

The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'06)

1) Making Pairwise Comparison The evaluation matrices are built up through pairwise comparing each decision factor under the topmost goal. The comparison results which are implemented by asking the questions: Which is more important? How much? are presented in square matrix A as

Table 2: Consistency index


Dimension RI 1, 2 0 3 0.58 4 0.9 5 1.12 6 1.24 7 1.32 8 1.41 9 1.45

A=

a a a (6) a a a a a a where ai j denotes the ratio of the ith factor weight to the jth factor weight. The fundamental 1 to 9 scale can be used to rank the judgments as shown in Table 1.

2) Calculating Weight Vector For the matrix A, we calculate its eigenvalue equation written as AW = max W , where W is non-zero vector called eigenvector, and max is a scalar called eigenvalue. After standardizing the eigenvector W, we regard the vector element of W as the local weight of each decision factor approximately, denoted as: (7) W j T = { w , w , w } 3) Checking for Consistency element in (6) satisfies the equations ai / j a j / i = 1 and ai / k ak / j = ai / j , the matrix A is the consistency matrix. Unfortunately, the evaluation matrices are often not perfectly consistent due to peoples random judgments. These judgment errors can be detected by a consistency radio (CR), which is defined as the radio of consistency index (CI) to random index (RI). CI can be achieved by CI = (max n) /(n 1) (8) where If every

C. Calculating Global Weights From above steps, we can obtain not merely the weights of decision factors towards the topmost goal from W j , but also the weights of alternative sensor nodes towards each factor. It is assumed that there are five nodes in such cluster. All the five weight matrixes of alternatives under three factors construct a 5 3 matrix, denoted as Wn / j . i w n1 / wn1 / w n1 / w n2 / w n2 / w n2 / (10) W ni / j = w n3 / w n3 / w n3 / w w n4 / w n4 / n4 / w n5 / w n5 / w n5 / The global weight of each senor node can be achieved through multiplying the local weight by its corresponding parent. So the final weight matrix in the symbol of Wn is i calculated as (11) Wn = W n / j W j
i i

where each alternative sensor node final weight is calculated as


w ni =

w
j =1

ni / j

wj

(12)

The larger the final weight of sensor node, the more importance it towards enhancing the network lifetime. Thus, the sensor node with the largest weight is selected as CH. IV. IMPLEMENTATION To perform the proposed algorithm, it is significant to figure out the operation of WSN and some necessary exchanging information. The operation of WSN is divided into rounds. Each round is separated into two phases: setup phase and steady-state phase. A. Setup Phase In the setup phase, a certain number of clusters must be confirmed and set in advance. Then the setup phase can be separated into following steps as depicted in Fig.3: 1) BS broadcasts an inquiry request message (InquiryREQ) using a non-persistent CSMA MAC protocol [12]. This massage is a short message, consisting of the number of clusters. 2) On receiving Inquiry-REQ, each node responds with current information messages (CInform), including its basic information, e.g. nodes ID, node category, node velocity etc. This message is much longer than Inquiry-REQ. 3) According to the collected CInform messages, BS makes a decision to choose appropriate CHs using the AHP algorithm in a centralized fashion. 4) BS broadcasts a notification message (NTF). It is again a small message containing a list of CHs ID.

max = (1 / n) ( AW )i / Wi
i =1

(9)

And RI is shown in Table 2 [11]. When CI 0.1 , the judgment errors are tolerable and the weight coefficients of the global weight matrix W j is the weights of decision factor under the topmost goal. Otherwise, the pairwise comparisons should be adjusted until matrix A satisfies the consistency check. Table 1: A fundamental 1 to 9 scale
Number Rating

Verbal Judgment of Preferences Equally preferred Moderately preferred Strongly preferred Very strongly preferred Extremely preferred

1 3 5 7 9

2, 4, 6, 8 indicate the medium value of above pairwise comparison.

