Sheet Music

Performance evaluation of a multiuser detection based MAC design for Ad Hoc networks

Description
Performance evaluation of a multiuser detection based MAC design for Ad Hoc networks
Categories
Published
of 5
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  Performance Evaluation of A Multiuser Detection Based MAC Designfor Ad Hoc Networks Jinfang Zhang, Zbigniew Dziong, Francois Gagnon and Michel Kadoch Department of Electrical Engineering, Ecole de Technologie SuperieureMontreal, Quebec, Canada, H3C 1K3  Abstract —  In general, the performance and radio resource uti-lization of Ad Hoc networks are limited by half-duplex opera-tion and possible collisions. In this paper, we propose a novelapproach for MAC design that practically eliminates collisionsand significantly increases the bandwidth utilization. The keyelement of this approach is the CDMA multiuser detection tech-nology that allows receiving several signals in parallel withoutinflictingself-interference. Thesefeaturesgiveapromiseofsig-nificant performance improvements. The main goal of this pa-peristoassesstherangeofthisgainwhencomparedtootherex-istingalternatives. Inparticular, wecompare theperformance of the proposed multiuser detection based MAC design with MACdesign based on IEEE 802.11 concept and with MAC designbased on multi-code CDMA with one signal reception by oneuser at a time. Key words:  CDMA Multiuser detection, MAC design, dis-tributed scheduling, Ad Hoc networks.I. I NTRODUCTION Multi-hop mobile Ad Hoc networks have recently been thesubject of extensive research. Nevertheless, the performanceand bandwidth utilization of the proposed solutions are signifi-cantly inferior when compared with the fixed wireless or wirednetworks. Apart mobility, the main reasons for these limitationsare half-duplex operation and possible collisions caused by hid-den/exposed terminal problems. While sophisticated adaptivedirectional antennas give some promise of improvement in thelong term, at the moment, this technology seems to be not ma-ture enough to be considered in a difficult Ad Hoc network en-vironment. In this paper, we consider another technology thatis more advanced and can significantly improve both the perfor-manceandthebandwidthutilization. Namely, recenttechnolog-ical advances allow integration of a CDMA multiuser detectionbased receiver on one chip and therefore, we consider applica-tion of this technology for new MAC design in multi-hop AdHoc networks.There are two main advantages of using multiuser detectionin Ad Hoc networks. First, multiple signals from different mo-biles can be received at the same time, which practically elim-inates the collision problem and can significantly reduce theend-to-end packet transfer delay. Second, the multi-user detec-tion avoids interference between the received signals and thisfeature can increase the bandwidth utilization by a large fac-tor [1][2][3]. Although multiuser detection is known for a longtime, most of the studies focus on the physical layer[3]. Ac-cording to our knowledge, there are no published studies on ap-plication of this technology to mobile Ad Hoc networks. Inthis context, the main goal of this paper is to evaluate potentialgain from application of CDMA multiuser detection comparedto other solutions proposed for Ad Hoc networks. In particu-lar, we use as references MAC design based on IEEE 802.11concept and MAC design based on multi-code CDMA, but withone signal reception by one user at a time.In order to achieve a gain from multiuser detection, a newMAC layer design is required. This design should realize threemain functions. First, distributed dynamic code assignmentthat avoids code collisions is needed. Then, an implementa-tion of distributed scheduling of transmissions and receptionsthat maximizes throughput within the required fairness criteriais required. Finally, the data transmission based on multiuserdetection has to be organized. Each of these functions consti-tutes a research topic sufficient for a separate publication of itsown. Therefore, in this paper, we give a high level presenta-tion that is focused mainly on an conservative assessment of thepotential gain from the application of multiuser detection in AdHoc networks. More detailed study of required protocols andalgorithms and their optimizations will be given in subsequentpublications.The remainder of the paper is organized as follows. In Sec-tion II, we review a multiuser detection model. In the sectionthat follows, we present a framework for the proposed MACdesign. It is based on a synchronous time division