IEEE b0Z.

11b Performance Evaluation: Convergence of Theoreti

cal, Simulation and Experimental Results
Rosario G Garroppo, Stefano Giordano, Stefano Lucetti, Franco Russo
Dept. oflnformation Engineering, Universit of Pisa, Via Caruso I -1-56122, Pisa, ITALY
Ab8rC
The paper presents the analysis of the eval uation of te maximum goodput achievable in a IEEE 802, lib Basic
Set Service (BSS) performed by means of theoretical, simulative and experimental approaches. With respect to
previous works on this topic, the presented study focuses on the comparison of the results obtained using the dif
ferent approaches, highlighting their convergence in the simple case of i deal radio channel. The assumption of an
ideal channel, although not always verified in real scenarios, is usefl to evaluate the upper bound of the network
capacity ofered by this technology in absence of losses and performance degradations due to factors not directly
related to the protocol. Furthermore, the adopted simulation tool has been preventively validated versus theoreti
cal results under varing working conditions and for diferent physical layers. It is noteworthy that the results ob
tained by each one of the three approaches are in accordance with those already presented in literature, although
they are derived by means of different tools, thus further confirming the accuracy of such result
1 1uIludBcl0u
In recent years, the design of wireless networks for
local area communications has focused the attention of
standardization bodies and network device manufac
turers. Among the diferent available technologies, the
IEEE 802.11 b standard [1 )[2] is the one that has
gained popularity and has been widely deployed in
enterprises mostly to extend the connectivity, already
provided by the wired infrastrcture, to mobile and
portable devices, such as Laptops, PDAs, etc .. The
wide success of thi s technology has driven the creation
of new IEEE Task Groups and the design of new stan
dards; in paricular the Tak Group 802.11a [3] and
802.llg [4] and recently te 80211n [5], are devoted
to provide higher maximum data rates. Although the
study group 802.11n has stared its works for the def
nition of a 100 Mbps data rate techno logy, and some
devices based on 802.lIa and 802.11 g standard at 54
Mbps are available, it is likely that 802.lIb deploY
ments will continue to operate in both enterprises as
well as the residential market for the next few years.
The fndamental mechanism used by the 802.11 pro
tocol to access the medium is called Distributed Coor
dination Function (DCF). For an overview on the
802.11 network architecture and on detailed protocol
principle of operati on, the reader can refer to [6].
In this paper, we investigate the maximum goodput
achievable by the basic access under the DCF scheme,
thus not considering the RTS/CTS mechanism. The
original contribution of our work is to present the joint
results obtained by meanS of theoretical, simulative
and experimental approaches, comparing the obtained
405
values under different working conditions and thus
validating experimentally the adopted simulative and
analyi cal models. As far as the theoretical approach is
concered, we extensively detail the analysis related
to the time-based evaluation of busy and idle periods
of the basic access scheme, derived from the 802.11
protocol description (IJ, also introduced in [7]. Such
analysis will be used to derive a general simple rela
tionship between the UDP payload size and the maxi
mum achievable goodput, for each of the four possible
data rates available in 802.11 b networks, i.e. I, 2, 5.5
and II Mbps. Simulation results on the 8
0
2.11 b per
formance have been already obtained under various
working conditions in [8] using the NS2 Network
Simulator, as well a in [9J by meanS f an ad hoc
simulation tool. In this paper, we present the results of
the performance analysis of 802.11 b carried out using
the OPNET Modeler network simulation environment
[10], which provides a comprehensive set of tools and
an i ntegrated model library to design, develop and
analyze any sort of communication network. In par
ticular, we validate the OPNET simulation model ver
sus the theoretical results .obtained in [6], which pre
sents a deep theoretical analysis referred to the FHSS
and the DSSS physical layers at I Mbps, confrmed by
means of a custom simulation tool. Afer the valida
tion of the simulation tool, we estimate the maximum
goodput obtainable with the UDP trafc varing the
UDP payload size, referring to the DSSS physical
