On-Demand Energy Efficient Clustering in Ad Hoc Networks Mayur Maheshwari [email protected]

ABSTRACT Routing in MANET (Mobile Ad Hoc Network) is one of the most crucial aspects for its effective deployment. But, one of the factor that limits its use is the limited energy available to a node. Therefore, special care should be taken to reduce the energy consumption or utilize the available energy resource to its maximum. In this paper, we propose a scheme of efficient clusters through inter-cluster and intra-cluster management to reduce the energy consumption. The routing protocol effectively utilizes network. The first step is the cluster-head selection and then selecting a subset of clusterheads as gateway hosts. The second and most important step is intra and inter cluster management. This is the step where our main algorithm works in on-demand manner. Our on-demand clustering algorithm makes the clustering energy efficient.

Saurabh Arora [email protected]

manner, or be connected to a larger network, e.g., the fixed Internet. But, inspite of its immense potentials its widespread deployment has suffered because of the serious inhibition caused by limited energy of the batteries used in the mobile nodes. Hence, a lot of literature has been devoted to the optimization of energy consumed in routing packets over the network. In this respect, significant amount of work have been done on topology control, efficient broadcasting etc. Some generalized approaches were reported in[7]. Another set of works have been done in unicast routing in[2,5]. Also, efficiency have also been tried to induce in broadcasting and multicasting of packets in[14]. Along with these works, quite a lot of work in topology control has also been done[16,3]. GPS Supporting Satellite

Keywords Ad hoc network, routing, energy efficiency

Requires No GPS support

General Terms

On-Demand Cluster Maintenance

INTRODUCTION

In the recent years, MANET (Mobile Ad Hoc Network have received significant attention due to its immense potential applications in various situations such as battlefield, emergency relief etc. A MANET is an autonomous collection of mobile users (nodes) that communicate over relatively bandwidth-constrained wireless links. Each node is equipped with wireless receivers and transmitters using antennas that may be omni-directional or highly directional. Due to nodal mobility, the network topology may change rapidly and unpredictably over time. The network is decentralized, where network organization and packet delivery must be executed by the nodes themselves, i.e., routing functionality should be incorporated into mobile nodes. Nodes must also contend with the effects of radio communication, including multi-user interference, energy constraints etc. A MANET may operate in a stand-alone

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission of the authors. .

Gateway Selection as Subset of Cluster-Head

Energy Aware Nodes Classification

Algorithm

1.

Initail Cluster-Head Selection Algorithm

Effective Utilization of Factors like Duplicate Gateways An Ad Hoc Network

Increase in Network Life

Figure 1: A Schematic Representation of Our Work. In this paper, we propose a new scheme of on-demand clustering in ad hoc networks. As depicted in figure 1, our works simultaneously supports many other factors like network life, effective utilization factors like duplicate gateways which are otherwise treated as problems. The initial cluster head selection algorithm also significantly reduce the bandwidth consumption because of its distributed nature and a significant amount of attention has been devoted for effective load distribution which includes division of energy loads equally amongst all the nodes which increases the network life. Through this paper, our main contributions are, namely ✩ Design of an efficient cluster formation algorithm. ✩ Gateway selection algorithm. ✩ Energy aware nodes classification.

✩ Cluster maintenance algorithm. ✩ Effective utilization of duplicate gateway problem. ✩ Battery power recovery algorithm. ✩ Increases the network lifetime. The remaining parts of the paper proceeds as follows. The next section 2 provides basics of the network model and energy model that we have assumed for the rest of the paper. The section 3 provides the preliminary of the paper. Section 4 provides the various details of the routing algorithm which is basically our clustering algorithm. Simulation and Previous works are described in section 5 and 6. Section 7 concludes the paper.

2.

