Computer Communications 39 (2014) 41–53

Contents lists available at ScienceDirect

Computer Communications

journal homepage: www.elsevier .com/locate /comcom

Towards voice/video application support in 802.11e WLANs:A model-based admission control algorithm

0140-3664/$ - see front matter � 2013 Elsevier B.V. All rights reserved.http://dx.doi.org/10.1016/j.comcom.2013.10.006

⇑ Corresponding author.E-mail addresses: nchendeb@ul.edu.lb (N. Chendeb Taher), ghamri@ensiie.fr

(Y. Ghamri Doudane), bachar_elhassan@ul.edu.lb (B. El Hassan), nazim.agoulmine@iup.univ-evry.fr (N. Agoulmine).

Nada Chendeb Taher a,⇑, Yacine Ghamri Doudane b, Bachar El Hassan a, Nazim Agoulmine c

a Lebanese University, Azm Center for Research and Biotechnologies, Mitein street, Tripoli, Lebanonb ENSIIE, 1 square de la Résistance, 91025 Evry, CEDEX, Francec University of Evry val d’Essonne, Bd. François Mitterrand, 91025 Evry, CEDEX, France

a r t i c l e i n f o a b s t r a c t

Article history:Available online 23 October 2013

Keywords:Admission controlQoS differentiationEDCAVoice/video services802.11e WLAN

Supporting emergent voice/video applications in all wireless technologies is a requirement in the NextGeneration Network (NGN) where Wireless Local Area Networks (WLANs) is a main component. For thistype of applications, QoS needs to be fully maintained in order to assure user satisfaction. Actually, QoScontrol in 802.11e WLANs to support real time voice/video services remains an open problem. All thesolutions that only aim to enhance the performance of the Enhanced Distributed Channel Access (EDCA)mechanism cannot resolve the performance degradation problem once the channel becomes saturated.Hence, an efficient admission control scheme in EDCA is the key to guarantee the QoS required byvoice/video services in WLANs. In this paper, we propose a model-based admission control algorithm thatis located within the QoS Access Point (QAP). An accurate analytical model is used to predict the QoSmetrics that can be achieved once a new flow is introduced in the WLAN. Based on this prediction andon the QoS constraints of already admitted (active) flows as well as of the new flow, the QAP takes theappropriate decision for the new flow. The proposed admission control scheme is fully compatible withthe legacy 802.11e EDCA MAC protocol. It is validated numerically and through simulations using severalrealistic usage scenarios.

� 2013 Elsevier B.V. All rights reserved.

1. Introduction

In order to support high throughput applications (video), a newgeneration of WLANs is going to be standardized in the upcomingyears. This shows the importance of supporting voice/video applica-tions by this kind wireless technology as it is the case of other wire-less mobile networks (NGN). Standards offering high throughput area requirement for the future of internet services. These standardscoupled to QoS aware mechanisms can constitute a solution to themajor problems in today’s standards. In 802.11e access mechanismEDCA, there is no guaranty in terms of throughput and delay assur-ance for voice/video services. Before the network gets saturated,there is no QoS problem. The problem arises once the network startsto reach saturation and a high number of flows share the limitedchannel resources. At each new flow arrival, the existing flows loosea certain degree of their already achieved performance. This is due tothe fact that channel resources have to be distributed among differ-ent active flows according to their priority. This means that voice/vi-deo services which are unable to adapt their flows to this resource

limitation cannot be supported correctly. Therefore, a resource man-agement solution, controlling the number of active flows in a givenWLAN is required. Hence, there is a compelling need for an efficientadmission control mechanism capable of maintaining the QoS re-quired by voice/video applications.

Many proposals for an admission control scheme exist in theliterature [3–17]. We mainly distinguish those based on measure-ments and those based on analytical models. The former are limitedby their inability to guaranty the required performance to flows interms of throughput and delay assurance. Indeed, measurement-based approaches are based on the current channel utilizationconditions measurements, which make them unable to map thesemeasurements to the required performance metrics. These mecha-nisms have also another important limitation related to the fact thatthey cannot measure the new channel conditions resulting from theintroduction of the new flow without really introducing this newflow. On the other side, the model-based admission control schemesare mainly limited by the analytical model upon which they are builtto make their decision. This later is either developed under the satu-ration conditions assumption and therefore it is not suitable to beused in an efficient admission control procedure as this one is sup-posed to be used to avoid such saturation, or it suffers from some lim-itations when it was developed and therefore has a low degree ofaccuracy (missing of one or more of the three EDCA differentiation

42 N. Chendeb Taher et al. / Computer Communications 39 (2014) 41–53

parameters, missing or partial modeling of some important featuresof the EDCA procedure, etc.).

In this article, we propose an admission control mechanism whichis based on our new analytical model developed in [1,2]. This later wasdeveloped with two main objectives: (1) to provide a sufficient degreeof accuracy and precision and (2) to have a low computational over-head in order to be suitable for the usage in an admission control algo-rithm. The analytical model we developed is applicable to all trafficconditions going from non-saturation to complete saturation. It canpredict the achievable throughput and the mean access delay per Ac-cess Category (AC) for any configuration of the three EDCA differenti-ation parameters: Arbitration Inter Frame Space (AIFS), ContentionWindow (CW) sizes, and Transmission Opportunity Limit (TXOPLim-it). Our analytical model outperforms already existing models in theliterature as it was demonstrated extensively in [1,2]. Note also thatparts of this article had also been published in [3]. Differently from[3], this article contains the complete study and procedure going fromthe design of the accurate analytical model and its validation results tothe design, discussion and validation of the admission control mech-anism. The validation part of the proposed admission controlalgorithm also contains more extensive results for wide range of real-istic usage scenarios.

After this introduction, in Section 2, we discuss the differentapproaches used in the conception of admission control schemesin EDCA, their advantages and their limitations. Some existingadmission control mechanisms in the literature are also discussedin this section. Our analytical model, on which the admission con-trol algorithm is based, is presented in Section 3 with the focus onthe main contributions that distinguish our model from the others.Section 4 contains the validation of the analytical model. In Section5, we present our approach for an admission control scheme basedon the analytical model. A set of realistic scenarios to validate theadmission control solution and to show its reaction in the pres-ence/absence of non-real time applications with the real timeapplications, are realized and the results are discussed in Section6. Conclusions and perspectives close this paper.

2. Related work and motivations

Existing admission control schemes for EDCA can be classifiedinto three categories: measurement-based, model-based and hy-brid. In the following, we discuss advantages and limitations ofeach of these three approaches while studying in details somemodel-based admission control proposals.

2.1. Measurement-based admission control

In this category, we cite the work of [4–8]. Using such mecha-nism, the QAP measures at each measurement interval the channelconditions. Based on the values obtained, it makes its decision toaccept or reject the upcoming flow. Such mechanisms have the fol-lowing two advantages:

� Complex numerical computation is avoided. Simple computa-tion of additional loads that could be generated by the activa-tion of the new flow is done. These additional loads are thencompared to the measured residual capacity of the channel.� Measurements give more precision than numerical estimations.

However, this approach suffers from multiple limitations whichcan be summarized by:

� The QAP cannot consider the real QoS requirements in terms ofthroughput and delay as a decision criteria. In fact, the measure-ments cannot be mapped to QoS metrics and there are nomeans to be sure that the QoS requirements will be guaranteed.

