This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING
Spatial Distribution and Channel Quality
Adaptive Protocol for Multi-Hop Wireless
Broadcast Routing in VANET
Michael Slavik, Student Member, IEEE, and Imad Mahgoub, Senior Member, IEEE
Abstract—Multi-hop wireless broadcast is an important component in
vehicular networks. Many applications are built on broadcast communi-
cations, so efficient routing methods are critical for their success. Here
we develop the Distribution-Adaptive Distance with Channel Quality
(DADCQ) protocol to address this need and show that it performs well
compared to several existing multi-hop broadcast proposals.
The DADCQ protocol utilizes the distance method to select forward-
ing nodes. The performance of this method depends heavily on the
value of the decision threshold, but it is difficult to choose a value
that results in good performance across all scenarios. Node density,
spatial distribution pattern, and wireless channel quality all affect the
optimal value. Broadcast protocols tailored to vehicular networking must
be adaptive to variation in these factors.
In this work we address this design challenge by creating a decision
threshold function that is simultaneously adaptive to the number of
neighbors, the node clustering factor, and the Rician fading parameter.
To calculate the clustering factor, we propose using the quadrat method
of spatial analysis. The resulting DADCQ protocol is then verified with
JiST/SWANS and shown to achieve high reachability and low bandwidth
consumption in urban and highway scenarios with varying node density
and fading intensity.
Index Terms—Wireless broadcast, statistical broadcast, broadcast
storm, VANET
1 I NTRODUCTION
Multi-hop wireless broadcast plays an important role in
the operation of wireless systems. Many point-to-point
ad-hoc networking protocols, such as AODV and DSR,
use broadcast to discover routes. Research in this area
has demonstrated that blindly retransmitting broadcast
packets (flooding) can lead to an explosive growth of
traffic called the broadcast storm problem [1], that attenu-
ates broadcast performance as a result of collisions and
congestion.
Further, emerging classes of networks use broadcast as
a primary data delivery mechanism. Vehicular Ad-hoc
NETworks (VANETs) are a class of wireless networks
in which mobile vehicles communicate with each other
in an ad-hoc fashion. VANET applications such as traf-
fic data dissemination use broadcast exclusively as the
M. Slavik and I. Mahgoub are with the Department of Computer and
Electrical Engineering and Computer Science, Florida Atlantic University,
Boca Raton, FL, 33431 USA e-mail: mslavik@fau.edu, mahgoubi@fau.edu.
Digital Object Indentifier 10.1109/TMC.2012.42 1536-1233/12/$31.00 © 2012 IEEE
1
means of communication, so solving the broadcast storm
problem in an efficient way is critical.
The distribution of nodes in VANETs can be highly
variable. Network density can range from one or two
neighbors to very high densities. Dedicated Short Range
Communications (DSRC), a layer 1 and 2 standard
based on IEEE 802.11 designed for vehicle-to-vehicle and
vehicle-to-infrastructure communications, allows trans-
mission ranges up to 1000m. Even with a more modest
transmission radius of 200m, the number of vehicles in
range of any point on a 4-lane road with cars separated
by 20m is 80.
The distribution pattern, independent of density, can
also vary. In isolated highways, nodes will be con-
strained along one dimensional paths. In an urban area,
nodes may appear more uniformly distributed in two
dimensions. Road crossings and suburban areas exhibit
many combinations in between. Broadcast protocols
should be able to function equally well in all these
scenarios.
Another problem in wireless communications gener-
ally and VANET in particular is loss of packets as they
traverse the medium. Multi-path fading causes a signal
to interfere with itself as it splits into multiple paths
due to being reflected off objects in the environment.
Packets can also be lost when different transmitted
signals interfere with each other, a phenomenon called
a collision. These effects vary in intensity across space
and time and can work against the design of multi-hop
broadcast protocols. If a broadcast protocol is designed
and tested in an environment that assumes perfectly
reliable communications, it could break down when
fading and collisions are introduced.
Often, broadcast protocols must accommodate packet
communication failures by increasing the number of
nodes that are rebroadcasting source messages. Ideally,
this increased usage of bandwidth is the minimum nec-
essary to retain the level of connectivity required by the
application. Practical VANET broadcast protocols should
accommodate fading and collisions while minimizing
redundant source message retransmissions.
Wireless broadcasting methods can be classified into
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING
at least statistical and topological methods [2]. Topo-
logical protocols use the network topology (knowledge
of the identity of nodes in the 2-hop neighborhood) to
select nodes to forward broadcast messages. The key
advantages of topological protocols are that they reduce
broadcast redundancy in a way that is agnostic to node
distribution and they can be proven to not degrade
reachability compared to flooding, provided communi-
cations are perfectly reliable [3].
However, topological protocols have disadvantages.
The density and speed of vehicles in VANET means that
the network topology will be highly variable. Consider
the rather modest scenario of a single 2-lane street in
isolation with vehicles traveling in both directions at
a speed of 50 kph (32 mph) and separated by 20m.
Half of each node’s neighbors (the ones traveling in the
same direction) are roughly constant. However, the set
of neighbors traveling in the opposite direction changes
every 715 ms. If a protocol needs accurate neighborhood
information to operate properly, messages used to dis-
cover the topological data will need to be sent at a high
rate to keep the local neighborhood information up-to-
date.
Statistical broadcast protocols typically do not use this
rapidly changing neighborhood information. Statistical
protocols measure the value of one or more locally
available variables and make a decision to rebroadcast
based on the measured value and a cutoff threshold. For
example, the statistical method used in this work, the
distance method, measures the distance to the nearest
neighbor from which a node has received the broadcast
message. If that distance is greater than a threshold
value, then the node rebroadcasts the message. The
threshold value is calculated as a function of more slowly
changing topological factors such as node density and
spatial distribution pattern.
Another qualitative advantage of statistical methods
is with respect to privacy. Privacy is a critical aspect
of vehicular networking systems. Data broadcasted in
VANET applications, such as vehicle speed and location,
must not be traceable. Therefore security mechanisms
need to be built in to obscure or hide node identifica-
tion. Multi-hop broadcast protocols built using statistical
methods ease the development of these mechanisms,
because statistical methods generally do not require
any neighborhood information and thus node identifiers
are not required to be sent with data and overhead
messages. By contrast, topological methods use node
identifiers extensively. Privacy preserving schemes may
still be required when using statistical methods, but the
statistical method will complement and reinforce the
privacy mechanism.
The downside of statistical protocols is that they are
stochastic in nature and usually cannot be proven to al-
ways give full reachability. Instead, simulation modeling
is done to verify that the protocols exhibit acceptable per-
formance in a target percentage of randomly distributed
networks.
2
It is possible to design a statistical protocol, as we do
in this work, that shows high efficiency while achieving
desired reachability. The key to accomplishing this is in
the design of the threshold value. If the threshold value
is too aggressive, the protocol will give low coverage.
If it is too conservative, the broadcast efficiency will
be low. An optimal value of the threshold must be
found that gives the best efficiency possible without
sacrificing reachability. Further, the optimal value of the
threshold varies with node density, distribution pattern,
and channel quality. A main objective of this work is to
demonstrate how to design a threshold function that is
adaptive to a wide range of these factors. This threshold
function is then used in the proposed DADCQ protocol.
The contributions of this work are summarized as
follows:
• DADCQ Protocol: The primary contribution of this
work is the proposed multi-hop broadcast proto-
col DADCQ. DADCQ combines local spatial dis-
tribution information and other factors with the
distance method heuristic to select rebroadcasting
nodes. Previous broadcast protocols proposed for
ad-hoc that make use of the distance method use
less comprehensive supplemental information. A
key insight proposed here is a methodology for
incorporating more information into the protocol.
This extra information is used to make the protocols
adaptive to more networking scenarios than many
previous proposals.
