[ieee 2012 seventh international conference on broadband and wireless computing, communication and...

6
DominoMAC: A Wireless Sensor Networks Medium Access Protocol Fayez Gebali, Senior Member, IEEE, Rami Sulaimani and Haytham Elmiligi Electrical and Computer Engineering University of Victoria, Victoria, BC, Canada Email: [email protected] ABSTRACT Wireless sensor networks have unique characteristics and restricted energy efficiency performance require- ments that dictate special medium access control (MAC) protocols. In this work we show that design of any MAC should mainly consider the proximal nodes near the sink since they carry the most traffic while being the most crucial for survival of the network. We provide a comprehensive analysis of sensor network traffic and its impact on the design of MAC protocols. Network per- formance is modelled using analysis and verified using numerical simulations. The MAC protocol we propose here has several attractive features: viz. is collision-free, allows smooth degradation of network due to battery discharge, has short active period to prolong network operational time. Index Terms: Sensor networks, Wireless MAC 1 I NTRODUCTION Wireless sensor networks (WSN) operate under se- vere energy budget which require design of specialized medium access control protocols (MAC). The situation is made worse by limiting the transmitted signal en- ergy [1], [2]. However energy consumption increases after retransmitting packets due to MAC collisions and bit errors from physical channel noise. Collaborative message forwarding implies that each node will carry extra traffic over and above the traffic it generates. As a result, nodes near the sink have to carry a heavier traffic load compared to farther nodes far away from the sink. Several researchers reportedthat the sensors closest to the sink tend to deplete theirenergy budget faster than other sensors [3], [4], [5], [6], [7]. As a result, an energy hole is created around the sink and the network stops functioning even though most of its nodes still have energy to transmit. An efficient MAC protocol must take into account this fact. Therefore the performance of the WSN system and of each node in the system will depend on forwarded traffic, error control protocol and the MAC protocol being used. Several protocols are at work in a WSN system such as error control, MAC, coding and modulation. Thus multi-layer model of a WSN is required to get a more representative analysis of the performance. We consider a simple sensor network with one in- terface to the outside world through a single sink. The sink has sufficient energy compared to the sensor nodes comprising the network. The nodes are assumed to have a limited energy budge and can control their power during packet transmission such that radiated energy consumption follows the equation [8]: E = d α + c (1) where d is the transmission distance, α 2 and c is a technology-dependent positive constant. 2 WSN NETWORK MODEL We assume wireless sensor network with N nodes which are randomly distributed around a sink in a square area of side length L as shown in Fig. 1(a). The sink is located at the center of the square area. The node density is given by ρ = N/L 2 . We further assume for simplicity that all nodes have the same transmission range r. Assuming that a node can adaptively adjust its power level and transmission range is not a common feature found in a sensor node. Such simplifying assumptions will allow us to derive simple expressions to guide our design of an energy- efficient MAC protocol. Define an annulus region i which is centered around the BS and has radii ir and (i 1)r. As an approximation, the nodes that lie in that annulus are identified as belonging to the set S i , with 1 i m. A node in S i is assumed to require i hops to send or receive a message to/from the sink. In that sense, a node in S i will communicate with a parent node in S i1 and will also communication with child nodes in S i+1 . The number of sets m is given by m = L r (2) 2012 Seventh International Conference on Broadband, Wireless Computing, Communication and Applications 978-0-7695-4842-5/12 $26.00 © 2012 IEEE DOI 10.1109/BWCCA.2012.12 8

Upload: haytham

Post on 24-Mar-2017

217 views

Category:

Documents


2 download

TRANSCRIPT

DominoMAC: A Wireless Sensor NetworksMedium Access Protocol

Fayez Gebali, Senior Member, IEEE, Rami Sulaimani and Haytham ElmiligiElectrical and Computer Engineering

University of Victoria, Victoria, BC, CanadaEmail: [email protected]

ABSTRACT

Wireless sensor networks have unique characteristicsand restricted energy efficiency performance require-ments that dictate special medium access control (MAC)protocols. In this work we show that design of any MACshould mainly consider the proximal nodes near thesink since they carry the most traffic while being themost crucial for survival of the network. We provide acomprehensive analysis of sensor network traffic and itsimpact on the design of MAC protocols. Network per-formance is modelled using analysis and verified usingnumerical simulations. The MAC protocol we proposehere has several attractive features: viz. is collision-free,allows smooth degradation of network due to batterydischarge, has short active period to prolong networkoperational time.

