This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts

for publication in the ICC 2008 workshop proceedings.

Energy Efficient Clustering Algorithms for Wireless Sensor Networks

Dali Wei, Shaun Kaplan and H Anthony Chan
Dep. of Electrical Engineering, University of Cape Town, South Africa E-mails: dlwei@crg.ee.uct.ac.za, shaunkaplan@gmail.com, h.a.chan@ieee.org
Abstract-Energy efficiency is a major concern in Wireless Sensor Networks (WSNs). Many clustering algorithms have been proposed for such a purpose. This paper investigates the existing clustering algorithms. The algorithms have been classified and some representatives are described in each category. After analyzing the strengths and the weaknesses of each category, an important characteristic of WSNs is pointed out for further improvement of energy efficiency for WSNs.

where ET and ER are the energy consumptions of transmitting and receiving data respectively. The energy dissipated in operating the transmitter radio, transmitter amplifier and receiver radio are expressed as et, eamp and er respectively. Furthermore, d is the distance between nodes and n is the parameter of power attenuation with (2 n 4). III. ALGORITHM CLASSIFICATION Energy efficiency is the primary challenge of WSNs. Clustering schemes strive to reduce power consumption in order to prolong the network lifespan. This section classifies the existing clustering algorithms into six categories [2, 8], as shown in Table I.
TABLE I Categories of Energy-efficient Clustering Algorithms. Category Objective Aggregating data To reduce the amount of data that need to be sent to the sink Rotating the role of CH To distribute the higher burden of a CH among the nodes Equalizing cluster size To balance the burden among the clusters Locating CH in the central To reduce the total power consumption of area of the cluster intra-cluster communication Assigning the lowest needed To minimize the power consumption of each power to the nodes hop Optimizing route using To select the route with the lowest energy to different power level clusters relay the data

I.

INTRODUCTION

It is essential to improve energy efficiency for wireless sensor networks (WSNs) as the energy supply for sensor nodes is usually extremely limited. Clustering, as potentially the most energy efficient organization, has witnessed wide application in the past few years [1, 2] and numerous clustering algorithms have been proposed for energy saving. In clustering algorithms, geographically close nodes are organized into groups and each group is referred to as a cluster. In each cluster, nodes are assigned roles, such as clusterhead (CH), gateway node, or member node. A CH acts as a coordinator for its cluster and usually performs the following tasks; (1) intra-cluster transmission arrangements, (2) cluster maintenance, (3) data aggregation, and (4) data forwarding. Gateway nodes, which are non-essential for clustered WSNs, are the nodes that belong to more than one cluster and are capable of forwarding data between clusters. Gateway and CH nodes form the backbone of a clustered WSN. Member nodes are non-CH nodes without any inter-cluster links [2-4]. This paper investigates the most recently proposed energyefficient clustering algorithms for WSNs that can be accessed. Section II describes the energy model adopted by most algorithms. Section III classifies the algorithms and describes some representatives for each category. Section IV analyzes these algorithms, addresses their weaknesses, and suggests a feasible method for further improvement. The paper is concluded in Section V. II. ENERGY MODEL The energy model used in this paper is similar to that used by most existing energy-efficient clustering algorithms and can be seen in (1) [5-7]. It shows the energy consumption E in sending a units of data between nodes:

A. Aggregating Data The basic operation in a WSN is the systematic collection of sensed data to be transmitted to a data sink for processing [9]. The key challenge in such data gathering is to conserve the sensor energy to maximize sensor lifetime and thus network lifetime [10]. Data aggregation has emerged as a basic tenet in WSNs. The key idea is to aggregate data from different sensors to eliminate redundant transmissions [11]. In a clustered WSN, two typical methods to aggregate data have been proposed: one is to require a CH to aggregate data after it has collected data from all member nodes before the inter-cluster communication occurs [5, 12-14]; another is to aggregate data over each passing hop [6, 10]. Low-Energy Adaptive Clustering Hierarchy (LEACH) [5] applies local data aggregation to reduce global communication. This algorithm requires each CH to aggregate data collected from member nodes at the end of each data transmission period before sending them to the sink.

E = ET + ER = a (et + eamp d n ) + a er

(1)

Manuscript received December 20, 2007. Work conducted with funding support by the University of Cape Town, Telkom, Siemens, and the National Research Foundation.