• Adaptive Threshold Function Design: This paper gives
a novel design strategy for a decision threshold
function. Threshold functions are a critical com-
ponent of many multi-hop broadcasting methods,
such as stochastic broadcast (gossiping), the counter
method, the distance method, and the location
method [4]. The proposed design scheme builds
a threshold function using three independent in-
put variables chosen to allow the threshold to be
adaptive to the environmental conditions of pri-
mary interest. These variables measure local node
density, the distribution pattern of nearby nodes,
and the wireless channel quality. Thus the resulting
threshold function of these three inputs causes the
protocol to operate efficiently across a broad range
of conditions.
• Quadrat Method for Spatial Distribution Characteriza-
tion: One of the factors used to compute the rebroad-
casting decision threshold in DADCQ is local node
distribution pattern. This work is the first to pro-
pose that the quadrat method of spatial statistics be
used to characterize the spatial distribution of nodes
for use in a multi-hop broadcast protocol. Because
distribution pattern may affect the behavior of many
multi-hop broadcast methods, this contribution may
be applicable in a wider context as well.
• Analysis of Behavior with Respect to Threshold Value:
This paper presents original results addressing the
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING
nature of the threshold value used in the distance
method. We show that system performance exhibits
a phase transition with respect to the threshold
value and suggest that an optimal threshold value
should cause the system to operate in the supercrit-
ical region of performance as close to the transition
region as possible.
In Section 2 we will discuss related work to describe
where this work fits into the literature. Section 3 de-
scribes the distance method and statistical broadcast
protocol design generally. Sections 4 and 5 show how
to build adaptivity to distribution pattern and channel
quality into a protocol that uses the distance method.
Section 6 describes multi-hop broadcast protocols that
will be used for comparison. Section 7 discusses the
collection of supplemental information used by protocols
to make rebroadcasting decisions. Section 8 presents sim-
ulation results and Section 9 gives concluding remarks.
2 R ELATED W ORK
Multi-hop broadcast in VANET mainly supports two
types of broadcast applications. One type is safety ap-
plications such as Collective Collision Avoidance (CCA)
where in the event of an emergency an alert message is
propagated as quickly as possible into oncoming traffic.
Broadcast protocols for these applications are usually
designed to propagate messages in one or two directions
with high reliability and minimum latency. Examples of
these proposals are [5], [6], and [7].
The focus of this work is the second type of VANET
broadcast application such as traffic data dissemination.
These applications require data to be disseminated to a
large area as efficiently as possible, with less stringent
reliability and delay requirements than CCA-type appli-
cations. Protocols to support these applications are more
like traditional MANET broadcast protocols, except they
must handle high node mobility, variable node distribu-
tion, and poor channel quality of VANET environments.
Multi-hop broadcast protocols can generally be clas-
sified into statistical and topological protocols by the
type of information used to select forwarding nodes.
The fundamental statistical methods (stochastic, counter,
distance, and location), are described in [1]. Several pro-
tocols have been built using these methods that are adap-
tive to node density. Several papers have addressed the
need to make stochastic broadcast adaptive to density
[8] [9]. Density-adaptive versions of counter are given
in [10] [11]. Density-adaptive versions of both counter
and location methods are given in [4]. We described a
method using the link between continuum percolation
and wireless broadcast to make stochastic broadcast
adaptive to density in [12].
For VANET in particular, several proposals for broad-
cast protocols have been made. Tonguz et al, have pro-
posed forwarding node selection algorithms based on
the fundamental statistical methods mentioned above
[13] [14] [15]. Of these, we have found p-persistence
3
gives the best performance so we use it as part of our
evaluations here. A more comprehensive study of these
methods in highway scenarios is rolled into a scheme
called DV-CAST [16]. DV-CAST has been shown to
have performance issues in urban environments, so the
authors propose a variant called UV-CAST designed for
urban environments [17]. Our work focuses on building
a multi-hop broadcast protocol that performs well in a
wide range of node distribution patterns.
Bako et al adopted a stochastic broadcast method orig-
inally proposed in [18] for VANET [19]. From that body
of work we use Advanced Adaptive Gossiping (AAG)
in our evaluations. AAG is a hybrid protocol that uses
2-hop topological data to set forwarding probabilities
used as in a traditional stochastic broadcast algorithm.
p-persistence and AAG are described in more detail in
Section 6. Simulation results comparing these protocols
to the proposed solution are given in Section 8.
Osafune et al propose MHVB [20] and E-MHVB [21],
which use the distance method with constant threshold
value to suppress rebroadcasts. The proposed protocol,
DADCQ, also makes use of the distance method, but
with a highly adaptive threshold value. Finally, Korkmaz
et al propose Urban Multi-hop Broadcast (UMB) [22],
which like the CCA broadcast protocols is fundamentally
uni-directional, but does contain provisions for branch-
ing at intersections to cover wider areas.
Multi-Point Relaying (MPR) [23] is a well-known
broadcast protocol in MANET that uses network topol-
ogy information to select rebroadcasting nodes. An ex-
tension of MPR that is more accommodating to high
mobility called E-MPR is proposed in [24]. We use
another variation of MPR, called Double Covered Broad-
cast (DCB) [3], in our evaluations here because it was
designed to accommodate transmission errors, which is
one of the topics of this paper.
3 S TATISTICAL B ROADCAST P ROTOCOL D E -
SIGN
The following three sections (3, 4, and 5) guide the reader
through the design of the proposed protocol, DADCQ.
First, we give background information on a heuristic
used by DADCQ called the distance method, including
an original analysis of system behavior with respect to
the forwarding decision threshold value. Also included
in this discussion is an overview of the general threshold
value design procedure and WiBDAT, a specialized tool
we designed for the purpose of statistical broadcast
protocol design [25].
Wireless broadcasting methods can be divided into at
least statistical and topological methods [2]. Statistical
methods use a priori knowledge of the impact of an
external variable (examples listed below) present at each
node to decide whether or not to rebroadcast. These
methods share a common framework where nodes mea-
sure the value of the variable at their location, calculate a
threshold value, and make a rebroadcast determination
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING
based on whether or not the value of the variable exceeds
the value of the threshold. Several local variables have
been proposed for use with this framework, most due to
Ni et al [1]. Here we use a technique called the distance
method that is designed to select nodes to rebroadcast
that will add the most coverage area (see Section 3.2).
3.1 Outline of Design Methodology
In the following sections we will present the design of
an adaptable multi-hop broadcast protocol built around
a statistical broadcast method. We begin with the basic
distance method heuristic, then build in adaptability to
node distribution pattern and channel fading intensity
to create the DADCQ protocol. The behavior of distance
method protocols is largely determined by a particular
protocol parameter called Dc . The following sections will
develop a function to map locally measured variables
to experimentally determined good values for Dc . This
function is referred to throughout the paper as the
threshold function. To reduce design time, we use a high-
level simulator called WiBDAT [25] (discussed further
below). It is important to note that this tool produces less
accurate results than a full network simulator such as ns-
2 (which in turn produces less accurate than real-world
experiments), so all values of Dc presented below are
approximations only. Further, the functional forms and
function parameters selected to model those values of Dc
(e.g., Equation 1) are themselves only approximations,
typically chosen to balance computational and curve
fitting complexity with accuracy in fitting the estimated
values. The result is that the final threshold function for
DADCQ is clearly only a rough model of the “true”
optimal values of Dc . In the simulation results section
(Section 8), we use JiST/SWANS, a full network simula-
tor, to verify the protocol in as accurate an environment
as possible and gauge the performance of the threshold
function built using these approximations.
3.2 Distance Method
At its core, the proposed DADCQ protocol utilizes the
distance method to select forwarding nodes. The dis-
tance method uses the minimum distance from sender
to receiver (1-hop distance) as the variable of discrim-
ination between rebroadcasters and non-rebroadcasters.
The method appeals to the intuition that if a node has
received a message from another node very close to it,
there is little benefit in terms of additional coverage
achieved by rebroadcasting. Nodes then should favor
rebroadcasting when this distance is large.
The algorithm is simple and is given here.
