Multiband CSMA/CA with RTS-CTS Strategy Baher MAWLAWI1,2,3, Jean-Baptiste DORE´1, Nikolai LEBEDEV2,3,4, Jean-Marie GORCE2,3 1 CEA-Leti Minatec, 17 rue des Martyrs, 38054 Grenoble Cedex 9, France 2University of Lyon, INRIA 3INSA-Lyon, CITI-INRIA, F-69621, Villeurbanne, France 4CPE Lyon, BP 2077, F-69616, France {baher.mawlawi, jean-baptiste.dore}@cea.fr {[email protected], [email protected]} 5 Abstract—We present in this paper a new medium access CSMA/CA to the multiband case [3] [4] [5] in order to 1 control (MAC) scheme devoted to orthogonal frequency division increasetheglobaldatarate.Intheseprotocols,usersaremul- 0 multiple access (OFDMA) systems which aims at reducing tiplexed through different channel while keeping the classical 2 collision probabilities during the channel request period. The n proposed MAC relies on the classical carrier sense multiple CSMA/CA strategy in each channel. a access/collision avoidance (CSMA/CA) protocol with RTS / CTS Other works tried to eliminate collisions between control J (”RequestToSend”/”ClearToSend”)mechanism.Theproposed and data packets by separating physically the control and the 7 method focus on the collision probability of RTS messages data planes: one band is reserved for control packets and the exploitingamulti-channelconfigurationforthesemessageswhile rest for data transmissions [6] [7] [8]. This scheme provides a ] using the whole band for data transmissions. The protocol may I be interpreted as an asynchronous frequency multiplexing of higher throughput compared to the classical protocol adopted N RTSmessages.Thismethodachievesstrongperformancegainsin in 802.11 standard. However, it suffers from two issues when . termsofthroughputandlatencyespeciallyincrowdednetworks. the network is crowded or lightly busy. In crowded situations, s c the classical CSMA/CA still runs on a common channel and [ suffersfromcollisionsbetweencontrolmessages.Inlowtraffic 1 IndexTerms—Carriersensemultipleaccess/collisionavoidance conditions, high rates users are penalized because they cannot v (CSMA/CA), multiband, throughput, MAC protocol. transmit simultaneously on several channels, even if several 0 ones are free. 3 I. INTRODUCTION 4 We propose in this paper to adapt the CSMA/CA with 1 In recent years, the fast increasing demand for high-speed RTS/CTS mechanism to address both issues. We prove by 0 wireless internet access motivated researchers to make efforts simulationsthattheoutcomeoftheproposedprotocolinterms . 1 for improving the efficiency of decentralized wireless net- of saturation throughput is better than the single band case 0 works.Thedevelopmentofnumerousnewservicesonwireless and it remains quasi-constant for dense networks. The system 5 terminals indeed lead to a strong expansion of the number of delay is improved as well. 1 : users causing an important deterioration of these networks in The paper is outlined as follows. We describe and justify v terms of throughput and system performance [1] [2]. the proposed protocol in Section II and the system model is i X The traditional single band CSMA/CA system has the derived. Section III presents different scenarios exploiting the r advantageofrequiringneithersignalingforbandwidthrequest proposed protocol. Simulation results are presented in Section a nor planned allocation. However, its effectiveness degrades IV and the protocol performance is analysed. Finally, section rapidly with the increasing number of simultaneous source V is reserved for conclusion. nodes. This limitation can be overcome by using multiple division access on different bands where several source nodes II. SYSTEMMODEL can transmit simultaneously. Sources are familiar with the As described in the introduction, we consider a CSMA/CA availability status of each band at each time instant. This protocol with RTS/CTS scheme [9]. Actually, the throughput multipleaccessondifferentbandsmayoperateswithOFDMA is closely related to the collision rate between users [10]. (OrthogonalFrequencyDivisionMultipleAccess)wherebythe Consideringanidealchannel,collisionsmayoccuronlyduring spectral resource (bandwidth) is divided into several orthog- RTSs transmissions. Sending RTS on orthogonal bands may onal sub-carriers. This set of sub-carriers is further split into help to reduce drastically the collision probability. In this subsets, each subset constituting one communication channel. paper, we consider orthogonal frequency multiplexing for Source nodes then compete for accessing and sharing these these RTS messages. resources