layer at II Mbps. All the simulation and analytical re
sults are then validated by means of measurements
carried out on an experimental test-bed based on two
diferent 802.11 b NICs, conducing to results analo-
gous to those obtained in similar experimental analy
si s, such as those presented in [I I ) and [12). All the
analyses have been carried out in the assumption of
ideal channel conditions. In the experimental analysis,
this condition was obtained situating the network
nodes in the same rom, with no physical obstacles
between them, verifing that no other station was
transmitting on interfering channels, and constantly
monitoring the driver supplied lin quality indicator
between the two nodes.
Tdata
Sender
SIFS
Recier
Trp
Fig. I Transmission of a DATA frame
011.
@11:/5S/1tM.
Fig.2 Overhead due to protocol encapsulations
Table I: IEEE 802.lIb DSSS phys!callayer paramt.rs
value
1
Parmeter Vaue No!es
SlotTime 20 v
SIS 10 1"
D1FS 501" (SIFS+ 2 Slotme)
CWmin JI (rult in 620 Ic)
CWmo 1023 (r,ullS in 20.46 msc)
1h8ur8tCal1ntatnuI
101.11 Ubb acty
The frst analysis is focused on the deterination of a
theoretical maximum goodput achievable by a single
station generating UDP trafc. To this aim, we present
Fig. I, which shows the timings related to the trans
mission of a single frame over the air interface, fom a
sender to a receiver, followed by the imediate ac
knowledgemenl (ACK) frame. Since the indicated
Data feld contains a number of headers (from the
UDP down to the Physical layer), it is straightforard
to understand that only a portion of the dat rate is de
voted to the transmission of "user-level" data.
The values of the parameters are specifc for te un
derlying physical layer; Table I co. lects the values for
the DSSS physical layer in IEEE 802.11 b. The Back
of Time is a random period, equal to R times the
SlotTime, where R is a random variable drawn from a
uniform distribution over the interal [O,CW], and
CW stands for Contention Window. The value CW is
an integer which can assume a finite set of values in
the range CWminCWCWmax; CWmin and
CWmax are defined by the standard for each physical
layer, and their values for DSSS are repored in Table
I. Initially CW is set equal to CWmin, and it assumes
the values 2n_1 (with n which ranges from 5 to 10)
each time there is an unsuccessfl event of the DATA
frame transmission, identified by the absence of an
ACK within a given timeout. To evaluate the upper
bound of the capacity of the IEEE 802.11b system, we
assume to have only one Station (STA) tansmitting.
This hyothesis implies that there are no collision
events and, hence, CW is equal to CWmin throughout
the transmission of the whole series of data fames.
The mean value of the Backof Time is then equal to
SlotTime*CWmin2, since R is uniforly distributed.
Fig. I leads to the following efciency rate defnition:
= , Tw ,
CW
TDm + SIFS + T + DIFS + _s/-.n-
(I)
To obtain the eficiency as it is seen by the user, it is
necessary to compute T DATA
including all the overhead
related to the various headers that are carried with the
user data (see Fig. 2). In absence of fagmentation
during the layers crossings, we have that for each
block of user-level bit to be transmitted, it is necessar
to spend time to transmit also: the UDP header, the IP
header, the LLC header, the MAC header and the
Frame Check Sequence feld and lastly, the Physical
Header PLCP, constituted by a Preamble and a
Header, which sum to 24 Bytes and are tansmitted at
a data rate equal to I Mbps, whereas all the other in
formation are transmitted at the data rate selected by
the user (I, 2, 5.5 or II Mbps). Supposing to have a
NIC data rate of II Mbps, T DATA results equal to:
UserDala _
To,,, . T + T. --1-) -+ 192(PLCP)+
64(UD} + 1601P) + 64(LLC+224(MC_
sec
II
(2)
Transmitting one application block of inforation
would incur in a fixed overhead due to the only encap
sulation equal to 46.5 lec (for the transmission of te
11 Mbps data) plus 192 IJsec (for the I Mbps PLCP
transmission). The impact of such overhead is obvi
ously greater as the size of the UserData block re
duces. The last value in Eq. I which has to be deter
mined is T'CK' An ACK frame is 14 Bytes long, trans
mitted at I or 2 Mbps (the lowest beteen these
va.ues and the NIC data rate), pl us a PLCP physical
header equal (in size and transmission rate) to the one
of the Data frame. Its overall transmission time T
ACK
is then equal to 56+ 192 lec, supposing to have a NIC
transmission rate of 2Mbps or higher. Substituting all
the va.ues in Eq. I, the resulting eficiency rate as
sumes the expression (in the I I Mbps case):