1) Initialize D = 1
2) When a message is received, set d to the distance
to the sender, D = min{D, d/r}, and set a random
backoff timer
3) If a message is received during the backoff, repeat
2 Dc (N ) = Dmax − βeαN
4
4) When the backoff expires, rebroadcast if D > Dc
The key to any statistical broadcast protocol is the
decision cutoff, in this case Dc . If the value is set too
high, reachability will be degraded. If the value is set
too low, the protocol will not prevent many nodes from
rebroadcasting. Thus protocol behavior exhibits a phase
transition relative to Dc [26]. In the supercritical phase,
when Dc is less than the critical value, reachability is
nearly one in all simulation runs. Around the critical
value, the reachability quickly transitions from one to
zero and the variance of the reachability spikes, indicat-
ing the reachability is highly variable from run to run. In
the subcritical phase, message propagation is suppressed
so that only nodes near the source will receive the
message.
The design trade-off is with bandwidth consumption.
When Dc is set deep into supercritical phase, the protocol
will propagate messages to the entire network with high
reliability, but it comes at a cost of decreased broadcast
efficiency. Our goal then is to set Dc as high as possible
while maintaining acceptable reachability.
This optimal value for Dc varies with node density
(among several factors). At low densities, Dc must be
lower to allow the message to propagate to the entire
network. When the node density is high, we can set
Dc more aggressively to eliminate excess transmissions.
Therefore, we use the following procedure to find Dc as
a function of N , the number of neighbors of the node
(also referred to as local density).
1) Find the optimal value of Dc at many values of
the global density λ. The optimal value of Dc is the
maximum value for which the reachability remains
acceptable
2) Plot the measured optimal values and find suitable
approximation function Dc (λ).
3) Substitute local density N for global density λ to
get Dc (N ).
4) Test Dc (N ) to ensure reachability is acceptable.
5) Adjust Dc (N ) as necessary.
Here we define acceptable reachability as broadcast
messages are propagated to at least 99% of all nodes
95% of the time. In previous work, we found that using
an exponential curve for Dc (N ) provided good perfor-
mance across the range of N [2] [27]. This determination
was made by empircally finding optimal values of Dc vs
N while holding node distribution pattern and channel
quality constant. For the range of node distributions and
channel qualities reasonable in VANET (see Section 8)
we find that the curve of Dc vs N roughly follows an
exponential form. This indicates that for low node den-
sities, the threshold should be low to allow many nodes
to retransmit. As the density increases, the threshold can
quickly become more aggressive to prevent unnecessary
rebroadcasts. We therefore use the following general
form for the threshold function.
(1)
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING
The values of the function parameters Dmax , β, and α
depend on external factors such as node distribution pat-
tern and channel quality. In Sections 4 and 5 we propose
methods of finding good values for these parameters
across a wide range of network scenarios.
3.3 The WiBDAT Design Tool
The threshold curve design procedure described above
requires that a multi-hop broadcast system be evalu-
ated in many different scenarios. We are unaware of
any purely analytical models capable of accurately de-
scribing the behavior of multi-hop wireless broadcast
systems, particularly when fading and collisions are
considered. Therefore, we will use simulation to evaluate
threshold values.
If a detailed simulator such as ns-2 [28] or
JiST/SWANS [29] were used to perform these evalua-
tions, the design process would be prohibitively long.
WiBDAT, a fast high level multi-hop broadcast simula-
tion tool, has been developed specifically to support this
design process [25]. WiBDAT uses stochastic models to
represent fading and collisions so that only the network
layer broadcast protocol is simulated directly, making
the simulator fast and scalable. A WiBDAT simulation
run places nodes in a field as specified then selects a
node near the center to be the source to originate a
broadcast. When the broadcast propagation is complete,
the simulation ends.
In Sections 4 and 5, we present threshold values for
various node densities, distribution patterns, and fading
models. These results are all obtained using WiBDAT.
Finally, in the evaluation section (Section 8) we switch
to JiST/SWANS to verify the results.
4 A DAPTATION TO D ISTRIBUTION PATTERN
In the next two sections we will build adaptivity into our
proposed protocol, DADCQ. First, this section focuses on
node distribution pattern. The next section extends the
results for channel quality adaptivity.
For the distance method, adaptivity to node distri-
bution pattern means that the rebroadcasting threshold,
Dc , somehow varies as needed when the distribution
pattern changes. We observed that a good functional
form for Dc is given in Equation 1, which depends on
three parameters: Dmax , β, and α.
We applied the design procedure described in Section
3.2 to find good values for these parameters in two dif-
ferent scenarios: once with nodes uniformly distributed
in the entire field, and once with nodes uniformly dis-
tributed along a line. In each scenario, we ran WiBDAT
simulations at several node densities to find the largest
value of Dc that provided at least 99% reachability in
95% or more of the simulation runs. Next we fit Equation
1 to these measured values by hand-selecting a value for
Dmax then performing a least-squares fit to calculate α
and β.
5
TABLE 1
Practical efficient values for Dmax , α, and β for 1-D and
2-D uniform distributions
Distribution Dmax α β
1-D Uniform 0.90 -0.039 1.0
2-D Uniform 0.90 -0.099 1.2
Fig. 1. Threshold value vs density for 1-D and 2-D uniform
distributions
Table 1 shows the calculated values of the function
parameters found using this method. Figure 1 com-
pares the threshold curves for one-dimensional and two-
dimensional distributions visually. Clearly, the threshold
value must be less aggressive (lower) when the network
is confined to one dimension. The parameter that differs
significantly between the two scenarios is the exponen-
tial rate multiplier α.
Since the 2-D threshold curve is more aggressive, if it
is used in linearly-oriented networks, reachability will
not meet the required target. On the other hand, the
1-D threshold curve is more conservative, so while it
should provide adequate reachability in both 1-D and 2-
D networks, the broadcast efficiency in 2-D networks will
be lower than it could be if a better threshold function
were used. Ideally, we would like a threshold curve that
provides optimal performance in both cases.
4.1 Measuring the Spatial Distribution
Since practical protocols to support a broadcast applica-
tion in VANET must support both one-dimensional and
two-dimensional distributions, we need to find a way
to adapt the threshold function dynamically to either
scenario. To do this, we first need a method for nodes
to dynamically measure the distribution pattern so they
can set the threshold function accordingly. The science
of analyzing spatial data is called spatial statistics or
geographical information systems. We propose using a
technique from this discipline called the quadrat method
to characterize the node distribution pattern.
The quadrat method attempts to measure how evenly
spaced a set of points are. A point distribution can either
be regularly spaced, such as in a lattice, positioned in
clumps, or somewhere in between. The quadrat method
addresses the problem by examining the frequency pat-
tern of the distribution [30]. There are two basic quadrat
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING
6

(a) Q < 1 (b) Q ≈ 1 (c) Q > 1

Fig. 2. Illustration of quadrat method statistic Q in 2-D fields. Small values of Q indicate even spacing while large
values of Q indicate clustering.

methods: quadrat censusing and quadrat sampling. Cen-

susing divides the space into equally sized cells and
counts the number of points in each cell. The result of
the census, the frequency array nm , indicates for each
count m, how many cells have m nodes in them. For
example, if n3 = 5, then 5 cells contain 3 nodes. In
sampling, a cell is placed at a random location and the
nodes within it counted some number of times. Like
censusing, sampling produces a result frequency array
nm .
We have found that in the wireless broadcast context, Fig. 3. Q vs node density for 1-D and 2-D uniform
when the area being measured (the node’s transmission distributions
area) and the number of sample points (neighbors) are
small, sampling seems to work better than censusing.
mean. If Q < 1, then the variance of the frequencies
Thus results shown here use sampling. Selecting values
is less than the mean, indicating the nodes are more
for the number of samples and the cell size is com-
evenly spaced than is expected in uniformly distributed
plicated and does not have a one-size-fits-all solution.
networks. If the nodes are clustered together, then the
Through experimentation, we have chosen a sample size
variance will be larger than the mean and the value
of 30 and a circular cell with radius r/5 (where r is the
of Q will be greater than one. Figure 2 illustrates these
transmission radius).
different ranges of Q in two-dimensional fields.
