CRDSA, CRDSA++ and IRSA: Stability and Performance Evaluation Alessio Meloni and Maurizio Murroni DIEE - Department of Electrical and Electronic Engineering University of Cagliari Piazza D’Armi, 09123 Cagliari, Italy Email: {alessio.meloni}{murroni}@diee.unica.it Abstract—In the recent past, new enhancements based on the of these new enhancements, also the study of the stability of 5 well established Aloha technique (CRDSA, CRDSA++, IRSA) thesenewtechniquesisofbiginterest.Thereforetheobjective 1 have demonstrated the capability to reach higher throughput of the study shown in this paper is the development of a 0 thantraditionalSA,inburstytrafficconditionsandwithoutany model able to evaluate the stability for these new techniques 2 needofcoordinationamongterminals.Inthispaper,retransmis- sionsandrelatedstabilityforthesenewtechniquesarediscussed. when retransmissions are considered. To do so, we refer to n A model is also formulated in order to provide a basis for the [6] in which a performance evaluation of Slotted Aloha was a J analysis of the stability and the performance both for finite and presented. In particular, the mentioned paper analyzes system 3 infiniteuserspopulation.Thismodelcanbeusedasaframework stabilitybehaviourthroughthedevelopmentofanalyticmodels for the design of such a communication system. 2 based on the so called equilibrium contour, that represents those points for which the expected number of successful ] I. INTRODUCTION packets is equal to the average number of newly generated T Since its birth more than 40 years ago, ALOHA [1] [2] packets, so that the overall communication is stable (i.e. the I . has established as one of the most well-known protocols expected number of packets sent is the same at any time). In s c for packet-switching wireless networks. Since then, several the followings, we adapt this model to our case to provide a [ papers have been published on this random access technique. usefultoolforthestudyofthestabilityinCRDSA,CRDSA++ 1 Some papers are concerned with the issue of improving the and IRSA. Moreover, a framework for the optimization will v throughput(e.g.providingslotsforsynchronizedtransmission be introduced. The starting point and the aim of this paper 9 from users [3] or providing diversity by sending the same are similar to those in [11]. However, while the graphical 0 packet more than once [4]) while other papers concentrate representation used by Kissling represents the drift of the 8 on the study of the stability [5] [6] and on the retransmission communication as the one in [5], our representation is based 5 0 policiestoensurethatthecommunicationisstable[7].Infact, on the one in [6]. The details for this different representation . when retransmissions are considered, a feedback is created choice will be clear throughout the paper, but the main point 1 0 betweenthetransmitterandthereceiver.Therefore,depending is basically the possibility to graphically distinguish a curve 5 on the arrival rate of packets to be transmitted and on the influenced by the retransmission probability and a load line 1 PacketLossRatio,thepossibilityofanoverloadedorsaturated determined by the number of users and their probability of v: channelispresent.Forthisreason,studiesonstabilityofsuch transmission so that these dependencies are separated and i a communication scenario and on possible policies to have it is easier to understand how the communication behaviour X a channel working in optimal conditions have been of big changes when changing those values. Moreover, in [11] the ar interest. Recently, a new technique named Contention Reso- goal of showing agreement between simulations and expected lution Diversity Slotted Aloha (CRDSA) has been introduced analytical drift is reached by averaging simulations results [8]. This technique allows to reach values of throughput up over a consistent number of trials for any possible initial to 0.55pkt using an approach similar to Diversity Slotted state (in terms of number of backlogged users). In this paper slot Aloha[4](i.e.everypacketissenttwiceindifferentslots)and instead, the result of single simulations with various settings adding a Successive Interference Cancellation (SIC) process are presented, aiming at showing the outcomes in terms of at the receiver in order to attempt restoring collided packets backlogged users, throughput and so on, frame after frame (details of how this is done will be illustrated in Section II). and starting from the