Dynamic bandwidth partition schemes for integrated voice, video, and data traffic in the IEEE...

INTERNATIONAL JOURNAL OF COMMUNICATION SYSTEMSInt. J. Commun. Syst. 2010; 23:391–412Published online 9 November 2009 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/dac.1077

Dynamic bandwidth partition schemes for integrated voice, video,and data traffic in the IEEE 802.11e distributed wireless LANs

Yang Xiao1,∗,†, Frank Haizhon Li2, Ming Li3, Jingyuan Zhang1 and Bo Li4

1Department of Computer Science, University of Alabama, AL, U.S.A.2Division of Computer Science, University of South Carolina, Upstate, SC, U.S.A.

3Department of Computer Science, California State University, U.S.A.4Department of Computer Science, Hong Kong University of Science and Technology, Hong Kong

SUMMARY

In mobile cellular networks, bandwidth is deterministic in terms of the number of channels by frequencydivision, time division, or code division. On the other hand, bandwidth partition schemes in the contention-based medium access control (MAC) in distributed wireless LANs are extremely challenging due to thecontention-based nature, packet-based network, and the most important aspect: only one channel available,competed by an unknown number of stations. In this paper, we study this challenging issue. We proposeand study four different bandwidth partition schemes for integrated voice/video/data traffic in the IEEE802.11e wireless LANs: a Static bandwidth Partition (SP) scheme, a Dynamic budget Partition (DP)scheme, a Dynamic bandwidth Partition with Finer-Tune (DP-FT) scheme, and a Dynamic bandwidthPartition with Reserved Region (DP-RR). The proposed schemes are compared and evaluated via extensivesimulations. Results show that the DP-FT scheme is the best scheme. Copyright q 2009 John Wiley &Sons, Ltd.

Received 6 May 2009; Revised 7 September 2009; Accepted 8 September 2009

KEY WORDS: bandwidth allocation; 802.11e; WiFi

1. INTRODUCTION

Almost all the end-user networks need a medium access control (MAC) layer. The MAC layeris very fundamental for wired/wireless networks, such as Ethernet and IEEE 802.11 WLAN [1].Ethernet and the IEEE 802.11 distributed WLAN have become widely deployed since these

∗Correspondence to: Y. Xiao, Department of Computer Science, University of Alabama, AL, U.S.A.†E-mail: [email protected]

Contract/grant sponsor: NSF; contract/grant numbers: CNS-0737325, CNS-0716211, CCF-0829827Contract/grant sponsor: RGC; contract/grant numbers: 615608, 616207Contract/grant sponsor: NSFC/RGC; contract/grant number: N HKUST603/07Contract/grant sponsor: HKUST; contract/grant number: RPC06/07.EG27

Copyright q 2009 John Wiley & Sons, Ltd.

392 Y. XIAO ET AL.

contention-based MAC protocols are simple, robust, and they allow fast installation with minimalmanagement and maintenance costs. The MAC of IEEE 802.11 WLAN employs a mandatorycontention-based channel access function called Distributed Coordination Function (DCF), andan optional, centrally controlled, channel access function called Point Coordination Function(PCF) [1]. The DCF adopts a carrier sense multiple access with collision avoidance (CSMA/CA)with binary exponential backoff. It is considered a wireless version of the most successful LANtechnology, i.e. IEEE 802.3 (Ethernet), which adopts a CSMAwith collision detection (CSMA/CD)with binary exponential backoff. Both the IEEE 802.11 DCF and IEEE 802.3 are very robustprotocols for the best-effort service. The popularity of the IEEE 802.11 WLAN is mainly due tothe DCF, whereas the PCF is barely implemented in today’s products due to many reasons.

On the other hand, centrally controlled MAC protocols and reservation-based protocols managequality of service (QoS) more easily, but they are hardly implemented in today’s products for amyriad of reasons: centrally controlled MAC protocols have higher complexity and are inefficientfor normal data transmission; reservation-based protocols have higher complexity, lack robustness,and make strong assumptions such as global synchronizations; and finally, users who set up andmaintain WLANs prefer contention-based protocols due to its simple configuration procedure. Thissituation resembles the well-known ‘war’ between ATM networks and IP networks several yearsago. It is likely that the contention-based wireless MAC protocols will be still widely adopted inthe future mostly due to similar reasons leading to the success of IP and Ethernet. In this proposedwork, we choose to focus on the most challenging choice, i.e. QoS bandwidth allocations at thecontention-based wireless MAC layer. Without QoS support at the MAC layer, QoS support fromhigher layers will be difficult and inefficient.

Although contention-based MAC protocols are very successful commercially and are robust forthe best-effort traffic, they are unsuitable for multimedia applications with QoS requirements [2–5].However, QoS is important and necessary for real-time traffic such as voice and video. If a stationhas to wait an arbitrarily long time to transmit a frame, real-time applications may suffer [5].One possible solution is to provide a good priority scheme for the DCF. Simple DCF priorityschemes can be easily designed with minor changes to the DCF, and they are quite effective [3, 4].To support the MAC-level QoS, the IEEE 802.11 Working Group has recently developed IEEE802.11e [6]. The emerging IEEE 802.11e standard provides QoS features and multimedia supportto the existing 802.11b/g and 802.11a WLANs, while maintaining a full backward compatibilitywith these legacy standards. The IEEE 802.11e MAC employs a channel access function, calledHybrid Coordination Function (HCF), which includes both contention-based channel access andcentrally controlled channel access mechanisms. The contention-based channel access mechanismis also referred to as Enhanced Distributed Channel Access (EDCA). The EDCA provides a priorityscheme by differentiating the inter-frame space as well as the initial and the maximum contentionwindow sizes for backoff procedures.

In the previous work in [2–5, 7–13], the main focus was on studying the EDCA mechanismsand differentiated services. However, without a good admission control mechanism and a goodprotection mechanism, the existing multimedia traffic cannot be protected and QoS requirementscannot be met. These schemes cannot provide guaranteed QoS, and the multimedia traffic cannotbe protected sufficiently. In previous works [14, 15], realistic two-level QoS protection and guar-antee schemes have been proposed so that the existing voice and video flows are protected fromthe new and other existing voice and video flows. However, in [14, 15], bandwidth allocationproblem has never been touched. In [16], bandwidth guarantee is provided to voice flows, whereasvideo flows are transmitted with best effort. In [17], we proposed bandwidth sharing schemes,

Copyright q 2009 John Wiley & Sons, Ltd. Int. J. Commun. Syst. 2010; 23:391–412DOI: 10.1002/dac

DYNAMIC BANDWIDTH PARTITION SCHEMES 393

which can be complementary schemes to the proposed schemes in this paper. In [18], the authorsstudy a feedback scheme for bandwidth allocation algorithm for the 802.11e. There are otherrelated schemes for 802.11, such as [19–22], and for 802.15.4, such as [23–27]. QoS guaranteeand bandwidth allocation schemes [28–34] have been well studied for mobile cellular networks,where bandwidth is deterministically allocated in terms of the number of channels by frequencydivision, time division, or code division. On the contrary, bandwidth allocation in contention-baseddistributed wireless LANs is extremely challenging due to the contention constraint, the packet-based network, and, most importantly, intense competition for the only one available channel. As aconsequence, both guaranteeing bandwidth and allocating bandwidth are challenging issues. Oneof the key challenges is to guarantee the different QoS requirements for different traffic classes,while simultaneously ensuring that the scarce bandwidth is utilized efficiently.

