Cooperation in Random Access Networks - Signals and Information ...
  • No tags were found...

Cooperation in Random Access Networks - Signals and Information ...

1694 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 30, NO. 9, OCTOBER 2012Cooperation in Random Access Networks:Protocol Design and Performance AnalysisAmr A. El-Sherif and K. J. Ray LiuAbstract—This paper tries to answer the questions of how toenable cooperative communications in random access networks.And, since cooperation introduces extra transmissions in thechannel, what are the benefits and possible tradeoffs associatedwith cooperation? To answer these questions, a novel cooperativeprotocol for wireless networks based on IEEE 802.11 randomaccess protocol is proposed. Cooperation is achieved throughthe deployment of a relay node that intelligently access thewireless medium without imposing any penalties in terms ofincreased collision probability. To help other network nodes,relay stores packets in its queue, and accesses the channel aftereach transmission attempt made by any network node. Using thisscheme, relay is guaranteed uncontested access to the channel. Tocapture the interactions between different network nodes and therelay, the network is modeled using Markov chains in conjunctionwith queuing analysis. Throughput and delay performances of theproposed protocol are characterized and compared to CSMA/CAwithout cooperation. The results demonstrate that significantgains are achieved by the proposed cooperative protocols.Index Terms—Cooperative communications, random access,queuing theory.I. INTRODUCTIONCOOPERATIVE communications is a concept that hasbeen recently introduced to improve reliability andthroughput in wireless networks [1], [2], [3]. The broadcastnature of the wireless medium is the key property that allowsfor cooperation among networking nodes. In cooperative networks,users cooperate by relaying each others’ message overmultiple paths. At the destination, the original and relayedsignals are combined to generate a signal with better quality,creating a new form of diversity which can significantlyimprove the system performance and robustness.Some works have focused on the impact and implementationof cooperation at higher network layers. A cognitivemultiple access protocol was proposed in [4]. The protocolbenefits from source burstiness to enable cooperation by allowinga relay to utilize the nodes’ periods of silence. In [5], theauthors have proposed a cooperative multiple access protocolfor speech networks. The protocol enables cooperation byexploiting the relatively long silent periods typical in speechcommunications. Cooperation in random access networks hasbeen considered in [6], where a distributed version of networkdiversity multiple-access (NDMA) [7] protocol was proposed.The main focus of [6] and [7] is on collision resolution usingconventional source separation techniques at the destination,the corrupted packets are retrieved.Manuscript received 20 January 2011; revised 20 June 2011.A. A. El-Sherif is with the Department of Electrical Engineering, AlexandriaUniversity, Alexandria 21544, Egypt (e-mail: J. Liu is with the Department of Electrical and Computer Engineering,University of Maryland, College Park, MD 20742, USA ( Object Identifier 10.1109/JSAC.2012.121013.0733-8716/12/$31.00 c○ 2012 IEEETo enable cooperation in a wireless network, transmissionsamong the cooperating nodes are generally needed in orderto coordinate their actions. In large random access networkswithout centralized scheduler like in IEEE 802.11 networks[8], these extra transmissions will increase the number ofpacket collisions. As such it is not clear if there is any benefitof using physical layer cooperation in this case.To answer the question of how to enable cooperationin a random access network without the increased numberof packet collisions, we begin by proposing a cooperativeprotocol in which a relay node is deployed to help differentnetwork nodes forward their packets to the access point (AP).In order not to increase the collision probability, the relayintelligently accesses the wireless medium at times when it isguaranteed that no other node is accessing the medium. In theIEEE 802.11 protocol and the CSMA/CA protocol, on whichit is based, after each packet transmission, and irrespective ofthe outcome of this transmission, all nodes wait for a randomamount of time before making a transmission attempt. Byallowing the relay to access the channel immediately after eachtransmission attempt, it is guaranteed to have an uncontestedaccess to the wireless medium.In a network operating as described above, all nodes’squeues are interacting, i.e., the service process of a givenqueue depends on the state of all other queues (whetherthey are empty or not). Interaction between queues is mainlybecause of the possible collisions that occur if multiple nodestry to access the channel at the same instant. And, for therelay, its own ability to access the channel is dependent onother nodes’ transmissions.In order to capture this queue interaction, and to be ableto analyze the performance of our protocol, two coupledMarkov models are used to describe the operation of therelay and other network nodes. Both Markov models togetherare able to completely describe the dynamics of the networkand interactions between different nodes. Moreover, queuinganalysis is used to analyze the delay performance of thenetwork, and the model is validated through simulation. Theresults presented reveal significant gains in terms of networkthroughput, delay, and the number of supported nodes, dueto cooperation and our proposed protocol. Furthermore, itis shown that, by virtue of the protocol design, collisionprobability has decreased rather than the expected increasedue to extra transmissions on the channel.The rest of the paper is organized as follows. In section II,the channel and network models are presented. The proposedcooperation protocol is presented in section III. Markov andqueuing models are developed and analyzed in sections IVand V, respectively. Network throughput and delay are characterizedin section VI. Section VII presents the numerical

