Content uploaded by Farshad Eshghi
Author content
All content in this area was uploaded by Farshad Eshghi on Feb 13, 2014
Content may be subject to copyright.
IEEE Communications Magazine • March 2005 107
0163-6804/05/$20.00 © 2005 IEEE
TOPICS IN ADHOC NETWORK
INTRODUCTION
Mobile computing continues to enjoy rapid
growth thanks to ongoing technological advances
in portable devices, which have led to lowering
power consumption and lengthening the battery
life span. Simultaneously, the need for continu-
ous connectivity while roaming has put ad hoc
networks at the center of attention of researchers
and the industry over the past few years. Porta-
bility and mobility are greatly consistent with the
infrastructureless nature of ad hoc networks.
This has made mobile ad hoc networks
(MANETs) a compelling research topic. On the
other hand, viewed as either a standalone net-
work or a last-mile wireless connection to the
Internet, ad hoc networks have the capability of
expansion through multihopping. Multihop
MANETs, in situations where they are supposed
to support multimedia, present a challenge in
many aspects. This is attributed to the fact that
applications of different natures demand differ-
ent treatments. Multihop MANETs [1, 2] have
features that distinguish them from wired and
infrastructured wireless networks. These features
can be enumerated as: unpredictable link prop-
erties, node mobility, limited battery life, route
maintenance, security, hidden terminal, exposed
terminal, and capture effect.
In a multihop ad hoc network, nodes commu-
nicate with each other using several wireless
links, and there is no fixed infrastructure such as
a base station. Each node in the network also
acts as a router, forwarding data packets to
other nodes. One of the important challenges in
the design of multihop ad hoc networks (not our
concern here) is the development of routing pro-
tocols that can efficiently find routes between
two communication nodes [3]. In multihop ad
hoc networks, due to the numerous variables
involved, system multidimensionality grows sig-
nificantly, thus making analytical modeling a
harder task. Turning to simulation enables us to
investigate a greater number of phenomena and
possibly more involved models regarding traffic,
channel, and so on. In such a complex system,
adequate selection of the system parameters can
lead to considerable improvement in perfor-
mance, especially for time-sensitive applications.
Among similar attempts in the literature, [4]
studies the performance of the multihop ad hoc
IEEE 802.11 wireless LAN (WLAN) in various
mobility and traffic volumes. It proposes a piggy-
back reservation protocol in which neighbors
that hear the data packet are blocked and avoid
colliding by returning acknowledgment (ACK).
Furthermore, they learn about the next packet
transmission time. Likewise, neighbors of the
receiver that hear the ACK will avoid transmit-
ting at the time when the receiver is scheduled
to receive the next data packet. Another related
work [5] compares two different feedback mech-
anisms, explicit (per hop) vs. end-to-end ACK,
in multihop scenarios. The concept of adaptive
clustering application to multihop ad hoc net-
works was introduced in [6]. In the proposed
architecture, nodes are organized into nonover-
lapping clusters. The clusters are independently
controlled and dynamically reconfigured as
nodes move. Adaptive clustering uses code sepa-
ration among different clusters. However, the
Farshad Eshghi, Ahmed K. Elhakeem, and Yousef R. Shayan, Concordia University
ABSTRACT
Ongoing technological advances in portable
devices, coupled with the need for continuous
connectivity while mobile, have made ad hoc
networks a compelling research and develop-
ment topic, particularly in a challenging multi-
media multihop scenario. The ability of IEEE
802.11’s ad hoc mode of operation, as a domi-
nating wireless local area network (WLAN) pro-
tocol, to serve multihop networks requires
thorough investigation. In this article, through
considering crucial real-life physical phenomena
and avoiding as many confining assumptions as
possible, system performance measures such as
delay and packet failure rate are evaluated. As a
result, the importance of adequate selection of
the system parameters toward performance
improvement is underscored. Moreover, the sim-
ulation results imply that by complementing
through priority provisions, coordination, route
reservation, clustering, and optimum channel
coding considerations, the IEEE 802.11 medium
access control (MAC) protocol can survive in a
multihop scenario. The custom simulation envi-
ronment developed features modularity, com-
prising traffic generator, mobility, wireless
channel, and IEEE 802.11 protocol modules,
and is capable of accommodating many more of
the physical phenomena involved.
Performance Evaluation of
Multihop Ad Hoc WLANs
ESHGHI LAYOUT 2/16/05 11:03 AM Page 107
IEEE Communications Magazine • March 2005
108
work is not particularly about the IEEE 802.11
protocol. A very recent related work [7] discuss-
es priority considerations that are implemented
through queuing and contention window manip-
ulations. The motivation for our work herein is
to compensate for the fact that all the past simu-
lation efforts have adopted too many restrictive
assumptions, and to find out if 802.11’s ad hoc
mode can serve as the core building block of a
multihop MANET. We have tried to develop a
thorough simulation environment spanning both
medium access control (MAC) and physical
(PHY) layers that considers mobility and varying
wireless channel as well as forward error correc-
tion (FEC) using turbo coding.
The simulation is arranged in a module-based
manner where each independently implemented
module copes with one unique process, while
exchanging data with other modules through some
defined function values. This makes module
replacement/augmentation task quite simple. Aim-
ing at time-sensitive applications, our goal is to
evaluate the performance measures of the system
such as end-to-end delay and packet failure rate in
different situations (e.g., traffic, mobility) and
against different choices of system parameters
(e.g., fragmentation factor, buffer length, retrans-
mission limit). The article is organized as follows.
A topological description of the system is present-
ed. Traffic generation, mobility, channel, and
802.11 protocol modules are introduced. Key
assumptions made throughout the simulation are
intermittently mentioned when discussing the mod-
ules. Computer simulation results are presented
afterward. Concluding remarks then appear.
SYSTEM DESCRIPTION
A number of ntot nodes with limited buffer size
are uniform-randomly located inside a rectangu-
lar area, laid diagonally between (xmin,ymin) and
(xmax,ymax). These upper and lower limits, togeth-
er with the transmitter’s wireless range, determine
whether a single-hop or multihop scenario is dealt
with for the source-destination communicating
pair of each call. All nodes broadcast their trans-
missions omnidirectionally. As mentioned earlier,
routing is the first concern that distinguishes sin-
gle-hop from multihop. Since our main emphasis
in this article is the interaction of PHY and MAC
layers, we try to make the routing algorithm as
simple and straightforward as possible. As such,
we assume that all nodes are aware of each other’s
geographical location. This can be achieved
through location information (probably acquired
by global positioning system, GPS) exchange
among the nodes and certainly results in some
bandwidth cost. The routing strategy is then
defined as follows. Each node with a packet ready
to be sent picks another node as its intermediate
destination, provided that it is inside the source
node’s coverage area and has the least Euclidean
distance to the packet’s final destination. While
the latter may not be the best routing strategy (in
terms of cost and reliability), in most cases it
leads to the least number of hops traversed and is
of course easy to implement. However, any other
routing algorithm [8] can easily be incorporated
in our simulation environment. The most impor-
tant point that will not be further discussed is the
bandwidth cost of whatever routing strategy is
used. This degrades system performance below
the presented results. In the initialization phase,
some nodes are uniform-randomly selected as
mobile and the rest are left as fixed nodes. The
mobility status will prevail for the whole simula-
tion time. The parameter
along with v, mobile node’s speed, represent the
mobility volume of the system. In a similar way,
uniform-randomly selected nodes are responsi-
ble for generating traffic, while all nodes might
serve as forwarding and final destination nodes
at any time during the simulation. A parameter
representing the population of traffic generating
(active) nodes is defined as
This parameter may serve as the system’s traffic
volume indicator as well. Each active node is
pre-assigned a node as its final destination that
remains unchanged throughout the course of
simulation. This is not considered a simplifying
or restrictive assumption since all the nodes are
randomly located over the network area.
SIMULATION MODULES
TRAFFIC GENERATION
Considering voice connection as a representative
of real-time applications, the packet generation
model employed is an on-off traffic model. This
model has been used extensively in the literature
to model voice communications (e.g., [9]), mostly
because it can efficiently imitate silence/talk spurt
periods of human conversation and can easily be
simulated. Figure 1a illustrates the traffic genera-
Pn
n
mobile active
tot
=.
Pn
n
mobile mobile
tot
=
nn
nn
Figure 1. a) On-off traffic generation model; b) Gilbert model for time variations of the channel.
ACT IDL
0
(a)
r b/s
α
1 – α
1 – β
βG B
(b)
α'
1 – α'
1 – β'
β'
Since our main
emphasis in this
article is the
interaction of
physical and MAC
layers, we try to
make the routing
algorithm as simple
and straightforward
as possible. As such,
we assume that all
nodes are aware of
each other’s
geographical
location.
ESHGHI LAYOUT 2/16/05 11:03 AM Page 108
IEEE Communications Magazine • March 2005 109
tion model. While in the active state (ACT in the
figure), the model acts as a constant bit rate
(CBR) source generating rb/s, and while in the
idle state (IDL in the figure), no packet is gener-
ated. Parameters αand βare self-loop transition
probabilities in IDL and ACT states, respectively.
The slot time of the model, tslot, is considered
long enough to accommodate the longest packet
size generated above the node’s link layer. The
average idle and active states dwell times are cal-
culated as TIDL = tslot(1 – α)–1 and TACT = tslot (1
– β)–1, respectively. Upon packet generation at
the source node, vector function
pkt(SEQ_NO,CP_VER,HOP_VER) is responsi-
ble for carrying over all the information pertinent
to the generated packet, such as generating node
identity, final destination node identity (specified
at the transport or higher layers), intermediate
node identity (specified at the network layer as a
result of routing strategy application), packet
location in the queue, and temporal information.
Each packet is uniquely identified by its 3-tuple
identifier (SEQ_NO,CP_VER, HOP_VER) as
explained shortly.
MOBILITY
A special case of a random walk mobility model
(with constant speed and limited movement
directions) was adopted to predict mobile nodes’
movements throughout the network. The model
features completely random and memoryless
mobility; as such, it might not be suitable for all
kinds of applications. A complete treatment of
mobility models in ad hoc networks and the
importance of model choice regarding perfor-
mance results can be found in [10]. In our model,
all nodes designated as mobile periodically (each
∆/vs where the input parameter ∆denotes dis-
placement step length in meters) update their
locations. This moves the mobile node as much
as ∆along either the x or y axis. Large-scale fad-
ing and shadowing phenomena are directly relat-
ed to the macro movements of the nodes, so
they are handled in the mobility module. Using a
two-slope curve model [11] and log-normal dis-
tribution for path loss and shadowing, respec-
tively, we have
(1)
In Eq. 1, nPL and λrepresent terrain-dependent
path loss exponent and wavelength, respectively,
and Xis a normal random variable in dB. Parame-
ter d0denotes the breakpoint distance below which
the environment has free space characteristics. For
simplicity we do not consider correlation in shad-
owing from one location to another. The displace-
ment, distance, and large-scale fading between all
pairs of nodes are calculated and used for channel
bit error rate (BER) evaluation as follows.
THE CHANNEL MODEL
In this module the small-scale behavior of the
channel, in particular multipath fading as its
major contributor, is simulated. The two impor-
tant parameters that quantify channel behavior
with regard to fading classes are coherence
bandwidth (Bc) and coherence time (Tc). Less
conservative expressions for the coherence band-
width and coherence time are [12]
(2)
In Eq. 2 στdenotes the delay spread of the wire-
less channel and fm, vm, fc, and Cdenote maxi-
mum Doppler frequency, maximum mobile
speed, carrier frequency, and light speed respec-
tively, which are related through fm= vmfc/C.
The power spectral density of the direct
sequence spread spectrum (DSSS) signal in
802.11 is limited to around 20–22 MHz using a
transmit spectrum mask [13]. This bandwidth
translates to a 9–10 ns delay spread through
applying Eq. 2. While this value of delay spread
may fall into the rural area range, suburban and
urban areas show higher delay spreads [14] and
subsequently lower coherence bandwidths. To
avoid frequency band variations of the signal
strength, at this point we adopt the flat fading
channel assumption, which may not be valid for
all terrain situations. Regarding the coherence
bandwidth and for the system under study, val-
ues of C= 108m/s, fc= 2.4 GHz, and v= 10
m/s give a coherence time of Tc= 5.29 µs. Com-
pared to the symbol period of 1 µs, the wireless
channel can conventionally be considered slow
fading. However, since we are concerned about
packet rather than symbol (bit herein), depend-
ing on the packet length, a packet may experi-
ence several channel conditions. So at any time
we consider the channel as a time-invariant sta-
tionary process with different statistics during
different coherence time units. In order to
model the time variations of the channel, a two-
state Markov model (Gilbert model) is adopted.
Markovian modeling of fading channels is exten-
sively found in the literature (e.g., [15]). Most of
these references agree on the suitability of the
first order Markov chain in modeling slow to
medium fading channels, while higher-order
Markov chains are needed for faster fading
channels. Moreover, limiting Markov states to
two as good and bad channel states has been
common due to its efficient trade-off between
accuracy and simplicity, and of course its ease of
computer simulation. However, good and bad
states have been defined in a variety of ways
such as having fixed low and high BERs, experi-
encing no and deep fade, or non- and erroneous
BTC
vf
cc
mc
≈≈
1
5
0 423
στ
,..
Pd Pd n d
dXdd
LL
() ( ) log (, );= +
+≥
010
0
10 0
σ
00
10
20 4
Pd d
L() log=
π
λ
;.dd<0
nn
nn
Figure 2. Random access and data transmission in IEEE 802.11 ad hoc
mode of operation using the optional RTS/CTS channel reservation scheme.
123 j
Tra nsmitter
DIFS
Slot time
RTS SIFS DATA
Receiver CTSSIFS
Others NAV(RTS)
NAV(CTS)
NAV(DATA)
Defer access
ACKSIFS
ESHGHI LAYOUT 2/16/05 11:03 AM Page 109
IEEE Communications Magazine • March 2005
110
packet delivery. We assume that a mobile node
along its movement alternatively experiences
either a line of sight (LOS) or Rayleigh fading
channel along its target node. Figure 1b shows
the embedded channel model in our simulation
with α′ and β′ denoting the self-loop probabili-
ties of bad (B) and good (G) states. By equating
Tcin Eq. 2 to the model’s slot time, the average
dwell times of the bad and good states are TB=
Tc(1 – α′)–1 and TG= Tc(1 – β′)–1, respectively.
In the good state the received power (PRG) is
calculated from the transmitted power (PT) as
PRG= PT– PL(d) where PL(d) is defined in Eq.
1 and with all powers in dB. In the bad state,
PRGserves as the average power of the received
signal with a Rayleigh distributed envelope from
which its power is calculated.
802.11 PROTOCOL IMPLEMENTATION
IEEE 802.11 protocol implementation in ad hoc
mode is discussed here. More emphasis is put on
the MAC layer while the PHY layer implemen-
tation is realized through adopting parameters
related to the specific choices of modulation,
spreading, and FEC coding.
The MAC Layer — The random backoff algo-
rithm is the contention resolution strategy in the
IEEE 802.11 WLAN protocol operating in ad hoc
or distributed coordination function (DCF) mode.
Figure 2 illustrates the channel access mechanism
through the optional request-to-send/clear-to-send
(RTS/CTS) channel reservation scheme in ad hoc
mode of operation. For a more detailed discussion
of the protocol, the reader is referred to [9, 13].
Real-time traffic, depending on the application,
requires a compromise between delay and loss.
Late packets are not worth more than lost ones.
To take care of this, and as an amendment to the
protocol toward real-time adaptability, unacknowl-
edged packets are retransmitted a limited number
of times (Rmax) before releasing the channel.
Moreover, Rmax unsuccessful retransmissions are
followed by a packet drop to avoid too much
latency of late packets. Other mechanisms that
result in packet elimination are blocking due to
the generating/target node’s full buffer, dropping
due to reaching some constraints such as time-to-
live, maximum hopping, retransmission limit,
duplication (explained below), route unavailability,
and successful reception. We also assume that
fragmentation is done at layers above MAC so we
only need to deal with single-fragment MAC pro-
tocol data units (MPDUs). The MPDU is then
prefixed with PHY layer overhead, coded, and
transmitted as the DATA packet (Fig. 2). In terms
of performance evaluation, the latter does not
seem to produce considerable discrepancy with
respect to the MAC fragmentation scheme [16].
Due to the ad hoc nature of the network, the
MAC protocol is implemented in a distributed
manner (concurrently in all nodes); thus, there are
some peculiarities that will be highlighted here.
The only verification mechanism regarding a
healthy DATA transmission is through ACK
exchange (Fig. 2), which itself may or may not fail.
Thus, situations may arise in which several copies
of the same packet are hopping to the same desti-
nation. Figure 3 illustrates such situations. In Fig.
3a node #1 sends packet A to node #3 through
intermediate node #2. Although the packet is
received correctly at node #2, its corresponding
ACK back to node #1 fails, which makes node #1
initiate a retransmission. At the same time, node
#2 will forward already queued packet A along its
route to the final destination. In Fig. 3b the same
event is illustrated, except that due to topology
changes (as a result of mobility) the retransmis-
sion of packet A is headed toward a different
intermediate node, node #4, which in turn will
forward it along its route to the final destination.
To better simulate the real-life situation and
account for the above possibilities, the following
measures are put in place:
• Only the originating node can manipulate
the function pkt(.),
• Upon correct reception at an intermediate
node, a new version of the packet with a dif-
ferent hop identification (HOP_VER) is pro-
duced (with this intermediate node as its
originator) and queued for further forwarding,
•At the time of retransmission, if the next
intermediate node is different from the pre-
vious transmission attempt, a new version
of the packet with a different copy identifi-
cation (CP_VER) is produced (at the origi-
nating node), and the old one is discarded.
With the above provisions, the multiversion
packet phenomenon is well simulated. Since all
versions of a packet have the same unique
sequence number assigned at generation time,
flooding can be partially prevented by discarding
the packet whose sequence number (SEQ_NO)
has already been met (duplication) at each
nn
nn
Figure 3. Upon an ACK failure, retransmission is forwarded to: a) the same
previous target node; b) a new target node due to mobility and topology
change.
(a)
HOP_VER is changed
X
pkt(SEQ_NO,CP_VER_1,HOP_VER_1)
pkt(SEQ_NO,CP_VER_1,HOP_VER_1)
pkt(SEQ_NO,CP_VER,HOP_VER_2)
1
Packet A
Packet A, RE_TX
Packet A
ACK
2
1 2
3
(b)
HOP_VER is changed
CP_VER is changed
X
pkt(SEQ_NO,CP_VER_1,HOP_VER_1)
pkt(SEQ_NO,CP_VER_2,HOP_VER_1) pkt(SEQ_NO,CP_VER_2,HOP_VER_2)
pkt(SEQ_NO,CP_VER_1,HOP_VER_2)
1
Packet A
Packet A, RE_TX
Packet A
Packet A, RE_TX
ACK
2
1 4 5
3
ESHGHI LAYOUT 2/16/05 11:03 AM Page 110
IEEE Communications Magazine • March 2005 111
receiving node. The vector function node
(ID_NO), throughout the course of simulation,
encompasses the static and dynamic information
for node #ID_NO.
PHY Layer — Spreading, modulation, and coding
schemes are considered here. According to FCC
regulations, a DSSS system shall provide a pro-
cessing gain of at least 10 dB. This is accom-
plished by chipping the baseband signal at 1 MHz
with an 11-chip PN code. The DSSS system uses
baseband modulations of binary phase shift key-
ing (BPSK) or differential binary phase shift key-
ing (DBPSK) to provide 1 Mb/s data rate. The
presence of fading in wireless environments moti-
vates the use of channel codes. Channel codes
mitigate the adverse effect of fading in wireless
channels by adding redundancy and memory to
the transmission. Turbo codes are a class of error
correction codes that enable reliable communica-
tions with power efficiencies close to the theoreti-
cal limit predicted by Claude Shannon. The good
error performance of turbo codes may be com-
promised by their bandwidth and computational
(hence energy consumption) costs in certain ad
hoc applications. For instance, they cannot be
afforded in most sensor networks where nodes
feature very limited battery resources. As an illus-
trative example of turbo code application to
WLANs, our turbo coding scheme features two
identical recursive systematic convolutional (RSC)
encoders with parameters n= 2, k= 1, K= 3,
G0= 7, and G1= 5, with half of the parity bits
punctured. The decoder is Log-MAP with 8 itera-
tions. Regarding the interleaver, we consider two
types: one, which shuffles fixed short size control
packets (RTS,CTS, and ACK), is a block inter-
leaver; the other, which deals with data packets, is
a random separated interleaver. While this coding
scheme consumes half the bandwidth in the first
place, it is hoped it will retrieve much more
through BER improvement. However, there is no
claim about the optimality of this coding scheme,
and no effort is made to evaluate its efficiency
since it is not in the main theme of this work.
Now the focus is turned toward how transmission
health is assessed in our simulation. Data packets,
depending on their length and the channel coher-
ence time (Tcin Eq. 2) they face, may experience
several channel qualities. The latter information
represented by BER1, BER2, …, BERxare carried
by pkt(.) function. On the other hand, short con-
trol packets always go through a single-quality
channel (represented by a single BER) except at
rarely occurring boundary situations. The corre-
sponding BERs are extracted from the BER vs.
SNR curve in the additive white Gaussian noise
(AWGN)/BPSK case with the above employed
coding scheme [17, Fig. 5.26]. Accordingly, the
packet error rate (PER) for all packet sizes is
(3)
where LTc denotes the equivalent bit length of
Tc, mod(a,b) accounts for the remainder of
dividing aby b, and xis the number of coher-
ence times fitted within one packet.
SIMULATION RESULTS
In this section the effects of different choices of
MAC layer parameters as well as other external
network parameters (e.g. mobility, traffic load,
etc.) on the system performance are simulated.
In particular, packet length, maximum lifetime
of a packet (time-to-live, TTL), maximum num-
ber of times a packet is retransmitted (ReTXmax),
and a node’s maximum buffer size (Qmax) are
examined. This is done in different network con-
ditions by varying the percentage of active nodes
(Pactive), percentage of mobile nodes (Pmobile),
and mobile node speed (v). However, due to
space limitations, only some of the produced
results are presented herein. Regarding other
parameters, the total number of nodes present
in the network (ntot), node traffic generating
rate, and network geographical dimensions
remain fixed at 30, 64 kb/s, and 400 m ×400 m,
respectively, throughout the simulation. Also,
throughout the simulation the following values
PER BER BER
x
LL
m
L
m
x
iTc Tc
=− − −
=
−
11 1
1
1
() ( );
mod( , ) ∏∏
= i RTS CTS ACK DATA,,, ,
nn
nn
Table 1. Parameters used in the simulation.
Network parameters
ntot 30
Node’s coverage radius 150 m
Network area 400 m ×400 m
Displacement step (∆)5 m
Simulation time 10.0 s
Source model parameters
Node’s traffic generation rate 64 kb/s
Avg. talk-spurt/silence period 1.0/1.35 s
Max. packet length (uncoded) 8000 bits
Channel parameters
Path loss exponent (nPL)3.2
Breakpoint distance (d0) 10 m
Standard deviation of 8 dB
shadowing (σ)
β′/α′ 0.967/0.9
IEEE 802.11 standard parameters for DSSS WLANs
LRTS/CTS/ACK 20/14/14 octets
OHMAC/PHY 34/24 octets
DIFS/SIFS/slot — time 50/10/20 µs
Channel rate 1 Mb/s
Miscellaneous parameters
Transmitter power (PT) 17 dBm
Noise floor (NF)–80 dBm
Simulation slot time 10 µs
Maximum hopping 4 hops
Since we are aiming
at real-time
application support,
a larger area means
more hops, which
in turn contributes
to unfavorable delay
accumulation. The
area is limited such
that the majority of
the source-to-
destination
transmissions can
take place with the
number of hoppings
less than
HOP_LMT = 4.
ESHGHI LAYOUT 2/16/05 11:03 AM Page 111
IEEE Communications Magazine • March 2005
112
were assigned to the above parameters unless
they are the parameter subject to variation; TTL
= 0.4 s, v= 30 m/s, Pactive = 0.5, Pmobile = 0.75,
ReTXmax = 4, and Qmax = 5. These values reflect
high traffic load and relatively fast mobility. To
give a sense of traffic volume, Pactive = 0.5
means, when counting in MAC/PHY/FEC over-
heads and idle/active periods, 15 nodes each
generating on average 60 kb/s continuously, and
this should be compared against 1 Mb/s channel
capacity. Moreover, in multihop situations, gen-
erated traffic is multiplied by a factor of number
of hops to final destination, which is lower than
the effective channel reuse factor. A per user
traffic rate of 64 kb/s can be representative of a
wide variety of real-time applications. A 400 m ×
400 m area together with a node’s 150 m cover-
age radius (which is a typical coverage radius for
wireless interfaces in the market) represent our
intended multihop scenario. Since we are aiming
at real-time application support, a larger area
means more hops, which in turn contributes to
unfavorable delay accumulation. The area is lim-
ited such that the majority of source-to-destina-
tion transmissions can take place with the
number of hoppings less than HOP_LMT = 4.
However, the simulation environment developed
has no limitation in handling areas of larger
sizes. Regarding the number of nodes, 30 users
in the area above would be a normal user popu-
lation (of course, in certain application scenar-
ios) and ensures reasonable availability of
intermediate nodes toward the final destination
as well. However, as mentioned earlier, in most
of the simulations Pactive was set to 0.5 to yield a
high traffic load scenario. Parameters used in
the simulation are tabulated in Table 1. The
parameters of the source and channel models
are chosen to be consistent with the correspond-
ing parameters in [9]. The other parameters are
partially acquired from [11] and Cisco WLAN
specification guides. The simulation time is fixed
at 10.0 s to accommodate a good number of gen-
erated packets and TTL periods, and for the
network to reach its steady state. In all the simu-
lation results, each point is acquired by applying
the batch means method [18] over 24 (6 ×4)
simulation runs to obtain a good level of confi-
dence. Two major measures considered herein
as performance indicatives are end-to-end (e.t.e.)
delay and packet failure rate (PFR), wherein
such a delay has been presented as normalized
with respect to the subject DATA packet trans-
mission time. The delay includes the time dura-
tion from the moment of packet generation until
its healthy reception. Not included are packeti-
zation delay, interleaving delay in turbo encod-
ing, propagation delay, delays related to turbo
decoder iterations, and other packet processing
times at intermediate and destination nodes.
However, these delays are both fixed and negli-
gible compared to random transmission and
queuing delays. In each simulation run, e.t.e.
delay is calculated by averaging over all healthy
received packets. Packet failure rate is the ratio
of the number of packets dropped to the total
number of generated packets throughout the
duration of simulation . Different limit violations
in the aggregate contribute to the total number
of dropped packets. Figures 4–6 illustrate how
the above mentioned performance measures are
affected by varying DATA packet length
(LDATA). In Fig. 4, a group of curves correspond-
ing to different values of TTL, 0.1–0.4 s, are
drawn. In the left plot, the normalized delay is
decreased by adopting longer packet lengths and
lower TTL values. However, this decrease
becomes less significant upon passing LDATA =
nn
nn
Figure 4.Performance measures of the system versus DATA packet length with time-to-live (TTL) as a parameter: a) normalized
delay; b) packet failure rate.
DATA packet length (uncoded,bits)
V = 30 m/s ; Pmobile = 0.75 ; Pactive = 0.5 ; ReTXmax = 4 ; Qmax = 5
(a)
20001000
0
20
Normalized delay (w.r.t. packet transmission time)
40
60
80
100
120
3000 4000 5000 6000 70008000
TTL = 0.1 s
TTL = 0.2 s
TTL = 0.3 s
TTL = 0.4 s
%80 Conf_int
DATA packet length (uncoded,bits)
V = 30 m/s ; Pmobile = 0.75 ; Pactive = 0.5 ; ReTXmax = 4; Qmax = 5
(b)
1000
0.6
0.65
Packet failure rate (PFR)
0.7
0.75
0.8
0.85
0.9
0.95
70008000
6000
4000
3000
20005000
TTL = 0.1 s
TTL = 0.2 s
TTL = 0.3 s
TTL = 0.4 s
%80 Conf_int
ESHGHI LAYOUT 2/16/05 11:03 AM Page 112
IEEE Communications Magazine • March 2005 113
4000 bits. The fact that the channel favors longer
packets all the way points to the effectiveness of
the FEC scheme employed, and that probably a
higher rate turbo coding might be also sufficient.
In the right plot, packet failure rate mostly
decreases with higher TTLs as may be expected.
However, TTL ≈0.3 s shows turning points in
4000- and 6000-bit packet lengths. By trading
delay for loss or vice versa, we may end up with
different choices in the ranges of 0.2 ≤TTL ≤
0.4 s and 4000 ≤LDATA ≤6000 bits. Figure 5
shows the effect of varying queue size and
DATA packet length on system performance
measures. Looking at the delay plot, all the
curves similarly descend by a climbing packet
length. The descents become less significant and
particularly indistinguishable beginning at the
4000-bit packet length. The closeness of the
curves for Qmax ≤3 suggests that beyond Qmax =
3, this parameter is not a decisive factor in delay
performance. On the PFR side, it seems that
6000-bit packet length is the best choice for all
queue lengths. Similar to the delay performance,
not much is gained by enlarging the queue
beyond Qmax = 3. This means that the parame-
ter pair (Qmax = 3, LDATA = 6000 bits) gives the
best combined delay-PFR performance. In the
last two figures, the two parameters subject to
variation control the dwelling time of the packet
in the system (in fact ReTXmax has a similar
effect). A longer stay of the packet in the system
(due to a higher TTL, queue size, or retransmis-
sion limit) on one hand may increase its chances
of successful delivery, while on the other hand it
decreases other packets’ chances by making
them violate the system parameters limits. The
multiple crossings of PFR plots at some points
stem from these conflicting effects.
The impact of mobility on system perfor-
mance can be evaluated by varying either the
number of mobile users in the system (Pmobile) or
the speed of mobile users (v). A higher Pmobile
(at fixed speed) is translated to deeper topology
changes since it involves more nodes. However,
since the topology changes take place in a com-
pletely random manner, considering the fixed
speed, in an average sense one should not expect
any wide fluctuations of performance measures
for different mobility volumes. On the other
hand, a higher node’s speed means faster and
more frequent topology changes of the network.
It also means that a particular long enough
packet may experience more channel qualities
due to shrinking channel coherence time.
Because of the powerful FEC scheme used,
many packets that would otherwise be complete-
ly faded can be recovered. The delay plot of Fig.
6 shows slightly higher delays for higher speeds
while maintaining the same descending pattern
with respect to packet length. Since the delay
measure only deals with successfully received
packets, it cannot clearly reflect the phenomenon
described above. The PFR plot illustrates how
longer packets may benefit from faster mobility.
It also shows that LDATA = 6000 bits can be a
good choice for different node speeds.
To better visualize the impact of adequate
selection of system parameters and traffic load
on performance, Fig. 7 illustrates performance
measures corresponding to another set of choic-
es: TTL = 0.25 s, v= 30 m/s, Pmobile = 0.75,
ReTXmax = 2, Qmax = 4, and different load con-
ditions. The selected values are intended to
improve the combined delay-PFR performance.
All the earlier graphs should be compared
against the diamonds (Pactive = 0.5) in Fig. 7. To
evaluate the performance under different traffic
loads, the ratio of active users (generating traf-
nn
nn
Figure 5. Performance measures of the system vs. DATA packet length with maximum queue length (Qmax) as a parameter:
a) normalized delay; b) packet failure rate.
DATA packet length (uncoded, bits)
(a)
Pactive = 0.5 ; TTL = 0.4 s ; Pmobile = 0.75 ; v = 30 m/s ; RE_TX_LIMIT = 4
700080001000
0
20
Normalized delay (w.r.t. packet transmission time)
40
60
80
100
120
6000
5000
4000
3000
2000
DATA packet length (uncoded, bits)
(a)
Pactive = 0.5 ; TTL = 0.4 s ; Pmobile = 0.75 ; v = 30 m/s ; RE_TX_LIMIT = 4
700080001000
0.55
0.6
Packet failure rate (PFR)
0.65
0.7
0.75
0.8
0.85
0.9
60005000400030002000
Qmax = 1
Qmax = 3
Qmax = 5
Qmax = 7
%80 Conf_int
Qmax = 1
Qmax = 3
Qmax = 5
Qmax = 7
%80 Conf_int
ESHGHI LAYOUT 2/16/05 11:03 AM Page 113
IEEE Communications Magazine • March 2005
114
fic) is varied over Pactive = 0.1–0.5. In Fig. 7, as
expected, all the plots in general show more
degradation with higher loads. Again, delay plot
favors longer DATA packets; interestingly, 6000-
bit packet could be the best choice for all traffic
volumes regarding the PFR performance.
Finally, in interpreting the quantities of the
performance measures, it should be noted that
the results pertain to pure basic 802.11 protocol,
and no extra efforts such as revising the proto-
col, optimizing the FEC scheme employed, or
clustering have been made.
CONCLUSION
In order to investigate the survivability and per-
formance of IEEE 802.11’s DCF and the signifi-
cance of adequate selection of the system
parameters in a multihop scenario, an inclusive
simulation was performed. The simulation, com-
nn
nn
Figure 6. Performance measures of the system vs. DATA packet length with the mobile nodes' speed (v) as a parameter:
a) normalized delay; b) packet failure rate.
DATA packet length (uncoded, bits)
(a)
TTL = 0.4 s ; Pmobile = 0.75 ; Pactive = 0.5 ; ReTXmax = 4 ; Qmax = 5
700080001000
0
20
Normalized delay (w.r.t. packet transmission time)
40
60
80
100
120
60005000400030002000
DATA packet length (uncoded, bits)
(b)
TTL = 0.4 s ; Pmobile = 0.75 ; Pactive = 0.5 ; ReTXmax = 4 ; Qmax = 5
700080001000
0.6
0.65
Packet failure rate (PFR)
0.7
0.75
0.8
0.85
60005000400030002000
v = 0.1 m/s
v = 10 m/s
v = 20 m/s
v = 30 m/s
%80 Conf_int
v = 0.1 m/s
v = 10 m/s
v = 20 m/s
v = 30 m/s
%80 Conf_int
nn
nn
Figure 7. Performance measures of the system vs. DATA packet length with the percentage of active nodes (Pactive) as a parameter and
a different choice of other parameters: a) normalized delay; b) packet failure rate.
DATA packet length (uncoded bits)
(a)
TTL = 0.25 s ; Pmobile = 0.75 ; v = 30 m/s ; ReTXmax = 2 ; Qmax = 4
70008000
1000
0
10
Normalized delay (w.r.t. packet transmission time)
20
30
40
50
60
70
6000500040003000
2000
Pactive = 0.1
Pactive = 0.25
Pactive = 0.5
%80 Conf_int
DATA packet length (uncoded bits)
(b)
TTL = 0.25 s ; Pmobile = 0.75 ; v = 30 m/s ; ReTXmax = 2 ; Qmax = 4
700080001000
0.2
0.3
Packet failure rate (PFR)
0.4
0.5
0.6
0.7
0.8
0.9
60005000400030002000
Pactive=0.1
Pactive=0.25
Pactive=0.5
%80 Conf_int
ESHGHI LAYOUT 2/16/05 11:03 AM Page 114
IEEE Communications Magazine • March 2005 115
prising traffic generation, mobility, wireless chan-
nel, and IEEE 802.11 MAC protocol, was intend-
ed to evaluate system performance measures,
such as delay and packet failure rate, of special
interest to real-time applications. Here we do not
consider throughput, since the maximum band-
width available cannot be clearly defined due to
the multichannel nature of multihop ad hoc
WLANs. Most of the results corresponding to the
performance measures above do not demonstrate
a monotonic trend against the parameters consid-
ered. While a higher TTL, queue size, and
retransmission limit allow a particular packet to
stay longer in the system and have more chances
of delivery, it contributes to a more crowded sys-
tem and in turn more chances of collision, queue
overflow (for later coming packets), TTL viola-
tion (for earlier arriving packets), and thus final
failure. These conflicting issues signify the impor-
tance and necessity of adequate selection of the
parameters involved. The only monotony seen is
the decrease in normalized delay vs. packet
length. The nonlinear declining behavior, general-
ly more profound for short packet lengths
although dependent on the other parameters, is
basically due to the effectiveness (or maybe over-
sufficiency) of the FEC scheme employed.
An important observation is that apart from
the maximum delay, which was constrained by
the TTL (i.e., chosen to fulfill the delay require-
ment of a variety of time-sensitive applications),
the results of other performance measurements
do not seem promising for real-time applica-
tions. This is not surprising, given the worst case
deployment scenario of this article. Multihop ad
hoc networking, high traffic load, lack of coordi-
nation among nodes, no facility for route reser-
vation or clustering, and other issues all
contribute to such a worst case. Nevertheless,
future advances in the aforementioned areas will
make the IEEE 802.11 ad hoc mode of opera-
tion a viable approach to handling real-time traf-
fic in wireless environments.
ACKNOWLEDGMENT
The authors would like to express their gratitude
to series editor Prof. Silvia Giordano and the
reviewers, whose constructive comments led to
the article’s quality upgrade and producing more
accurate simulation results.
REFERENCES
[1] H. Wu et al., “Speech Support in Wireless, Mobile, and
Multihop Networks,” Proc. Nat. Sci. Council, ROC(A),
vol. 24, no. 6, 2000, pp. 441–49.
[2] S. Xu and T. Saadawi, “Does the IEEE 802.11 MAC Pro-
tocol Work Well in Multihop Wireless Ad Hoc Net-
works,” IEEE Commun. Mag., June 2001, pp. 130–37.
[3] P. Mohapatra, J. Li, and C. Gui, “QoS in Mobile Ad Hoc
Networks,” IEEE Wireless Commun., June 2003, pp. 44–52.
[4] C. R. Lin and M. Gerla, “Real-Time Support in Multihop
Wireless Networks,” Wireless Networks, no. 5, 1999,
pp. 125–35.
[5] M. Kazantzidis and M. Gerla, “End-to-end vs. Explicit
Feedback Measurement in 802.11 Networks,” Proc. 7th
Int’l. Symp. Comps. and Commun., 2002, pp. 429–34.
[6] C. R. Lin and M. Gerla, “Adaptive Clustering for Mobile
Wireless Networks,” IEEE JSAC, Sept. 1997, pp. 1265–75.
[7] K. Sundaresan, H. Y. Hsieh, and R. Sivakumar, “IEEE 802.11
over Multihop Wireless Networks: Problems and New Per-
spectives,” Ad Hoc Networks, no. 2, 2004, pp. 109–32.
[8] J. Broch et al., “A Performance Comparison of Multi-
Hop Wireless Ad Hoc Network Routing Protocols,”
MobiCom, Dallas, Texas, USA, Oct.1998, pp. 85–97.
[9] B. P. Crow et al., “IEEE 802.11 Wireless Local Area Net-
works,” IEEE Commun. Mag., Sept. 1997, pp. 116–26.
[10] T. Camp, J. Boleng, and V. Davies, “A Survey of Mobil-
ity Models for Ad Hoc Network Research,” Wireless
Commun. & Mobile Comp., Special Issue on Mobile Ad
Hoc Networking: Research, Trends, and Applications,
vol. 2, no. 5, 2002, pp. 483–502.
[11] M. J. Ho et al., “RF Challenges for 2.4 and5 GHz
WLAN Deployment and Design,” WCNC, vol. 2, Mar.
2002, pp. 783–88.
[12] T. S. Rappaport, Wireless Communications: Principles
and Practice, 2nd ed., Prentice Hall PTR, 2002.
[13] Wireless Medium Access Control (MAC) and Physical
Layer (PHY)Specifications, ANSI/IEEE Std. 802.11,
LAN/MAN Standards Committee of the IEEE Computer
Society, 1999.
[14] T. A. Wysocki and H. J. Zepernick, “Characterization of
the Indoor Radio Propagation Channel at 2.4 GHz,” J.
Telecommun. and Info. Tech., no. 3–4, 2000, pp.
84–90.
[15] H. S. Wang and P. C. Chang, “On Verifying the First-
Order Markovian Assumption for a Rayleigh Fading
Channel Model,” IEEE Trans. Vehic. Tech., vol. 45, no.
2, May 1996, pp. 353–57.
[16] F. Eshghi, “Performance Evaluation of Multihop Ad
Hoc Wireless LANs,” Ph.D. dissertation, Concordia
Univ., Montreal, Quebec, Canada, Apr. 2004.
[17] L. Hanzo, T. H. Liew, and B. L. Yeap, Turbo Coding, Turbo
Equalisation and Space-Time Coding, Wiley, 2002.
[18] A. Leon-Garcia, Probability and Random Processes for
Electrical Engineering, Addison-Wesley, 1994.
BIOGRAPHIES
FARSHAD ESHGHI (farshade@ieee.org) received his Ph.D.
degree in electrical engineering from Concordia University,
Montreal, Canada, in 2004, during study toward which he
has been stydying the IEEE 802.11 WLAN protocol. Since
then he has been working on different issues regarding
wireless ad hoc networks as a post-doctoral fellow at Con-
cordia. Prior to starting the Ph.D. program, he worked on
various electronic circuit design projects in different com-
panies. His main research interests include wireless com-
munications, WLAN access protocols, routing, and
performance evaluation.
AHMED K. ELHAKEEM [SM] (ahmed@ece.concordia.ca)
received his Ph.D. degree from the Southern Methodist
University, Dallas, Texas, in 1979.He spent the next two
years working as a visiting professor in Egypt, after which
he moved to Ottawa, Canada, in 1982. He assumed
research and teaching positions in Carleton and Manitoba
Universities and later moved to Concordia University in
1983, where he is now a professor in the Electrical and
Computer Engineering Department. He has published
numerous papers in IEEE and international journals in the
areas of spread spectrum and networking. He is a well-
known expert in these areas and serves as a consultant to
various companies. His current research interests include
interconnected wireless LANs, error correction for IP multi-
cast,wideband networks, switching architectures , CDMA
networks, multiprotocols, software radios, and reconfig-
urable networks, He is a co-author of the book Fundamen-
tals of Telecommunications Networks (Wiley, 1994). He
has chaired and organized numerous technical sessions in
IEEE conferences, and was Technical Program Chairman
for IEEE Montech ‘86 in Montreal, Canada. More recently,
he was the key Guest Editor for four issues of IEEE Journal
on Selected Areas in Communications on CDMA, CDMA I,
II, III and IV, appearing in May and June 1994, and Octo-
ber and December 1996. He was the Communications
chair of IEEE Montreal and TCCC Representative to ICC
‘99, He served as an associate editor for IEEE Communica-
tions Letters (1996–1999), and is a professional engineer
of Ontario.
YOUSEF R. SHAYAN (yshayan@ece.concordia.ca) received his
Ph. D. in 1990 in electrical engineering from Concordia
University. Since 1988 he has worked in several wireless
communication companies. He has been with SR Telecom,
Spar Aerospace, Harris, EMS Technologies, and BroadTel
Communications as senior engineer, senior manager and
vise president of engineering. In 2001 he joined the
Department of Electrical and Computer Engineering of
Concordia University as associate professor. His research
interests include wireless communications, error control
coding, and modulation techniques.
An important
observation is that
apart from the
maximum delay,
which was con-
strained by the TTL,
the results of other
performance
measure do not
seem promising in
view of real-time
applications. This is
not surprising, given
the worst case
deployment
scenario of
this article.
ESHGHI LAYOUT 2/16/05 11:03 AM Page 115