ASSUMPTION AND MODELS

2.1 NETWORK MODEL In our paper, we have modeled ad hoc network as directed transmission graph Gt = (V, E). Here, V denotes the nodes (mobile devices) and E is the collection of all the connectivity which varies over time due to the mobility of individual nodes. Let T (u) be the set of all mobile devices which are within the transmission range of mobile device u. All the nodes within T (u) can communicate with each other. For any edge (u, v) ∈ E there is an associated cost which we denote as cu,v and every node has a variable power capacity pu .

In our work, network life is an important factor. In this respect, we have defined network life. According to us, for the given graph G(t) = (V (t), E(t)) representing an ad hoc network at given time t, where V (t) and E(t) varies over time and represents alive connections. Assuming that G(0) is connected, and N (t) = |V (t)| gives of the cardinality of the network at time t and N = N (0). Under these cases, the network life is defined as the minimum between t1 and t2 , where t1 is the time it takes for the cardinality of the largest connected component of G(t) to drop below c1 × n(t), t2 is the time it takes for n(t) to drop below c2 × n, and 0 ≥ c1 , c2 ≤ 1. Here, c1 and c2 defined the quality of the network. For ad hoc network with high quality, c1 and c2 tends to value 1. Other for ad hoc networks with lower quality, c1 and c2 tends to 0.

4. ROUTING ALGORITHM As depicted by figure 2, the ad hoc networks consists of 12 nodes. There are three region where nodes can communicate through local broadcast, namely region 1, 2 and 3. Nodes 7 and 9 serves the roles of gateway hosts.

1 3 2

12 5

4

10 7

11

9

2.2 ENERGY CONSUMPTION MODEL In the most common of the power-attenuation model, the signal power falls as 1/r α , where r is the distance from the transmitter antenna and α is a constant between 2 and 5 which depends on the wirelees transmission environment. This is typically called the path loss parameter. Apart from the transmitter-receiver separation, the path loss also depends on the heights of the transmitting antennas. We have assumed that all the mobile devices have similar antenna heights so that in this paper, we discuss the path loss that is dependent only on the distance. Thus, the energy required to support a link between two nodes u and v, separated by a distance r is rα , which is called the transmitting power of node u. So, we can see by simple geometry computing, that relaying a signal through intermediate nodes can result in lower energy consumption, than communication over a large distance directly within the two nodes because of the non-linear power attenuation.

3.

PRELIMINARIES

In most of the earlier ad hoc routing algorithms using clusters[9,1,8] were tested in simulated environment. And in those cases, accurate global information regarding connectivity relations, node locations are easily available. Unlike these cases, in actual environment, these information are hard to collect and collected through periodic broadcast of control packets. Our routing approach improves over this case because unlike those cases, our algorithm does not require GPS support and cluster heads are selected in a localized manner. Here, we have provided the symbols that we have used in rest of the paper. NODE ID provide unique ID for nodes, NEXT HEAD represent the NODE ID which will be the next cluster head, C ID represents a cluster ID, CL : Used to denote a cluster. In additions, CH denotes a cluster head, CM ( CL ) = {M C1 , · · · , M Cn } specifies cluster members for the cluster CL and Pstate is used to represent a power level of battery at given state.

6

8

Figure 2: The figure represents three local broadcast regions. They are as region 1: (1, 2, 3, 4, 5, 7), region 2: (6, 7 ,8 ,9) and region 3: (9, 10 ,11, 12).

4.1 Working Mechanism In this section, we have explained the working of routing protocol. Our explanation is accompanied by figure 2. Figure 2 gives a scenario of ad hoc network based on which we have provided our example. To explain the working mechanism, we have classified the nodes into four categories. They are ✩ Cluster Head Nodes : It is the head of the cluster to which it belongs, i.e. it is responsible of routing all the packets from the nodes within the cluster to other cluster heads and/or free nodes and vice versa. It is also responsible for the management process of the cluster. ✩ Cluster Slave Nodes : It is a simple node belonging to some cluster i.e. which works under the guidance of a cluster head. ✩ Free Nodes : A free node is responsible only for its ownself, and works like a common node in an ad hoc network devoid of any clustering. ✩ Dead Nodes : Those nodes which either move out of the range of any node, or puts his network interfaces to sleep is termed as a dead node.

