Cooperative communications with relay selection in wireless sensor networks

For Peer Review

Wirel. Commun. Mob. Comput. 00: 1–16 (2007)Published online in Wiley InterScience(www.interscience.wiley.com) DOI: 10.1002/wcm.0000

Cooperative communications with relay selection for wirelessnetworks: design issues and applications†

Xuedong Liang1 Min Chen2∗ Ilangko Balasingham3,4 Victor C.M. Leung1

Summary

Relay selection schemes for cooperative communications to achieve full cooperative diversity gains whilemaintaining spectral and energy efficiency have been extensively studied in recent research. These schemes selectonly the best relay from multiple relaying candidates to cooperate with a communication link. In this paper, wereview recently proposed cooperative communication protocols which integrate with relay selection mechanisms.The key design issues for relay selection mechanisms, e.g., relaying candidate selection, optimal relay assignment,and cooperative transmission, are identified. We further discuss the challenges of optimal relay assignment inmulti-hop wireless sensor networks, and present the potential applications of cooperative communications withrelay selection in such networks. Future research directions are outlined, e.g., the issues of service differentiationand system fairness in cooperative communication systems, and the joint use of game theory and machine learningtechniques in relaying candidate selection and optimal relay assignment mechanisms for efficient allocation ofnetwork resources. Copyright c© 2007 John Wiley & Sons, Ltd.

KEY WORDS: Cooperative communications; optimal relay assignment; quality of service; wireless sensornetworks.

1. Introduction

In recent years, cooperative communications [1, 2]have been proposed to exploit the spatial and timediversity gains in wireless networks by utilizingthe broadcast nature of the wireless medium.Users in cooperative communication systems workcooperatively by relaying data packets for each other,and thus forming multiple transmission paths orvirtual multiple-input-multiple-output systems to the

∗Correspondence to: Min Chen at [email protected]

destination without the need of multiple antennas ateach user [3].

A significant amount of work has been donein designing cooperative protocols which definethe relaying candidate selection [4], coding [5],cooperative transmission [6], and power allocationschemes. However, there lacks a comprehensivesurvey on the recent proposed cooperative protocols,especially for relay selection mechanisms that onlychoose optimal relays among multiple relayingcandidates to cooperate with communication links.In this survey, we aim at providing backgroundknowledge, potential applications, and key design

Copyright c© 2007 John Wiley & Sons, Ltd.Prepared using wcmauth.cls [Version: 2007/01/09 v1.00]

Page 2 of 19

http://mc.manuscriptcentral.com/wcm

Wireless Communications and Mobile Computing

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

For Peer Review

issues of cooperative communications, particularlyfor cooperative protocols that integrate with relayselection schemes.

In the literature, most of the research on cooperativecommunications [7, 8, 9, 10, 11] models the wirelesschannel as a narrow band Raleigh block fadingchannel with additive white Gaussian noise (AWGN)[12]. For any two nodes, e.g., i and j, the channelcoefficient hij which captures the effects of path-loss,shadowing, and fading, is modeled as a zero-meancircular symmetric complex Gaussian random variablewith the expectation of E(|hij |2) = 1. For the linksbetween a pair of nodes, the channel coefficients areassumed to be reciprocal, i.e., hij = hji. The channelcoefficients are constant for a given transmitted block,or a codeword, but are independent and identicallydistributed (i.i.d.) for different blocks [7]. For differentlinks, the channel fading coefficients are statisticallyi.i.d., which is a reasonable assumption as the nodesare usually spatially deployed [13].

source destination

relay

hs,d

hs,r hr,d

Fig. 1. A three node cooperative communication model

The concept of cooperative communications isillustrated in Fig. 1. The simple cooperative com-munication system consists of three nodes, thesource, relay and destination, which are all withineach other’s communication range. The cooperativecommunication protocol usually operates in twophases. In the first phase, i.e., the direct transmissionphase, the source transmits a message to thedestination. The signal received at the destination thattransmitted by the source is expressed as in (1),

Yd1= hs,dXs + ηs,d, (1)

where Xs is the information symbol transmitted bythe source, hs,d is the channel coefficient between thesource and the destination, and ηs,d is the additivenoise, which captures the effects of input noise at thereceiver and other interferences in the network and ismodeled as a zero-mean, circular symmetric, complexGaussian distribution with variance N0 [7].

The relay may overhear the message due to thebroadcast nature of the wireless medium. The signalreceived at the relay is expressed as in (2),

Yr = hs,rXs + ηs,r, (2)

where hs,r is the channel coefficient between thesource and the relay, and ηs,r is the additive noise.

In the second phase, i.e., the relaying transmissionphase, the relay can simply amplify and forward thereceived signal to the destination, or decode the signalfirst, then encode and forward the message to thedestination. The signal received at the destination thatretransmitted by the relay is expressed as in (3),

Yd2= hr,dR+ ηr,d = hr,df(Yr) + ηr,d, (3)

where R is the symbol transmitted by the relay, andR = f(Yr) is a function of the received signal Yr, hr,d

is the channel coefficient between the relay and thedestination, and ηr,d is the additive noise.

Thus, two paths, i.e, source-destination and source-relay-destination, are formed from the source tothe destination. The destination receives two copiesof the original signal, i.e., Yd1

and Yd2, which

are transmitted over the two independent paths andexperience different channel fading and shadowing.The destination may combine the signals, e.g.,applying maximum-ratio-combining [14] for optimalpacket decoding, or simply choose the signalwith a higher signal-to-noise-ratio (SNR) and thendecode it. Therefore, cooperative diversity gainscan be achieved, i.e., a packet transmission failureoccurs only when both of the two independentpaths experience deep channel fading or shadowingsimultaneously.

A variety of cooperative transmission schemes havebeen proposed to achieve cooperative diversity gainsand channel efficiency, which are described as follows.

• amplify-and-forward (AaF) [15]: the relayamplifies the received signal from the sourceand forwards it towards the destination.

• decode-and-forward (DaF) [15]: the relaydecodes the packet received from the sourceby either full decoding or symbol-by-symboldecoding, and then encodes and forwards thepacket to the destination.

• selection relaying [15]: whether a relay cooper-ates with a communication link or not dependson the measured metric |hs,r|2, which denotesthe channel gain between the source and therelay. If |hs,r|2 is lower than a certain threshold,the source simply continues its transmission tothe destination, in the form of repetition or more

Copyright c© 2007 John Wiley & Sons, Ltd.Prepared using wcmauth.cls

Wirel. Commun. Mob. Comput. 00: 1–16 (2007)DOI: 10.1002/wcm

Page 3 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

For Peer Review

powerful codes. If |hs,r|2 is higher than thethreshold, the relay forwards the signal receivedfrom the source, using either AaF or DaF, toachieve diversity gains. The calculation of thethreshold value depends on a number of factors,e.g., the source’s transmission power, data rate,channel bandwidth, and the noise level at thereceiver. More details on the calculation of thethreshold can be found in [15].

• incremental relaying [15]: to improve spectralefficiency, whether a relay cooperates witha communication link or not depends onthe feedback from the destination. If thefeedback indicates that the direct transmission issuccessful, the relay keeps idle. Otherwise, therelay retransmits the overheard signal from thesource towards the destination, either using AaFor DaF.

• coded cooperation [16]: the cooperation isintegrated with channel coding and works bysending different parts of each user’s codewordvia two independent fading paths. In codedcooperation, each user’s data is encoded into acodeword N that is partitioned into two parts,containing N1 bits and N2 bits, respectively.The first part of N1 bits itself is a valid codewordbut weakly coded, and the N2 bits in thesecond part are the puncture bits [1]. The codedcooperation operates in two phases. In the firstphase, each user transmits its first part of thecodeword which contains N1 bits, and alsodecodes its partner’s first part of the codewordcontaining N1 bits as well. In the second phase,a user calculates and transmits the second partof the codeword, i.e., the remaining N2 bits,for its partner, if it can decode the first partof codeword successfully; otherwise, the usertransmits its own second part of the codewordthat contains the remaining N2 bits. The basicidea of coded cooperation is that each usertries to transmit incremental redundancy for itspartners.