The process of calculating the frequency values nm
Here we are primarily interested in using Q to dis-
then is as follows:
tinguish 1-D oriented uniform distributions from 2-D
1) Initialize nm = 0 for m = [0, N ] oriented ones. Figure 3 shows Q vs density for the 1-D
2) Generate random point x, y within the node’s and 2-D uniform distribution patterns. As expected, the
transmission area, set m = 0 value of Q is nearly one across all densities for the 2-D
3) For each neighbor and the node itself, if it is within distribution. When viewing the field as 2-D, as is done
r
5 of x, y, increment m in the calculation of Q, nodes distributed along a line
4) Increment nm appear to be clustered together. Thus the 1-D oriented
5) Repeat steps 2 - 4 thirty times uniform distribution has Q > 1.
Once the frequency array is computed, we need to
create a single value to represent the result. If the distri- 4.2 Quadrat-Based Threshold Function Design
bution is 2-D uniformly random, then the values in nm
Finally, we will incorporate the spatial statistic Q into
will follow a Poisson distribution. One characteristic of
our protocol. We will use Q to compute a threshold
the Poisson distribution is that the mean is equal to the
value that works well for both 1-D and 2-D networks.
variance. Thus we define a metric that is the ratio of the
First, note that of the three parameters in Equation 1,
variance to the mean:
only α needs to vary based on the node distribution.
V [n] Specifically, it should be roughly 3 times greater in the
Q= (2) 2-D case than in the 1-D case (see Table 1).
E[n]
Our strategy then is to find a function of Q that is 1
When the distribution is 2-D uniformly random, the in the pure 2-D case and 3 in the pure 1-D case. We will
value of Q is near one, since the variance is equal to the divide the base α from the 2-D case (approximately −0.1)
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING
by that value to get the final α. Observe that the value
of Q for 1-D networks closely follows this equation.
Q1D ≈ 1 + 0.125N
We can use this approximation to create a linear
mapping (Equation 4) that will range from α2D in pure
2-D networks to α2D /3 pure 1-D networks. Here α2D is
the optimal α from the 2-D experiments (Table 1).
 −1
2(Q − 1)
α = α2D +1 (4)
0.125N
When the network is 2-D distributed, Q ≈ 1 and
α ≈ α2D . When the network is 1-D distributed, Q ≈
0.125N + 1 and α ≈ α2D /3. These values roughly match
the desired values in Table 1. Other values of Q will be
mapped linearly between the pure 1-D and 2-D value.
Using WiBDAT, we cannot easily generate random node
distributions with arbitrary values of Q, so we do not
have samples between the pure 1-D and pure 2-D cases.
Thus the linear form is chosen to fill the gap between
pure 1-D and pure 2-D due to its simplicity. The resulting
equation for Dc is shown by Equation 5.
−1
Dc (N, Q) = 0.90 − 1.2e−0.099( 0.125N +1)
2(Q−1)
N
5 A DAPTATION TO C HANNEL Q UALITY
Finally, this section incorporates adaptability to channel
quality into the proposed protocol. The design described
in the previous section assumes that the wireless channel
and the medium access control protocol deliver mes-
sages between nodes located within transmission range
of each other with perfect reliability. In practice, wireless
signals in the system interfere with themselves and with
each other in unpredictable ways, leading to apparently
non-deterministic message reception.
When two nodes transmit messages at the same time,
the wireless signals may interfere and cause one or both
of the messages to be lost at a destination node. Fading,
the phenomenon where multiple parts of the same signal
traveling along different paths interfere with each other,
degrades communications even when only a single node
is transmitting.
Multi-hop wireless broadcast protocols must be able to
operate effectively even when communication reliability
is poor. What follows in this section is the design of a
decision threshold function that is adaptive to both node
distribution pattern and channel quality, culminating in
the Distribution-Adaptive Distance with Channel Qual-
ity (DADCQ) protocol.
5.1 Modeling Fading
First, we need a way to model multi-path fading in
WiBDAT so that we can find good threshold values in the
presence of fading. In VANET, fading may be modeled as
Rician fading [31]. In Rician fading, the received signal is
assumed to have a strong component from line-of-sight
7
among scattered multi-path signals. Note that others
have disputed using a standard fading model such as
Rician or Nakagami fading in VANET [32]. Clearly,
(3) movement of vehicles causing temporary disruptions
in line-of-sight communications can have large fading
effects that are not captured by these standard models.
However, for our purposes we need to adopt a single
model of fading so that the fading intensity can be
quantified and used to adapt the broadcast threshold
value. The Rician model allows the fading intensity to be
easily quantified and has been accepted in the literature
as a reasonable model of fading in VANET [31] [33].
A parameter, K, quantifies the relative strength of
the line-of-sight signal in Rician fading. When K = 0,
also called Rayleigh fading, there is no strong line-of-
sight signal. When K = ∞, fading has no effect thus
all nodes within transmission range of each other can
reliably communicate.
Several studies have estimated the value of K in
VANET scenarios. In [31], the value is measured in the
range 5 to 11 for communications in the 900 MHz band.
At the 5.3 GHz carrier, researchers have measured K
between 8 and 13 [33]. Standardization of VANET com-
munications is coalescing around IEEE 802.11p (DSRC),
(5) which uses the 5.85 - 5.925 GHz band. Our evaluations
will use Rician fading with K from 0 to 20.
5.2 Designing a Threshold Function using K
In Section 4, we found good values of Dmax , α, and β
when nodes were laid out in a line and in a rectangle,
then fit the curve of Dc to these cases using Q to
differentiate them. We will extend that process here,
finding good values of Dc in 1-D and 2-D networks for
various values of K.
Previously, we found that α in the 2-D scenario is
roughly 3 times the value in the 1-D scenario. We will
reuse this result here and restrict the value of α in each
case to follow this pattern and then fit the value of Dmax
and β using least squares optimization. Table 2 shows the
resulting values of the threshold function parameters for
average neighbors ranging from 8 to 30 in the 2-D case
(Q ≈ 1) and 15 to 70 in the 1-D case (Q ≈ 3 to 10).
Next we need to find expressions for Dmax , α, and β
in terms of Q and K. Finding a perfect fit is difficult,
so we make two simplifications. First, the value of Dmax
is roughly the same for the 1-D and 2-D cases, so Dmax
can be independent of Q. Second, β is about 3 times
larger in the 2-D case than the 1-D case, so we can use
a similar linear mapping between them as we use for
α. These simplifications allow us to use a simple one-
dimensional least squares curve fit to find expressions
of Dmax and β. Results are shown in Equations 6 - 8.
 −1
2(Q − 1)
α = −0.200 +1 (6)
0.125N
 −1
−0.1K 2(Q − 1)
β = 6.25e +1 (7)
0.125N
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING
8

TABLE 2 parameter K. N and Q can be easily calculated using

Practical efficient values for Dmax , α, and β for 1-D and overhead messages exchanged between nodes (see also
2-D uniform distributions with varying channel quality Section 7). K, however, cannot be easily calculated by
nodes dynamically.
Distribution K Dmax α β
For dynamically measuring K, nodes in practice will
1-D Uniform 0 0.543 -0.067 2.25
have several options. First, designers may be aware of
1-D Uniform 5 0.561 -0.067 2.29
a worst case (minimum) value of K that nodes can use
1-D Uniform 10 0.634 -0.067 1.84
in all circumstances. Results in [31] and [33], for exam-
1-D Uniform 15 0.690 -0.067 2.01
ple, show K is almost always greater than 5. Statically
1-D Uniform 20 0.712 -0.067 1.99
employing this value of K, reachability should always
2-D Uniform 0 0.553 -0.200 6.30
exceed the application requirements, provided the true
2-D Uniform 5 0.604 -0.200 5.68
value of K is not less than the assumed minimum.
2-D Uniform 10 0.645 -0.200 4.95
However, since the value of K may often be higher
2-D Uniform 15 0.678 -0.200 4.31
than the minimum, this method may result in excessive
2-D Uniform 20 0.727 -0.200 4.19
rebroadcasts and thus inefficient use of wireless band-
width.