EL-SHERIF and LIU: COOPERATION IN RANDOM ACCESS NETWORKS: PROTOCOL DESIGN AND PERFORMANCE ANALYSIS 1695results and their discussions. Finally, the paper is concludedin section VIII.II. CHANNEL AND SYSTEM MODELSA. Channel modelA Rayleigh fading channel model is considered, the signalreceived at the access point or the relay is modeled as√y ij = Gr −γij h ijx + η j (1)where i is the source index, j ∈{A, R} is the access pointor the relay index, x is the transmitted signal, G is thetransmission power, assumed to be the same for all nodes,r ij denotes the distance between source node i and its destinationj, γ is the path loss exponent, h ij the channel fadingcoefficients, modeled as zero-mean complex Gaussian randomvariables with unit variance, and η j is an additive noise termat the destination, modeled as zero-mean complex Gaussianrandom variable with variance N 0 . We assume that the channelcoefficients are constant for the duration of the transmission ofone packet. In this work, without loss of generality, we onlyconsidered the case of a symmetric network, where all theinter-user channels are assumed to be statistically identical.This is possible for example when nodes are grouped in asmall cluster at fixed distance from the access point.Success and failure of packet reception is characterized byoutage events and outage probabilities. The outage probabilityis defined as the probability that the Signal to Noise Ratio(SNR) at the receiver is less than a given SNR threshold β.For the channel model in (1) the probability of outage can bewritten as,{Pijout = Pr | h ij | 2 < βN } (0r γ ij=1− exp − βN )0r γ ij.GG(2)B. Network ModelOur work in this paper focuses on wireless networks inwhich all nodes and the AP are within communication rangefrom each other. In other words, we are only considering asingle hop wireless network where each node is communicatingonly with the AP. Based on this assumption, this singlehop network does not suffer from the hidden node problemsince any node can overhear all other nodes. Without loss ofgenerality, this paper also focuses on the single relay case.Due to space limitation the case of a multi-hop network withmultiple relays is not considered in this paper. Extension of ourcooperation protocol design to the multiple-hop network case,how to deal with hidden nodes and the effect of cooperationon routing will be the subject of a future work.C. IEEE 802.11 DCF OperationThe distributed coordination function (DCF) is the fundamentalmedium access mechanism in the IEEE 802.11 protocol[8]. It is a random access scheme based on the CSMA/CA(Carrier Sense Multiple Access with Collision Avoidance)with binary slotted exponential backoff. As depicted in fig.1 a node with a packet to transmit invokes the carrier sensingmechanism to determine the busy/idle state of the channel. IfFig. 1. DCF basic access mechanism; numbers in figure represent node’sbackoff timer.the channel is sensed to be idle for a period of time equalto a Distributed Inter-Frame Space (DIFS), the node proceedswith packet transmission. Otherwise, it persists to monitor thechannel until it is measured idle for a DIFS. The node thendefers for a randomly selected backoff interval, initializing itsrandom backoff timer, which is decremented as long as thechannel is sensed idle and is frozen when a transmission isdetected.The time immediately following an idle DIFS is slotted,and a node is allowed to transmit at the beginning of a slottime if its backoff timer reaches zero. The slot duration, σ,is set equal to the time needed for any node to detect thetransmission from any other node. It depends on the physicallayer, and it accounts for the propagation delay, and the timeneeded to detect a busy channel.The random backoff interval is uniformly chosen in therange (0,w − 1). The value w is called the contentionwindow. At the first transmission attempt, w is set equal toa minimum contention window value CW min . After eachunsuccessful transmission w is doubled, up to a maximumvalue CWmax = 2 m CW min , m is the maximum backoffstage. Once w reaches CWmax, it remains at this value untilit is reset to CW min after a successful transmission.To signal the successful packet reception, an ACK istransmitted by the destination. The ACK is transmitted aftera period of time called short inter-frame space (SIFS). Asthe SIFS is shorter than a DIFS, no other station is able todetect the channel idle for DIFS until the end of the ACK.If the transmitting node does not receive the ACK within aspecified ACK Timeout, the packet is assumed to be lost andthe node reschedules the packet transmission according to thegiven backoff rules.III. RANDOM ACCESS COOPERATION PROTOCOLInherent wireless channel fading and transmission errorshave a significant impact on the network’s performance [9].In wireless networks, nodes are unable to detect collisionsby hearing their own transmission. Therefore, there is nodifferentiation between a packet loss due to a collision, and apacket loss due fading. Therefore, a source node will deal witha wireless channel induced packet loss in the same way it dealswith a collision induced one. Hence, doubling its contentionwindow and waiting for a random amount of time beforereattempting transmission. As a result of invoking the backoffprocedure in a non-congested channel, the network will sufferfrom an increased delay and lower achievable throughput [10].To combat the wireless channel impairments leading tothese problems, we propose the deployment of a cooperative