• CDMA-based cooperation [17, 18]: the coop-eration mechanism is implemented in a codedivision multiple access (CDMA) system, inwhich a user constructs a signal to transmitby combining its own signal and the signalreceived from its partner. Similarly, the user’spartner constructs its signal to transmit in asame fashion. Then, the user and its partnercooperate by sending both of their messages to

the receiver and use different spreading code toavoid interferences.

In wireless networks where nodes are denselydeployed, as shown in Fig. 2, there are usu-ally multiple relaying candidates available for thesource and destination. In a conventional multi-nodecooperative communication system, all the availablerelays actively participate in the communication byretransmitting signals towards the destination andthus form multiple paths between the source anddestination.

source destinationr2

r3

r1

ri

r5 rN

rMr4

hs,ri hri,d

...

Fig. 2. A multi-node cooperative communication model

The conventional multi-node cooperative commu-nication systems have the potentials of achieving fullcooperative diversity gains. For instance, for a pair ofsource and destination with N relays participating inthe cooperative communication, a packet transmissionfailure occurs only when all the N + 1 paths (source-destination plus source-relays-destination) experiencedeep channel fading or shadowing simultaneously.However, the channel efficiency of the multi-node cooperative communication system is muchlower than non-cooperative communication systems.The reason is that a total number of N + 1time slots is needed for the packet transmission,assuming carrier sense multiple access with collisionavoidance (CSMA/CA) or time division multipleaccess (TDMA) is used as the underlying mediumaccess control (MAC) protocol. Besides, packettransmissions suffer extra delays due to the receiverdeferring packet decoding until all of the relays havecompleted their transmissions. Moreover, the relays’multiple transmissions consume precious networkresources, e.g., spectrum, bandwidth, and energy,while increasing the probabilities of channel accesscontention and packet collision.

To achieve full cooperative diversity gains whilestill obtaining high channel efficiency and lowtransmission delay, selective single relay cooperativeschemes, in which only one optimal relay is selectedfrom multiple relaying candidates to cooperate with



Page 4 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

For Peer Review

the communication link, have been extensively studiedin recent research. A number of relay selectionmechanisms have been proposed for cooperativecommunication systems, in which various schemesand criteria are used in optimal relay assignment.

The rest of the paper is organized as fol-lows. The background information on cooperativecommunications with relay selection, key designissues of relay selection schemes, e.g., relayingcandidate selection, optimal relay assignment, andcooperative transmission, are identified in Section2. Section 3 compares and discusses the maindesign issues of optimal relay selection schemes.Cooperative communications in multi-hop wirelesssensor networks (WSNs) is further discussed inSection 4. In Section 5, we present the potentialapplications of cooperative communications withrelay selection in WSNs. Finally, we conclude thepaper and discuss future research directions in Section6.

2. Cooperative communications with relayselection: background and key design issues

Cooperative communications with single relay selec-tion schemes have been demonstrated to be effectivein achieving full cooperative diversity gains andobtaining high channel utilization efficiency. Asshown in [8, 10], in DaF cooperative scenarios, usingan optimal relay to cooperate with a communicationlink can achieve the same diversity gains asconventional cooperative protocols that employ all thepotential relaying candidates.

Adaptive relay selection schemes for cooperativeprotocols play important roles in cooperative com-munication systems, and have significant impacts ondiversity gains and network performance. The selectedoptimal relay should be the best one among all therelaying candidates, which can make the most ofcontributions in improving the network performance,in terms of packet outage probability and channelutilization efficiency [13, 15, 19]. The challenge ofoptimal relay selection is finding the best relay indynamic environments in wireless networks, wherethe network topology may change and the wirelessmedium is time-varying, due to the dynamic nature ofsuch networks.

In cooperative communication systems whereadaptive relay selection mechanisms are utilized,the cooperative protocols usually operate in threephases, namely, relaying candidate selection, relay

assignment, and cooperative transmission. We identifythe key design issues for adaptive relay selectionmechanisms and discuss them as follows.

2.1. Phase 1: relaying candidate selection

A number of nodes are determined as relayingcandidates for the communication link between thesource and destination in this phase.

2.1.1. Pre-assigned selection scheme

In pre-assigned selection scheme, the relayingcandidates are selected prior to data flow connection.

In [20, 21, 22], the relaying candidate selection isaccomplished by using the mechanism of cooperativemulti-hop mesh structure construction. The relayingcandidates are assigned in the procedure of multi-hop mesh route discovery and establishment. Thepre-assigned scheme is the simplest approach forrelaying candidate selection, in term of algorithmdesign complexity. However, the pre-assigned relayingcandidate selection scheme cannot deal with networkdynamics, e.g., node mobility, wireless channelvariation, and network topology changes, and thusdoes not fit in dynamic environments. Furthermore,the relaying candidate selection is based on thecooperative mesh structure, which is constructed fullyindependent from the data flow and thus incurssignificant communication overhead.

2.1.2. Adaptive selection scheme

Adaptive relaying candidate selection schemes aremore suited for cooperative communications overwireless networks than pre-assigned schemes, due tothe dynamic nature of such networks.

To reduce the communication overhead, adaptiverelaying candidate selection schemes often utilize thesignaling messages defined in the MAC protocol, orintegrate with the routing mechanism in the networklayer.

In [9, 10, 11], the signaling messages in the MAClayer, i.e., request-to-send (RTS) and clear-to-send(CTS) signals defined in the IEEE 802.11 standard[23], are used in relaying candidate selection. Thatis, when a node overhears a RTS signal from thesource and a CTS signal from the destination, the nodedetermines that it is a common neighboring node forboth of the source and destination, and could act as arelaying candidate.

Multiple-RTS (M-RTS) and multiple-CTS (M-CTS) signaling messages, extended from RTS and



Page 5 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

For Peer Review

CTS, are utilized in [24] to identify the relayingcandidates. When a node receives a M-RTS signalfrom the source, it considers itself to be a relayingcandidate, and will start the relay selection competingprocedure.

In [25], the selection of relaying candidates areimplemented by using Hello messages. By exchangingthe Hello messages, a set of nodes which are thecommon immediate neighboring nodes for both of thesource and destination are selected as the relayingcandidates.

In [26], the relaying candidate selection schemeis integrated with the route finding mechanism, i.e.,ad hoc on-demand distance vector (AODV) routingprotocol. In the route discovery procedure, for twoadjacent routers (1-hop sender and receiver), a nodedetermines that it is a relaying candidate for therouters, if it has heard both the route request (RREQ)signal transmitted by the sender and the route reply(RREP) replied by the receiver, and has not beenselected by the sender as the next hop router.

2.2. Phase 2: relay assignment

When a set of relaying candidates is selected, one ofthe relaying candidates should be chosen based onsome criteria to cooperate with the communicationlink between the source and destination. Thefrequently used criteria for relay assignment aredescribed as follows.

2.2.1. Pre-defined and random relay assign-ments

The simplest solutions for relay assignment areassigning the relays in advance, or choosing the relaysrandomly in runtime, as proposed in [7, 21]. The pre-defined and random schemes can reduce the designcomplexity and overhead of the schemes. However,such schemes cannot achieve optimal performance indynamic environments and lack of the capacity ofdealing with network dynamics.

2.2.2. Distance-based relay assignment

An intuitive scheme of optimal relay assignmentis using distance, either towards the source or thedestination, as the criterion of optimal relay selection.

The distance-based cooperative protocol [21]chooses a node which is the closest one to thedestination as the optimal relay. In [24], whena candidate starts the relay selection competingprocedure, it first sets a backoff timer with a value

proportional to the candidate’s distance to the source.Thus, the candidate node, which is the closest oneto the source, will expire its backoff timer first andwin the competition. The node which wins in thecompetition procedure will send a M-CTS signal toinform the source that it is ready to send, as well asnotifying other candidates to cancel their competingprocedures.