Since vehicles equipped with VANET capabilities will
Dmax = 0.75 − 0.237e−0.1K (8)
likely also have extensive mapping data used by on-
These equations, along with Equation 1, define Dc as board navigation systems, it may be possible to improve
a function of the three independent variables N , K, and on this static value of K by encoding minimum values of
Q, each measured independently at each node. N is the K at different geographic locations. For example, fading
number of neighbors, K is the Rician fading parameter will likely be worse in urban areas with tall buildings
(see Section 5.3), and Q is the spatial distribution statistic. than in flat open areas. These geographic influences on
Figure 4 shows various slices of this decision surface. K can be measured and encoded in mapping data for
Each subfigure shows a plot of Dc varying two param- use by the broadcast protocol.
eters while holding the third constant. At each node, Finally, although nodes probably cannot accurately
when the measured value of D (distance to the nearest measure K dynamically, they may be able to use the
neighbor from whom the broadcast message has been measured node distribution to provide some feedback
received) is above the curve, the node will retransmit to the value of K. For a given value of N , Q, and K,
the message. Thus higher values of Dc imply that less nodes will know about how many overhead messages
nodes in the system will rebroadcast. The plots clearly (beacons) can be expected from neighbors at various
show that increasing N and K causes Dc to increase, distances away. So if, for example, the number of beacons
indicating that as node density increases or the channel received from neighbors at the edge of the transmission
quality improves, less nodes need to rebroadcast. Q is range is much different than expected, the node can
the opposite; as the node distribution becomes more adjust its assumed value of K to reflect this observation.
clustered, more nodes need to rebroadcast. In the evaluations in later sections, we assume nodes
The Distribution-Adaptive Distance with Channel know K and can use it to calculate the rebroadcast
Quality (DADCQ) protocol can then be summarized threshold for DADCQ.
as follows. For every independent broadcast message,
nodes perform the following procedure:
6 B ROADCAST P ROTOCOLS U SED FOR C OM -
1) When a broadcast message is received for the first
PARISON
time, calculate d as the distance to neighbor the
message was received from, and initialize D = d/r. In this section we will briefly describe the three other
2) Set a random backoff timer with duration chosen protocols used in the evaluation with DADCQ: p-
uniformly between 0 and Tmax . persistence, DCB, and AAG. p-persistence and AAG
3) If the message is received again before the timer are both well-known broadcast protocols proposed for
expires, calculate d (distance to the sender) and VANET that have several papers published about them.
update D = min{D, d/r}. Cancel the timer and We choose DCB because it is a variant of MPR that better
goto 2. handles transmission errors. MPR with some extensions
4) When the backoff expires, calculate for mobility has been proposed for use in VANET [24].
t = Dc (N, Q, K) according to Equations 6 - 8
and 1 and rebroadcast if D > t 6.1 p-persistence
The first protocol we will use for comparison is p-
5.3 Notes on Finding K persistence, originally proposed by Tonguz et al [13].
DADCQ calculates the rebroadcasting threshold value p-persistence is interesting because it utilizes three dif-
as a function of N (the number of neighbors), Q (the ferent statistical methods simultaneously. Foremost, p-
distribution clustering metric), and the Rician fading persistence uses the counter method with a threshold
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING
(a) K = 12
Fig. 4. Illustration of DADCQ rebroadcast decision surface. Subfigure caption indicates the parameter being held
constant in the plot.
value of one. When a node receives a message, it first
sets a backoff timer and waits to see if the message is
received again. If it is, the retransmission is canceled. If
not, the node rebroadcasts.
Calculation of the backoff period involves two other
statistical methods. First, nodes decide if they are high
priority rebroadcasters according to a fixed probability p.
In the evaluations that follow, we follow Wisitpongphan
et al [15] and use p = 0.5. Nodes that do not elect them-
selves as high priority rebroadcasters set a backoff time
of T = Ns τ , where Ns and τ are protocol parameters.
Finally, nodes that elect themselves to have high prior-
ity use one-hop distance to the sender to bias the backoff
timer, similar to the distance method. Specifically, the
backoff time is calculated as T = Ns (1 − d/r)τ , where
d is the distance to the node from whom the message
was received and r is the transmission radius.
6.2 Double-Covered Broadcast
The next two protocols, DCB and AAG, use network
topology data to select forwarding nodes. To describe
these topological protocols, we need to provide some
basic definitions (this notation is adopted from [3]).
Nk (v) is the set of all nodes within k hops of node v,
including v itself. Hk (v) is the set of nodes exactly k
hops from v.
Double-Covered Broadcast (DCB) [3] is based on
Multi-Point Relaying (MPR) [23]. MPR is a dominant
pruning broadcast algorithm used as the broadcasting
mechanism in the popular Optimized Link State Routing
(OLSR) protocol [34]. In dominant pruning, forwarding
nodes, beginning with the source, select which of their
neighbors should be the next forwarding nodes (this
stands in contrast to self pruning, where nodes decide
for themselves whether or not to rebroadcast). This
forwarding node list is attached to each data message,
so when a node receives a message, it first checks if
it is in the forwarding node list. If it is, it calculates
which neighbors should also be forwarding nodes and
retransmits the message. Otherwise, it does nothing.
In MPR, when forwarding node v receives a message,
it calculates the next set of forwarding nodes F using a
9
(b) Q = 1 (c) N = 20
simple greedy approach called the Forwarding Node Set
Selection Process (FNSSP).
1) Initialize X = H1 (v), U = H2 (v), and F = ∅
2) Find w in X with maximum degree deg(w) =
|N1 (w) ∩ U |
3) Update F = F ∪ {w}, U = U − N1 (w), and X =
X − {w}
4) Repeat 2,3 while U = ∅
DCB uses the FNSSP procedure with modified values
for X and U to achieve various goals. X can be seen as
the set of nodes from which v chooses forwarding nodes,
and U is the set of nodes that need to be covered. Since
v will cover H1 (v) by broadcasting itself, only nodes
in H2 (v) need to be covered by the forwarding nodes
designated by v. Note that as long as the network is
connected, this algorithm will guarantee all nodes in the
network receive the broadcast message, provided there
are no transmission errors.
In practice, of course, there are often transmission er-
rors. DCB attempts to account for link errors by making
the protocol more reliable. It does this in two ways.
First, nodes use a double-coverage version of FNSSP to
select forwarding nodes. Node v sets X = H1 (v) and
U = N2 (v) − {v} and then executes the FNSSP, which
results in all nodes in H2 (v) being covered once by a
broadcast and all nodes in H1 (v) being covered twice.
Since each node will receive the message at least twice,
they have a better chance of receiving it.
Second, DCB introduces a retry mechanism. When
forwarding node v rebroadcasts a message P , it will
have selected a set of forwarding nodes F (v) ⊆ H1 (v).
After it transmits P , v starts a timer and waits to receive
messages. If it has not overheard the rebroadcast of
each of the forwarding nodes in F (v) by the time the
timer expires, it retransmits the message P again. It will
continue this process until it overhears all the nodes in
F (v) rebroadcast or it reaches the maximum of R retries.
Following the suggestions of the authors of DCB, we use
the value R = 1 in all evaluations.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING
6.3 Advanced Adaptive Gossiping
Advanced Adaptive Gossiping (AAG) [19] is a hybrid
between pure topological protocols like MPR and DCB
and a stochastic broadcast protocol. AAG uses 2-hop
topological information to calculate a rebroadcast prob-
ability p.
First, when node v receives a message from node u,
let C(v) = H1 (v)−H1 (u) be the set of neighbors of v that
are not covered by u, called the children of v. For each
child node c ∈ C(v), v then determines the parent set of
c, defined as P (c) = H1 (c) ∩ H1 (u), the set of nodes that
could potentially cover c by rebroadcasting.
Next, for each child node ci , v calculates a forwarding
probability pi = prob(|P (ci |) where prob is defined
below. The final forwarding probability used to make
the rebroadcasting decision is then calculated as p =
max{pi }. Like any stochastic broadcast protocol, AAG
rebroadcasts the message if a uniformly random number
between 0 and 1 is less than p.
