IJCSNS International Journal of Computer Science and Network Security, VOL.6 No.

12, December 2006

281

Hybrid adaptation of the maximum contention window (CWmax) and

minimum contention window (CWmin) for Enhanced Service
Differentiation in IEEE 802.11 Wireless Ad-hoc Networks*
Hassan Takabi, Ali H. Moghadam, and Ahmad Khonsari,,

Computer Engineering Department, Sharif University of Technology, Tehran, Iran

Department of Electrical and Computer Engineering, University of Tehran, Tehran, Iran

Institute for Studies in Theoretical Physics and Mathematics (IPM), Tehran, Iran
than a physical wire. The WLAN medium is also
unlicensed and is therefore subject to interference from
Summary
The IEEE 802.11e draft standard that proposed by IEEE Task
other devices. To further compound the problem, wireless
Group E, defines new MAC protocols for QoS in wireless
devices are generally constrained by size, weight and
networks, mainly HCF that combines functions from DCF and
battery size, limiting the processing power and the battery
PCF with enhanced QoS-specific mechanisms and frame types.
life. These factors further limit the capability of the
HCF has two modes of operation EDCA and HCCA. EDCA is a
network to provide an optimal solution. The main
contention-based mechanism and HCCA is a polling-based
objective of WLAN QoS is to optimize use of limited
mechanism for channel access. EDCA, in fact, is an extension to
bandwidth offered by a WLAN to address the issues noted
DCF and is a contention-based channel access method that
above.
provides prioritized access to the medium. In this paper, we
To optimize the best use of the resources and fulfill the
propose to extend EDCA using a hybrid adaptation algorithm of
the maximum contention window(CWmax) and minimum
resource requirements of different applications, QoS
contention
window(CWmin)
for
Enhanced
Service
provides mechanisms to control access and usage of the
Differentiation in IEEE 802.11 Wireless Ad-hoc Networks, that
medium based on the application. Each application has
enables each station to adjust size of CWmax and CWmin used
different needs in terms of bandwidth, delay, jitter, and
in its back-off algorithm at run time. The purpose of suggested
packet-error rate and, therefore, QoS must cater to each of
scheme is to reduce delay and jitter and increase the efficiency of
these needs. Applications requiring low delay (e.g., voice)
the transmission channel. Priorities between access categories are
may be given higher priority to use the medium, whereas
prepared by updating the size of the CWmax and CWmin
applications requiring higher bandwidth may be assigned
according to application requirements and network conditions.
longer transmit times (e.g., video). Other traffic may
The performances of the IEEE 802.11e EDCA, enhanced with
our hybrid adaptation algorithm, are inquired by simulations. Our
require high reliability (e.g., email and data) and must be
results show that the hybrid adaptation algorithm outperforms the
delivered with low packet-error rate.
802.11e EDCA standard in terms of channel throughput, packet
Quality of Service (QoS) support is critical to multimedia
delay and utilization, specially at high traffic load conditions.
applications [14]. These applications, including audio,
Indeed our proposed scheme increases the total throughput for
video conferencing, voice, etc, require some specified
high priority access category by up to 27% and reduces the delay
bandwidth, delay, jitter and error rate guarantee to support
for high priority traffic more than 49%. Furthermore, channel
a certain Quality of Service (QoS). Guaranteeing these
utilization ratio also increases at least 23%. Moreover,
QoS requirements is a challenging task with regard to
throughput for medium and low priority access categories
802.11 WLAN [3], [8], [9] protocols and Medium Access
remains stable.
Control (MAC) functions [5].
Key words:
IEEE 802.11e, QoS, EDCA, back-off algorithm, wireless ad-hoc
In order to support QoS in 802.11 WLAN, several priority
networks, hybrid adaptation algorithm, CWmax, CWmin
schemes has been developed [1], [4], [15], [16]. There are
some priority schemes under discussion currently [7], [8].
The IEEE 802.11 Task Group E is working on the support
1. Introduction
of QoS in a new standard. It is defining enhancements to
the 802.11 MAC access methods to support QoS,
One of the drawbacks of wireless networks in comparison
providing the classes of service, enhanced security and
to wired networks is that they are generally less efficient
authentication mechanism [6]. These enhancements are
and unpredictable. Wireless has limited bandwidth, high
defined in 802.11e draft [7] which introduces a main
packet overheads, and is more prone to environmental
access method, HCF that combines functions from DCF
factors such as obstructions, interference, and weather and
and PCF with enhanced QoS-specific mechanisms. HCF
so on. The wireless medium (air) is much harder to control

* This research was in part supported by a grant from I.P.M. (No. CS1385-2-02).
Manuscript received December 5, 2006.
Manuscript revised December 25, 2006.