The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'06)

For each round, the length of every exchanging message is set in Table 3. Table 3: Length of exchanging message Message Inquiry-REQ, NTF, ADV, Join-REQ, MEM CInform, collected sensed data Length 25 bits 200 bits

Figure 3: Message exchanging during the setup phase 5) Once a node e.g. node i finds its ID in the list, node i is notified to become CH for the current round. 6) Each selected CH broadcasts an advertisement message (ADV). This message is a small message containing each CHs ID and a header that distinguishes this message from an announcement message. 7) On receiving ADVs, each common node decides to which cluster it belongs based on the received ADV signal strength and then transmits a join-request message (Join REQ) back to the chosen CH. This massage is a short message carrying the nodes ID and the CHs ID. 8) On receiving Join-REQs, each CH broadcasts a member message (MEM) including the list of its cluster members. And then its involved cluster is formed. B. Steady-State Phase The steady-state phase is divided into pieces of frames. Each common node collects data according to TDMA schedule and sends the collected data to the corresponding CH in its time slot per frame. The length of colleted data is constant and as long as CInform messages. Once CH received all the colleted data, it performs data aggregation to reduce the uncorrelated noise among the signals [4]. Thus data compressions happen during the data aggregation. The resultant data then are sent to BS via CHs. Once the selected CHs die or move to other clusters, a new round begins. V. SIMULATION A. Simulation Environment In the simulation environment, the WSN consists of 100 sensor nodes randomly deployed over a monitoring area of 50 50 meters. There exist two types of sensor nodes: high energy node (HEN) and low energy node (LEN) with initial energy of 10J and 1J respectively. The propagation radius of LEN is 5 meters while HEN is 10 meters. The number of HEN is 20 percent of the total number of sensor nodes. BS is fixed and located at the position (300,100). The mobility of nodes is classified into three degrees: 3m/s, 1m/s and 0 m/s respectively and their moving direction is randomly assigned.

For (1) and (3), the path loss exponent is set to 2 and 2.5 for intra-cluster communication and transmission to BS respectively. And the compression coefficient c in (3) is set to 5%. Message exchanging flows in each round are implemented according to Section IV. Since the limited pages of this paper, the AHP matrix is not presented here, and anyone can contact the authors via email for it as well as the source code of the whole simulation platform. B. Simulation Environment Three metrics of evaluating network lifetime of WSN are proposed in [5]. They are First Node Dies (FND), Half of the Nodes Alive (HNA) and Last Node Dies (LND). We run the operation of WSN for 500 times to obtain the network lifetime. One of these simulation results is shown in Fig. 4. We can see that the nodes with AHP approach remain alive for longer time than that of using LEACH [3]. In other words, AHP approach performs much better than LEACH in prolonging the network lifetime for all metrics. The differentiated performance lies in the fact that LEACH algorithm only tries to evenly distribute the energy load and ignores other factors (e.g. mobility and distance to the involved cluster centroid) which also influence the network lifetime. Therefore, the 80% nodes died quickly before 120 rounds and only HEN nodes stay active. Compared with Fuzzy Logic Control (FLC) approach [7, 8], AHP cluster head selection approach still exhibits better improvement in prolonging network lifetime especially for the FND metric. The average network lifetime of 500 times simulation is shown in Table 4. For LND, 57.14% and 3.57% improvements are accomplished using AHP compared with LEACH and FLC algorithm. For HNA, the improvements are 162.62% and 2.18%. And for FND, the improvements are up to 360% and 112.9%. We also simulate the case of homogenous network. The initial energy of all the nodes is set to 5J. One of the 500 simulation results is displayed in Fig. 5. We can see that our AHP approach gets the first place in HNA and LND, second in FND. That is to say, the AHP is as better as other approaches in homogenous network condition but acts the best in heterogeneous networks (the initial energies of nodes are not equal). Moreover, we simulate the algorithm sensitivity to the energy ratio of HEN to LEN. The simulation result shown in Fig. 6 illustrates that the higher initial energy ratio of HEN to LEN, the better AHP approach performs compared with other algorithms to make an appropriate cluster head selection.