1696 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 30, NO. 9, OCTOBER 2012relay node into the coverage area of the wireless network. Thecooperative relay node will help combat the channel fadingthrough the introduction of spatial diversity into the network.Relay node will help source nodes forward their packets byoperating in an incremental decode-and-forward mode [1]. Inthis mode, in case of a packet loss at the access point, therelay attempts to decode the received packet, and in case ofsuccessful detection at the relay, it forwards a regeneratedversion of the packet to the AP.The relay makes use of the AP’s ACK packet to know ifa packet is successfully received by the AP. In case the relaysuccessfully receives a packet, but the AP does not receivethat packet (ACK Timeout occurs), the relay stores it in itsqueue and sends an ACK packet over the channel to informother nodes that the packet was received successfully. Uponreceiving the relay’s ACK packet, the packet owner drops itfrom its queue and the delivery of that packet becomes therelay’s responsibility. Because of the relay’s ACK packet, thenode with the lost packet will reset its contention windowCW min .The challenging part in the design of our cooperationprotocol is to enable the relay to gain access to the wirelesschannel without increasing the number of collisions, and hencerendering its existence useless. To deal with this issue wepropose the following relay channel access protocol:• Following a transmission attempt from one or moresource nodes (outcome of the transmission attempt isirrelevant), the relay node attempts to transmit the packetat the head of its queue immediately after the AP’s ACK,or after the ACK Timeout.• For the relay not to be totally dependent on other nodes’transmission attempts, the relay also maintains a singlestage backoff counter with contention window size CW r• When the relay’s backoff counter reaches zero, it willattempt to transmit the packet at the head of its queuelike other network nodes.• Like other network nodes, the relay will invoke thebackoff procedure after each transmission attempt, theonly difference is that the relay has a single backoff stageas opposed to m stages for other nodes.By accessing the channel after each transmission attempt therelay has the ability to serve the packets in its queue withoutcausing any collision.IV. MARKOV MODELS AND ANALYSISA number of models have been proposed to study theperformance of IEEE 802.11 DCF in the saturated [11], [12],unsaturated [13], [14] traffic conditions, and in the presenceof channel impairments [10]. To analyze the performance ofthe proposed cooperative protocol, we start from the discretetimeMarkov model for non-saturated sources developed in[14], and incorporate the channel effects and relay operationinto the model. We consider two separate Markov chains, thefirst chain models source nodes while the second models therelay node.We assume that the network consists of N contendingnodes in addition to the relay node. Each node has an infinitequeue to store packets. Each node receives packets fromupper layer based on a Poisson arrival process with arrivalFig. 2.0,0 e 0,1 e0, 2 e0,0 0,1 0, 2i ,0m,0m ,1 m,2Source node’s Markov model.s0, W0 2e0, W 2s0s0, W01e0, W 1mW , m 2 mW , 1smrate λ s packets/sec (The super(sub)script s or r are used todifferentiate between source node and relay parameters), andfixed packet size L. The queuing model used will be discussedin details in section V.A. Source NodesFig. 2 represents the discrete-time Markov chain used tomodel the operation of source nodes. Each node is modeledbyapairofintegers(i, k). The backoff stage i, starts at 0 at thefirst attempt to transmit a packet and is increased by 1 everytime a transmission attempt fails, up to a maximum value m.It is reset after a successful transmission. At any backoff stagei ∈ [0,m], the backoff counter, k, is initially chosen uniformlybetween [0,Wiss− 1], whereWi =2 i W0 s , 0 ≤ i ≤ m, isthe range of the counter, and W0s is the parameter CW minspecified in the IEEE 802.11 standard. The backoff counter isdecremented by 1 in each idle time slot of duration σ, andthenode transmits when the backoff counter k =0.States (0,k) e , k ∈ [0,W0 s − 1] are introduced to representthe state of the node when it has an empty queue after asuccessful transmission. Note that i =0in these states becauseif i>0 then a failed transmission should have occurred, so apacket must be awaiting.The fundamental assumption in our model is that, at eachtransmission attempt, and regardless of the number of retransmissionssuffered, each packet fails with a constant andindependent probability, Pfs or P f r , for the source nodes orrelay node, respectively [11], [14].Let τ s and τ r be the probability that a source node or therelay transmit in a given slot, respectively. Now we are readyto write the Markov chain’s transition probabilities, for 0 ≤i ≤ mP {(i, k)|(i, k +1)} = P i ,0 ≤ k ≤ Wi s − 2P {(i, k)|(i, k)} =(1− P i ),0 ≤ k ≤ Wi s P {(0,k)|(i, 0)} = (1 − q s)(1 − Pf s)W0s , 0 ≤ k ≤ W0 s − 1P {(0,k) e |(i, 0)} = q s(1 − Pf s)W0s , 0 ≤ k ≤ W0 s − 1(3)s0