The crude form of the intial cluster formation algorithm is described in algorithm 1. In this algorithm, initially all the nodes are able to communicate only through local broadcast. Multi-hop communication is not possible initially in the network. According to our algorithm, all nodes first broadcast HELLO packets along with their NODE ID s. After this step, all node will have neighbouhood information of the nodes. If Ni be any node, based on the model described in section 2, each nodes will have T (Ni ). For another node Nj , cNi ,Nj is considered uniform and energy level of nodes are assumed to be equal . Next step is that node Ni will broadcast the T (Ni ) to all neighbouring nodes. This makes the two-hop information available to all the nodes. Next we choose those nodes to be cluster-heads which have atleast two disconnected neighbours. This is the crude form of the clustering algorithm. Formally, the two phases of algorithm that leads to cluster formation can be stated as below: [htb]

particular node, but also helps in redeeming some amount of energy as mentioned in [4]. Second task of the cluster head is to effectively utilizing duplicate gateways. The cluster-head maintains list of gateways obtained during gateway selection phase. One of the task of cluster head performs is regular change of the gateways. This is done in effect of increasing the network life by not allowing any node to bear the major energy load. Moreover, one of the problem solved by this approach is that the purpose of maintaining cluster is solved without significant overhead of regular broadcast of control packets. We have represented the transfer the responsibility as gateways in on-demand manner. The packets maintains an extra field that can be ON or OFF . Whenever a node receives a packet containing extra field to be OFF , it turns the node status of gateway to non-gateway status. But at the same time, it turns some other node which are member of gateway list maintained by cluster-heads to be gateway through delivery of packets as ON field. This process reduces the following set of problems:

: Algorithm for initial cluster formation A. Cluster-Head Selection Algorithm: [1]: Each node broadcasts HELLO packets along with their NODE ID s. This provides each of nodes with list of NODE ID . [2]: Each nodes send ACK along with the list of their neighbours based on NODE ID . All nodes after this step gains total two-hop neighbourhood information. [3]: Based on two hop information, we select those nodes to be cluster head which have atleast two disconnected neighbours. [4]: This leads to the formation of cluster-heads. B.Gateway Generation Algorithm: [5]: During packet transfers, those cluster-heads which transfers packets from two cluster-head neighbours are notified. [6]: Those nodes are notified with category gateway. [7]: They are now gateways. [8]: Since the cluster heads are reduced to nodes and gateways to clusters heads, we use same optimization algorithm for gateway generation like in cluster heads. The formed cluster selection algorithm may result in a lot of nodes that are redundant in respect of working as cluster-head. The cluster-head selection algorithm is thus crude form not much efficient in bringing efficiency into the system. The following two steps are required for optimizing the cluster formation algorithm:

✩ Duplicate Gateways: It may happen during the gateway selection phase described in algorithm 1, multiple gateway may result. By our approach, duplicate gateway problem is removed with less overhead. This brings efficiency into routing and hence saves energy. This also makes the broadcasting efficient and hence contribute to network lifetime.

✩ We consider two connected nodes Ni and Nj in the generated set of clusters. If we exclude Nj from the neighbours list of Ni and it is a subset of the neighbours of Nj , then we consider the node Ni as redundant and removed from the list of cluster-head. ✩ Taking Ni and Nj as neighbours. Taking a third node Nk and, excluding Ni and Nj from its neighbours set, a new set of neighbours are generated. If the newly generated set of neighbours is subset of union of the set of neighbours of Ni and Nj , then exclude Nk from the set of cluster heads. After the initial cluster formation, network management process begins. It is divided into two parts. One is called intra-cluster management while the other is called inter-cluster management.