However, it is well understood that communicationsbetween senders and receivers with similar distancesmay have significant differences in terms of receivedsignals’ SNRs, due to interference, shadowing andmulti-path fading effects on the wireless links.Therefore, the use of distance as the criterion ofrelay assignment cannot reflect the channel stateappropriately.

2.2.3. SNR (channel gain)-based relay assign-ment

The most intuitive solution of optimal relay assign-ment is to choose the relay that has the highest SNRs,or the maximum wireless channel gains with both thesource and destination. Since both of the two paths,i.e., source-relay and relay-destination, are importantfor the end-to-end performance, each relay shouldevaluate the link qualities of both paths.

In [25], for a link between the source s anddestination d, the source considers the immediateneighboring nodes of both s and d as relayingcandidates. The source maintains a list which containsthe candidates’ MAC addresses and their link qualitiesto s and d. The source uses a metric γr, as shown in(4),

γr = min (SNR(s, ri), SNR(ri, d)) , (4)

to sort the candidates, where SNR(s, ri) andSNR(ri, d) denote the SNR between the link of s-ri and ri-d, respectively. Then,the source selects thetwo candidates which have the two highest of theminimum SNR of the relay channels, i.e., from s tothe relay, and from the relay to d, as the optimal relays.Each candidate measures the SNRs locally and sendsthe information to the source in a fixed time interval(every one second in the paper), so the source canalways have the up-to-date information on the list. Theperformance evaluation has shown that the selectedrelays have high link qualities with both s and d basedon computer simulations.

In [7], the source chooses N relays, whosereceived signals’ SNRs are the N highest among



Page 6 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

For Peer Review

all the relaying candidates, to cooperate withthe communication link between the source anddestination.

In [9], the relaying candidates use RTS and CTSmessages to assess the link qualities of the source-relay and relay-destination. The transmission of theRTS from the source allows for the estimationof the instantaneous wireless channel coefficienths,ri between the source and the relay ri; thetransmission of the CTS from the destination allowsfor the estimation of the instantaneous wirelesschannel coefficient hri,d between the relay ri and thedestination.

The channel coefficients hs,ri , hri,d at each relay,describe the quality of the wireless path betweensource-relay-destination for the relay. Each relayassesses the link qualities between the source-relay-destination, and uses the following policies todetermine the best relaying candidate distributedly.Under policy I, as defined in (5), the minimum of thetwo is selected, while under policy II, as defined in (6),the harmonic mean of the two is chosen.

• Policy I:

Hri = min{|hs,ri |

2, |hri,d|

2}

(5)

• Policy II:

Hri =2

1

|hs,ri |2+ 1

|hri,d|2=

2 |hs,ri |2 |hri,d|

2

|hs,ri |2+ |hri,d|

2

(6)

After receiving the CTS, each relay will start atimer with an initial value of Ti, which is set inverselyproportional to the end-to-end channel quality Hri .Therefore, the best relay’s timer will expire first andstart retransmitting the signal towards the destination.

The optimal relay assignment scheme [11] isintegrated with the power control mechanism inphysical layer. The relaying candidates use RTS andCTS messages to assess the link qualities and computeindividually the required transmission power that canmeet the desired link qualities. Different from [9], thesource also participate in the competition procedure, ifit believes that it has the potential of being an optimalrelay.

The authors in [8] proposed an adaptive relayselection scheme for cooperative communicationprotocols, based on the channel state information(CSI) at the source and the relays. The optimal relay is

the node which has the maximum instantaneous scaledharmonic mean function of its source-relay and relay-destination channel gains.

The relay’s metric βm, as defined in (7), denotesthe scaled harmonic mean function and gives aninstantaneous indication about the relay’s ability tocooperate with the source.

βm = μH

(q1 |hs,ri |

2, q2 |hri,d|

2)

=2q1q2 |hs,ri |

2 |hri,d|2

q1 |hri,d|2+ q2 |hs,ri |

2 (7)

μH(.) is the standard harmonic mean function. q1and q2 can be calculated as in (8).

q1 =A2

r2, q2 =

B

r (1− r)(8)

r = P1

P1+P2is the power ratio, where P1 and P2

are the transmission power of the source and relay,respectively.

For multiple phase shift Keying modulation, A andB can be calculated as in (9) and (10), respectively.

A =1

π

∫ (M−1)πM

0

sin2 θdθ =(M − 1)

2M+

sin( 2πM )

4π(9)

B =1

π

∫ (M−1)πM

0

sin4 θdθ =3(M − 1)

8M

+sin( 2πM )

4π−

sin( 4πM )

32π(10)

The relay whose metric is equal to β∗ =max {β1, β2, . . . , βM}, is chosen as the optimal relay.

The signaling messages, RTS and CTS, are alsoutilized in the single relay selection scheme to assesslink quality under aggregate power constraints [10].For DaF protocol with reactive relay selection, theoptimal relay is the candidate which has the maximuminstantaneous channel gain between the relay anddestination, as shown in (11).

r∗i = argmaxri∈D

|hri,d|2 (11)

where r∗i is the optimal relay, D is the set of relayswhich can decode the received message successfullyduring the source-destination transmission phase.



Page 7 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

For Peer Review

For DaF protocol with proactive relay selection,the best relay is chosen prior to the transmissionsof source-destination and relays-destination. The bestrelay is the candidate that can maximize the minimumof the weighted channel strengths between the sourceand destination, as shown in (12).

r∗i = arg maxri∈K

min{ζ |hs,ri |

2, (1− ζ) |hri,d|

2}(12)

where K is the set of relaying candidates, ζ and(1− ζ) ∈ [0, 1] denote the fractions of the total powerallocated to the source transmission and overall relaytransmission, respectively.

For AaF protocol with proactive relay selection, theoptimal relay is the candidate which can maximize themutual information, defined as follows.

r∗i = argmaxWrii∈K

(13)

where

Wri =|hs,ri |

2 |hri,d|2

ζ1−ζ (1 +

1γsri

)Ωsri + |hri,d|2 (14)

where Ωsri denotes the average channel gainbetween the source and relay ri, γsri is the averageSNR for the link between the source and relay ri.

2.2.4. Game theory-based relay assignment

Game theory provides a set of tools for modelingthe interactions of a finite number of decision-makerswhose actions often have mutual effects on each other.Thus, game theory is inherently suitable for modelingand analyzing wireless networks, where nodes sharelimited network resources, e.g., spectrum, time slots,both cooperate and/or contend with each other, andtheir actions have different outcomes.

There are a number of game theory basedschemes in the literature that model the process ofoptimal relay selection as a non-cooperative game,in which the relaying candidates are modeled asplayers. Each candidate plays the game against theother candidates in a distributed manner. Whether acandidate cooperates with a communication link or notdepends on the payoff it may achieve, which is usuallydefined as the difference between the benefit and costof acting as a relay.

In most of the game theory based schemes, thenodes are modeled as rational players, which means

the nodes are expected to follow a set of strategies andchoose actions from the strategies to maximize theirutilities. In the meantime, the nodes behaves selfishly,i.e., a node always chooses action to maximize itsown payoff, without considering the utilities of othernodes. Most of the papers do not consider cooperativegames, as additional signaling messages between thedecision-makers needed to be implemented to achievecommon agreements, which make it more difficult torealize in wireless networks.

In [27], game theory is used to model a wirelessnetwork consisting of selfish nodes, wherein a credit-based micro-economical model involving exchangeof virtual currencies is proposed to manage nodeinteractions. Whether a node involves a cooperativelink depends on the credit that can be earned andthe resource needed for relaying packet. The authorsin [28] proposes a relay selection and power controlscheme based on a two-level Stackelberg game, inwhich the source node and relay nodes are modeledas buyer and sellers, respectively. The proposedscheme jointly considers the benefits of both thebuyer and the sellers and can achieve the best systemperformance with minimum power consumption. [29]presents a game-theoretic analysis of DaF cooperativecommunications over AWGN and Rayleigh fadingchannels. The analysis shows that a mutually NashEquilibrium exists if proper power control is utilizedand users care about their long-term performance.

