Efficient Rate Adaptation with QoS Support forWireless Networks

Khoder Shamy, Chadi Assi and Jad El-NajjarConcordia University

Montreal, Quebec, Canada, H3G 1M8Email: {k shamy, assi, j eln}@ece.concordia.ca

Abstract— Wireless LANs (WLANs) suffer degrading perfor-mance when operating within domestic areas due to multiplereasons such as: multi-path, fading, path loss and user mobility.To overcome this, transmission rate is usually adjusted to amore error-resistant rate. In this paper, we first present a novelrate adaptation algorithm for IEEE 802.11 that can efficientlyidentify the threshold frame error rate (FER) at which linkadjustment is required, based on a simple throughput analysis atthe MAC layer. Then, we extend our rate adaptation algorithmto support IEEE 802.11e quality of service (QoS) requirements.When a real-time stream with QoS requirements is admitted,critical constraints such as delay bound and maximum packetdrop count are integrated in the selection of the most convenienttransmission rate that best respects the flow requirements.Moreover, we use dynamic bandwidth allocation rather than thedefault transmission opportunities (TXOPs) in a way that bestoffers a flow the required time for retransmissions due to packetfailure based on the variant loss rate present in the channel. Wevalidate our proposed rate adaptation algorithms via simulationresults where the efficiency and effectiveness of the algorithm arenoticed for both best effort and QoS flows.

I. INTRODUCTION

The 802.11 standard provides several rates for data trans-mission, where the maximum is 11 Mbps for 802.11b andcan reach up to 54 Mbps for 802.11a. These data rates utilizeefficient but different physical modulation schemes. In idealchannel conditions where nodes are within transmission range,the higher the data rate, the better the throughput attained atthe receiver. However, when operating in error-prone channels,lower data rates can overcome the channel conditions toachieve more reliable transmissions than the higher data rates;this is due to the relatively narrow channels used at lowerrates, and the frequent change in transmitting frequencieswhich yield more robust transmission links. Recent studieshave shown that link errors are very common in domesticenvironments [1], where multi-path and fading signals arevery common reasons for the degraded wireless performance.Hence, in order to have an efficient rate adaptation algorithm,nodes need to recognize the right time at which it is beneficialto decrement (or increment) the transmission rate for attainingthe optimal performance based on the channel conditions.Additionally, when real time flow with quality of service re-quirements (such as delay bound and maximum tolerable droprate) are admitted to the medium, link adjustment algorithmsshould integrate these parameters for QoS enhancements. It isessential as well to dynamically allocate bandwidth for QSTAs

through the assigned transmission opportunities (TXOPs) ina way that reflects the additional time needed to retransmitpackets due ACK timeouts caused by either collision orchannel noise.

In this paper, we introduce a novel approach for rateadaptation based on precise identification of the thresholdframe error rate (FER) at which link adjustment is beneficial.Additionally, QoS flows requirements are treated differentiallywith the proposed rate adaptation technique, where criticalconstraints such as delay bound and maximum packet dropcount are used to select the most convenient transmissionrate at specific channel conditions. Finally, our QoS approachutilizes dynamic bandwidth allocation rather than the defaulttransmission opportunities (TXOPs) in a way that best offers aflow the required time for retransmissions due to packet failurebased on the variant loss rate present in the channel. The rateadaptation approach is validated via simulations where theefficiency and effectiveness of the algorithm are noticed forboth best effort and QoS flows.

Priority AC Traffic Type1 0 Best Effort2 0 Best Effort0 0 Best Effort3 1 Video Probe4 2 Video5 2 Video6 3 Voice7 3 Voice

TABLE I

ACCESS CATEGORIES (AC) DEFINED IN IEEE 802.11E

II. RELATED WORK

The most common and basic rate adjustment algorithm isthe Automatic Rate Fallback (ARF) currently implementedin IEEE 802.11 wireless cards. ARF alternates between thesubsequent data rates based on a timing function and missedacknowledgment (ACK) frames [2], assuming that link failureresulted from packet corruption. Similar work was performedin [3]. Here, two counters are kept updated; one for successfultransmissions and one for failed transmissions with no dif-ferentiation between the failure of losses. Hence, the successcounter is incremented if a frame is successfully transmitted,and the failure counter reset to zero. Similarly, if an ACKtimeout occurs, then the failure counter is incremented by