initial state with no backlogged users. Afterwards,thesameconcepthasbeenextendedtomorethan This allows to empirically understand how and how much the twoinstances1perpacket(CRDSA++)[9]andalsotoirregular communication values move around the expected behaviour. number of packet repetitions (IRSA) [10]. As a consequence Finally, an average packet delay analysis is also introduced. (cid:13)c2012 IEEE. The IEEE copyright notice applies. DOI: 10.1109/ASMS- II. SYSTEMOVERVIEWANDPROBLEMSTATEMENT SPSC.2012.6333080 Consider a multi-access channel populated by a total num- 1Theterminstancesisusedtowardsthepaperwiththemeaningof”total numberloftransmissionsforthesamepacketindifferentslots”. ber of users M (finite or infinite). Users are synchronized so unsuccessful packet in one of the successive frames with USER 1 probability p . In the remainder of this paper, we analyze r USER 2 the conditions that ensure stability of the overall transmission USER 3 when using such a policy and we give a simple yet effective USER 4 tool to design such a communication scenario in relation to theexpectedthroughputandpacketdelaydistributiongathered N f from the model. Fig. 1: Example of frame at the receiver for CRDSA (2 instances III. STABILITYMODEL perpacket).Plainslotsindicatethatatransmissionoccurredforthat Considertheaforementionedrandomaccesscommunication user in that slot. system.Eachusercanbeinoneoftwostates:Thinking(T)or Backlogged (B). Users in T state are idle users that generate a packet in a frame interval with probability p ; if they do, that the channel is divided into slots and N consecutive slots 0 f no other packets are generated until successful transmission are grouped to constitute a so called frame. The probability for that packet has been acknowledged. Users in B state are that, at the beginning of a frame, an idle user has a packet users that failed transmitting their packet, therefore they are to transmit is p . When a frame starts, users having a packet 0 waiting either to retransmit their unsuccessfully transmitted to transmit place l instances of their packet over the N slots f packetwithprobabilityp atthebeginningofeachframe(thus of the frame. The number of instances l can be either the r geometrically distributed) or to receive a feedback about the same for each packet (regular burst degree distribution2) [8] outcome of their retransmission. Moreover, we assume that [9] or not (irregular burst degree distribution) [10]. Packet usersareacknowledgedaboutthesuccessoftheirtransmission instances are nothing else than redundant copies as in [4] at the end of the frame. except for the fact that each instance contains a pointer to Let’s define the location of the other instances. These pointers are used in order to attempt restoring collided packets at the receiver by • NBj : backlogged packets at the end of frame j mFiegaunrseo1f,SreupcrceessesnivtienIgntaerffrearmeneceatCtahnecreellcaetiivoenr(fSoIrCt)h.eCocansseidoerf • GjB = NB(jN−f1)pr : expected channel load of frame j due to users in B state 2 instances per packet (CRDSA). Throughout the paper, we assume perfect interference cancellation and channel estima- • GjT : expected channel load of frame j due to users in T tion, which means that the only cause of disturbance for the state correct reception of packet’s instances is interference among • GjIN =GjT +GjB : expected total channel load of frame them. Moreover, FEC and possible power unbalance are not j considered. Each slot can be in one of three states: • no packet’s instances have been placed in a given slot, • PLRj(GjIN,Nf,d,Imax):expectedaveragepacketloss ratio of frame j, with dependence on the expected total thus the slot is idle; channel load G , the frame size N , the burst degree • only 1 packet’s instance has been placed in a given slot, IN f distribution d and the maximum number of iterations for thus the packet is correctly decoded; the SIC process I • more than 1 packet’s instance has been placed in a given max slot,thusresultingininterferenceofallinvolvedpackets. • GjOUT = GjIN(1−PLR(GjIN,Nf,d,Imax)) : part of Ifatleastoneinstanceofacertainpackethasbeencorrectly load successfully transmitted in frame j, i.e. throughput. received (see User 4), the contribution of the other instances Our aim is to find the equilibrium contour in the (N ,G ) of the same packet can be removed from the other slots. B T plane[6],definedasthelocusofpointsforwhichatanyframe, This process might allow to restore the content of packets the expected channel load due to users in T state is equal to in slots where intereference occurred (see User 2) and in an the expected throughput. Thus iterativemannerotherpacketsmaybecorrectlydecoded,upto a point in which no more packets can be restored or until the G =G =G (1−PLR(G ,N ,d,I )) (1) T OUT IN IN f max maximumnumberofiterationsfortheSICprocessisreached. where the frame number j has been omitted, since in equilib- Depending on the design choice, the packets that have rium state this condition must hold for any frame. Moreover, not been decoded at the end of the SIC process are either being in an equilibrium point implies that also the expected discarded or a retransmission process is accomplished. In the number of backlogged users remains the same frame after lattercase,afeedbackisneededinordertoinformusersabout frame. Therefore the eventual failure of their transmission so that a certain retransmission policy can be applied. In the retransmission NB =NB(1−pr)+GINNfPLR(GIN,Nf,d,Imax) (2) policy considered in this paper, each user retransmits its from which G PLR(G ,N ,d,I )N 2The burst degreedistribution isdefined asthe probabilitydistribution to N = IN IN f max f (3) haveacertainnumberofinstancesforacertainpacket. B p r 1000 500 500 800 450 450 400 400 Channel saturation 350 600 350 point NB 240000 pr=pr0=.50.75 NB112230505000000 Channel load lineoCphpearoanitnnitnelg NB112230505000000 Channel load line oCphpearoanitnnitnelg p=1 50 50 r 0 00 0.2 0.4 0.6 00 0.2 0.4 0.6 0 0.1 0.2 0.3 0.4 0.5 G G T T G T (a)Stablechannel (b)Unstablechannel(finiteM) Fig. 2: Equilibrium contours for CRDSA with N = 100 slots, f I =20 max p =0.143 p =0.143 0 0 p =0.5 p =1 Equations (1) and (3) completely describe the equilibrium r r contour. In fact, once the configuration parameters N , d, M =350 M =350 f I of the system are set and the retransmission probability max (GG,NG)=(0.49,8.2) (GS1,NS1)=(0.495,3.87) is chosen, the curves are completely described plotting G T B T B T and N for different values of G 3 apart from the M (GU,NU)=(0.44,42.74) B IN T B and p considered. Figure 2 displays examples of equilibrium 0 (GS2,NS2)=(6·10−3,346.4) contour for various p values in the CRDSA case. As we can T B r see, when the probability of retransmission on a given frame decreases, the equilibrium contour moves upwards so that the the maximum expected throughput is obtained for a bigger 500 500 mean number of backlogged users NB. 450 450 A.LDetefiunsitiostnudoyf Stthaebilciotynditions under which the described 345000 d line 345000 sCathpuaornaintniteoln tedcshyxoeespsntcesrertecmialtbnaeetidtisdo(cisnb.hteyaab.bnestlhntteawee.tliIesoneionnntapchrGuaeyltTlfeidondulaplneocudwhtt)oaNi.nnOugnBsnseeclfwroesloreMiaancdocTanlenisrnsdittedaa,pietn0rewMcahcoraiecmnahdmnberdefiuepnpneri0encedtasti,oetritenobhltnyees NB112230505000000 oCphearantninelg Channel loa NB112230505000000 Channel load line scenario. For the finite population case, the channel load line point 50 50 can be defined as 0 0 0 0.2 0.4 0.6 0 0.2 0.4 0.6 M −N G G G = Bp (4) T T T N 0 (c)Unstablechannel(infiniteM) (d)Overloadedchannel f while for M → ∞ the channel input can be described as a λ=0.4 p =0.18 0 Poisson process with expected value λ [2] so that G = λ p =0.5 p =1 T r r for any N , i.e. the expected channel input is constant and B M →∞ M =350 independent on the number of backlogged packets. Consider Figure 3, representing various scenarios for (GS1,NS1)=(0.4,1.7) (GS,NS)=(6·10−3,346.4) T B T B CRDSA with N = 100 slots and I = 20. Equilibrium f max (GU,NU)=(0.4,107.32) contours divide the (N ,G ) plane in two parts and each T B B T channel load line can have one or more intersections with (GS2,NS2)=(0.4,∞) T B the equilibrium contour. These intersections are referred to 3Concerning the values used for the Packet Loss Ratio, it is known from Fig. 3: Examples of stable and unstable channels for CRDSA with the literature [8] [10] that the relation between PLR(GIN) and GIN can Nf = 100 and Imax = 20. Stable equilibrium points are marked not be easily modelled in an analytical manner. For this reason PLR values with a black dot. usedinthisworkaretakenfromsimulations. as equilibrium points. The rest of the points of the channel in the design phase. Finally Figure 3d shows the case of an load line will belong to one of two sets: those on the overloaded channel. In this case there is only one equilibrium left of the equilibrium contour represent points for which point corresponding to the channel saturation point. As for G > G , thus situations that yield to decrease of the unstable channels with finite M, this channel can be rendered OUT T backlogged population; those on the right represent points for stable decreasing p . Now, from the equilibrium contour we r which G < G , thus situations that yield to growth of know that the point for the maximum expected throughput is OUT T the backlogged population. (Gmax,Nmax), therefore the communication channel can be T B From the considerations above, we can gather that an designed using Equation (4): intersection point where the channel load line enters the left part for increasing backlogged population corresponds to a M −Nmax Gmax = B p (5) stable equilibrium point, since it acts as a sink. In particular, T N 0 f iftheintersectionistheonlyone,thepointisagloballystable equilibrium point (indicated as GGT,NBG), while if more than given that (M/Nf) · p0 ≥ Gmax must hold since the one intersection is present, it is a locally stable equilibrium slope (known in the literature as m) of the channel load line point (indicatedasGST,NBS).Ifanintersectionpointentersthe must be a negative value. As an example, if p0 is a fixed right part for increasing backlogged population, it is said to and not modifiable design constraint, the number of users M be an unstable equilibrium point (indicated as GU,NU) in the that ensures maximum throughput can be calculated from (5). T B sensethatassoonasastatisticalvariationfromtheequilibrium However the stability and the average packet delay need to be point occurs, the communication will diverge in one of the computed in order to verify they fulfill the design constraints. two directions of the channel load with equal probability (as The former can be verified as illustrated above, the latter can claimed in [11] ). be computed as shown in the next section. Figure 3a shows a stable channel. The globally stable equilibrium point can be referred as channel operating point IV. PACKETDELAYMODEL in the sense that we expect the channel to operate around that point. With the word around we mean that due to statistical Assuming that a channel is operating at its channel op- fluctuations, the actual G and N may differ from the erating point, we would like to know what is the expected T B expected value, however numerical results will show that distribution and mean delay associated to successfully trans- averaging over the entire history of the transmission, values mittedpackets.Thiscanbeentirelydescribedusingadiscrete- close to the expected ones are obtained. Figures 3b and 3c time Markov chain with two states (Figure 4). The two show unstable channels respectively for finite and infinite states represent the state of a generic user that has a packet number of users. Analyzing this two figures for increasing to transmit at the end of each frame. Therefore a frame number of backlogged packets, the first equilibrium point is a duration is our discrete time unit for this Markov chain. The stable equilibrium point. Therefore the communication will edges emanating from the states represent the state transi- tend to keep around it as for the stable equilibrium point tions occurring to users. These transitions depend on pr and in Figure 3a, and we can refer to it once again as channel PLR(GOIN,Nf,d,Imax). If a packet transmitting for the first operating point. However, this is not the only point of equi- time receives a positive acknowledgement (ACK) the user librium since more intersections are present. Therefore, due stays in T state while in case of negative acknowledgement to the abovementioned statistical fluctuations, the number of (NACK) the user switches to B state until successful retrans- backloggeduserscouldpassthesecondintersectionandreturn missionhasbeenachievedandacknowledged.Therefore,here to the right part of the plane, causing an unbounded increase the packet delay Dpkt is considered as the number of frames of the expected number of backlogged users in the case of thatelapsefromthebeginningoftheframeinwhichthepacket infinite M (Figure 3c) or an increase till a new intersection was transmitted for the first time, till the end of the one in pointisreachedinthecaseoffiniteM (Figure3b).Inthelatter which the packet was correctly received. case, this third intersection point is another stable equilibrium point known as channel saturation point, so called because it 1-PLR PLR (1-p)+(PLR p) is a condition in which almost any user is in B state and r r G approaches zero. In the former case of infinite M, OUT N will increase indefinitely and we can say that a channel T B B saturation point is present for N → ∞. Notice that if M B is finite, a stable channel can always be achieved using a sufficiently small value of p (the channel in Figure 3a has (1-PLR)p r r the same parameters as the one in Figure 3b except for p r that is 0.5 instead of 1). However, when p gets smaller, the Fig. 4: Markov Chain for the Packet Delay analysis r corresponding average packet delay could get larger as we will show in Section V. Therefore a tradeoff between stability According to the definition given above, the delay distribu- of the channel and average packet delay need to be faced tion is entirely described by 0.7  1−PLR, for n=1 0.65  0.6 Pr{D =n}= pkt ·P[1L−Rp[pr+(P1L−RPpLR]n)−]·2 , for n>1 ]OUT0.55 r r G (6) ut [ 0.5 while the average expected delay is ghp 0.45 u o (cid:88)∞ Thr 0.4 Av[D ]= n·Pr{D =n} (7) pkt pkt 0.35 n=1 Av[G ]=0.53 Av[G ]=0.486 IN OUT Figure5showssomeexamplesofdelaydistribution.Inthis 0.3 Av[PLR]=0.075 Av[N]=8.4 particular case, when pr gets smaller the distribution spreads 0.25 B over bigger values of delay and even though the probability 0 200 400 600 800 1000 Frame sequence number [j] that a packet is successfully transmitted at the first attempt increases, the expected average packet delay gets higher. This Fig. 6: Simulated throughput for CRDSA with Nf = 100 slots, can be also explained noting that in this example, when Imax =20, p0 =0.143, pr =0.5, M =350 decreasing p the throughput decreases while the number of r backlogged packets increases so that a packet will require more time to be successfully transmitted. Let us call Gmax duetotheaforementionedstatisticalvariations,thethroughput OUT the maximum throughput achievable. If M and p are such oscillates around a certain value. This value is the equilib- 0 that (M ·p )/N ≤ Gmax, the average packet delay at the rium point. In fact the horizontal line represents the value 0 f OUT channel operating point will always increase when decreasing Av[GOUT] = 0.486, that is the value obtained averaging the p , since the equilibrium contour moves upwards. throughput over the entire simulation and is, as expected, a r value really close to the one claimed in Figure 3a. Moreover, also the related average packet delay has been found to agree 100 with the analytical results.4 p=1 , PLR=0.07 , Av[D ]=1.075 r pkt p=0.5 , PLR=0.067 , Av[D ]=1.143 r pkt 0.8 p=0.15 , PLR=0.055 , Av[D ]=1.388 r pkt 0.7 10−2 } 0.6 Pr{Dpkt NB2300 G]OUT0.5 10 ut [ 0.4 p 10−4 0.44 0.46 0.48 0.5 0.52 gh GT ou 0.3 hr T 0.2 2 4 6 8 10 12 14 16 18 Dpkt 0.1 Fig. 5: Delay distribution for CRDSA with N =100 slots, f 0 I =20, M =350 and p =0.143 0 200 400 600 800 1000 max 0 Frame sequence number [j] Fig. 7: Simulated throughput for CRDSA with N = 100 slots, f V. NUMERICALRESULTS I =20, p =0.143, p =1, M =350 max 0 r Inthissection,theresultsofsimulationsareshowninorder to validate the stability model described above. The simulator Figure 7 illustrates an outcome of simulations for the has been built according to the system description given communication scenario of the unstable channel described in in Section II, therefore perfect interference cancellation and Figure 3b. In this case, the initial behaviour of the channel channel estimation have been assumed. Moreover, neither the is such that the throughput oscillates around the channel possibility of FEC nor power unbalance have been considered for our simulations. 4In the simulations only the average packet delay has been computed. In Figure 6 shows the result of simulations for a communi- fact,regardingthedelaydistribution,amorecomplicatedsimulatorisrequired in order to trace the identity of each packet so that each individual packet cation scenario with the same parameters as the example of delay is known. Even though this can be accomplished, we have not built stable channel described in Figure 3a. It can be seen that suchasimulatoryet. operatingpoint.However,differentlyfromthepreviousexam- 140 ple, this is not a globally stable equilibrium point. Therefore statistical fluctuations will sooner or later cause divergence 120 fromthechanneloperatingpointandthesubsequentsaturation ]B of the channel with an average throughput that approaches N 100 zero. The time it takes for the channel to diverge from the ers [ channel operating point varies from simulation to simulation. us 80 d In literature [6] the expected time of divergence for Slotted ge AlohaisknownasFirstExitTime(FET).Inourcasethisvalue klog 60 c is not easily computable since a Markov chain describing the a B 40 communication and considering all the transition probabilities needs to be computed. This is computationally costly (as 20 already claimed in [11]) and we have not carried on methods and formulas for its calculation yet. 50 100 150 200 250 300 350 Frame sequence number [j] 0.65 Fig. 9: Number of backlogged users for CRDSA with N = Av[G ]=0.42 Av[G ]=0.41 f 0.6 IN OUT 100slots,Imax =20,λ=0.4,pr =0.5,M →∞withdivergence from the operating point after a certain time Av[PLR]=0.031 Av[N]=2.6 0.55 B ]T OU 0.5 considers the tradeoff between stability, throughput and delay G ut [ 0.45 in order to provide an efficient and stable communication. As hp also suggested by reviewers (that we gratefully acknowledge g ou 0.4 for their suggestions and comments), future work will regard hr T the analysis of a broader range of cases, the use of a different 0.35 number of replicas than 2 in order to better understand this dependency and the study of such a communication scenario 0.3 when introducing FEC codes as well as the possibility of 0.25 packet power unbalance. 0 200 400 600 800 1000 Frame sequence number [j] REFERENCES Fig. 8: Simulated throughput for CRDSA with N = 100 slots, f [1] N. Abramson, ”The aloha system: Another alternative for computer Imax = 20, λ = 0.4, pr = 0.5, M → ∞ in the case before communications”,inProceedingsofthe1970FallJointComput.Conf., divergence from the operating point occurred AFIPSConf.,vol.37,Montvale,N.J.,1970,pp.281-285. [2] N. Abramson, ”The throughput of packet broadcasting channels”, IEEE Trans.Comm.,vol.25,pp.117-128,Jan.1977. Finally, Figure 8 illustrates an outcome of simulations for [3] L.G. Roberts, ”ALOHA packet systems with and without slots and the case illustrated in Figure 3c, that is the case of infinite capture”,ARPANETSystemNote8(NIC11290),June1972. population. In particular, this outcome shows an occurrence [4] G.L. Choudhury and S. S. Rappaport, ”Diversity ALOHA - A random accessschemeforsatellitecommunications”,IEEETrans.Comm.,vol.31, in which divergence from the channel operating point has pp.450-457,Mar.1983. not occurred yet. As we can see, as long as the commu- [5] Carleial, A.; Hellman, M.; , ”Bistable Behavior of ALOHA-Type Sys- nication takes place around the channel operating point, the tems,” Communications, IEEE Transactions on , vol.23, no.4, pp. 401- 410,Apr1975 same throughput and delay considerations as for the stable [6] Kleinrock,L.; Lam, S.;, ”PacketSwitching ina MultiaccessBroadcast channel are valid. This example highlights that depending on Channel:PerformanceEvaluation,”Communications,IEEETransactions the communication parameters, even though the channel is on,vol.23,no.4,pp.410-423,Apr1975 [7] Lam, S.;Kleinrock, L.;, ”PacketSwitching ina MultiaccessBroadcast unstable, the FET could be so big that instability might be Channel: Dynamic Control Procedures,” Communications, IEEE Trans- acceptabledependingonapplication.However,ifthisisnotthe actionson,vol.23,no.9,pp.891-904,Sep1975 case,thecommunicationwillsoonexitthestabilityregionand [8] Casini, E.; De Gaudenzi, R.; Herrero, Od.R.; , ”Contention Resolution Diversity Slotted ALOHA (CRDSA): An Enhanced Random Access thenumberofbackloggeduserswillgrowfastandindefinitely SchemeforSatelliteAccessPacketNetworks,”WirelessCommunications, as shown in Figure 9. IEEETransactionson,vol.6,no.4,pp.1408-1419,April2007 [9] De Gaudenzi, R.; del Rio Herrero, O.; , ”Advances in Random Access VI. CONCLUSIONS protocols for satellite networks,” Satellite and Space Communications, 2009. IWSSC 2009. International Workshop on , vol., no., pp.331-336, In this paper a model for CRDSA based on equilibrium 9-11Sept.2009 contourandextandablealsotoCRDSA++andIRSAhasbeen [10] Liva,G.;,”Graph-BasedAnalysisandOptimizationofContentionRes- proposed. This model allows to study and predict the stability olutionDiversitySlottedALOHA,”Communications,IEEETransactions on,vol.59,no.2,pp.477-487,February2011 and to calculate the expected throughput as well as the aver- [11] Kissling, C.; , ”On the Stability of Contention Resolution Diversity age and expected distribution delay at the channel operating Slotted ALOHA (CRDSA),” Global Telecommunications Conference point, thus allowing a design that based on design constraints (GLOBECOM2011),2011IEEE,vol.,no.,pp.1-6,5-9Dec.2011