282

IJCSNS International Journal of Computer Science and Network Security, VOL.6 No.12, December 2006

has two modes of operation, Enhanced Distribution

Coordinate Access (EDCA) and HCF Controlled Channel
Access (HCCA). EDCA is a contention-based mechanism
and HCCA is a polling-based mechanism for channel
access. EDCA is an extension to DCF and is a contentionbased channel access method that provides prioritized
access to the medium [2], [7], [11]. EDCA is part of HCF
for infrastructure networks. However, it may be used as a
separate coordination function for wireless ad-hoc
networks. The EDCA channel access method defines 4
access categories (ACs) based on the IEEE 802.11D
standards and provides differentiated service, distributed
access to the wireless medium for them [7]. A traffic
category (TC) in 802.11 is defined as the application
traffic related to a special user priority (UP) specified in
IEEE 802.11D. The mapping between traffic categories
(TCs) and access categories (ACs) is presented in IEEE
802.11e draft standard [2].
The proposed EDCA channel access mechanism uses at
most 4 prioritized output queues on each QoS station
(QSTA), one for each Access Category (AC). Figure 1
illustrates the different queues for different priorities.
Different from a station, a QoS-supporting Access Point
(QAP) should support at least 4 Access Categories (ACs).
Service differentiation is provided by changing the size of
the contention window (CW). EDCA uses the contention
window to assign priority to each access category. It
assigns a short contention window (CW) to an access
category (AC) that should have higher priority in order to
ensure that in most cases, high priority ACs will be able to
transmit before the low priority ones. Indeed, the CWmin
and CWmax parameters can be set differently for different
access categories, yielding higher priority ACs with
smaller values of CWmin and CWmax.

Figure 1. Queue-based EDCA fvs. basic DCF

For further differentiation, in 802.11e different IFS (Inter

Frame Space) can be used according to traffic categories.
Instead of DIFS, an Arbitration IFS (AIFS) is used. The
AIFS for a given category should be a DIFS plus some

(possibly zero) time slots. Categories with the smallest

AIFS will have the highest priority as it is shown in figure
2.

Figure 2. Some IFS Relationship

After any unsuccessful transmission (i.e. collision) to

reduce the probability of a new collision, a new contention
window (CW) is calculated with the help of the
persistence factor PF[ACi] as shown in equation (1).
Whereas in legacy 802.11 [9], CW is always doubled after
any unsuccessful transmission (i.e., PF[ACi]=2). EDCA
uses the PF to increase the CW different for each access
category. Note that in the latest 802.11e draft [7] PFs
differentiation per access category are no longer
considered, i.e., for all access categories PFs equals to 2.
CW new [ AC i ] = ((CW old [ AC i ]) PF [ AC i ]) 1 (1)

In addition, the CW never exceeds the maximum

possible value for contention windows associated with
each access categories, i.e., CWmax[ACi].
In this paper, we focus on the hybrid adaptation of the
maximum contention window (CWmax) and minimum
contention window (CWmin) for different channel
conditions. We compare the performance of our proposed
scheme with basic EDCA related to 802.11e proposed
standard.
The rest of this paper is organized as fallows. In
section 2, we present the proposed hybrid adaptation
algorithm scheme. In section 3, the simulation topology
and parameters are described. Simulations results and the
performance of our proposed scheme are detailed in
section 4. Finally in section 5, we conclude and describe
our outlining future works.

2. HYBRID ADAPTATION ALGORITHM

In the rest of this paper, we assume that n stations are
sending packets through the wireless media. The flows
sent by each station may belong to different category of
service with various priority levels. In each station and for
each category i, the scheme maintains: the current
contention window value (CW[i]), the minimum

IJCSNS International Journal of Computer Science and Network Security, VOL.6 No.12, December 2006

