Modelling Capture Behaviour in IEEE Radio Modems

Transcription

1 Modelling Capture Beaviour in IEEE Radio Modems Cristoper Ware, Joe Cicaro, Tadeusz Wysocki 20t February Abstract In tis paper we investigate te performance of common capture models in terms of te fairness properties tey reflect across contenting idden connections We propose a new capture model,, as a means of providing an accurate description of experimental data Using two fairness indices we undertake a quantitative study of te accuracy wit wic eac capture model is able to reflect experimental data Standard capture models are sown to be unable to accurately reflect te fairness properties of te experimental data Te capture model is sown to provide a good estimate of actual system performance in varying signal strengt conditions 1 Introduction Te IEEE wireless Pysical Layer (PHYS) and Medium Access Control (MAC) protocols ave lead to te widespread adoption of local wireless area networking over recent years However, recent experiment [1] as indicated tat in many conditions, te potential exists for significant unfairness at te MAC layer In tis paper, we investigate te ability of capture models presented in literature [2, 3, 4] to provide a realistic representation of an IEEE radio modem We consider te fairness properties of simulation traces, generated using te network simulator ns, compared against te experimental data [1] Our results illustrate te difficulty in developing accurate models wic describe te beaviour of real IEEE modems An intuitive definition of fairness is employed in tis paper Hosts sould be able to acieve relatively equal transmission rates, and no ost sould be able to prevent oters from gaining access to te cannel for a sustained period In tis context, fairness is an important parameter for a wireless MAC protocol Te ability to provide fairness over te sortest possible time scale is necessary to prevent jitter in TCP acknowledgement arrivals, known to ave a significant impact on TCP performance [5] Using bot TCP and UDP trace data, we illustrate tat te standard capture models are unable to accurately reflect te fairness properties evident in eac trace A new capture model based on te pysical operation of an IEEE radio modem [6] is presented, termed Tis model is derived from work investigating te impact of multiple access interference [7, 8] and parallel receiver structures [9] We investigate te fairness properties of traces generated using tis model Te network model considered is one involving idden terminals over a semi-slotted MAC/PHYS layer, illustrated in Figure 1 All nodes employ a common spreading code wit no power control Te ns package contains an PHYS/MAC layer model, as well as providing excellent implementations of iger layer protocols suc TCP/IP, UDP, FTP etc Te cannel model employed is an Additive Wite Gaussian Noise (AWGN) Two-Ray Ground model Two fairness indices are employed, Jain s fairness index, and a new index first proposed in [10], te Kullback-Leibler Index Te remainder of tis paper is organised as follows: Section 2 presents details of current capture models Section 3 presents details of te Reception model Section 4 presents an investigation of te fairness properties of te trace data, wile Section 5 concludes te paper 2 Capture Models Te development of models describing te initial capture of a frame by a radio modem represents a significant body of literature [2, 3, 4] Te common goal of eac model is to determine te probability wit wic a given frame may be captured by te receiver, as a function of te number of active stations Tere are two significant stages in te successful reception of a frame by a radio modem Initially, te frame must be successful detected and subsequently captured by te receiver Following tis, te frame must ten be successfully received in te presence of interference, bot from oter transmissions and external noise sources Most literature [2, 3] as considered te probability wit wic successful detection and capture of a frame at te start of a transmission slot occurs Te second aspect requires an understanding of te impact multiple access interference will ave on te captured frame [7, 8, 11] and depends significantly on te modulation tecnique and spreading codes employed Capture models are often used wen simulating te performance of wireless networks Experimental data presented in [1] owever, suggest a more complex capture beaviour resulting in te significant unfairness evident in te traces Furter, in cases were idden nodes are likely (eg a mobile ad oc network) tere is a strong possibility of late starting transmissions colliding wit oter signals at te common receiver In a scenario were all nodes are able to sense carrier, slot boundaries are easily identified and defined, tereby reducing significantly te probability Host 2 Host 3 Host 1 Connection A Connection B Figure 1: Topology