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 1

one and the success counter is reset to zero. The rate is thenadjusted when the counter reaches a specified threshold forfailure or success. However, packet loss could have resultedfrom collisions; decrementing the data rate in this case wouldincrease the contention due to the longer durations of pack-ets transmission. Hence, both approaches are inefficient aschannel quality is not taken into consideration. Authors of[4] recognized this deficiency and accordingly proposed a rateadaptation method that differentiates between the causes offailed transmissions, whether from instantaneous transmissions(collisions) or from poor channel conditions (noise). Hence,a transmitter would decrement its rate solely for subsequentfailed ACK frames due to transmission errors. However,Collision Aware Rate Adaptation (CARA) presented in [4]uses RTS/CTS (Request to Send/Clear to Send) exchangemechanism to distinguish packet losses in addition to ClearChannel Assessment (CCA); here a failed RTS/CTS exchangemay be considered as a collision and hence no link adaptationwill be invoked. However, if multiple ACK frames failed, thendata frames are considered lost due to bad channel qualitywhich requires using lower bit rate. However, in addition tothat RTS/CTS handshake adds overhead to the transmittingtime, RTS/CTS is switched off in commercial network cards.Other related studies in [5] and [6] require a receiver to initiatethe adjustment of the link rate based on channel conditionobserved; the receiver will rely either on RTS/CTS exchangeor Signal-to-Noise Ratio (SNR) values to calculate the frameerror rate (FER). Here, FER values are derived from SNRmeasurements. However, several real-scenarios studies haveproven that it is likely impractical to correlate loss probabilitywith SNR and Bit Error Rate (BER) [7]. On the other hand,challenges facing link adaptation while providing QoS supportfor real-time multimedia flows were tackled so far by [8]and [9]. Authors of [9] suggested link adjustment based onservice requirements of the flows, yet, no practical mechanismwas provided but rather an extension to [6]. Threshold valuesfor the FER critical to QoS constraints were considered in[8]. However, in addition to that the impractical correlation ([7]) between receiver sensitivity(Pr), SNR and BER valueswere used to derive the FER thresholds (QFER), no flowrequirements were taken into consideration such as requireddelay bound or maximum drop limit.

III. PROPOSED RATE ADAPTATION ALGORITHM

Implementing an efficient link rate adaptation algorithmrequires a methodology for wireless nodes in order to decidethe most suitable transmission rate. For example, a node maydecide to switch to a lower rate if the throughput (Xlow) is atleast equal or more than that (Xhigh) obtained at higher ratewhere Xlow and Xhigh represent the throughput values at-tained at lower and higher data rates respectively. Throughputhowever, depends on the duration spent by a station to transmitits packets successfully after applying the DCF mechanism.This time can be expressed as follows:

Ts = TB + DIFS + (TP + TH) +L

R+ TACK (1)

where TB represents the time spent in backoff, DIFS isthe DCF Interframe space, TP , TH and TACK representthe transmission time of preamble, header and ACK framerespectively, L is the payload size and R is the bit rate used.From the previous equation, we see that decreasing the datarate would result in a longer duration for transmitting a packetwhich in turn leads to a lower throughput and more collisions.Hence, stations normally transmit at the maximum possibletransmission rate; 11 Mbps in case of PHY 802.11b and 54Mbps for PHY 802.11(a/g). However, this is true only whennodes communicate over a medium with ideal characteristics.Nevertheless, many experimental studies (e.g. [1]) have shownthat the wireless channel exhibits time varying characteristicsdue to multiple causes (noise, attenuation, multipath, and hostmobility) that results in poor transmission conditions whichleads to erroneous frames at the receiving side. Hence, in orderto maintain an acceptable performance, adjusting transmissionrates is crucial when nodes face highly impaired channelconditions. The frame error rate (FER) is considered a reliablemeasurement of the channel quality at the node level; whenlink rate adjustment is performed for new transmission, oneexpects that the FER suffered at the lower rate (el) to be muchless than that experienced at the higher rate (eh) due to themore robust modulations. Accordingly, a useful expression forthe throughput in an error-prone medium can be used [10]:

{Xlow = rlσl(1 − el)Xhigh = rhσh(1 − eh) (2)

where rl (rh) is the low (high) bit rate, σl and σh are theportions of the useful throughput that is independent from thetransmission rate used. Assuming σl = σh, the FER thresholdvalue (emax) at which a node should decrement its bit rate is:

emax = eh = 1 − rl

rh(1 − el) (3)