Index Terms: Sensor networks, Wireless MAC

1 INTRODUCTION

Wireless sensor networks (WSN) operate under se-vere energy budget which require design of specializedmedium access control protocols (MAC). The situationis made worse by limiting the transmitted signal en-ergy [1], [2]. However energy consumption increasesafter retransmitting packets due to MAC collisions andbit errors from physical channel noise. Collaborativemessage forwarding implies that each node will carryextra traffic over and above the traffic it generates. Asa result, nodes near the sink have to carry a heaviertraffic load compared to farther nodes far away from thesink. Several researchers reportedthat the sensors closestto the sink tend to deplete theirenergy budget fasterthan other sensors [3], [4], [5], [6], [7]. As a result, anenergy hole is created around the sink and the networkstops functioning even though most of its nodes stillhave energy to transmit. An efficient MAC protocol musttake into account this fact. Therefore the performance ofthe WSN system and of each node in the system willdepend on forwarded traffic, error control protocol andthe MAC protocol being used. Several protocols are at

work in a WSN system such as error control, MAC,coding and modulation. Thus multi-layer model of aWSN is required to get a more representative analysisof the performance.

We consider a simple sensor network with one in-terface to the outside world through a single sink. Thesink has sufficient energy compared to the sensor nodescomprising the network. The nodes are assumed to havea limited energy budge and can control their powerduring packet transmission such that radiated energyconsumption follows the equation [8]:

E = dα + c (1)

where d is the transmission distance, α ≤ 2 and c is atechnology-dependent positive constant.

2 WSN NETWORK MODEL

We assume wireless sensor network with N nodes whichare randomly distributed around a sink in a square areaof side length L as shown in Fig. 1(a). The sink is locatedat the center of the square area. The node density is givenby ρ = N/L2. We further assume for simplicity that allnodes have the same transmission range r. Assumingthat a node can adaptively adjust its power level andtransmission range is not a common feature found in asensor node.

Such simplifying assumptions will allow us to derivesimple expressions to guide our design of an energy-efficient MAC protocol.

Define an annulus region i which is centered aroundthe BS and has radii ir and (i−1)r. As an approximation,the nodes that lie in that annulus are identified asbelonging to the set Si, with 1 ≤ i ≤ m. A node inSi is assumed to require i hops to send or receive amessage to/from the sink. In that sense, a node in Si willcommunicate with a parent node in Si−1 and will alsocommunication with child nodes in Si+1. The number ofsets m is given by

m = �Lr� (2)

2012 Seventh International Conference on Broadband, Wireless Computing, Communication and Applications

978-0-7695-4842-5/12 $26.00 © 2012 IEEE

DOI 10.1109/BWCCA.2012.12

8

��� ��� � �� ��

���

���

��

��

Fig. 1. A WSN network. (a) Physically nodes are ran-domly distributed. (b) Logically nodes are assigned tocommunicating sets.

We should note here that the above equation givesonly an approximate estimate of the number of sets.The actual number of sets will depend on the physicaldistribution of the nodes and the transmission/receptionrange of each node. Nodes in S1 have the sink as theirparent node and we call these nodes proximal nodes.Nodes in Sm have no children and we call these nodesdistal nodes.

The network can also be represented by a graph G =(V, E) [9] where V is the set of vertices correspondingto WSN nodes and E is the set of edges. The edgesrepresent the communication links between the nodes.Two nodes share a common edge if they are locatedat a distance less than r. We found it more useful todefine our WSN network as a set of communicatingclusters where each node defines a set of parents, siblingsand children as discussed below. We organize the nodesinto communicating sets as shown in Fig. 1(b). Thealgorithm for clustering the nodes into sets is explainedin Algorithm 1. A node in set Si stores the following:

1) Node ID or MAC address which is unique to eachnote not unlike the Ethernet ID assigned to eachEthernet card.