4.2 INTRA-CLUSTER MANAGEMENT Intra-cluster management basically deals with two tasks where the cluster head plays an active part. One of the tasks is to sleep off the node whose battery power level has dipped below the threshold power level which we have explained in detailed in part 4.5. Whenever, any node’s battery level dips below the threshold battery level, then it sends a packet intimating the head of his decision to interface in sleep state, and the cluster head removes him from his neighbourhood list. This not only helps in saving battery of that

✩ Load Maintenance: Through the maintenance of two fields: ON and OFF , we have distributed the load between gateways. This inherits benefits from existence of multiple gateways rather and finally increases network life-time as defined in section 3. So, any members of the gateway list can get the responsibility of gateway host which satisfies the threshold value of the battery level. The cluster head through the use of gateway changing capability increases the life of the network. : Intra-Cluster Management Algorithm A. Initial Gateway Selection Algorithm [1] : All nodes are initially in state as resulted form clustering algorithm. [2] : Gateways in a cluster informs cluster-head about its existence. [3] : Only one randomly selected gateway is marked ON by the cluster-head. B. Duplicate Gateway Utilization Algorithm GatewayP ower < Pthreshold Inform cluster-head to select another gateway. C. Gateway Selection Algorithm [1] : Current cluster-head is informed to change working gateway. [2] : Send packets with the extra field set to OFF [3] : Randomly choose another gateway and send packets with the extra field set to ON .

5. INTER-CLUSTER MANAGEMENT After the initial cluster formation, the nodes can communicate efficiently. But the mobility of nodes can result in new positions of all nodes and a new topology. These changes can result in a disconnected network having all the clusters restricted to a small area of the whole network. Such a situation can result in degradation of network performance, as well as the energy consumption. We term such a situation as cluster clogging. Hence we propose an algorithm to remove cluster clogging. The management is based upon the energy-aware classification of nodes and in on-demand manner. The explanation we have provided is through the help of the figure 3. This section details the intricacies involved in the management of the clusters in the network. We consider the effect of movement

1

to other nodes, then immediately all nodes in the free node set can communicate to the same.

3

2

5.0.4 DEAD NODES

14 11 5

4

7

15 9 6 13 12 8

10

Figure 3: The figure represents three local broadcast regions. They are as region 1: (1, 2, 3, 4, 5, 7), region 2: (6, 7 ,8 ,9, 12), region 3: (8, 9, 10 , 12), region 4: (10, 13), region 5: (15), region 6: (14) and region 7: (5, 11)

of various nodes on the network and the reformation steps taken thereafter. Firstly we consider the effect of movement of cluster slave nodes. As long as the cluster slave node is under at least one of the cluster heads it wont do anything special. If a cluster slave node comes under more than one cluster heads then it will become a part of probable gateway nodes list. But if it goes out of the range of all the neighbouring cluster heads then it will transmit a packet to sense its neighbours and take appropriate actions. For the purpose of keeping a heuristic idea of its belonging, the cluster head keeps a timer which is initiated when to joins the network i.e., initial time. Whenever, the node receives packets from its cluster head (participating in the network activity), it re-initiates the timer to the initial value. If the time reduces to 0, that is the node receives no packet from any cluster head for that period of time, then it is said to be in the doubt state. Whenever in doubt state, the node broadcasts a packet telling to be replied by any cluster head in its local broadcast range. Under this case, four cases may arise based upon our energy aware classification, they are:

5.0.1 COMMUNICATE WITH OTHER CLUSTER HEAD If some CH exists in the local broadcast zone of the node, then it replies and updates it to his CH list. One case that may arise here is the multiple cluster head may consider it under neighbourhood list. This creates no problem because the node may become part of multiple gateway list. Like in node 10 in the figure 3 where 12 and 13 are cluster heads and 10 is in the neighbourhood list of both.