978-1-4244-2052-0/08/$25.00 2008 IEEE

236

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2008 workshop proceedings.

Chen and Megerian [6] propose an algorithm where cluster sizes are based on correlation levels of sensor data. High levels of correlation will result in larger clusters that are likely to have multi-hop routes to the CH. Data is aggregated at each hop on its route to the CH to reduce the amount of data transmitted. B. Rotating the Role of the CH In a clustered WSN, a CH has a higher burden than that of member nodes making it drain energy much more quickly than member nodes. Rotating the CH role distributes this higher burden among the nodes, thereby preventing the CH from dying prematurely [5, 7, 12, 13, 15-19]. LEACH divides the network lifetime into rounds. A round comprises of a set up phase (cluster organization, CH role rotation) and a steady state phase (data collection, data aggregation and data forwarding) [5]. During the set up phase, each node locally determines whether it will become a CH. The CH selection is based on the residual energy of the node. Those with higher residual energy have the advantage during the CH competition [5]. Clusters are created by non-CH nodes choosing to join a CH based on the signal strength of advertisements received from CHs. It has been shown that LEACH reduces energy consumption and improves network lifetime compared to fixed CH schemes. Liu and Lin [7] propose the Cluster Head Election Protocol (CHEP) that bases CH selection on the remaining energy of member nodes. Once a cluster has been created with a nominal CH, each member node will send the value of its remaining energy to the CH, which will select the node with the highest remaining energy to be its successor. This node will be the CH for a cluster cycle. At the end of each cycle the current CH will select the node with the most remaining energy to be its replacement. Muruganathan et al. [12] propose a centralized clustering algorithm to better organize clusters. The task of dividing the network into clusters and selecting CHs is performed by the base station (the sink), which is assumed to have enough computational power and resources. The algorithm consists of setup and communication phases. During the set up phase each node sends the value of its current remaining energy to the base station. The base station will determine the nodes that have more than average remaining energy. Out of these nodes a specified number will be become CHs. The high burden of being CH is distributed by repeating this process. C. Equalizing Cluster Size Rotating the CH role balances power consumption among the nodes. Power consumption can be further balanced if the clusters have similar burdens. Many clustering algorithms therefore assign (approximately) the same number of nodes to each cluster [12, 20-24]. The Base-Station Controlled Dynamic Clustering Protocol (BCDCP) by Muruganathan et al. [12] makes use of the base station to divide the network into equally sized clusters. After the CH set has been determined as described in the previous

section, the base station will split the network into two clusters, selecting two nodes (from the CH set) with the greatest separating distance to be the CH of each cluster. Nodes will be allocated to each cluster based on distance before cluster balancing is applied to attain approximately the same number of nodes in each cluster. Each cluster is split and this process is continued until the required number of CHs is allocated. Krishnan and Starobinski [22] propose two algorithms, rapid and persistent, for message efficient clustering with local cluster size bounds. Each cluster initiator (CH) in both algorithms has a budget for cluster members. In the rapid algorithm, the CH splits the budget up between its neighboring nodes, less one for itself. The CHs neighboring nodes will then attempt to use their budgets by mimicking the CH and forwarding the remaining budget to their neighbors. This algorithm can result in smaller than preferred cluster sizes if nodes cannot allocate their full budget. The alternative algorithm, persistent, takes into account that neighboring nodes may not be able to allocate their full node budget. It thus allows nodes to pass their remaining budget back to the CH to be redistributed to better balance the cluster sizes. D. Locating the CH in the Central Area of the Cluster The total power consumption of a cluster is also related to the location of the CH in the cluster. Selecting a node in the central area of a cluster as CH [6, 20, 25-27] reduces the total power consumption of intra-cluster communication. In a multihop cluster, in which a member node may send data to the CH by a multi-hop route, the total number of hops between the member nodes and the CH is minimized, thereby saving energy for relaying data. In a single-hop cluster, in which all member nodes send data to the CH by a single-hop route, the sum of squares of the distances between the member nodes and the CH is reduced, thereby reducing the power consumption of transmitting data. Chen and Megerian [6] propose a clustering and routing scheme for aggregating data, as mentioned previously. They propose an algorithm to select CHs that are central within their cluster and suitably dispersed within the network. Each node in the cluster determines the maximum number of minimum hops (max-min hops) required to communicate with any other node in the cluster. The nodes with the lowest maxmin hops values are eligible to become CHs. This can be seen in Fig. 1. The numbers represent the number of hops (max-min hops) required to reach any other node. A cost function making use of the max-min hops and distance between CHs is used to keep suitable distances between CHs. Ghiasi, et al. [20] consider optimizing cluster formation by calculating the total square of the distances between the CH and all the nodes that the CH can support for different cluster formations. Clusters are selected based on the minimum total square of the distances between the CH and its member nodes. Energy consumed during the process when the member nodes send data to the CH will be reduced significantly according to (1) when n= 2.

