See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3417306

Comments on IEEE 802.11 saturation throughput analysis with freezing of

backoff counters

Article in IEEE Communications Letters · March 2005

DOI: 10.1109/LCOMM.2005.02008 · Source: IEEE Xplore

CITATIONS READS
144 197

2 authors, including:

Chuan Heng Foh

University of Surrey
213 PUBLICATIONS 5,609 CITATIONS

SEE PROFILE

All content following this page was uploaded by Chuan Heng Foh on 14 February 2014.

The user has requested enhancement of the downloaded file.

130 IEEE COMMUNICATIONS LETTERS, VOL. 9, NO. 2, FEBRUARY 2005

Comments on IEEE 802.11 Saturation Throughput

Analysis with Freezing of Backoff Counters
Chuan Heng Foh, Member, IEEE, and Juki Wirawan Tantra, Student Member, IEEE

Abstract— This letter presents an accurate model for the Backoff for station A = 6

performance analysis of the IEEE 802.11 saturation throughput

with freezing of the backoff counter. The model corrects the DIFS DIFS DIFS
existing model presented by Ziouva and Antonakopoulos which station A station B station A
assumes that the channel access probability and station collision
6 5 4 3 3 2 1 0
probability are independent of channel status. Simulation results
show the accuracy of the new model.
Fig. 1. IEEE 802.11 basic access method.
Index Terms— Wireless local area networks, performance anal-
ysis, saturation throughput analysis.
II. S ATURATION T HROUGHPUT A NALYSIS
I. I NTRODUCTION The mechanism of the IEEE 802.11 Distributed Coordi-
nation Function (DCF) with basic access method is shown
T HE bi-dimensional Markov Chain modeling introduced
by Bianchi [1] for the analysis of the IEEE 802.11
saturation throughput has become a common method to study
in Fig. 1. It differs from the model presented in [1] in the
decrement of the backoff counter. The IEEE 802.11 standard
[2] specifies that a station freezes its backoff counter when it
the performance of the IEEE 802.11 Medium Control Access
detects a transmission on the channel (note that the backoff
(MAC) protocol [2] and its enhancements. The model was
counter is not decreased during the channel busy period).
later refined to capture further details of the IEEE 802.11
This backoff freezing procedure directly affects the probability
protocol operations. Among the refinements, one is due to
that a station accesses the channel, and this probability also
Ziouva and Antonakopoulos [3] aiming to capture the freezing
depends on whether the previous period is busy or idle.
of backoff counters when the broadcast channel is sensed busy
To understand this, we first consider the channel access
by a station. Precisely, when a channel turns idle from busy
event after a busy period due to a collision. After a collision,
due to, for example, a Distributed InterFrame Space (DIFS),
since stations that did not participate in this collision had
Bianchi’s model assumes that each station immediately reac-
frozen their backoff counters, they will not access the channel
tivates and decrements its counter, whereas the IEEE 802.11
after the busy period; only those suffered a collision may
standard specifies that a backoff counter is decremented only
access the channel if their newly chosen backoff counter is
after the channel continues to remain idle for a predefined slot
zero. Hence, only a group of stations rather than all stations
time. The refinement reported in [3] was, however, introduced
may access the channel after a busy period. In case of a
without realizing that the two key probabilities governing the
successful transmission, only one station, which performed
performance, namely the channel access probability, τ , and
the successful transmission, may access the channel after the
the station collision probability, p, depend on the channel
successful transmission period. As opposed to the case of a
status. This inaccuracy in the model affects several important
busy period, after an idle period, all stations whose backoff
measures including the probabilities that a particular time
counters are decremented to zero will access the channel;
period on the broadcast channel is an idle, a successful
hence, it is obvious that the channel access probability actually
transmission, or a transmission collision period. Owing to the
depends on whether the previous period is idle or busy.
large difference in the duration of the different types of time
This is not modeled in [3] where the derived channel access
periods, the inaccuracy does not reflect significantly in the final
probability is not conditioned upon the type of the previous
saturation throughput results in most common cases; however,
time period.
fundamentally, the model is in error.
The new model describing the backoff process of a station
In this letter, we present a new model correcting that of [3]
for the saturation network condition is presented in Fig. 2.
by evaluating the channel access probabilities and the station
Here we consider a network of n saturated stations. The state
collision probabilities conditioned upon the channel status. We
{i, j, k} represents the state of a station at a particular time
then show the accuracy of our results via computer simulation,
period, where i indicates the type of the previous period, either
and demonstrate the errors if such details are ignored.
idle or busy (i = 0 or i = 1 respectively); j indicates the
Manuscript received June 4, 2004. The associate editor coordinating the current backoff stage (j = 0, 1, . . . , m); and k indicates the
review of this letter and approving it for publication was Prof. Jinwoo Choe. current backoff counter (k = 0, 1, . . . , Wj −1). Two important
The authors are with the School of Computer Engineering, Nanyang protocol parameters describing the backoff process are the
Technological University, Singapore (e-mail: [email protected],
[email protected]). minimum backoff window denoted by W0 , and the maximum
Digital Object Identifier 10.1109/LCOMM.2005.02008. backoff stage denoted by m. The backoff window of a station
1089-7798/05$20.00