The game theory based schemes in the literatureoften assume that players has complete informationof the game, i.e., a player has the full knowledge ofthe game, i.e., the other players’ identities, strategies,payoffs, and utility functions [28]. Furthermore, thegame’s history, e.g., the actions of each player inprevious stages, is also assumed to be known to allplayers in a multi-stage game [27]. However, thisassumption is not always hold in realistic scenarios,as nodes in wireless networks usually only havelocally observed information and limited knowledgesof others’ behavior. Therefore, adaptive learning,e.g., estimating payoffs may obtain by taking certainactions, predicting the other players’ strategies, shouldbe involved in the game designs [30].

2.2.5. Reinforcement learning-based relayassignment

Reinforcement learning [31, 32] provides a frameworkin which an agent can learn control policies in dynamicenvironment based on experiences and rewards. Avariety of reinforcement learning algorithms haves



Page 8 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

For Peer Review

been implemented for nodes to learn adaptive policiessuch as protocols for MAC [33], routing [34, 35,36] layers, QoS provisioning [37], and cooperativecommunications [21, 22, 26, 38, 39].

In [26], reinforcement learning is used to learn theoptimal policy of relay selection in multi-hop wirelesssensor networks. For a pair of sender and receiveralong the established route, the sender chooses a relayfrom all the potential relays to cooperate with a linkbetween the sender and receiver in an arbitrary mannerin the beginning of the procedure of optimal relayselection, as the sender does not have knowledge of thechannel gains between the sender-relays and relays-destination. After the selected relay’s transmission,the sender evaluates the quality of relay selectionbased on the feedback from the receiver, e.g., theSNR improvement on the link. After a series of trial-and-error interactions with the network, the sendercan learn an optimal policy of relay selection, andtend to choose a relay which can make the biggestimprovement on the link quality in a long-term run. Toavoid being stuck in a sub-optimal solution, the relayselection algorithm uses the ε greedy method [31] toexplore the environment with a certain probability.That is, with the probability of 1-ε, the sender choosesa relay that is expected to be able to make thebiggest improvement on the link quality. And withthe probability of ε, the sender randomly chooses arelay to cooperate with the communication link (ε isset to 0.1 in the algorithm). By doing so, the sendercontinues exploring the dynamic environment.

2.3. Phase 3: Cooperative transmission

The relays can cooperate with the communication linkbetween the source and destination for any packettransmission by retransmitting the overheard signals.For instance, in CRP [25], two selected optimal relaysretransmit each message overheard from the sourcetowards the destination. Therefore, the destinationmay receive three copies of the signals for any packettransmitted from the source.

To reduce network overhead and increase channelefficiency, the cooperative communication can betriggered when it is necessary, e.g., the packettransmission fails in the direct transmission phase,or the direct link between the source and destinationcannot meet the desired QoS requirements, as theprotocol proposed in [26].

In [13], a new automatic repeat request (ARQ)mechanism is introduced in the cooperative com-munication protocol. That is, if the destination can

successfully decode the signal transmitted by thesource in the direct transmission phase, it sends backan acknowledgment (ACK) and the relay keeps idle.Otherwise, if the destination cannot decode the signalsuccessfully, it sends back a negative acknowledgment(NACK). In the latter case, cooperative transmissionwill be invoked, i.e., the relay which received thesignal in the direct transmission phase forwards thesignal to the destination.

In [8], the source computes the ratio βs,d

βmaxand

compares it to a cooperation threshold α, which isoften defined as 1. If the βs,d

βmax≥ α, then the source

use direct transmission only; otherwise, if βs,d

βmax< α,

the source will choose an optimal relay to retransmitthe signal. The mechanism can be interpreted as thatthe source s will pick up a relay ri to cooperatewith the communication link between the source sand destination d, if the link quality that definedas the modified harmonic mean function of channelgains between s-ri-d is higher than the channel gainbetween s-d. Otherwise, the source will choose directtransmission only.

3. Comparison and discussion

In this section, we categorize recent proposedmechanisms of relay selection and list their featuresin Table I, then we compare and discuss the maindesign issues of optimal relay selection for cooperativecommunication protocols.

3.1. Optimal relay selection criterion

In the literature, most of the relay selection schemesfor cooperative communications utilize SNR orchannel gain and its variations as the unique criterionfor optimal relay assignment, and assume that full orpartial CSI is available at the source, destination andall of the potential relays. However, the use of SNRas the unique relay selection criterion is not sufficientin dynamic wireless networks. It has been shown in[7] that received SNR based selection scheme behavessimilarly to random selection scheme, or even slightlyworse in some scenarios. Furthermore, significantcommunication overhead is incurred in acquiringand disseminating of CSI to all of the cooperativeparticipants, especially for the cooperative protocols,as in [8, 9], that instantaneous CSI is required at allthe potential relays for relay selection.



Page 9 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

For Peer Review

Table I. Relay selection schemes for cooperative communications

Relaying candidate Optimal relay Cooperative Relayselection assignment criterion transmission scheme selection

Random selection [7] N/A (1-hop network) Random Continuous ReactiveReceived SNR selection [7] N/A (1-hop network) SNR Continuous ReactiveFixed priority selection [7] N/A (1-hop network) Pre-defined Continuous ReactiveScheme in [8] N/A (1-hop network) Weighted channel gain Triggered/threshold ReactiveOpportunistic DaF-1 [10] RTS and CTS Channel gain Continuous ReactiveOpportunistic DaF-2 [10] RTS and CTS Weighted channel gain Continuous ProactiveOpportunistic AaF [10] RTS and CTS Mutual information Continuous ProactiveEECC [11] RTS and CTS Channel gain Continuous ReactiveMISO [24] M-RTS and M-CTS Distance Continuous ProactiveCRP [25] Hello message SNR Continuous ReactiveQoS-RSCC [26] Route finding mechanism Outage probability, Triggered/ARQ Reactive

channel efficiencyGame theoretic scheme [27] N/A (1-hop network) Payoff (benefit minus cost) Continuous Reactive

3.2. Centralized vs. distributed mechanisms

In centralized mechanisms of relay selection, acoordinator, usually the source, is responsible to selectthe relay. As the mechanisms in [8], each relayingcandidate measures its channel gains of source-relayand relay-destination, calculates their harmonic meanfunctions, and sends this metric to the source. Then,the source selects a relay which has the highest metricamong all the relaying candidates and broadcasts acontrol signal to all the relays and the destinationto indicates its decision of relay assignment. Similarprocedures can also be found in [25].

In distributed mechanisms, relay-competition-timeris often used to decide which candidate should actas the optimal relay. In the mechanisms proposedin [9, 11, 24], each relaying candidate measuresits source-relay and relay-destination link qualities,e.g. using channel gain as a metric, and thenstarts a relay-competition-timer which is set inverselyproportional to the metric. The candidate with thehighest metric will have its timer reduced to zerofirst, as its timer is set with the lowest value. Then,the candidate will broadcasts a flag message to theother candidates informing that it wins in the relay-competition procedure, and will cooperate with thecommunication link.

Most of the game theory based approaches arealso of distributed mechanisms. For instance, as in[27], a relaying candidate estimates its opponent’spossible strategy, e.g., cooperating or remaining silent,and then take the best response to its opponent’sstrategy. That is, the candidate will cooperate withthe communication link, if it estimates that theopponent’s probability of cooperating is lower than athreshold; otherwise, the candidate will remain silent.

The threshold is defined as the cooperating probabilityof candidates at the Nash Equilibrium, which can beinterpreted as a steady network state in the context ofwireless networks, as none of the nodes will intend todeviate from the strategy profile to increase its payoff[40].