The prob(n) function used to calculate forwarding
probabilities in AAG is shown in Equation 9. This func-
tion depends on three parameters: τarp , δ, and D. τarp
is an application defined reachability target, which we
set to be 0.99. δ is the network diameter (in number of
hops) and D is the number of neighbors of v.
 find node distribution pattern changes more slowly than
 1/δ 1/n
0.5
prob(n) = 1 − 1 − τarp −
1 + e−0.3(D−14)
7 OVERHEAD B EACONING
Next we will describe the mechanism for communicating
topological information between nodes, called beacon-
ing. The algorithm is very simple, but results in efficient
transfer of information in the network. Nodes broadcast
a short message every Tb seconds called a beacon. Unlike
the multi-hop broadcast messages that get propagated to
the entire network, beacons are broadcast from a node
to its neighbors only and are never retransmitted. Nodes
send beacons at a constant fixed rate, so in any interval
of Tb seconds, every node in the network will send
exactly one beacon. The timing of beacons is in no way
synchronized between nodes, except that all nodes use
the same value of Tb .
Individual nodes set a timer for Tb seconds and track
all the beacons received from neighbors during that time.
At the end of the interval, they compile all the beacons
and transmit their own beacon, which may contain
information from the most recently received beacons.
The information contained in beacons depends on the
propagation algorithm. p-persistence requires no addi-
tional information and so does not send beacons at all.
DADCQ needs to calculate N and Q, so nodes transmit
their current location in the beacon. Nodes estimate N
by simply counting the number of beacons received in
an interval and calculate Q using the relative locations
of all the received beacons. The topological protocols
10
used here, AAG and DCB, require 2-hop neighborhood
information, so they transmit their ID and their most
recent list of neighbors in the beacon.
With DADCQ, beacons contain a location, which we
define to be 12 bytes. We assume network layer ad-
dresses are four bytes long, so in AHBP and DCB, bea-
cons contain 4 bytes for the node’s address and 4N bytes
for the addresses of all the node’s neighbors. Additional
bandwidth is used by MAC and network layer packet
framing.
If the network is static, the beacon interval Tb can be
very long and nodes will still have reasonably accurate
topological data. However, if nodes are mobile, the bea-
con interval needs to be reduced so that beacons received
near the beginning of an interval are not stale by the end
of the interval. VANETs of course are not static, so Tb
needs to be short enough that vehicles maintain accurate
information.
We have found that a good value of Tb for DADCQ is
30 seconds while for AAG and DCB is 1 second. The
difference stems from the varying temporal nature of
the external data needed by the protocols. AAG and
DCB use topological information, so they need up-to-
date data on the identities of nodes in their 2-hop
neighborhood. DADCQ requires measurements of the
node distribution pattern in the form of N and Q. We
network topology, so overhead messaging in statistical
(9) protocols such as DADCQ can be lower than in topo-
logical ones like AAG and DCB.
Finally, it is currently unknown whether the over-
head messaging used by multi-hop broadcast protocols
in VANET will be unique to the broadcast protocol
or whether it will be shared across many components
in the platform. If this information is available from
other subsystems, then beacons should not be considered
additional overhead of the broadcast protocol. If the
information is not available, either because other systems
do not use the same amount of information (ex. 2-hop
topological data) or other systems use different transmit
power, then beacons should be understood as overhead.
Evaluations that follow present results to accommodate
both perspectives.
7.1 Protocol Adaptability Rate
Another important issue when considering the choice of
Tb is the rate at which the protocol can adapt to sudden
changes in node density and distribution pattern. Nodes
continue to use collected topological information that is
up to Tb seconds old, so in the worst case a node will not
be fully aware of a sudden change in topology for Tb sec-
onds. However, beacon reception is randomly staggered
between all nodes, so if a vehicle transitions into an area
with a dramatically different node distribution pattern, it
will begin to receive updated information immediately.
Its internal aggregate measurements of topology will
slowly be updated and improve in accuracy until after
Tb seconds the topological information is up-to-date.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING
Fig. 5. Urban scenario map with overlaid maximum
transmission range
In our experimental results using simulated vehicular
traffic patterns on both real and contrived road maps,
we find that both node density and node distribution
pattern change slowly relative to value Tb = 30s used
for DADCQ. However, it is easy to imagine a sce-
nario where a vehicle moves suddenly from an area
with sustained topology A to an area with sustained
topology B, where A and B are very different. (Note
that nodes do experience dramatic temporary changes
in density and distribution pattern in the simulation
scenarios presented below, but these tend to average out
and do not have significant impact on broadcast protocol
performance.) In cases of rapid change between long-
term average topology, it can be expected that protocol
performance will be negatively impacted for a period of
Tb , with performance slowly improving from when the
change occurs until topological information is entirely
updated. We suggest that sudden dramatic changes in
sustained topology are likely to be rare in practical
vehicular scenarios, so we defer the issue to future study.
8 S IMULATION R ESULTS
In this section we present our comparative simula-
tion analysis of three existing propagation protocols, p-
persistence (“pper” in the results), AAG, and DCB, and
the proposed protocol DADCQ. We use the distribution
of JiST/SWANS [29] produced by Ulm University along
with VanetMobiSim [35], a realistic vehicular mobility
generator. Many other traffic mobility pattern generators
are available, including [36]. We use VanetMobiSim be-
cause it is easy to configure and is commonly used in
vehicular networking research (e.g., [37], [38]).
We evaluate the protocols under two different mobility
scenarios. The first is an urban scenario based on a 2 km
square region of Washington, D.C. shown in Figure 5,
built using TIGER data published by the U.S. Census
Bureau. The second is a highway scenario that restricts
node movements to a 1800m square. Node movement
uses Intelligent Driver Model with Lane Changing with
a given maximum speed. Mobility traces of these sce-
narios created with VanetMobiSim are then used by
JiST/SWANS to generate node movement.
These two mobility scenarios also serve to test a
wide range of node distribution patterns. The highway
11
scenario presents nodes that are primarily aligned in
one dimension and thus appear as clustered to the
two-dimensional quadrat statistic used in DADCQ. The
urban scenario includes both linear sections and node
distributions closer to two-dimensional uniformly ran-
dom.
The network simulation begins with a 60 second warm
up period followed by 1000 seconds of broadcast ac-
tivity. During the active period, a new 128-byte broad-
cast message is originated about every 1 second by a
randomly chosen node. Signal propagation is modeled
with two-ray ground propagation with Rician fading.
Radio parameters are set to mimic IEEE 802.11p (5.89
GHz, 3 Mbps) [39] with about 400m transmission range.
JiST/SWANS models both packet collisions and chan-
nel fading. Nodes use single-channel 802.11 MAC layer
signaling and IPv4 layer 3 addressing.
The maximum RAD time used in DADCQ is 150 ms.
The beacon interval for DADCQ is 30 seconds and for
AAG and DCB is 1 second. Beaconing is disabled for
p-persistence. DCB uses a retry time of 500 ms and max-
imum retries R = 1. p-persistence protocol parameters
are set to p = 0.5, Ns = 5, and τ = 5ms. With AAG,
τarp = 0.99.
In each set of results we show three graphs. The first,
reachability, shows the average fraction of nodes that
receive a broadcast message. The second plot shows the
number of rebroadcasts (not including beacons) per re-
ceiving node. For example, a value of 0.5 indicates there
were 0.5 retransmissions on average for each node that
received the source message. This metric considers how
many nodes were selected to rebroadcast and ignores
overhead due to beacons. The third plot is the number of
bytes sent (including beacons) per node that received the
message. The reachability and rebroadcasts per covered
node plots include 95% confidence intervals.
8.1 Effect of Node Density
The first set of results evaluates the effect of the node
density on the performance of the multi-hop broadcast
protocols. Here we vary the number of nodes from 200
to 400, resulting in densities in the range from 25 to 50.
The urban scenario maximum speed is set to 15 m/s
and the highway maximum speed is set to 30 m/s. The
Rician fading parameter is set to K = 10.