2) Node index Si which identifies the node requires ihops to communicate with the sink

3) Parent vector Vp is a list of all nodes in set Si−1 thatare within transmit/receive radius of the node.

4) Sibling vector Vs is a list of all nodes in set Si thatare within transmit/receive radius of the node.

5) Child vector Vc is a list of all nodes in set Si+1 thatare within transmit/receive radius of the node.

6) Any control parameters such as duration of ac-tive/sleep periods and other physical and mediumaccess layer communication parameters.

This information is shared with neighboring nodes dur-ing network initial configuration and after detected net-work changes.

We should note that in Algorithm 1 a hello() messagecontains, among other things, the sender’s ID and theindex value of the set that the sender node belongs to.

For example, the BS sends a hello() message with ID =0 and the set index value is 0.

Algorithm 1 is distributed and iterative. Steady stateis reached when every node has been reached andconfigured. The assignment of nodes to sets occurs in aripple effect starting with set S1 until all nodes have beenallocated to m sets; where m depends on the geographi-cal distribution of the nodes. By construction, a node inset Si can communicate only with neighboring nodes insets Si−1, Si and Si+1 only. Nodes in Si require i hops toreach the BS. The clustering of nodes into sets helps bothmessage routing and designing an energy-efficient MACprotocol as will be explained in sequel. The clusteringshown in Fig. 1(b) is slightly correlated with the physicaldistance of each node from BS. We should point outhere that the presented clustering algorithm is basedon the geographical location of the nodes relative toeach other. As a simplifying approximation, we can statethat set Si contains all the nodes that lie in the annuluscentered at BS and the two radii (i−1)r and ir. It shouldbe mentioned that this clustering of nodes was doneintentionally in this work. In other systems, the nodesare already clustered as dictated by the neighbors theycan reach; e.g. S-MAC [10].

One fact becomes apparent in any WSN system with aBS: due to message forwarding, traffic carried by nodesin set Si is due to locally generated traffic and all trafficsgenerated by nodes in sets Sj (where i < j < m). Thussurvival of proximal nodes closer to BS is more urgentcompared to survival of distal nodes more distant formBS. The former are not crucial to forwarding messageswhile the latter are very crucial to forwarding traffic. Forexample, if all nodes in set Si all fail then the networkeffectively shrinks to a radius of iR only.

The procedure for reconfiguring the network dependson whether a node runs out of power or a new node isadded [11]. Let us deal with the former situation first.Before a node leaves the network or senses that it isrunning out of power, it could informs all its neighborsthat it will soon be leaving the network. All neighborswill then delete this node update their respective Vp,Vs and Vc vectors. That change is local and does notneed to propagate to other nodes. Alternatively, BS couldperiodically send an update message update() whichinitiates an algorithm similar to Algorithm 1 so that eachnode updates its Vp, Vs and Vc vectors.

When a node joins the network it first listens to thetraffic and monitors the associated node indices. Thephysical parameters of the node and Algorithm 1 ensurethat the new node will only sense node indices withadjacent values such as i−1, i , and i+1. The new nodewill also create three new vectors V1, V2 and V3 vectors,which will become eventually the Vp, Vs and Vc vectors.The new node will insert the indices it detected into thenew vectors such that equal indices are added to thesame vector. That node will then assign its node indexto the intermediate value i and assign the labels theparent, sibling and child to the three vectors it created.

9

Algorithm 1 WSN initialization or reconfiguration algorithm.Procedure InitializeNetRequire: s = −∞ for all nodes in net /* set assignment is initially undefined */Require: Vp, Vs, Vc = 0 for all nodes in net

1: S(0) = BS /* Set S0 contains the sink ID */2: BS send hello() message /*BS initializes all neighboring nodes */3: for all nodes receiving hello() message do4: if S(receiver) = −∞ then5: S(receiver) = S(sender) + 1 /* receiver index set value is incremented relative to sender */6: receiver sends hello() message /* message contains receiver’s ID and its set index value */7: Add sender_ID to receiver’s Vp vector8: else if S(receiver) > S(sender) then9: Add sender_ID to Vp vector

10: else if S(receiver) = S(sender) then11: Add sender_ID to receiver’s Vs vector12: else if S(receiver) < S(sender) then13: Add sender_ID to Vc vector14: end if15: end for

A special case will occur if the new node happens tofind itself in set Sm. In that case only two node indexvalues will be observed i and i + 1 and the new nodewill assign itself to the larger index value and declareall nodes with the lesser index to be its parents. Havingdone that, the new node will then issue hello() messageso that the neighboring nodes will update their lists.