5.0.2 COMMUNICATE THROUGH INTERMEDIARY NODE Under this case, any other node that is communicable by the node performs the initial algorithm. That is, the intermediary node act as gateway and the considered node becomes cluster-head. This brings the connectivity to other part of the network because intermediary node node must belong to some cluster head.

5.0.3 FREE NODES This case represents the scenario where two or more nodes are connected to each other but disintegrated from the original network neighbourhood. Maintaining free node brings an important point in the sense that if any node in the free node set can communicate

The last and the worst scenario is the case where the node is unable to communicate to any node in the network. Under these circumstances, network interface may be put in sleep state and timer based process is used to test connectivity in later stages. Then, we consider the result of the movement of gateway nodes. If the gateway which was being used is no more reachable then a new gateway node will be chosen from amongst the list of probable gateway list. If the cluster head node has no gateway connecting it to any other cluster head then it is said to be in the doubt state. In the normal transmission of a packet, an acknowledgement packet is sent back, to the source node. If the source node does not receives the packet from the gateway, which it expected to(through its routing table), it realizes that the particular gateway has either died r moved away from its range. So, it will send a packet to another node from amongst the probable gateway list telling it to act as a new gateway node. Now, we consider the effect of movement of clutser head nodes. The movement of cluster heads results in a situation analogous to the situation of the movement of cluster slave nodes and the situation of movement of gateway nodes. Its movement will either result in some cluster slave nodes out of its range or some gateway nodes no more in a situation of acting as gateways. Both of the situation is remedied in the way above two situations have been handles. Cluster head also makes use of the packets coming through gateway nodes to update their information. The information collected by the cluster head helps it in maintaining a neighbourhood list. This list helps him in keeping track of the gateway nodes. If more than one gateway node between the same pair of cluster heads are present, then the cluster head sends a packet telling the duplicate gateway node to sleep off. All the data packets which arrive at a node contain the destination node. Before forwarding any packet forward, all the cluster head compares the destination node with their neighbourhood list. If the destination node exists in their neighbourhood list then the to-be-forwarded field is set to false, otherwise it is set to be false. All the packets received by the destination node are replied by them with an acknowledgment packet. This helps in further maintenance of the cluster network as well as in ensuring packet received by the destination node and thus saves retransmission and hence helps in reducing energy consumption. If the acknowledgment is not received by the cluster head then it strikes off that node from its neighbourhood result and retransmits the packet with the to-be-forwarded field as true.

5.1 REMOVAL OF GRAY ZONE PROBLEM The problem of gray zone[10] is a common problem that arises in ad hoc networks. Gray zone as depicted in figure 4 refers to an area. All the nodes belonging to this area are although present in the neighbourhood list of the cluster head, but are not reachable by normal data packets. This results in repeated transmissions of the same data packet, resulting in energy loss. One of the main reasons for the gray zone problem is the use of HELLO packets for neighbour sensing. This problem of gray zones arises because HELLO packets are transferred at low bit rates in IEEE 802.11b, and low bit transmissions are more reliable and also since the small packets are generally less prone to bit errors, hence the area where these HELLO packets can reach may be unreachable by the normal data packets. But our cluster maintenance makes use of normal data packets sent at normal transmission rates, hence the problem

of gray zones doesn’t happens.

6. SIMULATION

Reach of Broadcast

Reach of Unicast

GW

C1

C2 MN

Gray Zones

Gray Zone Traversal

Figure 4: The figure represents the gray zone problem that are typically encountered in ad hoc networks.