Figures 6 and 7 show the performance of the four pro-
tocols in the urban and highway scenarios, respectively.
The first plot shows the reachability, defined here as the
average fraction of nodes to receive broadcast messages.
Here we see AAG and DADCQ give high reachability
in all cases. p-persistence is somewhat lower, showing
80% of nodes receive broadcast messages in the highway
scenario. DCB is lower still, failing to reach more than
40% on average in the highway scenario.
The second and third plots show the bandwidth usage
measured as the fraction of rebroadcasting nodes and
the number of bytes sent per covered node. In the urban
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING
12

Fig. 6. Performance vs number of nodes of the broadcast protocols in the urban scenario

Fig. 7. Performance vs number of nodes of the broadcast protocols in the highway scenario

Fig. 8. Performance vs K of the broadcast protocols in the urban scenario

Fig. 9. Performance vs K of the broadcast protocols in the highway scenario

 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING
scenario, AAG and DCB select the most rebroadcasters
and transmit the most bytes. On the highway, DCB
selects too few rebroadcasters and so the reachability
is low. p-persistence and DADCQ show similar band-
width consumption, with p-persistence being a little
better across all results. Considering the reachability and
bandwidth results together, DADCQ has an advantage,
showing high reachability while using a similar or less
amount of bandwidth than the other protocols.
8.2 Effect of Fading Intensity
The second set of results investigates the performance of
the broadcast protocols with different levels of fading,
specifically the Rician fading parameter K. Here the
number of nodes is fixed at 250, the urban scenario
maximum speed is set to 15 m/s, the highway maximum
speed is set to 30 m/s. The fading parameter K is varied
from 0 (most intense fading) to 20.
Figures 8 and 9 show the protocol performance in the
two mobility scenarios. Results are in the same pattern
as the density results. The reachability of DCB and the
rebroadcast efficiency of AAG are poor. DADCQ and p-
persistence perform well in both measures, although p-
persistence has fairly low reachability in the highway
scenario. Interestingly, the reachability of p-persistence
seems to be better when fading is more intense.
Here, we see that DADCQ is highly adaptive to the
intensity of fading, which is expected since we designed
the DADCQ decision surface to account for it. When the
fading is most disruptive (small values of K), DADCQ
selects more nodes to rebroadcast in order to compensate
for the reduced connectivity. When K = 0, DADCQ
selects even more nodes to rebroadcast than AAG in
the highway scenario. As the fading intensity reduces
(K gets larger), the decision threshold becomes more
aggressive and the bandwidth consumption is reduced.
9 C ONCLUSION
We have proposed the Distribution-Adaptive Distance
with Channel Quality (DADCQ) protocol for multi-hop
broadcasting in VANET. VANETs exhibit wide variabil-
ity in node density, distribution pattern, and channel
quality. In response, we have shown how to design a
broadcast protocol utilizing the distance heuristic that
is adaptive to these factors. In the past, the distance
method has been made adaptive to a range of node
densities, but not to be efficient across a range of
these three factors simultaneously. DADCQ adopts the
quadrat method of spatial analysis to characterize the
distribution pattern at each node. The resulting metric,
the Q statistic, is combined with local node density N
and the Rician fading parameter K to create a decision
surface used by the distance method.
Simulation results with JiST/SWANS demonstrate
DADCQ provides high reachability and efficient use
of bandwidth in both urban and highway scenarios
with varying node density and fading intensity. DADCQ
13
achieves better than 95% reachability across all network
scenarios while showing efficient use of bandwidth. In
most cases, protocols used for comparison either fail to
match this level of reachability or do so using a higher
level of bandwidth consumption. There are some sce-
narios where DADCQ performance could be improved,
notably in very poor channel quality, and its use requires
positional information, but overall it is shown to be
highly adaptable and efficient.
The performance of protocols such as DADCQ that
use statistical broadcasting methods like the distance
method is highly dependent on the rebroadcasting deci-
sion threshold function. In the past, this dependency has
been a key disadvantage of statistical protocols, as it is
difficult to choose a value for the decision threshold that
works well across a wide range of network scenarios.
In this work we addressed this challenge directly and
proposed a methodology to design a decision threshold
surface dependent on several external factors so the re-
sulting protocol is adaptive to many scenarios. Bridging
this statistical broadcast protocol design gap is important
for VANET as it now opens up a new class of multi-hop
broadcast protocols that are well-suited to the challenges
of VANET.
ACKNOWLEDGMENTS
This work is an extension of papers presented at
GLOBECOM 2010 (adaptivity to channel quality [2]) and
WiMOB 2010 (adaptivity to node distribution pattern
[27]).
R EFERENCES
[1] S.-Y. Ni, Y.-C. Tseng, Y.-S. Chen, and J.-P. Sheu, “The broadcast
storm problem in a mobile ad hoc network,” in MobiCom ’99:
Proceedings of the 5th annual ACM/IEEE international conference on
Mobile computing and networking. New York, NY, USA: ACM,
1999, pp. 151–162.
[2] M. J. Slavik and I. Mahgoub, “Statistical broadcast protocol design
for unreliable channels in wireless ad-hoc networks,” in Proceed-
ings of IEEE Globecom 2010 - Wireless Networking Symposium (GC10
- WN), Miami, Florida, USA, December 2010.
[3] W. Lou and J. Wu, “Toward broadcast reliability in mobile ad
hoc networks with double coverage,” IEEE Transactions on Mobile
Computing, vol. 6, no. 2, pp. 148–163, 2007.
[4] Y.-C. Tseng, S.-Y. Ni, and E.-Y. Shih, “Adaptive approaches to
relieving broadcast storms in a wireless multihop mobile ad hoc
network,” Computers, IEEE Transactions on, vol. 52, no. 5, pp. 545–
557, May 2003.
[5] S. Biswas, R. Tatchikou, and F. Dion, “Vehicle-to-vehicle wireless
communication protocols for enhancing highway traffic safety,”
IEEE Communications Magazine, vol. 44, no. 1, pp. 74–82, Jan. 2006.
[6] F. Ye, R. Yim, J. Guo, J. Zhang, and S. Roy, “Prioritized broadcast
contention control in vanet,” in Communications (ICC), 2010 IEEE
International Conference on, May 2010, pp. 1 –5.
[7] Y. Bi, L. Cai, X. Shen, and H. Zhao, “A cross layer broadcast
protocol for multihop emergency message dissemination in inter-
vehicle communication,” in Communications (ICC), 2010 IEEE In-
ternational Conference on, May 2010, pp. 1 –5.
[8] J. Cartigny, D. Simplot, and J. Carle, “Stochastic flooding
broadcast protocols in mobile wireless networks,” LIFL
Univ. Lille 1, Tech. Rep., May 2002. [Online]. Available:
http://citeseer.ist.psu.edu/525199.html
[9] X.-Y. Li, K. Moaveninejad, and O. Frieder, “Regional gossip
routing for wireless ad hoc networks,” Mob. Netw. Appl., vol. 10,
no. 1-2, pp. 61–77, 2005.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
IEEE TRANSACTIONS ON MOBILE COMPUTING
[10] S. al Humoud, L. Mackenzie, and J. Abdulai, “Neighbourhood-
aware counter-based broadcast scheme for wireless ad hoc net-
works,” in GLOBECOM Workshops, 2008 IEEE, 30 2008-Dec. 4
2008, pp. 1–6.
[11] A. Mohammed, M. Ould-Khaoua, L. Mackenzie, and J.-D. Abdu-
lai, “Dynamic probabilistic counter-based broadcasting in mobile
ad hoc networks,” in Adaptive Science Technology, 2009. ICAST
2009. 2nd International Conference on, 2009, pp. 120–127.
[12] M. Slavik and I. Mahgoub, “Stochastic broadcast for vanet,”
Proceedings of the 2010 Consumer Communications and Networking
Conference, January 2010.
[13] O. Tonguz, N. Wisitpongphan, J. Parikh, F. Bai, P. Mudalige, and
V. Sadekar, “On the broadcast storm problem in ad hoc wireless
networks,” Broadband Communications, Networks and Systems, 2006.
BROADNETS 2006. 3rd International Conference on, pp. 1–11, Oct.
2006.
[14] O. Tonguz, N. Wisitpongphan, F. Bai, P. Mudalige, and V. Sadekar,
“Broadcasting in vanet,” in 2007 Mobile Networking for Vehicular
Environments, May 2007, pp. 7 –12.
[15] N. Wisitpongphan, O. Tonguz, J. Parikh, P. Mudalige, F. Bai, and
V. Sadekar, “Broadcast storm mitigation techniques in vehicular
ad hoc networks,” Wireless Communications, IEEE, vol. 14, no. 6,
pp. 84–94, December 2007.
[16] O. Tonguz, N. Wisitpongphan, and F. Bai, “Dv-cast: A distributed
vehicular broadcast protocol for vehicular ad hoc networks,”
Wireless Communications, IEEE, vol. 17, no. 2, pp. 47 –57, april
2010.
[17] W. Viriyasitavat, F. Bai, and O. Tonguz, “Uv-cast: An urban
vehicular broadcast protocol,” in Vehicular Networking Conference
(VNC), 2010 IEEE, dec. 2010, pp. 25 –32.
[18] P. Kyasanur, R. Choudhury, and I. Gupta, “Smart gossip: An
adaptive gossip-based broadcasting service for sensor networks,”
IEEE International Conference on Mobile Adhoc and Sensor Systems
Conference, vol. 0, pp. 91–100, 2006.
[19] B. Bako, F. Kargl, E. Schoch, and M. Weber, “Advanced adap-
tive gossiping using 2-hop neighborhood information,” in Global
Telecommunications Conference, 2008. IEEE GLOBECOM 2008. IEEE,
nov. 2008, pp. 1 –6.
[20] T. Osafune, L. Lin, and M. Lenardi, “Multi-Hop Vehicular Broad-
cast (MHVB),” in ITS Telecommunications Proceedings, 2006 6th
International Conference on, jun. 2006, pp. 757 –760.
[21] M. Mariyasagayam, T. Osafune, and M. Lenardi, “Enhanced
multi-hop vehicular broadcast (mhvb) for active safety appli-
cations,” in Telecommunications, 2007. ITST ’07. 7th International
Conference on ITS, jun. 2007, pp. 1 –6.
[22] G. Korkmaz, E. Ekici, and F. Ozguner, “Black-burst-based multi-
hop broadcast protocols for vehicular networks,” Vehicular Tech-
nology, IEEE Transactions on, vol. 56, no. 5, pp. 3159 –3167, sep.
2007.
[23] A. Qayyum, L. Viennot, and A. Laouiti, “Multipoint relaying for
flooding broadcast messages in mobile wireless networks,” Hawaii
International Conference on System Sciences, vol. 9, 2002.
[24] C. Wu, K. Kumekawa, and T. Kato, “A novel multi-hop broadcast
protocol for vehicular safety applications,” Journal of Information
Processing, vol. 18, pp. 110–124, 2010.
[25] M. J. Slavik, I. Mahgoub, and M. Rathod, “Statistical broadcast
protocol design with wibdat: Wireless broadcast design and
analysis tool,” in Proceedings of IEEE Wireless Communications and
Networking Conference (WCNC 2010), Cancun, Mexico, March 2011.
[26] M. J. Slavik and I. Mahgoub, “Designing statistical multi-hop
wireless broadcast protocols using confidence levels from stochas-
tic models of reachability,” in The International Wireless Communi-
cations and Mobile Computing Conference (IWCMC-2011), Istanbul,
Turkey, 7 2011.
[27] ——, “Adapting statistical broadcast to linearly oriented networks
for VANETs,” in Proceedings of IEEE International Conference on
Wireless and Mobile Computing, Networking and Communications
(WiMob’2010), Niagara Falls, Canada, Canada, October 2010.
[28] I. S. Institute, “The network simulator - ns-2,” July 2006. [Online].
Available: http://www.isi.edu/nsnam/ns/
[29] R. Barr, “Swans - scalable wireless ad hoc network
simulator user guide,” March 2004. [Online]. Available:
http://jist.ece.cornell.edu/
[30] R. Thomas, An Introduction to Quadrat Analysis. Geo Abstracts
Ltd., 1977.
[31] I. Davis, J.S. and J. Linnartz, “Vehicle to vehicle rf propagation
measurements,” in Signals, Systems and Computers, 1994. 1994
14
Conference Record of the Twenty-Eighth Asilomar Conference on, vol. 1,
oct. 1994, pp. 470 –474 vol.1.
[32] M. Boban, T. Vinhoza, M. Ferreira, J. Barros, and O. Tonguz,
“Impact of vehicles as obstacles in vehicular ad hoc networks,”
Selected Areas in Communications, IEEE Journal on, vol. 29, no. 1,
pp. 15 –28, january 2011.
[33] A. Domazetovic, L. Greenstein, N. Mandayam, and I. Seskar,
“Propagation models for short-range wireless channels with pre-
dictable path geometries,” Communications, IEEE Transactions on,
vol. 53, no. 7, pp. 1123 – 1126, jul. 2005.
[34] T. Clausen and P. Jacquet, “Optimized link state
routing protocol (olsr),” October 2003. [Online]. Available:
http://www.ietf.org/rfc/rfc3626.txt
[35] M. Fiore, J. Harri, F. Filali, and C. Bonnet, “Vehicular mobility
simulation for vanets,” in Simulation Symposium, 2007. ANSS ’07.
40th Annual, mar. 2007, pp. 301 –309.
[36] O. K. Tonguz, W. Viriyasitavat, and F. Bai, “Modeling
urban traffic: a cellular automata approach,” Comm. Mag.,
vol. 47, pp. 142–150, May 2009. [Online]. Available:
http://dx.doi.org/10.1109/MCOM.2009.4939290
[37] Z. Jin, N. Yan, and L. Bing, “Reliable on-demand
geographic routing protocol resolving network disconnection
for vanet,” in Proceedings of the 5th International
Conference on Wireless communications, networking and
mobile computing, ser. WiCOM’09. Piscataway, NJ, USA:
IEEE Press, 2009, pp. 3963–3966. [Online]. Available:
http://portal.acm.org/citation.cfm?id=1737966.1738429
[38] G. Wolny, “Modified dmac clustering algorithm for vanets,”
in Proceedings of the 2008 Third International Conference on
Systems and Networks Communications. Washington, DC, USA:
IEEE Computer Society, 2008, pp. 268–273. [Online]. Available:
http://portal.acm.org/citation.cfm?id=1473252.1474527
[39] IEEE, “Dedicated short range communica-
tions working group.” [Online]. Available:
http://www.leearmstrong.com/DSRC/DSRCHomeset.htm
Michael Slavik is a Computer Engineering
Ph.D. Candidate in the Department of Computer
and Electrical Engineering and Computer
Science, at Florida Atlantic University. He
was previously a software engineer for
Motorola, Inc in the iDEN and WiMAX handset
development group. His research interests
include vehicular networking, multi-hop wireless
broadcast, wireless sensor networks, distributed
computing, and machine learning. His email
address is <mslavik@fau.edu>.
Imad Mahgoub is a full professor of Computer
Science and Engineering and director of
Mobile Computing Laboratory at Florida Atlantic
University. His research interests include mobile
computing, vehicular ad hoc networks, wireless
sensor networks, and parallel and distributed
systems. He has over 100 publications including
four books. He received research funds from
federal government agencies and the industry.
He is in the editorial board of the International
Journal of Communication Systems. He served
on the editorial board of the International Journal of Computers
and Applications and the Encyclopedia of Wireless and Mobile
Communications. He is a member of the steering committee of the
International Symposium on Performance Evaluation of Computer and
Telecommunication Systems (SPECTS). He served as vice chair, track
chair, posters chair, publicity chair, and program committee member for
several international conferences and symposia. He is a senior member
of the IEEE. He is also a member of the IEEE Computer Society, and
the ACM. His email address is <mahgoubi@fau.edu>.