As a result of above algorithm, some nodes might beisolated from the network. This arises if a node is locatedsuch that no nodes are present within an r radius. Whenthis situation occurs, these nodes will not be used sincethey are simply too far located to be of any use to theWSN system. Correcting this situation will result in amore complex node clustering algorithm and even morecomplicated MAC protocol.

3 PROPOSED DOMINO MAC PROTOCOL

Most WSN protocols are based on S-MAC and theproblem with this protocol is the long listen period toaccommodate forwarding packets from the most distantnode to the sink. Thus any energy-aware MAC protocolmust not use time division multiple access (TDMA)among the nodes since this requires proximal nodes toremain active for relatively very long periods and willthus reduce their lifetime. The protocol must use someother form of multiplexing node traffic such as time divi-sion duplex (TDD) for uplink/downlink directions, codedivision multiple access (CDMA), orthogonal frequencydivision multiple access (OFDMA) or even by compress-ing/dropping or postponing some node transmissionstill later frames.

The MAC protocol we propose here is distributedbut its operational phases are initiated by the sink. Thephysical channel is divided into MAC frames of durationτf . A frame consists of long sleep period and a shorteractive period that consists of control, listen and talk

phases with dynamically variable duration that is set bythe control messages at the start of the active period. Theprotocol uses OFDMA extensively using a cascading ordomino-like activation schedule as discussed below.

Figure 2 shows the Domino MAC active period struc-ture. The arrows indicate the direction of data/controlflow. An up-arrow indicates flow of information from thecurrent node to its parent nodes toward BS (uplink). Adown-arrow indicates flow of information from the cur-rent node to its children nodes away from BS (downlink).Note that the node is in transmit mode for the phases:Talk1, Control2 and Talk2. These phases are adjacent toreduce delay due to switching the radio transmit/receiverepeatedly during the active period. The active period of

Listen1

Control1

Node active period duration (�a)

Listen2Talk1

Control2

Talk2Node in set Si

To parent node in Set Si-1

To child nodes in Set Si+1

Fig. 2. Domino MAC active period structure.

a node in set Si is divided into two components: datacommunication between the node and its parents in setSi−1 followed by data communication between the nodeand its children in set Si+1. First component starts withthe first control phase (Control1) where control messagesarrive from the parent node. The parent node informs thechild node of information such as the ODFM subcarriersto use, the current values of the frame and the active

10

period duration. The second component starts with thesecond control phase (Control2). Note from the figure thatif there were transmission errors or collisions then thechild will retransmit the message in the next frame. TheBS and nodes in Sm have truncated phases as shown.

Figure 3 shows the Domino MAC overlappingsleep/active periods for the nodes in the set cluster Si.All nodes in the system follow a cascading or domino-like

Set

Time

BS

1

2

...

...

...

m

m-1

Frame Duration (�f)

Control phase

Talk phase

Listen phase

Fig. 3. Domino MAC overlapped sleep/active periods.

sleep/active periods where each active period is brokendown into sequential control, listen and talk phases.

The BS starts the frame by broadcasting a controlmessage to all its children in set S1 to setup the relevantcommunication parameters. BS then goes into Talk thenListen phases. The control message triggers the Listen1

and Talk1 phases in the child nodes in set S1. The nodesin S1 then broadcast control messages to their childrenin set S2 and switch to Talk2 and Listen2 phases. Thesame sequence of events is repeated by all nodes in thenetwork as shown.