2 HOST 1 HOST 2 HOST 3 Sense Slot Time Error DIFS Receive DATA Transmit DATA Clear Cannel RTS Collision Figure 2: Slot Time Error Time of a new transmission interfering wit an ongoing transmission Wen carrier cannot be sensed and a node as no knowledge of an ongoing idden transmission, an interfering transmission may arrive at a common receiver at any time during a slot Tis is due to differences in te slot time boundaries observed by bot idden nodes As illustrated in Figure 2, Host 3 as commenced a data transfer prior to Host 1 (being idden from Host 3) commencing a cannel sense On sensing a clear cannel, Host 1 defers for a DIFS period (Distributed co-ordinate function Inter- Frame Space) ten transmits an RTS message Tis collides wit te data frame from Host 3, illustrating te potential for a late starting transmission to interfere wit an ongoing transmission Tis is furter complicated by te semi-slotted nature of Rigid slot boundaries are not maintained, requiring nodes to infer slot boundaries from te beginning and end of surrounding transmissions Data transmissions are able to occupy multiple slot times Guard times are inserted between sensing an idle cannel and transmitting (te DIFS), or returning management frames (te Sort Inter-Frame Space, SIFS) to maintain te semi-slotted cannel Te lack of carrier from an opposing idden node owever, allows tat node to transmit at wat appear random times to te common node In te following sections we briefly review te significant capture models considered in literature, wit te reception model outlined in Section 3 21 Capture capture originally described by Davis and Gronemeyer [3], enables te capture of a frame in a given timeslot, provided no oter frame arrives witin a given capture time, of te initial frame Only te initial frame is able to be received Frame arrivals are assumed uniformly distributed on te interval Te initial frame arrives at time, and may be captured by te receiver provided tat, were is te arrival time of te frame Tis model is ciefly controlled by te parameter, governing te lengt of time te receiver requires to detect, correlate wit, and lock onto te received signal Te larger te frame ratio, te less effective te modem is at capturing a 22 Capture capture, originally described wit Rayleig fading, and constant transmitter power [4], is described by te following in- : equality over te interval! #"$&%(' ) +* Te model allows a frame to be captured provided te received power of te frame wit te largest power,,-" is greater tan te sum of te power of all oter received frames,, times te capture ratio, % Te received signals are assumed to ave pase terms varying quickly enoug to allow te summation of te power of eac received frame Tis model is te most commonly employed in simulation of radio modems, allowing te first arriving frame in a slot to be received provided no oter frame arrives witin te capture time, aving a power violating (1) In te case were (1) is violated, no frame is captured 23 Capture Te ybrid model was originally proposed by Ceun and Kim [2] Te power capture effect is used to increase te capture probability of te first arriving frame in a given timeslot, even toug te delay model would oterwise indicate capture as not occurred Capture occurs wen te following inequality olds: %(' ) +* / Te total accumulated energy must be less tan te energy received from te first packet, over te capture interval Tis model results in a greater capture probability, reflecting te ability of a direct sequence spread spectrum receiver to correlate wit te initially detected frame and reject oter transmissions as noise 3 Te Capture Model An enanced capture tecnique is described in [6] wic allows a modem to successfully receive a signal tat would oterwise be considered lost by te previous models Te modem implements a Message In Message process, wose function is to monitor te energy received on eiter antenna during reception %465 of a frame If an increase in energy beyond a given tresold, is observed, te modem attempts to syncronise wit and demodulate te new energy as a potential new signal If tis is acieved a retraining process allows te modem to prepare to receive tis new frame once te prior transmission as finised Tis ability implies tat eac of te capture models previously described will result in a pessimistic capture probability for a frame over a given duration Te message retraining ability of te modem also extends te time scale over wic capture must be considered Retraining may take place at any time during frame reception, as opposed to te delay, power and ybrid capture models wic consider only a sort duration at te start of a frame (1) (2)