c 2005 IEEE
FOH and TANTRA: COMMENTS ON IEEE 802.11 SATURATION THROUGHPUT ANALYSIS WITH FREEZING OF BACKOFF COUNTERS 131

1/W0 1/W0 1/W0 … 1/W0 1/W0

1 1-p1 1-p1 1-p1 1-p1
1,0,0 1,0,1 1,0,2 1,0,W0-2 1,0,W0-1
1-p1 p1 p1
p0
1-p0 1-p0 1-p0 1-p0 …
0,0,0 0,0,1 0,0,2 1-p0 0,0,W0-2
p0
1/W1
1/W1 1/W1 1/W1 … 1/W1
1-p1 1-p1 1-p1 1-p1 1-p1
1,1,0 1,1,1 1,1,2 1,1,W1-2 1,1,W1-1
p1 p1 p1
1-p1 p0
1-p0 1-p0 1-p0 1-p0 …
0,1,0 0,1,1 0,1,2 1-p0 0,1,W1-2
p0
…
1/W2
…

p1 p0
1/Wm
p1 1/Wm 1/Wm 1/Wm … 1/Wm
1-p1 1-p1 1-p1 1-p1 1-p1
1,m,0 1,m,1 1,m,2 1,m,Wm-2 1,m,Wm-1
p1 p0 p1
p0 1-p1
1-p0 1-p0 1-p0 1-p0 …
0,m,0 0,m,1 0,m,2 1-p0 0,m,Wm-2
s1 a1
S1 s1 a1 A1 S1 s1 a2 A1
represents
S2 s2 a2 s2 a1
A2 S2 A2
s2 a2

Fig. 2. The Markov chain representation of the new model.

at the j-th backoff stage is Wj , or precisely 2j W0 . where

⎧
Two key probabilities governing the backoff process are first ⎪
⎪ 1 if j = 0,
defined, they are, p0 (p1 ), the probability that, from a station’s ⎪
⎪
⎪
⎪ p0 (W0 − 1)
⎪
⎪ if j = 1,
point of view, at least one of the other stations transmit during ⎪
⎨ W1
a slot after an idle (a busy) period. We further define q0 (q1 ) ψj = p0 (W0 − 1)
to be the probability that the broadcast channel remains idle ⎪
⎪ πj if j = 2, 3, . . . , m − 1,
⎪
⎪ W1
after an idle (a busy) period. ⎪
⎪
⎪
⎪ p0 (W0 − 1) Wm
Given that the channel is either an idle or a busy period, let ⎪
⎩ πj if j = m,
W1 Wm − p1 − p0 (Wm − 1)
Pi be the probability that a particular period on the channel
is idle, then Pi = q0 Pi + q1 (1 − Pi ), which gives and
j  
p1 p0
πj = + (Wx−1 − 1)
q1 x=2
Wx Wx
Pi = . (1)
1 − q0 + q1 with ⎡ ⎤
m Wj −2 Wj −1
  
⎣ b0,j,k + b1,j,k ⎦ = 1.
Let bi,j,k be the stationary distribution of the described j=0 k=0 k=0
Markov Chain. Each of the state stationary probability can
be expressed in terms of b1,0,0 as Define τi and τb to be the probabilities that a station
accesses the broadcast channel after an idle and a busy period
respectively. Similar to [3] and [1], each of τi and τb can be
b1,j,0 = ψj b1,0,0 for j = 1, 2, . . . , m, expressed as a function of the stationary probabilities. They
1 + p0 (Wj − 1 − k) are given by
b1,j,k = ψj b1,0,0 m
1 − p1 j=0 b0,j,0
(2) τi = q1 ,
for k = 1, 2, . . . , Wj − 1 and j = 0, 1, . . . , m, 1−q0 +q1
m (3)
b0,j,k = (Wj − 1 − k)ψj b1,0,0 j=0 b1,j,0
τb = .
for k = 0, 1, . . . , Wj − 2 and j = 0, 1, . . . , m, 1 − 1−qq01+q1
132 IEEE COMMUNICATIONS LETTERS, VOL. 9, NO. 2, FEBRUARY 2005