3.1 MAC Data Multiplexing SchemesThe proposed MAC scheme uses a combination ofTDMA, OFDMA and TDD to achieve several goals

1) Collision-free operation to reduce wasted delayand energy during retransmissions.

2) Reduce the time of active period.3) Maintain equal active period duration for all nodes

for equal node lifetime.4) Support the traffic carried by the nodes in all sets.5) Allows smooth degradation. When a node senses

its battery charge is low, it transmits at lower powerand reduces the number of transmitted subcarriersto preserve the same transmission range.

The MAC protocol borrows OFDMA concepts from theIEEE 802.16 WiMax protocol [12], [13], [14]. Each nodeis allocated a subchannel to transmit its data. The allo-cation of OFDMA subchannel gives any node a unique,collision-free access to the medium. The restriction thatall nodes have equal-duration active periods implieswe can not allocate multiple TDMA slots for a nodesince this implies that some nodes will stay active forlonger periods. This restriction does not apply to BS since

it presumably has no energy restrictions. Thus we areforced de rigueur to use TDMA slots among all the nodesin set S1 only. The advantage now is that every nodein S1 can use all available K subcarriers to distributeamong its child nodes. All nodes in S2 — Sm must followone of the TDMA slots that belong to their parents.

The sink and the nodes in S1 set a lower bound onframe duration τf . Using Fig. 3, we obtain:

τf >1

2N1(m+ 1)τa (3)

where N1 = πρr2 is the number of nodes in S1.The node active period duration is determined using

Fig. 3 as:

τa = 2τc + 4λτfR (4)

where τc is the duration of the control phase and R isthe data modulation rate. Central to the above equationis the requirement that if any station has more data totransmit, it would use OFDM to increase the bandwidthor postpone its transmission instead of using more TDDslots.

Now every node in S1 has access to the full set of Ksubcarriers of the OFDMA physical channel. A node inS1 also knows the number of children associated withit. Let us assume that the number of child nodes in setS2 associated with parent node in S1 is N2. Dependingon the traffic load and battery power level in the childnodes, the parent node could schedule the child nodesaccording to the following options:

1) Postpone transmission of a node for the next suc-ceeding frames

2) Allocate 0 < k ≤ K subcarriers to each child nodeSuch subcarrier allocation scheme is “optimal” in thesense that it is based on local decisions and does notrequire global knowledge of the status of all the nodesin the network. It also accommodates the frame-to-frametraffic variations in the nodes.

Each child node could be connected to none, one, ormore parent node depending on its physical locationand transmission/reception range. There is an optimalbinding of children to parent nodes that ensures uniformtraffic loading distribution for a given physical distri-bution of the nodes. However, such optimal bindingresults after global knowledge of the network status.Practically, however, nodes bind to their parents usinglocal knowledge only. We follow this policy for ournumerical simulations. The following lemma discussescollisions due to transmissions from nodes in set S1.

Lemma 1: The nodes in set S1 can not collide whilecommunicating with the BS or their children in set S2.Proof: The sink allocates a TDMA scheme for the nodesin S1 and hence transmissions from these nodes do notoverlap and no collisions are possible.

The following lemma discusses collisions due to trans-missions from nodes in set Si belonging to the sameparent in set Si−1, where 1 < i ≤ m.

11

Lemma 2: The child nodes in set Si, where 2 ≤ i ≤ m,communicating with the same parent node in set Si−1

can not collide.Proof: The parent node ensures collision-free communi-cations among its children through two approaches:

1) Postpone transmission of a node for the next suc-ceeding frames

2) Allocate different subcarriers or CDMA sequencesto each child node

The following lemma discusses collisions due to trans-missions from nodes in set Si belonging to the differentparent nodes in set Si−1, where 1 < i ≤ m.

Lemma 3: The child nodes in set Si, where 2 ≤ i ≤ m,communicating with different parent nodes in set Si−1

can not collide.Proof: According to Lemmal 2, All parent nodes in set Si

are themselves child nodes for parent nodes in set Si−1.Carrying this relation all the way up to parent nodes inset S1, we have one of two situations:

1) The nodes ultimately belong to same parent in setS1

2) The nodes ultimately belong to different parents inset S1

The former case implies that the child nodes are allo-cated different OFDM subcarriers. The latter case impliesthat the child nodes are allocated different TDMA timeperiods In either case, no collision is possible.