In this paper, we propose and study four different bandwidth partition schemes for integratedvoice/video/data traffic in the IEEE 802.11e wireless LANs: a Static bandwidth Partition (SP)scheme, a Dynamic budget Partition (DP) scheme, a Dynamic bandwidth Partition with Finer-Tune (DP-FT) scheme, and a Dynamic bandwidth Partition with Reserved Region (DP-RR). Eachscheme contains both admission control and data control. In admission control mechanisms, videoand voice flows are accepted or rejected based on the available budget, and a guard period isproposed to prevent bandwidth allocation from over-provisioning. In the data control mechanism,best-effort data parameters are dynamically controlled based on traffic load condition. In a dynamicbandwidth partition scheme, bandwidth is dynamically partitioned among voice and video basedon current voice/video/data traffic load condition. In the DP-FT scheme, bandwidth of voiceand video is partitioned proportionally to voice traffic load and video traffic load in the previousmeasurement interval, and a Finer-Tune (FT) method is adopted to handle some extreme casesto avoid starvations and over-provisioning for another real-time traffic. The FT scheme is definedas borrowing some bandwidth/budget from another Access Category (AC) if available beforerejecting a flow. In the DP-RR scheme, bandwidth of voice and video is partitioned proportionallyto voice traffic load and video traffic load in the previous measurement interval and two ReservedRegions (RRs) are reserved for voice and video, respectively. The proposed schemes are comparedand evaluated with extensive simulations.

The remainder of paper is organized as follows. We briefly introduce the IEEE 802.11 DCFand the 802.11e EDCA in Section 2. The SP partition scheme with a guard concept is proposed inSection 3. Dynamic partition schemes are proposed in Section 4. Performance studies are carriedout in Section 5 with extensive simulation results. We conclude our paper in Section 6.

For convenience of readers, we include a list of acronyms used in this paper in Table I.

2. IEEE 802.11 DCF AND IEEE 802.11e EDCA

We briefly introduce the IEEE 802.11 DCF and the IEEE 802.11e EDCA, an earlier 802.11edraft, i.e. IEEE 802.11e/D4.3, for differentiated services of the EDCA in subsection 2.1 andsubsection 2.2, respectively.

2.1. IEEE 802.11 DCF

The IEEE 802.11 MAC employs a mandatory DCF and an optional PCF. In a long run, time isdivided into repetition intervals called superframes. Each superframe starts with a beacon frame,


394 Y. XIAO ET AL.

Table I. Acronyms.

MAC Medium access controlSP Static bandwidth PartitionDP Dynamic budget PartitionDP-FT Dynamic bandwidth Partition with Finer-TuneDP-RR Dynamic bandwidth Partition with Reserved RegionDCF Distributed Coordination FunctionPCF Point Coordination FunctionCSMA/CA Carrier sense multiple access with collision avoidanceCSMA/CD Carrier sense multiple access with collision detectionQoS Quality of ServiceWLAN Wireless LANEDCA Enhanced Distributed Channel AccessFT Finer-TuneAC Access CategoryRRs Reserved RegionsCFP Contention-free periodCP Contention periodRTS Request-to-sendCTS Clear-to-sendTBTTs Target beacon transmission timesDIFS Distributed inter-frame spaceCW Current window sizeACK Acknowledgment frameHCF Hybrid Coordination FunctionQAP QoS Access PointQSTAs QoS StationsQPSE QoS Parameter Set ElementTXOPBudget[i] The additional amounts of time available for AC I during the

next beacon intervalSurplusFactor[i] The ratio of over-the-air bandwidth reserved for AC i to the

required bandwidth of the transported frames for successfultransmission

TxUsed[i] The amount of time occupied on-air by transmissions, irrespectiveof success or not, from AC i of this station, including associatedSIFS and ACK times if applicable

TxSuccess[i] The transmission time for successful transmissions from AC i ofthis station

TxLimit[i] The transmission time limit from AC i of this station

and the remaining time is further divided into a contention-free period (CFP) and a contentionperiod (CP). The DCF works during the CP and the PCF works during the CFP. If the PCF isnot active, superframes do not exist. However, the beacon frames are periodically transmittedirrespectively. The beacon frame is a management frame for synchronization, power management,and delivery of network operation parameters. Beacon frames are generated in regular intervalscalled target beacon transmission times (TBTTs).

The DCF defines a basic access mechanism and an optional request-to-send/clear-to-send(RTS/CTS) mechanism. Under the DCF, a station with a frame to transmit monitors the channelactivities until an idle period equal to a distributed inter-frame space (DIFS) is detected. Aftersensing an idle DIFS, the station waits for a random backoff interval before transmitting.



The backoff time counter is decremented in terms of slot time as long as the channel is sensedidle. The counter is suspended when a transmission is detected on the channel, and resumed withthe old remaining backoff interval when the channel is sensed idle again for a DIFS interval. Thestation transmits its frame when the backoff timer reaches zero. For each new transmission attempt,the backoff interval is uniformly chosen from the range [0,CW −1] in terms of timeslots, whereCW is the current contention window size. At the very first transmission attempt, CW equals theinitial backoff window size CWmin. After each unsuccessful transmission, CW is doubled until amaximum backoff window size value CWmax is reached. The station keeps track of the number ofretry for each frame. If the number of retry reaches the retry limit, which is a predefined WLANparameter, the station drops this frame. After the destination station successfully receives theframe, it transmits an acknowledgment frame (ACK) following a short inter-frame space (SIFS)time. If the transmitter station does not receive an ACK within a specified ACK Timeout, itreschedules the frame transmission according to the backoff rules discussed above.

2.2. IEEE 802.11e EDCA

IEEE 802.11e provides a channel access function, called HCF to support applications with QoSrequirements. The HCF includes both contention-based channel access and centrally controlledchannel access schemes. The contention-based channel access of the HCF is also referred to asEDCA.

The EDCA works with four Access Categories (ACs), which are virtual DCFs, and each ACachieves a differentiated channel access. This differentiation is achieved through varying theamount of time a station would sense the channel to be idle and the length of the contentionwindow for a backoff. The EDCA supports eight different priorities, which are further mappedinto four ACs, where AC VO (or AC 3), AC VI (or AC 2), AC BE (or AC 1), and AC BK(or AC 0) correspond to voice, video, best-effort, and background traffic, respectively. We, in thiswork, assume that data traffic is served via AC 0. Differentiated ACs are achieved by varying thearbitration inter-frame space (AIFS), the initial window size, and the maximum window size. Thatis, for AC i(i=0, . . . ,3), the initial contention window size is CWmin[i], the maximum contentionwindow size is CWmax[i], and the arbitration inter-frame space is AIFS[i]. For 0�i< j�3, we haveCWmin[i]�CWmin[ j], CWmax[i]�CWmax[ j], and AIFS[i]�AIFS[ j]. In other words, the EDCAemploys AIFS[i], CWmin[i], and CWmax[i] (all for i=0, . . . ,3) instead of DIFS, CWmin, andCWmax, respectively. If one AC has a smaller AIFS/CWmin/CWmax, its traffic has a better chanceto access the wireless medium earlier.

Four transmission queues are implemented in a station, and each queue supports one AC,behaving roughly as a single DCF entity in the original IEEE 802.11 MAC. It is assumed thata payload from a higher layer is labeled with a priority value, and put into the correspondingqueue according to the mapping. Each queue acts as an independent MAC entity and performsthe channel access with a different inter-frame space (AIFS[i]), a different initial window size(CWmin[i]), and a different maximum window size (CWmax[i]). Each queue has its own backoffcounter (BO[i]), which acts independently in the same way as the original DCF backoff counter.If there is more than one queue finishing the backoff at the same time, the highest AC frame ischosen to transmit by the virtual collision handler. Other lower AC frames whose backoff countersalso reach zero will increase their backoff counters with CWmin[i] (i=0, . . . ,3), accordingly.