Centralized mechanisms often need exchangesof additional signaling messages to perform relayselection that inevitably introduces overhead to thenetwork. Furthermore, the need of centralized controllimits the scalability of such mechanisms, especially inmulti-hop wireless networks. In contrast, informationexchange is usually not necessary in distributedmechanisms, and a relaying candidate decides whetherto cooperate with a communication link or not ina distributed manner. Distributed mechanisms basedalgorithms often scale well in large-scale networks,however, it may happen that more than one candidatedecide to cooperate with a communication link whichwill result in a packet collision, or no candidatechooses to act as the optimal relay. For instance, in[11], if two candidates set their relay-competition-timer with a same value, both of them will reduce thetimer to zero and start retransmitting simultaneouslyand a packet collision happens. And in [27], ifthe candidates make inaccurate estimations of theiropponents’ strategies, relay selection failure, e.g.,collision or no candidate decides to act as the optimalrelay, may happen.

3.3. Proactive vs. reactive relay assignment

The relay can be assigned prior to source-destinationtransmission, which is called proactive relay assign-ment; or selected after source-destination transmis-sion, named as reactive relay assignment [10].



Page 10 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

For Peer Review

Proactive relay assignment has the advantage ofenergy efficiency, because only the selected relayneeds to be in the receiving mode during the source’stransmission, and the unselected relaying candidatescan switch to power-down mode for energy saving.Furthermore, proactive relay assignment schemessimplify the algorithm deign and overall networkoperations, as well as reducing the probabilityof channel access contention. However, proactiverelay selection schemes cannot guarantee optimalperformance in dynamic environments. As shown in[8], to achieve full diversity gains, the best relay mustbe chosen at each time instant of packet transmissionbetween source-destination. In proactive assignmentmechanisms, assigning a relay before the start oftransmission cannot ensure that the relay is the realbest one, as the wireless channel is assumed to bevarying over time.

3.4. Continuous cooperation vs. triggeredcooperation

Continuous cooperation schemes can guarantee diver-sity gains by always choosing relays to participate inthe communication. However, due to the shared andcontention nature of the wireless medium, the use ofcontinuous cooperation increases the probabilities ofchannel access contention and packet collision, andthus leads to low channel utilization efficiency.

In triggered cooperation mechanisms, feedback,e.g., ACK or NACK messages from the destination isoften used to trigger the cooperative communication.That is, an ACK message sent by the destinationindicates that the source-destination transmissionis successful; and a NACK message will invokea cooperative transmission, and a relay, eitherselected by the source or determined distributedlyby the relaying candidates, will cooperate with thecommunication link between the source-destination.

Triggered cooperation has the advantage of spec-tral efficiency, because the relaying transmissionis invoked only when the direct link betweensource-destination experiences deep channel fading,shadowing, or interference. However, the use oftriggered cooperative scheme increases the coopera-tive algorithm design complexity and computationaloverhead, as signaling messages are needed to indicatewhether the direct transmission between the sourceand destination is successful or not. Trade-off shouldbe considered between network performance andalgorithm complexity [41, 42, 43].

3.5. Impact of relaying candidate’s state oncooperative communications

Although the relaying candidates in the wirelessnetworks are assumed to be functionally equivalent(in terms of radio communications and signalprocessing), their individual states, e.g., incomingtraffic, duty cycle, and processing and queuingdelays may vary. Relaying candidates’ states couldhave significant impacts on the performance ofcooperative communications, and should be taken intoaccount in the optimal relay selection. Intuitively,it should be avoided that choosing nodes whichhave already involved in other data flows, e.g.,acting as intermediate routers or relays. The reasonis that assigning too many tasks will pose heavycomputational and communication burden on thenodes, which may become bottlenecks of the network,as well as resulting in severe network trafficimbalance.

Only a few papers in the literature consider node’sstate as a metric of relay selection, as these parametersare difficult to measure or even estimate. Rein-forcement learning could be a promising approachto address this issue, as the reinforcement learningbased approaches [26, 38] choose the relays throughexperience and rewards without actual measuring thenodes states, which is similar to the procedure ofhuman or animal learning in a dynamic environmentfrom scratch. That is, in the beginning of the procedureof relay selection, relays are selected randomly.After a series of trial-and-error interactions, theoptimal selection can be strengthened and sub-optimalselections are weakened by utilizing the reinforcementlearning algorithm.

3.6. Applicabilities of relay selection schemes inwireless networks

To apply relay selection schemes in dynamic wirelessnetworks, the schemes should have the capabilityof dealing with network dynamics, e.g., networktopology changes, varying wireless link qualities.

Usually, pre-defined relay selection schemes indynamic networks do not work well, as the fixedassignment of relays cannot adapt to dynamicnetworks. Distance based schemes cannot guaranteethat the selected relays are the optimal ones, asdistance is not the only factor that has effect ona communication link, and other factors, such asinterference, shadowing, and fading also affect thelink qualities. Furthermore, distance based schemesrequire that the distance information, i.e., from the



Page 11 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

For Peer Review

source to relays, or from relays to the destination,is available at each relay to make decision of relayselection, which is more difficult to implement indynamic environments. In SNR (channel gain) basedrelay selection schemes, optimal relays are chosenby measuring the SNRs of the signaling messagesthat transmitted prior to the transmission of datapackets. To adapt to the varying wireless channel,the measurements are often conducted on a packet-by-packet basis, which inevitably introduces a highcommunication overhead. Reinforcement learning andgame theory based schemes might be the mostadaptive schemes, as prior knowledge of networkmodel and link qualities are not necessary, and thepolicy of relay selection are cooperatively learnedvia a series of trial-and-error interactions by therelaying candidates. However, the convergence speedsof the relay assignment algorithms might limit theapplicabilities of such schemes. The reason is thatrelaying candidates, regarded as agents, needs acertain number of interactions to learn the optimalpolicy and then jointly adjust their behavior to achievea system level optimal performance. It is shown in[27, 38, 44] that the number of interactions, oftenbetween 20 and 50, depending on the network scale,topology, and channel variations, is needed for thealgorithms to reach convergence. Compared with theabove mentioned adaptive relay selection schemes,random relay selection schemes have the advantagesof lower computation and communication overhead,and still can achieve a moderate network performance[7].

Another important factor needs to be consideredis the cost of utilizing cooperative communications.As we know, cooperative communications is effectivein improving the network performance in terms oftransmission reliability, robustness, adaptivity, net-work throughput and lifetime, by exploiting the spatialdiversity of the wireless medium. However, the useof cooperative communications also associates withcertain costs because of nodes conducting extra tasksof signal processing, packet receiving and retransmit-ting. Furthermore, using cooperative communications,particularly for cooperative communication protocolsintegrated with relay selection schemes, also increasesnetwork operations, as optimal relays need to assignedeither by a coordinator or in a distributed manneron a packet-by-packet basis. Therefore, both thebenefits and cost should be considered when designingcooperative communication systems.

4. Cooperative communications in multi-hopWSNs

In multi-hop WSNs, as shown in Fig. 3, thecommunication between a source and its destinationusually involves a number of nodes, which act asintermediate routers and establish a multi-hop routefor packet transmission.

rc1

rcircN

ml

relaying candidates RCs={rci}, i= 1...N

destinationrc4

rc3

rc2

source

two adjacent routers: l, m

...

Fig. 3. Cooperative communications in multi-hop WSNs

Due to the lack of centralized control in multi-hop WSNs, the cooperative communication protocolshould work in a distributed manner, i.e., for eachhop of the route, the cooperative partner assignmentand cooperative transmission scheme should bedetermined locally without the need of global networkstate information.

In multi-hop WSNs, nodes may play multipleroles, e.g., source, destination, relays and intermediaterouters. For instance, nodes which are selectedas optimal relays for a data flow, may also bechosen as intermediate routers, or relaying candidatesfor other data flows. Therefore, nodes may havevarying incoming traffic, computational burden, andprocessing and queuing delays. This feature increasesthe design and analysis complexity of cooperativecommunication protocols.