This implies that as long as the FER exhibited observed at aspecific bit rate is lower than emax, it is beneficial not to adjustthe transmission rate. However, if the FER reaches a value thatis greater than or equal to emax, then decrementing the bit ratewould be essential to achieve better performance. We notehere that we utilized a recent study [11], in order to developan online estimator that enables the nodes to calculate thepacket error and collision probabilities with no modificationto the currently implemented standard. The authors presentedan efficient estimation method that works at the MAC layerthrough keeping a track of both busy slots (other transmits) aswell as the idle slots during a predetermined time intervalfrom which collision probability (pc) and error probability(pe) are then estimated, (the details of the estimation methodcan be found in [11]). Based on the events defined for thethresholds at which transmission rate should be adjusted, wehave implemented a link adaptation algorithm that operatesat the MAC layer as follows; initially active nodes wouldbe listening to the channel regardless of whether a node isin transmitting, receiving or idle states. Hence, continuousestimation of the channel condition (i.e., FER) using the online

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 2

error estimator is possible, since the status of all slots can bedetermined at all times. A node wishing to transmit will applythe DCF contention mechanism first and when the mediumis idle, it will start transmitting using the highest availablephysical data rate. The algorithm assumes that the FER facedat the next lower rate (el) is initially zero, as the lower ratehas not been used yet, and hence the node could not runthe estimator to determine the FER value. If the transmissionwere successful, then the node proceeds with using that rate.Otherwise, if the ACK timeout duration expired, and thecurrent measured FER (ecurrent) is greater than the calculatedFER threshold (emax), then this indicates that the transmittedframes are subject to excessive corruption that is beyond thethreshold, and hence a retransmission is scheduled with thenext lower rate. Here, another extension we added is thatthe node would not double the contention window CW butmaintain it after decrementing the rate due to a transmissionfailure, since there is no need to increase the CW on the lowerrate. This is because the cause of the transmission failure waserroneous in the channel. Therefore, the error faced at thelower rate will be lessened greatly and consequently, there isno need to use larger CW for the next transmission. Similarly,the process is repeated for lower bit rates and the estimatedvalues of the FER are saved in an array for each bit rate as theyare required later in the calculation of the FER threshold (el)which was set initially to zero. For increasing the transmissionrate, each node keeps track of the successful ACK frames;then, a sequence of consecutive acknowledgments (e.g., 10consecutive ACKs) would indicate that the channel conditionhas improved and the next higher rate could be used.

IV. QOS−BASED RATE ADAPTATION STRATEGY

A. Preliminaries.

Wireless LANs present unique challenges to multimedia ap-plications of strict QoS requirements; this is due to the limitedcapacity and shared medium that is subject to unpredictableerroneous transmissions. Today’s real-time applications re-quire service differentiation when admitted to transmissionmedium, on contrast with best effort flows. Hence, it isessential to integrate QoS requirements in any rate adaptationalgorithm to ensure real-time flows are treated differentially.QoS streams differs from other data flows, mainly, in thedelay requirements they are obliged to meet; here, the receiverdemanding real−time traffic consider late packets as lost andwill eventually discard them. Accordingly, minimizing thedelay time of QoS streams is a key issue in today’s applicationswhose boundaries differ from one application to another, andare usually included in the table of specifications (TSPEC).The delay associated with packet transmission in WLANs canbe divided mainly into queueing delay and medium accessdelay. The queueing delay resembles the time the packetspent since it entered first the transmission queue till timeit got on the head of queue and start contending for channelaccess. Medium access delay, however, is composed of thetotal time the packet waited while contenting for idle mediumtime till it gets transmitted. Hence, the total delay is the

summation of the queueing delay and the medium accessdelay and is usually referred to as end−to−end delay as thepropagation time of packet is neglected. We note here that, themaximum retransmission limit (Rmax

i ) for an unacknowledgedpacket (due to either collision or noise) of priority class ito be dropped greatly influences the medium access delayand is bounded by 0 ≤ Ri ≤ Rmax

i . For instance, Ri =0 corresponds to the case where the queuing delay is veryhigh and the system cannot guarantee the delay requirementof the arriving traffic stream. Here, a packet is not allowedany retransmission and it is dropped upon the first failure.Alternatively, if the queueing delay is negligible (e.g., whenthe load is light), then Ri converges towards its limit, Rmax

i ,for which an exact expression may be derived. Using the factthat a packet is dropped when it reaches the last backoff stageand experiences another collision; the average time to drop apacket is estimated [15] as:

E[Tdrop] =CWmin,i(2Rmax

i +1 − 1) + (Rmaxi + 1)

2E[slot]

(4)CWmin,i represents the lower bound of the CW for priority

class i and E[slot] is the average slot time (whether an idleslot, a busy slot with a successful transmission or a busy slotwith a failed transmission), whose value can be measured bya station. Subsequently, if an arriving flow is of type CBR anddelay bound Di, the station will use the measured E[slot] inorder to estimate the retry limit for the incoming flow. Bysetting E[Tdrop] = Di, one can solve for the maximum retrylimit:

Rmaxi =

⌊log2

{2Di

CWmin,iE[slot]+ 1

}− 1

⌋(5)

for priority class i with delay bound Di and slot time E[slot].It is to be noted that when the load is not high, the queueingdelay is relatively negligible (for example, it has been shownin [13] that only when the traffic per AC[3] (and AC[2])per station exceeds 2Mbps, the queuing delays start to getconsiderably large) and the retry limit is very close to Rmax

i ;otherwise, the retry limit is upper bounded by Rmax

i . Besidesthe strict delay requirements of real−time streams, a droppingthreshold may be specified to ensure the delivery a guaranteedQoS flow. A packet is usually dropped at the MAC laterwhen the retransmission limit is reached. Hence, after severalconsecutive transmission failures, due to either collision ortransmission corruption, the packet is discarded and dropped.Depending on the type of the real-time traffic, drop rate canbe tolerated up to a fixed threshold. Certain value of droppedpackets may be tolerable for most multimedia traffic flows,however, if the dropped packets portion surpassed certaindegree, QoS flow will not achieve its requirements and adegrading performance will be noticed on the receiving side.Although the drop rate is not included in TSPEC table of IEEE802.11e, applications can specify it in their traffic requests.The probability that a given packet of traffic flow of priorityi is dropped after Ri attempted transmissions is given by:

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 3

pdropi = (1 − (1 − perror

i )(1 − pcoli ))Ri+1 (6)

where perrori and pcol

i represent the packet error and packetcollision probabilities respectively.

Using the relation in (5), one can easily derive the retrans-mission limit (Ri) from the delay requirements. Additionally,given that pcol

i is always less than 1, and with a specifiedmaximum allowable packet drop pmax drop

i , one can obtainthe value of the maximum tolerable noise level from (6):

emax =(Ri+1)

√pmax drop

i − pcoli

1 − pcoli

(7)

Subsequently, as channel conditions deteriorate, more pack-ets are dropped threatening the QoS flows requirements.Hence, it is crucial to diagnose such degrading factors. Here,equation (7) identifies the maximum corruption rate at whichdelay and maximum dropping rate are critical. This threshold(7) is used by our QoS rate adaptation technique to adjust thetransmission rate when the faced corruption rate is beyond thisvalue as the state flow diagram in Figure 1 shows.

B. Dynamic Bandwidth Allocation

Transmission opportunity (TXOP) is a new concept that wasintroduced in the recent IEEE 802.11e draft [14]. The TXOPrepresents the allowable time period at which nodes associatedwith a QoS access point (QAP) can initiate transmission duringa service interval. Two modes of TXOP are defined; theinitiation of the TXOP period at which the medium permitsaccess followed by multiple frames transmission exchange andis often called contention free bursting (CFB) as there is no im-mediate contention between packets transmission. By default,TXOP values are assigned fixed values by QAPs according tothe traffic type requirements. As frames transmissions over thewireless medium are usually associated with missed ACKs thatare lost due to either collision or channel erroneous, the Sur-plus Bandwidth Allowance (SBA) parameter was introducedin the TXOP assignment to ensure the requesting QoS station(QSTA) is allocated a minimum amount of excess time by theQAP such that the retransmissions can take place within thesame service interval. The SBA factor is calculated by eachQSTA based on the loss probability and then sent through themanagement frame to the QAP which in turn will considerwhen assigning the TXOP limit for the specific QSTA. TheIEEE 802.11e draft suggests the SBAi for traffic of AC i becalculated as SBAi = Nallocated