The values of AIFS[i], CWmin[i], and CWmax[i] (all for i=0, . . .,3), are referred to as the EDCAparameters, which will be announced by the QoS Access Point (QAP) via periodically transmitted


396 Y. XIAO ET AL.

beacon frames. The QAP can also adaptively adjust these EDCA parameters based on the networktraffic conditions.

3. STATIC PARTITION

To define bandwidth in a contention-based channel is not as simple as in a cellular network.In this paper, measurements are conducted during each regular time interval, which can be abeacon interval or several beacon intervals. We define bandwidth T as the time interval betweentwo measurements, and it is a constant value. If a portion of bandwidth is partitioned, each ACoccupies one partition of the whole portion, although the partition may be changed dynamicallybased on traffic load if it is a dynamic bandwidth partition scheme. On the other hand, if a portionof bandwidth is reserved, one particular AC occupies the whole portion.

A general equation of bandwidth allocation is given as follows, where � is budget, T is band-width, and � j is allocation.

�T +∑j

� j T =T (1)

In a fixed bandwidth partition scheme as shown in Figure 1(a), the total bandwidth is dividedinto three portions: a portion (�1T ) for video only, a portion (�2T ) for voice only, and a guardperiod (�T ) to prevent bandwidth allocation from over provisioning and for best-effort data traffic,where �+�1+�2=1. The guard period will be explained in the later subsection. Some bandwidthcan be reserved for either voice flows or video flows if needed. Parameters such as � and � aresystem tunable parameters, which are chosen based on system’s requirements such as how muchpercentage of the bandwidth is reserved for video and voice. Note that a partition scheme has noforward/backward issue, which is the order of using bandwidth.

We propose four different bandwidth partition schemes, shown in Figure 1 for integratedvoice/video/data traffic in the IEEE 802.11e wireless LANs: a Static bandwidth Partition (SP)scheme, a Dynamic budget Partition (DP) scheme, a Dynamic bandwidth Partition with Finer-Tune(DP-FT) scheme, and a Dynamic bandwidth Partition with Reserved Region (DP-RR). In DP-RR scheme, as shown in Figure 1(d), �2 and �3 represent the portion of bandwidth dynamically

T

1T

2T

1T

2T

T T

1T

2T

3T

4T

1B

2B

B

(a) (b) (c) (d)

Figure 1. Bandwidth partition schemes: (a) fixed partition; (b) dynamic partition with finer-tune;(c) dynamic partition; and (d) dynamic partition with reserved region.



allocated for video and voice, respectively. Moreover, �1 and �4 represent the portion of band-width reserved for video and voice, respectively. Figure 1(a) is a special case of Figure 1(d) when�2=�3=0. The DP, DP-FT, and DP-RR schemes are given in the next section.

Any bandwidth partition scheme consists of two parts: (1) admission control for voice and videoflows and (2) data control for data traffic.

3.1. Static partition

The Static Partition (SP) scheme is shown in Figure 1(a). We only discuss admission control parthere, and data control is discussed in the later subsection. In this subsection, an AC stands foreither voice AC or video AC.

The distributed admission control is developed to protect active QoS flows, i.e. voice and videoflows. The QAP announces the transmission budget via beacon frames, and the budget is sharedby both voice and video. The budget indicates the allowable transmission time in addition to howmuch is being utilized. QoS Stations (QSTAs) determine an internal transmission limit per AC foreach beacon interval, based on the transmission count during the previous beacon period and thetransmission budget announced from the QAP. The local voice/video transmission time per beaconinterval shall not exceed the internal transmission limit per AC. When the transmission budget isdepleted, new flows will not be able to gain transmission time, whereas existing flows will notbe able to increase the transmission time per beacon interval, which they are already using. Thismechanism protects existing flows.

(1) Procedure at QAP: The QoS Parameter Set Element (QPSE) provides information neededby QSTAs for a proper operation of the QoS facility during a CP. The QPSE includesCWmin[i], CWmax[i], AIFS[i], for (i=0, . . . ,3), and TXOPBudget[i] for (i=1,2,3),and SurplusFactor[i] for (i=1,2,3). These are global variables in the sense that theyare maintained by QAP and transmitted to QSTAs via beacon frames. The first threevariables/parameters were already discussed in the previous sections. TXOPBudget[i] spec-ifies the additional amounts of time available for AC i , respectively, during the next beaconinterval, and SurplusFactor[i](>1) represents the ratio of over-the-air bandwidth reservedfor AC i to the required bandwidth of the transported frames for successful transmission.Note that with the SurplusFactor, we reserve a bandwidth more than the minimum requiredto compensate potential transmission failures, e.g. due to collisions. The SurplusFactor iscalculated by the QAP for each beacon interval and embedded into the next beacon frame.The QAP shall measure the amount of time occupied by transmissions from each AC

during the beacon period, including associated SIFS and ACK times if applicable. The QAPshall maintain a set of counters TxTime[i], which shall be set to zero immediately followingthe transmission of a beacon. For each data frame transmission (either uplink or downlink),the QAP shall add the time, equal to the frame transmission time and all overhead involvedsuch as SIFS and ACK, to the TxTime counter corresponding to the AC of that frame. TheQAP determines TXOPBudget[i] by

TXOPBudget[3] =max(�2T −TxTime[3]×SurplusFactor[3],0)TXOPBudget[2] =max(�1T −TxTime[2]×SurplusFactor[2],0)

where �1T and �2T are defined in Figure 1(a) (�1+�2+�=1) and AC=1 is not used sincewe assume that data traffic is at AC=0. How to choose SurplusFactor is well studied in theprevious work [15].


398 Y. XIAO ET AL.

(2) Procedure at Each QSTA: When the transmission budget is depleted, new QSTAs cannotgain transmission time, whereas existing QSTAs cannot increase the transmission time perbeacon interval, which they are currently utilizing. Accordingly, this mechanism protectsexisting flows.

Each QSTA has to maintain the following local variables for each AC: TxUsed[i], TxSuccess[i],TxLimit[i], TxRemainder[i], and TxMemory[i]. These variables are local in the sense that theyare updated locally at each station based upon those information related to the station itself.In other words, local variables are those related to a particular station and obtained from theviewpoint of this station, whereas global variables are related to all stations in the commonwireless channel, and obtained from the viewpoint of the AP. TxUsed[i] counts the amount oftime occupied on-air by transmissions, irrespective of success or not, from AC i of this station,including associated SIFS and ACK times if applicable. TxSuccess[i] counts the transmission timefor successful transmissions. A station shall not transmit a data frame if doing so would resultin the value in TxUsed[i] exceeding the value in TxLimit[i], where how to determine this valueis presented below. If the QSTA is prevented from sending a frame for this reason, it may carryover the partial frame time remainder to the next beacon interval, by storing the remainder inTxRemainder[i], where TxRemainder[i]=TxLimit[i]−TxUsed[i]; Otherwise, TxRemainder[i]=0.TxMemory[i] ‘memorizes’ the amount of resource that AC i of this station utilized during a beaconinterval. Let f denote the damping factor whose function will be explained below. Let B[i] denotea predefined budget threshold. Note that B[i] is also referred as to an inside guard period, whichwill be discussed in a later subsection. B[i] is also referred as to non-zero-budget in [14, 15] toprevent over-provisioning, and it is essential to provide a stable quality of video and voice. It issimilar to the guard period, and it has been well studied in [14,15]. At each TBTT, the TxMemory,TxLimit, and TxSuccess variables are updated according to the following procedure:

• If TXOPBudget [i]<B[i],◦ Both TxMemory[i] and TxRemainder[i] shall be set to zero for new QSTAs which starttransmission with this AC in the next beacon interval. All other QSTAs’ TxMemory[i]remains unchanged;

• Else

◦ For new QSTAs, which start transmission with this AC in the next beacon interval, an initialvalue for TxMemory[i] is assigned a number between 0 and TXOPBudget[i]/SurplusFactor[i]. All other QSTAs’ TxMemory[i] are updated according to the following procedure:

– TxMemory[i]= fxTxMemory[i]+(1− f )×(TxSuccess[i]×SurplusFactor[i]+TXOPBudget[i]);

• TxSuccess[i]=0;• TxLimit[i]=TxMemory[i]+TxRemainder[i];Note that in the above procedure, only TXOPBudget[i] and SurplusFactor[i] are global variables,

and the others are local variables. From the above procedure, when the transmission budget for anAC becomes zero,

• Its TxLimit[i] will become zero for new STAs, and hence AC iof any new QSTA will not beable to gain a transmission time in the next beacon interval.

• The existing QSTAs’ TxMemory[i] remains unchanged, and hence the existing QSTAs’TxLimit[i] remains basically unchanged. In other words, existing stations will not be able to



increase the transmission time above what they are currently using. Note that this mechanismprotects existing flows.

From the above procedure, as long as the transmission budget is larger than zero, bothTxMemory[i] and TxLimit[i] need be adjusted periodically. The new TxMemory[i] value is aweighted average of the old TxMemory[i] value and the sum of the successful transmission timeand the budget. The value TxSuccess[i]×SurplusFactor[i]+TXOPBudget[i] is the target to whichTxMemory converges. The TxLimit equals TxMemory plus a possible capped remainder, whereTxMemory ‘memorizes’ the amount of time that a specific AC of the QSTA has been able toutilize per beacon interval. Once the budget is depleted (i.e. TXOPBudget[i] hovers around 0),TxMemory converges to TxSuccess, which is the lower limit. This ensures that a QSTA cancontinue consuming the same amount of time in subsequent beacon intervals. The damping allowsfor some amount of fluctuation to occur. However, TxMemory cannot grow any further in thesaturated state. This prevents new flows from entering a specific AC when it is saturated.

The damping factor does not affect the entrance of a newly entered flow into the system whenan enough budget is available because the decreased TXOPBudget[i] is offset by an increasedTxSuccess instantaneously, and thus TxMemory does not change a lot. In other words, for a newlyentered flow, TXOPBudget[i] is decreased due to this new entrance, and TxSuccess is increasedsince it is changed from zero to a positive value so that the sum of these two in the algorithmabove does not change a lot. The damping factor does affect TxMemory when a new flow startsup in a QSTA, which does not have an existing flow of the corresponding AC. In such a case, thedecreased TXOPBudget[i] is not offset by an increased TxSuccess, and the TxMemory convergesto the lower target value consequently. QSTAs shall not increase their TxLimit[i] if they did nottransmit traffic of AC i during the previous beacon interval.

For each video/voice flow, a Leaky-Bucket algorithm plus a Token-Bucket algorithm can bealso implemented at the QSTA to control the flow rate.

Note that in the equations, we cannot see signal-noise-ratio (SNR) since the proposed schemesare measurement-based schemes, which measure successful frame-transmissions so that impactsof inference and noise are absolved in the measurement results.

3.2. Data control

Since too many data transmissions can degrade the performance of existing voice and video flows,we propose a retry-based data control mechanism in this paper to dynamically adjust data trafficparameters based on traffic condition. In the proposed approach, stations dynamically adjust theEDCA data parameters based on the behavior of one or more frame transmission(s). During eachframe transmission, whenever the number of retries ever reaches a threshold K , the next frame’sinitial window size is increased by CWmin[0]=�×CWmin[0] where � is a constant and �>1;whenever there are L consecutive successful transmissions, the next frame’s initial window size isdecreased by CWmin[0]=CWmin[0]/�. Note the above changes should be within the data EDCAparameter’s range, e.g. they are changed into integers, CWmin[0]�CWmin[1] holds all the time,and otherwise, no change should be made.

3.3. Guard period and data traffic

In each sharing scheme, a guard period, shown in Figure 1, is needed for two reasons: (1) as aguard bandwidth reserved partially for collision and idle time to prevent bandwidth allocation from


400 Y. XIAO ET AL.

over provisioning and (2) as a period reserved partially for best-effort data traffic. The guard periodhas the function as a guard bandwidth reserved partially for collision and idle time to preventbandwidth allocation from overprovisioning, and the guard period is partially for best-effort datatraffic, referred as to data traffic thereafter. When the minimum number of active stations N0=20,the grad period � is around 22–24%, and when N0=5, and � is around 18–23%. In our laterextensive simulations, we choose �=20% under which the performance is very good. The abovechoices are based on [35–38].

4. DYNAMIC PARTITION SCHEMES

In a dynamic bandwidth partition scheme, bandwidth is dynamically partitioned among voiceand video based on current voice/video/data traffic load condition. We propose and study threedifferent dynamic bandwidth partition schemes: DP, DP-FT, and DP-RR as shown in Figure 1.Since data control and guard period of the proposed dynamic schemes are the same as the SPscheme, this section only focuses on how to implement the dynamic partition concept. Furthermore,due to the limited space, we cannot present the whole schemes in details, but we only presentmajor differences in each scheme, and many functions are similar to Section 3.1 with revisions tocertain parameters (e.g. TXOPBudget).

4.1. Dynamic partition with finer-tune

In the DP-FT scheme, shown in Figure 1(b), bandwidth of voice and video is partitioned propor-tionally to voice traffic load and video traffic load in the previous measurement interval, anda Finer-Tune (FT) method is adopted to handle some extreme cases to avoid starvations andover-provisioning for another real-time traffic. The FT scheme is defined as borrowing somebandwidth/budget from another AC if available before rejecting a flow.

An intuitive approach is that bandwidth of voice and video is partitioned proportionally to voicetraffic load and video traffic load in the previous measurement interval. Let us define the followingnotations for the previous measurement interval:

U [i] = TxTime[i]×SurplusFactor[i]A[i] = TXOPBudget[i]+U [i]

We should have∑

i A[i]=T . For the next measurement interval, we should define A[i]=(U [i]/∑i U [i])T for proportional (which is better than (TxTime[i]/∑i TxTime[i])T verified bysimulations, omitted) such that ∀i, A[i]−U [i]�0. However, the above scheme may cause starva-tions or over-provisioning for another real-time traffic under some extreme cases. For example,(1) in the previous measurement interval, if U [3]=0 (voice), U [1]=U [0]=0, U [2]>0 (video),it may cause starvations for voice traffic later; (2) based on our simulations, when one AC (e.g.voice)’s remaining budget is not large enough for acceptance of another voice flow and there is alarge portion of available video’s budget, the newly arrived voice flow may be rejected due to alarge difference of the required bandwidth for a video flow and the required bandwidth for a voiceflow—but a good approach should be that voice can borrow some bandwidth/budget from videoso that the newly arrived voice flow is accepted. The bandwidth borrowing idea in case (2) can



also avoid the problem of case (1). Therefore, we propose the FT method as a fix to the problems,where the FT scheme is defined as to try to borrow some bandwidth/budget from another AC(voice form video or video from voice) if available before rejecting a real-time flow.