CDMA (TD-CDMA) structure where each frame is divided into two controlslots and a continuous data transmission slot. The control slotsare used by a code assignment protocol and a data transmissionscheduling protocol. Section IV describes a distributed schedul-ing mechanism that provides local fairness and priorities forreal time traffic based on time deadlines. The numerical results,presented in Section V, indicate that the proposed MAC designincreases the throughput by more than an order of magnitudewhen compared with MAC designs based on IEEE 802.11 con-cept and by around  100%  when compared with MAC designsbased on multi-code CDMA. At the same time, the performancecharacteristics, such as packet delay and loss, are significantlyimproved. Section VI gives concluding remarks and issues forfuture work.II. L INEAR MULTIUSER DETECTOR Multiuser detection improves the performance of spread-spectrum systems by exploiting the structure of the multi-accessinterference when demodulating the signal of a user. Fig. 1demonstrates the throughput improvement multiuser detectorover one user detector when two users are transmitting simul-taneously. When there is only one communication, the maxi-mum reliable transmission rate is  R 1 . But when there are two  simultaneous communications, with multiuser detection, the re-liable communication is possible at rate pair  ( R 1 ,R 2 )  whichoutperforms the one user communication with  R 1  +  R 2  >max ( R 1 ,R 2 ) . 012345012345 User 1 transmission rate R 1    U  s  e  r   2   t  r  a  n  s  m   i  s  s   i  o  n  r  a   t  e   R    2 multiuser detectorone user detectorthroughput gain Fig. 1. Throughput improvement of multiuser detection We consider a minimum mean square error (MMSE) detectoras it is an optimal linear detector that maximizes the signal-to-interference-plus-noise (SINR) ratio. Under random spreadingsequences, the SINR at node  k  from node  i  is [1] γ  ( i,k ) =  p i σ 2 +  1 L  M j =1 ,j  = i  I  (  p j ,p i ,γ  ( i,k )) (1)where I  (  p j ,p i ,γ  ( i,k )) =  p j  p i  p i  +  p j γ  ( i,k ) is the effective interference from user  j .  p i ,p j  are the receivedpowers from node  i,j  at node  k ;  σ 2 is the noise power at node k ;  L  is the processing gain; and  M   is the number of signals be-ing processed by node  k . This equation holds when the numberof nodes ( M  ) and the processing gain ( L ) both go to infinity,with  M/L  →  θ , a constant. In a real system, the SINR ex-pression is an approximation to the actual SINR. It is noted thatthe approximation becomes more accurate as the system scalesup. This model also gives explanation why the received signalsdo not inflict normal self-interference and this feature providessignificant increase of data transmission throughput.III. F RAMEWORK FOR MULTIUSER DETECTION BASED MAC  PROTOCOLS We assume that each node is equipped with a half-duplexCDMA multiuser detector and that each node is allocated adedicated code for transmitter-based data transmission. When anode enters into the system, it is allocated a dedicated code froma set of predefined dedicated code channels that are not used inthe node’s neighborhood. This allocation can be changed whentwo nodes with the same code are entering the same neighbor-hood in order to avoid a code collision.Due to the nature of multiuser detection, we assume that thetransmission times are synchronized at packet level with thehelp of either in-band signal exchanges or out-of-band solu-tions, such as GPS. The proposed frame structure of such asynchronous time division CDMA (TD-CDMA) system is illus-trated in Fig. 2 where three slots are defined. The first two slots connectivity update N u Scheduling N s data transmissionconnectivity update N u (a)Scheduling N s data transmissionconnectivity update KN u Scheduling N s data transmission(b)Scheduling N s data transmission1 super frame (K frames)K normal frames Fig. 2. Synchronous transmission are associated with control functions and are called connectivityupdate and scheduling slots, respectively. The third slot servesfor a continuous data transmission in the CDMA channels thatare selected during the scheduling slot.The connectivity update slot is used by the connectivity up-date protocol. In this phase, each node broadcasts its identityinformation on the common signaling channel so that the neigh-boring nodes can be detected and code assignment and reas-signment can be realized. Since the connectivity update can beexecuted in a significantly longer time period than the frame cy-cle, there are two ways to do connectivity update. One is