contention window value (CWmin[i]), and the maximum

contention window value (CWmax[i]). Note that ACi is
the ith access category, with i varies between 0 and 3 and
that the high priority level is 0 and low priority is 3.
Both in the legacy DCF [9] and EDCA [7], after a
successful transmission, the basic EDCA mechanism sets
the contention window (CW) size of the corresponding
category to its minimum contention window (CWmin)
regardless the channel conditions. The problem is that
when the transmission channel is high loaded or in a
congested state, such aggressive reduction of the CW
could cause more collisions. Also after an unsuccessful
transmission (i.e. collision), the basic EDCA mechanism
increases the contention window (CW) size of the
corresponding category to its maximum contention
window (CWmax) regardless the network condition. In
this case the new contention window is calculated with the
help of the persistence factor PF[ACi] whereas in legacy
DCF [9], always we have PF[ACi]=2, but EDCA [7] uses
the PF to increase the CW different for each access
category. Note that in the latest 802.11e draft standard [7],
like DCF [9], for all access categories PFs equals to 2.
Furthermore, the CW never exceeds the maximum
possible value for contention windows associated with
each access categories, i.e., CWmax[ACi]. The problem is
that when setting CWmax[i] too small the back-off growth
stops too soon and delay an jitter increases (because of a
higher number of collisions) and if the CWmax[i] is too
large, the back-off growth stops too late and results in
greater delay and jitter.
Previous researches focus on the adaptive enhanced
distributed coordination function [14], the dynamic tuning
of the minimum contention window (CWmin) value after
successful transmissions [5] and the dynamic adaptation of
the maximum contention window (CWmax) value after
unsuccessful transmissions [6].
In this paper, we focus on hybrid adaptation of the
CWmax and CWmin in both successful and unsuccessful
transmissions. The main idea behind hybrid adaptation of
the CWmax and CWmin is to adapt the CWmax[i] and
CWmin[i] values for a certain access category i to traffic
load and channel conditions. We believe, that by hybrid
adaptation of CWmax and CWmin in both successful and
unsuccessful transmissions for a certain access category i
to the traffic load and channel conditions, we can improve
parameters of QoS such as delay, jitter, throughput and
channel utilization.2. Tables, Figures and Equations

2.1 Scheme Description

In the basic EDCA scheme for ad-hoc networks [7], [11],
the CWmax[i] and CWmin[i] are statically set for each
priority level. After each successful transmission the
contention window is reduce to CWmin[i]. After each

283

unsuccessful transmission (i.e. collision) the contention

window is doubled, i.e. with an exponential back-off, and
if it reaches or higher than the CWmax[i], so CW[i]
remains at this value.
So, we propose that, after successful transmission update
CWmax[i] and after collision update CWmin[i], both
according to the traffic load and channel conditions.
We note that we use both a hybrid adaptation mechanism
for CWmax[i] and CWmin[i] according to traffic load and
also we differentiate between ACs while updating
CWmax[i] and CWmin[i] for different priority levels.
In the next sub-sections detail how the contention window
of each priority level i is set after each successful
transmission and also after each collision.

2.2. Setting CW after Each Successful Transmission

After each successful transmission, the basic EDCA
mechanism simply sets the contention window of the
corresponding category to its minimum contention
window regardless the network conditions, which
probably lead to bursty collisions. In our hybrid adaptation
scheme, we propose that each access category updates
CWmin[i] and CWmax[i] parameters in an adaptive way
using the estimated collision rate in each station at regular
update period T update expressed in time slots. We re-use
the method defined in [14] to estimate the average
collision rate as seen by a station p. Instantaneous collision
j
at the j th update period T update is
rate f curr
calculated using the number of collisions and the total
number of packets sent during that period. The value of
collision rate is given by equation (2).
E (collisions j [ p ])
j
=
f curr
E (data _ sent j [ p])
(2)
Where E(collisionsj[p]) is the number of collisions of
station p during period (update priod or step) j, and
E(data_sentj[p]) is the total number of packets that have
been sent during the update period T update . Note that the
j
above ratio f curr
is always in the range of [0, 1].
In order to minimize the bias against transient collisions,
we use an estimator of Exponentially Weighted Moving
Average (EWMA) to smoothen the estimated values. The
value of collision rate at step j is given by equation (3).
j = (1 ) f j + f j 1
f avg
curr
avg
(3)
th
update period T update and
Where j refers to the j
j
stands for the instantaneous collision rate, is
f curr

284

IJCSNS International Journal of Computer Science and Network Security, VOL.6 No.12, December 2006

the weight (also called the smoothing factor) in the range

[0, 1].
For adaptive calculation of the values of CWmax and
CWmin, we re-use the methods that defined in [5]. Since,
after each successful transmission, contention window sets
to minimum contention window, in this case, we calculate
the value of CWmin in an adaptive way by equations (4),
(5).
We can calculate the value of CWmin in an adaptive way,
using the equations (4).
j
DCW min [i ] = (1 f avg ) CW min [i ] +
j (
f avg
CW max [i ] CW min [i ]) 2 i 2
(4)
Where DCWmin[i] shows the adaptive value of contention
window minimum for an access category i, CWmin[i] is
the minimum contention window (according to EDCA)
j
repassigned for the same access category i and f avg
resents the estimated collision rate at step j.
The dynamic contention window minimum for AC i
obtained in equation (4) varies between a lower bound of
CWmin[i], when the collision rate equals to 0, and an
upper bound of (CW max [i ] CW min [i ]) 2 i 2 , when
the collision rate equals to 1. Thus, this upper bound
depends on the priority level of the access category.
Indeed, this upper bound of DCWmin[i] is smaller for
high priority traffic and greater otherwise. In order to
ensure that the adaptive contention window minimum has
an upper bound, the derived formula (in equation 4) uses
the static value of CWmax according to EDCA along with
the following formula:

DCW min [i] = min( DCW min [i], CW max [i])

(5)

2.3. Setting CW after Each Collision

The basic EDCA, after each unsuccessful transmission of
packet of class i, the CW of this class is doubled, while
remaining less than the maximum contention window
CWmax[i]. In our hybrid adaptation scheme, we propose
that each access category updates CWmin[i] and
CWmax[i] parameters in an adaptive way using the
estimated collision rate in each station at regular update
period T update expressed in time slots.
For adaptive calculation of the values of CWmax and
CWmin, we re-use the methods that defined in [6]. Since,
after each unsuccessful transmission, contention window
sets to maximum contention window, in this case, we
calculate the value of CWmax in an adaptive way, using
the equations (6), (7).
The value of maximum contention window CWmax is
calculated as bellows:

j )5 2 i
DCW max [i ] = ()i + 1) ( f avg
(CW max [i ] CW min [i ])

newCW max [i ] = 2 i + 3 CW min [i ] +

DCW max [i ]

(6)

(7)

Where newCWmax[i] shows the adaptive value of

contention window maximum for an access category i
(AC), CWmin[i] is the contention window minimum
assigned for the same AC i and DCWmax[i] is a dynamic
value of newCWmax.
In equation (6), CWmax[i] represents the static value of
the contention window maximum according to EDCA and
j
is the estimated collision rate. The new contention
f avg
window value in equation (7) consists of two parts, a static
part and a dynamic part given by DCWmax[i].
In order to ensure that the adaptive contention window
maximum has an upper bound, the derived formulas (6)
and (7) use the static value of CWmax according to EDCA
along with the following formula:
newCW max [i ] = min(newCW max [i ], max PHYCW lim)
(8)
Where maxPHYCWlim is the maximum size of contention
window limited by the physical layer, e.g., we use here a
maximum value of the contention window of a 1023 slots.
In the adaptive contention window maximum defined in
equation (6) and (7) there is a minimum backoff of at least
i+3 times if there is no collision in the past (i.e., lower
bound of the adaptive contention window maximum).
With, the increase in the collision rate the value of
newCWmax[i] increases and so with the value of i.
In order to ensure that the adaptive contention window
maximum has an upper bound, the derived formulas (6)
and (7) use the static value of CWmax according to EDCA
along with the equation (8).
In the adaptive contention window maximum defined in
equation (6) and (7) there is a minimum backoff of at least
i+3 times if there is no collision in the past (i.e., lower
bound of the adaptive contention window maximum).
With, the increase in the collision rate the value of
newCWmax[i] increases and so with the value of i.
After each unsuccessful transmission the contention
window for an access category i is set as the following:
CW new [i ] = max(newCW max [i ], PF [i ] CW old [i ]) (9)

In our scheme we are using a PF[i]=2 for all access

categories, so that, the contention window is doubled

IJCSNS International Journal of Computer Science and Network Security, VOL.6 No.12, December 2006

while remaining less than the maximum adaptive

contention window, i.e., newCWmax[i].

3. SIMULATION METHODOLOGY
We have implemented our proposed scheme under the
network simulator ns-2 [13] using the Atheros semipackage to support QoS enhancements features. We report
in this section part of simulations we have done with
different network topologies and source characteristics.
Indeed this section presents the generic simulation
topology used in order to evaluate the performance of the
hybrid adaptation of the CWmin and CWmax scheme as
well as a detailed analysis for a proper selection of and
T update parameters used in the proposed scheme. An
analysis of performance is presented in detail.