3 8 Te model allows te modem to receive a transmission arriving randomly trougout te reception of anoter frame, provided te power of te new transmission is ig enoug to allow successful detection, syncronisation, and demodulation of te frame preamble Results presented in [7], indicate tat te energy associated wit te new transmission will ave a significant impact on te BER observed at te correlator output for te original frame Tese results indicate te initial frame will be unintelligible if te signal power difference between te new and existing transmission is greater tan a tresold of 3-5 db Te model accounts for tis by dropping te initial frame if a new frame is detected wit a signal power greater tan te current by te tresold %465 Successful reception of a frame, 7 8 will occur provided tat over te duration of tis transmission: %495 ) ' :*;#< = * ;2> 8 (3) Tis model results in te successful reception of te strongest frame transmitted trougout its own duration ie 7 8 will be successfully received provided no oter frame arrives over te duration of 7 8 wit a power greater tan 8?%465 (measured in dbm) As te purpose of tis paper is te evaluation of capture models via simulation, a more detailed analytic study of tis model in terms of probability of successful reception of a frame is considered in future studies Te probability of successful capture of a given frame is given by te probability tat no frame arrives (during te transmission time of te captured frame) wit a received power greater tan te captured frame 4 Fairness Study To make a quantitative comparison of eac capture model wit te experimental data, a fairness metric is required In tis context we require tat eac node is able to access te cannel witout sustained delay, and tat no node is able to monopolise te radio cannel at te expense of oter nodes Tis sould be independent of te pysical network topology In [1], experiments controlling te signal power on contending idden connections (Figure 1) illustrate tat signal power is a significant factor in determining te distribution of cannel access We consider two of te experiments in tis analysis Te first involved a constant signal power on eac connection trougout te data transfer Connection A as a Signal to Noise Ratio (SNR) of 25dB, wit Connection B at 20dB Te second, a controlled signal power experiment, commences wit te same signal power for eac connection, ten at 5 seconds, te signal power on Connection A is reduced by 8dBm, bringing te SNR down to 17dB Tese trials are performed using bot TCP and UDP Examples of te experimental data from te stationary signal power trials are sown in Figure 3 and 4 In eac case, is is evident tat te weaker ost is prevented from obtaining a fair sare of te available cannel capacity Te reader is referred to [1] for furter detail on eac experiment Following [10], we employ two fairness indices : Jain s Fairness Index, and a new index proposed in [10], te Kullback- Leibler Fairness Index In eac case, a sliding window metod is Data Transmitted (kbytes) Connection A - 25dB Connection B - 20dB Time (sec) Figure 3: Stationary Signal UDP Data Transmitted (kbytes) Connection A - 25dB Connection B - 20dB Time (sec) Figure 4: Stationary Signal TDP used to calculate te fairness over a specified orizon Te window slides along te packet sequence indicating wic node as successfully gained access to te cannel, calculating an instantaneous value for eac index Te average value is ten calculated across te entire trace We present curves illustrating te fairness as a function of window size In te TCP trial, te trace records successfully acknowledged data Terefore tese results give an indication of te fairness associated wit te data transfer at te transport layer, including effects from te MAC and PHYS layers We calculate fairness in tis manner, as TCP is te most common transport protocol in use today, and any wireless PHYS/MAC protocol must be expected to support competing TCP streams witout imposing additional fairness caracteristics 41 Jain s Fairness Index Tis index as been used widely in te literature to describe te fairness caracteristics in bot congestion control [12] and wireless MAC protocols [10] An ideal fair distribution of cannel access would result in a value for tis index, toug values above A BDC are typically considered to indicate excellent fairness

4 O P M M E E O _ ) ' G G ) ' H N N I g g properties Te index is defined in (4) 7 8FE :*; H JI KL' ) +* H (4) were K H is te fractional sare acieved by te connection, and is te number of active connections A value of would imply tat 30% of nodes were suffering significant unfairness 42 Kullback-Leibler Fairness Index Te Kullback-Leibler Fairness Index was first proposed in [10] Te tecnique considers te distribution of cannel access for eac node as a probability distribution, M Te Kullback-Leibler distance OQP NSR+R NT, an entropy measure of te distance between two probability distributions, is calculated between te desired Tis measure N distribution, and te measured distribution, M provides an indication of te fairness in te system A value of corresponds to a perfectly fair system, wit values below A DC typically indicating a system wit excellent fairness properties NUR:R N T H :*; H H a`+bdc AVAWA HYX K[ Jde&`:bfc K K again, is te number of nodes, and H acieved by te node 43 K^] (5) (6) te fractional sare Simulation trials of te stationary and controlled signal power experiments were undertaken, and bot fairness indices calculated as a function of te sliding window size Figures 5, 6, 7, and 8 present bot indices for eac capture model, te experimental data, and a simulation trial employing no capture Te window size in eac case does not extend beyond 1000 frames, as tis represents alf te number of frames transferred on eac connection As expected, te fairness improves as te orizon is increased Te stationary signal power trials, Figures 5 and 6, illustrate ow te,, and models significantly overestimate te fairness performance acieved in eac trial Te experimental data exibits significantly worse unfairness tan eac of tese models indicate In Figure 5 wit TCP, te alternates between under and over estimation of te fairness evident in te trace Tis can be attributed to a large timeout in te experiment data [1] wic did not occur in te Message Retraining trace Te experimental data reaces a fairness peak at a window of 500 frames wic may be related to te timeout event in te trace data Figure 6 wit UDP also illustrates te significant difference between te capture models and experiment Te model provides a pessimistic indication fairness according to bot indices In te stationary signal strengt experiment te model follows Figure 5: Stationary TCP 5 Figure 6: Stationary UDP