todistribute connectivity update function into  K   frames and eachframe contains  N  u  minislots for connectivity update that givein total  KN  u  minislots as shown in Fig. 2 (a). Alternatively,one can introduce a connectivity update slot with  KN  u  minis-lots for every  K   normal frames as illustrated in Fig. 2 (b). Inthis case, the normal frame only have scheduling and data trans-mission slots. Connectivity update slot achieves exclusive codeassignment in neighborhood to avoid hidden/exposed terminalproblems. The broadcasting and code assignment scheme is outof the scope of this paper. The interested readers can refer to [4]for details.Theschedulingslotisassociatedwiththeschedulingprotocoland its objective is to select transmissions that can be realized inthe subsequent data transmission slot. In this phase, the trans-mitting and receiving nodes have to be chosen together withthe data packets to be transmitted and the channel parametersto be used (like power and transmission rate). The transmis-sion selection should take into account the packet priority, theaccess fairness and throughput objectives. A simple schedulingmechanism is presented in Section IV. Note that the schedulingprotocol can use dedicated code channels that are established inthe connectivity update phase. In this way, the problem of col-lisions at receivers which is typical for existing solutions can beeffectively eliminated.As mentioned before, during the data transmission slot, sev-eral transmissions are executed in the same neighborhood bytaking advantage of CDMA multiuser detection. Here we as-sume that multicode CDMA is deployed to support variablerate transmission. When a node needs to transmit at  m -rate,it first converts its date stream, serial-to-parallel, into  m  basic-  rate streams. Each basic-rate stream is spread with a differ-ent code generated by the so called “sub-code concatenation”scheme[5] and superimposed before spread spectrum modula-tion. The sub-code concatenation generated codes are unique toeach mobile.IV. D ISTRIBUTED NEIGHBORHOOD SCHEDULING To facilitate the presentation of the scheduling algorithm, wefirst review the traffic model considered in this paper. We con-sider twoclassesof traffic: realtimevoice and non realtimedata.Eachnoderequirestwobuffersforeachofitsneighboringnodesto accommodate bursty voice and data packets arrivals. Eachof the traffic class has different packet generation rate, SINR,packet loss rate (PLR) and delay requirements. The data trans-mission slot length corresponds to the transmission time of onevoice packet with basic transmission rate on a single code chan-nel. We assume that the data packets can be fragmented or as-sembled to form transmission packets with the same transmis-sion duration as the voice packets. It is also assumed that eacharrivalpacketthatenterstheschedulingsystemisallocatedade-lay bound (time-out) value expressed in number of frames. Thisvalue is decreased by one for every frame interval till the packetis scheduled to transmit or till the value is zero. In the lattercase, the packet is discarded. The rate at which voice packetsare discarded is referred to as the voice packet loss rate. In thefollowing, we refer to the packets with time-out value equal toone as the most urgent packets (MUPs), the packets with time-out value equal to two as the second MUPs, and so on.We assume that the voice packets have higher priority thandata packets and that, among the same class packets, the small-est time-out value gives precedence for transmission. Based onthese rules, each node selects a candidate packet or a set of can-didate packets from the same class and destined to one of itsneighboring nodes for a possible transmission in the next trans-mission slot. Then the node priority for transmission contentionwith other neighboring nodes is determined. Here again, theprecedence is given for voice packets. Then, for nodes withcandidate voice packets, the precedence is given for nodes withhigher overall voice packet loss rate. Analogously, for nodeswith candidate data packets, the precedence is given for nodeswith higher overall data packet delay.To resolve the contention for transmission and select the re-ceiving nodes, we apply exchange of RTS/CTS control mes-sages conducted on dedicated channels. Namely, at the begin-ning of each scheduling slot, each node, with a set of candidatepackets for transmission, sends out an RTS message. A RTSmessage carries the intended receiver ID, the sender ID, a bunchof subcodes which represent the variable transmission rate, andthe node priority which contains