dynamic data rate and transmit power adjustment in ieee 802.11

Dynamic Data Rate and Transmit Power Adjustment

in IEEE 802.11 Wireless LANs

Pierre Chevillat,1Jens Jelitto,

1,2and Hong Linh Truong

1

In this paper a novel link adaptation algorithm is proposed that is capable of adjusting thetransmit power level and the data rate jointly to the radio channel conditions. The proposed

method relies solely on link quality information available at the transmitter by employing thereception or non-reception of the acknowledgment frames as a measure of the channel qualitywith respect to the power level and data rate. The method is fully compatible with the 802.11

wireless LAN standard. In contrast to many other proposals, it neither relies on the RTS/CTSprotocol nor requires a feedback channel to transmit link-quality estimates from the receiverto the transmitter. Different strategies for optimizing the data rate and power level are given.

These depend on the scenarios considered, the number of active stations, and the servicerequirements. The two main strategies are either to drive the system towards the highestpossible data rate and adjust the rate and power levels accordingly (‘‘high-performance’’

mode) or to focus on power saving, possibly trading this for other performance criteria such asthroughput or delay performance (‘‘low-power’’ mode). Other special cases, such as power orrate only adaptation, are also discussed. It can be shown that in most cases the best choice forachieving low transfer times, maximizing throughput, and alleviating the hidden terminal

problem is to transmit at the highest possible rates and with high power levels. This ‘‘high-performance’’ mode of operation also minimizes the transmission times, which in turn max-imizes the time for putting idling components into a sleep mode, thereby minimizing the

overall power consumption.

KEY WORDS: 802.11 Wireless LANs (WLANs); dynamic link adaptation; power and rate adaptation;throughput optimization; power efficiency.

1. INTRODUCTION

Broadband wireless LANs (WLANs) based onthe IEEE 802.11a [1] and 802.11g [2] physical layerstandards support multiple data rates, which enableswireless stations (WSTAs) to select the appropriatetransmission rate depending on the required qualityof service and the radio channel conditions. Tomaximize the throughput or minimize the transmis-sion delay, using a high data rate is the right choice in

many cases. However, if the distance to the receiver islong, using a high data rate may lead to an excessivenumber of retransmissions owing to the low signal-to-noise ratio (SNR). This may result in performancedegradation or even in a total loss of communication.In such a situation, a more robust but lower data ratemay achieve a better throughput. Other parametersthat affect overall system performance also need to beconsidered. For battery-powered devices such aslaptops and PDAs, transmission power awareness iscrucial to save energy and prolong battery life.Reducing the transmit power level, in particularwhen the distance to the receiver is small, also helpsreduce interference with neighboring WLANs. Toachieve the highest possible system performance, it is

1 Zurich Research Laboratory, IBM Research GmbH, Saumerstr.

4/Postfach, CH-8803 Ruschlikon, Switzerland.2 Tel.: +41-1-724-84-96; Fax: +41-1-724-89-55; E-mail: jje@

zurich.ibm.com

International Journal of Wireless Information Networks, Vol. 12, No. 3, July 2005 (� 2005)DOI: 10.1007/s10776-005-0006-x

1231068-9605/05/0700-0123/0 � 2005 Springer Science+Business Media, Inc.

therefore mandatory to use an automatic link adap-tation algorithm that allows a station to adapt itstransmission parameters to the actual conditions ofthe wireless channel.

The interest in designing a link adaptationmechanism for 802.11 WLANs is large because thebasic 802.11 MAC standard [3] does not specify anyprocedure for rate switching or power level selection.Many papers have appeared on this topic, most ofthem only dealt with either rate adaptation [4–11] ortransmit power control [12–15]. A few papersdescribed algorithms that combine rate adaptationwith power control [16–18].

The classical way of performing link adaptationis to rely on feedback from the receiver. In thisapproach the channel quality is estimated from theSNR, the received signal strength, or packet errorrate measurements, and the transmit rate or powerlevel to be used in future transmissions is derived.This information is then sent back to the transmit-ter over a feedback channel [19,20]. Unfortunately,the 802.11 MAC standard does not provide anyprotocol means for the receiver to inform thetransmitter about the link quality or the transmis-sion rate to be used. Link adaptation methodsrelying on a feedback channel therefore cannot beemployed or require a modification of the 802.11standard.

A number of algorithms proposed make use ofthe so-called ‘‘RTS/CTS’’ protocol to exchange theinformation needed for the adaptation. The RTS/CTS protocol is defined in the 802.11 standard tocombat the well-known ‘‘hidden’’ terminal problem.Before sending a data frame, a transmitting stationmakes a reservation for the wireless channel bysending a short ‘‘Ready-To-Send’’ (RTS) frame. Thereceiving station replies with a ‘‘Clear-To-Send’’(CTS) frame. The transmitting station proceeds withthe transmission of the data frame after havingreceived the CTS frame. All other stations in thevicinity of the transmitter and the receiver that havereceived these two frames will defer their owntransmissions until the end of the reservation. Thus,the RTS/CTS protocol can be exploited to exchangelink adaptation parameters such as rate and packetsize [4], or interference margins and transmit powerlevels [12].

An interesting idea is described in [9], in whichthe access point (AP) broadcasts special beaconframes that allow the wireless stations (WSTAs) todetect their locations relative to the AP and todetermine the transmit rates to be used.

Another alternative is to measure the receivedsignal strength and derive the transmit parametersfrom that measurement [11,21]. The transmit powerlevel can also be included in the packets, such as inthe method described in [17], in which a combinedrate and power adaptation for 802.11a WLANsoperating under the Point Coordination Function(PCF) is described. The main idea is to includethe information about the transmit power level in theSERVICE field of a MAC frame that is sent by thePoint Controller (PC). This information will allowthe receiver of the frame to estimate the path lossbetween itself and the PC, and to determine the bestrate–power pair to be used for future transmissions.A similar method is defined in [18] for the 802.11Distributed Coordination Function (DCF). It usesthe RTS/CTS frames sent at full power prior to anydata transmission to estimate the path loss. TheWSTAs determine the best power–rate combinationfrom a table indexed by the packet length, the pathloss condition, and the retry counters of the DCFprocedure. The table entries are calculated off-line.

All algorithms described so far require either amodification of the 802.11 standard or the coopera-tion of the peer station, e.g. when making use of theRTS/CTS procedure. These drawbacks can beavoided by employing only information that isavailable at the transmitter. In [5], the transmitterswitches between two fixed data rates, with the higherrate being the default operating mode. After twoconsecutive transmission errors, it uses the lower rate,and returns to the high one after 10 successfultransmissions or after a time out. This techniquewas enhanced in [10] to handle multiple transmissionrates and varying channel conditions. In [22] themechanism of [10] is employed to adapt the transmitpower levels to the channel conditions; however, thetransmission rate is maintained at a fixed value.

In this paper we extend the ideas proposed in[10] and [22] to a new link adaptation algorithm thatis capable of adjusting jointly the transmit powerlevel and the data rate to the radio channel condi-tions. The method is fully compatible with the 802.11wireless LAN standard. It neither relies on the RTS/CTS protocol nor requires a feedback channel totransmit link-quality estimates from the receiver tothe transmitter. Instead, the proposed scheme reliessolely on information available at the transmitter byusing the ACK frames.