40
UUiDalabifj
II
-.+ 46.S + .,+ 10 + (56+ 192)+ SO + J 1 0
UwrDalafbil] UserDaIByle]
U$eIDatabItJ+ 9421.5;:Use,OIaByteJ+ t 178
(3)
Fig. 3 Theoretcal goodput vs Payload Sie for the 4
diferent data rat ..
Table II: KPR for the difereDt S02.lIb data rat .
,
Dt mlc DR [Mps] 2 5,5 I I
114 214 589 1178
The maximum godput achievable at the user lever is
then equal to the NIC data rt (II Mbps in our case)
multiplied by the efciency rate. Fig. 3 shows the be
havior of the goodput for the four available data rates
as a function of the application payload size, fom 0 to
the maximUm allowed in 802.llb networks, which is
2312 Bytes (Maximum MAC Payload) minus 36
Bytes of UDP+IP+LLC headers, i.e. 2276 Byte's. Ie
general expression of the achievable goodput as a
fnction of the NIC Dat Rate (DR) is reported in Eq.
4, where the parameters KDR are collected in Table II.
cooao(bkDR,
UserD",tByl.j
(4)
UserDalt By/e] + K
It is noteworthy that the main cause of the reduction of
the goodput with respect t the nominal data rate is
onstituted by the PLCP physical preamble and
header, which ae always tnsmited at the lowest
dat rate (I.Mbps), and hence is particularly relevant
when the tansmission of the data occus at the higher
data rates. To reduce this degradation, the amendment
supplement b [2J to the 802.11 standard has intro
duced, as optionalg the concept of Short PLCP, which
halves the time needed for such header= Furtherore,
in a infastrctred network, a last cause of perfor
ance reduction is the channel occupation due to bea
cons transmission by the A. Usually the inter
repetition time between to beacons is about 100
msec, so the performance degradation introduced by
them is limited.
XcuSVcY8d8uu uhc
bmu8luu1uul
The perforance evaluation by means of the simula
tion approach is based on the 802.11 b model devel
oped in te OPNET Modeler simulation tool [IOJ. We
carried out an extensive validation of the IEEE
407
S02.lIb model implemented in OPNET Modeler, be
fore it use. The goal of the validation of a simulation
tool is of paramount importance, since it allows to
substitute part of the experimental analysis wt simu
lation one, typically less expensive and easier to set
up. Furthermore. a simulative approach allows t per
for "What-if' analyses without te need of efec
tively deploying the new fnctionalities.
The IEEE 802.lIb simUlation model available as a
standard library in OPNET Modeler is almost com
pletely compliant wt te IEEE specifcations [1][2].
We frst validate te simulation tool considering the
saturation throughput analyially evaluated at MAC
layer on the 802.11 Markovian model developed in [6]
as the reference result; aCCOrding to it, the s311ration
throughput of the BSS can be obtained as a fnction of
the number n of active stations in the BSS
the probability r that a transmission occurs in a ran
dom time slot
the parameter W equal to I +CWmin
the maximum backof stage m, defned by the fol
lowing relation: CWma 2m'(I+CWmin)-l
physical layer dependant constants, such as the time
slot and the propagation delay
Note that r is a function of the probabilit p that a
packet collides (assumptions on the independence of
such collisions have been raised in [6]) which, in tm,
is a fnction of .itself.
It is terefore necessa to solve the nonlinear system
jp ^ 1 (l " 1)"-'
1 ....
|"I
I'
for all the relevant couples of values W m.,
In order to compare the satumtion throughput evalu
ated using the teoretical ad simulative approaches,
we have set up our discrete events simulations accord
ing to the corresponding physical layer asumptons
made in [6]. In particular, Fig. 4 contains the compari
son of the saturatiorthroughput at.MAC layer for a
sigle BSS composed by a variable nuber of stations
transmitting IP packets of fxed size equal to 1023
bytes in the case ofa I Mbps FHSS physical layer. The
abscissa axis represents, on a 10& scale, te value of
the parameter W, which is directly related to the initial
value of the contenton window CWmin, whereas the
maimum backoffstage m has been fxed equal to 6.
The values of the saturation throughput obtained, by
simulation are ver close to those obtained by the
theoretical stdy; indeed, the maimum diference be
tween the two values is, below 3%, also for a high
number of contending stations, i.e. 50. The exposed
results highlight the goodness of the simulation model
accuracy, particularly with reference to the dynamics
of the MAC moel and specifically of the backof
procedure, given the variable number of contending
sttions.
r-. ----------
BW
0
&
8#
...
.:: :