EL-SHERIF and LIU: COOPERATION IN RANDOM ACCESS NETWORKS: PROTOCOL DESIGN AND PERFORMANCE ANALYSIS 1697where q s is the probability that the node’s queue is emptyupon a departure (see section V). P i is the probability that thechannel is sensed idle by the source node (i.e., all the remainingN − 1 source nodes and the relay node are not attemptingto transmit), and is given by P i =(1− τ s ) N−1 (1 − τ r ).Asource node’s transmission attempt is considered successfulif the channel is idle (i.e., no collision) and either the APor the relay correctly receive the transmitted packet, in otherwords if either the source-AP or the source-relay channel is notin outage. Let Pfs be the probability of a failed transmissionattempt, since a failed transmission is the complement event ofa successful transmission (a transmission is successful if thechannel is idle and either the source-AP link or the sourcerelaylink is not in outage), then Pf s =1−P i (1 − PsA outPsR out),where PsAout outand PsRare the outage probabilities of thesource-AP and source-relay links, respectively.The first and second equations in (3) accounts for the factthat at the beginning of each idle slot time the backoff counteris decremented by one, and that the counter remains at itscurrent state if the channel is not idle. The third and forthequations account for the fact that following a successfulpacket transmission, backoff stage i is reset to 0, and thus thebackoff is initially uniformly chosen in the range [0,W0 s − 1].In case of an unsuccessful transmission at backoff stagei−1, the backoff stage is increased, and the new initial backoffcounter is initially chosen in the range [0,Wis − 1]. Oncethe backoff stage reaches the value m s , it is not increasedin subsequent packet transmissions, then we haveP {(i, k)|(i − 1, 0)} = P s f /W si ,P {(m, k)|(m, 0)} = P s f /W s m. (4)Given that the node’s queue is empty and the chain is instate (0,k) e , in case of a packet arrival, the backoff counter isdecremented and the chain makes a transition into the (0,k−1)state if the channel is idle, and to state (0,k) if the channel isnot idle. In the case of an idle channel but no packets arrivesto the queue the chain transits into (0, (k − 1) e ). When thebackoff timer reaches zero, the node remains in state (0, 0 e )as long as the queue is empty. If a packet arrives, then thenode moves into state (0,k), wherek is uniformly chosen inthe range [0,W s 0 − 1]. Therefore we haveP {(0,k) e |0, (k +1) e } = P i (1 − a i ), 0 ≤ k ≤ W0 s − 2P {(0,k)|(0, (k +1) e )} = P i a i , 0 ≤ k ≤ W0 s − 2P {(0,k e )|(0,k e ))} =(1− P i )(1 − a b ), 0 ≤ k ≤ W0 s − 1P {(0,k)|(0,k e )} =(1− P i )a b , 0 ≤ k ≤ W0 s − 1P {(0, 0 e )|(0, 0 e )} =1− (P i a i +(1− P i )a b ) ,0 ≤ k ≤ W0 s − 1P {(0,k)|(0, 0 e )} = P ia i +(1− P i )a bW0s , 0 ≤ k ≤ W0 s − 1(5)where a i and a b are the probabilities of at least one packetarrival during an idle or a busy slot, respectively. From thePoisson arrival assumption, these probabilities are given bya i =1− e −λsσ and a b =1− e −λsT b,whereσ is the idleslot duration, and T b the busy slot duration (for simplicitywe neglect the difference in durations between successful andFig. 3.e0Relay node’s Markov model.12unsuccessful transmission attempts). Typically, σ =20μs, andT b = 2160.4μs, based on 11 Mbps channel rate and packetsize L = 2312 octets [8].Let π s (i, k) denote the stationary probability of beingin state (i, k). To solve for the stationary distribution ofthis Markov chain we used balance equations [15] tofind expressions for all the stationary probabilities as afunction of π s (0, 0). Imposing the normalization condition∑ m ∑ Wsi −1i=0 k=0π s (i, k) + ∑ W s 0 −1k=0π s (0,k) e =1, we can calculateπ s (0, 0), hence, all the steady state probabilities. Fullderivation of the closed form expressions is omitted because ofthe limited space, however a similar procedure to our analysiscan be found in [11].Finally, since a node will make a transmission attempt ina given slot time if the Markov chain is in state π s (i, 0) fori ∈ [0,m], then, τ s , the probability that a source node makesa transmission attempt in a given slot time can be expressedas τ s = ∑ mi=0 π s(i, 0).B. Relay NodeA relay node operating as described in section III will bemodeled using the Markov chain model of Fig. 3. The modelhas a single backoff stage represented by states k ∈ [0,W r −1]. The backoff counter is uniformly chosen in that range, andthe relay makes a transmission attempt when in state 0. Therelay node makes a transition to state e if its queue becomesempty after a successful transmission. Finally, the chain is instate t when the relay is attempting to transmit following abusy channel.Again we have the assumption that, at each transmissionattempt, and regardless of the number of re-transmissionssuffered, each packet fails with a constant and independentprobability Pfs or P f r , for the source nodes or relay node,respectively [14], [11]. Now we are ready to write the Markovchain’s transition probabilities. At the beginning of each idleslot time, the backoff counter is decremented, thenWr 2Wtr1P {k|k +1} = P i , 0 ≤ k ≤ W r − 1, (6)where P i is the probability that the channel is sensed idle bythe relay node (i.e., all N source nodes are not attempting totransmit), and is given by P i =(1− τ s ) N .Since the relay Markov chain has a single backoff stage,following an unsuccessful transmission attempt or a successfulattempt that leaves the relay queue non-empty, the backoffcounter is initially uniformly chosen in the range [0,W r − 1].