237

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2008 workshop proceedings.

Number: 4 3 4 3

3 4 4

power level. Energy consumption of routing data is therefore reduced. IV. ANALYSIS OF ALGORITHMS The following algorithms described in the previous section improve the energy efficiency of WSNs: Algorithms that aggregate data reduce the data to be sent, thereby reducing the energy of routing data to the CH or to the sink. CHs have higher burdens than member nodes, therefore, rotating the role of the CH shares the burden and thus extending the useful lifetime of those clusters. Clusters with many nodes have a higher burden than clusters with fewer nodes as the CHs for those larger-sized clusters have to receive, aggregate and transmit more data. Some nodes are likely to be further away from the CH than would normally be the case and will require messages to be multi-hopped to the CH or transmitted with a higher transmission power to reach the CH. Therefore cluster sizes should be equalized. Selecting CHs in the central area of a cluster reduces the power consumption of intra-cluster communication. However many algorithms achieve this with frequent reclustering of the nodes with a ripple effect. This additional re-clustering can consume significant energy [5]. Power consumption can be reduced by assigning the lowest necessary transmission power to the nodes in networks where the nodes are aware of location information and are able to adjust their transmission power. In these networks it is possible to only consume the minimum amount of power required to transmit data over a link. Organizing the nodes into clusters with different power levels in WSNs with uneven node distribution improves energy efficiency by optimizing the different power level hops and also increases network capacity. All the proposed algorithms have improved energy efficiency for WSNs, yet these algorithms have usually overlooked an important characteristic of WSNs: directional data traffic towards the data sink. Further improvement can be achieved if this characteristic can be carefully considered [2]. A. The Impact of Directional Data Traffic on Lifetime of WSNs As the processing capability of sensor nodes is usually extremely limited, sensor nodes in a WSN need to send their data to the data sink, resulting in directional data traffic towards the sink. If the network is comprised of identical nodes, those with the higher burdens will die prematurely. There are two methods of data transmission from CHs to the data sink: single hop communication and multi-hop communication. The attributes of these methods can be clearly seen from the linear network shown in Fig. 2. We assume the initial energy of each node in Fig. 2 is 0.5 J and that the lifetime of the linear network is measured in rounds, where each round consists of each node sending 50 byte data to the sink. The node is considered dead when its

3 2 4 3 4 3

Figure 1. Nodes with maximum hop number to any other node (Adapted from [6]).

E. Assigning the Lowest Needed Power to the Nodes The power consumption of each link for the algorithm in this category is minimized by assigning to the member nodes or the CHs the lowest power needed for intra-cluster communication; and to the gateway nodes the lowest power needed for intercluster communication. The algorithm in [28] assumes that each node has the capability of adjusting the transmission power so that the lowest necessary power of each link can be assigned to each node. It first creates clusters by grouping closer nodes in the network to reduce the transmission ranges. The simulated annealing algorithm is used for this purpose. Then it assigns each node in the cluster with the lowest possible transmission power while keeping the intra-cluster connection. The pair of nodes that have the minimum distance between any two clusters are chosen as gateway nodes. The communication between the CHs is realized through these nodes. The last step is to assign the gateway nodes the lowest possible transmission power to maintain the connection of the backbone network. By assigning these minimum needed transmission powers to the nodes, the total power consumption of the route is reduced. F. Optimizing Route Using Different Power Level Clusters Most clustering algorithms assume a WSN with uniform node distribution. The sensor nodes, however, are not always uniformly dispersed in the network. Different areas in the sensor network may have different node density. The nodes in one area may access each other by using a low transmission power through multi-hops. However they may not be able to access nodes in another area using a low transmission power. The nodes may therefore be organized into clusters with different power levels. By optimizing the routes consisting of different power levels, energy can be saved. Kawadia and Kumar [29] propose the CLUSTERPOW for non-homogeneously spaced networks. There are no CHs in this protocol. Nodes are simply grouped based on energy levels. Nodes are likely to belong to multiple clusters of different transmission energy levels. Messages are routed from the source to the destination by the source selecting a path to the destination (or a hop closer) using the lowest power possible such that no further hops exceed this transmission