7. PREVIOUS WORKS

5.2 BATTERY POWER RECOVERY Other than decreasing the power consumption within the network through effective clustering technique, our routing algorithm is also customized to increase the network life through effective energy management. In this respect, we have utilized the fact that for batteries the power recovery occurs if it is not involved in activities like transmission and receiving signals. Under this criteria, we have made certain advantages like each cluster-head need not have the knowledge of battery levels of cluster members. Also, we assume that initially, all nodes have initial battery levels and power consumption occurs based upon the energy model in 2.2. Each node maintains a value called battery threshold, denoted by Pthreshold . In our scheme, each node notifies its cluster-head if its battery level goes down bellow the threshold value. The amount of energy that a battery can recover during a slot is given by Erecovery = a0 e−gN (Be −b)

(1)

where Be is the initial charge and b is the current battery level, a0 and gN are parameter that depends on the recovery capability of the battery. We have used this fact to increase the network life. The battery energy mode is based upon [4]. When a node goes below its threshold value, i.e., PID ≤ Pthreshold , then the node informs its cluster-head regarding this effect and cluster-head removes it from the list of its cluster members, i.e., CM = CM − {NID }. When the node regains the battery power, it is then treated as free node and it joins the cluster based on previous description. This is described in algorithm 3.

5.3 ENERGY SAVING IMPLICIT IN OUR ALGORITHM : Battery power recovery algorithm A. If power goes below threshold value P owerID < Pthreshold Set the status as dead node. Inform cluster-head with ID CM = CM − {NID } B. If node power goes above threshold value. P owerID ≥ Pthreshold Set status as free node. Run algorithm for free node. Most of the clustering algorithms proposed earlier assume the knowledge of network topology. This assumption is based on the use of GPS. But for all practical purposes, they don’t provide exact configuration. And, moreover their use adds on to a lot of energy overhead. Our protocol does not uses GPS, and hence saves a lot of energy on that part. And the other schemes used like duplicate gateway off and removal of gray zone helps in saving energy and increasing overall network lifetime.

The problem of minimizing energy requirements in ad hoc networks has recieved significant attention in recent years [6,11]. Most of the works have been devoted to design of routing protocols. Many application level schemes are designed to minimize energy requirements. Most studies are categorized to energy-efficient routing and power aware routing. Our routing scheme works on both properties: energy efficiency and power-aware communications. The difference between energy efficient communications and power aware communications were explained in [15]. In another paper, author have provided various power-aware parameters to determine feasible routes [12]. Some of routing protocols are also dependent on GPS like systems. Along with GPS support, a routing protocol has been proposed in [13] In comparison to existing routing protocols, our clustering approach wins in respect of bringing energy efficiency into the clusters. Most of the existing clustering solution have been simulated assuming complete neighbourhood information. But, in real situation attaining complete information of the network is highly impossible, even if we use techniques like GPS, which again adds to the energy consumption. Gateway Selection Algorithm Evaluation 100

80

60

40

20

0

Crude Clustring Algorithm First Optimization Second Optimization

Number of Cluster Head and Gateway Hosts

Covered Area by Ad Hoc Network

Some amount of simulation was performed to investigate the performance of the system. Major part of the simulation was devoted to cluster formation algorithm evaluation. One of the major factor that we have studied is the size of the cluster heads. The topology used in the simulation process was random topology. We have evaluated the conditions under two conditions, namely, one is under sparse ad hoc networks and other one is dense ad hoc networks. As depicted by figure 5, we have evaluated the clustering algorithm for sparse networks. In the crude form, the clustering algorithm performance is low. In almost all the cases, the number of cluster heads and gateways approximates to total number of mobile devices. In the second simulation conducted, we have studied the clustering algorithm for dense ad hoc networks. From the results as shown in the figure 6, we can conclude that our works performs much better in dense ad hoc networks.

0

20

60 40 Number of Mobile Devices

80

100

Figure 5: Simulation Results of the Clustering Process for Sparse Networks.

Gateway Selection Algorithm Evaluation 100

60

40

20

0

Number of Cluster Head and Gateway Hosts

80

Crude Clustring Algorithm First Optimization Second Optimization

0

[7] [8]

[9]

[10] 20

60 40 Number of Mobile Devices

80

100

Figure 6: Another Simulation Results of the Clustering Process for Dense Ad Hoc Networks.