1698 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 30, NO. 9, OCTOBER 2012ThereforeP {k|0} = (1 − q r)(1 − Pf r)W r + P frW r , 0 ≤ k ≤ W r − 1P {k|t} = (1 − q r)(1 − Pf ′r)W r + P f′rW r , 0 ≤ k ≤ W r − 1(7)where q r is the probability that a departing packet will leavethe relay queue empty, Pfr is the probability of a failed relaytransmission attempt out of state 0 (failure due to collisionor channel error), and Pf′r is the probability of a failed transmissionattempt out of state t (failure can be caused only bychannel errors), and are given by Pf r =1−(1−τ s) N (1−PRA out)and Pf′r = P RA outout,wherePRA is the outage probability of therelay-AP link.A successful transmission that leaves the relay queue emptyleads to a transition to state e, thenwehaveP {e|0} = q r (1 − Pf r )P {e|t} = q r (1 − P f ′r ). (8)Transitions into state t occur when the relay attempts totransmit a packet immediately after any transmission attempton the channel, thusP {t|e} = Nτ s (1 − τ s ) N−1 PsA out out(1 − PsR )=aP {t|k} =1− P i , 0

EL-SHERIF and LIU: COOPERATION IN RANDOM ACCESS NETWORKS: PROTOCOL DESIGN AND PERFORMANCE ANALYSIS 1699The next component is the time spent in a backoff stage ibefore making a transmission attempt (i.e., before the backoffcounter reaches zero). At stage i the counter is initializeduniformly in the range k ∈ [0,Wis − 1], therefore thedistribution of the time spent in stage i is characterized bythe PGF, F i (z) = ∑ Wi s −1 F k (z)k=0 W.isFinally, the PGF for the service time S s can be written as⎡m−1∑i∏G s (z) =(1− Pf s )z T b ⎣ (Pf s z T b) i F j (z)+(P s f z T b) mm ∏j=0i=0F j (z)j=0⎤∞∑(Pf s z T b) i Fm(z)i ⎦i=0⎡m−1∑i∏=(1− Pf s )zT b ⎣ (Pf s zT b) i F j (z)i=0j=0+ (P f szT b) m ∏ mj=0 F ]j(z)1 − Pf s , (11)zT b Fi m (z)where the term outside the brackets accounts for the busyslot in which the packet is successfully delivered, the firstterm inside the brackets accounts for the possible number offailures a packet encounters (hence, the number of backoffstages it goes through), its composed of the time spent in thebackoff counter decrements and the time spent transmittingthe packet. Finally, the second term accounts for the amountof time spent at the maximum backoff stage m, which isdecomposed into the time spent to reach this state, the time forcounter decrements, and the time for packet transmission. Theservice rate can then be calculated by differentiating G s (z)and setting z =1, μ −1s = E[S s ]= dGs(z)dz∣ .z=1B. Relay Node Arrival RateThe time, A r , between packet arrivals to the relay queueis composed of the following components: (i) Idle periods inwhich no node (source or relay) is transmitting. This periodshave a length σ and probability P i =(1− τ s ) N (1 − τ r ). (ii)Busy periods of duration T b , which occur if the relay queueis empty and the transmission attempt does not result in anarrival at the relay. This occurs with probabilityP b1 = π r (e)[1 − (1 − τs ) N−1 − (N − 1)τ s (1 − τ s ) N−2 PsA out (1− PsR out ) ] .(iii) Busy periods of duration 2T b not resulting in a relayarrival, which occur if the relay queue is not empty whena source node makes a transmission attempt. Thus has aprobabilityP b2 =[1− (1− τs ) N−1 − (N − 1)τ s (1 − τ s ) N−2 PsA out out(1−PW∑r −1π r (k).k=1sR )](iv) Finally, a busy period during which a packet enters therelay queue. This will always have a duration T b , and has[a probability P a = π r (e)+ ∑ ]W r −1k=1π r (k) (N − 1)τ s (1 −τ s ) N−2 PsA out out(1 − PsR ).Given the above mentioned probabilities, we can write thePGF of A r as follows,∞∑ i∑ ∑i−j[G a (z) =P a z T i!bj!k!(i − j − k)!i=1 j=1 k=0· (P i z σ ) j (P b1 z T b) k (P b2 z 2T b) i−j−k] . (12)The arrival rate can then be calculated by differentiatingG a (z) and setting z =1, λ −1r = E[A r ]= dGa(z)dz∣ .z=1C. Relay Node Service RateSimilar to the way the source nodes service rate was calculated,we will start the calculation of the relay node service rateby defining the different components that constitute a packet’sservice time S r . We note that, as opposed to source nodes,the relay can leave the backoff stage after any source node’stransmission attempt on the channel (Fig. 3). Therefore, thetime the packet at the head of the queue spends in the backoffstage can be split into two components: (i) The time before thebackoff counter (initialized uniformly between 0 and W r − 1)reaches 0, which in the relay case is composed only of idleslots. The PGF characterizing the distribution of that time isthen given byF 0 =W∑r −1k=0P ki zkσW r . (13)(ii) The time spent in the backoff stage before the Markovchain reaches state t, which is composed of a single busyperiod and a maximum of W r − 1 idle slots. The PGFcharacterizing the distribution of that time is then given byF t =(1− P i )z T bW∑r −2k=0P ki zkσW r . (14)Finally, let the probability that a packet enters relay queuebe a = Nτ n (1 − τ n ) N−1 PsA out out(1 − PsR ), the PGF for theservice time S r can be written asG r (z) =π r (e)a(1 − P ′rf )zT b+·∞∑i=0 j=0(aP ′rf zT b+W∑r −1k=0π r (k)[(F0· (z)(1 − Pf r )z T b+ F t (z)(1 − P f ′r )z T )b⎤i∑( ) i (F0(z)Pf r )jzT b j (Ft (z)P f ′r ) zT b i−j⎦ ,(15)which accounts for the case when a packet is immediatelyserved by the relay after it enters the queue (if queue wasempty at packet arrival), the possible number of failuresa packet encounters getting transmitted from either state 0or state t, and finally, the periods at which the packet is)