238

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2008 workshop proceedings.

residual energy is less than 0.01 J. Equation (1) is used to evaluate the lifetime.
60m Data Sink 1 60m 2 60m 3 60m 4 60m 5

Figure 2. One-dimensional linear wireless sensor network.

Table II shows the lifetime in rounds of the nodes in the linear network when eamp = 50 pJ/b/m2. The nodes can send data to the sink either by a single-hop route or by a multi-hop route. In case the node uses a multi-hop route, it can increase its transmission power to reach the next available node if its neighbor were to die.
TABLE II Node Lifetime in Rounds when eamp = 50 pJ/p/m2. Node Single hop Multi-hop with variable necessary tx 1 2 3 4 5 5327 1591 734 419 270 908 986 1073 1187 1396

The system in Fig. 2 has a total energy of 2.5 J as each node has 0.5J energy. Each node needs to relay a different data amount from others. This data amount is relevant to the node burden. If this 2.5J can be freely re-allocated to each node according to its burden, the node lifetime in the system as shown in Fig. 2 can be equalized. Table III shows the equalized lifetime of each node and the energy allocated to each node in the multi-hop system. The lifetime of the network when all nodes have the same initial energy is 1073 rounds when 50% dies. Yet this lifetime becomes 1551 rounds after the node lifetime is equalized. It can be seen that the lifetime of the network is extended substantially. In a clustered WSN, such energy allocation can be realized by allocating different number of nodes to different clusters according to their burdens.
TABLE III Equalized Node Lifetime in Rounds and re-allocated Energy (J) when eamp = 50 pJ/p/m2. Node Lifetime Allocated energy 1 1564 0.8545 2 1559 0.6772 3 1551 0.5000 4 1533 0.3227 5 1474 0.1456

It can be seen that in a single-hop network, the first node dies at 270 rounds whereas in a multi-hop network the first node dies at 908 rounds. If we assume that the network dies when 50% of the area of the network is out of coverage, as the performance of the network under these conditions will be greatly reduced, then the network dies at 734 rounds for singlehop scenario and 1073 rounds for multi-hop scenario. Therefore choosing to use multi-hop route when available will extend the life of the network. If the nodes lifetimes are equalized then no node will die prematurely. Sensor nodes have limited maximum transmission ranges; therefore it is unreasonable to expect all nodes to be able to reach the data sink in a single hop in many networks. Residual energy in the nodes that can no longer reach the sink because the distance between themselves and (the next node to) the sink is greater than its maximum transmission range is wasted. If each node in Fig. 2 represents a cluster then the lifetime of different clusters can be equalized by allocating energy to the cluster in accordance with its burden. This can be realized by assigning different numbers of nodes to each cluster in a clustered WSN. In a homogeneous WSN with identical nodes and uniform node distribution, equalization of the cluster lifetime will result in clusters of different sizes. B. Improvement Achieved by Equalizing Node Lifetime Using P to express the total power consumption of a node and Einitial to express the initial energy of that node, to equalize node lifetime, (2) must be observed, in which const is a constant.