two fields. The first field istraffic type, either voice or data; and the second field is the mea-sured packet loss rate for voice traffic and the largest delay fordata traffic. The distribution of RTS messages in the neighbor-hood is not straightforward since the targeted RTS receiver mayalso have an RTS message to be sent. To solve this problem,we propose a protocol that requires only three minislots. In thisapproach, the nodes are divided into black and white nodes insuch a way that in the neighborhood there are at least one black and one white node. Then, in the first minislot, the white nodestransmit the RTS messages on the dedicated channels. Thesemessages are received by all black nodes due to multiuser de-tection. In the second minislot, the black nodes transmit theirown RTS messages together with the RTS messages received inthe first minislot. Finally in the third minislot, the white nodestransmit the RTS messages received from the black nodes in thesecond minislot. Due to the space limitation, we will providefull description of this protocol in a forthcoming publication.After three minislots of RTS exchanges, each node has thetransmission status, either transmission or reception, of itsneighboring nodes. A node that receives at least one RTS mes-sage determines whether it will receive or transmit in the subse-quent transmission slot by comparing its own node priority withthe other contending nodes’ priorities. In case at least one of thereceived RTS messages has higher priority, the node becomes areceiver to all contending neighboring nodes; otherwise, it willtransmit its own candidate packets. Then, the nodes selected asreceivers send out the CTS messages to the senders of receivedRTS messages in one minislot by means of dedicated channels.After the first cycle of RTS/CTS exchange, a second roundof RTS/CTS exchange is executed. Since this time the receiv-ing nodes are determined, the exchange can be executed in twominislots. This additional exchange has several goals. The firstgoal is associated with the fact that in the second round theRTS messages are received directly from its source nodes. Itgives the selected receivers possibility to estimate the propaga-tion loss related to each connection. Also, it allows estimat-ing the interference induced by connections destined to othernodes. In this way, local power control can be implemented atreceivers. Let  M  v  and  M  d  denote the number of neighboringvoice and data connections which have the same destination.Define  α v  =  M  v /L,α d  =  M  d /L . Let  γ  v  and  γ  d  denote theSINR requirements and  p v  and  p d  denote the minimum requiredreceived power to achieve  γ  v  and  γ  d , respectively. If we replaceEq. (1) with these parameters and after some manipulation, theSINR expressions for voice and data traffic can be specified as  γ  v  =  p v σ 2 +( α v − 1 L ) I  (  p v ,p v ,γ  v )+ α d I  (  p d ,p v ,γ  v )+  1 L P i  I  (  p i ,p v ,γ  v ) γ  d  =  p d σ 2 + α v I  (  p v ,p d ,γ  d )+( α d − 1 L ) I  (  p d ,p d ,γ  d )+  1 L P i  I  (  p i ,p d ,γ  d ) . (2)The last item in the divisor is the effective interference fromneighboring nodes whose destinations are other nodes. Forlarge networks,  γ  v /p v  and  γ  d /p d  are a constant[1].Based on Eq. (2), the minimum required received power toachieve  γ  v  and  γ  d  for voice and data traffic can be estimatedat each receiver. With the estimated propagation attenuation,the minimum required transmission power for each subcodechannel is estimated. Under maximum transmission power con-straint, the maximum number of packets that can be supportedin each connection can also be obtained. The transmission per-missions which include the confirmed number of subcodes andthe transmission power for each packet transmission are passedvia the second CTS message to the senders.The second goal of the second RTS/CTS round is to allocatea function to the stranded nodes. The stranded nodes are the  nodes that do not receive CTS messages addressed to them inthe first round since their RTS messages are sent to the nodesthat win the contention for transmission. Note that after the firstround of RTS/CTS messages, each stranded node knows whichfunction was allocated to each node in the neighborhood (re-ceiver or transmitter or stranded). Based on this information,two options are considered. In the first one, each stranded nodeselects new candidate packets from the packets destined to oneofthereceivingnodes. Basedonthisnewselection, thestrandednodesendsanewRTSmessagedirectlytotheselectedreceivingnode. In the second