1700 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 30, NO. 9, OCTOBER 2012delivered successfully. The service rate can then be calculatedby differentiating G r (z) and setting z =1, μ −1r = E[S r ]=∣ .z=1dG r(z)dzVI. PERFORMANCE MEASURESA. Network ThroughputLet S be the normalized network throughput, defined as thefraction of time the channel is used to successfully transmitpayload bits to the AP, which can be expressed as S = Ps·TpT s,where P s is the probability of a successful transmission to theAP (by source or relay nodes), T p is the time to transmit thepayload part of a packet, of course this is less than T b ,thetotal transmission time of a packet including the headers andthe AP ACK packet. And T s is the expected slot duration.To calculate the probability P s , we identify the events thatresult in a successful packet delivery to the AP, which are:(i) If the relay queue is empty, a source node successfullytransmits a packet to the AP, or that packet fails to reachthe AP but was successfully received and forwarded by therelay. This event has a probability Ps1 = π r (e) [ Nτ s (1 −τ s ) N−1 ((1 − PsA out out out out)+PsA (1 − PsR )(1 − PRA )) ] . (ii) Ifthe relay queue is not empty, a source node transmissionfails to reach the AP (due to fading or collision), andthe relay successfully transmits the packet at the head ofits queue to the AP. This occurs with probability Ps 2 =[1 − (1 − τs ) N − Nτ s (1 − τ s ) N−1 (1 − PsA out)]∑ W rk=0 π r(k).(iii) If the relay queue is not empty and both source nodetransmission and the immediately following relay nodetransmission were successful. This occurs with probabilityPs3 = Nτ s (1 − τ s ) N−1 (1 − PsA outout)(1 − PRA ) ∑ W rk=0 π r(k).(iv) The relay succeeds in transmitting a packet whenits backoff counter reaches 0, which has a probabilityPs 4 = τ r (1 − τ s ) N (1 − PRA out ). Finally, the probabilityP s = Ps 1 + P s 2 +2P s 3 + P s 4. The factor of 2 before P s3accounts for the fact that the associated event results in thesuccessful delivery of two packets to the AP.The average length of a randomly chosen[slottime is given by T s = (1− τ s ) N (1 − τ r ) σ + τ r +(π r (e) 1−(1−τ s ) N −Nτ s (1 − τ s ) N−1 PsAout out(1 − PsR )]T ) b +2π r (t)T b , which accounts for the three possible slot durations.1) Idle slots of duration σ, in which neither the relay norany other node attempts to transmit a packet, thus having aprobability (1 − τ s ) N (1 − τ r ). 2) Busy slots of duration T b inwhich either the relay transmits or a source node transmissionis not followed by a relay transmission. The relay transmitsif its backoff counter reaches zero, which occurs withprobability τ r . A source node transmission is not followed bya relay transmission when the relay queue is empty (whichoccurs with probability π r (e) ) and no arrivals to the relayoccur during source transmission. An arrival to the relayoccurs when a source node transmission fails to reach thedestination (due to channel outage) but reaches the relay, thisevent has a probability Nτ s (1 − τ s ) N−1 PsAout out(1 − PsR ).The event of no relay arrival is then the complement of thatevent, and we must take care to exclude the event of nosource transmissions from that complement event. 3) Busyslots of duration 2T b , in which a relay transmission followsa source node transmission, hence the factor of 2, such anoccurs with probability π r (t).Based on an 11 Mbps transmission rate, and payload oflength L = 2312 octets, typical slot duration are σ =20μs,T p = 1681.5μs, andT b = 2160.4μs.B. DelayIn the proposed cooperation protocol, a packet can encountertwo queuing delays; the first in the source node’squeue and the second in the relay’s queue. If a packetsuccessfully transmitted by a source node arrives to the AP,then this packet is not stored on the relay’s queue. Let P adenote the probability of this event. Then the total delayencountered by a packet can be modeled as{Ds , w.p. PD =a(16)D s + D r , w.p. 1 − P awhere D s and D r are the queuing delays in the source andrelay queues, respectively. We can elaborate more on (16) asfollows. For a given packet in the source node’s queue, if thepacket is delivered from the source to the AP directly (withoutrelay help), then the delay encountered by this packet is onlythe queuing delay in the source nodes queue. On the otherhand, if the packet is delivered to the AP through the relay,then the packet will encounter the following delays: queuingdelay in the source nodes queue in addition to the queuingdelay in the relays queue.First, we find the queuing delay in either the source node orthe relay queue, as both are modeled as M/G/1 queues, withthe difference being in the average arrival and departure rates.For an M/G/1 queue, the mean waiting time in queue is givenby the Pollaczek-Kinchin formula [16], E[W i ]= λiE[S2 i ]2(1−λ , i/μ i)where i ∈ (s, r), λ i is the average arrival rate, μ i the averageservice rate, and S i the service time. From the mean waitingtime, one immediately gets the mean queuing delay as D i =E[W i ]+E[S i ].The probability P a , that, for any packet, the first successfultransmission from the source node’s queue is to the AP is1−Pgiven by P a =outsA(1−PsA out )+(1−P outsR )−(1−P outsA )(1−P out).AndthesRaverage delay is thus given by D = D s +(1− P a )D r .VII. RESULTS AND DISCUSSIONSWe compare the performance of the cooperative protocoland the CSMA/CA protocol without cooperation. We setthe SNR threshold β =15dB and the path loss exponentγ = 3.7. The distance between any node and AP is 120m, and between any node and the relay 70 m, and betweenrelay and AP 50 m. Transmission power is 100mW , andnoise variance N 0 =10 −11 . Source node’s initial contentionwindow W0s = 32 with m = 5 backoff stages, and relaynode’s contention window size W0r = 32. To validate themodel, we have built a custom packet-based simulator, thatclosely follows all the IEEE 802.11 protocol details for eachsource node, and follows the details of our proposed protocolat the relay node. Simulation results are based on an 11 Mbpstransmission rate, and payload of length L = 2312 octets, slotduration σ =20μs, DIFS=128μs, andACK Timeout=300μs.