� The measurements can only give the channel utilization condi-tions. They cannot give the achievable values of throughput andaccess delay.� It is very hard to choose the measurement interval value. This

value must be sufficiently high to reflect the steady state func-tioning regime, and at the same time, it must be sufficiently lowto reflect any change in the channel conditions. This compro-mise is difficult to reach and there is no precise solutionexplaining what the best measurement interval value is.� It is often necessary to use signaling between the QAP and the

QSTAs (QoS STAtions) to share measured information.

2.2. Analytical-model-based admission control

Using this approach, the QAP is based on numerical computa-tion to predict the performance metrics before making any deci-sion. For these mechanisms, we note:

� From one side, a high response time when the analytical modelis computationally complex.� From the other side, an accuracy problem if the model is not

accurate, it is based on severe approximations or it does notreflect correctly the protocol behavior.

This is despite the following advantages:

� The decision and the processing are made only within the QAP.This later has no need to gather information from the QSTAs.Hence, there is no need for signaling that increases the controlload in a limited resources network.� The decision is based on the real QoS requirements of flows. In

fact, the QAP uses the predicted performance metrics if the newflow is admitted. Based on these predictions and on the QoSrequirements of the new flow and the already active flows,the QAP decide to reject or admit the new flow.� The QoS metrics are predicted analytically without the need to

introduce the new flow.

In the following, we discuss more in detail different solutionsthat had been proposed in this category. First, the solution of[9–11] use an analytical model limited by the saturation condi-tions. These admission control mechanisms are not efficientbecause they are based on the achievable values of QoS metricsin the saturation conditions. These are the asymptotic values thatare far from the optimal achievable values in the network as dem-onstrated in [1,2].

For the works that that are not limited by the saturationassumption, we have [12–14]. In [12], the decision criteria is theaccess delay. The objective of the proposed admission control isto maintain the network far from the saturation state. The authorsof [12] affirm that the used analytical model overestimates theaccess delay and therefore the decisions are made with a certainsecurity margin. We agree that this margin permits to be sure thatthe network is still far from the saturation state and therefore per-forms better. But, this overestimation – if large – may cause therejection of new flows even if the network can serve them.

Similarly, in [13], the delay constitutes the decision criteria. Infact, the authors suppose that an admission control based on theachievable access delay is better than one which is based on theachievable throughput. This is because the need of bandwidthcomes after the need of low delays. This affirmation does not holdin all situations, we think that it depends on different factors:mainly traffic types, EDCA parameters configuration and QoS con-straints of active flows. We argue that it is not sufficient to use onlythe access delay as decision criteria. It is necessary that the

N. Chendeb Taher et al. / Computer Communications 39 (2014) 41–53 43

admission control mechanism is capable of considering the twoperformance metrics at the same time: the access delay and theachievable throughput. Hence, it can be applicable to all flowtypes; those that are delay sensitive (voice) and those that requirea large bandwidth (video).

As in [12], the objective of [13] is to avoid reaching the saturationstate. For that, the admission control mechanism is divided in twophases: in the first phase, the QAP makes a preliminary decision,and then in the next phase, it monitors the network to ensure thatthe admittance of the new flow does not lead to a saturation state.This temporary admittance of a flow, then its possible rejection inthe second phase if its activation violates the required QoS con-straints, is also considered as a drawback of the proposed scheme.In fact, there is no mean in 802.11e EDCA to stop or kill a flow afterit was associated and admitted in the WLAN. Therefore, the authorskeep this problem open. Additionally, in the proposed scheme, thedecision is based on the estimation of the transmission probability.

Finally, Bellata in its work [14], propose an admission controlwith the main objective to adapt the EDCA parameters. This adap-tation procedure and the proposed admission control are based onthe analytical model developed by the same author [15]. This ana-lytical model contains some approximations in order to be simpleenough to achieve the adaptation task in a satisfactory responsetime. These simplifications led to a loss of accuracy.

2.3. Measurement and model-based admission control

In this category, the QAP achieves at each measurement intervalthe measurements of some defined variables. It uses the valuesmeasured in the analytical model to predict the performance met-rics and make the decisions. These mechanisms couple the advan-tage of the two above mechanisms. They however keep somelimitations. One of their major limitations is that they are basedon the actual measured values of the channel conditions (busychannel probability and collision probabilities). These values areupdated using Exponential Weighted Averaging (EWA) and Expo-nential Weighted Moving Averaging (EWMA) techniques to predictthe performance metrics that may result from the activation of thenew flow. This may of course lead to non-completely accurate pre-dictions. These mechanisms cannot have the real measurementsvalues without really activating the new flow.

In this category, we can cite [16–18]. The first two are based onan analytical model developed in saturation conditions and there-fore they suffer from the inherent limitation as explained earlier.The third one, despite the use of a non-saturation model, suffersfrom limitations related essentially to the accuracy of the usedmodel. Actually, the model used is an extension of a model devel-oped for DCF1 combined with a flow homogenization technique. Theauthors chosen to sacrifice the accuracy of the model to minimizethe numerical computation overhead as it is stated in the followingsentence extracted form [18]: ‘‘. . .we opt for sacrificing some accuracyby developing an approximation model based on the non-saturationanalysis of the DCF. . .’’. However, the effect of this approximationon the accuracy of the model was not discussed.

Other solutions which are not completely compatible with theEDCA standard such as [19] and others exist in the literature. Thisdemonstrates how this field of research is in continuous progressand researchers are trying to solve the problem of QoS assuranceby different ways. The authors in [20] have introduced an admis-sion control scheme which is based on analytical modeling andgame theory. Here the new users play an important role in the pro-cedure with the AP and the performance parameters are the end-to-end delay and the loss probability.

1 Distributed Coordination Function, the mandatory access mechanism for 802.11standard.

2.4. Summary

To sum up, we say that to map the real QoS requirements ofvoice/video applications to decision criterion, a measurement-based admission control cannot be efficient. A hybrid admissioncontrol based simultaneously on measurements and analyticalmodel can be used, however, we have to sacrifice a certain levelof prediction precision caused by the weight of old measurementsvalues on the decisions. Finally, a model-based admission control isgood if the analytical model is characterized by a good accuracyand by a low numerical computation overhead.

3. The EDCA analytical model

Before presenting our proposed admission control procedure, itis first necessary to describe the analytical model on which it isbased. As will be shown throughout this section, our objective isto reach a high accuracy level and a low computational overhead.