i
3 4 6

7 $
-
Fig. 4 Comarison oh
/
mulated and aaaJ)tical satra-
tin througbput for a single BSS (1M bps FSS PH.
u---- -
BW --

i ~ L ~- + -
1W W Q 1& W
fW
Fig. 5 Comparison of simulated and aoalytkalsatura
tln tbrugput for a single BS versus packet si
(IMbp. DSSPH
UD,naw
I,PSIu, . ij&d%"""
---

AI Sl.lIb
ST A 11.l1b
Fig. 6 Refereact Senario ror tbe EIprimenta
Meauremenl
Not being possible to show all the resut obtained
during our reseach work, we present only one other
fige t stss the good correspondence between the
simulations r wt OPNET and !he expected !eo
retical results.
Fig. 5 collects the MAC layer satuation tughput
values of a BSS populated by 10 or 50 active stations
with a I Mbps DSSS physical layer (also in tis casc,
the physical layer has been chosen to perfonn a coher
ent comparson with the results prescnted in [6]) plot
ted versus the MAC payload size. The figure high
light a good corspondence between the to values,
with an error limited within 3% for the case of 10 ac
tive stations, and abut 4 to 8% in the 50 STA cae. As
a conclusion, the presented case studies, as well as
others not doumented here, highlight that te accu
racy of the results achievable using OPNET Modeler
simulation tool is adequte t adopt it as a tool to in
vestigate performance evaluation of 802.lIb wireless
access networks.
408
4 XQ6Om60f808$5
The last anlysis presented is based on the experimen
t approach. To perfonn the analysis, we set up an
experimentl testbed, depicted in Fig. 6, to verif te
corespondence between the teoretical goodput
achievable by a single UDP soure and the measured
one. The testbed is compsed by