EL-SHERIF and LIU: COOPERATION IN RANDOM ACCESS NETWORKS: PROTOCOL DESIGN AND PERFORMANCE ANALYSIS 1701Aggregate arrival rate350340330320310300290CSMA/CACoop.1−λ r/μ r10.99950.9990.99850.9982800 10 20 30 40 50Number of nodes "N"0.99750 5 10 15 20 25Number of nodes "N"Fig. 4. Maximum achievable aggregate arrival rate vs number network nodes.Queuing delay (Sec.) Sim.Coop.Coop. Sim.00 5 10 15 20 25Number of nodes "N"Fig. 5. Queuing delay vs. number of nodes for λ s =15.Fig. 6. Probability that relay queue is empty vs. number of nodes for λ s =15.Collision Probability0. Sim.Coop.Coop. Sim.00 5 10 15 20 25Number of nodes "N"Fig. 7. Collision probability vs. number of nodes for λ s =15.Fig. 4 depicts the maximum aggregate arrival rate (sumof the arrival rates of all network nodes) supported by thenetwork while maintaining queues stability versus the numberof network nodes. We can observe that, for a given numberof nodes, the proposed cooperative protocols resulted in a7% average increase in the maximum supported aggregatearrival rate. This increase is due to the fact that the relaynode provides a more reliable path to the AP leading to ahigher packet delivery rate. Therefore, source nodes are able toempty their queues at a faster rate, thus, freeing the channel forrelay access, and for additional nodes that the network mightaccommodate. As the number of nodes increase the supportedarrival rates start to decrease since the network starts to getcongested and the queues’ stability can no longer be supportedwithout a decrease in arrival rates.Fig. 5 shows the delay performance of our cooperativeprotocol compared to the non-cooperative CSMA/CA protocol.It is clear that our protocol outperforms CSMA/CA interms of queuing delay. This is mainly because most of therelay’s transmission attempts are made just after source nodes’transmissions, and not by waiting for the backoff counter toreach 0. Therefore, the relay is guaranteed a high degree ofuncontested channel access. Moreover, as the network loadincreases, the average number of source nodes’ transmissionattempts increase, which offers the relay more channel accessopportunities to service its queue that now has a higher arrivalrate. To prove this, the quantity (1 − λ r /μ r ), which fromqueuing theory is the probability that the relay queue is empty,is plotted in Fig. 6. It can be seen that there is less than 1%variation in the probability over the range of supported numberof nodes.Fig. 7 compares between the collision probability ofCSMA/CA and our cooperative protocol. Another merit ofour cooperation protocol and its channel access mechanism isthat, the introduction of the relay node in the network does notresult in an increased collision probability as it is the case withany random access protocol. We further notice a decrease inthe collision probability, which is because of the second pathto the AP the relay offers to the network nodes. This secondpath helps the different nodes empty their queues at a fasterrate, hence, nodes do not have to access the channel as oftenas in the case without cooperation, which reduces the collisionprobability. Finally, Fig. 8 demonstrates the effect of cooper-

1702 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 30, NO. 9, OCTOBER 2012Service rate "μ s"350300250200150CSMA/CACoop.1000 5 10 15 20 25Arrival rate "λ " sFig. 8. Source node’s service rate vs. arrival rate for N =15.ation on how fast nodes’ queues get empty by comparing thesource node’s service rates under both cooperative and noncooperativeprotocols. An average increase of 28% is observedin the service rate, which interprets the reduction achieved inthe collision probability.VIII. CONCLUSIONSIn this paper, we have proposed a novel cooperative protocolfor IEEE 802.11 based wireless random access networks.Through cooperation, the proposed protocol mitigates thedetrimental effects of wireless channel errors on the performanceof CSMA/CA random access protocol. Cooperation isachieved by deploying a relay node that will help differentnetwork nodes to forward their packets to the AP. By virtue ofthe relay’s channel access mechanism, the increase in collisionprobability associated with the addition of more nodes to thenetwork is mitigated.The protocol’s performance is thoroughly investigated andcompared to the non-cooperative CSMA/CA protocol. Performancecharacterization is achieved through the developmentof a Markov model coupled with queuing analysis of thenetwork operation. The developed Markov model accuratelydescribed the network dynamics in the presence of relay, andcaptured the interactions between different network nodes.Results revealed a significant improvement in terms of themaximum achievable arrival nodes’s rates, delay, and thenumber of nodes supported by the network.REFERENCES[1] J. N. Laneman, D. N. C. Tse, and G. W. Wornell, “Cooperative diversityin wireless networks: efficient protocols and outage behavior,” IEEETrans. Inf. Theory, vol. 50, pp. 3062–3080, Dec. 2004.