3.1. Simulation Topology

We use a generic topology (circular routing scenario)
shown in figure 3, which consists of n stations indexed
from 1 to n. Each station generates three types of data
streams, labeled with high, medium and low according to
their priorities. These data streams belong to the three
traffic categories (TCs), respectively, audio (high), video
(medium) and background traffic (low). Station n sends
packets to station 1 and station i sends packets to station
i+1. The highest priority access category in each station
generates packets at sending rate of 64Kbps (PCM audio
flow) which corresponds to a packet size of 160 bytes and
an inter-packet arrival of 20ms. The medium priority
access category generates packets at sending rate of
1024Kbps which corresponds to a packet size of 1280
bytes and inter-packet interval of 10ms. The low priority
access category sending rate is 128Kbps which represents
a packet size of 200 bytes and an inter-arrival packet of
12.5 ms. To increase the load of the system, we gradually
increase the number of stations.

Figure 3. Simulation Topology

285

All the stations are located within an Independent Basic

Service Set (IBSS), such that each station can detect the
transmission from any other station. The different nodes
are uniformly spread out of 500 500m 2 dimensions in
2D space. Table 1 shows the different MAC parameters
for the three access categories (0, 1, and 2) used in the
different simulation scenarios.
Table 1.MAC parameters for the three TCs

Parameters
CWmin
CWmax
AIFS(s)
PF
Packet
size
(bytes)
Packet Interval
(ms)
Sending
rate(Kbps)

High
7
200
34
2
160

Medium
15
500
43
2
1280

Low
31
1023
52
2
200

20

10

12.5

64

1024

128

This table presents as well as the parameters (e.g., packet

size, sending rate and packet interval) of the three traffic
categories associated with the defined three access
categories. Table 2 presents the 802.11a PHY/MAC
parameters.
In the following simulations, we assume that each QSTA
operates at IEEE 802.11a PHY mode 6 [9] (i.e.,
modulation 16-QAM, coding rate of 3/4, data rate of 36
Mbps).
Table 2. 802.11a PHY/MAC parameters used in simulations

SIFS
DIFS
ACK size
Data rate
Slot time
CCA time
MAC header
Modulation
Preamble Length
RxTxTurnaround time
PLCP header length

16s
34s
14 bytes
36 Mbps
9s
3s
28 bytes
16-QAM
20s
1s
4s

IJCSNS International Journal of Computer Science and Network Security, VOL.6 No.12, December 2006

286

3.2. Impact of T update (Update Period) and

(Smoothing Factor) Parameters
As described in Section 2.1, our proposed scheme uses an
update period T update (defined in number of time slots)
and a smoothing factor . In order to observe the effect
of two mentioned parameters on the delay and on the
goodput performances, we have done several set of
simulations. First, to analyse the effect of the smoothing
factor we vary the value of in the range of [0, 1] and
we set the update period to T update =8000 time slots and
we run simulations for a fixed number of stations, i.e. 25
stations. We run 20 simulations and results are averaged
over these simulations. Goodput is defined as the total
application layer received bytes divided by the total
simulation time. Two performance criteria are used, the
total goodput (or throughput) and the mean audio delay.
The effect of smoothing factor on total goodput and mean
audio delay shown in figure 4 and 5.

Figure 4 shows that a value of in the range of [0.55,

0.65] achieves a higher goodput with a maximum of
goodput for =0.6. Also, we have higher goodput for
values of in the range [0, 0.2]. Since, small values
could contribute to random fluctuations we consider only
values in the range of [0.55, 0.65].
From figure 5, we can see that, a value of in the range
[0.6, 0.9] achieves a lower delay and lowest delay is for
=0.9. Thus, we can see that values of in the range of
[0.6, 0.7] achieve a best trade-off between higher total
goodput and lower mean audio delay. So, in the following
simulations we set to 0.6.
Figure 6 and 7 show the variations of total goodput and
mean audio delay as a function of the update period values
expressed in time-slots. The selection of the value of
update period, T update , should take into account that
higher values make adaptations less useful and smaller
ones could hurt the adaptation scheme since high frequent
updates of CWmax and CWmin could be influenced by
channel fluctuations.

Impact of smoothing factor on goodput

Impact of update period on goodput

1520

1940
1920

1500

Goodput(KBps)

1490
1480
1470
1460

1900
1880
1860
1840
1820

1450
0.9

Sm oothing factor

35
00
0

0.8

29
00
0

0.7

25
00
0

0.6

21
00
0

0.5

17
00
0

0.4

13
00
0

0.3

70
00

0.2

50
0

0.1

40
00

1800

1440

20
00

Goodput (KBps)

1510

Update Period

Figure 4. Impact of smoothing factor on goodput

Figure 6. Impact of update period on goodput

Impact of smoothing factor on mean audio delay

Impact of updare period on mean audio delay

5

32
00

0.9

0.8

27
00

0.7

0.6

23
00

0.5

0.4

sm oothing factor

19
00

0.3

0.2

15
00

0.1

10
00

70
00

0.5

40
00

2
1.5
1

20
00

3
2.5

6
5.9
5.8
5.7
5.6
5.5
5.4
5.3
5.2
5.1
5

50
0

4
3.5

Mean audio delay (ms)

mean audio delay (ms)