Furthermore, there are often multiple data flowscoexist in the network, which may have impactson the performance of the source-destination dataflow, because the network resources, e.g., spectrum,bandwidth and energy, are shared by all nodesin the network. Thus, it is necessary to jointlyconsider cooperative communications with networkoptimization [45, 46].

5. Applications of cooperativecommunications with relay selection

The main benefit of applying cooperative com-munications in wireless networks is to achieve



Page 12 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

For Peer Review

diversity gains, without the need of maintainingmultiple antennas at each user. Moreover, spectralefficiency is still guaranteed by employing adaptiverelay selection techniques. Therefore, cooperativecommunications with relay selection may findvarious potential applications, especially in resource-constrained WSNs.

5.1. Reliable and energy-efficient datadissemination

In mission-critical WSN applications, e.g., battlefieldsurveillance, medical care and disaster response, thenetwork used for communication must ensure that datapackets can be delivered to the data processing centerreliably and efficiently.

Multi-path routing has been proposed for reliabledata dissemination, in which important data packetsare delivered to the destination through multiplepaths to achieve fault-tolerance. However, signifi-cant computational and communication overhead isincurred in the multi-path routes establishment anddata transmitting. Besides, the energy consumptionof multi-path routing is much higher than uni-pathrouting, due to the redundant packet transmissionsin multiple paths. Cooperative communications withrelay selection can be an effective approach forreliable and energy-efficient data dissemination inWSNs, by exploiting the spatial diversity gains, i.e.,choosing nodes to help in the packet delivering in casedeep channel fading, shadowing or interference occursin the multi-hop route from the source to its destination[20, 21].

5.2. QoS provisioning in WSNs

Due to low-cost node platforms, self-organizingmanner and ease of deployment, WSNs have numer-ous potential applications, e.g., medical care, battle-field surveillance, wildlife monitoring, and disasterresponse. In these mission-critical applications, a setof quality of service (QoS) requirements, e.g., delay,packet delivery ratio, network lifetime, throughput andcommunication bandwidth, on network performancesmust be satisfied [47]. However, providing guaranteedQoS is almost impossible in dynamic WSNs [48, 49],due to the dynamic network topology, time-varyingwireless medium, and severe constraints on powersupply, computation power, and communicationbandwidth [50, 51, 52, 53, 54, 55].

Thus, it is more practical to provide soft QoS[56] than guaranteeing hard QoS in multi-hop WSNs

[48, 49]. In soft QoS provisioning, when a QoS-support route is established, and the data flow is intransmission, there may exist a transient amount oftime that the QoS requirements cannot be met. Thelevel of soft QoS provisioning can be quantified bythe fraction of total disruption time over the totalconnection time. The ratio should not be higher thana threshold, which is determined by user applications.

For a QoS-support route, QoS violations may occurbecause the intermediate routers cannot fulfill the QoSattributes, that they have been assigned or promised inthe QoS route discovery and establishment procedure,which might be caused by network topology change,concurrent transmission interferences, thermal noise,shadowing and multi-path fading. For instance, asshown in Fig. 3, for the two adjacent routers l and m,which are the immediate routers along the establishedroute, the link between l and m may experiencechannel fading and thus cannot meet the assigned QoSattributes. Retransmitting the packet, e.g., using ARQmechanism, from l to m might not be effective in thiscase, since the link between l and m may remain indeep fading or shadowing for a long period in a slowlyvarying channel [13].

The channel fading and shadowing for differentlinks are assumed to be statistically independentin WSNs, because the nodes in WSNs are usuallyspatially well separated [57]. Therefore, there mightexist a node, e.g., node rci, which is a neighboringnode for both l and m, overhear the packettransmission between l and m, due to the broadcastnature of the wireless medium. Node rci mayhelp in the packet delivering between l and m

by retransmitting the packet to m, even it hasnot been assigned any routing task in the routediscovery and establishment procedure. This is knownas spatial diversity gain and has been demonstratedto be effective in improving network performance.In the context of cooperative communications, theneighboring node rci acts as a cooperative partnerfor the communication between the intermediaterouters l and m. When QoS violations happen, thecooperative partners may help in the packet deliveringby retransmitting the signals and thus reassuring theQoS attributes. Therefore, the transient amount oftime of QoS violation can be minimized and thesatisfied level of soft QoS provisioning is increased, byapplying cooperative communications with adaptiverelay selection in WSNs.



Page 13 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

For Peer Review

6. Conclusions and future researchdirections

In the paper, we have reviewed the relay selectionschemes for cooperative communication protocols,identified the key design issues for the adaptiverelay selection schemes, and discussed the potentialapplications of cooperative communications withrelay selection in WSNs. Compared with conventionalcooperative communication protocols, cooperativeprotocols integrated with adaptive relay selectioncan be more effective in improving the networkperformance by exploiting diversity gains, whilestill achieving channel efficiency. However, the useof cooperative communications with relay selectionincurs computational and communication overhead, aswell as increasing the design and analysis complexityof cooperative communication systems. Depending onuser applications and QoS demands, trade-off shouldbe made in designing cooperative communicationsystems to achieve optimal network performance.

In future research, service differentiation andsystem fairness could be interesting topics in thedevelopment of cooperative protocols. The distribu-tion of relaying tasks are important in resource-constrained WSNs, where multiple data flows coexist.In most of the current research, it has been assumedthat sensor nodes are passive in the sense thatthey are chosen passively by the source node(s) asoptimal relays, without consideration of their taskpriorities and willingnesses of being relays. In orderto provide differentiated network services, achievesystem fairness, and prolong the network lifetime, thetrade-off between task priority and system fairnessshould be further investigated. Moreover, the useof game theory can be promising in the design ofcooperative communication protocols for WSNs [58].In particular, the process of optimal relay assignmentcan be modeled as a mixed-strategy game, whereeach player plays the game with other players in adistributed manner. By taking different actions, e.g.,relaying a packet or remaining silent, and calculatingthe benefit and cost achieved, each player can evaluatethe actions’ qualities and then adjusts its probabilitiesof taking different actions in a dynamic environment.Optimal network performance can be achieved byby encouraging cooperation and discouraging selfishbehavior, e.g., using the Pricing mechanism [40]to regulate selfish players strategies. To solve theproblem that the available information in a game isoften incomplete and inaccurate, as well as adapting tothe dynamic environments, learning algorithms, e.g.,

fictitious learning, reinforcement learning should beinvolved in game designs. [27, 30].

Acknowledgement

This research was supported by the Canadian NaturalSciences and Engineering Research Council undergrant RGPIN 44286-09. The work was also supportedin part by The NAP of Korea Research Councilof Fundamental Science & Technology; The MKE(The Ministry of Knowledge Economy), Korea, underthe ITRC (Information Technology Research Center)support program supervised by the NIPA (NationalIT Industry Promotion Agency) (NIPA-2010-(C1090-1011-0004)). Part of the research is in the contextof the EU project IST-33826 CREDO: Modelingand analysis of evolutionary structures for distributedservices (http://www.cwi.nl/projects/credo/).

References

1. A. Nosratinia, T.E., Hunter, and A. Hedayat. Cooperativecommunication in wireless networks. IEEE CommunicationsMagazine, 2004, vol. 42, 74-80.

2. Yao-Win Hong, Wan-Jen Huang, Fu-Hsuan Chiu and C.-C. JayKuo. Cooperative Communications in Resource-ConstrainedWireless Networks. IEEE Signal Processing Magazine, 2007,vol. 42, 47-57.

3. Xia Chen, Honglin Hu, Haifeng Wang, Hsiao-Hwa Chen andMohsen Guizani. Double Proportional Fair User Pairing Algo-rithm for Uplink Virtual MIMO Systems. IEEE TransactionsOn Wireless Communications, 2008, vol. 7, no. 7, 2425-2429.