1
The proposed model is the average time that the channel is sensed busy due to
Modified Ziouva’s model
0.9 IEEE 802.11 Simulation a collision. These quantities for the basic and the RTS-CTS
0.8 access methods are given by (14) and (17) in [1].
0.7 The probabilities Pi , Ps and Pc describe the probabilities
that a particular period on the channel carries no transmission
0.6 Pi
(idle), a successful data frame transmission, and two or more
Probability

0.5 transmissions (collision), respectively. Probability Pi is given

0.4 by (1) and probabilities Ps and Pc can be expressed as
0.3 Ps Ps = nτi (1 − τi )n−1 Pi + nτb (1 − τb )n−1 (1 − Pi ),
(7)
0.2 Pc = 1 − P i − P s .
0.1 Pc
III. N UMERICAL R ESULTS
0
5 10 15 20 25 30 35 40 45 50 In Figs. 3 and 4, numerical results for the saturation
Number of Stations
throughput along with several important probabilities obtained
Fig. 3. Probabilities of an idle, a successful transmission, and a collision of from our model are plotted and compared with that of [3]1 . We
various models for W = 16 and m = 6. use the same protocol parameters as [1] for this comparison.
0.85
The numerical results are obtained using the fixed point
(0)
iteration technique [5]. In brief, initial guessing for τi and
(0)
RTS-CTS τb were first made, then p0 and p1 were computed by
0.8
(5), and later used to calculate bi,j,k using (2). After that,
new values for τi∗ and τb∗ were obtained by (3) and (4).
Saturation Throughput

0.75 (1) (1)

We finally updated τi and τb for the next iteration by
(1) (0) (1) (0)
τi = 0.5τi + 0.5τi and τb = 0.5τb + 0.5τb∗ . The
∗
0.7
simulation results (shown with symbols) are obtained with
a 95% confidence interval lower than 0.001. As in [1], our
0.65
model assumes that after the completion of a transmission,
basic each station detects a collision after a DIFS period. This
0.6 makes the collided stations synchronized to other stations. This
The proposed model
Modified Ziouva’s model assumption is relaxed in the simulation, where we consider
IEEE 802.11 simulation
0.55 each station reveals a collision after the acknowledgment
5 10 15 20 25 30
Number of Stations
35 40 45 50
timeout expires, which is 0.3 ms.
As can be seen in Figs. 3 and 4, our model gives accurate
Fig. 4. Saturation throughput of various models for W = 16 and m = 6. results for the performance of IEEE 802.11 with freezing of
backoff counters. For the model of [3], only small errors
in saturation throughput are produced for the case of the
Having obtained τi and τb , q0 (q1 ) can be determined based RTS-CTS method as data frame transmission duration is
on the fact that the broadcast channel remains idle after an significantly longer than other time periods, where the model
idle (a busy) period if no station accesses the channel. Thus of [3] appears to be accurate if not carefully studied.
we get
q0 = (1 − τi )n ,
(4) R EFERENCES
q1 = (1 − τb )n .
[1] G. Bianchi, “Performance analysis of the IEEE 802.11 distributed coor-
A particular station finds a slot to be busy if at least one of dination function,” IEEE J. Select. Areas Commun., vol. 18, pp. 535-547,
Mar. 2000.
the other stations access the channel. Hence we have [2] Wireless LAN Medium Access Control (MAC) and Physical Layer Spec-
p0 = 1 − (1 − τi )n−1 , ification, IEEE Std. 802.11, 1997.
(5) [3] E. Ziouva and T. Antonakopoulos, “CSMA/CA performance under high
p1 = 1 − (1 − τb )n−1 . traffic conditions: throughput and delay analysis,” Computer Communi-
cations, vol. 25, pp. 313-321, Feb. 2002.
The system throughput S, the fraction of time used for [4] Y. Xiao, “Performance analysis of IEEE 802.11e EDCF under saturation
successful payload transmission, can be expressed as condition,” Proc. ICC, Paris, France, June 2004.
[5] L. V. Fausett, Numerical Methods Algorithm and Applications. Engle-
Ps E[P ] wood Cliffs, NJ: Prentice Hall, 2003.
S= , (6)
Pi σ + P s T s + Pc T c
1 Two modifications were applied to the original model, which are (i) the
where E[P ] is the average payload length, σ is the duration special state, {−1, 0}, modeled in [3] is said to be inconsistence with the
of an empty slot time, Ts is the average time that the channel standard [4], it was removed; and (ii) equation (8) in [3] is corrected as
is sensed busy because of a successful transmission, and Tc follows: pb = 1 − (1 − τ )n−1 .

View publication stats