4.2. Dynamic partition with reserved region

In the DP-RR scheme, shown in Figure 1(d), bandwidth of voice and video is partitioned propor-tionally to voice traffic load and video traffic load in the previous measurement interval, and twoReserved Regions (RRs) are reserved for voice and video, respectively. The difference of DP-FTand DP-RR is the difference of FT and RR, i.e. how to fix some extreme cases. For the RR method,each AC (voice/video) reserves at least a minimum budget good enough for one flow.

One drawback of the DP-RR scheme is that it may be possible that some bandwidth/budget iswasted for reserving for a non-existing flow. But this can be alleviated since the ‘wasted’ budgetcan be used by best-effort data. However, this is not very efficient in some cases.

One advantage of the DP-RR scheme is that if some applications need some guarantee for oneof or both voice and video flows (e.g. in a WLAN that is configured to guarantee providing atleast 10 voice flows no matter what traffic pattern it has), the DP-RR scheme is useful. In such asense, the PP-FT and DP schemes can also be revised to include reservation regions to handle theabove case, and we can call the PP-FT-RR scheme and the DP-RR scheme, respectively.

4.3. Dynamic partition

In the DP scheme, shown in Figure 1(c), instead of partitioning bandwidth, budget of voice andvideo is partitioned proportionally to voice traffic load and video traffic load. For the previousmeasurement interval, we need to calculate the total budget (B)

B=∑i

(A[i]−TxTime[i]×SurplusFactor[i])

For the next measurement interval, we have

TXOPBudget[3] = �2×B for voice

TXOPBudget[2] = �1×B for video

A[i] = TXOPBudget[i]+TxTime[i]×SurplusFactor[i]

Note that in this paper AC=1 is not used.

5. PERFORMANCE EVALUATION

In this section, we conduct performance evaluation for the proposed partition schemes via extensivesimulations. Our results show that the DP-RR scheme is not efficient. We adopted IEEE 802.11a[1] and IEEE 802.11e draft [2] in these simulations. We implemented a simulation program usingdiscrete event simulation with Java application. In addition, to simplify our study without loss ofgenerality, channel noise and multi-rate are not considered. Three traffic types are considered inour simulations: voice (AC 3), video (AC 2), and best-effort data (AC 0).


402 Y. XIAO ET AL.

5.1. Performance metrics and simulation setup

We adopt the following performance metrics in our simulations: (1) average throughput per voiceflow, video flow, or data station, (2) total throughput, (3) Txlimit, (4) TxBudget, (5) number ofaccepted and active flows (NAAF) in the system per AC, and (6) throughput square relative differ-ence (SRD). Throughput SRD is proposed to characterize the normalized difference of achievedthroughput and required throughput. Let K (t) and Ti (t) denote the number of flows in an AC(AC>0) and the average throughput, respectively, at the t th measurement interval. Let Ti denotethe required throughput for flow i (i=1,2, . . .,K (t)). Throughput SRD at the t th measurementinterval for this AC is defined as

SRDT (t)=k(t)∑i=1

(Ti (t)−Ti

Ti

)2

We assume that the transmitted traffic is no larger than required throughput on average; otherwise,a token bucket algorithm can be also implemented to control the traffic rate. Throughput SRD canbe only applied to voice and video traffic, but not data traffic.

The defaults EDCA access parameters used for our simulations are listed as follows:CWmin[3]=16; CWmax[3]=256; AIFS[3]=25�s; CWmin[2]=32; CWmax[2]=2048; AIFS[2]=25�s; CWmin[0]=256; CWmax[0]=51200; AIFS[0]=34�s; and queue size is 30 frames for eachAC (voice, video, data). For other parameters, the following values are adopted unless statedotherwise: beacon interval is 100ms; damping factor is 0.9; each voice flow is 0.0832Mbps,which is generated by a constant inter-arrival time 20ms with a fixed payload size of 208bytes, corresponding to G.711-coded VoIP over RTP/UDP/IP/SNAP [39]. Each video flow is4.68Mbps, which is generated by a constant inter-arrival time 2.5ms with a mean payload sizeof 1464 bytes. It corresponds to a traffic-shaped CBR video flow. Each station generates dataframes with an exponential distribution with a mean inter-arrival time 12ms and a fixed payloadsize of 1500 bytes. IEEE 802.11a is adopted and parameters are listed as follows: the data rateis 54Mbps; the control rate is 24Mbps; retry limit is 7; SIFS time is 16�s; Slot time is 9�s;physical layer’s preamble is 16�s, physical header time is 4�s; and a symbol time is 4�s. Weassume that all the stations are within the transmission range. Data control parameter �=1.3. Wealso have �2=�1=0.4, �=0.2.

5.2. SP vs DP

This subsection compares the SP scheme with the DP scheme under the following traffic pattern:at 5 s, 1 voice flow, 1 video flow, and 1 data are added; then 1 voice flow is added at each 5 suntil the number of voice flows is 10, 1 video is added at each 5 s till the number of video flowsis 5, and 1 data station is added at each 5 s until the number of data stations is 10; the simulationtime is 80 s.

Figure 2 shows the NAAF and the average throughput for the SP and DP schemes. The DPscheme outperforms the SP scheme by accepting two additional video flows (4 video flows forthe DP scheme vs 2 video flows for the SP scheme at 80 s) while maintaining a similar averagethroughputs for voice and video. Figure 2(c) and (d) shows that both schemes provide desired andstable average throughput for voice and video flows. This is due to the admission described at theend of Section 1. Moreover, the average throughput of best-effort data is degraded more in DPscheme in Figure 2(e). After two additional video flows are accepted, thanks to dynamic partition



0 20 40 60 800

2

4

6

8

10N

AA

FSP dataSP voiceSP video

0 20 40 60 800

2

4

6

8

10

NA

AF

DP voiceDP dataDP video

0 20 40 60 800

0.02

0.04

0.06

0.08

0.1

Time (s) Time (s) Time (s)

Time (s) Time (s)

Ave

rage

Thr

ough

put (

Mbp

s)

SP voiceDP voice

0 20 40 60 800

1

2

3

4

5

6A

vera

ge T

hrou

ghpu

t (M

bps) SP video

DP video

0 20 40 60 800

0.5

1

1.5

Ave

rage

Thr

ough

put (

Mbp

s)

SP dataDP data

(a) (b)

(c) (d) (e)

Figure 2. Comparisons of NAAF and average throughputs.

scheme, some bandwidth is shifted from best-effort data to the additional video flows due to theproposed data control mechanism.

Figure 3 shows that (1) the throughput SRDs are almost the same (near zero, indicating goodQoS); (2) the numbers of collisions are almost the same; (3) the total throughput of the DP schemeis better than that of the SP scheme. In other words, the DP scheme is better than the SP schemeby accepting more video flows, therefore achieving a better total throughput with the similar QoS.

Figure 4 shows ATL (i.e. A[i]) and TxBudget. For the SP scheme, ATLs of voice and videoare both 40ms fixed, where ATL standards for Allocation Time Limit. For the DP scheme, ATLsof voice and video are dynamically changed but the total ATLs is fixed 80ms since 20ms forbest-effort data and guard period. Budgets behave a little differently for ATLs.

For the SP scheme, Figure 5(a) shows that all 10 voice flows are accepted (also shown inFigure 2(a)), and Figure 5(b) shows that 2 video flows are accepted (also shown in Figure 2(a)). Forthe DP scheme, Figure 5(c) shows that all 10 voice flows are accepted (also shown in Figure 2(b)),and Figure 5(d) shows that 4 video flows are accepted (also shown in Figure 2(b)). Therefore, DPoutperforms SP with a higher acceptance ratio for video flows.