4. Xia Chen, Honglin Hu, Shengyao Jin, and Hsiao-HwaChen. Broadcast Reserved Opportunity To Achieve DiversityRelaying And Its Performance Evaluation. EURASIP JournalOn Wireless Communications And Networking, vol. 7, no. 2,2008

5. Mugen Peng, Jie Zhang, Wenbo Wang and Hsiao-Hwa Chen.Heterogeneous cooperative relay selection with maximal-ratiocombining for multi-radio access networks. InternationalJournal of Communication Systems, 2010, vol. 23, no. 6-7, 732-750.

6. Lie-Liang Yang and Hsiao-Hwa Chen. Error probability ofdigital communications using relay diversity over Nakagami-mfading channels. IEEE Transactions on Wireless Communica-tions, 2008, vol. 7, no. 5, 1806-1811.

7. Todd E. Hunter and Aria Nosratinia. Distributed Protocolsfor User Cooperation in Multi-User Wireless Networks. TheIEEE 47th annual Global Telecommunications Conference(GLOBECOM’04), 2004, 3788-3792.

8. A.S Ibrahim, A.K. Sadek, Weifeng Su and K.J.R Liu.Cooperative communications with relay-selection: when tocooperate and whom to cooperate with? IEEE Transactions onWireless Communications, 2008, vol. 7, no. 7, 2814-2817.

9. Aggelos Bletsas, Ashish Khisti, David P. Reed and AndrewLippman. A Simple Cooperative Diversity Method Based onNetwork Path Selection. IEEE Journal on Selected Areas inCommunications, 2006, 659-672.

10. Aggelos Bletsas, Hyundong Shin and Moe Z. Win. CooperativeCommunications with Outage-Optimal Opportunistic Relaying.IEEE Transactions on Wireless Communications, 2007, vol. 24,no. 3, 3450-3460.



Page 14 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

For Peer Review

11. Zhong Zhou, Shengli Zhou, Jun-Hong Cui and ShuguangCui. Energy-Efficient Cooperative Communication Based onPower Control and Selective Single-Relay in Wireless SensorNetworks. IEEE Transactions on Wireless Communications,2008, vol. 7, no. 8, 3066-3078.

12. Marvin K. Simon and Mohamed-Slim Alouini. DigitalCommunication over Fading Channels: A Unified Approach toPerformance Analysis. John Wiley & Sons, Inc, 2000.

13. A.K. Sadek, Yu Wei and K.J.R. Liu. When Does CooperationHave Better Performance in Sensor Networks?. IEEE The3rd Sensor and Ad Hoc Communications and Networks(SECON’06), 2006, 188-197.

14. D. G. Brennan. Linear Diversity Combining Techniques.Proceedings of the IEEE, 2003, vol. 91, no. 2, 331-356.

15. J. Nicholas Laneman, David N. C. Tse, and Gregory W.Wornell. Cooperative Diversity in Wireless Networks:EfficientProtocols and Outage Behavior. IEEE Transactions onInformation Theory, 2004, vol. 50, 3062-3080.

16. Todd E. Hunter and Aria Nosratinia. Cooperation Diversitythrough Coding. IEEE 2002 International Symposium onInformation Theory (ISIT’02), 2002, 220.

17. Andrew Sendonaris, Elza Erkip and Behnaam Aazhang.User Cooperation Diversity-Part I: System Description. IEEETransactions on Communications, 2003, vol. 51, 1927-1938.

18. Andrew Sendonaris, Elza Erkip and Behnaam Aazhang. UserCooperation Diversity-Part II: Implementation Aspects andPerformance Analysis. IEEE Transactions on Communications,2003, vol. 51, 1939-1948.

19. Xu Feng, F.C.M. Lau, Q.F. Zhou and Yue Dian-Wu. OutagePerformance of Cooperative Communication Systems UsingOpportunistic Relaying and Selection Combining Receiver.IEEE Signal Processing Letters, 2009, vol. 16, no. 2, 113-136.

20. M. Chen, T. Kwon, S. Mao, Y. Yuan and V. C.M. Leung.Reliable and Energy-Efficient Routing Protocol in DenseWireless Sensor Networks. International Journal on SensorNetworks, 2008, vol. 4, no. 12, 104-117.

21. Min Chen, Xuedong Liang, Victor C.M. Leung and IlangkoBalasingham. Multi-Hop Mesh Cooperative Structure basedData Dissemination for Wireless Sensor Networks. Interna-tional Journal on Sensor Networks, 2009, 102-106.

22. Xuedong Liang, Min Chen, Yang Xiao, Ilangko Balasinghamand Victor C.M. Leung. A Novel Cooperative CommunicationProtocol for QoS Provisioning in Wireless Sensor Networks.The 5th International Conference on Testbeds and ResearchInfrastructures for the Development of Networks & Communi-ties and Workshops (TridentCom’09), 2009, 1-6.

23. IEEE. IEEE Standard 802.11: Wireless LAN Medium AccessControl (MAC) and Physical Layer (PHY) Specification. IEEEStandard, 1997.

24. Gentian Jakllari, Srikanth V. Krishnamurthy, Michalis Falout-sos, Prashant V. Krishnamurthy and Ozgur Ercetin. A Cross-Layer Framework for Exploiting Virtual MISO Links in MobileAd Hoc Networks. IEEE Transactions on Mobile Computing,2007, vol. 6, 579-594.

25. Yuxia Lin, Joo-Han Song and Vincent W.S. Wong. CooperativeProtocols Design for Wireless Ad-Hoc Networks with Multi-hop Routing. Mobile Networks and Applications, 2009, vol. 14,no. 2, 143-153.

26. Xuedong Liang, Ilangko Balasingham and Victor C.M. Leung.Cooperative Communications with Adaptive Relay Selectionfor QoS Provisioning in Wireless Sensor Networks. The 52ndIEEE Global Communications Conference (GLOBECOM’09),2009, 1-8.

27. Simone Sergiilia, Fabrizio Pancaldi and Giorgio Matteo Vitetta.A game theory approach to selection diversity in wireless ad-hoc networks. The 8th 2009 IEEE international conference onCommunications (ICC’09, 2009, 4184-4189.

28. Beibei Wang, Zhu Han and K.J. Ray Liu. DistributedRelay Selection and Power Control for Multiuser Cooperative

Communication Networks Using Stackelberg Game. IEEETransactions on Mobile Computing, 2009, vol. 8, no. 7, 975-990.

29. Yingda Chen and Shalinee Kishore. A Game-TheoreticAnalysis of Decode-and-Forward User Cooperation. IEEETransactions on Wireless Communications, 2008, vol. 8, no. 7,1941-1951.

30. Beibei Wang, Yongle Wu and K.J. Ray Liu. Game theory forcognitive radio networks: An overview. Computer Networks,2010, vol. 54, no., 142537-2561.

31. Richard S. Sutton and Andrew G. Barto. ReinforcementLearning: An Introduction. MIT Press, 1998

32. Leslie Pack Kaelbling, Michael L. Littman and Andrew P.Moore. Reinforcement Learning: A Survey. Journal of ArtificialIntelligence Research, 1996, vol. 4, 237-285.

33. Zhenzhen Liu and I. Elhanany. RL-MAC: A QoS-AwareReinforcement Learning based MAC Protocol for WirelessSensor Networks. The 2006 IEEE International Conference onNetworking, Sensing and Control (ICNSC’06), 2006, 768-773.

34. Shaoqiang Dong, P. Agrawal and K. Sivalingam. Reinforce-ment Learning Based Geographic Routing Protocol for UWBWireless Sensor Network. IEEE Global TelecommunicationsConference 2007 (GLOBECOM’07), 2007, 652-656.

35. Ping Wang and Ting Wang. Adaptive Routing for SensorNetworks using Reinforcement Learning. The IEEE 6thInternational Conference on Computer and InformationTechnology (CIT’06), 2006, 219-225.