Npayload, where Nallocated represents

the number of allocated packets that can be sent during aservice interval and Npayload represents the actual packetsneeded to be transmitted during the service interval. Nallocated

can be represented as “Npayload + Packets in excess”. Somerelated studies suggest the QAP administrator assign the SBAan approximate fixed value (e.g. 1.1 in [16]). However, this isnot accurate as the wireless channel conditions are not constantdue to multiple reasons, (e.g. QSTAs mobility). In addition tothat, the standard suggests the SBA be set by the QSTAs rather

than the QAP. Therefore, the SBA should reflect the varianttransmission failures at each QSTA. Accordingly, the authorsof [15], have suggested calculating the SBA as follows:

Pic = 1 − 1

SBAi(8)

where Pic is the collision probability for the flow of AC i.

Here, the authors assume ideal channel conditions and hencethey neglected the packet loss due channel noise. In practice,however, erroneous transmissions constitute a major portionof the failed transmissions. Hence, it is of great interest toextend the SBA calculation to include not only the collisionprobability of the QoS flows, but also the error probability Pi

e

as well, which can then be expressed as:

SBAi =1

(1 − Pic)(1 − Pi

e)(9)

Consequently, QSTAs would be calculating their SBAi

during each service interval (SI). This dynamic SBA willreflect their need for the extra retries needed to overcome thefailed transmissions due to either collision or channel noise.Afterwards, the QAP can assign dynamically the correspond-ing TXOPi for each QSTA as follows:

TXOPi = TXOPidefault × SBAi (10)

Using the dynamic TXOP combined with the previous rateadjustment technique, nodes will have a relatively fair channelaccess duration, since the assigned TXOPs provide betterchance for the reimbursement of the failed transmissions betterthan if the SBA had the same constant value for all QSTAs.The later condition would make stations facing poor channelconditions retain the same retrial chances as any other stationwith ideal medium.

V. PERFORMANCE ANALYSIS

Extensive experiments through computer simulations weregenerated in order to validate the correctness and effectivenessof the proposed rate adjustment (RA) methodology. Initially,we simulate scenarios with saturated best effort flows of 1050bytes payload size (no QoS) using QualNet simulator [17].We consider 5 nodes, 30 meters initially far from the AP ina Ricean fading channel with K factor 10−4, and Dopplerfrequency of 20 Hz; we compare the performance of Auto RateFallback (ARF) [2], our rate adaptation algorithm, in additionto the optimal fixed PHY rate. Performance of the systemthroughput is then recorded as nodes move 45 m and then 60m away from the AP. The performance results are shown infigure 2, where we notice a remarkable out performance ofthe proposed link adaptation algorithm over the standard ARFmethod (close to that of the fixed rate). This is due to thefact that ARF relies on certain number of consecutive missedacknowledgments which results from transmission failures duenot only to channel conditions but also to instantaneous trans-missions (collisions); our rate adaptation algorithm, however,is capable of measuring the noise threshold at which it iscrucial to adjust the transmission rate without confusion with

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 4

QoS Flow?NO YES

Current Pe >Threshold

Current Pe > QoS_Threshold

NO YES

NO YES

Retrieve Error & Collision Probabilities From estimator

Calculate QoS Threshold

Decrement Tx Rate& Maintain Same CW

Calculate non-QoS Threshold

Decrement Tx Rate& Maintain Same CW

Receive Packet From Queue

Apply DCF/EDCA

Transmit Wait for ACK

YES

NO

ReTx Exceeded YES Drop Packet

NO

Fig. 1. Rate Adaptation Algorithm State Flow Diagram

30 45 600

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

x 106

Average Distance Between Communicating Nodes (m)

Thro

ughp

ut A

ttain

ed (b

ps)

5 Nodes performance as distances separting them increase

Our RA (Best Effort)Fixed RateARF

11 Mbps

5.5 Mbps

2 Mbps

Fig. 2. Performance as function of distance.

collisions and thus better performance is noticed when thelatter method is used. Additionally, when a node running ourRA chooses to decrement its transmission rate upon reachingthe critical corruption threshold, it refrains from doubling itsCW as the cause of transmission failure (channel noise) isdiminished; this minimizes unnecessary further backoff delay.Moreover, as distance increases between transmitting nodesand access point, the difference between throughput of bothmechanisms decreases; this is due to the lower rates used thatwill increase the contention and limit the available bandwidth.