3.1. IEEE 802.11e EDCA MAC access mechanism

EDCA defines four access categories (ACs). Three parameters areused in EDCA to implement AC specific traffic prioritization, i.e.Arbitration Inter Frame Space (AIFS), Contention Window (CW)sizes, and Transmission Opportunity Limit (TXOPLimit). In EDCA,each AC within a wireless station behaves like a virtual station; itmust sense the medium before initiating a transmission. If the med-ium is sensed as being idle for a time interval equal to the AC’s spe-cific AIFS (i.e. AIFS [AC]), then the AC chooses randomly a value a in[0,CW] and waits for an additional time interval equal to a timeslots. This is called the backoff process. At any time slot during thebackoff process, if the AC senses the channel busy, it freezes its back-off count down and waits for the channel to be idle again for a com-plete AIFS before re-starting the count-down. Once the backoffcounter reaches zero and the channel is still idle, the AC transmitsinto the channel. Once getting the right to transmit into the physicalchannel, an AC is granted this right for time duration equal to TXOP-Limit. So, it continues its transmission till either its TXOPLimit ex-pires or its queue becomes empty. Acknowledgements are used tonotify the sending station about the successful reception of frames.If no acknowledgement is received, the sending station considersthat there is a collision or channel error. In this case, it waits foran AIFS and schedules a retransmission. Other stations, which arenot involved in the collision, wait for a specific Extended Inter FrameSpace (EIFS [AC]) which is greater than the AIFS [AC]. This is calledthe post-collision, it is necessary to give the sending station the pri-ority to retransmit as soon as possible. In addition to physical colli-sions (i.e. collisions of frames originating from different stations),virtual collisions may occur among different ACs within the samestation. In this case the higher priority AC is the one that is grantedthe opportunity for physical transmission.

After each unsuccessful transmission, the CW is doubled untilthe CWmax [AC] is reached. The starting value of CW is CWmin[AC]. In case the maximum number of allowed retransmissions,called Retry Limit, is reached without success, the failing frame isdropped. After each successful transmission, the transmitting ACinitiates another backoff, even if there is no other pending packetto be delivered. This is often referred to as post-backoff, as thisbackoff is done after, not before a transmission. After achievingthe post backoff, if the station has no data to be transmitted, itstays in an idle state waiting for the arrival of new packets.

3.2. Motivations for a new model

Several analytical models had been proposed for 802.11e EDCAin the few last years. A comparative study and the detailed

44 N. Chendeb Taher et al. / Computer Communications 39 (2014) 41–53

discussion of the value added and limitations of these models canbe found in [1,2]. This comparison is based on the differentiationparameters considered by the proposed models (CW, AIFS andTXOPLimit), the baseline features of EDCA (retry limit, backofffreeze, internal collision) and the calculated performance metrics(throughput and access delay). This study allowed drawing twomain conclusions. First, even though almost all EDCA models takeinto account AIFS and CW differentiation, they unfortunately ne-glect the TXOPLimit differentiation parameter despite its notice-able effect on the global performance as shown in [21]. This canbe considered as a clear gap in these models. Second, almost allEDCA models do not consider accurately the modified backoffcountdown operation in the 802.11e EDCA. This one should coop-erate with the AIFS countdown operation. After each frozen coun-ter in the backoff, the channel must be sensed idle during thewhole AIFS period before the backoff countdown procedure startsagain. This cooperation process is neglected in almost all the ana-

PB;0;0;0 ¼ð1�peÞpc ð1�pm

c Þ1�pc

1þ ½Tc� þ 1pb

1ð1�pbÞ½Tpc � � 1� �� �

þ 12ð1�pbÞ

½N�pb þ 1ð1�pbÞ½A�

� � Xm

j¼1

pjcwjð1� peÞ þw0

!

þ1þ ½W�pe þ 1�ð1�pbÞ½A�

pbð1�pbÞ½A�þ ½Ts� 1� pmþ1

c ð1� peÞ � pcpe

� �

26664

37775�1

ð1Þ

lytical models we have studied in the literature. Actually, all thesemodels are only two-dimensional Markov chains: one dimension isfor the backoff stage and the second one is for the backoff counter[1,2]. This leads to the simplification of the AIFS and backoff count-down cooperation. Therefore, in addition to the integration of theTXOPLimit parameter, we propose in our model a three-dimen-sional Markov chain. This dimension is the minimum required toaccurately model the AIFS and backoff co-operation process as wellas the different EDCA procedures (AIFS countdown, backoff freeze,virtual collision between ACs of the same station, external collisionwith other stations, post collision, post backoff and retry limit). Tobe used in an admission control algorithm, our model should alsohandle general traffic conditions and not be limited to saturationconditions.

3.3. The analytical model

We present here quickly our analytical model: the associatedMarkov chain as well as the main formulas. These are the basisof our admission control algorithm. More details about the deriva-tion are in [1,2].

To be able to integrate all EDCA features in our model, ourapproach consists of following the states in which an AC may beduring its transmission cycle. This approach is the key that makesour model complete and therefore differentiates it from existingmodels. In our model, each state represents an AC in a time slot.The states that an AC can occupy at a randomly chosen time slotare grouped into seven ‘‘periods’’ representing EDCA operations:Idle period, AIFS period, backoff period, frozen period, collision period,post collision period and transmission period.

The proposed Markov chain is a discrete four dimensional onedrawn for each AC. The first dimension p(t) indicates the periodin which the AC is at time t. The second dimension s(t) representsthe backoff stage and the third one b(t) denotes the backoff countervalue. Finally, the fourth dimension r(t) indicates the remainingtime to leave the current period. So, (p(t),s(t),b(t), r(t)) is a discretetime Markov chain based on the assumption that the transitionprobabilities are constant.

At each time slot, the state of each AC is determined by (i, j, k,d);i = I stands for idle, A for AIFS, F for frozen, B for backoff, C for

collision, PC for Post Collision and T for Transmission.j = 0,1,2, . . . ,m; m being the retry limit; k is uniformly chosen from[0,wj] where wj depends on the backoff stage and satisfies wj+1 = 2-wj + 1 when wj < wmax (w0 = wmin), and wj + 1 = wj when wj = wmax.Finally, d, the time remaining to leave the period, it depends onthe value of i (i.e. on the period itself). The entire Markov chainis shown in Fig. 1. In this figure the rectangular blocs contain aset of states with their transition (frozen and AIFS blocs), the con-tent of these blocs is presented in Fig. 2.The transition probabilitiesbetween states are based on pc (collision probability), pb (channelbusy probability) and pe (probability of empty queue). To resolvethe Markov chain system, we consider Pi,j,k,d the steady state prob-ability of (i, j,k,d), then we calculate the probabilities of all stateswith respect to one probability which is here PB,0,0,0. Finally, toget PB,0,0,0, we resolve the normalization condition:

Pi;j;k;dPi;j;k;d ¼ 1.

Eq. (1) gives a relation between PB,0,0,0, pc, pb, pe and othersystem parameters. [Tc], [Tpc], [N], [A], [W] and [Ts] are respec-tively the collision time, the post collision time, the frozen timethe AIFS time, the waiting time and the transmission time. Theseparameters are namely EDCA and network related parameters.Therefore to resolve the system, we have to compute mainlypc, pb and pe.