36. N. Ouferhat and A. Mellouk. Optimal QoS and adaptativerouting in Wireless Sensor Networks. The 2nd IEEEInternational Conference on Information and CommunicationTechnologies (ICTTA’06), 2006, 2736-2741.

37. C. Pandana and K.J.R. Liu. Near-optimal reinforcementlearning framework for energy-aware sensor communications.IEEE Journal on Selected Areas in Communications, 2005, vol.23, no. 4, 788-797.

38. Xuedong Liang, Min Chen, V. C. M. Leung and IlangkoBalasingham. Soft QoS Provisioning for Wireless SensorNetworks: a Cooperative Communications Approach The5th International ICST Conference on Communications andNetworking in China (ChinaCom’10), 2010, 1-8.

39. G. Naddafzadeh-Shirazi, P.Y. Kong and C.K. Tham. DistributedReinforcement Learning Frameworks for Cooperative Retrans-mission in Wireless Networks. IEEE Transactions on VehicularTechnology, 2010, vol. 59, no. 8, 1-10.

40. M. Felegyhazi and J.P. Hubaux. Game Theory in WirelessNetworks: A Tutorial. Technical Report: LCA-REPORT-2006-002, EPFL, 2006.

41. Peng Cheng, Zhaoyang Zhang, Guanding Yu, Hsiao-HwaChen and Peiliang Qiu. Cross-layer bandwidth and powerallocation for a two-hop link in wireless mesh network. WirelessCommunications and Mobile Computing, 2009, vol. 9, no., 2,155-167.

42. Honglin Hu, Jijun Luo, and Hsiao-Hwa Chen. Radio resourcemanagement for cooperative wireless communication systemswith organized beam-hopping techniques. IEEE WirelessCommunications, 2008, vol, 15, no. 2, 100-109.

43. Li Li, Honglin Hu, Xia Chen, and Hsiao-Hwa Chen. AdaptiveCooperative Coding with Power Allocation in Wireless RelayNetworks. IEEE Transactions on Wireless Communications,2009, vol. 8, no. 9, 4604-4615.

44. Xuedong Liang, Min Chen and Victor C.M. Leung. RelayAssignment for Cooperative Communications over DistributedWireless Networks: a Game-theoretic Approach. TechnicalReport: UBC, 2009, vol. 16, 1-8.

45. Wei Wang, Dongming Peng, Honggang Wang, Hamid Sharif,and Hsiao-Hwa Chen. Cross-Layer Multirate Interaction WithDistributed Source Coding In Wireless Sensor Networks. IEEETransactions On Wireless Communications, 2009, vol. 8, no. 2,787-795.



Page 15 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

For Peer Review

46. Honggang Wang, Dongming Peng, Wei Wang, Hamid Sharif,and Hsiao-Hwa Chen. Cross-layer routing optimization inmultirate wireless sensor networks for distributed sourcecoding based applications. IEEE Transactions On WirelessCommunications, 2008, vol. 7, no. 10, 3999-4009.

47. Ji Li, Lachlan L. H. Andrew, Chuan Heng Foh, MosheZukerman, and Hsiao-Hwa Chen. Connectivity, Coverage andPlacement in Wireless Sensor Networks. Sensors, 2009, vol. 9,7664-7693.

48. Qian Zhang and Ya-Qin Zhang. Cross-Layer Design for QoSSupport in Multihop Wireless Networks. Proceedings of theIEEE, 2008, vol. 19, no. 10, 64-76.

49. Prasant Mohapatra, Jian Li and Chao Gui. QoS in Mobile AdHoc Networks. IEEE Wireless Communications, 2003, vol. 10,no. 3, 44-52.

50. Goran Djukanovic, Milan Sunjevaric, Natasa Gospic andHsiao-Hwa Chen. Dynamic guard margin CAC algorithm withensured QoS and low CDP in heterogeneous wireless networks.Computer Communications, 2010, vol. 33, no. 14, 1645-1650.

51. Kun Yang, Alex Galis and Hsiao-Hwa Chen QoS-AwareService Selection Algorithms for Pervasive Service Composi-tion in Mobile Wireless Environments. Mobile Networks andApplications, 2010, vol. 15, no., 4, 488-501.

52. Caoxie Zhang, Xinbing Wang, Xinping Guan and Hsiao-HwaChen. Quality-of-Service in Cognitive Radio Networks withCollaborative Sensing The 52nd IEEE Global CommunicationsConference (GLOBECOM’09), 2009, 1-5.

53. Honglin Hu, Yan Zhang and Hsiao-Hwa Chen. An EffectiveQoS Differentiation Scheme for Wireless Mesh Networks.IEEE Network 2008, vol. 22, no. 1, 66-73.

54. Xi Zhang, Jia Tang and Hsiao-Hwa Chen. Space-time diversity-enhanced QoS provisioning for real-time service over MC-DS-CDMA based wireless networks. Wireless Communications andMobile Computing, 2008, vol. 8, no. 6, 733-744.

55. Wei Wang, Dongming Peng, Honggang Wang, HamidSharif, and Hsiao-Hwa Chen. Energy-Constrained QualityOptimization for Secure Image Transmission in WirelessSensor Networks. Hindawi Advances in Multimedia Journal,2008, vol. 7, no. 1, 1680-1687.

56. A. Veres, A.T. Campbell, M. Barry and Sun Li-Hsiang.Supporting service differentiation in wireless packet networksusing distributed control. IEEE Journal on Selected Areas inCommunications, 2001, 2081-2093.

57. Ahmed K. Sadek, Weifeng Su and K. J. Ray Liu. MultinodeCooperative Communications in Wireless Networks. IEEETransactions on Signal Processing, 2007, vol. 55, 341-355.

58. Zhaoyang Zhang, Jing Shi, Hsiao-Hwa Chen, Mohsen Guizaniand Peiliang Qiu. A cooperation strategy based on Nashbargaining solution in cooperative relay networks. IEEETransactions On Vehicular Technology, 2008, vol. 57, no. 4,2570-2577.

Authors’ Biographies

Xuedong Liang is a Postdoctoral Research Fellowat the Department of Electrical and ComputerEngineering, University of British Columbia, Van-couver, Canada. He received his Ph.D degree in

the Department of Informatics, University of Osloin 2009. His research interests are in the areas ofwireless communication protocols, formal modelingand validation of wireless networks, QoS provisioningin wireless sensor networks.

Min Chen received the Ph.D degree in ElectricalEngineering from South China University of Tech-nology in 2004, when he was 23 years old. He hasbeen a Research Associate in the Dept. of ComputerScience at University of British Columbia (UBC) forhalf year, and worked as a Post-Doctoral Fellow inDepartment of Electrical and Computer Engineeringat UBC for three years, where his faculty advisor isProf. Victor C.M. Leung. Before joining UBC, hewas a Post-Doctoral Fellow at SNU for one and halfyears. He received the Best Paper Runner-up Awardfrom QShine 2008. He was interviewed by ChineseCanadian Times where he appeared on the celebritycolumn in 2007. He is the author of OPNET NetworkSimulation (Tsinghua University Press, 2004). He hasmore than 60 technical publications.

Ilangko Balasingham received the Siv.Ing (MSc)and Dr.Ing. (PhD) degrees both in signal processingfrom Norwegian University of Science and Tech-nology (NTNU), Trondheim, Norway in 1993 and1998, respectively. He has been with the InterventionalCenter, Rikshospitalet University Hospital, Oslo,Norway as a Research Scientist and Head of theBiomedical Sensor Network Research Group. He wasalso appointed as an Adjunct Professor of SignalProcessing in Medical Technology at NTNU in 2006.His research interests are in areas of biomedicalsensors, wireless sensor network and its applicationsin medical care.

Victor C.M. Leung received the B.A.Sc. (Hons.)degree and the Ph.D. degree in from the Universityof British Columbia (UBC) in 1977 and 1981,respectively, both in electrical engineering. Dr. Leung



Page 16 of 19



123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

Documents

Cooperative communications with relay selection in wireless sensor networks