in both time and frequency. We consider a spectrum divided into N bands. We assume Different works already proposed to generalize the that RTS messages have the same time duration for all users Fig.2. MultibandCSMA/CA. which station is allowed to transmit. The bandwidth of CTS messages is N times the bandwidth of RTS messages. The chosen station (STA) sends its data and waits for Acknowledge (ACK) from the AP. Both data and ACK mes- Fig.1. Flowchartoftheproposedprotocol. sages are sent using all the available bandwidth. Upon receipt of all transmitted data (successful transmission), and imme- diately, after a SIFS duration (”Short Inter-Frame Space”), present in the network and that all transmitters (TX) and the destination node sends an ACK (for ”Acknowledgment”). receivers(RX)havetheknowledgeofthebandsizeandcentral The Contention window (CW) is an integer between CWmin positionsofeachband.Wefurtherassumethatthesenodesare and CWmax. CW is initially set to the minimum value: able to work simultaneously on these bands which is made CW = CW . Whenever a source node is involved in min possible by the use of software radio transceivers. a RTS collision, it increases the transmission waiting time The proposed scheme is used to avoid collisions between by doubling the CW, up to the maximum value CWmax. multiple users (source nodes) requesting simultaneously an Conversely, in the case of a successful RTS transmission, the access to the channel. According to this protocol, a source source node reduces its CW to CWmin. node wishing to transmit data should first listen to the com- Figure 2 provides an exemple with four stations: STA0, munication channel. A flow chart of the proposed protocol is STA1, STA2 and STA3, and a single AP. Each STA tries to depicted in Figure 1. send an RTS on a band randomly chosen. STA0 and STA1 If the channel is busy, a period (expressed in number of respectively choose band 2 and band 1 while STA2 and STA3 time slots) of a waiting counter (known as ”backoff counter”) chooseband3.Atthereceiversideacollisionoccursonband is chosen randomly in the interval [0, CW-1], where CW is 3 but the AP detects both RTS from STA0 and STA1. The a contention window. The channel is declared busy if there AP chooses randomly STA0 and sends CTS over all bands exists a signal on at least one band. The backoff counter indicating that STA0 has won the channel access. All STAs is decremented by one each time the channel is detected receive and decode the CTS and only STA0 tries to send its to be available for a Distributed Inter-Frame Space (DIFS) packets during a defined amount of time (several time slots). duration. The wait counter freezes when the channel is busy, ThecommunicationissaidsuccessfulwhenSTA0receivesthe and resumes when the channel is available again for at least ACK from the AP. DIFS time. Whenthebackoffcounterreacheszero,thesourcerandomly chooses one band over the N to send a permission request III. MULTIBANDCSMA/CA-RTS/CTSCASESTUDY message (RTS) to the destination node. It waits for receiving an authorization message (CTS) from the destination node We now explore the benefits and potential issues of this before transmitting data. The destination (AP) listens simul- proposed MAC in regards of the classical problems arising taneously all the bands. If one or more RTS is detected, the with CSMA/CA, such as hidden [11] and exposed nodes [12] APbroadcastsaCTSmessageoverthewholeband,indicating [13]). A. Hidden node The hidden node problem refers to a configuration of three X radio range Y radio range nodes X, R and Y. X can hear R but not Y and Y can hear R but not X. A ”hidden node” scenario results when Y attempts X R Y to transmit while X is transmitting to R, since Y has sensed the channel idle. The node configuration is depicted in Figure R radio range 3. This classical problem is resolved by the handshaking X R Y mechanism(RTS/CTS).Theuseofavirtualcarriersense(also (a) Hidden node scenario knownasNetworkAllocationVector(NAV)scheme)provides D radio range awaytodealwithhiddennodeproblem.WhenaRTSorCTS E S radio range E D radio range isreceivedbynontransmittingnodes,theydefertheirbackoff during a time specified into the RTS/CTS messages. In the D S S D E E case of the proposed protocol no additional mechanisms are required at the MAC layer. At the physical layer the receiver must be able to analyze each band independently for RTS S radio range messages but also to be able to decode the whole band. This is not an issue with OFDM systems. DE SE S D (b) Exposed