[2] W. Su, A. K. Sadek, and K.J.R. Liu, “Cooperative communicationsin wireless networks: Performance analysis and optimum power allocation,”Wireless Personal Communications, vol. 44, no. 2, pp. 181–217,Jan. 2008.[3] K.J.R.Liu,A.K.Sadek,W.Su,andA.Kwasinski, CooperativeCommunications and Networking, Cambridge University Press, 2008.[4] A. K. Sadek, K. J. R. Liu, and A. Epheremides, “Cognitive multipleaccess via cooperation: Protocol design and performance analysis,”IEEE Trans. Inf. Theory, vol. 53, no. 10, pp. 3677–3696, Oct. 2007.[5] A. A. El-Sherif, A. Kwasinski, A. K. Sadek, and K. J. R. Liu,“Content-aware multiple access protocol for cooperative packet speechcommunications,” IEEE Trans. Wireless Commun., vol. 8, no. 2, pp.995–1005, Feb. 2009.[6] R. Lin and A. P. Petropulu, “A new wireless medium access protocolbased on cooperation,” IEEE Trans. Signal Process., vol. 52, no. 12,pp. 4675–4684, Dec. 2005.[7] M. K. Tsatsanis, R. Zhang, , and S. Banerjee, “Network-assisteddiversity for random access wireless networks,” IEEE Trans. SignalProcess., vol. 48, pp. 702–711, March 2000.[8] IEEE Computer Society LAN MAN Standards Committee, WirelessLAN medium access control (MAC) and physical layer (PHY) specifications,IEEE Standard 802.11-1999, New York, NY: IEEE, 1999.[9] P. Chatzimisios, A.C. Boucouvalas, and V. Vitsas, “Performance analysisof ieee 802.11 dcf in presence of transmission errors,” in Proc. Intl.Conf. on Comm. (ICC), June 2004, pp. 3854–3858.[10] Y. Zheng, K. Lu, D. Wu, and Y. Fang, “Performance analysis of ieee802.11 dcf in imperfect channels,” IEEE Trans. Veh. Technol., vol. 55,no. 5, pp. 1648–1656, Sep. 2006.[11] G. Bianchi, “Performance analysis of the ieee 802.11 distributedcoordination function,” IEEE J. Sel. Areas Comm., vol. 18, no. 3, pp.535–547, Mar. 2000.[12] Y. Xiao, “Saturation performance metrics of the 802.11 mac,” in Proc.IEEE Veh. Technol. Conf., Oct. 2003, vol. 3, pp. 1453–1457.[13] K. Duffy, D. Malone, and D. J. Leith, “Modelling of 802.11 distributedcoordination function in non-saturated condition,” IEEE Commun. Lett.,vol. 9, no. 8, pp. 715–718, Aug. 2005.[14] D. Malone, K. Duffy, and D. Leith, “Modelling the 802.11 distributedcoordination function in nonsaturated heterogeneous conditions,”IEEE/ACM Trans. Netw., vol. 15, no. 1, pp. 159–172, Feb. 2007.[15] G. R. Grimmett and D. R. Stirzaker, Probability and Random Processes,Oxford University Press, 2001.[16] Ronald W. Wolff, Stochastic Modeling and the Theory of Queues,Prentice Hall, 1989.Amr El-Sherif (S’ 00, M’ 08) received the (with highest Honors) and the M.Sc. degreein electrical engineering form Alexandria University,Alexandria, Egypt in 2002 and 2005, respectively.He received the Ph.D. degree in electrical engineeringfrom the University of Maryland, College Park,in 2009.He is currently an Assistant Professor in theElectrical Engineering Department at AlexandriaUniversity, Egypt. His research interests includecooperative communications and networking, crosslayerdesign for wireless networks, multiple access technologies for wirelessand sensor networks, and spectrum sharing and cognitive radio systems.K. J. Ray Liu (F’03) is named a DistinguishedScholar-Teacher of University of Maryland, CollegePark, in 2007, where he is Christine Kim EminentProfessor in Information Technology. He serves asAssociate Chair of Graduate Studies and Researchof Electrical and Computer Engineering Departmentand leads the Maryland Signals and InformationGroup conducting research encompassing broad aspectsof wireless communications and networking,information forensics and security, multimedia signalprocessing, and biomedical engineering.Dr. Liu is the recipient of numerous honors and awards including IEEESignal Processing Society Technical Achievement Award and DistinguishedLecturer. He also received various teaching and research recognitions fromUniversity of Maryland including university-level Invention of the YearAward; and Poole and Kent Senior Faculty Teaching Award and OutstandingFaculty Research Award, both from A. James Clark School of Engineering.An ISI Highly Cited Author in Computer Science, Dr. Liu is a Fellow ofIEEE and AAAS.Dr. Liu is President and was Vice President - Publications of IEEE SignalProcessing Society. He was the Editor-in-Chief of IEEE Signal ProcessingMagazine and the founding Editor-in-Chief of EURASIP Journal on Advancesin Signal Processing.

More magazines by this user
Similar magazines