The major hurdle in achieving complete network information is node mobility. Most of the existing solution use broadcasting of neighbourhood information or they use GPS in building up the network information. Unlike all these case, our approach uses only two hop information in on-demand manner.

8.

CONCLUSION

In this paper, we have presented a clustering scheme for ad hoc network that is well managed in terms of energy efficiency. We have tried to introduce on-demand behaviour rather than relying on periodic characteristics of most of existing clustering algorithm. This not only reduces the energy consumption but also results in increased network life of ad hoc network as gateway responsibility is rotated among duplicate gateway. This rotation on responsibility also helps in retrieval of certain amount of energy. Through our approach, we have introduced the novel concept of energy efficient and energy managed clustering in ad hoc network.

REFERENCES [1] ANERJEE,S.,AND KHULLER,S. A clustering scheme for hierarchical control in multi-hop wireless networks. In INFOCOM(2001), pp.1028-1037 [2] ANERJEE,S.,AND MISRA,A. Minimum energy paths for reliable communication in multi-hop wireless networks.ACM MobiHOC 2002 (June 2002). [3] HEN,B.,JAMIESON,K.,BALAKRISHNAN,H.,AND MORRIS,R. Span:An energy-efficient coordination algorithm fo topology maintenance in ad hoc wireles networks. In ACM/IEEE Mobicom 2001. [4] HIASSERINI, C.F., AND RAO, R. Energy efficient battery management. In Proceedings of INFOCOM 2000(2000),vol.2, pp. 396-403. [5] LBATT, T., KRISHNAMURTY, S., CONNORS, D., AND DAO, S. Power management for throughput enhancement in wireless ad hoc networks. In ICC 2000. [6] EINZELMAN, W., CHANDRAKASAN, A., AND

[11] [12]

[13] [14] [15]

[16]

BALAKRISHNAN, H. Energy-efficient communication protocol for wireless microsensor networks. In Proceedings of the 33rd Hawaii International Conference on Systems Sciences(HICSS 2000)(January 2000). AJARAMAN, R. Topology control and routing in ad hoc networks: A survey. ACM SIGACT News 33,2(June 2002). RISHNA,P., CHATTERJEE, M., VAIDYA,N.H.,AND PRADHAN, D.K. A cluster-based approach for routingin ad hoc networks. In USENIX Symposium on Location Independent and Mobile Computing(April 1995), pp. 1-8. RISHNA,P., CHATTERJEE, M., VAIDYA,N., AND PRADHAN, D. A cluster-based apoproach for routing in dynamic networks. ACM SIGCOMM Computer Communication Review(April 1997), 49-65. UNGREN, H., NORDSTRM, E., AND TSCHUDIN, C. Coping with communication gray zone in ieee 802.11b based ad hoc networks. In WoWMoM02, September 28, 2002, Atlanta, Georgia, USA(2002). ODUPLO, V., AND MENG, T.H. Minimum energy mobile wireless networks. IEEE Journal on Selected Areas in Communications 17,8(August 1999). INGH, S., WOO, M., AND RAGHAVENDRA, C.S. Power-aware routing in mobile ad hoc networks. In ACM International Conference on Mobile Computing and Networking(MOBICOM)(1999). TOJMENOIVIC, I., AND LIN, X. Power-aware localized routing in wireless networks. IEEE Transactions on Parallel and Distributed Computing 12, 10(October 2001). IESELTHIER, J., NGUYEN, G., AND EPHREMIDES, A. Energy Efficient broadcast and multicast trees in wireless networks. Mobile Networks and Applications(2002). IESELTHIER, J.E., AND NGUYEN, G.D. Enerfy-aware wireless networking with directional antennas: The case of session-based broadcasting and multicasting. IEEE transactions on Mobile Computing 1,3(July-Sept 2002), 176-191. U, Y., HEIDEMANN, J.S., AND ESTRIN, D. Geography-informed energy conservation for ad hoc routing. In ACM/IEEE Mobicom 2001(2001),pp.70-84.