1. Computation of pci and pb

In the following we will append the index i to the collision prob-ability pci and to the empty probability pei to denote the AC number(1 6 i 6 4). These two probabilities are AC dependent ones but pb isnot. Let us see now how a collision may occur, and when the chan-nel is busy. We differentiate between virtual and external colli-sions. We assume a fixed number of active stations M in thesame radio range. In each station, all ACs are active. Let si be theprobability for an ACi attempt to access the channel in a randomtime slot. According to our model:

si ¼ PI;0;0;1 þXm

j¼0

PB;j;0;0

¼ ð1þ peiÞ þpci 1� pm

ci

� �ð1� peiÞ

1� pci

� �PB;0;0;0 ð2Þ

From the viewpoint of a station, the probability s that it acces-ses the channel is the probability that at least one of its ACs tries toaccess the channel:

s ¼ 1�Y3

i¼0

ð1� siÞ ð3Þ

The internal collision probability for ACi is the probability that atleast one of the higher priority ACs tries to access the channelsimultaneously with it, so Pcinti

¼ 1�Q

j>ið1� sjÞ, the external colli-sion probability is the probability that at least one of the other(M � 1) stations tries to access the channel at the same time slot,Pcexti

¼ 1� ð1� sÞM�1. So, the total collision probability is given by:

Fig. 1. Markov chain of the model.

N. Chendeb Taher et al. / Computer Communications 39 (2014) 41–53 45

Pci ¼ 1� ð1� sÞM�1Yj>i

ð1� sjÞ ð4Þ

For the busy channel probability, let vi be the probability for thechannel to be occupied by the ACi, this is the probability for it to bein transmission or external collision state:

v i ¼ ½Tsi� 1�pmþ1ci ð1�peiÞ�pcipei

� �þ½Tci�ð1�peiÞ

pci�pmþ1ci

1�pcipcexti

� �PB;0;0;0 ð5Þ

The probability for the channel to be occupied by a station is theprobability for it to be occupied by one and only one AC from thisstation, so:

v ¼X3

i¼0

v i

Yj–i

ð1� v jÞ ð6Þ

Finally, the channel is considered as idle if none of the stationsis using it. Thus:

Pb ¼ 1� ð1� vÞM ð7Þ

1. Computation of pei and [Wi]

Fig. 2. Frozen and AIFS blocs (j,k); 0 6 j 6m, 1 6 k 6wj.

To calculate pei and [Wi], we assume a Poisson arrival process ofrate k (packet/s). [Wi] being the mean waiting time during the Idleperiod. Even though the Poisson assumption may not be realistic, itprovides insightful results and allows the model to be tractable.

For this consideration, qi ¼ kiDi is the probability that the queuecontains data to be handled, so the probability that the queue isempty is:

pei ¼ 1� qi ¼ 1� kiDi ð8Þ

where Di is the mean service time, which is here the mean accessdelay of ACi. For the average idle time, it is obvious that an AC tran-sits to idle state after the completion of the last transmission andthe post backoff, while the queue remains empty. So the averagewaiting time depends on these times as well as on the packet in-ter-arrival time using the following equation:

½Wi� ¼ 1=ki � Di � ½Ts� � TPB ð9Þ

1=ki is the packet inter-arrival time, and TPB is the time needed toachieve the post-backoff procedure.

3.4. Throughput analysis

The average achievable throughput for each AC can be obtainedby:

Si ¼ peiSinsat þ ð1� peiÞSisat ð10Þ

where Sinsat is the non-saturation throughput and Sisat is the satura-tion throughput.

Sinsat ¼MpeiE½P�qi=1� qi

TPBi þ ½Wi� þ pciDi þ ð1� pciÞð11Þ

Sisat ¼psiE½P�NTXOPi

ð1� pbÞ þ pb

P3j¼0psjTsi þ pb 1�

P3j¼0psj

� �Tc

ð12Þ

In this last equation:psi ¼ MP½Ts �

d¼1pT;0;0;dð1� vÞM�1Qj>ið1� v jÞ

NTXOPi ¼ TXOPiTs1þSIFS

h i; Tsi ¼ NTXOPiðTs1 þ SIFSÞ; E[P] is the average packet

payload size, Ts1 is the transmission time for one frame, Tc is thecollision time.

The effect of the TXOP parameter appears in our model in thecomputation of the transmission time Tsi. In fact, during a TXOP-Limit, a station may be allowed to transmit multiple data framesfrom the same AC with a Short Inter Frame Space (SIFS) betweenan ACK and the subsequent data frame as illustrated in Fig. 3. Azero value for TXOPLimit indicates that a single frame may betransmitted. Tsi computation takes this into consideration.

3.5. Access delay analysis

We define Di;j;k;d the delay needed to go from the state (i, j,k,d) tothe successful transmission state. The fourth dimension of ourMarkov chain allows us to compute recursively the delays for allstates. The final equation that gives the access delay in saturationand non-saturation conditions is therefore equal to:

Fig. 3. Transmission during TXOPLimit.

All EDCA parameters ac�vated

0

500

1000

1500

2000

2500

3000

20 40 50 100 200 400 500 800 1000 1500 1800

Arrival rate (Kb/s)

Thro

ughp

ut (k

b/s)

AC0-Mod AC1-Mod AC2-Mod AC3-ModAC0-Sim AC1-Sim AC2-Sim AC3-Sim

Fig. 4. Achievable throughput per AC – all parameters activated.

TXOPLimit = 0 for all ACs, CFB deac�vated

0

500

1000

1500

2000

2500

3000

20 40 50 100 200 400 500 800 1000 1500 1800Arrival rate (Kb/s)

Thro

ughp

ut (K

b/s)

AC0-Mod AC1-Mod AC2-Mod AC3-ModAC0-Sim AC1-Sim AC2-Sim AC3-Sim

Fig. 6. Achievable throughput per AC – TXOPLimit deactivated.

TXOPLimit = 0 for all ACs, CFB deac�vated

0,00

0,02

0,04

0,06

0,08

0,10

0,12

0,14

0,16

20 40 50 100 200 400 500 800 1000 1500 1800

Mea

n ac

cess

del

ay (s

)

All EDCA parameters ac�vated

0,00

0,02

0,04

0,06

0,08

0,10

0,12

0,14

0,16

20 40 50 100 200 400 500 800 1000 1500 1800

Arrival rate (Kb/s)

Mea

n ac

cess

del

ay (s

)

AC0-Mod AC1-Mod AC2-Mod AC3-ModAC0-Sim AC1-Sim AC2-Sim AC3-Sim

Fig. 5. Access delay per AC – all parameters activated.

46 N. Chendeb Taher et al. / Computer Communications 39 (2014) 41–53

Di ¼ peipci þ 1� peið ÞDA;0;0;½A� þ peið1� pciÞ ð13Þ

where DA,0,0,[A] is the access delay of the first state of the Markovchain.

3.6. Model resolution

The analytical model proposed is a non-linear equation system((1), (4), (7) and (8)). It can be resolved by means of numericalmethods. Once resolved, transition probabilities will be known.These probabilities are then used in the throughput and delay com-putation using two simple equations ((10) and (13)). Eqs. (10) and(13) represent the closed-form expressions developed to simplifythe computational costs.

So, the computational complexity of our model resides only inthe non-linear system resolution. This contains 5 nested ‘‘forloops’’, one for pb and four for pc (i.e. one for each AC). In orderto reduce the computational complexity, we suggested searchingthe probability value that leads to the smallest estimation errorin a given interval, and then we re-calculate around this value tillwe get the solution. This method greatly reduces the calculationtime and gets the exact solution with a very small estimation error(�10�6) and in a short period of time.