a Lucent Orinoo A
)ODD Access Point a Linux baed Pentum II class
desktop PC which acts a a UDP sink, and a Linux
baed Pentiumll class laptop which acts a te UDP
Source. We chose MGEN [13} as the tafc genera
tion application, because of its wide confgurbility, in
ter of te types of statistcal properties of the gen
erated tafc (stochastic inter-ee times and
packet size). The A, connected directly t the UDP
sink, has an auto sensing 1011 OMbps auto duplex
Etheret interfce. The UDP sin is equipped with an
analogue NrC, hence t t cards autoconfgured for
a 100Mbps fll duplex connection.
4.1 Testng of trafc generator and meas
uring 1001
The first set of measurements was aimed at verifing
whether the imposed generation rate was actually pro
duced by MGEN, and at evaluatng the conditon
where the meaured rate was accurate with te equip
ments (mainly PC ad NIC) adopted in the experimen
tl scenario. To this aim, a simple pOint to point wired
conection between the UDP source a a UDP sink
has ben set up, and the packets originte at the
tansmittig applicaton have been dumped at the dat
lin layer. At the other extemity of te link, the re
ceived packet were duped at both the applicaton
and the data link layer, recording their timestmps.
The ue of a wired point-to-point connection permits
to evaluate the perfonnance of the used tools, without
the influence of the WLAN components. In order to
verif any limitaton of the generation rate due to the
application itself and the adopted devices, a piecewise
step fnction rate has been configd for generation,
raising the arrival rate of constant-sied packets at
fxed time interals. When considering large packets
(e_g. 612 Byes) we have not obsered any diference
beteen the aoWt of trafc set at generator side aod
that obsered at the receiving side until a tafic rate of
the order of the data rate of the IEEE 802-1lb tech
nology. The situation becomes ver diferent when us
ing ver small packets.
We have repeated the same test adopting the smallest
possible packet size generable by MGEN, i.e. 28
bytes, setting a nominal tansmitd dat rate which
increases of 250 kbps ever 30 seconds_ The result of
te meaurement are collected in Fig. 7; the highest
curve is the transmission rate, as it is measured at the
Data Link level of the source host whereas the other
two represent the rate measured at the receiver.
I

I
|
Fig. 7 Tansmission rat e of an UDP fow generated by
MGEN over a wired link (packet payload:28 bytes)
l
..
!a111M1mt
"t,"
--~ -, . . -"-- _ , : .. .
l
..~.-1 ....
;
.... .. - . .. -
... _.) . . ,...,,,, ... :,_, . ....- :.u,_
.`- : .. ,'-.
. . . . ~ +. ..... . :. .
.
.. J. ...L .
._... . . L ..
,
(" . .+ -
u....J ,
,
TV*I
. t=.