4.5

Update period

Figure 5. Impact of smoothing factor on mean audio delay

Figure 7. Impact of updare period on mean audio delay

IJCSNS International Journal of Computer Science and Network Security, VOL.6 No.12, December 2006

We can see that trade-off between higher goodput and

lower mean audio delay can achieve by selecting an
update period T update in the range of [5000, 8000].
In the following simulations, we set smoothing factor
and update period T update values to 0.6 and 6000 time

287

categories) are started at around 3.0 seconds with small

individual offsets to have accurate CDFs (Cumulative
Distribution function) of the delay. The simulation
duration is 18 seconds. In order to have confidence in
results obtained by simulations, we run 20 simulations and
obtained results are averaged on theses simulations.

4.1. Throughput

4. SIMULATION RESULTS
In order to evaluate the performance of the hybrid
adaptation of the CWmax and CWmin scheme, in this
section we inquiry the impact of traffic load and compare
it to the basic EDCA scheme. The different type of traffic
(associated with access categories) used for simulations
are described in Table 1. We simulate various loads of the
system by instantiating the simulation topology in figure 3
for different number of stations. All the stations are
located within the same independent basic service (IBSS),
so that, every station is able to detect the transmission
from any other station.
The following QoS metrics are used to evaluate the
performance of the proposed scheme:
Gain of goodput: This metric stands for the gain
(in %) on the average goodput (AG) of the proposed
scheme compared with basic EDCA:
Gain _ of _ goodput =

AG DCW min max AG EDCA

AG EDCA

Figure 8 shows the gain in goodput for hybrid adaptation

scheme over EDCA.
The throughput improves in hybrid adaptation scheme and
the gain of throughput for hybrid adaptation over EDCA is
up to 27%. Furthermore, the gain increases when traffic
load is greater than 10 stations as shown in figure 8. So,
according to system throughput hybrid adaptation
outperforms EDCA. The higher performance in
throughput for hybrid adaptation is due to the increase in
channel utilization because of the hybrid adaptation
algorithm.
hybrid adaptation

Gain of goodput (%)

slots.

20
18
16
14
12
10
8
6
4
2
0
5

10

13

15

17

20

25

30

35

40

45

num ber of stations

Mean delay: stands for the average delay of all the

flows that have the same priority in the different
stations. This metric is used to evaluate how well the
scheme can accommodate real-time flows. However,
real-time flows also require low average delay and
boundd delay and jitter.
Collision rate: represents the average number of
collisions that occurs per second. In WLAN collisions
increase the delay that station should wait, before
initiating a new transmission attempt and it causes
more delays.
Medium utilization(Mu): the medium utilization
represents the percentage of time used for the
transmission of data frames and it is given by:

Mu=

Totaltime Collisiontime Idletime

100%
Totaltime

For the different simulation scenarios used in this section,

all the traffic categories (associated with the access

Figure 8. Throughput for hybrid adaptation scheme over EDCA

4.2. Packet delay

In this subsection, we compare the average packet delay
under EDCA and our proposed scheme.
Figure 9 shows the mean audio delay as a function of
traffic load for both hybrid adaptation scheme and EDCA.
The mean audio delay improves significantly in hybrid
adaptation compared to EDCA. Indeed, hybrid adaptation
scheme maintains a lower audio delay than EDCA even in
low traffic load environment, i.e., for a number of stations
less than 10. As the load traffic increases, hybrid
adaptation is able to maintain a lower delay than EDCA.
The mean audio delay in hybrid adaptation scheme is up to
34% lower than in EDCA for a traffic load of 30 stations

IJCSNS International Journal of Computer Science and Network Security, VOL.6 No.12, December 2006

288

and 49% lower for a traffic load of 45 stations and results

in lower delay and jitter for high priority access categories.
EDCA

Hybrid adaptation

Mean audio delay (ms)

30
25
20
15

4.3. Medium utilization

Figure 11 shows the collision rate for hybrid adaptation
scheme and EDCA. The collision rate is the same for
hybrid adaptation and EDCA for a very low traffic load,
i.e., 5 stations. As the traffic increases, the collision rate in
hybrid adaptation, starting from a system load of 10
stations, maintains a lower increase than in EDCA. It can
be seen that, for 25 stations, the collision rate in hybrid
adaptation is 40% lower than in EDCA.