4. Model validation

To validate our model, we implemented it in Matlab. To decideabout the accuracy of our model, we compare the numerical resultsto simulation results. The simulation is realized using the ns-2 [22]enhanced with the TKN’s EDCA implementation [23]. The selectedphysical protocol for validations is 802.11b.

The simulation topology consists of four different wirelessnodes, one AP and one wired station. All wireless nodes send theirdata to the wired node via the AP and are all situated in the sameradio range. Poisson distributed traffic, consisting of 800-bytespackets, was generated at equal rate to each of the four ACs. Thearrival rates go from 20 Kb/s (non-saturated) to 1800 Kb/s (fullysaturated) for each AC at each node. Our goal is to prove the accu-racy of our model concerning the three parameters modeling: AIFS,CW and TXOPLimit. Also, we aim to show the effect of each of themon the overall performance. To do this, we run four simulations.The parameters for these simulations are summarized in Table 1.

In simulation 1, EDCA default parameters are used. Figs. 4 and 5show the results obtained. We can clearly see in Fig. 4 how AC1 (vi-deo) gains more throughputs in the saturation zone in favor of AC2

(BE). We can also see in Fig. 5 how the delays of AC0 (voice) and AC1

(video) remain relatively low as compared to AC2 and AC3 satisfy-ing their applications constraints.

In simulation 2, the TXOPLimit differentiation parameter isdeactivated. Figs. 6 and 7 show the corresponding results. Com-pared to simulation 1, this case shows the effect of TXOPLimit onthe global performance. We can see in Fig. 6 how the saturatedthroughput of AC1 cannot reach the same value obtained in Fig. 4and how AC2 can gain more throughputs as compared to the de-fault values case.

In simulation 3, the AIFS differentiation parameter is deacti-vated. Figs. 8 and 9 show the results. In this case, AIFS is set to 2

Table 1EDCA parameters for different simulation scenarios.

AIFSN CWmin CWmax TXOPLimit

(AC0, AC1, AC2, AC3)

Sim. 1 (2,2,3,7) (7,15,31,31) (15,31,1023,1023) (3264,6016,0,0)Sim. 2 (2,2,3,7) (7,15,31,31) (15,31,1023,1023) (0,0,0,0)Sim. 3 (2,2,2,2) (7,15,31,31) (15,31,1023,1023) (3264,6016,0,0)Sim. 4 (2,2,3,7) (7,7,7,7) (15,15,15,15) (3264,6016,0,0)

Arrival rate (Kb/s)AC0-Mod AC1-Mod AC2-Mod AC3-ModAC0-Sim AC1-Sim AC2-Sim AC3-Sim

Fig. 7. Access delay per AC – TXOPLimit deactivated.

for all ACs. Therefore, AC2 and AC3 will have the same EDCA param-eter values. It is obvious in this case that we get the same QoS met-rics for these two ACs. And this is what is obtained as noticed inFig. 8 and 9.

AIFS = 2 for all ACs, AIFS differenc�a�on deac�vated

0

500

1000

1500

2000

2500

3000

20 40 50 100 200 400 500 800 1000 1500 1800Arrival rate (Kb/s)

Thro

ughp

ut (K

b/s)

AC0-Mod AC1-Mod AC2-Mod AC3-ModAC0-Sim AC1-Sim AC2-Sim AC3-Sim

Fig. 8. Achievable throughput per AC – AIFS deactivated.

AIFS = 2 for all ACs, AIFS differenc�a�on deac�vated

0,00

0,02

0,04

0,06

0,08

0,10

0,12

0,14

0,16

20 40 50 100 200 400 500 800 1000 1500 1800

Arrival rate (Kb/s)

Mea

n ac

cess

del

ay (s

)

AC0-Mod AC1-Mod AC2-Mod AC3-ModAC0-Sim AC1-Sim AC2-Sim AC3-Sim

Fig. 9. Access delay per AC – AIFS deactivated.

Cwmin = 7, Cwmax = 15 for all ACs, CW differen�a�on deac�vated

0,00

0,02

0,04

0,06

0,08

0,10

0,12

0,14

0,16

20 40 50 100 200 400 500 800 1000 1500 1800Arrival rate (Kb/s)

Mea

n ac

cess

del

ay (s

)

AC0-Mod AC1-Mod AC2-Mod AC3-ModAC0-Sim AC1-Sim AC2-Sim AC3-Sim

Fig. 11. Access delay per AC – CW deactivated.

N. Chendeb Taher et al. / Computer Communications 39 (2014) 41–53 47

In simulation 4, the CW differentiation parameter is deacti-vated. Figs. 10 and 11 show the results obtained. In this case,CWmin is set to 7 and CWmax is set to 15 for all ACs. So, AC1 be-comes more prioritized than AC0. In fact, with this configuration,AC1 is privileged by the TXOPLimit parameter. For this reason, itgets more throughput than AC0, and approximately the same ac-cess delay as AC0.

Finally, the nearly-exact match between the analytical and thesimulation curves of the figures above proves that our model givesa very good estimation of the performance metrics for any of theconfigurations of EDCA parameters. If we go further into detailsin these curves and if we distinguish between the three regions:non saturation, saturation and transition regions, we can extractthe following interpretations:

a. In the non saturation region (<200 Kb/s), the achievablethroughput for each AC is equal to its arrival rate. The modelis extremely accurate.

Cwmin = 7, Cwmax = 15 for all ACs, CW differen�a�on deac�vated

0

500

1000

1500

2000

2500

3000

20 40 50 100 200 400 500 800 1000 1500 1800Arrival rate (Kb/s)

Thro

ughp

ut (K

b/s)

AC0-Mod AC1-Mod AC2-Mod AC3-ModAC0-Sim AC1-Sim AC2-Sim AC3-Sim

Fig. 10. Achievable throughput per AC – CW deactivated.

b. In the saturation region (>800 Kb/s) the QoS metrics reachthe asymptotic values and remain constant. The total achiev-able throughput is distributed among the ACs according totheir priorities. The model provides also a very good accu-racy in this region.

c. Within the transition region (between 200 Kb/s and 800 Kb/s),we notice an estimation error between analytical and simula-tion results. This is due to two effects. First, it is in this regionthat the effect of the queue model assumption is introduced.This effect is already analyzed in [21], and we proved that thiserror does not affect the accuracy. Second, our approach ofusing an interpolated value based on pe can have an impacton this error. This error being low, does not affect the globalaccuracy.

Some other common and general conclusions about the behav-ior of the EDCA can also be drawn:

1. In the saturation region, the throughput and the delay remainconstant. Hence, for a fixed number of active stations, thepacket arrival rate does not affect the achievable throughputand delay.

2. The saturated throughput is not the maximum achievablethroughput. Some ACs get more throughputs before reachingthe saturation region.

3. Finally, the performance metrics for each access category arehighly influenced by the choice of the EDCA parameters. A cor-rect parameter’s setting is a very important task when all ACsco-exists in the network.

5. The proposed admission control mechanism

As shown above, noticeable effort was done to develop an accu-rate analytical model that predicts on the best the accurate valuesof the achievable performance metrics. Likewise, we tried to sim-plify and decrease on the best the numerical computation over-head and therefore the response time. This was obtained by theuse of a resolution algorithm faster than the classic one. All of theseefforts are done with the objective to get round of the two possiblelimitations of model-based admission control mechanisms.