Now, we present the simulation results for the proposed rateadaptation scheme in the presence QoS flows. As QualNetdoes not support the bandwidth allocation extension of IEEE802.11e, we used a modified version of ns-2, in order toperform the required verifications. We consider 5 QoS stations(QSTA), each carrying two almost saturated video flows

(AC=2) of payload size 900 bytes. We run the simulations fora duration of 300 seconds, which in turn was split into threeintervals according to different Ricean channel conditions:during the first interval (0-99s), the nodes are close to theQAP (∼ 20m), i.e., erroneous transmissions are negligible.Afterwards, the distance is increased to 50 meters in the nextinterval (100-199s), where the corruption rate is enormous.Finally, the distance between the QAP and QSTAs is decreasedto ∼ 35 meters at the last interval (200-300s) in order toenhance the medium conditions. Figure 3(a) plots the indi-vidual node throughput versus simulation time for three linkadjustment methods: the standard ARF, the proposed QoSrate adaptation method (QoS-RA) while using the defaultset TXOPs in addition to the QoS-RA while integratingthe proposed dynamic bandwidth allocation technique. Aswe expected, we notice that individual nodes throughput isoptimal during the first simulation time interval in the threecases when QSTAs are close to QAP. Then, it drops down tobelow 50 packets per second as channel conditions deterioratedue to QSTAs mobility away from the QAP. Finally, nodesreattain better performance as the channel noise lessens. Wethen compare the mean nodes throughput for the three linkadaptation methods as show in figure 3(b). It is clear that theQoS-RA proposed method outperforms the ARF throughoutall the experiment. This is expected and is similar to theprevious scenarios (with no QoS flows) as ARF is incapable ofdifferentiation between causes of the transmissions failure. Onthe other hand, it is worthwhile to notice the outperformanceof QoS-RA when dynamic bandwidth is enabled over thatwhen the default preset TXOP limits are used. This is mainlyapparent in the second interval. When the FER faced byQSTAs is negligible (first time interval), both the dynamicand constant bandwidth allocation QoS-RA achieve almost thesame performance. However, as the corruption rate increases(second and last time interval), it is apparent that the rateadaptation algorithm with dynamic bandwidth allocation out-perform the one with the default TXOP values. This is becausewhen the loss rate is low, the dynamically set SBA factor hasinsignificant effect. However as the corruption rate increases,the need for extra bandwidth becomes essential in order toachieve the desired performance while overcoming the failedretransmissions. Here, the dynamic SBA allocation takes thisextra retransmission into consideration as we already presentedand longer transmission opportunities will be assigned for thatpurpose opposed to the case where the constant default SBAvalues are used all the time. Average end-to-end delay anddelay latency are affected as well. It can be easily noticedin figure 4(a) that the proposed rate adjustment techniquealong with the dynamic TXOP assignment achieve a lowerdelay than the standard method. This is mainly due to tworeasons; the QoS-RA decrements the transmission rate only ifthreshold FER is reached rather than missed ACKs whetherfrom channel noise or collisions which in turn create highercontention in channel as in the case of ARF. Moreover, afterQSTAs choose to decrement the transmission rate, CW valueis maintained rather than doubled because the transmission

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 5

0 50 100 150 200 250 3000

50

100

150

Pac

kets

/sec

Individual nodes throughput while running ARF

0 50 100 150 200 250 3000

50

100

150

Individual throughput while running QoS RA and Default TXOP

Pac

kets

/sec

0 50 100 150 200 250 3000

50

100

150

Individual throughput while running QoS RA and Dynamic TXOP

Simulation Time (sec)

Pac

kets

/sec

Node 1Node 2Node 3Node 4Node 5

(a) QoS flows under variable channel characteristics

0 50 100 150 200 25025

40

60

80

100

120

140

Simluation Time (sec)

pa

cke

ts /

se

c

Nodes mean throughput while using different rate adaptation methods.

using ARFQoS RA with Default TXOPQoS RA with Dynamic TXOP

(b) Comparing mean QSTAs throughput

Fig. 3. QoS flows performance while using different RA methods.

0 50 100 150 200 2500

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

Simulation Time (sec)

End

−to−

End

Del

ay (

sec)

using ARFusing QoS RA

(a) Average end-to-end delay

0 50 100 150 200 250 3000.015

0.02

0.025

0.03

0.035

0.04

0.045

0.05

0.055

Simulation Time (sec)

Del

ay L

aten

cy (

sec)

using ARFusing QoS RA

(b) Packets delay latency

Fig. 4. Delay and delay latency

failure cause is lessened. Therefore, less time will be wastedin backing off while using a smaller contention window.