5.3. DP vs DP-FT

This subsection compares the DP scheme and the DP-FT scheme under the following traffic pattern:at 5 s, 1 voice flow, 1 video flow, and 1 data are added; then 1 voiceflow is added at each 5s till


404 Y. XIAO ET AL.

0 20 40 60 800

5

10

15

20

25

30

Tota

l Thr

ough

put (

Mbp

s)

DPSP

0 20 40 60 800

50

100

150

200

Num

ber

of C

ollis

ions

DPSP

0 20 40 60 800

0.01

0.02

0.03

0.04

0.05

Time (s) Time (s)

Time (s) Time (s)

Thr

ough

put S

RD

SP voiceDP voice

0 20 40 60 800

0.01

0.02

0.03

0.04

0.05

Thr

ough

put S

RD

SP videoDP video

(a) (b)

(c) (d)

Figure 3. Comparisons of other metrics.

the number of voice flows is 50, and 1 data station is added at each 5s till the number of datastations is 10; the simulation time is 250 s.

Figure 6 shows the NAAF and the average throughput for the DP and DP-FT schemes. TheDP-FT scheme outperforms the DP scheme by accepting 49 more additional voice flows (50 voiceflows for the DP-FT scheme at 250 s vs 1 voice flow for the DP scheme at 80 s) while maintaininga similar average throughputs for voice and video. In Figure 6(c) and (d), both DP and DP-FTschemes provide desired and stable average throughput for voice and video flows. This is due tothe admission described at the end of Section 1. Figure 6(e) shows that the average throughput ofbest-effort data is degraded more in DP-FT scheme. Since more voice and video flows are accepted,thanks to the fine tune method, some bandwidth is shifted from best-effort data to the additionalvoice and video flows. This is bandwidth shift is due to the proposed data control mechanism. Thereason is that the DP scheme still can cause starvation in this special traffic case due to the largedifference of the required budget for a video flow and the required budget for a voice flow. In otherwords, access time of one voice flow is very small compared with access time of one video flowso that the ATL for voice is very small after proportional partition in certain traffic conditions.

Figure 7 shows that (1) the throughput SRDs are almost the same (near zero, indicating goodQoS); (2) the number of collisions for the DP-FT is larger than that of the DP scheme since 49more voice flows (small frames) are accepted; (3) the total throughput of the DP-FT scheme is



0 20 40 60 800

10

20

30

40

50

60

70

80

(a) (b)

(c) (d)

ATL

(ms)

0 20 40 60 800

10

20

30

40

50

60

70

80

ATL

(ms)

0 20 40 60 800

10

20

30

40

50

60

70

80

TxB

udge

t (m

s)

0 20 40 60 800

10

20

30

40

50

60

70

80

TxB

udge

t (m

s)

Time (s) Time (s)

Time (s) Time (s)

DP videoSP video

DP videoSP video

DP videoSP video

DP videoSP video

Figure 4. ATL (i.e. A[i]) and Txbudget.

0 20 40 60 800

5

10

15

20

25

30

35

40

45

(a)

TxL

imit

(ms)

Time (s)

0 20 40 60 800

5

10

15

20

25

30

35

40

45

TxL

imit

(ms)

0 20 40 60 800

5

10

15

20

25

30

35

TxL

imit

(ms)

0 20 40 60 800

10

20

30

40

50

60

70

80

TxL

imit

(ms)

DP voice DP video SP voice SP video

(b) Time (s) (c) Time (s) (d) Time (s)

Figure 5. Txlimit.

a little better than that of the DP scheme. In other words, the DP-FT scheme is better than theDP scheme by accepting more voice flows, having a little better total throughput, and havingsimilar QoS.


406 Y. XIAO ET AL.

0 50 100 150 200 2500

2

4

6

8

10

(a) (b)

(c) (d) (e)

NA

AF DP data

DP voiceDP video

0 50 100 150 200 2500

10

20

30

40

NA

AF

DP-FT voiceDP-FT dataDP-FT video

0 50 100 150 200 2500

0.02

0.04

0.06

0.08

0.1

Ave

rage

Thr

ough

put (

Mbp

s)

DP voiceDP-FT voice

0 50 100 150 200 2500

1

2

3

4

5

6A

vera

ge T

hrou

ghpu

t (M

bps)

DP videoDP-FT video

0 50 100 150 200 2500

0.5

1

1.5

Ave

rage

Thr

ough

put (

Mbp

s)

Time (s) Time (s)

Time (s) Time (s) Time (s)

DP dataDP-FT data

Figure 6. Comparisons of NAAF and average throughputs.

Figure 8 shows ATL and TxBudget. For the DP scheme, ATLs of voice and video becomefixed values, but the total ATL is 80ms. For the DP-FT scheme, ATLs of voice and video aredynamically changed, but the total ATL is fixed to 80ms with 20ms for best-effort data and guardperiod. Budgets behave a little differently from ATLs.

For the DP scheme, Figure 9(a) shows that 1 voice flow is accepted (also shown in Figure 6(a)),and Figure 9(b) shows that 1 video flow is accepted (also shown in Figure 6(a)). For the DP-FTscheme, Figure 9(c) shows that all 50 voice flows are accepted (also shown in Figure 6(b)), andFigure 9(d) shows that 1 video flow is accepted (also shown in Figure 6(b)).

5.4. Comments

We conducted two simulation experiments, i.e. two traffic patterns, in this paper. In subsection 5.2,one traffic pattern is used to compare DP with SP. In subsection 5.3, another traffic pattern isdesigned to compare DP-FT with DP. In each simulation, the exactly same traffic pattern is appliedto two different schemes.

Owing to dynamic partition algorithm, the DP scheme admits more voice/video flows intoWLAN than SP scheme does. Indeed, compared with SP scheme, the DP scheme permits a betterexploitation of the bandwidth of WLAN, with the price of the sacrifice of best-effort data. Similarly,due to fine tune method, the DP-FT scheme admits more voice/video flows into WLAN than theDP scheme does. Compared with the DP scheme, the DP-FT scheme permits a better exploitation



DP-FTDP

DP-FTDP

DP voiceDP-FT voice

DP videoDP-FT video

0

5

10

15

20

25

30

Tota

l Thr

ough

put (

Mbp

s)

0

0.01

0.02

0.03

0.04

0.05

Time (s) Time (s)

Time (s) Time (s)

Thr

ough

put S

RD

0 50 100 150 200 2500 50 100 150 200 250

0 50 100 150 200 250 0 50 100 150 200 250

(a) (b)

(c)

0

5

10

15

20

25

30

Tota

l Thr

ough

put (

Mbp

s)

0

0.01

0.02

0.03

0.04

0.05

Thr

ough

put S

RD

(d)

Figure 7. Comparisons of other metrics.

of the bandwidth of WLAN, with the price of the sacrifice of best-effort data. This is exactly thedesign goal of the proposed DP scheme and the DP-FT scheme.

6. CONCLUSIONS

In this paper, we study a challenging issue, i.e. bandwidth allocation for contention-based MACwith QoS guarantee. We propose and study four different bandwidth partition schemes: the SP, DP,DP-FT, and DP-RR schemes. In the DP-FT scheme, bandwidth of voice and video is partitionedproportionally into voice traffic load and video traffic load in the previous measurement interval, andthe FT method is adopted to handle some extreme cases to avoid starvations and over-provisioningfor another real-time traffic, via borrowing some bandwidth/budget from another AC if availablebefore rejecting a flow. Our simulations show that the DP scheme is better than the SP scheme,and the DP-FT scheme is better than the DP scheme. The DP scheme still has some starvationsin some particularly designed special cases, and this can be avoided by either adopting the DP-FT scheme or the DP2-RR scheme (i.e. including reservation region in the DP scheme). Someschemes’ details, many results, and analysis are omitted due to the limited space.