node scenario Last but not least note that if the classical RTS/CTS mech- anism avoids collisions in the hidden node scenario, it cannot Fig.3. Illustrationofthehiddenandexposednodeproblem deal with collisions between RTS messages themselves. The channel is kept clear only when the CTS has been sent. B. Exposed node no data packet arrives ...), we propose to use the destination RTS/CTS handshake mechanism was introduced to deal identity field already present in the RTS message in order to with the hidden node issue. However this mechanism intro- detect what we call virtual RTS collision. When two or more duces a new problem, known as exposed node. The issue RTS can be decoded, the destination analyzes the identity of of exposed node is depicted in Figure 3. Exposed node S the destination node. If at least two different identities are E can hear the RTS and DATA packets sent out from node S detectedthenacollisionisdeclaredandnoCTSisbroadcasted to D. Consequently, through the virtual carrier sensing, S overthecell.Thiscasedoesnotexistinthecontextofper-AP E can not initiate transmission despite being out of range of single band CSMA/CA-RTS/CTS with frequency reuse since the receiver D. Consequently, the transmission between S each AP send its CTS over its own band. E and D is differed introducing a lost in capacity. The same E problem exists with the proposed protocol but dealing with IV. PROTOCOLAPPLICATION this issue is kept out of the scope of this paper. It is worth In this section we discuss a real application to exhibit the mentioning that some mechanisms have been proposed in the motivation for this work. literature to face the exposed node problem and they could We consider an uplink scenario with a random distribution be transposed to the multiband RTS/CTS CSMA/CA protocol ofuserssharingthesamebandwidthinacellusingCSMA/CA (see [12] for instance). with a RTS/CTS mechanism. As we know, the signal to C. New pathologic case noise ratio (SNR) depends on the user position relatively to the access point. Each user experience a different SNR, and The frequency multiplexing of RTS introduces a new issue accordingly a specific capacity. that can be easily solved by a basic rule. Let us consider the Suppose as described in Figure 4 that there are three users following scenario including four nodes, two sources and two ready to transmit data (backoff equals zero) to the AP. User1 destinations. Source A sends a RTS to node B using band i is close to the AP, user3 is far from the AP and user2 is and at the same time, source C sends a RTS to node D using in the middle. In order to keep the system working properly, bandi+1.NodeB canhearbothAandC,whilenodeD can the time of RTS should be the same regardless of the users hear A or C only1. In this case no RTS collision occurs since channel capacity. Thus, all users are penalized by the farthest RTS messages are sent on different bands. Without particular one since the duration of the RTS should be kept equal to rulethetwodestinationswillrespondCTS.Insomecases,this ensure the proper behavior of the protocol. The fact that users scenario can introduce a CTS collision (since CTS messages with a high SNR does not exploit their whole capacity for are broadcasted over all the bands). To prevent the CTS RTS represents a spectral efficiency loss. collision and its consequences (watchdog timer is required if But with the multi-bands protocol, the duration of the RTS 1whenallnodescanheareachotherthesameproblemoccurs can be kept small by allowing distant users to transmit their Packetpayload 8184bits MACheader 272bits PHYheader 128bits ACKlength 112bits+PHYheader RTSlength 160bits+PHYheader CTSlength 112bits+PHYheader ChannelBitRate 72.2Mbit/s PropagationDelay 1µs SIFS 10µs SlotTime 9µs DIFS 28µs TABLEI PHYLAYERPARAMETERSFOR802.11N 0.7 #bands=1 #bands=2 Fig.4. Multiuserswithuplinkcommunications. 