option, the stranded nodes organize com-munication between them. Namely, based on the known neigh-boring stranded nodes priorities, each node determines whetherit becomes a transmitter or a receiver and the transmitters sendRTS message directly to the associated receiver.The scheduling approach proposed in this section is not nec-essarily optimal but is sufficient for a conservative assessmentof the gain from application of the multiuser detection. Moresophisticated schemes will be presented in subsequent publica-tions.V. N UMERICAL RESULTS We consider three major performance measures of the pro-posed neighborhood scheduling scheme. i) voice packet lossrate; ii) average data packets transmission delay; iii) through-put. Average data packets transmission delay is formulated as ( N  loss × D bound  +  m n trans. × D delay ) /N  , where  N  loss  isthe total number of data packet loss,  D bound  is the delay boundfor data packets, n trans.  is the number of successfully transmit-ted data packets with delay of   D delay  in frames. Throughput isdefined as the accumulated number of packets that are success-fully received in one hop’s transmission.The numerical results are obtained by means of a discreteevent simulation that models an Ad Hoc network working at 450 MHz frequency band with the following parameters. Thespreading gain is set to  512 . We focus on a circle simulationarea with radius of   350 m. A mobility model which mimics hu-man and vehicle movement behavior is applied [6]. The speedlimit is  120 km/h. To avoid boundary effect, the nodes movingout of the circle will reenter the simulation area again. We as-sume a reliable wireless communication and a free space propa-gation mode so that the signal attenuation is caused exclusivelyby transmission distance. To promptly update its connectivityinformation for neighborhood identification, we apply the radiostructure as shown in Fig.2 (b) for connectivity update with aninterval of   10 × normal frame length. Each node has the maxi-mum transmission power of   7 w. The neighborhood threshold isset to  10 − 6 w and the noise power is set to  10 − 7 w. Each nodeaccommodates two types of traffic, voice and data with SINRrequirements of   7 dB and  10 dB, respectively. Both the voiceand the data packet flows are simulated by an ON-OFF model.The delay bound for a packet entering a node is set to  2  and  100 frames for voice and data packets, respectively. The simulationruns  10000  frames to obtain the results.Sincethegoalofthepresentedstudyistoassesstheefficiencyof the proposed scheduling scheme, the objective of the appliedpacket generation model is to achieve large and uniform loadingof each node. Therefore, we do not consider the routing algo-rithm and the packets are generated in each node with indicationof the neighboring node to which they should be sent. The ONand OFF durations of packet flows follow an exponential dis-tribution with mean of   1  and  49  frames for voice and  5  and  25 frames for data traffic, respectively. The voice and data pack-ets generation rate at each node is proportional to the numberof neighboring nodes and is assumed to be Poisson distributedwith mean of   0 . 1  and  1  times the number of identified neigh-boring nodes. Every bunch of packets generated in a frame israndomly assigned to a destination chosen from its neighboringnode list. As a destination node may move out of the transmis-sion range of a sender, waiting packets with this destination arereassigned to a destination from the updated neighboring nodelist.We compare the performance of the proposed MAC designbased on multiuser detection with two existing concepts: multi-ple CDMA connections in the neighborhood [7], and one con-nection in the neighborhood as in IEEE 802.11. The three com-munication scenarios are demonstrated in Fig. 3. Note that in    c     2  c 3                        c                                1     c     2                       c                                1     c     2  a. Multiuser reception b. Parallel reception c . One user reception Fig. 3. Simulation scenarios the first two scenarios, there are several parallel connectionsin the neighborhood, but only in the multiuser detection casea node can receive several connections simultaneously. In orderto facilitate the comparison, we assume that all three commu-nication scenarios use the same frame structure as proposed insection III. This corresponds to an assumption that on average,the ratio between the data communication time and the time as-sociated with exchange of control packets is the same in all sce-narios. 