3.2 Domino MAC Features

Several attractive/salient features of domino MAC areimmediately seen:

1) Domino MAC is collision-free.2) Domino MAC is distributed where control of any

node is relegated to its set of parent nodes.3) Domino MAC is adaptive where the duration of

talk/listen phases can change based on node bat-tery power and the later talk/listen phases willstart whenever the parent nodes indicates that.

4) Domino MAC could be contention-free orcontention-based. The parent node determineswhich mode it should employ based on thenumber of child nodes and the number of logicalchannels it can allocate to the nodes.

5) Domino MAC is self-stabilizing to changes in thenetwork [9]. A changes in the network, e.g. nodedeletion or addition, affect the nodes communicat-ing with the departing or arriving node.

6) The operation of the nodes does not rely on a globaltime reference. Synchronization of any node iscontrolled by the parent node. In fact, the durationof the active period and its Talk/Listen phases canbe locally and dynamically changed between pairsof communicating sets.

7) Each node receives messages during the listenphase and transmits these messages on the next

talk phase which occurs after one frame durationτf . This gives each node enough time to compressthe data of incoming messages and do the requiredprocessing tasks.

8) Network performance such as its throughput andlifetime can be changed dynamically when the sinkinforms nodes in set S1 of the upcoming frameperiod. All other nodes in the network will beprogressively informed of that duration in a rollingfashion.

9) Domino MAC allows for different chanellizationschemes. For example, random access could beaccomplished by making the nodes access themedium using any random subchannel.

4 SIMULATING OFDMA/TDD IN WIRELESSSENSOR NETWORKS

We used MATLAB R2009b for our numerical simulationand the nodes were defined using object-oriented designstrategy.

We simulated three attributes of Domino MAC:1) Node assignment to sets Si

2) Traffic distribution in the network3) OFDMA subcarrier assignment to the nodes

The parameters used are shown in Table 1

TABLE 1Parameters used for OFDMA/TDD WSN network

Parameter Symbol Values UnitsNumber of users N 200Network range L 50 mNode range r 5 mNode traffic λτf 1 kbConnectivity pc 0.998

Figure 4 illustrates the clustering of nodes into setsaccording to the Algorithm 1. For the given parameterswe had 12 sets in total (i.e. m = 12) with set S1 being theproximal set and S12 being the distal set. The ’X’ in eachfigure indicates the location of BS. Figure 4(a) shows theoriginal random distribution of the nodes. Figure 4(b)shows the nodes belonging to set S1. Note that thesenodes are confined to a circle centered at BS whoseradius is r. Figure 4(c) shows the nodes belonging toset S2. These nodes are confined to an annulus centeredat BS whose radii are r and 2r. Figure 4(d) shows thenodes belonging to set S3. We expect these nodes to beconfined to an annulus centered at BS whose radii are 2rand 3r. In fact, only few of the nodes fall in that region.The majority of the nodes of S3 fall in the same annulusas that of the nodes in S2. The reason for this is lack ofconnectivity between these nodes and nodes in set S1.Our clustering algorithm correctly classifies them into S3

since they require three hops to reach BS.Figure 5 shows minimum, average and maximum

number of children associated with the nodes in a set.

12

��� ��� � �� ��

���

���

��

��

��� ��� � �� ��

���

���

��

��

�����

(a) (b)

��� ��� � �� ��

���

���

��

��

������������

��� ��� � �� ��

���

���

��

��

�����������

(c) (d)

Fig. 4. Clustering of nodes in a WSN network usingDomino MAC. (a) Original random distribution. (b) Nodesin Set S1. (c) Nodes in Set S2. (d) Nodes in Set S3. L = 50,N = 200, and r = 5.

1 3 5 7 9 11 13

1

2

3

4

5

6

7

8

9

10

Set Index

Num

ber

of C

hild

ren

Fig. 5. Minimum, average and maximum number ofchildren associated with the nodes in each set of theWSN network. Upward-facing triangles represent maxi-mum number of children. Downward-facing triangles rep-resent minimum number of children. Solid circles repre-sent average number of children.