Besides the network lifetime extension, the performance of delivering data to the sink is also improved. When some nodes die prematurely, the area monitored by these nodes will be out of coverage and thus the data that the nodes deliver to the sink are only from the partial network and cannot convey the full information of the network. Equalizing the node lifetime makes more data from the entire network be sent to the sink. C. Related Work Some existing clustering algorithms also take the directional data traffic into account and design clusters to make them have similar lifetimes. Ye et al. [13] improve LEACH [5] by assigning fewer nodes to the CHs farther away from the sink so that the higher burden, caused by the required higher transmission power to reach the CH, is compensated. Unfortunately, to send the data to the sink by a long single hop is infeasible or not energy efficient, especially in a large sized sensor network. As described previously in this paper, Muruganathan et al. [12] developed a protocol that creates clusters of similar size and uses multi-hop routing between CHs to the sink. The CH which does the last hop forwarding is randomly selected from the set of CHs to reduce the burden on CHs that are located nearer the sink. This algorithm will not scale to large networks as CHs located a significant distance from the sink will not have a high enough transmission power to reach the sink in a single hop. The directional data traffic towards the sink is also accommodated in [16] by using a chessboard configuration for heterogeneous WSNs formed by different sensor nodes. The network is separated into clusters (grids) as shown in Fig. 3. Each grid has a supernode (CH) which has more initial energy than the member nodes. Within each grid, a member node

Einitial / P = const

(2)

239

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2008 workshop proceedings.

nearer the CH will have a higher burden than the one farther away from the CH since it needs to relay more data from other member nodes. The chessboard configuration separates the network lifetime into two periods. In the first period, it schedules the CHs in the grids with darker (lighter) shade to be active. After these CHs run out of energy, the first period expires. The CHs in the grids with lighter (darker) shade are active when the second period initiates. The nodes which are closer to the CH with higher burden in the first period are further away from the CH in the second period, thereby distributing the higher burden. This algorithm is suitable for heterogeneous WSNs with some nodes more powerful than others to be CHs.

CH

Figure 3. Chessboard configuration (Adapted form [16]).

V. CONCLUSION This paper investigated the most recently proposed energyefficient clustering algorithms for WSNs. These algorithms were classified into: aggregating data, rotating the role of CH, equalizing cluster sizes, locating the CH in the central area of the cluster, assigning the lowest needed power to the nodes and optimizing routes using different power level clusters. The key characteristics of each category were summarized and some representative algorithms were described. The strengths and weaknesses of each were analyzed thereafter. The important feature of directional data traffic towards the data sink in sensor networks was analyzed and emphasized. Our analysis shows that energy efficiency of clustered WSNs can be further improved by equalizing the cluster lifetime by taking into account that the directional data traffic burdens the clusters differently. Besides the energy efficiency, another conclusion that can be reached from the analysis is that the performance of delivering data to the sink can also be optimized by equalizing cluster lifetime. This will prevent premature node/cluster death and the data that are delivered to the sink will carry the information of the entire network. REFERENCES
[1] [2] [3] [4] N. Vlajic and D. Xia, Wireless Sensor Networks: To Cluster or Not to Cluster? Proc. of WoWMoM06, 2006. D. Wei, Clustering Algorithms for Sensor Networks and Mobile Ad Hoc Networks to Improve Energy Efficiency, PhD thesis, University of Cape Town, Sep. 2007. J. Yu and P. Chong, A Survey of Clustering Schemes for Mobile Ad Hoc Networks, IEEE Communication Surveys, 7(1): 32-48, March 2005. D. Wei and H. A. Chan, 'Clustering Ad Hoc Networks: Schemes and Classifications, Proc. of IEEE SECON'06 under the track of IEEE IWWAN 2006, pp. 920-926, June 2006.