Contrarily to other model-based admission control that useonly the access delay, or only the achievable throughput in thedecision making. In our solution, we will use these two metricsas decision criterion. In fact, although real time services are delaysensitive, and have a strict requirement in terms of access delay,the use of the two performance metrics as decision criterion meansthat the proposed admission control scheme can be applied at thesame time to video applications which have high constraints interms of throughput assurance, and to voice applications whichare sensitive to the two.

48 N. Chendeb Taher et al. / Computer Communications 39 (2014) 41–53

Concerning the saturation, the non-saturation and the transi-tion regions, the analytical model is applicable to all functioningregions. This is to the QAP to decide, based on the predicted valuesif the network is permitted to attain the saturation state or it mustremain in the non-saturation region. This depends, of course on theQoS requirements of the active flows as well as of the new flow.This is also related to the network capacity which in turn dependson the EDCA parameters configuration of different ACs.

5.1. The procedure at each QSTA

The 802.11e standard defined a simple admission control proce-dure between the QAP and the QSTAs while keeping the specifica-tion and the implementation of the admission control algorithm tomanufacturers.

According to the procedure defined by the standard, a QSTAsends an access request ADDTS Request (ADD Traffic Stream re-quest) containing the priority and the QoS requirements in theTSPEC (Traffic Specification) field. TSPEC define the nominal framesize (MSDU size), the average arrival rate (Rmean), the minimaland maximal arrival rate (Rmin and Rpeak) and the tolerated delaybound (Delay bound). The QAP studies this request using theavailable admission control algorithm and sends an ADDTS Re-sponse (ADD Traffic Stream Response) containing the decision ofadmission or rejection [24]. This procedure is schematized inFig. 12.

So, each QSTA, sends an ADDTS request for each new flow belong-ing to a specified AC to the QAP while specifying its QoS require-ments (in our case, we are interested to the mean achievablethroughput and the mean access delay). If the request is accepted,the flow becomes active and starts the channel contention proce-dure. Once the transmission is achieved, a DELETS (Delete TrafficStream) message is sent by the QSTA to the QAP aiming to inform thislater that it can delete this flow from the set of active flows.

5.2. The procedure at the QAP

At the reception of the ADDTS Request, the QAP extracts thespecifications of the new flow from the TSPEC field (AC, framesize, mean arrival rate and tolerated delay). Using the analyticalmodel, it predicts the new system probabilities while consideringthis new flow active. Using these probabilities, the QAP computethe achievable throughput and the access delay for the alreadyadmitted flows as well as for this new flow. If the QoS require-ments can be met, the new flow is admitted and an ADDTS Re-sponse with the response ‘Accept’ is sent to the requestingstation. This new flow and its QoS requirements are added tothe set of active flows. In the other case, an ADDTS Response withthe response ‘Reject’ is sent to the station.

Two sets of flows have to be managed by the QAP: ‘Wait-ing_Flows’ and ‘Admitted_Flows’. ‘Waiting_Flows’ contains theflows waiting for a response from the QAP. These are the flows that

Accept or Reject

CAC Algorithm

ADDTS Response

ADDTS Request (TSPEC)

QSTA QAP

Fig. 12. Admission control procedure in 802.11e EDCA.

have already sent the ADDTS Request but not treated yet by theQAP and therefore did not received the ADDTS Response. ‘Admit-ted_Flows’ contains the list of already admitted flows that did notsend the DELETS message yet. It is very necessary that the QAP keepinformation concerning the already admitted flows and their speci-fications. This is required to check if the admittance of any new flowmay violate the QoS constraints of the previously admitted flows.Once the DELETS message is received by the QAP, it deletes the cor-responding flow from the set of ‘Admitted_Flows’.

The pseudo code of the proposed admission control algorithmwithin the QAP is presented in Algorithm 3. According to thisalgorithm, the QAP extracts the AC of the new flow, updates thenumber of active stations, the number of active access categoriesin each QSTA and the total arrival rate of each AC while takinginto consideration the new flow. These variables are used to re-solve the non-linear system equations of the analytical modelpresented in III. After the resolution, the transition probabilitiesof the Markov chain are known, they are used to compute theachievable throughput and access delay per AC by the use ofEqs. (10) and (13) respectively. These calculated values are thencompared to the required values of all the flows belonging tothe corresponding AC. For the throughput constraint, the sum ofthe required throughputs for all the flows belonging to a givenAC must be less than the achievable throughput for this AC to ad-mit the new flow. For the access delay constraint, the calculatedaccess delay for a given AC must be less than the maximal toler-ated access delay of flows belonging to this AC. It is obvious thatthese conditions must be verified for the four ACs to admit thenew flow.

Algorithm 3: Admission control algorithm within the QAP

for each ADDTS_Request from Waiting_Flows doFi = New_Flow (ADDTS_Request)ACi = Get Access Category Fi)N = Nb_wireless_stations (Admitted_Flows & Fi)MN = Nb_ACs_per_Station (Admitted_Flows & Fi)TAR = Total_Arrival_Rate_per_AC (Admitted_Flows & Fi)Resolve system equations (N, MN, TAR)for each Access Category AC in (VO, VI, BE, BK) do

Calculate Achievable_Throughput (AC)Calculate Access_Delay (AC)if (AC – ACi) then

if (Calculated_Throughput(AC) < sum(Requested_Throughput(Admitted_Flows(AC)) or

Calculated_Delay(AC) > max(Requested_Delay(Admitted_Flows(AC))) then

Reject (Fi)Send ADDTS_Response(reject); go to ⁄⁄⁄

end ifelse/⁄ This AC is the requested flow’s AC, ACi

⁄/if (Calculated_Throughput(AC) – sum

(Requested_Throughput(Admitted_Flows(AC)) <Requested_Throughput(Fi) or

Calculated_Delay(AC) > max (Requested_Delay(Admitted_Flows(AC) & Fi)) then

Reject (Fi)Send ADDTS_Response(reject); go to ⁄⁄⁄

end ifend ifend for/⁄ end testing all ACs ⁄/Admit(Fi)Send ADDTS_Response(accept)Admitted_Flows = Admitted_Flows & Fi⁄⁄⁄/⁄ go to the next waiting flow ⁄/

end for

N. Chendeb Taher et al. / Computer Communications 39 (2014) 41–53 49

6. Admission control validation

Fig. 14. Effect of CAC on access delay – voice only.

To validate our admission control algorithm, we will considerthree types of applications: voice (AC_VO), video (AC_VI) and data(AC_BE). In each QSTA, we run one and only one instance of theseapplications. So, in this context, we have video stations, voice sta-tions and data stations. Hence, the addition of a new flow in theWLAN is equivalent to the addition of a new station (flow is hereequivalent to station). This choice was done in order to simplifythe flow identification and to make the explanation of the realizedevaluation scenario more comprehensive. It is however importantto note that this choice does not affect the global validity of theproposed mechanism. Note also that this choice reflects the caseof a realistic WLAN home network scenario in which we may findvideo receptors (TV), dedicated machines for VoIP and computersfor web navigation and data transfer.

For the simulation scenarios, the topology used is constituted ofa variable number of QSTAs contending for channel access, oneQAP and one wired station connected to the QAP. All the QSTAssend and receive their data to the wired station via the accesspoint. These QSTAs and the AP are all situated in the same radiorange. The physical layer parameters used are defined by the802.11b standard while the MAC parameters are the default EDCAparameters defined by the 802.11e standard [24].

All flows belonging to the same AC have the same specificationsand the same QoS constraints. The mean arrival rate and the aver-age payload size of the three types of applications are summarizedin Table 2. The QoS constraints of these applications have beenchosen with respect to ITU-T recommendations [25], [26] for thevoice and video traffic. For best effort data applications, we assume

Table 2Specifications of the selected applications in the admission control validation.

Application, AC Mean arrival rate Mean payload size

Data, BE 100 Kb/s 500 octetsVideo, VI 600 Kb/s 7568 octetsVoice, VO 64 Kb/s 600 octets

Table 3QoS requirements of the selected applications in the admission control validation.

Application, AC Required throughput Maximal tolerated delay

Data, BE 0 Kb/s InfinityVideo, VI 600 Kb/s 400 msVoice, VO 64 Kb/s 150 ms

Fig. 13. Effect of CAC on achievable throughput – voice only.

Fig. 15. Effect of CAC on achievable throughput – video only.

Fig. 16. Effect of CAC on access delay – video only.

that there are no QoS constraints in terms of bandwidth and delay.The set of QoS requirements for these applications are given inTable 3.

The admission control algorithm is implemented in Matlab. Thisalgorithm provides the number of admitted flows until the firstrejection with the reason of rejection. The simulation is conductedusing network simulator version 2 (ns-2). It provides a comparisonbetween the performance metrics achieved without and withadmission control.

In our tests, we aim to validate our admission control mecha-nism in different situations. We are seeking to highlight the impactof the co-habitation of different priorities on the decisions of theadmission control algorithm. Hence, we examine the results of

50 N. Chendeb Taher et al. / Computer Communications 39 (2014) 41–53

the admission control mechanism where we have only voice sta-tions in the WLAN, video only, voice and data, video and data, voiceand video, and finally voice, video and data stations. In the follow-ing, we present the results for each of these cases.

Fig. 19. Effect of CAC on. throughput – video then data.

6.1. Voice only

In this case, we consider only voice stations. Each 10s, a newflow requests admission. Figs. 13 and 14 illustrate the results forthis case. 38 flows are admitted and the 39th is rejected. We cannote that the admission control does not permit the WLAN to reachthe saturation region. When analyzing more deeply the reason ofthe 39th flow rejection, we found out that it was due to a violationof the access delay constraint. In fact, it is clearly seen in the curvespresented in Figs. 13 and 14 that without admission control, theaccess delay increases dramatically to reach very high values com-pared to the tolerated delay and the achievable throughput de-creases considerably. The admission control algorithm detectsthis and limits the number of active flows to respect the QoS con-straints of voice flows which are sensitive to delay increase.

Fig. 20. Effet or CAC on access delay – video then data.

6.2. Video only

In this scenario, a new video flow asks for admission each 10s.10 flows are admitted and the 11th one is rejected because ofbandwidth constraint violation. Ns-2 simulation results are shownin Figs. 15 and 16.

From these two figures we can conclude the following: Thebandwidth constraint is more significant than the delay constraintin the case of video flows. The admission control permitted theWLAN to reach the saturation region. It is very clear in Figs. 15and 16 that the throughput achieved at the last admitted flow is

Fig. 17. Effect of CAC on. throughput – voice then data.

Fig. 18. Effet or CAC on access delay – voice then data.

Fig. 21. Effect of CAC on. throughput – data then voice.

Fig. 22. Effet or CAC on access delay – data the voice.

Fig. 23. Effect of CAC on throughput – data then video.

Fig. 24. Effet of CAC on access delay – data then video.

Fig. 25. Effect of CAC on. throughput –voice then video.

Fig. 26. Effet or CAC on access delay – voice then video.

Fig. 27. Effect of CAC on. throughput –video then voice.

Fig. 28. Effet or CAC on access delay – video then voice.

N. Chendeb Taher et al. / Computer Communications 39 (2014) 41–53 51

the maximal achievable throughput that occurs exactly at the sat-uration limit region. On the other hand, from our simulations wenoticed that the video delay did not increased considerablyalthough in the saturation region (cf. Fig. 16).

6.3. Voice and data; video and data

For these scenarios, we fix the number of voice (respectively, vi-deo) stations to an admissible one based on the results we get fromthe first two scenarios. i.e. 20 voice stations (respectively, 3 videostations). Then after 100s of simulation time, we add a new dataflow each 10s. What we noticed in this case is that the admissioncontrol algorithm continued to admit data flows without limit.The ns-2 simulations results presented in Figs. 17 and 18 (respec-tively, Figs. 19 and 20) show that the admission of data flows,

whatever is their number, does not degrade the performance of al-ready active voice/video flows in the WLAN.

Also in the reversed scenarios (data then voice/video), the num-ber of admitted voice/video flows remains similar to the ones ob-tained in the case when they were the only active flows in theWLAN (see Figs. 21–24). These results can be explained by twofacts. First, we do not have QoS constraints on data flows in termsof throughput and delay. So, whatever will be the achievablethroughput and the access delays of these flows, the admissioncontrol algorithm will not reject them. Second, the default EDCAparameter configuration prohibited the data flows from degradingthe performance achieved by voice/video flows. For these later, theachievable throughput remains all the time stable and the accessdelay always low. The best effort data flows take the residualcapacity of the saturation region with very high access delays.

The vertical lines that sometimes appears in the delay curveswithout admission control schematize the high increase of the

Fig. 29. Effect of CAC on. throughput – data then voice and video.

Fig. 30. Effet or CAC on access delay – data then voice and video.

52 N. Chendeb Taher et al. / Computer Communications 39 (2014) 41–53

access delay toward a very high values compared to the rangetraced in these curves. In general they represent the delay valuesgreater than 1 s. We chosen to trace only significant values of theaccess delay for the voice and the video (vertical axis), because ifwe trace all the values we obtained, the low values will not be vis-ible and thus it will not be possible to clearly depict the effect ofthe admission control algorithm on maintaining a low access delayfor voice and video.

6.4. Voice and video

In the scenario that corresponds to Figs. 25 and 26, we have 15voice stations during 100s, then each 10s a new video station re-quests the access. In this case, 7 video stations have been admittedand the 8th one is rejected because of the violation of bandwidthconstraint. This decision leads to protect active voice and videoflows as shown in Figs. 25 and 26.

Figs. 27 and 28 illustrate the results of the reversed scenario, inwhich we have three active video stations before the first voice sta-tion requests the access. In this case, 27 voice stations were admit-ted. Here also, our admission control algorithm proved to be usefulby stopping flow admission before entering the saturation region.

6.5. Voice, video and data

In the last scenario, we fix the number of data stations to 5.Then we add a video station followed by a voice station alterna-tively. 10 s always separate two arrivals. Six video stations wereadmitted and after that the voice stations continue to be admitteduntil their number reaches 16 stations. We can notice that theadmission of the 16th voice station generated small transitory per-formance perturbations for voice and video flows as shown in Figs.29 and 30. This denotes the vicinity of the saturation conditionwhich naturally led to the rejection of the 17th voice station.

6.6. Results discussion and interpretation

From the set of results obtained, we can draw the followinggeneral conclusions:

� The admission control algorithm we proposed succeeded tomaintain and guaranty the QoS requirements for voice andvideo applications. It succeeded to avoid the network fromfunctioning in a severe saturation state in which high perfor-mance degradation take place for all application types espe-cially voice/video applications. Hence, the main objective ofour work was achieved.� Voice applications are very sensitive to delay. The admission of

any new voice station near the saturation region increases thedelay considerably. The admission control predicts this delayincrease and rejects the request. We can conclude that, torespect the QoS constraints of voice applications, we have inter-est to keep the WLAN functioning slightly before the saturationregion.� Video applications have different behavior. The delays remain

at an acceptable level for the video and we can reach the begin-ning of the saturation region without violating the QoS con-straints of this type of applications. The bandwidth constraintviolation is the one limiting the admission of video flows.� The presence of data stations in the WLAN does not have an

impact on the admission of voice and video applications. Thisis assured by two factors: first, the default EDCA parameter con-figuration that protects high priority flows and second theabsence of QoS constraints imposed on data applications.

7. Conclusion

In this article, we proposed an admission control mechanismbased on an accurate analytical model we also proposed. After giv-ing an overview of our analytical model and after validating it, wedescribed the admission control procedure in each QSTA and with-in the QAP. We presented and explained the algorithm to be imple-mented in the QAP and which is responsible for the decisionmaking at the arrival of a new request. Then, we validated, numer-ically as well as by simulation, the proposed solution using a set ofrealistic scenarios. In the validation part, we were interested tostudy the behavior of the admission control in different situations:with presence/absence of different application types (voice, videoand best effort data services). Our objective was to check if theproposed admission control mechanism is able to maintain andguaranty the QoS requirements for voice/video applications. Fromthe obtained results, we can affirm that the proposed admissioncontrol mechanism is capable of maintaining and guarantyingthe QoS for voice/video services. The use of the default EDCAparameters as suggested by the 802.11e standard helps to limitthe impact of best effort data flows on the admission of voice/videoservices in the WLAN. The enhancement of the proposed algorithmby an EDCA parameters optimization module constitutes one of theperspectives of this work.

References

[1] N. Chendeb Taher, Y. Ghamri Doudane, B. El-Hassan, A complete and accurateanalytical model for 802.11e EDCA under saturation conditions, in: AICCSA,Rabat, Morroco, 2009.

[2] N. Chendeb Taher, Y. Ghamri Doudane, B. El-Hassan, N. Agoulmine, An accurateanalytical model for 802.11e EDCA under different traffic conditions withcontention free bursting, J. Comput. Syst. Networks Commun. (2011).

[3] N. Chendeb Taher, Y. Ghamri-Doudane, B. ELHassan, N. Agoulmine, An efficientmodel-based admission control algorithm to support voice and video servicesin 802.11e WLANs, in: GIIS 2009, Tunisia, 2009.

[4] S. Garg, M. Kappes, Admission control for VoIP traffic in IEEE 802.11 networks,in: Globecom, San Francisco, USA, 2003.

N. Chendeb Taher et al. / Computer Communications 39 (2014) 41–53 53

[5] D. Gu, J. Zhang, A new measurement-based admission control method forIEEE802.11 wireless local area networks, in: PIMRC, Beijing, China, 2003.

[6] L. Zhang, S. Zeadally, HARMONICA: enhanced QoS support with admissioncontrol for IEEE 802.11 contention-based access, in: RTAS, Toronto, Canada,2004.

[7] H. Yoon, J. Kim, D. Shin, Dynamic admission control in IEEE 802.11e EDCA-based wireless home network, in: CCNC, Las vegas, Nevada, USA, 2006.

[8] R. Garroppo, L. Tavanti, S. Lucetti, S. Giordano, Extension and application of thenetwork utilization characteristic metric to IEEE 802.11e, in: ICC, Glascow,Scotland, 2007.

[9] Y.-L. Kuo, C.-H. Lu, E. Wu, G.-H. Chen, An admission control strategy fordifferentiated services in IEEE 802.11, in: Globecom, San Francisco, USA, 2003.

[10] J. Zhu, A.O. Fapojuwo, A new call admission control method for providingdesired throughput and delay performance in IEEE802.11e wireless LANs, IEEETrans. Wireless Commun. 6 (2) (2007) 701–709.

[11] S. Oh, J. Shin, D. Kwak, C. Kim, A novel call admission control scheme for theIEEE 802.11e EDCA, in: ICACT, Hyderabad, India, 2008.

[12] X. Chen, H. Zhai, X. Tian, Y. Fang, Supporting QoS in IEEE 802.11e wirelessLANs, IEEE Trans. Wireless Commun. 5 (8) (August 2006) 2217–2227.

[13] A. Bai, T. Skeie, P. Engelstad, A model-based admission control for 802.11eEDCA using delay predictions, in: IPCCC, New Orleans, Louisiana, USA, 2007.

[14] B. Bellata, Flow-level QoS guarantees in IEEE 802.11e-EDCA based WLANs,Spain, 2006.

[15] B. Bellalta, C. Cano, M. Oliver, M. Meo, ‘‘Modeling the IEEE 802.11e EDCA forMAC Parameter Optimization, in: Hets-Nets, Bradford, UK, 2006.

[16] D. Pong, T. Moors, Call admission control for IEEE 802.11 contention accessmechanism, in: Globecom, San Francisco, USA, 2003.

[17] M. El Masri, G. Juanole, S. Abdellatif, Hybrid admission control algorithm forIEEE 802.11e EDCA: analysis, in: ICN, Cancun, Mexico, 2008.

[18] B. Bensaou, Z.-N. Kong, D. Tsang, A Measurement-assisted, model-basedadmission control algorithm for IEEE 802.11e, in: I-SPAN, Sydney, 2008.

[19] A. Nafaa, A. Ksentini, On Sustained QoS guarantees in operated IEEE 802.11wireless LANs, IEEE Trans. Parallel Distrib. Syst. 19 (8) (2008) 1020–1033.

[20] J. Hu, G. Min, W. Jia, M. Woodward, Admission control in the IEEE 802.11eWLANs based on analytical modelling and game theory, in: GLOBECOM 2009,Honolulu, HL, 2009.

[21] N. Chendeb, Y. Ghamri-Doudane, B. El-Hassan, Effect of transmissionopportunity limit on transmission time modeling in 802.11e, in: IPOM, SanJosé, California, 2007.

[22] The Network Simulation – ns-2, [Online]. Available: <http://www.isi.edu/nsnam/ns/>.

[23] [Online]. Available: <http://www.tkn.tu-berlin.de/research/802.11e_ns2/>.[24] Wireless medium access control (MAC) and physical layer (PHY)

specifications: medium access control (MAC) quality of service (QoS)enhancements, IEEE P802.11e/D13.0, 2005.

[25] 06 2006. [Online]. Available: <http://www.itu.int/rec/T-REC-H.323-200606-I/en>.

[26] 11 2007. [Online]. Available: <http://www.itu.int/rec/T-REC-G.113/en>.