The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC'06)

Table 4: Simulation result comparisons with initial energy of HEN = 10J and LEN = 1J Algorithm LEACH (p=0.05) FLC AHP Network Lifetime (Rounds) FND HNA LND 15 107 203 62 275 308 69 281 319

VI. CONCLUSION In this paper, we propose a centralized cluster head selection mechanism based on AHP approach. With the goal of prolonging the network lifetime, three factors are considered and they are energy, mobility and the distance to the cluster centroid. The trigger conditions of CH re-selection are adaptive based on the mobility and remaining energy of the nodes. We also introduce the message exchanging procedures to implement the CH selection. The simulation results demonstrate that the proposed CH selection approach can improve the network lifetime remarkably, especially for differentiated initial energy of nodes. ACKNOWLEDGMENT This work has been sponsored by Beijing Natural Science Foundation Program in 2006 (4062023). REFERENCES
[1] O. Younis and S. Fahmy, HEED: a hybrid, energy-efficient distributed clustering approach for ad hoc sensor networks, IEEE Trans. Mobile Computing, vol. 3, pp. 366379, Oct. 2004.

Figure 4: Number of nodes alive over time for one simulation with initial energy of HEN =10J and LEN = 1J

[2] W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan, Energyefficient communication protocol for wireless microsensor networks, in Proc. of the 33rd Annual Hawaii International Conference on System Sciences (HICSS), Maui, HI, pp. 3005 3014, Jan. 2000. [3] C. Ragusa, A. Liotta and G. Pavlou, An adaptive clustering approach for the management of dynamic systems, IEEE Journal on Selected Areas in Communications, pp. 2223 2235, Dec. 2005.

[4] W. B. Heinzelman, A. P. Chandrakasan, and H. Balakrishnan, An application specific protocol architecture for wireless microsensor networks, IEEE Trans on Wireless Communication. vol. 1, pp. 660670, Oct. 2002. [5] M. J. Handy, M. Haase and D. Timmermann, Low energy adaptive clustering hierarchy with deterministic cluster-head selection, in Proc. 4th International Workshop on Mobile and Wireless Communications Network, pp. 368 372, Sept. 2002. [6] M. Chatterjee, S. K. Das and D. Turgut, An on-demand weighted clustering algorithm (WCA) for ad hoc networks, IEEE GLOBECOM, vol. 3, pp. 16971701, Nov. 2000. [7] I. Gupta, D. Riordan and S. Sampalli, Cluster-head election using Fuzzy Logic for wireless sensor network, in Proc. of the 3rd Annual Communication Networks and Services Research Conference (CNSR05), pp. 255 260, May. 2005. [8] Q. Liang, Clusterhead election using for mobile ad hoc wireless network, in Proc. 14th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, (PIMRC), pp. 1623 1628, Sept. 2003. [9] P. Tillapart, T. Thumthawatworn, P. Pakdeepinit, T. Yeophantong, S. Charoenvikrom and J. Daengdej, Method for cluster heads selection in wireless sensor networks, in Proc of IEEE Aerospace Conference, vol. 6, pp. 3615 3623, Mar. 2004. [10] T. L. Saaty, Fundamentals of Decision Making and Priority Theory with the Analytic Hierarchy Process, RWS Publications, U.S.A., 2000. [11] Q. Y. Song and A. Jamalipour, A network selection mechanism for next generation networks, IEEE Int. Conf. Communication 2005 (ICC2005), vol. 2, pp. 1418 1422, May. 2005.

Figure 5: Number of nodes alive over time for one simulation with initial energy set to 5J

Figure 6: Algorithm sensitivity to the energy ratio of HEN to LEN

[12] K. Pahlavan, and A. Levesque, Wireless Information Networks. New York, Wiley, 1995.

S-ar putea să vă placă și