10
5

EDCA

Hybrid adaptation

0
10

15

20

25

30

35

40

Figure 9. Mean audio delay for hybrid adaptation scheme and EDCA

This gain in delay for hybrid adaptation scheme can be

explained by the hybrid adaptation algorithm that
performs better than static CWmax and CWmin values
and especially for medium and high loaded environment.
As it can be seen in figure 10, there is an improvement of
the mean video delay (medium priority traffic) in hybrid
adaptation compared to EDCA. Both EDCA and hybrid
adaptation have the same mean video delay when the
traffic load is low, i.e., less than 13 stations. However, the
delay improves in hybrid adaptation as the traffic load
increases. The video delay is 75% lower in hybrid
adaptation scheme than in EDCA for a system load of 35
stations. This can be explained by the hybrid adaptation
algorithm used to adjust the size of CWmax[i] and
CWmin[i] that performs better than a static ones in
medium and high loaded channel system.
EDCA

Hybrid adaptation

4000
3500
3000

1600
1400
1200
1000
800
600
400
200
0
5

10

15

20

25

30

35

40

45

Num ber of stations

Figure 11. Collision rate for hybrid adaptation scheme and EDCA

We believe that the hybrid adaptation has contributed to

reduce the number of collisions in the IBSS. Figure 12
shows the medium utilization for hybrid adaptation
scheme and EDCA. It can be seen that under most system
loads, hybrid adaptation scheme has much better channel
utilization than EDCA. The capacity in hybrid adaptation
is higher than EDCA (maximum channel utilization in
EDCA is reached for 16 station while in hybrid adaptation
corresponds to 27 stations). Therefore, the channel
capacity is 23% higher than in EDCA.This is, because
hybrid adaptation adjusts the size of CWmax [i] and
CWmin[i] according to the network conditions.

2500
2000

EDCA

Hybrid adaptation

1500
80

1000
500
0
5

10

15

20

25

30

35

40

45

Num ber of stations

Figure 10. Mean video delay for hybrid adaptation scheme and EDCA

Medium utilization (%)

Mean video delay (ms)

1800

45

Num ber of Stations

Collision rate (Collisions/s)

70
60
50
40
30
20
10
0
5

10

15

20

25

30

Num ber of stations

35

40

45

IJCSNS International Journal of Computer Science and Network Security, VOL.6 No.12, December 2006

Figure 12. Medium utilization delay for hybrid adaptation scheme and
EDCA

From the simulation results, we can conclude that hybrid

adaptation scheme outperforms EDCA scheme in light,
medium and high system load.
The total throughput increases by up to 27% for high
priority traffic and remain stable for medium and low
priority access categories. Furthermore, the delay for high
priority access category reduces more than 49%. Moreover,
the channel capacity improves and is 23% higher than in
EDCA.

5. CONCLUSIONS
In this paper we have proposed a new dynamic scheme for
the hybrid adaptation of the contention window maximum
(CWmax) and the contention window minimum (CWmin)
in order to enhance the service differentiation for 802.11
WLANs. We have extended the basic EDCA scheme by
an algorithm that enables each station to adjust the size of
the CWmax and CWmin used in its back-off algorithm at
run time.
The adjusting is differentiated for each access category i
and performed according to the channel traffic conditions.
Simulation results demonstrated that our scheme achieves
better performance of throughput, delay and jitter than
basic EDCA, specially for high priority traffic.
Results are validated by analyzing the impact of sources
and network dynamics on the performance metrics and
compared with the basic 802.11e EDCA. On one hand,
results have shown that audio delay associated with high
priority access category, improves greatly and decreases
by up to 49%. Furthermore, the channel capacity improves
and is 23% higher than in EDCA. Moreover, we have
observed the total throughput increases by up to 27% for
high priority traffic and remain stable for medium and low
priority access categories.
Further work could include implementation of the
proposed hybrid adaptation algorithm for infrastructure
networks.

References
[1] Aad, Imad and Castellucia,C. "Differentiation mechanisms
for IEEE 802.11", In Proceedings of IEEE Infocom 2001, A,
Alaska, April 2001.
[2] ANSI/IEEE Standard 802.1D, 1998 ed., "IEEE standard for
information
technology-Telecommunications
and
information exchange between systems-Local and
metropolitan area networks-Common specifications. Part 3:
media access control (MAC) bridges", Institute of Electrical
and Electronics Engineers, 1998.