In the next section we will discuss somefundamental link adaptation issues. In Section 3the proposed joint power and rate adaptation

124 Chevillat, Jelitto, and Truong

algorithm is presented. A comprehensive evalua-tion of the performance of the proposed scheme isgiven in Section 4. Finally, Section 5 concludes thepaper.

2. LINK ADAPTATION PRINCIPLES

Our goal is to optimize the throughput andtransmit power for a given channel, i.e., to transmitas many bits with as little energy as possible. Theoptimization strategy depends on several factors,such as the service requirements (e.g. target data rate,delay constraints), power consumption, and batterylifetime, but also on coexistence and fairness consid-erations in a wireless network.

The parameters that can be controlled by a linkadaptation algorithm are the transmit power PD, thedata rate rD, and the packet length lD. Although thethroughput depends also on the packet length [23], ithas been shown in [22] that fragmenting large packetsas defined in the 802.11 MAC standard [3] is not anappropriate means for increasing throughput becauseevery fragment is acknowledged separately. Hence,the overhead increases linearly with the number offragments. For this reason, only transmit power anddata rate adaptation will be considered in this paper.Let us define the effective throughput of a WLAN as

TP ¼ Nb

T; ð1Þ

where Nb is the total number of successfully trans-mitted data bits in the observed time interval T. Thetime interval T includes the transmit times TD of thedata packets and overhead times TOH such as DIFS(Distributed Interframe Space), SIFS (Short Inter-frame Space), acknowledgment times, backoff timesand other idle times (e.g. waiting times for newpackets in a system that is not in saturation). Inaddition, let us define the average transmit power as

�Ptx ¼PK�1

k¼0 PD;kTD;k þPL�1

l¼0 POH;lTOH;l

T; ð2Þ

where PD;k is the transmit power of the kth dataframe transmitted, TD;k the transmit time of the kthdata frame, and K the total number of data framestransmitted, including erroneous frames. POH;l is thepower associated with the lth overhead time TOH;l

(e.g. to transmit an ACK frame, or the power con-sumed during idle times DIFS, SIFS), and L the totalnumber of overhead events. In the simulation resultspresented in the next section, the power consumed

during idle times (e.g. DIFS, SIFS) is not consideredand set to zero. Hence, as long as no RTS/CTSmechanism is enabled in the system, the overheadenergy consumption is determined by the transmis-sion power PA and transmission time TA of the ACKframes to acknowledge successful data transmissions.Note that even in this case, K and L are not equalbecause of unsuccessful transmission attempts. Fromthese two definitions the power efficiency can bederived as

gTP ¼TP�Ptx

¼ NbPK�1

k¼0 PD;kTD;k þPL�1

l¼0 POH;lTOH;l

:

ð3Þ

The denominator describes the overall transmit en-ergy consumption associated with the successfuldelivery of Nb data bits. To maximize the powerefficiency, the times TD and TOH and the powers PD

and POH should be kept as small as possible for thegiven service requirements.

The transmission times TD(lD,rD) and TA(lA,rA)depend on the frame lengths lD and lA and thetransmission rates rD and rA of the data and ACKframes, respectively. For 802.11a, the time to trans-mit a data frame of lD bytes at rate rD is [1]

TDðlD; rDÞ ¼ TPA þ TSIG þ TSYM �16þ 8 � lD þ 6

NDBPSðrDÞ

� �

;

ð4Þ

where TPA, TSIG and TSYM are the times to transmitthe preamble, the signal field and one OFDM symbolof the 802.11a OFDM PHY layer, respectively.NDBPS is the number of data bits encoded within oneOFDM symbol and depends on rD. The operator d ereturns the smallest integer value greater than orequal to its argument value. In 802.11a, theseparameters have the values TPA=16 ls, TSIG=TSYM=4 ls, NDBPS=4Æ rD. Possible transmissionrates are rD=6, 9, 12, 18, 24, 36, 48, and 54 Mbps. Ifwe assume a packet length of lD=1000 bytes, thetransmission times for the different data ratesare TD(lD=1000 bytes, rD ={6,9,12,18,24,36,48,54}Mbps)={1360,912,692,468,356,244,188,172} ls.Hence, for lD=1000 bytes, the transmission time at arate of 54 Mbps is approximately 1/8 of the timerequired at a rate of 6 Mbps.

The transmit power necessary for a certainservice requirement depends on the transmissionrate, the distance between communicating devices,the channel characteristics and other factors; thereexists no simple relationship. Assuming a fixed

125Dynamic Data Rate and Transmit Power Adjustment in Wireless LANs

distance between transmitter and receiver, a roughestimate of the relative transmit power requirement asa function of the transmission rate to support a desiredpacket error rate (PER) can be extracted from Fig-ure 1. For an AWGN channel and a desired PER of10)2, let the necessary transmit power bePD(rD=6 Mbps)=P0. The power levels required forthe different rates are roughly PD(rD={6,9,12,18,24,36,48,54} Mbps)={1,2,2,4,8,16,32,64}Æ P0. Hence, apower difference of about 18 dB, or 64 times the powerlevel, is required to maintain the same PER at a rate of54 Mbps as at 6 Mbps. This power ratio of 18 dB isalso confirmed by [24].

The transmit rates rA for the ACK frames aredefined by the standard, hence the times TA aregiven. As the immediate ACK procedure and itsreliability are crucial for the system, it is assumedthat all ACK frames are transmitted at the highestpower level.

From the above results, several observations canbe made. If the transmit power is fixed, one shouldtry to transmit at the highest possible rate. Thisincreases the power efficiency and minimizes the airtime. This in turn increases the system throughputand minimizes the collision probability. If, on theother hand, the rate is fixed (as might be required by aservice), the lowest possible power level should beused to save energy. However, this strategy mightworsen the hidden terminal problem. If both thetransmit power and data rate are adaptive, thecomparison of the transmit time ratio (1/8) andthe transmit power ratio (64/1) for 54 and 6 Mbpssuggests to transmit at the lowest possible data rateand then optimize the transmit power accordingly.However this statement is only valid if we considerthe power efficiency for a single communication link.There are several issues that need to be taken intoaccount:

–2 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 3410– 4

10 –3

10 –2

10 –1

10 0

E /N

Pac

ket E

rror

Rat

e

6 Mbps

6 Mbps

6 Mbps9 Mbps

9 Mbps

9 Mbps

12 Mbps

12 Mbps

12 Mbps

18 Mbps

18 Mbps18 Mbps

18 Mbps

24 Mbps

24 Mbps

24 Mbps

36 Mbps

36 Mbps

36 Mbps

48 Mbps

48 Mbps

48 Mbps

54 Mbps

54 Mbps

54 Mbps

AWGN

ETSI ’A’

Fig. 1. Packet error rate as function of Es/N0 and the 802.11a data rates for AWGN (solid) and ETSI ‘‘A’’ (dashed) channel (18 taps,

Rayleigh fading, rrms=50 ns, rmax=400 ns).


• The power efficiency does not take intoaccount the energy consumption of the base-band and MAC layer signal processing. High-er data rates allow the radio front-ends as wellas the PHY and MAC engines to be put intoan energy-saving sleep mode for longer timeperiods.

• The reduced transfer times associated withhigher data rates result in a reduced collisionprobability. This will result in a reducedretransmission rate and better throughputperformance than for a low data rate trans-mission.