W. R. Heinzelman, A. Chandrakasan and H. Balakrishnan, Energyefficient Communication Protocol for Wireless Microsensor Networks, Proc. of the 33rd IEEE Hawaii International Conference on System Science, Jan. 2000. [6] H. Chen and S. Megerian, Cluster Sizing and Head Selection for Efficient Data Aggregation and Routing in Sensor Networks, Proc. of IEEE WCNC 2006. [7] J.-S. Liu and C.-H. R. Lin, Energy-Efficiency Clustering Protocol in Wireless Sensor Networks, Ad Hoc Networks, vol. 3, pp. 371-388, May 2005. [8] D. Wei, H. A. Chan and K. V. N. Kameri, Circular-Layer Algorithm for Ad Hoc Sensor Networks to Balance Power Consumption Proc. of IEEE SECON'06 under the track of IEEE IWWAN 2006, pp. 945-950, June 2006. [9] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam and E. Cayirci, Wireless Sensor Networks: A Survey, Computer Networks, vol. 38, pp. 393-422, 2002. [10] K. Dasgupta, K. Kalpakis and P. Namjoshi, An Efficient ClusteringBased Heuristic for Data Gathering and Aggregation in Sensor Networks, Proc. of IEEE WCNC, pp. 1948-1953, 2003, [11] B. Krishnamachari, D. Estrin and S. Wicker, The Impact of Data Aggregation in Wireless Sensor Networks, Proc. of Intl. Workshop on Distributed Event-Based Systems, 2002. [12] 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, pp S8-S13, March 2005. [13] M. Ye, C. Li, G. Chen and J. Wu, EECS: An Energy Efficient Clustering Scheme in Wireless Sensor Networks, Proc. of IPCCC 2005, pp. 535540, April 2005. [14] Y. He, Y. Zhang, Y. Ji and X. Shen, A New Energy Efficient Approach by Separating Data Collection and Data Report in Wireless Sensor Networks, Proc. of ACM IWCMC06, pp. 1165-1170, July 2006. [15] O. Younis and S. Fahmy, HEED: A Hybrid, Energy-Efficient, Distributed Clustering Approach for Ad Hoc Sensor Networks, IEEE Transactions on Mobile Computing, 3(4): 366-379, 2004. [16] X. Du and Y. Xiao, Increasing Network Lifetime by Balancing Node Energy Consumption in Heterogeneous Sensor Networks, Wiley Journal of Wireless Communications and Mobile Computing (WCMC), 2006, DOI: 0.1002/wcm.452, in press. [17] G. Pei and C. Chien, Low Power TDMA in Large Wireless Sensor Networks, Proc. of IEEE MILCOM, pp. 347-351, October 2001. [18] W. Wang and A. Jantsch, An Algorithm for Electing Cluster Heads Based on Maximum Residual Energy, Proc. of ACM IWCMC06, pp. 1465-1470, July 2006. [19] Z. El-Bazzal, M. Kadoch, A. L. Agba, F. Gagnon and M. Bennani, An Efficient Management Algorithm for Clustering in Mobile Ad Hoc Network, Proc. of PM2HW2N06, pp. 25-31, Oct. 2006. [20] S. Ghiasi, A. Srivastava, X. Yang and M. Sarrafzadeh, Optimal Energy Aware Clustering in Sensor Networks, Sensors, pp. 258-269, Jul. 2002. [21] S. Bandyopadhyay and E. J. Coyle, An Energy Efficient Hierarchical Clustering Algorithm for Wireless Sensor Networks, Proc. of IEEE INFOCOM, pp. 1713-1723, 2003. [22] R. Krishnan and D. Starobinski, Efficient Clustering Algorithms for Self-Organizing Wireless Sensor Networks, Ad Hoc Networks, 4(1): 3659, Jan. 2006. [23] L. Ramachandran, M. Kapoor, A. Sarkar and A. Aggarwal, Clustering Algorithms for Wireless Ad Hoc Networks, Proc. of the 4th International Workshop on Discrete Algorithms and Methods for Mobile Computing and Communications, pp. 54-63, August 2000. [24] G. Gupta and M. Younis, Load-Balanced Clustering of Wireless Sensor Networks, Proc. of IEEE ICC, pp. 1848-1852, 2003. [25] W. B. Heinzelman, Application-Specific Protocol Architectures for Wireless Networks, PhD Thesis, Massachusetts Institute of Technology, June 2000. [26] Y. Zhou, M. Hart, S. Vadgama and A. Rouz, A Hierarchical Clustering Method in Wireless Ad Hoc Sensor Networks, Proc. of IEEE ICC 2007. [27] P. Wang, C. Li and J. Zheng, Distributed Minimum-Cost Clustering Protocol for UnderWater Sensor Networks (UWSNs), Proc. of IEEE ICC 2007. [28] K. S. Manousakis and J. S. Baras, Clustering for Transmission Range Control and Connectivity Assurance for Self Configured Ad Hoc Networks, Proc. of IEEE MILCOM, pp. 1042-1047, 2003. [29] V. Kawadia and P. R. Kumar, Power Control and Clustering in Ad Hoc Networks, Proc. of IEEE INFOCOM, pp. 459-469, 2003.

[5]

240