2468101214160123456x 10 4 Average node degree    V  o   i  c  e  p  a  c   k  e   t   t   h  r  o  u  g   h  p  u   t   (  p  a  c   k  e   t  s   )  Multiuser receptionParallel receptionOne user reception Fig. 4. Voice packet throughput  2468101214160123456x 10 5 Average node degree    D  a   t  a  p  a  c   k  e   t   t   h  r  o  u  g   h  p  u   t   (  p  a  c   k  e   t  s   )  Multiuser receptionParallel receptionOne user reception Fig. 5. Data packet throughput 24681012141610 −3 10 −2 10 −1 10 0 Average node degree    V  o   i  c  e  p  a  c   k  e   t   l  o  s  s  r  a   t  e  Multiuser receptionParallel receptionOne user reception Fig. 6. Voice packet loss rate 24681012141650556065707580859095100 Average node degree    A  v  e  r  a  g  e   d  a   t  a  p  a  c   k  e   t  s   t  r  a  n  s  m   i  s  s   i  o  n   d  e   l  a  y   (   f  r  a  m  e  s   )   Multiuser receptionParallel receptionOne user reception Fig. 7. Average data packets transmission delay Figs. 4 and 5 show the voice and data packet throughput asa function of average node degree. Node degree is defined asthe number of first hop neighboring nodes. As average nodedegree increases, the proposed MAC design with multiuser de-tection can provide a total throughput that greatly outperformsthe other two solutions. In particular, this gain is in the rangeof   100%  when compared to multiple CDMA channels case andaround  15  times when compared to one channel case when theaverage node degree reaches  16 . Obviously, this gain is realizedlargely by the data traffic since the voice traffic constitutes onlyaround  9%  of the total traffic. It is important to underline that atthe same time, the multiuser detection scheme provides a lowervoice packet loss rate and a smaller data packets transmissiondelay as illustrated in Figs. 6 and 7. Another observation of Figs. 6 and 7 is that as the average node degree increase, thevoice packet loss rate and the average data packet transmissiondelay increase as well because the increase of network densityresults in more contentions among neighboring nodes.VI. C ONCLUSIONS A new MAC design based on multiuser detection is pre-sented in this paper. By using a linear MMSE multiuser de-tector at each node, the proposed solution can significantly out-perform the solutions based on multiple CDMA channels andIEEE 802.11 concepts. Moreover, with described receiver-based power control, each node can save power consumptionand reduce the interference induced in other nodes, which canaccordingly allow greater spatial reuse.The presented work is part of a larger project that is sup-ported by both the government and a private company fundsand the work is being continued in several directions in bothphysical and networking layers. Concerning work related tomultiuser detection, currently we are working on applicationof distributed fair queuing algorithm for combined fairness andthroughput optimization. Also, a more sophisticated propaga-tion model with consideration of shadowing and fading will betaken into account in opportunistic scheduling for better radioresource utilization.R EFERENCES[1] D. Tse and S. Hanly, “Linear multiuser receivers: Effective interference,effective bandwidth and user capacity,”  IEEE Transactions on InformationTheory , vol. 45, pp. 641–657, March 1999.[2] R. Lupas and S. Verdu, “Linear multiuser detectors for synchronous code-divisionmultiple-accesschannels,”  IEEETransactionsonInformationThe-ory , vol. 35, pp. 123–136, January 1989.[3] Z. Xie, R. T. Short, and C. K. Rushforth, “A family of sub optimum detec-tors for coherent multi-user communications,”  IEEE Journal On Selected  Areas in Communications , vol. 8, pp. 683–690, May 1990.[4] J. Zhang, Z. Dziong, M. Kodach, and F. Gagnon, “Enhanced broadcastingand code assignment in multihop mobile Ad Hoc networks,” in  The 11thworldmulti-conferenceonsystemics, syberneticsandinformatics: WMSCI  ,July 8-11, Orlando, Florida 2007.[5] I. Chih-Lin and R. D. Gitlin, “Multi-code CDMA wireless personal com-munications networks,” in  Communications, 1995 IEEE International Con- ference On , vol. 2, pp. 18–22, June 1995.[6] J. Zhang, J. M. Mark, and X. Shen, “An adaptive resource reservation strat-egy for handoff in wireless cellular CDMA networks,”  Canadian J. Elec-trical and Computer Engineering , vol. 29, pp. 77–83, Jan./April 2004.[7] K. T. Jin and D. H. Cho, “Multi-code MAC for multi-hop wireless Ad Hocnetworks,” in  Vehic. Tech. Conf. IEEE  , vol. 2, pp. 1100–1104, 2002.
Search
Similar documents
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks