- Email: [email protected]
Performance Evaluation of the IEEE 802.11 WLAN Supporting Quality of Service
Performance Evaluation of the IEEE 802.11 WLAN Supporting Quality of Service Adel BEDOUI*, Kamel BARKAOUI**, Karim DJOUANI*** * Laboratoire SYSCOM, ENIT, Tunis, Tunisie (Tel:+216 95540040; e-mail: [email protected]) **Laboratoire CEDRIC, CNAM, Paris, France (e-mail:[email protected]) *** Laboratoire LISSSI, Univ-Paris12, Paris, France F’SATIE at TUT, Pretoria, Republic of South Africa (e-mail: [email protected])} Abstract: This paper presents a performance evaluation of a new solution for the support of QoS by IEEE 802.11 WLAN. It aims at the improvement of the Medium Access Control (MAC) by taking into account information from both physical and network layers for packets differentiation scheduling. An analytical model based on stochastic Petri nets representing the dynamics of the new MAC is evaluated. Then, OPNET simulations are carried out based on the new MAC in order to evaluate its performance. Comparison between analytic and simulation results show the interest of our Cross-Layer approach which allows improving or at least keeping the same network performances while supporting Quality of Service. Keywords: MAC, Quality of Service, Stochastic Petri nets, modeling, performance evaluation, simulation, wireless LAN.
1. INTRODUCTION IEEE 802.11 specifies the Medium Access Control (MAC) and the physical layer (PHY) for wireless connectivity within a local area [IEEE 802.11, 1999]. The goal is to allow interoperability between WLANs systems of the various manufacturers and meet a need increasingly important for wireless communications. Two types of wireless local area network architectures exists [Mühlethaler, 2002]: ad hoc architecture and architecture with access point AP (or infrastructure). The MAC sublayer provides the control of the medium access, deliver data in a reliable way and protect data. It defines two access methods: the Distributed Coordination Function (DCF), which is mandatory, and the Poling Coordination Function (PCF), which is optional. However PCF function, supposed to support real time services, is not implemented in the majority of the commercial 802.11 products. Moreover, the co-operation between PCF and DCF modes leads to performance degradation [Visser et al., 1995]. Besides, DCF can be used in ad hoc network as in infrastructure network, whereas PCF is used only in infrastructure network. So, we choose a distributed control for the medium access ensured by DCF. The fundamental access method is a DCF. It is known as carrier sense multiple accesses with collision avoidance (CSMA/CA). It is based on the listen-before-talk scheme. The 802.11 standard defines a random Backoff time following a busy medium condition in order to reduce the
probability of collisions. In addition, an immediate positive acknowledgment is used to confirm reception. Two access types are proposed by DCF: basic access and RTS/CTS (Request To Send / Clear To Send). Compared to the former, the latter sends RTS/CTS frames of small sizes before sending the data frames, thus making it possible to minimize the collisions. Moreover, RTS/CTS mechanism is used to improve the access control and solve the problem of hidden stations. RTS frame is sent before each data frames transmission. In answer, the receiving station sends a CTS frame to confirm being ready to receive. Thus, all stations update their network allocation vector (NAV) according to RTS/CTS. A station shall sense the medium to decide if another station is transmitting. If the medium is free during a Distributed Inter Frame Space (DIFS), the station waits for an additional random time (Backoff time) to avoid collision with other "waiting" frames then it sends the frame and awaits an acknowledgment (ACK). If the medium is determined to be busy, the sending is differed according to the Backoff procedure. During the contention window CW (CWmin ≤ CW ≤ CWmax), the Backoff time is decreased as long as the medium is sensed idle. If the latter is sensed busy, Backoff time is frozen. When it reaches zero, the frame is transmitted. The medium state (idle/busy) is indicated by the carrier sense function which is performed both through physical and virtual mechanisms. Backoff time is given by the following formula:
BackoffTime = Random() * aSlotTime
(1)
Where aSlotTime depends on the physical layer parameters and Random( ) is a uniform value of distribution in [0, CW]. CW is an entire in [CWmin , CWmax] defined by the physical layer. When there is a collision and no ACK is received, the Backoff time increases according to the following formula:
n ← n +1
CW (n) = (aCWmin + 1) * 2 n − 1
(2)
In order to improve the MAC level according to the medium conditions indicated by the PHY layer, many studies were undertaken. [Qiao et al., 2002] used combination of the SNR ratio, the test frames count and average load as metrics for the connection adaptation algorithm. The latter is based on a pre-established table of a better transmission rate for future transmission attempts. [Chevillat et al., 2003] took into account the acknowledgment obtained by observation to evaluate the medium quality. Thus, if the number of successive succeeded transmissions exceeds S, then rate increases. Otherwise, if the number of failed successive transmissions exceeds F, then the rate decreases. [Barry et al., 2001] proposed a method based on modification of the maximum and minimum bounds of the contention window in order to be able to support two service classes, with high priority and "Best effort". The emulation of the MAC sub layer (Virtual MAC) and application of virtual sources (VS) make possible observation and medium state evaluation by following real packets in parallel and in passive way. In [Pavon et al., 2003], the approach is based on the level of the received signal (RSS) as a decision metric in order to adapt dynamically the transmission rate knowing that the emitted signal level is constant and that relation between RSS and SNR (Signal to Noise Ratio) is linear. [Lampe et al., 2002] proposed the prediction of Packet Error Rate (PER) as decision criterion for the connection adaptation. This is done thanks to the temporary medium transfer function, and the SNR C/I (Carrier/Interference) ratio. In addition, many approaches were developed in order to improve the MAC level according to the medium conditions. In a certain way, the proposed approaches improve the performance of the 802.11 standard but only from a network charge point of view [Bedoui et al., 2005]. However, the increasing number of multimedia applications such as VoIP and video streaming, and the expansion of the Internet causes the taking into account of real time constraints. New multimedia flows are as demanding in quality of service as it is insufficient to be limited to only one layer to support the QoS. Thus, any proposed approach for QoS improvement must deal with the performance metrics relating to physical and network layers, such as packets loss and throughput. 2. AN APPROACH FOR THE SUPPORT OF QOS Many researches concerning IEEE 802.11 WLAN proposed solutions to improve performance of the MAC sublayer without really taking into account the mutual interaction between MAC sublayer on a side and Network layer and of the PHY layer on the other side. We proposed a solution for MAC QoS improvement based on the integration of information resulting from the
Network layer and from the PHY layer of IEEE 802.11 which is the base of IEEE 802.11 a/b/g. 2.1 PHY and MAC interaction Based on the information resulting from the PHY layer, we distinguish between Measured Value Parameters (BER, SNR, etc.) and those with Values defined by the standard according to the concerned physical layer (aSlotTime, aCWmax, aCWmin, DIFS, SIFS, PIFS, etc.). Bit Error Rate (BER) indicates quality and state of medium but its value is calculated only at the receiving station. Then, the transmitting station is considered "blind" in meaning of BER and it will be informed of the medium quality only if it will becomes receiving station. Signal to Noise Rate SNR is a PHY layer parameter which concerns the useful rate depending on data packets computing. This parameter is calculated not only on the level of the receiving station but also at the transmitting station. Thus, we choose SNR parameter: Emitted / received _ signal _ power (3) SNR = Noise _ power According to medium quality, this parameter will indicate to the MAC what kind of traffic to send. Assume that (SNR)BE is the acceptable limit of SNR for the “Best Effort” traffic and (SNR)QoS is the acceptable limit of SNR for the “QoS” traffic. The following condition must be observed: • •
If (SNR) < (SNR)BE, then station differs transmission ; If (SNR)BE ≤ (SNR) < (SNR)QoS, then station differs transmission if we have QoS traffic, otherwise it emits the Best Effort traffic ; • If (SNR) ≥ (SNR)QoS, then station transmits in both cases of Best effort or QoS traffic. This proposal shows clearly the advantage of interaction between layers and the narrow dependence between decision to emit or differ the transmission on MAC, SNR ratio on PHY and traffic classes on Network layer. It’s possible to consider only QoS traffic allowed if the SNR is unfavourable. Although this increase the likelyhood that QoS messages are delivered timely, it reduce overall traffic which represents a considerable disadvantage. 2.2 MAC and Network interaction With an aim of supporting the multimedia applications, IP networks use Diffserv architecture ensuring service differentiation. To be treated in a particular way by various network nodes, packets are classified and marked. This classification is not known by traditional MAC sublayer since it supposes that there is no difference between traffic flows. In order to improve the MAC level to be able to support the QoS, we propose to jointly exploit information obtained from PHY and Network layers to deliver packets following their priorities according to QoS requirements. In a WLAN, frames must be differentiated according to priority classes indicated by high layers in order to ensure point-to-point QoS based on DiffServ. When a data frame MSDU (MAC Service Data Unit) arrives at MAC sublayer, it
is encapsulated in a MPDU (MAC Protocol Data Unit) by addition of a "MAC Heading" field and of a "Frame Control Sequence" field [Pavon et al., 2003].
“Duration” information contained in RTS/CTS frames while it makes this field carrying two important information: "Duration" and "Priority".
The 802.11e is an IEEE standard supporting QoS on the MAC level. It introduces a new function HCF (Hybrid Coordination Function) which defines two medium access mechanisms [IEEE 802.11 amendment 8, 2005][Mangold et al., 2003]: Contention access and Controlled access. The first mechanism refers to EDCF (Enhanced Distributed Coordination Function) which provides distributed and differentiated accesses for wireless medium to users with priority during contention periods. The controlled access is based on election principle with QoS support for free contention periods. It refers to Controlled HCF.
So, information delivered by the network layer can be integrated through RTS/CTS frames which belong to MAC sublayer. And, being able to reach all network stations, these frames carry information on priority of the data frame to be sent. Thus, other stations which are listening to medium can decide to send or delay their frames by comparing their priority level to that of the RTS/CTS. Moreover, the hidden stations, which can not realize that medium is occupied, will be informed not only that medium is busy but also about packet priority.
Knowing that we deal with distributed approaches, we consider in the following EDCF function. It is based on differentiated priorities where the traffic must be delivered according to four access categories representing virtual DCFs [Xiao, 2004]. So, information on the flow priority arriving from high layers will be transmitted to the MAC. A set of WLAN stations can support various types of applications. When a station will transmit a priority flow, other stations must be informed in order to delay their access requests if their flow does not have priority. This action is not currently possible. Thus, we propose that RTS/CTS frames indicate this information. Indeed, being used by MAC level to reserve medium for the required period to data exchange, these control frames reach all network stations even the hidden ones. Also, if a collision is experienced on the RTS, the source station will not receive CTS and then the reservation of required resources failed. So, RTS will be retransmitted in accordance with IEEE 802.11 procedures. The "Duration" field, part of RTS/CTS frames, indicates time, in microseconds, needed to transmit the waiting data or management frames, plus a CTS frame, plus an ACK frame, plus three SIFS intervals. Based on interaction between layers, our proposition is to use "Duration" field (16 bits) to convey "Priority" information from PHY to MAC. The idea is to subtract or not 1 of the field value to distinguish priority packets from non priority ones according to whether the "Duration" values are even or odd. Differently, in the case of a priority flow, the "Duration" value will be odd ("1" is subtracted of this value if it is even). If flow does not have priority, the "Duration" value will be even ("1" is subtracted of this value if it is odd). We use the subtraction to respect the maximum value of "Duration" field defined by IEEE 802.11. Thus, for other stations, the packet has priority if and only if its "Duration" field value is odd. So, "Priority" is memorised (or hidden) in "Duration" which then carry two informations. "Duration" value is defined as follows (in µs):
TD = TCTS + TACK + 3 * TSIFS + T( D ,G )
(4)
TD > TCTS + TACK + 3 * TSIFS With T(D, G) : time in µs of data or management frame. TD is always superior to the sum of CTS, ACK and 3*SIFS times. The subtraction of 1 µs doesn’t have notable effect on
Only two priority levels are considered. They correspond to the two values for the last bit of "Duration" field. We can define four levels by considering the last two bits of the "Duration" field and so on...if necessary. Then, we propose introducing the taking into account of the priority into the Backoff procedure. Assume that: aCWp = (aCWmax - aCWmin )/n where n is the number of priority levels. For n = 2, there will be two intervals at which CW will belong: •
priority flows: CW ∈ [aCWmin , aCWmin
•
non priority flows: CW∈ [aCWmin + aCWp p , aCWmax].
+
aCWp[,
So, the probability of reaching the medium will be lower for non priority flows. The latency (backoff time) of priority flows will be lower and the time needed to access to the medium will be better. 3. SPN MODEL Studies which proposed a modelling of 802.11 networks using Stochastic Petri Nets are not numerous. [German et al., 1999] proposed an interesting contribution where the dynamics of DCF function are represented by a compact and analytically tractable SPN model. This model is defined assuming some simplifications. Station subnets are folded so traffic sources are superposed and the buffer size is set to 1. The Backoff procedure is approximated by an exponential transition with infinite server semantics, its rate corresponds to the mean delay of the first Backoff: 2 × #Pbackoff / (aCWmin × aSlotTime). Therefore, performance measures of the compact model are defined as follows [German et al., 1999]: • Virtual load V = λL /B, normalized to the bit rate ; • Throughput S = E{#gen}L/B normalized to the bit rate ; • Mean waiting time: W = (N - E{#idle}) / E{#gen}. These measures are defined for each station and can be average over all stations. They are expressed in terms of rate and impulse rewards. E{#P} gives the expected number of tokens in place P and E{#T} the expected throughput of transition T. The virtual load represents the buffer capacity. Throughput indicates the quantity of transmitted data by the time unit. In other words, it can be defined as [Bianchi, 2000][Robinson et al., 2004]:
Payload _ Information _ in _ a _ slot _ time (5) E [Length _ of _ a _ slot _ time ] The mean waiting time represents the time between data generation and the end of transmission. Given that the IEEE 802.11 standard doesn’t support the quality of service, the model represents the DCF medium access without traffic differentiation so that priority flow will be seen as ordinary flow. In order to represent our approach in the model, we assume that this differentiation will be indicated by the Backoff Time. Priority traffic will spend less time in Backoff procedure. In accordance with IEEE 802.11, the Backoff time T is given as follows: T = Tbackoff = (2 # bc * (aCWmin + 1) − 1) * aSlotTime) (6) S=
We will not be limited to the first Backoff as done by [German et al., 1999] but we will be conform to the standard and consider all values of bc which initialized with zero and incremented before a repeated Backoff procedure for a pending frame. It can grow up to a maximum value which corresponds to aCWmax. For FHSS WLANs, bcmax = 6. Figure 1 shows the compact SPN model of the IEEE 802.11 DCF supporting QoS.
enabled according to the assigned guards and the token is put either in place Pvuln or Pbackoff. Tokens in place Pvuln represent the number of transmissions in the vulnerable period. When more then one token is in this place, a collision occurs. If only one token is in the place frames, the immediate transition succ is enabled. Otherwise, all tokens passed towards the place Pbackoff through transitions coll, Ttxcoll and Ttimeout. A token is put into place Ptimeout following the firing of Ttxcoll. The firing of Ttxsucc puts a token back to place idle. The backoff procedure is represented by Tbackoff. The firing of this transition puts a token back to place sense and the backoff counter represented by the place bc is updated. Knowing that the arc from Tbackoff to bc is markingdependent, a token is added only if bc contains fewer than bcmax = 6. Also, due to the marking-dependant multiplicity of the arc from bc to Ttsucc, all tokens are moved from bc, which represents the reset of the backoff counter. Assume that: Ti = (2 # i * (aCWmin + 1) − 1) * aSlotTime
α=
N N
gen
defer sense access #Ptxcoll+#Ptxsucc=0 Pvuln
#Pvuln>0
6
coll
succ
Ptxcoll
Ptxsucc
#bc #Ptxcoll>0
Ttxsucc
bc Ptimeout
Ttimeout
Ptxcoll
∑ m ≥ 4, the total number of WLAN stations. i
So, the rate of transition representing Backoff procedure is defined as follows: τ b = 2 × (# Pbackoff ) / α (8) Thus, it is possible to specify several priority levels depending on i values. Since we adopt two priority levels, we consider two backoff times T1 and T0 for respectively priority and non priority traffic. So, the exponential transition rate is as follows: τ b = 2 × (# Pbackoff ) /(n0T0 + n1T1 ) with n0 + n1 ≥ 4 (9) 4. PERFORMANCE EVALUATOION
frames
#frames>1
(7)
i i
i =0
idle
Tvuln
∑m T i =0
with N =
#Ptxcoll+#Ptxsucc>0
6
1
In order to evaluate our Cross-Layer approach and its impact on the improvement of medium access control, an evaluation of performance is carried out using two methods: analytic method based on the SPN model defined in 3 and by simulation. Performances will be evaluated in terms of Throughput and Mean Waiting Time for the first method and in terms of Throughput, Load and Backoff for the second method. We consider the PHY parameters derived from the standard and given by Table 1.
#Ptimeout>0
#Ptxcoll+#Ptxsucc=0
Pbackoff
#bc<6
Tbackoff
Fig. 1. SPN model of the DCF supporting QoS Since traffic sources are superposed, stations with no pending packet to transmit are represented by tokens in place idle. Transition gen represents the data packet generation. After sensing (sense) the channel, access or defer transition is
Table 1. FHSS PHY Parameters Parameter Slot Time aCCATime aRxTxTurnaroundTime SIFS DIFS Packet Maximum Length aCWmin aCWmax
Value 50 µs 27 µs 20 µs 28 µs 128 µs 4095 bytes 15 1023
Moreover, the following parameters independent of the physical layer are considered: RTS=160 bits, CTS=112 bits, ACK = 112 bits, aAirProgrationTime=1 µs, TimeOut=300 µs, bit rate=2 Mbps.
300
250 Without support of QoS 200
4.1 Analytic results
With support of QoS
150
Performance results are obtained by using SPNica tool [German, 2000]. Packet generation rate is determined by λ = V * (2B /N * MaxFrameBody). Thus, virtual load V will be considered as the model parameter. Performance indices will be evaluated for Poisson loads. The scenario is based on two configurations: the first relates to a WLAN network made up of stations in conformity with IEEE 802.11 standard and thus does not support the quality of service. The second describes a WLAN network including at least one priority station generating QoS traffic in accordance with our approach. The performance indices will be analyzed by comparing the results obtained for the two networks (with and without support of QoS). The network is made up of 30 wireless stations including 2 QoS stations. 0,86
100
50
0 1
2
3
4
5
6
7
8
9
10
Virtual load V (Mbps)
Fig.3. Mean waiting time vs. virtual load Figure 3 shows that mean waiting time is practically the same one for the two networks (with and without QoS). Therefore, the presence of priority flows as indicated by our approach did not degrade the network performance. It is as to notice as the curve tends about a maximum time (W≈270ms). Indeed, the contention between a more significant number of wireless terminals led to a more important latency. 4.2 Simulation results
0,85 0,84 0,83
Wit hout suppor t of QoS
0,82
Wit h support of QoS
0,81 0,8 0,79 0,78 0,77 1
2
3
4
5
6
7
8
9
10
V irtual load V (M bps)
Fig.2. Throughput vs. virtual load Figure 2 shows that Throughput increased depending on V and reached its maximum (S ≈ 0,85) for V=2. Then, the values decreased slowly when V increased until reaching a limiting throughput tending towards 0,84 for V=10. This scenario records a performance degradation for load V>2. This is explained by the fact that the more important the number of stations is, the more saturation of the network is fast. We also note that before reaching the maximum, the throughput with support of QoS was higher than that without support of QoS. Then, it is the throughput without QoS which becomes higher. What leads us to conclude that in presence of priority flows, the network is powerful for loads that do not exceed the defined bit rate.
The network modelling and simulation tool OPNET [OPNET modeler] is used to obtain simulation results. These simulations are carried out after adding extensions to the OPNET WLAN module to support the QoS in accordance with our approach. These extensions concerned (1) the coding of the "Priority" in the field "Duration" for a transfer of data (sending terminal), (2) the decoding of this "Priority" for a reception of data (receiving terminal), and (3) the taking into account of this priority by the terminals non concerned by the transfer so that they delay their access to the channel. Simulations concerned many scenarios describing different configurations based on small, average and big size IEEE 802.11 WLANs. Results presented after relate to the scenario of 30 stations since (1) study of other scenarios lead to the same conclusions and (2) it is the same scenario studied by analytic method. The network supporting QoS includes two priority terminals (node_3 and node_7). Results of 60 mn simulation give almost the same performance values. Thus, we will present only the results of four stations including QoS nodes. Figure 4 and figure 5 show Throughput without and with support of QoS.
Fig.4. Throughput without support of QoS
Without support of QoS, all nodes have on average the same load during the simulation time (Load ≈ 5000). In the other case, QoS nodes (node_3 and node_7) have maximum loading for longer periods (Load ≈ 8000). So, priority nodes generate more important traffic than ordinary ones. Figure 8 and 9 show backoff without and with support of QoS.
Fig.5. Throughput with support of QoS Knowing that Throughput is equal to information (in bits) received by MAC from PHY and to be sent to network layer, we notice that QoS nodes (node_3 and node_7) have on average the similar Throughput than other nodes. Indeed, receiving stations, whether with or without QoS, have almost the same Throughput. Figure 6 and 7 indicate load without and with support of QoS.
Fig.8. Backoff without support of QoS
Fig.9. Backoff with support of QoS Fig.6. Load without support of QoS
Being given that Backoff time is defined by the random number of time units (slots) that node must pass to still dispute the access to the medium or to reach it, Figure 8 indicates that all nodes have the same average Backoff time. But, QoS nodes in Figure 9 indicate a number of Backoff slots slightly higher than of ordinary ones. During the majority of simulation time, the number of Backoff slots of QoS nodes is with its maximum. It is the result of the new CW proposed by our approach allowing priority nodes to dispute more the medium based on a contention window close to CWmin. 4.3 Comparaison
Fig.7. Load with support of QoS Load corresponds to the number of bits received by the MAC from the upper layer per unit of time. It relates to data to be sent through the medium.
Performance evaluation was carried out being given what follows: (1) For the analytical method, the performance parameters were evaluated according to a virtual load and a 30 nodes network including 2 QoS nodes;
(2) Performance parameters studied by simulation were analyzed according to duration of simulation and a 30 nodes network including 2 QoS nodes; (3) Analytical method involves simplifications. Comparing the two methods is based on the overall performance of IEEE 802.11 network and on the performance of priority stations. Relaying on the results obtained by the two methods, we can deduce that network supporting QoS in accordance with our Cross-Layer approach shows the following characteristics: • The overall performances of the network are generally improved or in some cases preserved; • Priority stations present better performances for the majority of the parameters evaluated without degradation on average of the ordinary stations performances; • The resources are better used and the access to the wireless medium is optimized. 5. CONCLUSION A performance evaluation of the IEEE 802.11 WLAN supporting QoS has been presented. The quality of service is supported thanks to Cross-Layer design involving the interaction between adjacent layers to the MAC. The goal is to ensure a QoS in real time multimedia applications by ability to adapt to the dynamic medium characteristics. The taking into account of priority and its integration in the Backoff procedure gives an enhanced MAC integrating QoS requirements. The new IEEE 802.11 MAC was evaluated by analytic method based on a stochastic PN model with SPNica and by simulation with OPNET. Analysis and comparison between the two methods results showed that the support of QoS in accordance with our Cross-Layer approach improved or at least preserved the network performances while respecting the IEEE 802.11 requirements. Thus, priority flows can be favoured in a network which misses resources (wireless medium with limited capacity) without however handicapping ordinary flows since the global network quality is preserved. REFERENCES Barry M., Campell A. T., Veres A. (2001). Distributed control algorithms for service differentiation in wireless packet networks. IEEE INFOCOM proceeding, Alaska. Bedoui A., Barkaoui K., Djouani K. (2005). Global solution for the support of QoS by IEEE 802.11. The 10th IFIP International Conference on Personal Wireless Communications, pp 252-260, Imperial College Press, London. Bianchi G. (2000). Performance analysis of IEEE 802.11 distributed coordination function. IEEE Journal on Selected Areas in Communications, volume 18, n°3. Chevillat P., Jelitto J., Barreto A. N., Truong H. L. (2003). A dynamic link adaptation algorithm for IEEE 802.11a wireless LANs. IEEE ICC’03, volume 2, pp 1141-1145. German R. (2000). Performance Analysis of Communication System: Modelling with Non-Markovian Stochastic Petri Nets. John Wiley & Sons, ISBN 0-471-49258-2, England.
German R., Heindl A. (1999). Performance evaluation of IEEE 802.11 wireless LANs with stochastic Petri nets. Proceedings of the Eighth International Workshop on Petri Nets and Performance Models, pp 44-53, Spain. IEEE Standard for Information technology Telecommunications and information exchange between systems - Local and metropolitan area networks-Specific requirements. Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications. ANSI/IEEE Std 802.11, 1999 Edition. IEEE Standard for Information technology Telecommunications and information exchange between systems - Local and metropolitan area networks-Specific requirements. Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications. Amendment 8: Medium Access Control (MAC) Quality of Service Enhancements, September 2005. Lampe M., Rohling H. and Eichinger J. (2002). PERPrediction for Link Adaptation in OFDM Systems. OFDM Workshop, Germany. Mangold S., Choi S., Hiertz G., Klein O., Walk B. (2003). Analysis of IEEE 802.11e for QoS support in wireless LANs. IEEE Wireless Communications. Mühlethaler P. (2002). 802.11 et les réseaux sans fil. Eyrolles edition, ISBN 2-212-11154-1. OPNET modeler, http://www.opnet.com Pavon J. P., Choi S. (2003). Link adaptation strategy for IEEE 802.11 WLAN via received signal strength measurement. IEEE ICC’03, volume 2. Qiao D., Choi S., Shin K. G. (2002). Goodput analysis and link adaptation for IEEE 802.11a wireless LANs. IEEE Trans. Mobile Comp., volume 1. Robinson J. W., Randhawa T. (2004). Saturation throughput analysis of IEEE 802.11e enhanced distributed coordination function. IEEE Journal on Selected Areas in Communications, volume 22, n° 5. Visser M. A., Zarki M. E. (1995). Voice and data transmission over an 802.11 wireless networks. PIMRC proceeding, Canada. Xiao Y. (2004). IEEE 802.11e: QoS provisioning at the MAC layer. IEEE Wireless Communications.
1. INTRODUCTION IEEE 802.11 specifies the Medium Access Control (MAC) and the physical layer (PHY) for wireless connectivity within a local area [IEEE 802.11, 1999]. The goal is to allow interoperability between WLANs systems of the various manufacturers and meet a need increasingly important for wireless communications. Two types of wireless local area network architectures exists [Mühlethaler, 2002]: ad hoc architecture and architecture with access point AP (or infrastructure). The MAC sublayer provides the control of the medium access, deliver data in a reliable way and protect data. It defines two access methods: the Distributed Coordination Function (DCF), which is mandatory, and the Poling Coordination Function (PCF), which is optional. However PCF function, supposed to support real time services, is not implemented in the majority of the commercial 802.11 products. Moreover, the co-operation between PCF and DCF modes leads to performance degradation [Visser et al., 1995]. Besides, DCF can be used in ad hoc network as in infrastructure network, whereas PCF is used only in infrastructure network. So, we choose a distributed control for the medium access ensured by DCF. The fundamental access method is a DCF. It is known as carrier sense multiple accesses with collision avoidance (CSMA/CA). It is based on the listen-before-talk scheme. The 802.11 standard defines a random Backoff time following a busy medium condition in order to reduce the
probability of collisions. In addition, an immediate positive acknowledgment is used to confirm reception. Two access types are proposed by DCF: basic access and RTS/CTS (Request To Send / Clear To Send). Compared to the former, the latter sends RTS/CTS frames of small sizes before sending the data frames, thus making it possible to minimize the collisions. Moreover, RTS/CTS mechanism is used to improve the access control and solve the problem of hidden stations. RTS frame is sent before each data frames transmission. In answer, the receiving station sends a CTS frame to confirm being ready to receive. Thus, all stations update their network allocation vector (NAV) according to RTS/CTS. A station shall sense the medium to decide if another station is transmitting. If the medium is free during a Distributed Inter Frame Space (DIFS), the station waits for an additional random time (Backoff time) to avoid collision with other "waiting" frames then it sends the frame and awaits an acknowledgment (ACK). If the medium is determined to be busy, the sending is differed according to the Backoff procedure. During the contention window CW (CWmin ≤ CW ≤ CWmax), the Backoff time is decreased as long as the medium is sensed idle. If the latter is sensed busy, Backoff time is frozen. When it reaches zero, the frame is transmitted. The medium state (idle/busy) is indicated by the carrier sense function which is performed both through physical and virtual mechanisms. Backoff time is given by the following formula:
BackoffTime = Random() * aSlotTime
(1)
Where aSlotTime depends on the physical layer parameters and Random( ) is a uniform value of distribution in [0, CW]. CW is an entire in [CWmin , CWmax] defined by the physical layer. When there is a collision and no ACK is received, the Backoff time increases according to the following formula:
n ← n +1
CW (n) = (aCWmin + 1) * 2 n − 1
(2)
In order to improve the MAC level according to the medium conditions indicated by the PHY layer, many studies were undertaken. [Qiao et al., 2002] used combination of the SNR ratio, the test frames count and average load as metrics for the connection adaptation algorithm. The latter is based on a pre-established table of a better transmission rate for future transmission attempts. [Chevillat et al., 2003] took into account the acknowledgment obtained by observation to evaluate the medium quality. Thus, if the number of successive succeeded transmissions exceeds S, then rate increases. Otherwise, if the number of failed successive transmissions exceeds F, then the rate decreases. [Barry et al., 2001] proposed a method based on modification of the maximum and minimum bounds of the contention window in order to be able to support two service classes, with high priority and "Best effort". The emulation of the MAC sub layer (Virtual MAC) and application of virtual sources (VS) make possible observation and medium state evaluation by following real packets in parallel and in passive way. In [Pavon et al., 2003], the approach is based on the level of the received signal (RSS) as a decision metric in order to adapt dynamically the transmission rate knowing that the emitted signal level is constant and that relation between RSS and SNR (Signal to Noise Ratio) is linear. [Lampe et al., 2002] proposed the prediction of Packet Error Rate (PER) as decision criterion for the connection adaptation. This is done thanks to the temporary medium transfer function, and the SNR C/I (Carrier/Interference) ratio. In addition, many approaches were developed in order to improve the MAC level according to the medium conditions. In a certain way, the proposed approaches improve the performance of the 802.11 standard but only from a network charge point of view [Bedoui et al., 2005]. However, the increasing number of multimedia applications such as VoIP and video streaming, and the expansion of the Internet causes the taking into account of real time constraints. New multimedia flows are as demanding in quality of service as it is insufficient to be limited to only one layer to support the QoS. Thus, any proposed approach for QoS improvement must deal with the performance metrics relating to physical and network layers, such as packets loss and throughput. 2. AN APPROACH FOR THE SUPPORT OF QOS Many researches concerning IEEE 802.11 WLAN proposed solutions to improve performance of the MAC sublayer without really taking into account the mutual interaction between MAC sublayer on a side and Network layer and of the PHY layer on the other side. We proposed a solution for MAC QoS improvement based on the integration of information resulting from the
Network layer and from the PHY layer of IEEE 802.11 which is the base of IEEE 802.11 a/b/g. 2.1 PHY and MAC interaction Based on the information resulting from the PHY layer, we distinguish between Measured Value Parameters (BER, SNR, etc.) and those with Values defined by the standard according to the concerned physical layer (aSlotTime, aCWmax, aCWmin, DIFS, SIFS, PIFS, etc.). Bit Error Rate (BER) indicates quality and state of medium but its value is calculated only at the receiving station. Then, the transmitting station is considered "blind" in meaning of BER and it will be informed of the medium quality only if it will becomes receiving station. Signal to Noise Rate SNR is a PHY layer parameter which concerns the useful rate depending on data packets computing. This parameter is calculated not only on the level of the receiving station but also at the transmitting station. Thus, we choose SNR parameter: Emitted / received _ signal _ power (3) SNR = Noise _ power According to medium quality, this parameter will indicate to the MAC what kind of traffic to send. Assume that (SNR)BE is the acceptable limit of SNR for the “Best Effort” traffic and (SNR)QoS is the acceptable limit of SNR for the “QoS” traffic. The following condition must be observed: • •
If (SNR) < (SNR)BE, then station differs transmission ; If (SNR)BE ≤ (SNR) < (SNR)QoS, then station differs transmission if we have QoS traffic, otherwise it emits the Best Effort traffic ; • If (SNR) ≥ (SNR)QoS, then station transmits in both cases of Best effort or QoS traffic. This proposal shows clearly the advantage of interaction between layers and the narrow dependence between decision to emit or differ the transmission on MAC, SNR ratio on PHY and traffic classes on Network layer. It’s possible to consider only QoS traffic allowed if the SNR is unfavourable. Although this increase the likelyhood that QoS messages are delivered timely, it reduce overall traffic which represents a considerable disadvantage. 2.2 MAC and Network interaction With an aim of supporting the multimedia applications, IP networks use Diffserv architecture ensuring service differentiation. To be treated in a particular way by various network nodes, packets are classified and marked. This classification is not known by traditional MAC sublayer since it supposes that there is no difference between traffic flows. In order to improve the MAC level to be able to support the QoS, we propose to jointly exploit information obtained from PHY and Network layers to deliver packets following their priorities according to QoS requirements. In a WLAN, frames must be differentiated according to priority classes indicated by high layers in order to ensure point-to-point QoS based on DiffServ. When a data frame MSDU (MAC Service Data Unit) arrives at MAC sublayer, it
is encapsulated in a MPDU (MAC Protocol Data Unit) by addition of a "MAC Heading" field and of a "Frame Control Sequence" field [Pavon et al., 2003].
“Duration” information contained in RTS/CTS frames while it makes this field carrying two important information: "Duration" and "Priority".
The 802.11e is an IEEE standard supporting QoS on the MAC level. It introduces a new function HCF (Hybrid Coordination Function) which defines two medium access mechanisms [IEEE 802.11 amendment 8, 2005][Mangold et al., 2003]: Contention access and Controlled access. The first mechanism refers to EDCF (Enhanced Distributed Coordination Function) which provides distributed and differentiated accesses for wireless medium to users with priority during contention periods. The controlled access is based on election principle with QoS support for free contention periods. It refers to Controlled HCF.
So, information delivered by the network layer can be integrated through RTS/CTS frames which belong to MAC sublayer. And, being able to reach all network stations, these frames carry information on priority of the data frame to be sent. Thus, other stations which are listening to medium can decide to send or delay their frames by comparing their priority level to that of the RTS/CTS. Moreover, the hidden stations, which can not realize that medium is occupied, will be informed not only that medium is busy but also about packet priority.
Knowing that we deal with distributed approaches, we consider in the following EDCF function. It is based on differentiated priorities where the traffic must be delivered according to four access categories representing virtual DCFs [Xiao, 2004]. So, information on the flow priority arriving from high layers will be transmitted to the MAC. A set of WLAN stations can support various types of applications. When a station will transmit a priority flow, other stations must be informed in order to delay their access requests if their flow does not have priority. This action is not currently possible. Thus, we propose that RTS/CTS frames indicate this information. Indeed, being used by MAC level to reserve medium for the required period to data exchange, these control frames reach all network stations even the hidden ones. Also, if a collision is experienced on the RTS, the source station will not receive CTS and then the reservation of required resources failed. So, RTS will be retransmitted in accordance with IEEE 802.11 procedures. The "Duration" field, part of RTS/CTS frames, indicates time, in microseconds, needed to transmit the waiting data or management frames, plus a CTS frame, plus an ACK frame, plus three SIFS intervals. Based on interaction between layers, our proposition is to use "Duration" field (16 bits) to convey "Priority" information from PHY to MAC. The idea is to subtract or not 1 of the field value to distinguish priority packets from non priority ones according to whether the "Duration" values are even or odd. Differently, in the case of a priority flow, the "Duration" value will be odd ("1" is subtracted of this value if it is even). If flow does not have priority, the "Duration" value will be even ("1" is subtracted of this value if it is odd). We use the subtraction to respect the maximum value of "Duration" field defined by IEEE 802.11. Thus, for other stations, the packet has priority if and only if its "Duration" field value is odd. So, "Priority" is memorised (or hidden) in "Duration" which then carry two informations. "Duration" value is defined as follows (in µs):
TD = TCTS + TACK + 3 * TSIFS + T( D ,G )
(4)
TD > TCTS + TACK + 3 * TSIFS With T(D, G) : time in µs of data or management frame. TD is always superior to the sum of CTS, ACK and 3*SIFS times. The subtraction of 1 µs doesn’t have notable effect on
Only two priority levels are considered. They correspond to the two values for the last bit of "Duration" field. We can define four levels by considering the last two bits of the "Duration" field and so on...if necessary. Then, we propose introducing the taking into account of the priority into the Backoff procedure. Assume that: aCWp = (aCWmax - aCWmin )/n where n is the number of priority levels. For n = 2, there will be two intervals at which CW will belong: •
priority flows: CW ∈ [aCWmin , aCWmin
•
non priority flows: CW∈ [aCWmin + aCWp p , aCWmax].
+
aCWp[,
So, the probability of reaching the medium will be lower for non priority flows. The latency (backoff time) of priority flows will be lower and the time needed to access to the medium will be better. 3. SPN MODEL Studies which proposed a modelling of 802.11 networks using Stochastic Petri Nets are not numerous. [German et al., 1999] proposed an interesting contribution where the dynamics of DCF function are represented by a compact and analytically tractable SPN model. This model is defined assuming some simplifications. Station subnets are folded so traffic sources are superposed and the buffer size is set to 1. The Backoff procedure is approximated by an exponential transition with infinite server semantics, its rate corresponds to the mean delay of the first Backoff: 2 × #Pbackoff / (aCWmin × aSlotTime). Therefore, performance measures of the compact model are defined as follows [German et al., 1999]: • Virtual load V = λL /B, normalized to the bit rate ; • Throughput S = E{#gen}L/B normalized to the bit rate ; • Mean waiting time: W = (N - E{#idle}) / E{#gen}. These measures are defined for each station and can be average over all stations. They are expressed in terms of rate and impulse rewards. E{#P} gives the expected number of tokens in place P and E{#T} the expected throughput of transition T. The virtual load represents the buffer capacity. Throughput indicates the quantity of transmitted data by the time unit. In other words, it can be defined as [Bianchi, 2000][Robinson et al., 2004]:
Payload _ Information _ in _ a _ slot _ time (5) E [Length _ of _ a _ slot _ time ] The mean waiting time represents the time between data generation and the end of transmission. Given that the IEEE 802.11 standard doesn’t support the quality of service, the model represents the DCF medium access without traffic differentiation so that priority flow will be seen as ordinary flow. In order to represent our approach in the model, we assume that this differentiation will be indicated by the Backoff Time. Priority traffic will spend less time in Backoff procedure. In accordance with IEEE 802.11, the Backoff time T is given as follows: T = Tbackoff = (2 # bc * (aCWmin + 1) − 1) * aSlotTime) (6) S=
We will not be limited to the first Backoff as done by [German et al., 1999] but we will be conform to the standard and consider all values of bc which initialized with zero and incremented before a repeated Backoff procedure for a pending frame. It can grow up to a maximum value which corresponds to aCWmax. For FHSS WLANs, bcmax = 6. Figure 1 shows the compact SPN model of the IEEE 802.11 DCF supporting QoS.
enabled according to the assigned guards and the token is put either in place Pvuln or Pbackoff. Tokens in place Pvuln represent the number of transmissions in the vulnerable period. When more then one token is in this place, a collision occurs. If only one token is in the place frames, the immediate transition succ is enabled. Otherwise, all tokens passed towards the place Pbackoff through transitions coll, Ttxcoll and Ttimeout. A token is put into place Ptimeout following the firing of Ttxcoll. The firing of Ttxsucc puts a token back to place idle. The backoff procedure is represented by Tbackoff. The firing of this transition puts a token back to place sense and the backoff counter represented by the place bc is updated. Knowing that the arc from Tbackoff to bc is markingdependent, a token is added only if bc contains fewer than bcmax = 6. Also, due to the marking-dependant multiplicity of the arc from bc to Ttsucc, all tokens are moved from bc, which represents the reset of the backoff counter. Assume that: Ti = (2 # i * (aCWmin + 1) − 1) * aSlotTime
α=
N N
gen
defer sense access #Ptxcoll+#Ptxsucc=0 Pvuln
#Pvuln>0
6
coll
succ
Ptxcoll
Ptxsucc
#bc #Ptxcoll>0
Ttxsucc
bc Ptimeout
Ttimeout
Ptxcoll
∑ m ≥ 4, the total number of WLAN stations. i
So, the rate of transition representing Backoff procedure is defined as follows: τ b = 2 × (# Pbackoff ) / α (8) Thus, it is possible to specify several priority levels depending on i values. Since we adopt two priority levels, we consider two backoff times T1 and T0 for respectively priority and non priority traffic. So, the exponential transition rate is as follows: τ b = 2 × (# Pbackoff ) /(n0T0 + n1T1 ) with n0 + n1 ≥ 4 (9) 4. PERFORMANCE EVALUATOION
frames
#frames>1
(7)
i i
i =0
idle
Tvuln
∑m T i =0
with N =
#Ptxcoll+#Ptxsucc>0
6
1
In order to evaluate our Cross-Layer approach and its impact on the improvement of medium access control, an evaluation of performance is carried out using two methods: analytic method based on the SPN model defined in 3 and by simulation. Performances will be evaluated in terms of Throughput and Mean Waiting Time for the first method and in terms of Throughput, Load and Backoff for the second method. We consider the PHY parameters derived from the standard and given by Table 1.
#Ptimeout>0
#Ptxcoll+#Ptxsucc=0
Pbackoff
#bc<6
Tbackoff
Fig. 1. SPN model of the DCF supporting QoS Since traffic sources are superposed, stations with no pending packet to transmit are represented by tokens in place idle. Transition gen represents the data packet generation. After sensing (sense) the channel, access or defer transition is
Table 1. FHSS PHY Parameters Parameter Slot Time aCCATime aRxTxTurnaroundTime SIFS DIFS Packet Maximum Length aCWmin aCWmax
Value 50 µs 27 µs 20 µs 28 µs 128 µs 4095 bytes 15 1023
Moreover, the following parameters independent of the physical layer are considered: RTS=160 bits, CTS=112 bits, ACK = 112 bits, aAirProgrationTime=1 µs, TimeOut=300 µs, bit rate=2 Mbps.
300
250 Without support of QoS 200
4.1 Analytic results
With support of QoS
150
Performance results are obtained by using SPNica tool [German, 2000]. Packet generation rate is determined by λ = V * (2B /N * MaxFrameBody). Thus, virtual load V will be considered as the model parameter. Performance indices will be evaluated for Poisson loads. The scenario is based on two configurations: the first relates to a WLAN network made up of stations in conformity with IEEE 802.11 standard and thus does not support the quality of service. The second describes a WLAN network including at least one priority station generating QoS traffic in accordance with our approach. The performance indices will be analyzed by comparing the results obtained for the two networks (with and without support of QoS). The network is made up of 30 wireless stations including 2 QoS stations. 0,86
100
50
0 1
2
3
4
5
6
7
8
9
10
Virtual load V (Mbps)
Fig.3. Mean waiting time vs. virtual load Figure 3 shows that mean waiting time is practically the same one for the two networks (with and without QoS). Therefore, the presence of priority flows as indicated by our approach did not degrade the network performance. It is as to notice as the curve tends about a maximum time (W≈270ms). Indeed, the contention between a more significant number of wireless terminals led to a more important latency. 4.2 Simulation results
0,85 0,84 0,83
Wit hout suppor t of QoS
0,82
Wit h support of QoS
0,81 0,8 0,79 0,78 0,77 1
2
3
4
5
6
7
8
9
10
V irtual load V (M bps)
Fig.2. Throughput vs. virtual load Figure 2 shows that Throughput increased depending on V and reached its maximum (S ≈ 0,85) for V=2. Then, the values decreased slowly when V increased until reaching a limiting throughput tending towards 0,84 for V=10. This scenario records a performance degradation for load V>2. This is explained by the fact that the more important the number of stations is, the more saturation of the network is fast. We also note that before reaching the maximum, the throughput with support of QoS was higher than that without support of QoS. Then, it is the throughput without QoS which becomes higher. What leads us to conclude that in presence of priority flows, the network is powerful for loads that do not exceed the defined bit rate.
The network modelling and simulation tool OPNET [OPNET modeler] is used to obtain simulation results. These simulations are carried out after adding extensions to the OPNET WLAN module to support the QoS in accordance with our approach. These extensions concerned (1) the coding of the "Priority" in the field "Duration" for a transfer of data (sending terminal), (2) the decoding of this "Priority" for a reception of data (receiving terminal), and (3) the taking into account of this priority by the terminals non concerned by the transfer so that they delay their access to the channel. Simulations concerned many scenarios describing different configurations based on small, average and big size IEEE 802.11 WLANs. Results presented after relate to the scenario of 30 stations since (1) study of other scenarios lead to the same conclusions and (2) it is the same scenario studied by analytic method. The network supporting QoS includes two priority terminals (node_3 and node_7). Results of 60 mn simulation give almost the same performance values. Thus, we will present only the results of four stations including QoS nodes. Figure 4 and figure 5 show Throughput without and with support of QoS.
Fig.4. Throughput without support of QoS
Without support of QoS, all nodes have on average the same load during the simulation time (Load ≈ 5000). In the other case, QoS nodes (node_3 and node_7) have maximum loading for longer periods (Load ≈ 8000). So, priority nodes generate more important traffic than ordinary ones. Figure 8 and 9 show backoff without and with support of QoS.
Fig.5. Throughput with support of QoS Knowing that Throughput is equal to information (in bits) received by MAC from PHY and to be sent to network layer, we notice that QoS nodes (node_3 and node_7) have on average the similar Throughput than other nodes. Indeed, receiving stations, whether with or without QoS, have almost the same Throughput. Figure 6 and 7 indicate load without and with support of QoS.
Fig.8. Backoff without support of QoS
Fig.9. Backoff with support of QoS Fig.6. Load without support of QoS
Being given that Backoff time is defined by the random number of time units (slots) that node must pass to still dispute the access to the medium or to reach it, Figure 8 indicates that all nodes have the same average Backoff time. But, QoS nodes in Figure 9 indicate a number of Backoff slots slightly higher than of ordinary ones. During the majority of simulation time, the number of Backoff slots of QoS nodes is with its maximum. It is the result of the new CW proposed by our approach allowing priority nodes to dispute more the medium based on a contention window close to CWmin. 4.3 Comparaison
Fig.7. Load with support of QoS Load corresponds to the number of bits received by the MAC from the upper layer per unit of time. It relates to data to be sent through the medium.
Performance evaluation was carried out being given what follows: (1) For the analytical method, the performance parameters were evaluated according to a virtual load and a 30 nodes network including 2 QoS nodes;
(2) Performance parameters studied by simulation were analyzed according to duration of simulation and a 30 nodes network including 2 QoS nodes; (3) Analytical method involves simplifications. Comparing the two methods is based on the overall performance of IEEE 802.11 network and on the performance of priority stations. Relaying on the results obtained by the two methods, we can deduce that network supporting QoS in accordance with our Cross-Layer approach shows the following characteristics: • The overall performances of the network are generally improved or in some cases preserved; • Priority stations present better performances for the majority of the parameters evaluated without degradation on average of the ordinary stations performances; • The resources are better used and the access to the wireless medium is optimized. 5. CONCLUSION A performance evaluation of the IEEE 802.11 WLAN supporting QoS has been presented. The quality of service is supported thanks to Cross-Layer design involving the interaction between adjacent layers to the MAC. The goal is to ensure a QoS in real time multimedia applications by ability to adapt to the dynamic medium characteristics. The taking into account of priority and its integration in the Backoff procedure gives an enhanced MAC integrating QoS requirements. The new IEEE 802.11 MAC was evaluated by analytic method based on a stochastic PN model with SPNica and by simulation with OPNET. Analysis and comparison between the two methods results showed that the support of QoS in accordance with our Cross-Layer approach improved or at least preserved the network performances while respecting the IEEE 802.11 requirements. Thus, priority flows can be favoured in a network which misses resources (wireless medium with limited capacity) without however handicapping ordinary flows since the global network quality is preserved. REFERENCES Barry M., Campell A. T., Veres A. (2001). Distributed control algorithms for service differentiation in wireless packet networks. IEEE INFOCOM proceeding, Alaska. Bedoui A., Barkaoui K., Djouani K. (2005). Global solution for the support of QoS by IEEE 802.11. The 10th IFIP International Conference on Personal Wireless Communications, pp 252-260, Imperial College Press, London. Bianchi G. (2000). Performance analysis of IEEE 802.11 distributed coordination function. IEEE Journal on Selected Areas in Communications, volume 18, n°3. Chevillat P., Jelitto J., Barreto A. N., Truong H. L. (2003). A dynamic link adaptation algorithm for IEEE 802.11a wireless LANs. IEEE ICC’03, volume 2, pp 1141-1145. German R. (2000). Performance Analysis of Communication System: Modelling with Non-Markovian Stochastic Petri Nets. John Wiley & Sons, ISBN 0-471-49258-2, England.
German R., Heindl A. (1999). Performance evaluation of IEEE 802.11 wireless LANs with stochastic Petri nets. Proceedings of the Eighth International Workshop on Petri Nets and Performance Models, pp 44-53, Spain. IEEE Standard for Information technology Telecommunications and information exchange between systems - Local and metropolitan area networks-Specific requirements. Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications. ANSI/IEEE Std 802.11, 1999 Edition. IEEE Standard for Information technology Telecommunications and information exchange between systems - Local and metropolitan area networks-Specific requirements. Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications. Amendment 8: Medium Access Control (MAC) Quality of Service Enhancements, September 2005. Lampe M., Rohling H. and Eichinger J. (2002). PERPrediction for Link Adaptation in OFDM Systems. OFDM Workshop, Germany. Mangold S., Choi S., Hiertz G., Klein O., Walk B. (2003). Analysis of IEEE 802.11e for QoS support in wireless LANs. IEEE Wireless Communications. Mühlethaler P. (2002). 802.11 et les réseaux sans fil. Eyrolles edition, ISBN 2-212-11154-1. OPNET modeler, http://www.opnet.com Pavon J. P., Choi S. (2003). Link adaptation strategy for IEEE 802.11 WLAN via received signal strength measurement. IEEE ICC’03, volume 2. Qiao D., Choi S., Shin K. G. (2002). Goodput analysis and link adaptation for IEEE 802.11a wireless LANs. IEEE Trans. Mobile Comp., volume 1. Robinson J. W., Randhawa T. (2004). Saturation throughput analysis of IEEE 802.11e enhanced distributed coordination function. IEEE Journal on Selected Areas in Communications, volume 22, n° 5. Visser M. A., Zarki M. E. (1995). Voice and data transmission over an 802.11 wireless networks. PIMRC proceeding, Canada. Xiao Y. (2004). IEEE 802.11e: QoS provisioning at the MAC layer. IEEE Wireless Communications.