408 Y. XIAO ET AL.

DP-FT voiceDP voice

DP videoDP-FT video

DP-FT voiceDP voice

DP videoDP-FT video

0 50 150100 200 250

0 50 150100 200 250 0 50 150100 200 250

0 50 150100 200 2500

10

20

30

40

50

60

70

80

(a) (b)

(c) (d)

ATL

(ms)

0

10

20

30

40

50

60

70

80

ATL

(ms)

0

10

20

30

40

50

60

70

80

TxB

udge

t (m

s)

0

10

20

30

40

50

60

70

80

TxB

udge

t (m

s)

Time (s) Time (s)

Time (s) Time (s)

Figure 8. ATL (i.e. A[i]) and Txbudget.

0 100 2000

5

10

15

20

25

30

(a) (b) (c) (d)

TxL

imit

(ms)

0 100 2000

10

20

30

40

50

60

70

80

TxL

imit

(ms)

0 100 2000

5

10

15

20

25

30

TxL

imit

(ms)

0 100 2000

10

20

30

40

50

60

70

80

TxL

imit

(ms)

DP voice DP voice DP-FT video DP-FT video

Time (s) Time (s) Time (s)Time (s)

Figure 9. Txlimit.

This paper is about one-hop MAC QoS, but does not consider end-to-end QoS (such as end-to-end delay and jitter). In future work, we may consider end-to-end QoS that will be a cross-layerQoS design among MAC QoS, routing layer QoS, and TCP layer QoS together.



ACKNOWLEDGEMENTS

This work was supported in part by the US National Science Foundation (NSF) under the grant numbersCNS-0737325, CNS-0716211, and CCF-0829827. Dr Bo Li’s research was supported in part by grantsfrom RGC under the contracts 615608, and 616207, by a grant from NSFC/RGC under the contractN HKUST603/07, by a grant from HKUST under the contract RPC06/07.EG27.

We thank the anonymous reviewers’ constructive comments’ which significantly improved the qualityof our paper.

REFERENCES

1. IEEE 802.11-1999. Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specification,Standard. IEEE: New York, 1999.

2. Choi S, Prado J, Mangold S, Shankar S. IEEE 802.11e contention-based channel access (EDCF) performanceevaluation. Proceedings of IEEE ICC’2003, Anchorage, AK, May 2003.

3. Xiao Y. A simple and effective priority scheme for IEEE 802.11. IEEE Communications Letters 2003; 7(2):70–72.4. Xiao Y. Performance analysis of priority schemes for IEEE 802.11 and IEEE 802.11e wireless LANs. IEEE

Transactions on Wireless Communications 2005; 4(4):1506–1515.5. Deng D-J, Chang R-S. A priority scheme for IEEE 802.11 DCF access method. IEICE Transactions on

Communications 1999; E82-B(1):96–102.6. IEEE 802.11e/D4.3. Draft Supplement to Part 11: Wireless Medium Access Control (MAC) and physical layer

(PHY) specifications: Medium Access Control (MAC) Enhancements for Quality of Service (QoS). IEEE Std.802.11e/D4.3, May 2003.

7. Pallot X, Miller LE. Implementing message priority policies over an 802.11 based mobile ad hoc network.Proceedings of IEEE MILCOM’2001, Washington, DC, 2001.

8. Aad I, Castelluccia C. Differentiation mechanisms for IEEE 802.11. Proceedings of IEEE INFOCOM’2001,Anchorage, AK, 2001.

9. Veres A, Campbell AT, Barry M, Sun L-H. Supporting differentiation in wireless packet networks using distributedcontrol. IEEE Journal of Selected Area in Communications (JSAC) 2001; 19(10):2081–2093.

10. Mangold S, Choi S, May P, Kein O, Hiertz G, Stibor L. IEEE 802.11e wireless LAN for quality of service.Proceedings of European Wireless’ 2002, Florence, Italy, 2002.

11. Romdhani L, Ni Q, Turletti T. Adaptive EDCF: enhanced service differentiation for IEEE 802.11 wireless adhoc networks. Proceedings of IEEE ICC’2003, Anchorage, AK, May 2003.

12. Choi S, Prado J, Mangold S, Shankar S. IEEE 802.11e contention-based channel access (EDCF) performanceevaluation. Proceedings of IEEE ICC’2003, Anchorage, AK, May 2003.

13. Bianchi G, Tinnirello I. Analysis of priority mechanisms based on differentiated inter frame spacing in CSMA-CA.Proceedings of IEEE VTC’2003-Fall, Orlando, U.S.A., October 2003.

14. Xiao Y, Li H, Choi S. Protection guarantee for voice and video traffic in IEEE 802.11e wireless LANs.Proceedings of IEEE INFOCOM 2004, 2004.

15. Xiao Y, Li H, Choi S. Two-level protection and guarantee for multimedia traffic in IEEE 802.11e distributedWLANs. ACM/Springer Wireless Networks 2009; 15(2):141–161.

16. Guo T, Cai J, Foh C, Zhang Y. Improving videophone transmission over multi-rate IEEE 802.11e networks.Proceedings of IEEE ICC 2008, 2008.

17. Xiao Y, Li H, Li B. Bandwidth sharing schemes for multimedia traffic in the IEEE 802.11e contention-basedWLANs. IEEE Transactions on Mobile Computing 2007; 6(7):815–831.

18. Boggia G, Camarda P, Grieco LA, Mascolo S. Feedback-based bandwidth allocation with call admission controlfor providing delay guarantees in IEEE 802.11e networks. Computer Communications 2005; 28(3):325–337.

19. Zhao M, Yang Y, Zhu H, Shao W, Li VOK. Priority-based opportunistic MAC protocol in IEEE 802.11 WLANs.International Journal of Sensor Networks 2008; 3(2):84–94.

20. Musaloiu R-E, Terzis A. Minimising the effect of WiFi interference in 802.15.4 wireless sensor networks.International Journal of Sensor Networks 2008; 2(1):43–54.

21. Lin X, Ling X, Zhu H, Ho P, Shen X. A novel localised authentication scheme in IEEE 802.11 based wirelessmesh networks. International Journal of Security and Networks 2008; 3(2):122–132.

22. Zhang X, Tan L, Li J, Zhao S, Chen H. ATBAS: an efficient fair bandwidth allocation approach for multihopwireless ad hoc network. International Journal of Sensor Networks 2008; 3(2):134–140.


410 Y. XIAO ET AL.

23. Krishnamurthy V, Sazonov E. Reservation-based protocol for monitoring applications using IEEE 802.15.4 sensornetworks. International Journal of Sensor Networks 2008; 4(3):155–171.

24. Haffiz Shuaib A, Aghvami AH. Dynamic topology control for the IEEE 802.15.4 network. International Journalof Sensor Networks 2009; 6(3/4).

25. Misic J, Shafi S, Misic VB. Real-time admission control in 802.15.4 sensor clusters. International Journal ofSensor Networks 2006; 1(1/2):34–40.

26. Misic J, Amini F, Khan M. Performance implications of periodic key exchanges and packet integrity overheadin an 802.15.4 beacon enabled cluster. International Journal of Sensor Networks 2008; 2(1):33–42.

27. Wang K-C, Tang L, Huang Y. Transmission error analysis and avoidance for IEEE 802.15.4 wireless sensors onrotating structures. International Journal of Sensor Networks 2009; 6(3/4).