Figure 6 shows the minimum, maximum and averagetraffic carried by the nodes in each set for the case N =50, L = 20, r = 5 and λτf = 1. We had four sets intotal (i.e. m = 4) with set S1 being the proximal set andS4 being the distal set. We note that the average trafficcarried by each node increases for nodes closer to BS.

2 4 6 8 10 12 140

50

100

150

200

Set Index

For

war

ded

Tra

ffic

Fig. 6. Minimum, average and maximum forwarded trafficcarried by the nodes in each set of the WSN network.Upward-facing triangles represent maximum forwardedtraffic. Downward-facing triangles represent minimum for-warded traffic. Solid circles are average forwarded traffic.

REFERENCES[1] Zhen Tian, Dongfeng Yuan, and Quanquan Liang, “Energy

efficiency analysis of error control schemes in wireless sensornetworks,” in International Wireless Communications and MobileComputing Conference ( IWCMC ’08), 2008, pp. 401– 405.

[2] M. C. Vuran and I. F. Akyildiz, “Cross-layer analysis of er-ror control in wireless sensor networks,” in 3rd Annual IEEECommunications Society on Sensor and Ad Hoc Communicationsand Networks (SECON’06), 2006, vol. 2.

[3] W. Heinzelman, A. Chandrakashan, and H. Balakrishnan, “Anapplication-specific protocol architecture for wireless micro- sen-sor networks,” IEEE Trans. Wireless Comm., vol. 1, no. 4, pp.660670, 2002.

[4] F. Ingelrest, D.Simplot-Ryl, and I.Stojmenovic, “Targettransmis-sion radius over LMST for energy-efficient broadcast protocol inad hoc networks,” in IEEE Int. Conf. Comm., Paris, 2004.

[5] X. Wu, G. Chen, and S.K. Das, “Avoiding energy holes in wirelesssensor networks with nonuniform node distribution,” IEEE Trans.Par. Dist. Sys., vol. 19, no. 5, pp. 710–720, 2008.

[6] J. Lian, K. Naik, and G. B. Agnew, ,” International Journal ofDistributed Sensor Networks, vol. 2, pp. 121–145, 2006.

[7] V. Mhatre and C. Rosenberg, “Design guidelines for wirelesssensor networks: communication, clustering and aggregation,”Ad Hoc Networks, vol. 2, pp. 4563, 2004.

[8] S Olariu and I Stojmenovic, “Design guidelines for maximizinglifetime and avoiding energy holes in sensor networks withuniform distribution and uniform reporting,” in IEEE Int. Conf.Comp. Comm., Barcelona, SPAIN, 2006, pp. 2505–2516.

[9] C. Busch, M. Magdon-Ismail, F. Sivrikaya, and B. Yener,“Contention-free MAC protocols for wireless sensor networks,”in Ann. Conf. Dist. Comp., Oct 2004, pp. 245–259.

[10] W. Ye, J. Heidemann, and D. Estrin, “Medium access controlwith coordinated adaptive sleeping for wireless sensor networks,”IEEE/ACM Trans. Net., vol. 12, no. 3, pp. 493–506, 2004.

[11] I. Demircol, C. Ersoy, and F. Alagoz, “MAC protocols for wirelesssensor networks: A survey,” IEEE Comm. Mag., pp. 115–121, Apr.2006.

[12] “IEEE standard for local and metropolitan area networks part 16:Air interface for fixed broadband wireless access systems,” IEEE,, no. 802.16, 2004.

[13] C. Eklund, R.B. Marks, K.L. Stanwood, and S. Wang, “IEEEstandard 802.16: A technical overview of the wirelessMAN airinterface for broadband wireless access,” IEEE Comm. Mag., vol.40, pp. 98–107, June 2002.

[14] Y.-J. Choi, Suho Park, and Saewoong Bahk, “Multichannel ran-dom access in ofdma wireless networks,” Selected Areas inCommunications, IEEE Journal on, vol. 24, pp. 603–613, Mar 2006.

13