VI. CONCLUSION

We presented a novel and practical approach for linkadaptation based on channel quality measurement. A basicversion of the rate adaptation algorithm was introduced foradjusting the transmission rates based on threshold valuesof throughput enhancement along with maximum tolerablecorruption rate in case of having best effort flows. Moreover,we integrated QoS requirements of delay and maximum droprate along with dynamic bandwidth allocation in the linkadaption process in a way that best serves the needs ofreal−time applications. Finally, we validated the algorithmsthrough extensive simulations.

REFERENCES

[1] D. B. Faria, “Modeling Signal Attenuation in IEEE 802.11Wireless LANs− Vol. 1”, Computer Science Department, Stanford University.

[2] A. Kamerman and L. Monteban “WaveLAN−II: A High−PerformanceWireless LAN for the Unlicensed Band”, in Bell Labs Technical Journal,Summer 1997.

[3] P. Chevillat, J. Jelitto, A. N. Barreto, and H.L. Truong, “A Dynamic LinkAdaptation Algorithm for IEEE 802.11a Wireless LANs”, in Proc of ICC’03, Seatle, USA, May, 2003.

[4] J. Kim, S. Kim, S. Choi, D. Qiao, “CARA: Collision−Aware RateAdaptation for IEEE 802.11 WLANs”, in Proc. of INFOCOM, 2006.

[5] M. R. Souryal and N. Moayeri “Joint rate adaptation andchannel−adaptive relaying in 802.11 ad hoc networks”, in Proc. MIL-COM, October. 2006.

[6] G. Holland, N. Vaidya, and P. Bahl “A rate-adaptive MAC protocol formulti-hop wireless networks”, in Proc. MobiCom, 2001, pp. 236 251.

[7] D. Aguayo, J. Bicket,S. Biswas, G. Judd, and R. Morris, “Link-LevelMeasurements from an 802.11b Mesh Network”, in Proc. of ACMSIGCOMM, Boston, USA, August, 2004.

[8] T. Lee, A. Marshall and B. Zhou “A QoS−based Rate Adaptation Strategyfor IEEE a/b/g PHY Schemes using IEEE 802.11e in Ad−hoc Networks”,in Proc of the Int. conference on Networking and Services, 2006.

[9] J. Kim, H. Tode, and K. Murakami “Service−based Rate AdaptationArchitecture for IEEE 802.11e QoS Networks”, in Proc. of IEEE GLOBE-COM, St. Louis, USA, 2005.

[10] M. Heusse, F. Rousseau, R. Guillier, and A. Duda, “Idle Sense: AnOptimal Access Method for High Throughput and Fairness in RateDiverse Wireless LANs”, in Proc. of ACM SIGCOM, August, 2005.

[11] D. Malone, P. Clifford, D. J. Leith, “MAC Layer Channel QualityMeasurement in 802.11”, IEEE Comm. Letters, Vol. 11, No. 2, 2007.

[12] P. Chatzimisios, A.C. Boucouvalas, V. Vitsas, “IEEE 802.11 PacketDelay - A Finite Retry Limit Analysis”, in GLOBECOM, Vol. 2, pp.950-54, 2003.

[13] Engelstad, P.E., sterb, O.N., “Analysis of the Total Delay of IEEE802.11e EDCA and 802.11 DCF”, in Proc. of ICC, June, 2006.

[14] IEEE 802.11e WG, “Wireless LAN Medium Access Control (MAC)Enhancements for Quality of service (QoS)”,IEEE 802.11e Draft 8.0.

[15] Y. Liu, C. Assi, and A. Agarwal “Enhanced Per-Flow AdmissionControl and QoS Provisioning in IEEE 802.11e Wireless LANs”, in IEEETransactions on Vehicular Technology, to appear, 2008.

[16] K. K. Gavini, V. Apte and S. Iyer,“PLUS-DAC: A distributed admissioncontrol scheme for 802.11e WLANs”, in the Proc. of ICON, 2005.

[17] SNT, “QualNet Simlator”, http://www.scalable-networks.com/.

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.978-1-4244-2324-8/08/$25.00 © 2008 IEEE. 6