5 g g Figure 7: Controlled TCP te same trend as te trace data, yet maintains a consistent offset Tis may be due to a lack of variation in signal strengt, and te model allowing a stronger connection to capture a cannel for a longer period tan is evident in te trace data In te controlled signal power experiments, Figures 7 and 8, te Message retraining model is able to follow experimental data, were te oter models over estimate te fairness values Wit TCP, te fairness was over-estimated (ie experimental data exibited worse fairness properties), wile wit UDP it was underestimated Wen compared wit te,, and models, is able to matc te fairness time scale present in te experimental data quite closely Differences between te simulation models and protocol implementations must be considered wen interpreting tese results Wile ns is an excellent simulation platform, tere are still differences and implementation issues wic may lead to varying results In particular, subtle differences between protocol timers and tose in ns will result in deviation between simulation and experimentally obtained data Furter, cannel variations not accounted for in simulation will also ave an impact on te experimental data Tese quantitative results provide a positive indication tat te capture model is able to reflect, wit reasonable accuracy, te fairness properties tat may be obtained by a real system wen varying signal strengt conditions and idden terminals exist Furter investigation of te model is required owever, to account fully for te overestimation in te stationary signal power UDP experiment Figure 8: Controlled UDP 5 Conclusions In tis paper we ave investigated te performance of several capture models in terms of te fairness tey indicate a system may provide We ave proposed and investigated te fairness properties of a new capture model,, as a means of providing an accurate description of experimental data Using two fairness indices we undertake a quantitative study of te accuracy wit wic eac capture model is able to reflect experimental data Te capture model is sown to provide a good estimate of actual system performance in varying signal strengt conditions Understanding te fairness orizon associated wit a PHYS/MAC protocol is important in acieving good performance for real time multimedia traffic flows, and smooting te flow of TCP acknowledgements Te model can be employed in situations were varying signal strengt is expected to impact on system performance Tis as specific relevance were nodes in a given topology are unable to sense carrier from near neigbours Te model may also ave application in te development of quality of service mecanisms for te IEEE wireless MAC protocols Acieving a MAC layer free from unfairness arising at te pysical layer is paramount if reliable quality of service is to be offered by te MAC protocol Mecanisms to acieve tis require development Furter development and analysis of te capture model is necessary in order to matc experimental data more closely, toug te current results presented ere are very encouraging

6 Refer i ences [1] C G Ware, J Judge, J F Cicaro, and E Dutkiewicz, Unfairness and capture beaviour in adoc networks, in International Conference on Communications, vol 1, (New Orleans), IEEE Press, 2000 [2] K Ceun and S Kim, Joint delay-power capture in spread-spectrum packet radio networks, IEEE Transactions on Communications, vol 46, no 4, pp , 1998 [3] D H Davis and S Gronemeyer, Performance of slotted aloa random access wit delay capture and randomised time of arival, IEEE Transactions on Communications, vol COM-28, pp , 1980 [4] J C Arnbak, Capacity of slotted aloa in rayleig fading cannels, JSAC, vol SAC-5, pp , Feb 1987 [5] H Balakrisnan, V N Padmanaban, S Sesan, and R H Katz, A comparison of mecanisms for improving tcp performance over wireless links, IEEE/ACM Transactions on Networking, vol 5, no 6, pp , 1997 [6] R Mud, Wireless LAN Wit Enanced Capture Provision US Patent No US [7] C G Ware, T Wysocki, and J Cicaro, Hidden terminal jamming problems in IEEE mobile ad oc networks, in To Appear:International Conference on Communications, (Helsinki), 2001 [8] M B Pursley, Performance evaluation for pase coded spread spectrum multiple access communication - part 1: System analysis, IEEE Transactions on Communications, vol COM-25, no 8, pp , 1977 [9] D I Kim, I K Kim, and R A Scoltz, Counting collision free transmissions in common-code ssma communications, IEEE Transactions on Communications, vol 43, February / Marc / April 1995 [10] C E Koksal, H Kassab, and H Balakrisnan, An analysis of sort term fairness in wireless media access protocols: Extended version of sort paper, in ACM Sigmetrics, 2000 [11] M Sorousnejad and E Geraniotis, Probability of capture and rejection of primary multiple access interference in spread spectrum networks, IEEE Transactions on Communications, vol 39, no 6, pp , 1991 [12] R Jain, D Ciu, and W Hawe, A quantitative measure of fairness and discrimination for resource allocation in sared computer systems, Tec Rep TR-301, DEC, 1984