0.6 #bands=3 #bands=4 0.5 #bands=5 y bilit RTS over several bands. As an example, Figure 4 shows that oba0.4 r P user3 (highest SNR) uses one band over five and user2 uses on 0.3 two bands over five. ollisi For example, if user1 send RTS on the fifth band and user2 C0.2 send its RTS on first two bands (considering user3 does not 0.1 transmit),theAPwillbeabletodecodethetwomessagesand choose the qualified user to establish communication. In this 0 0 10 20 30 40 50 60 70 80 90 100 caseweachieveasuccessfultransmission.OnesuccessfulRTS Number of Mobile Stations transmission (band contains only 1 RTS) leads to successful communication. Fig.5. CollisionprobabilityforMultiRTSbands. In this context, if user 3 with the lowest SNR also sends its RTS to the AP, it can use the whole band for its RTS and the protocol becomes equivalent to the classical one. But high to access a common destination. Figure 5 depicts the simula- SNR users may take advantage of the multiband protocol. As tionresultsforthecollisionprobabilitybetweenRTSmessages described in the system model, CTS messages are sent from as a function of the number of mobile stations present in the the AP over all the bands anyway in order to be detectable network for various RTS bands values. and decodable by all users regardless their SNR. These results demonstrate that the collision probability increaseswiththenumberofusersbutisinverselyproportional V. SIMULATIONRESULTS to the number of RTS bands. For a single band CSMA/CA Inthissection,duetothelackofplace,werestrictthestudy to the case where all users have a SNR good enough to be able to use the multiband protocol. We focus our study on the impact of the number of RTS bands on the system performance. A home-made event- driven simulator was used to model the protocol behavior. The protocol and channel parameters are reported in Table I and correspond to those of 802.11n standard. The minimal contention window (W ) has been chosen constant and min equal to 16. It is worth mentionning that as the study focuses on the MAC mechanisms, an ideal physical layer (no path loss, no fading, no shadowing, ...) is considered. A. Collision Probability As the system performance is related to RTS collision Fig. 6. Saturation throughput (bits/sec) vs. number of mobile stations probability,itisinterestingtostudytheimpactoftheproposed consideringmultiRTSbands. band division. We consider different number of sources trying with 50 users, the probability of collision is around 50%. For a two bands protocol the probability of collision is reduced to 25%.When5bandsareconsideredtheprobabilityofcollision is less than 10%. As we discussed before, the proposed protocol reduces drastically the RTS collision probability. As collisions happen only during RTS transmissions (con- sidering perfect channel conditions), the proposed MAC im- proves the global system performance in terms of throughput and latency. B. Saturation Throughput In this sub-Section we study the throughput in saturation mode,sowesupposethateachstationhasalwaysinitsbuffer atleastonepacketreadyfortransmission.Figure6depictsthe saturation throughput as a function of the number of mobile Fig. 7. CDF of access delay with 100 stations for single and multi RTS band.Delayisexpressedinsecond. stations present in the network for various RTS bands values. It shows that increasing the number of RTS bands in the system improves as well the saturation throughput. Global CDF #RTSBands ProposedProtocolDelay(ms) Gain(%) system performance is improved by having the possibility to 99% 2 1.83 69.73 99% 3 1.61 94.46 detectsimultaneousRTSevenifthesystemcandealwithonly 99% 4 1.53 104.65 one RTS. This is due to the reduction of the RTS collision 99% 5 1.48 109.61 probability. 98% 2 1.62 65.29 98% 3 1.39 93.72 The improvement is significant for low and high number 98% 4 1.33 102.19 of users. Table II illustrates the gain introduced in the multi- 98% 5 1.30 105.15 band context. It is demonstrated that the gain becomes more 95% 2 1.28 62.35 95% 3 1.13 85.44 important in loaded networks. This protocol brings more than 95% 4 1.09 92.00 50% of gain (comparing multiband to single band in terms of 95% 5 1.05 97.61 saturationthroughput)whenthenumberofRTSbandsexceeds 90% 2 1.02 61.98 90% 3 0.92 78.45 four in charged mode (loaded network). 90% 4 0.87 88.34 90% 5 0.86 89.21 SaturationThroughput #Stations #RTSBands Gain(%) TABLEIII (Mbits/sec) DELAYGAINWITHTHEPROPOSEDMACFORDIFFERENTRTSBANDS 10 2 24.56 3.57 NUMBERWITH100USERS. 10 3 24.90 5.00 10 4 25.05 5.64 10 5 25.17 6.12 50 2 23.08 13.09 50 3 24.13 18.22 access delay is better for the multiband scheme. For instance, 50 4 24.66 20.84 50 5 25.06 22.77 99% of packets are transmitted with at most 3.13ms by the 100 2 21.73 29.84 single band protocol while they are sent with at most 1.53ms 100 3 23.53 40.56 by our proposed MAC protocol with 4 RTS bands. 100 4 24.51 46.42 100 5 25.11 50.04 Different gain values introduced by the proposed MAC are TABLEII reportedintableIII.Thegainiscomputedbycomparingboth SATURATIONTHROUGHPUTGAINWITHTHEPROPOSEDMACFOR single and multiband CDF. As seen in Table III, using multi- DIFFERENTMOBILESTATIONSANDRTSBANDSNUMBER. bandprotocol,thedelayisreducedbyhalfinloadednetworks (considering more than 3 RTS bands). This improvement is explained by the fact that the proposed protocol reduces the C. Statistical Delay Study collision probability between RTS, hence packets wait less To complete the study we go forward to simulate the before to be transmitted. delayintroducedbytheproposedMACprotocol.Thedelayis Figure 8 depicts the saturation throughput and delay gains defined as the duration needed to transmit a packet. In order (%)vs.thenumberofRTSbandsandmobilestations.Itshould to compare the delay between the two strategies (single and be noticed that the gain in terms of saturation throughput and multiband),weextractfromsimulationthecumulativedensity delay are always positives and becomes much important in function (CDF) of the delay for one network scenario and for the case of loaded networks. Increasing the number of RTS manynumberofusers.Figure7illustratesthattheCDFofthe bands improves the system performance (as the RTS collision [6] S.Basagni,C.Petrioli,R.Petroccia,andM.Stojanovic,“Multiplexing data and control channels in random access underwater networks,” in OCEANS2009,MTS/IEEEBiloxi-MarineTechnologyforOurFuture: GlobalandLocalChallenges,2009,pp.1–7. [7] N. Jain, S. Das, and A. Nasipuri, “A multichannel csma mac protocol with receiver-based channel selection for multihop wireless networks,” inComputerCommunicationsandNetworks,2001.Proceedings.Tenth InternationalConferenceon,2001,pp.432–439. [8] J. Deng, Y. S. Han, and Z. Haas, “Analyzing split channel medium accesscontrolschemes,”WirelessCommunications,IEEETransactions on,vol.5,no.5,pp.967–971,2006. [9] G. Bianchi, L. Fratta, and M. Oliveri, “Performance evaluation and enhancementofthecsma/camacprotocolfor802.11wirelesslans,”in Personal,IndoorandMobileRadioCommunications,1996.PIMRC’96., SeventhIEEEInternationalSymposiumon,vol.2,oct1996,pp.392– 396vol.2. [10] M. H. Manshaei and J.-P. Hubaux, “Performance analysis of the ieee 802.11distributedcoordinationfunction:Bianchimodel,”March2007. [11] F. Tobagi and L. Kleinrock, “Packet switching in radio channels: Part ii–thehiddenterminalproblemincarriersensemultiple-accessandthe busy-tone solution,” Communications, IEEE Transactions on, vol. 23, no.12,pp.1417–1433,dec1975. [12] A.Qayyum,M.U.Saleem,Tauseef-Ul-Islam,M.Ahmad,andM.Khan, Fig. 8. Saturation throughput and Delay vs. number of mobile stations “Performanceincreaseincsma/cawithrts-cts,”inMultiTopicConfer- consideringvariousnumberofRTSbands. ence,2003.INMIC2003.7thInternational,2003,pp.182–185. [13] J.Yao,T.Xiong,andW.Lou,“Eliminationofexposedterminalproblem usingsignaturedetection,”inSensor,MeshandAdHocCommunications andNetworks(SECON),20129thAnnualIEEECommunicationsSociety Conferenceon,2012,pp.398–406. probability is reduced). Moreover, this protocol improves the system latency since the collision probability and at the same time the number of contented users reduce. VI. CONCLUSION In this paper, we proposed an innovative scheme exploiting arandomfrequencydivisionmultiplexingofRTSmessagesin a CSMA/CA RTS/CTS access method. This technique is characterized by considering a spec- trum which is divided into several bands of known size. We demonstrated that the proposed MAC is very interesting especially in crowded networks. By considering a frequency division multiplexing of RTS messages, the probability of RTS collisions is decreased significantly. We achieved a gain of about 50% in terms of saturation throughput and 109% in terms of delay. Due to these good properties in crowded scenario, the proposed protocol is also a good candidate for wireless Machine to Machine (M2M) applications in which latency is critical. REFERENCES [1] G.Bianchi,“Performanceanalysisoftheieee802.11distributedcoordi- nationfunction,”SelectedAreasinCommunications,IEEEJournalon, vol.18,no.3,pp.535–547,march2000. [2] ——, “Ieee 802.11-saturation throughput analysis,” Communications Letters,IEEE,vol.2,no.12,pp.318–320,dec.1998. [3] H. Kwon, H. Seo, S. Kim, and B. G. Lee, “Generalized csma/ca for ofdma systems: protocol design, throughput analysis, and implementa- tion issues,” Wireless Communications, IEEE Transactions on, vol. 8, no.8,pp.4176–4187,august2009. [4] J.W.Chong,Y.Sung,andD.K.Sung,“Rawpeach:Multibandcsma/ca- basedcognitiveradionetworks,”CommunicationsandNetworks,Jour- nalof,vol.11,no.2,pp.175–186,2009. [5] H. Kwon, S. Kim, and B. G. Lee, “Opportunistic multi-channel csma protocolforofdmasystems,”WirelessCommunications,IEEETransac- tionson,vol.9,no.5,pp.1552–1557,2010.