289

[3] Cali, F., Conti, M. and Gregori, E. "IEEE 802.11 Protocol:

Design and Performance Evaluation of an Adaptive Backoff
Mechanism", In IEEE JSAC, vol. 18, no. 9, September 2000.
[4] Deng, D-J. and Chang, R-S. "A priority scheme for IEEE
802.11 DCF access method", IEICE Transactions on
Communications, E82-B(1), January 1999.
[5] Gannoune, L. and Robert, S. "Dynamic Tuning of the
Contention Window Minimum (CWmin) for Enhanced
Service Differentiation in IEEE 802.11 Wireless Ad-Hoc
Networks", To appear in IEEE PIMRC'04
[6] Gannoune, L. and Robert, S. "Results on Dynamic
Adaptation of the Contention Window Maximum (CWmax)
for Enhanced Service Differentiation in IEEE 802.11 AdHoc Networks", Accepted for publication in IEEE VTC Fall
2004
[7] IEEE WG, "Draft Supplement to Standard for
Telecommunications and Information Exchange Between
Systems-LAN/MAN specific requirements -Part 11:
Wireless Medium Access Control (MAC) and Physical
Layer specifications: Medium Access Control (MAC)
Enhancements for Quality of Service (QoS)", IEEE
802.11e/Draft D5.0, July 2003.
[8] IEEE WG home page: http://grouper.ieee.org/groups/802/11
[9] IEEE WG, "Part 11: Wireless LAN Medium Access Control
(MAC) and Physical Layer (PHY) Specifications, IEEE
802.11 Standard", 1999.
[10] Lindgren, A. "Evaluation of Quality of Service of Service
Schemes for IEEE 802.11", In Proc. Of IEEELCN,
Nonember 2001.
[11] Mangold, S., Choi, S., May, P., Klein, O., Hiertz, G. and
Stibor, L. "IEEE 802.11e Wireless LAN for Quality of
Service", in Proc. European Wireless '02, Florence, Italy,
February 2002.
[12] Ni, Q. "QoS issues and enhancements for IEEE 802.11
Wireless LAN", INRIA Research Report No.4612,
November 2002.
[13] Ns-2
network
simulator
homepage:
http://www.isi.edu.edu/nsnam/ns
[14] Romdhani, L., Ni, Q., and Turletti, T., "Adaptive EDCF:
Enhanced Service Differentiation for IEEE 802.11 Wireless
Ad-Hoc Networks", in the Proceedings of the IEEE
WCNC03.
[15] Vaidya, N.H., Bahl, P. and Gupta, S. "Distributed fair
scheduling in wireless LAN. In Sixth Annual International
Conference on Mobile Computing and Networking", Boston,
August 2000.
[16] Veres, A., Campbell, A.T., Barry, M. and Sun, L-H.
"Supporting service differentiation in wireless packet
networks using distributed control", IEEE Journal of
Selected Areas in Communications (JSAC), Special Issue
on Mobility and Resource Management in Next-Generation
Wireless Systems, vol. 19, No. 10, pp. 2094-2104, October
2001.

290

IJCSNS International Journal of Computer Science and Network Security, VOL.6 No.12, December 2006

Hassan Takabi received his B.Sc.

degree in Software Engineering
from AmirKabir University of
Technology (Tehran Polytechnic),
Tehran, Iran in 2004. He is
currently an M.Sc. student in the
Department
of
Computer
Engineering, Sharif University of
Technology, Tehran, Iran. He also
works as a research assistant in
Sharif Network Security Center from 2005. His main research
interests include Information and Network Security; Computer,
and Communication Networks; and Wireless, Adhoc and Sensor
Networks.

Ali H. Moghadam is currently an

M.S. student in the Software
Quality Research Lab. in Sharif
University of Technology, Theran,
Iran. He recieved his B.S. degree
in Software Engineering from
AmirKabir
University
of
Technology (Tehran Polytechnic),,
Tehran, Iran. His research
interests are Software Reliability
Modeling, Software Engineering,

and Wireless Networks.

Ahmad Khonsari received

the BSc degree in electrical
and computer engineering
from
Shahid-Beheshti
University, Iran, in 1991, and
MSc degree in computer
engineering from the Iran
University of Science and
Technology, Iran, in 1996 and
PhD degree in computer
science from the University of
Glasgow, UK, in 2003. He is
currently an assistant professor in the Department of
Electrical and Computer Engineering, University of
Tehran, Iran and a researcher in School of Computer
Science, Institute for Studies in Theoretical Physics and
Mathematics (I.P.M.), Iran. His research interests are
performance modelling/evaluation, mobile and ubiquitous
computing, communication networks and distributed
systems, and high performance computer architecture.