28. Ho C, Lea C. Improving call admission policies in wireless networks. Wireless Networks 1999; 5:257–265.29. Ramjee R, Towsley D, Nagarajan R. On optimal call admission control in cellular networks. Wireless Networks

1997; 3:29–41.30. Haung Y-R, Lin Y-B, Ho JM. Performance analysis for voice/data integration on a finite mobile systems. IEEE

Transactions on Vehicular Technology 2000; 49(2):367–378.31. Xiao Y, Chen CLP, Wang Y. Optimal admission control for multi-class of wireless adaptive multimedia

services. IEICE Transactions on Communications, Special Issue on Mobile Multimedia Communications 2001;E84-B(4):795–804.

32. Xiao Y, Chen CLP, Wang B. Bandwidth degradation QoS provisioning for adaptive multimedia in wireless/mobilenetworks. Computer Communications 2002; 25(13):1153–1161.

33. Xiao Y, Li H, Chen CLP, Wang B, Pan Y. Proportional degradation services in wireless/mobile adaptivemultimedia networks. Wireless Communications and Mobile Computing (WCMC) 2005; 5(2):219–243.

34. Li B, Li L, Cao X-R. On handoff performance for an integrated voice/data cellular system. Wireless Networks2003; 9:393–402.

35. Tinnirello I, Bianchi G, Xiao Y. Refinements on IEEE 802.11 distributed coordination function modelingapproaches. IEEE Transactions on Vehicular Technology, DOI: 10.1109/TVT.2009.2029118.

36. Xiao Y. Backoff strategies for demand re-registration in PCS database failure recovery. Computer Communications2004; 27(5):400–411.

37. Tanenbaum AS. Computer Networks (4th edn). Prentice-Hall: Englewood Cliffs, 2002.38. Bianchi G. Performance analysis of the IEEE 802.11 distributed coordination function. IEEE Journal on Selected

Areas in Communications 2000; 18(3):535–547.39. Collins D. Carrier Grade Voice over IP (2nd edn). McGraw-Hill: New York, 2002.

AUTHORS’ BIOGRAPHIES

Yang Xiao worked in industry as a MAC (Medium Access Control) architect involvingthe IEEE 802.11 standard enhancement work before he joined the Department ofComputer Science at The University of Memphis in 2002. Dr Xiao is currently with theDepartment of Computer Science (with tenure) at The University of Alabama. He was avoting member of IEEE 802.11 Working Group from 2001 to 2004. He is an IEEE SeniorMember. He is a member of American Telemedicine Association. He currently serves asEditor-in-Chief for International Journal of Security and Networks (IJSN), InternationalJournal of Sensor Networks (IJSNet), and International Journal of Telemedicine andApplications (IJTA). He serves as a panelist for the U.S. National Science Foundation(NSF), Canada Foundation for Innovation (CFI)’s Telecommunications expert committee,and the American Institute of Biological Sciences (AIBS), as well as a referee/reviewerfor many national and international funding agencies. He serves on TPC for more than

100 conferences, such as INFOCOM, ICDCS, MOBIHOC, ICC, GLOBECOM, WCNC, etc. He serves asan associate editor for several journals, e.g. IEEE Transactions on Vehicular Technology. His research areasinclude security, telemedicine, robot, sensor networks, and wireless networks. He has published more than 300papers in major journals, refereed conference proceedings, book chapters related to these research areas. DrXiao’s research has been supported by the U.S. National Science Foundation (NSF), U.S. Army Research,Fleet and Industrial Supply Center San Diego (FISCSD), and The University of Alabama’s Research GrantsCommittee. Dr Xiao is a Guest Professor of Jilin University (2007–2012), and was an Adjunct Professor ofZhejiang University (2007–2009).



Frank Haizhon Li is an assistant professor of Computer Science at the University ofSouth Carolina Upstate. He received his PhD degree from the University of Memphis.Prior to pursuing his career in Computer Science, he has worked as a chemical engi-neer for four years. His research interests include modeling and analysis of mobile,wireless, ad hoc and sensor networks, QoS and MAC enhancement for IEEE 802.11wireless LANs, adaptive network services, multimedia services and protocols overwireless networks, and Computer Security. Frank has served as a technical programcommittee member for many conferences such as IEEE ICC and IEEE WCNC. Email:[email protected]

Ming Li has been a faculty in the Department of Computer Science, California StateUniversity, Fresno, since August 2006. He received his MS and PhD degrees in ComputerScience from The University of Texas at Dallas in 2001 and 2006, respectively. Hisresearch interests include QoS strategies for wireless networks, robotics communications,and multimedia streaming over wireless networks. He is the recipient of the Best StudentPaper Award in the First IEEE International Workshop on Next Generation WirelessNetworks (WoNGeN’05). He has served as the TPC co-chair of The First InternationalWorkshop on Pervasive Computing Systems and Infrastructures (PCSI 2009), The SecondInternational Workshop on Sensor Networks (SN’09), Multimedia Networking trackin ICCCN’08, the IEEE International Work-shop on Data Semantics for MultimediaSystems and Applica-tions (DSMSA’08), and the Third IEEE International Workshopon Next Generation Wireless Networks (WoNGeN’08). He is a guest editor of a special

issue on Recent Advances in Sensor Integration for International Journal of Sensor Networks (IJSNet), a specialissue onData Semantics forMultimedia Systems in SpringerMultimedia Tools and Applications, and a special issuein Journal of Multimedia. He has served the technical program committees of several international conferencessuch as ICC, ICCCN, ISM, ROBOCOMM, VTC, and ChinaCom. Ming Li is a member of ACM and IEEE.

Jingyuan Zhang received his PhD degree in Computer Science from Old DominionUniversity in 1992. He is currently an associate professor with the Department ofComputer Science at the University of Alabama. Prior to joining the University ofAlabama, he had been a principal computer scientist with ECI Systems and Engineering,an assistant professor with Elizabeth City State University, and an instructor with NingboUniversity. Dr Zhang’s current research interests include wireless networks, mobilecomputing, and collaborative software.

Bo Li is a professor in the Department of Computer Science and Engineering, HongKong University of Science and Technology, where he has been affiliated with since1996. He was with IBM Networking System, Research Triangle Park, U.S.A., between1993 and 1996. He was an adjunct researcher at Microsoft Research Asia (MSRA)(1999–2005), in which he spent his sabbatical leave (2003–2004). with MicrosoftAdvanced Technology Center (ATC) in the summers of 2007 and 2008. His workshave resulted in over 200 publications. He has made original contributions on Internetproxy placement, capacity provisioning in wireless networks, routing in WDM opticalnetworks, Internet video streaming. He is best known for the series of works on asystem called Coolstreaming (Google entries over 1 000 000 in 2008), which attractedmillions of download and was credited as the first large-scale Peer-to-Peer live video


412 Y. XIAO ET AL.

streaming system in the world. He received the Young Investigator Award from Natural Science Foundationof China (NFSC) in 2004. He has been an editor or guest editor for 17 IEEE/ACM journals and involvedin organizing 50 conferences. He was the Co-TPC Chair for IEEE Infocom 2004. He was a DistinguishedLecturer in IEEE Communications Society (2006–2007). Bo Li received his B Eng Degree in the ComputerScience from Tsinghua University, Beijing, and his PhD degree in the Electrical and Computer Engineeringfrom the University of Massachusetts at Amherst.


Dynamic bandwidth partition schemes for integrated voice, video, and data traffic in the IEEE...

Documents

Transcript of Dynamic bandwidth partition schemes for integrated voice, video, and data traffic in the IEEE...