• Higher power levels associated with higherdata rates lead to a less pronounced hiddenterminal problem, as more stations can ‘‘hear"a station transmitting at a higher power level.

• To optimize the aggregate throughput in a cell,the best strategy is to use the highest possibledata rates for data transmission.

• The transfer times depend on the data rate,backoff times, and retransmissions needed.Hence, as long as higher data rates do notconsiderably increase the retransmission prob-ability, using the highest possible data rateswill result in the shortest transfer times.

The arguments given above suggest that the beststrategy is to select the highest possible data rate andadjust the transmit power accordingly.

3. JOINT POWER AND DATA RATE

ADAPTATION

3.1. Proposed Algorithm

The link adaptation algorithm proposed heremakes use of the immediate ACK strategy defined inthe IEEE 802.11 MAC protocol [3]. Error-free dataframes are immediately acknowledged by the recei-ver. Frames are protected against errors (due totransmission errors or collisions) by means of a framecheck sequence (FCS) field containing a 32-bit cyclicredundancy code (CRC) and a simple send-and-waitARQ mechanism. If the receiver detects a CRC errorthe frame is discarded. Otherwise, the receiver sendsan ACK frame back to the transmitter. If no ACKframe is received within a specified time, the frame isre-sent after a random back-off time. This process isrepeated until the transmitter receives an ACK, apacket lifetime value is exceeded, or a maximum

number of retries has been reached. In the latter twocases, the transmitter discards the frame.

As no channel state information (CSI) is avail-able at the transmitter, the proposed algorithmexploits the ability of the immediate ACK procedureto detect frame loss. If the transmitter does notreceive an ACK within a specified time interval, thealgorithm concludes that link quality was insufficientand that a lower data rate or a higher transmit powershould be used. On the other hand, if the transmittersucceeds in sending multiple data frames, it assumesthat the link quality is sufficient and that a higher rateor a lower transmit power can be used.

This adaptation scheme can be implemented inthe transmitter with a pair of counters for everydestination MAC address: one for successful trans-missions s and one for failed transmissions f. If aframe is successfully transmitted, counter s is incre-mented and counter f reset to zero; similarly, if atransmission fails, counter f is incremented andcounter s reset. If the success counter s reaches acertain threshold Smax, then the data rate is increasedor the power decreased, and both counters are resetto zero. Similarly, if the failure counter f reaches acertain threshold Fmax, then the data rate is decreasedor the power increased, and both counters are reset tozero.

3.2. Dynamic Threshold Adaptation

The values of Fmax and Smax are critical for theperformance of the link adaptation scheme [10].Throughout the paper we assume that Fmax=1. Thisconservative choice corresponds to an immediateadaptation towards a more robust operating point bydecreasing the data rate and/or increasing the powerin the case of a failed transmission. This strategyprevents unnecessary retransmissions at the price of aslightly suboptimal steady-state performance.

The optimum choice for Smax depends on thechannel dynamics. Fast-changing channels require asmall value of Smax, so that the transmission param-eters can keep up with the channel variations. On theother hand, for slowly changing channels, largevalues of Smax avoid ineffective switching to higherrates or lower power levels when the channel has notimproved.

Exploiting these observations, the basic algo-rithm described in Section 3.1 is enhanced with asimple but powerful method that estimates the speedof link-quality variations and switches dynamicallybetween two success threshold values, namely S1 and


S2, with S1<S2. In regions of high Doppler spread orfast-changing channel conditions, a low threshold(Smax=S1) improves the tracking performance of thealgorithm. For low Doppler spread or slowly chang-ing channel conditions, a higher threshold (Smax=S2)prevents throughput degradation due to prematureswitching.

The state transition diagram of the completeadaptation algorithm including dynamic thresholdadaptation is shown in Figure 2. As indicated in thefigure, a transmitting node may be in any one of thethree states—High, Low, and Spread?. The statesHigh and Low reflect the current node’s assumptionabout the changing speed of the channel quality.After Smax successful transmissions, the node adjustsits transmission parameters to a higher rate or to alower power level and enters the state Spread?. In thisstate, it waits for the result of the next transmission todecide whether it should move to the High or to theLow state. If the next transmission succeeds, the nodeassumes that the link quality is improving rapidly,enters state High, and sets Smax to the small value S1

to react quickly to the changing link quality. If, onthe other hand, the next transmission fails, the linkquality is assumed to either change slowly or not atall, so that the former decision to change thetransmission parameter was premature. Therefore,the state Low is entered and Smax is set to the largervalue S2.

3.3. Strategies for Data Rate and Power Level

Selection

From the discussion in Section 2, it is apparentthat there is no simple analytical approach for

deriving a procedure to select the optimum adapta-tion parameters, especially in an environment wheremultiple stations are active. Therefore, we introducetwo simple adaptation strategies with focus on (i)compliance with the 802.11 standard and (ii) lowimplementation complexity. We differentiate betweentwo basic adaptation modes, a ‘‘High-Performance’’(HP) mode and a ‘‘Low-Power’’ (LP) mode.

In HP mode, the optimization goal is to transmitat the highest possible transmit rate. The transmitpower is of secondary importance. This mode isapplied when the highest possible rate is required or ifthe overall throughput should be optimized. The LPmode performs the dual operation to the HP mode.Here, the optimization strategy is to work at thelowest possible power level, and the data rates areadjusted accordingly. It can for example be used inportable devices to save battery power and in caseswhere the data rate or delay requirements of theapplication are sufficiently low.

3.3.1. ‘‘High-Performance’’ (HP) Mode

The flow diagram for this mode is shown inFigure 3. As mentioned above, the primary goal inthis mode is to support the highest possible data rate.Therefore, after Smax successful transmissions (i.e. itis assumed that the link quality has improved), thedata rate is increased until the maximum rate isreached. Once the data rate cannot be increasedfurther, the transmit power level is reduced accord-ingly. In the case of Fmax failed transmissions, thealgorithm first attempts to increase the power level tokeep the rate at its current value. The rate is onlyreduced if the transmit power has already reached itsmaximum level.

Fig. 2. Transition diagram of the link adaptation algorithm.


no

Start

pcnt=0

s=0;f=0

rate_crit=0

pow=max

rate=maxInit

WAIT

ACK

read

success?

ACK=

s++;f=0

s=S_max?

s=0;f=0f=0;s=0pcnt--

f=F_max?

f++;s=0

rate=max?

rate_crit=0?

pcnt=P_max?

pow--

WAIT WAIT WAIT WAIT WAIT WAIT

pcnt++

pow-- rate_crit=rate

rate -- pow++

pow=max?

rate_crit=0]

OR

[rate_crit++

pcnt=0

rate=rate_crit

pow=max

yes

no

no

yes

yes

nono

yes

yes

no

no

yes

rate++

yes

Fig. 3. Joint power and rate adaptation algorithm. Flow diagram for HP mode.


A variable ratecrit is introduced to mark a criticaldata rate, namely, the lowest rate that cannot besupported because of the current channel conditionseven at the maximum power level. When this variableis set, indicating it does not make sense to increase thedata rate to ratecrit, it is more appropriate to reducethe transmit power after Smax successful transmis-sions. The number of successful power level reduc-tions is tracked in the counter pcnt. If the data ratebelow ratecrit can still be supported after decreasingthe power level Pmax times, it is assumed that channelconditions have improved. Hence, the algorithmincreases the data rate to ratecrit using the highestpossible transmit power level, and ratecrit is incre-mented or reset to zero to allow operation at themaximum data rate.