-..' 1 .
- .. ., , ". - , => -. . ..
Fig. 8 Comparison of tbe goodput obtained witb a
Orinoco Gold (above) and aD-Link DWL6S0 (below)
(payload=64 byt e. per packet)
Table Il: Measured Goodputs
UDP Payload
Siz [Bye]
6
576
1472
Onoo Gold
NIC[Mbpsl
0.569
3.55
6.06
DW50
NC [Mbpsl
0.59
3.7
6.2
Ihe lowest curve is obtined fom the measurements
petfmed at the Data Link level; lastly, the middle
one derives from the dump of the received packets at
the application level. It is straightforward to note that
the Data Link capture sufers of a severe limitation,
highlighted also by the number of discarded (i.e. not
dumped) frames reponed by the packet capture li
brry, i.e. libpcap. This may be due to the very high
packet rate necessar to generate a data rate of over I
Mbps using 28 byes long packets (note that each 250
kbps hig step corresponds to 1116 packets per sec
ond). Ihe d|0erence beteen the transmitted and the
received rate may be imputable to losses inside the
tansmitter network ampter, or to inaccuracies of the
dumping procedure a the receiver side. This topic is
40
being frher investigated adopting. diferent packet
generation and captre tools. However, as a result of
this preI|m|aapanalysis we can deduce that a feasible
number of measurable packets per seconds in oUr sce
nario is about 2000. Indeed, Fig. 7 h|ghlights that the
application level measurements allow to captre wit
out losses up to the secondstep.
4.2 Measurements in the WLAN sce
nario
In our experimental analysis, the laptop has been
equipped alterativelywith one of the following two
IEEE 802. lib NICs; a Lucent Orinoco Gold and a D
Link DWL650A baSed on Prism-chipset. The meas
urement campaign, frthertn evaluate the perorm
ance of the IEEE802.llb WA standard, permits to
compare the diferent performance achievable by
these to diverse couples of IEEE 802.11 b NICs and
Linux drivers. The first results seem to prove that a
limited but noticeable diference exists among these
two netork adapters. In more detail, considering the
plots repored in Fig. 8, the to graphs represent the
UDP transmitted and received rate using the two con
sidered PCMClA 802.lIb adapters in the UDP source
host. The plots are obtained varing the nominal
transmission rate of an UDP CBR fow, generated wt
MGEN, from 500 kbps t 5.5 Mbps, and measurng
tbe efective throughput at the Data Link level (at both
the tansmitter and rece|ver side) and at the applica
tion level at the receiver side, in order to verif the
presence of possibldosses inside the UP sink host.
These experiments have been caried out for diferent
UDP payload sizes (the plots of Fig. 8 refer to an UDP
payload size of 64 bytes), considering the limit of
2000 on te nominal packet generation rate found with
the analysis presented in the above paragraph.
The uppermost gaph is related to te perorce
achieve by the Lucent Orinoco Gold network
adapter, while the other one is obtained using a D
Link DWL650 based on Prism2.5chipset. The Ori
noo Gold NIC allows approach the maximum
theoretical value of 0.57 Mbps very closely; indeed,
the tsmission rate, averaged over the period 210-
330 sec, is about 0.569 Mbps. On the other hand, the
D-Link DWL650 allows to reach a slightly higher
throughput, even higher than the theoretical one
(about 0.59 Mbps in te period 60300 sec).
Note that a value of 64 bytes has been chosen instead
of 28 because of the detected instabil|t of the genera
tion process, as indicated in the subsection 4.1; in
deed, the packet rt related to the generation of a
600kbps rt is below 1200, which is lower than tbe
2200 limit obtained above. Table III sumarizes the
measured goodputs obta|ned with the three diferent
UDP payload sizes and the to considered
IEEE802.11b NICs, confrming the slightly higher
values obtained by the Prism2.5 based NIC wt re
spect to the Orinoco one.
Comparison of results
As the conclusion of our study, we compare the results
obtained considering the three approaches presented.
The results are qualitatively summarized in Fig. 9.
where we present the theoretical cure of goodput vs.
packet size when th" data rate at II Mps is consid
ered. In the same fgure. we also indicate the results
for 3 diferent packet sizes. obtained experimentally
using either the Orinoco Gold or the D-Link DWL650
NICs, or by simulation using OPNET (in this case the
analysis considers also a fourh value of packet size).
As we can observe from the picture, we have an en
couraging converence of the saturation goodput ob
tained using the diferent approaches which allows us
to assume the three approaches almost equivalent from
the results point of view. Table IV summarizes the
quantitative diference among the results prduced by
the diverse approaches. In particular, we can note the
slightly pessimistic results obtained using OPNET.
while minor differences can be noted comparing the
experimental and the theoretical ones.
.0

,$
, ..
,
I
I
Y
r
I
J.
I V
I
.'
I
10'
I
-
I I
.
I
,

, I
I
I
r
.
I ,
I
1
.....

..
IOoGdI1'1I
" E C1 J1

I
I
.
t

r

I
I
3
T 512 MD 7 10'115:IZ1401531M41721922211&
U)I"ZIlb)
Fig. 4 Comparison of analytcal, experimental and
simulated saturation god put for a .ingle UDP suurce
Table IV: Measured Goodputs
UP
Theoretic.1
Orinoo
OWL650 Simula-
Payload
Value
Gold
NIC tion
Siz
[Mbps]
NIC
[Mbps] (OPNET
[BJeJ [Mb),]
6 0.567 0.569 0.59 0.54
576 3.61 3.55 3.7 3.47
1472 6.1 t 6.06 6.2 5.94
Conclusions
The paper presents the analysis of the maximum
goodput achievable with an IEEE 802.11 b Basic Ser
vice Set, carried out following a theoretical, a simula
tive and an experimental approach. In particular, we
have taken into consideration a commercial discrete
event simulation tool, OPNET Modeler, analyzing the
802.11 model available as one of its standard libraries.
An extensive set of simulation rns has allowed us to
validate the simulation model versus theoretical re
sults known in literature, obtained using a Markovian
410
model. Such model accounts for all the exponential
backof protocol details. and is based on the assump
tion of constant and independent collision probability
of a packet transmitted by each station, regardless of
the number ofperfoned retransmissions.
Furhennore, an analytical evaluation of the saturation
goodput of a single IEEE 802.11 b tansmitter has been
conducted perfoning a timing budget of busy and
idle periods over the air interface. The most relevant
outcome of this work has been the pointing out of the
convergence towards such values of the results obtain
able with the simulative and, more considerably, ex
perimental approaches with a relative error of only
few percent, almost negligible fom an engineering
point of view. As a side result, the measurements pre
sented in Ihe paper, which have been conducted Oi our
experimental test-bed, highlight a limited, although
noticeable, diference between the maximum goodput
achievable using diferent IEEE 802.11 b compliant
wireless network interfaces, such as Ornoco and
Prism2.5 based ones.
7 Acknowledgements
This work has been carried out under the fnancial
suppor of te Italian Ministry for Education and Sci
entific Research (MIUR) in the famework of the
VICOM project.
References
[ I ] ANS!EEE SIr801.11 "Pa II: Wireless LAN Medium Access
Contol (MC) and Physicl Lye (PHY) Spc .... 1999 E.
[2] ANS!EEE St 802.llb "Par 11: Wirel ... LN Medium Access
Contol (MAC) ad Ph1"ic1 Lay (PHY) Spe.: Higher.Speed
Ph)ical uyer Ext. in the 2.4 OHz Bad", Sept. 1999
[3] http://standards.ieee.orgigetieee8021802.II.html.
[4] httpJ/grouper.ieee.orgigroups802l11IRepnstgg_update.htm
[5J hnp:l/groupr.ieee.otlgroup802l11IReptsltg_update.htm
[6] G. Bianchi, "Peronance Analysis of the IEEE 802.11 Distrib
uted Cc)rdination FunctiOll". IEEE Joural on Selecte Areas in
CQllmunications. Vol. 18, N. 3, March 200, pp. 535-541
[7) G. Anastasi. E. Borgi M. Conti. E. Gregori. "IEEE 802.1 t Ad
Ho NClwOrks; Peromlance Measu.-elllerlt:", Worksho on Mo
bile and Wireless Networks (MWN 2003) May 19. 20} Provi
dence, Rhode Island USA.
[8J D. P. Hole, f. A Tob.gi. "Capacity of M IEEE 802.llb Wireless
LAN supporing VoIP", submitted to ICC 2004
[9] G. Aatai, L Lenzini. "QoS provide by the IEEE 802.11
WLAN to advanced data applicalions a simulation study", Wire
les! Netos. Vol. 6, J.c. Bahzcr AG, Science Publish.rs. 200
rIO] OPNET Moeler web site www.opnet.corh
fll] s. Gar! a M. Kappes, "Al exprimental study oflhrougput
for VOP and Vol !raffle ;n IEEE 802.11 b networks", IEEE
Willess Communicat ions ad Nelworkmg 2003
[12] M. G. Arn. R. Aguero, L.Muioz and p, Mahonen, "Behaviour
of VOI'Based Application, over IEEE 802.1 t Wireless Net
wOTks", Pro. Qf te I Zth IEEE Interational Symposium on Per
,on.llndoo, and Mobile Radio Comm .. PIMRC 2001
[131 MGEN - htlp:llranimac.itd.nrLn.vy.miIlMGENI