3.3.2. ‘‘Low Power’’ (LP) Mode

In LP mode, we try to use the lowest possibletransmit power level, and only adjust the transmitrate if we can no longer decrease the power level. Theflow diagram for this mode can be derived fromFigure 3, as it performs the dual operation of the HPmode with the following dualities: rate is replaced bypow, the increment operator ++ by the decrementoperator ) ), and min by max. In addition, insteadof locking ratecrit the power powcrit is locked, pcnt isreplaced by the rate counter rcnt, and the counterlimit Pmax is replaced by Rmax.

The key operations in LP mode can be describedas follows: if a certain power level does not result in asuccessful transmission even for the lowest possibledata rate, then the unsuccessful power level is storedin powcrit and the power level is increased. If thepower level above the critical, currently unsupported,level powcrit still results in successful transmissionsafter increasing the data rate Rmax times, it is assumedthat the channel conditions have improved. Hence,the algorithm tries to decrease the power level topowcrit using the lowest possible data rate. To furtherdecrease the minimum power level, the critical valuepowcrit is either decreased or reset to zero to allowoperation at the minimum power level.

The two strategies described above for HP andLP mode, respectively, rely on the assumption thatthe discrete step sizes available for selecting the datarate and adjusting the power level provide sufficientoptimization potential to move the operating point asclosely as possible to the optimum power efficiencyvalue. This will be verified in Section 4 by means ofsimulation results.

3.3.3. Single-Parameter Link Adaptation

Two special cases can be obtained if either thedata rate or the power level is fixed. In [10] a ‘‘Rate-Only’’ (RO) mode was studied, in which the transmitpower level is fixed and the data rate is adjusted usingthe dynamic threshold adaptation scheme shown inFigure 2. Another possible optimization strategy wasconsidered in [22]. Here, the data rate is fixed and thepower level is adjusted. The data rate is increasedonly if the lowest possible power level has beenreached. The performance of single-parameter linkadaptation will be addressed in Section 4, togetherwith that of the proposed joint data rate and poweroptimization schemes.

4. PERFORMANCE EVALUATION

To study and understand the behavior andperformance of WLANs with the link adaptationalgorithm proposed in this paper, we have imple-mented a discrete-event simulator that models thePHY and MAC layers of 802.11a WLANs in detail.A comprehensive description of the simulator is givenin [10].

We focus first on a simple WLAN link with onlyone AP and one WSTA (point-to-point WLAN), andstudy the impact of the various parameters of the linkadaptation algorithm on its throughput and delayperformance. Then, infrastructure WLANs withmultiple WSTAs will be considered, in Section 4.2for Poisson traffic sources, and in Section 4.3 withmore realistic source models for data, telephony, andvideo services. Furthermore we distinguish betweensaturated and non-saturated networks. In a saturatednetwork, the traffic at the nodes is generated in such away that the nodes always have packets to send. Insuch a network the throughput is the most importantperformance value. On the other hand, in a non-saturated network, the total generated load is rathersmall compared with the capacity of the WLAN, anddelays experienced by the data packets are moreimportant than the achievable throughput.

Unless stated otherwise, the following parametervalues are used for the simulation runs:

• Fixed Doppler spread of fd=5 Hz (which isconsistent with measurement results in officebuildings [25]);

• constant packet length of 1000 bytes;• negative-exponential distributedpacket arrivals;


• transmit power levels max=+10 dBm,min=)10 dBm;

• transmit power step sizes up=+5 dB,down=)2 dB (the choice of these parametervalues will be explained later on);

• success thresholds S1=3 and S2=10; and• failure threshold Fmax=1.

4.1. Point-to-Point Wireless LAN

4.1.1. Saturated Point-to-Point Wireless LAN

In this section we study the performance of an802.11a WLAN with two stations as a function of thedistance between the two nodes and the adaptationschemes. The diagrams are obtained by having bothstations saturated with a traffic load of 30 Mbps. Forall adaptation schemes, the transmit power level isfixed at +10 dBm, except for the HP and LPalgorithms, for which it varies between ±10 dBmwith step widths of +5 and )2 dB. As reference thecurves for the RO and the ‘‘genie-aided’’ adaptationschemes are also shown in the diagrams. The ROscheme is the one in which only the data rates areadapted, while the power level PD remains fixed. Theperformance of this mode was studied in detail in

[10]. In particular it was shown [10] that this modeprovides performance values very close to thoseobtained with a ‘‘genie-aided’’ adaptation scheme inwhich the transmitter knows the CSI a priori.

The throughput of this point-to-point WLAN isshown in Figure 4. As expected, it decreases withincreasing distance between the nodes; furthermore,low data rates support larger ranges than high datarates do. It can be seen that the HP algorithm (uppersolid line) provides throughput values very close tothose of the ‘‘genie-aided’’ and the RO modes(dashed lines). The small throughput degradationwhen compared to those two modes is due to thealgorithms’ continuous attempt to reduce the trans-mit power level. This is also the reason it provides lessthroughput than the 54 Mbps fixed-rate mode does inthe case of a short distance. Otherwise, it alwaysachieves much higher throughput than any of thefixed-rate modes do.

In addition, we note that the LP mode (lowersolid line) achieves high throughput values only atshort distances. Its throughput performance degradesrapidly with increasing distances. However, itremains as good as the 6 Mbps fixed-rate mode atlarge distances. This behavior results from its strategy

0 10 20 30 40 50 60 70 80 90 1000

5

10

15

20

25

30

Distance (m)

Thr

ough

put (

Mbp

s)

6 Mbps fixed rate12 Mbps fixed rate24 Mbps fixed rate36 Mbps fixed rate54 Mbps fixed rategenie aidedRate OnlyHigh PerformanceLow Power

Fig. 4. Throughput in a saturated 802.11a point-to-point WLAN as function of the distance between the two nodes.


of using the lowest possible transmit power level andonly optimize (i.e. increase) the transmit rate whenthe minimum power level has been reached.

It is interesting to observe how the throughputefficiency of the fixed-rate modes deteriorates withhigher data rates, even if the distances are short.Whereas the 6 Mbps fixed-rate mode achieves amaximum throughput of approximately 5 Mbps,which corresponds to an efficiency of approximately83%, the efficiency of the 24 Mbps rate is reduced toapproximately 70%, and that of the 54 Mbps dropsto 50%. This is due mainly to the backoff and ACKprocedures of the 802.11 MAC layer: these overheadprocedures are independent of the rates used. At highdata rates, the transmission time of a MAC frame isreduced but not those of the overhead procedures,thus leading to a higher throughput loss. An exhaus-tive discussion of the impact of the MAC overheadon the 802.11 throughput performance can be foundin [23].

Figure 5 demonstrates how well the adaptationalgorithms can manage the transmit power problem.In particular, at distances of less than 20 m, both theHP and LP modes (solid lines) reduce the transmitenergy significantly compared with the RO (dashed

line) and the fixed-rate modes (dotted lines). Asmentioned in Section 2, the power consumed duringthe idles times (e.g. DIFS, SIFS) is set to zero in allsimulation results. Thus the average transmit power�Ptx displayed in Figure 5 contains only the energyconsumed during frame transmissions. For the ROmode, �Ptx first increases with increasing distance, thenremains almost constant. As the transmit power forthat mode is fixed to +10 dBm, the initial increase of�Ptx is due to the switch to lower transmit rates athigher distances, which results in longer transmissiontimes and thus in higher energy consumption. Sim-ilarly, the increase of �Ptx for the HP and LP modes(solid lines) is caused by both the rate reduction andthe power increase to support larger distances. Asexpected, at long distances, the �Ptx of the HP, LP, andRO modes approach the values of the 6 Mbps fixed-rate mode because the adapted transmission rate hasreached the lowest value of 6 Mbps and the transmitpower level the highest value of +10 dBm. There issimply no adaptation potential left.

It is interesting to observe the general behaviorof the fixed-rate modes. With increasing distance,their �Ptx values first decrease and then settle atconstant values at distances where their throughput

0 10 20 30 40 50 60 70 80 90 1000

1

2

3

4

5

6

7

8

9

10

Distance (m)

Ave

rage

Tra

nsm

it P

ower

(m

W)

6 Mbps fixed rate24 Mbps fixed rate54 Mbps fixed rateRate OnlyHigh PerformanceLow Power

Fig. 5. Average transmit power in a saturated 802.11a point-to-point WLAN as function of the distance between the two nodes.


approaches zero. This behavior corresponds to thefact that with increasing distance between nodes theprobability of erroneous transmissions increases.Consequently, nodes spend more time in the backoffmode than in the transmit mode; hence the averagetransmit power is reduced. At very large distances,almost all transmissions are erroneous, leading to aconstant average transmit power in the regions wherethe throughput is zero.

An attentive reader may notice the slightdecrease of the average transmit power for the LPmode at distances around 10 m. When the distancebetween the two stations is very short, a highthroughput is achieved by means of the highest datarate combined with the lowest transmit power levelfor the data frames. However, as mentioned before,we select the strategy of sending the ACK frameswith the highest possible power level because of theirvital importance for the adaptation algorithms: wewant to avoid retransmissions of data frames due toerroneous ACK frames. At very short distances, theenergy used to send ACK frames outweighs the oneused to send data frames. The ratio of this ACKenergy fraction to the one used to transmit dataframe is then reduced as the distance increases and

lower data rates (i.e. larger transmission times andhigher energy consumption) have to be used.

The power efficiencies gTP of the various adap-tation modes are compared in Figure 6. The higherthe gTP is, the more efficient is the correspondingadaptation mode. As expected, the LP mode achievesthe best power efficiency, followed by the HP mode.Both schemes are very efficient, in particular fordistances up to 20 m. In this region the LP modeoutperforms the HP mode because of its strategy offirst reducing the power level and only increasing therate once the minimum power level has been reached.This superiority however comes at the price of veryfast throughput degradation with distance, as shownin Figure 4. When the two nodes are very close toeach other, the efficiency of both modes is almost thesame, because in this case both modes operate at themaximum transfer rate and the lowest power level.Compared with the HP and LP modes, the remainingschemes are less efficient.

Until now we have fixed the power step sizesused by the HP and LP modes, namely to +5 dBwhen increasing the power level (the up size) and to)2 dB when decreasing it (the down size). Figure 7shows the impact of different step sizes on the

0 5 10 15 20 25 30 35 400

5

10

15

20

25

30

35

Distance (m)

Pow

er E

ffici

ency

(M

bps/

mW

)

6 Mbps fixed rate24 Mbps fixed rate54 Mbps fixed rateRate OnlyHigh PerformanceLow Power

Fig. 6. Power efficiency in a saturated 802.11a point-to-point WLAN as function of the distance between the two nodes.


performance of the wireless link. Whereas thethroughput achieved by the LP mode changes onlymarginally when the step sizes are varied, theperformance of the HP mode could be negativelyaffected by an improper step size selection. Takingthe curve with the equal up-down sizes of ±2 dB asreference (solid line), we can see that a high downsize of )5 dB dramatically reduces the throughputof the link (dotted line), whereas the same value inthe opposite direction (up=+5 dB, down=)2 dB)improves the performance (dashed line). Thisbehavior is due to the fact that a large decreasein transmit power level results in additional trans-mission errors, whereas a large power levelimproves the probability of successful transmissions.A further increase of the up size to +8 dB does notyield further improvement, because the power levelsused in the system considered were limited to±10 dBm.

4.1.2. Non-Saturated Point-to-Point Wireless LAN

In this section we study the performance of thevarious adaptation modes if the point-to-point wire-less link is not saturated. The distance between the

two nodes is fixed to 20 m, and the total offered loadgenerated at both stations varies between 0 and20 Mbps. We are interested in the transfer times ofthe data packets, which is an important performancemeasurement for delay-sensitive applications such astelephony or interactive video. The transfer time isdefined as the duration from the packet’s arrival atthe transmitting node until its successful receipt bythe destination node. Thus, it includes the waitingtime the packet experiences at the source node and allits transmission and re-transmission times.

The results for the average transfer time areshown in Figure 8. It can be observed that the ROmode (dashed line) has the best delay performancefollowed by the HP mode. Whereas the fixed-ratemodes can at best support an offered load of16 Mbps before becoming overloaded (see the curvefor the 36 Mbps fixed-rate mode), the HP and ROadaptation schemes still provide finite mean transfertimes for loads as high as 20 Mbps. The LP mode canonly supports loads up to approximately 4 Mbps. Itis interesting that for the channel type consideredhere the 24 Mbps fixed-rate mode achieves lowerdelay values than the 36 or 54 Mbps ones do even forrather low loads from 2 to 14 Mbps. This behavior is

0 5 10 15 20 25 30 35 400

5

10

15

20

25

30

Distance (m)

Thr

ough

put (

Mbp

s)

up= +2dB, down=–2dBup= +2db, down=–5dBup= +5dB, down=–2dBup= +8dB, down=–2dB

HP mode

LP mode

Fig. 7. Throughput of a saturated 802.11a point-to-point WLAN as a function of the distance between the two nodes. Impact of the power

step size.


due to the large variance of the transmission time athigh data rates: As the data rate increases, thetransmission time of a frame becomes shorter if theframe is sent without error. However, if the frame isdisturbed, its transmission time is increased bybackoff times, which are independent of the datarates. Hence the variance of the transmission timeincreases with higher data rates, which in turn leadsto larger mean waiting and mean transfer times.

The average transmit power and the powerefficiency of the point-to-point wireless link areshown in Figure 9. Again, similar to the non-satu-rated case, the LP mode provides the best powerefficiency at the price of high transfer delays. The HPand RO schemes achieve almost the same efficiency atlow offered loads. The advantage of the HP modebecomes more apparent in the region of higherloads; the more data are to be sent, the better theHP mode can adapt the transmission parameters tothe changing link states.

4.2. Infrastructure Wireless LAN with Multiple

WSTAs

Until now we have studied the performance of arather simple point-to-point WLAN. In the following

subsections we increase the system complexity byconsidering WLANs with multiple WSTAs.

4.2.1. Saturated Wireless LAN

Figure 10 shows the throughput performance ofa saturated 802.11a WLAN with 10 WSTAs placedrandomly within a circle of 60 m radius and the AP inthe center. A traffic load of 3 Mbps is generated ateach WSTA, and no traffic is generated at the AP.For each adaptation mode, 20 independent simula-tion runs were performed using different randomgenerator seeds. All WSTAs employ the same adap-tation scheme during a given simulation run. Eachpoint in Figure 10 represents one WSTA. There are20 runs *10 WSTAs=200 points for each adaptationmode.

It can be observed that for the 54 Mbps fixed-rate mode as one extreme case, the generated load of3 Mbps can only be supported by stations that arenot more than 30 m away from the AP. WSTAslocated farther away cannot communicate with theAP. In the other extreme case, where all WSTAscommunicate with the 6 Mbps fixed-rate mode, allstations can still communicate with the AP but theyachieve only a throughput of approximately0.5 Mbps. The LP mode provides a slightly better

0 2 4 6 8 10 12 14 16 18 200

5

10

15

20

25

30

35

40

45

50

Offered Load (Mbps)

Mea

n T

rans

fer

Tim

e (m

s)

6 Mbps fixed rate12 Mbps fixed rate24 Mbps fixed rate36 Mbps fixed rate54 Mbps fixed rateRate OnlyHigh PerformanceLow Power

Fig. 8. Mean transfer time in an 802.11a point-to-point WLAN as function of the offered load (20 m distance between the two nodes).


0 2 4 6 8 10 12 14 16 18 200.5

1

1.5

2

2.5

3

3.5

4

Offered Load (Mbps)

Pow

er E

ffici

ency

(M

bps/

mW

)


0 2 4 6 8 10 12 14 16 18 200

1

2

3

4

5

6

7

8

9

10

Offered Load (Mbit/s)

Ave

rage

Tra

nsm

it P

ower

(m

W)


(a)

(b)

Fig. 9. (a) Average transmit power and (b) power efficiency in an 802.11a point-to-point WLAN as function of the offered load (20 m distance

between the two nodes).


throughput than the 6 Mbps fixed-rate mode canachieve, the HP improves the throughput further,and the RO mode has the best performance. For

distances smaller than 30 m, the three modes HP, LP,and RO achieve lower throughput values than the54 Mbps fixed-rate mode does. This throughput

0 10 20 30 40 50 600

0.5

1

1.5

2

2.5

3

3.5

Distance (m)

Thr

ough

put (

Mbp

s)

6 Mbps fixed rate54 Mbps fixed rateHigh PerformanceLow PowerRate Only

0 10 20 30 40 50 600

5

10

15

20

25

Distance (m)

Pow

er E

ffici

ency

(M

bps/

mW

)


(a)

(b)

Fig. 10. (a) Throughput and (b) power efficiency in an 802.11a infrastructure WLAN with 10 WSTAs as function of the distance to the AP.


reduction is because now WSTAs located far awayfrom the AP can communicate when using theseadaptation schemes. Thus, the total throughput ofthe system is more fairly distributed to all WSTAs bythe three modes HP, LP, and RO than by the fixed-rate modes.

Compared with the power efficiency resultsobtained for the point-to-point case shown in Fig-ure 6, Figure 10(b) illustrates that while all otherschemes achieve almost the same power efficiency asin the point-to-point scenario, the LP mode suffers asignificant efficiency degradation for stations that areclose to the AP. This performance loss is due to thewell-known hidden station issue: Stations that areclose to the AP transmit with low power. Thereforetheir transmissions cannot be detected by the otherstations, which results in an increase in the number ofcollisions.

4.2.2. Non-Saturated Wireless LAN

The transfer delay performance of a non-satu-rated 802.11a infrastructure WLAN with 10 WSTAsdistributed within a circle of 20 m is shown inFigure 11. Each WSTA carries a traffic load of0.3 Mbps. Hence, even the 6 Mbps fixed-rate modeshould be able to support the total generated load of3 Mbps. As can be seen from Figure 11(a), theshortest transfer delay is provided by the RO scheme,followed by HP and the 24 Mbps fixed-rate modes.The results for the 54 Mbps mode shows a steepincrease of the mean transfer time with increasingdistance from the AP. The LP mode provides lowtransfer delays only to WSTAs that are close to AP;otherwise it achieves almost as long delays as the6 Mbps fixed-rate mode does, but with a significantlylower average transmit power, see Figure 11(b). It isinteresting that the hidden station issue as experi-enced in the former scenario (Figure 10(b)) is notapparent in this figure. The reasons are that theWSTAs now carry only 1/10 of the traffic load, i.e.the probability that two or more WSTAs have data tosend at the same time is smaller.

4.3. Infrastructure Wireless LAN with Realistic

Traffic Models

So far all simulation runs were done with trafficgenerated by a Poisson process for the packetarrivals. In this section we will consider a scenariowith more realistic traffic sources. It still is aninfrastructure WLAN with 1 AP and 10 WSTAs

distributed within a circle of 20 m radius, but now thestations support three types of services, namely best-effort data, interactive telephony, and real-timevideo. The best-effort data service is supported byall WSTAs and is modeled by sources generatingpackets having exponentially distributed inter-arrivaltimes. The packet-length distribution is taken from areal LAN trace, with a mean packet length of501 bytes [26]. An up/downlink ratio of 20 kbps/100 kbps has been chosen to emulate the asymmet-rical behavior of Web-browsing-like services.

Besides the data service, one half of the WSTAssupports telephony and the other half video services.Telephony voice signals are modeled by a two-stateON/OFF process, emulating talk spurts followed bysilence periods. The talk-spurt and silence durationtimes are exponentially distributed with a mean valueof 1 and 1.35 s, respectively [27]. During the talkspurts, the voice signals are encoded with the ITU-TG.711 speech codec [28], which generates 160-byte-long packets every 20 ms, corresponding to a con-stant bit-rate (CBR) of 64 kbps. Furthermore, aheader compression has been assumed that reducesthe 40 bytes of the RTP/UDP/IP header to 4 bytes[29].

The video service emulates a MPEG downlinkstreaming service delivered to the end user from astorage database, e.g. a DVD player or a Web VideoContent Server. We assume that the video flows aregenerated by a CBR source with a rate of 500 kbpsand a fixed packet length of 1500 bytes.

Figure 12 displays the uplink and downlinkmean transfer times for the voice service. It can beseen that the RO mode achieves the best delayperformance, followed by the HP and the 24 Mbpsfixed-rate schemes. The LP mode provides similarresults as the 6 Mbps fixed-rate mode. While the54 Mbps fixed-rate scheme exhibits quite small valuesin the uplink direction, its results for the downlinkdirection are not visible in Figure 12(b) because oftheir very high values. In general, the downlinkperforms worse than the uplink because of the natureof the infrastructure WLAN in which all traffic has togo via the AP, thus the AP has to carry a highertraffic load than the WSTAs. Note that the points onthe x-axis of Figure 12 correspond to those WSTAsthat do not have telephony service. The same appliesto Figure 14.

The results for the throughput of the best-effortdata service are shown in Figure 13(a) and (b). Withthe exception of 54 Mbps fixed-rate scheme, allmodes are capable of supporting the offered data


0 2 4 6 8 10 12 14 16 18 200

0.5

1

1.5

2

2.5

3

Distance (m)

Tra

nsfe

r T

ime

(ms)

6 Mbps fixed rate24 Mbps fixed rate54 Mbps fixed rateHigh PerformanceLow PowerRate Only

0 2 4 6 8 10 12 14 16 18 200

0.1

0.2

0.3

0.4

0.5

Distance (m)

Ave

rage

Tra

nsm

it P

ower

(m

W) 6 Mbps fixed rate


(a)

(b)

Fig. 11. (a) Mean transfer times and (b) average transmit power in an 802.11a infrastructure WLAN with 10 WSTAs as function of the

distance to the AP.


0 2 4 6 8 10 12 14 16 18 200

0.5

1

1.5

2

2.5Uplink Voice Transfer Time

Distance (m)

Tra

nsfe

r T

ime

(ms)


0 2 4 6 8 10 12 14 16 18 200

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5Downlink Voice Transfer Time

Distance (m)

Tra

nsfe

r T

ime

(ms)


(a)

(b)

Fig. 12. (a) Uplink and (b) downlink voice transfer times in an 802.11a infrastructure WLAN with 10 WSTAs as a function of the distance to

the AP.


0 2 4 6 8 10 12 14 16 18 200.014

0.015

0.016

0.017

0.018

0.019

0.02

0.021

0.022Uplink Data Throughput

Distance (m)

Thr

ough

put (

Mbp

s)


0 2 4 6 8 10 12 14 16 18 200.07

0.075

0.08

0.085

0.09

0.095

0.1

0.105Downlink Data Throughput

Distance (m)

Thr

ough

put (

Mbp

s)


(a)

(b)

Fig. 13. (a) Uplink and (b) downlink data throughput in an 802.11a infrastructure WLAN with 10 WSTAs as a function of the distance

to the AP.


traffic for distances of up to 20 m. Similar results areobtained for the throughput of the downlink videostream, see Figure 14, in which the video throughputis slightly reduced when the WLAN operates in the54 Mbps fixed-rate mode.

The average transmit power and the powerefficiency of the WSTAs are shown in Figure 15Again, it can be seen that in terms of power efficiencythe two adaptation modes HP and LP provide thebest values compared with the other modes withoutpower regulation. Note that for each mode there aretwo sets of points, corresponding to WSTAs with andwithout telephony traffic; here, WSTAs with tele-phony traffic have a better efficiency. The more data astation has to send, the more efficiently it works.Furthermore, those that receive video streams spenda lot of power acknowledging the video framescoming from the AP.

5. CONCLUSION

In this paper, we have introduced new andsimple but powerful dynamic link adaptation

schemes for 802.11a and 802.11g WLANs. Theseschemes utilize the immediate ACK strategy foradjusting the link parameters. Hence, the algorithmswork autonomously in the transmitter; no additionalfeedback is required from the receiver side, which inturn provides a fully standard-compliant solution.The required additional implementation complexityis very low: Because the approach only countssuccessful and unsuccessful transmissions, there isno channel estimation or other expensive operationinvolved. In contrast to most of the approachesdiscussed at the beginning of the paper, no RTS/CTSscheme is required here. It can, however, be appliedas an independent option if necessary.

Throughout the paper, we have analyzed severalpossibilities to adapt the link parameters power anddata rate dynamically in dependence of the channelconditions. It has been shown that the optimizationstrategy depends on the scenario considered, thenumber of active stations in a system, and the servicerequirements. However, it turns out that in manycases the strategy to transmit at the highest possiblerates with rather high power levels, as is done in theHP mode, is the best choice. The most important

0 2 4 6 8 10 12 14 16 18 200

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

Downlink Video Throughput

Distance (m)

Thr

ough

put (

Mbp

s)


Fig. 14. Downlink video throughput in an 802.11a infrastructure WLAN with 10 WSTAs as a function of the distance to the AP.


0 2 4 6 8 10 12 14 16 18 200

0.02

0.04

0.06

0.08

0.1

0.12

0.14WSTA Average Transmit Power

Distance (m)

Ave

rage

Tra

nsm

it P

ower

(m

W) 6 Mbps fixed rate


0 2 4 6 8 10 12 14 16 18 200

0.5

1

1.5

2

2.5

3

3.5

4

4.5WSTA Power Efficiency

Distance (m)

Pow

er E

ffici

ency

(M

bps/

mW

)


(a)

(b)

Fig. 15. (a) Average transmit power and (b) power efficiency of a WSTA in an 802.11a infrastructure WLAN with 10 WSTAs as function of

the distance to the AP.


arguments to support this statement are the achiev-able low transfer times, the higher overall throughputsupported, and the relaxation of the hidden terminalproblem. Moreover, when also taking into accountthe overall power consumption during idle times, thestrategy to keep the transmission times as small aspossible provides the greatest potential for puttingidle components into a low-power mode.

The HP mode suffers some throughput degra-dation compared with the RO mode because ofunsuccessful switching attempts; it is however muchmore power-efficient for low distances between thetransmitter and the receiver. The LP mode should beconsidered only if energy is the most critical issue andthe required data rates are low.

REFERENCES

1. IEEE Std 802.11a-1999, Supplement to IEEE Standard forInformation Technology—Telecommunications and Informa-tion Exchange between Systems—Local and Metropolitan AreaNetworks—Specific Requirements—Part 11: Wireless LANMedium Access Control (MAC) and Physical Layer (PHY)Specifications: High-speed Physical Layer in the 5 GHz Band,IEEE 1999.

2. IEEE Std 802.11g-2003, IEEE Standard for Information Tech-nology—Telecommunications and Information Exchangebetween Systems—Local and Metropolitan Area Net-works—Specific Requirements—Part 11: Wireless LAN Med-ium Access Control (MAC) and Physical Layer (PHY)Specifications—Amendment 4: Further Higher Data RateExtension in the 2.4 GHz Band, June 2003.

3. IEEE Std 802.11-1999, Information Technology—Telecommu-nications and Information Exchange Between Systems—Localand Metropolitan Area Networks—Specific Require-ments—Part 11: Wireless LAN Medium Access Control (MAC)and Physical Layer (PHY) Specifications, 1999.

4. G. Holland, N. Vaidya, and P. Bahl, A Rate-Adaptive MACprotocol for multi-hop wireless networks. In Proc. MOBI-COM’01, Rome, Italy, July 2001.

5. A. Kamerman and L. Montean, WaveLAN-II: A High-Per-formance Wireless LAN for the Unlicensed Band, Bell LabsTechnical Journal, pp. 118–133, Summer 1997.

6. D. Qiao and S. Choi, Goodput Enhancement of IEEE 802.11aWireless LAN via Link Adaptation. In Proc. ICC’2001, Vol. 7,pp. 1995–2000, Helsinki, Finnland, 2001.

7. A. Forenza and R. W. Heath Jr., Link Adaptation and ChannelPrediction in Wireless OFDM Systems. In Proc. Globecom’02,Vol. 3, pp. 211–214, 2002.

8. S. T. Sheu, Y. H. Lee, and M. H. Chen, Providing MultipleData Rates in Infrastructure Wireless Networks. In Proc. GlobalTelecommunications Conf. 2001, Vol. 3, pp. 1908–1912,San Antonio, TX, USA, 2001.

9. Y. J. Cheng, Y. H. Lee, and S. T. Sheu, Multi-rate Transmis-sions in Infrastructure Wireless LAN based on IEEE 802.11bProtocol. In Proc. 54th VTC 2001, Vol. 4, pp. 2609–2612,Atlantic City, NJ, USA, 2001.

10. A. Noll Barreto, P. Chevillat, J. Jelitto, and H. L. Truong, ADynamic Link Adaptation Algorithm for IEEE 802.11a Wire-

less LANs. In Proc. ICC 2003—IEEE 2003 Int’l Conf. onCommunications, Vol. 2, pp. 1141–1145, Anchorage, AK, USA,May 2003.

11. J. del Prado Pavon and S. Choi, Link Adaptation Strategy forIEEE 802.11 WLAN via Received Signal Stregth Measurement.In Proc. ICC 2003—IEEE 2003 Int’l Conf. on Communications,Anchorage, AK, USA, May 2003.

12. M. Krunz and A. Muqattash, A Power Control Scheme forMANETs with Improved Throughput and Energy Consump-tion. In Proc. of the 5th International Symposium on WirelessPersonal Multimedia Communications, Vol. 2, pp. 771–775,October 2002.

13. C. R. Lin and C. Y. Liu, Enhancing the Performance of IEEE802.11 Wireless LAN by Using a Distributed Cycle StealingMechanism. In Proc. 4th IEEE Conference on Mobile andWireless Communications Network, pp. 564–568, Stockholm,Sweden, September 2002.

14. S. Agarwal et al., Distributed Power Control in Ad Hoc Wire-less Networks. In Proc. IEEE PIMRC’01, 2: F-59–F-66, 2001.

15. J.-P. Ebert and A. Wolisz, Combined Tuning of RF Power andMedium Access Control for WLANs, Mobile Networks Appl.,Vol. 5, No. 6, pp. 417–426, 2001.

16. A. Kraemling et al., Interaction of Power Control and LinkAdaptation for Capacity Enhancement and QoS Assistance. InProc. of 13th IEEE International Symposium on Personal Indoorand Mobile Radio Communications (PIMRC 2002), Vol. 2,pp. 697–701, Lisbon, Portugal, September 2002.

17. D. Qiao, S. Choi, Amjad Soomro, and K. G. Shin, Energy-Efficient PCF Operation of IEEE 802.11a Wireless LAN. InProc. IEEE INFOCOM02, Vol. 2, pp. 580–589, New York,USA, June 2002.

18. D. Qiao, S. Choi, A. Jain, and K.G. Shin, Adaptive TransmitPower Control in IEEE 802.11a Wireless LANs. In Proc. 57thIEEE Semiannual Vehicular Technology Conference, Vol. 1,pp. 433–437, Jeju, Korea, April 2003.

19. S. Simoens and D. Bartolome, Optimum Performance of LinkAdaptation in HIPERLAN/2 Networks. In Proc. of VTC 2001,Vol. 2, pp. 1129–1133, 2001.

20. Z. Lin, G. Malmgren, and J. Torsner, System PerformanceAnalysis of Link Adaptation in HIPERLAN Type 2. In Proc. ofthe VTC 2000 Fall, Vol. 4, pp. 1719–1725, 2000.

21. S. L. Wu, Y. C. Tseng, C. Y. Lin and J. P. Sheu, A Multi-Channel MAC Protocol with Power Control for Multi-HopMobile Ad Hoc Networks, Computer J., Vol. 45, No. 1, pp.101–110, 2002.

22. J. Jelitto, A. Noll Barreto, and H. L. Truong, ‘‘Power and RateAdaptation in IEEE 802.11a Wireless LANs,’’ in Proc. 57thIEEE Semiannual Vehicular Technology Conference, Vol. 1,pp. 413–417, Jeju, Korea, April 2003.

23. J. Jelitto and H. Linh Truong, ‘‘A Multi-MAC Approach toOvercome the Throughput Limits of 802.11 WLANs,’’ in Proc.2004 International Zurich Seminar on Communications, pp. 32–35, Zurich, Switzerland, February 2004.

24. R. Nee and R. Prasad, OFDM Wireless Multimedia Communi-cations, Chapter 10.5.6,Artech House, Boston, London, 2000.

25. H. Hashemi, ‘‘The Indoor Radio Propagation Channel,’’ Proc.IEEE, Vol. 81, No. 7, pp. 943–968, July 1993.

26. B. H. Walke et al., ‘‘IP over Wireless Mobile ATM—Guaran-teed Wireless QoS by HiperLAN/2,’’ Proc. IEEE, Vol. 89,pp. 21–40, January 2001.

27. D. J. Goodman and H. Xie, ‘‘PRMA System Performance withVariable Number of Calls,’’ Technical Report WINLAB-TR-36,June 1992.

28. ITU-T Recommendation G.711, Pulse Code Modulation(PCM) of Voice Frequencies, 1988.

29. K. Svanbro, ‘‘Lower Layer Guidelines for Robust RTP/UDP/IP Header Compression,’’ RFC 3409, December 2002.


Pierre R. Chevillat ([email protected]) received a Dipl.

El-Ing. degree fromETHZurich, Switzerland, andMSEE and Ph.D.

degrees from Illinois Institute of Technology, Chicago, USA. He

joined the IBM Zurich Research Lab in 1976, where he has done

research in signal processing, coding algorithms, and protocols for

wireless and wireline communication systems. As a manager (since

1985) he has led research projects on voiceband data modems,

detectors for magnetic recording, radio and infrared wireless LANs,

wideband CDMA for 3G cellular systems, and sensor networking.

Dr. Chevillat’s has received IBMOutstanding Technical Innovation

and Achievement Awards. He is an IEEE Fellow (’95) and received

the IEEEMillenniumMedal.He has been anAssociateEditor for the

IEEETransactions on Information Theory and is currently an editor

of the International Journal of Wireless Information Networks. He

serves on the scientific boards of the ‘‘Deutsche Forschungsgeme-

inschaft’’ and the Swiss National Competence Center on Mobile

Communication, andhewas amember of the senate committee of the

‘‘Deutsches Zentrum fur Luft- und Raumfahrt’’ (DLR).

Jens Jelitto ([email protected]) received his M.Sc./Dipl.-

Ing. and Ph.D. degrees from the Dresden University of Technol-

ogy, Germany, in 1995 and 2001, respectively. From 1995 to 1996,

he worked in the field of speech recognition at the Institute for

Acoustics and Speech Communication in Dresden. In July 1996 he

joined the Mannesmann Mobilfunk Chair for Mobile Communi-

cations Systems at the Dresden University of Technology, Ger-

many, to work towards his Ph.D. degree, where his main research

interests included digital signal processing, smart antennas and

spatial dimension reduction problems. Since March 2001 he is a

research staff member at the IBM Zurich Research Laboratory,

Ruschlikon, Switzerland, working in the field of digital signal

processing for wireless LANs and for magnetic recording.

Hong Linh Truong ([email protected]) received the M.S.

degree in 1977 and the Ph.D. degree in 1981, both in Electrical

Engineering from the University of Stuttgart, Germany. He joined

the IBM Research Division in 1990 at the Zurich Research

Laboratory, where he is currently involved in the design and

development of communication architecture for wireless sensor

networks. He has also worked on architectures for ATM, IP

Telephony, and Wireless LANs. In particular, he was the main

inventor of LAN Emulation, a method for running applications on

ATM networks. Before joining IBM he was with the Institute of

Switching and Data Techniques at the University of Stuttgart until

1982, where he worked in the field of performance analysis of data

communication networks. From 1982 to 1990, he was employed by

SEL Alcatel, Stuttgart, Germany, where he was responsible for the

definition and specification of signaling protocols in ISDNs and

mobile networks.


dynamic data rate and transmit power adjustment in ieee 802.11

Documents