Packet Forwarding in VANETs, the Complete Set of Results

Marc Torrent-Moreno

, Felix Schmidt-Eisenlohr

,
Holger Fler

, Hannes Hartenstein

Institute of Telematics, University of Karlsruhe, Germany

{torrent, fschmidt}@tm.uni-karlsruhe.de, [email protected]

Computer Science IV, University of Mannheim, Germany

[email protected]
Technical Report 2006-2
ISSN 1432-7864
Dept. of Computer Science
Universitt Karlsruhe (TH)
Abstract
The discrepancy between real-world radio channel behavior and its standard modeling in simulations
(Unit Disk Graph) is a major reason for protocols to perform different often worse than predicted
when deployed in a real-world setup. As researchers having to deal with real ad hoc networks are aware
of, assuming a xed border for a nodes communication range might not only lead to inaccurate results
but also to a wrong judgment on the comparison between different protocols. We have set up a simu-
lation study to investigate the effects of realistic channel characteristics on packet forwarding strategies
for vehicular ad hoc networks. The contributions of this technical report are threefold: i) We provide
a performance evaluation of various routing/forwarding strategies under the realistic non-deterministic
Nakagami radio propagation model and compare the results with the ones obtained using the standard
Two-Ray-Ground model. Validated German highway movement patterns are used to model node mobil-
ity. ii) We demonstrate that realistic channel conditions present an opportunity and not only a drawback
for some forwarding strategies. More specically, we show that for contention-based forwarding (CBF)
techniques, realistic channel characteristics provide a positive impact in terms of an increased average
hop distance. iii) We provide an analytical derivation of the expected hop distance for CBF that provides
a basis to optimally adjust CBF parameters.
1 Introduction
Vehicular ad hoc networks (VANETs) are a specic type of mobile wireless ad hoc networks that are cur-
rently attracting the attention of researchers around the globe. The joint efforts of governments, standard-
ization bodies, car manufacturers and academia (among others) in several national/international initiatives
(DSRC [1] in USA, C2CCC [2] in Europe, InternetITS [3] in Japan or Network on Wheels [4] in Germany,
to name a few) aim to make possible that, in a near future, vehicles can benet from spontaneous wireless
communications.
Clearly, safety-related applications will be the most important class of applications in VANETs requir-
ing, primarily broadcast messages. In addition, certain applications for trafc information or access to sta-
tionary nodes are envisioned demanding geo-cast communications, i.e., addressing/requesting information
to/from a specic region that is few hops away.
VANETs differ from other wireless networks in several aspects. Some of them stress standard routing
challenges, such as high node mobility due to the high potential speeds of their nodes. Others, though, can
favor the message forwarding strategies, such as pre-dened topologies (limited to the shape of the roads),
an unlimited energy source in every node and the increasing availability of positioning systems.
Additionally, vehicular scenarios can present particularly adverse environmental phenomena, not nec-
essarily common in all ad hoc scenarios. Basically, a VANET is composed by multiple static and mobile
reecting nodes and obstacles that can disturb the amplitude of a receiving signal due to multi-path. How-
ever, most of previous studies on the mobile world analyze and adjust communication protocols under
1
simplistic assumptions, i.e., Unit Disk Graph [5]. As any researcher that has implemented his/her design
in a real testbed knows, to assume a xed received signal strength for a xed distance does not reect real-
ity. While this modeling is very useful for understanding and explaining protocols, it is a major reason for
unexpected protocol behavior when moving from theory to reality.
In our aim to determine an appropriate routing/forwarding strategy for VANET highway scenarios we
use both realistic highway movements and a realistic meaning probabilistic radio propagation model. In
addition, we have improved the 802.11 medium access control and changed the 802.11 physical model of
ns-2 to reect 802.11p, which is a certain candidate for use in VANET systems. As candidate geo-addressing
schemes we have selected two promising position-based strategies, greedy forwarding with beacons (PBF
for Position-Based Forwarding), and CBF (Contention-Based Forwarding). When analyzing the perfor-
mance of these protocols though, the comparison with a non-position (routing) based approach appears to
be interesting. For completeness, we selected Ad hoc On-Demand Distance Vector Routing (AODV) as one
of the routing protocols for mobile networks with experimental RFC-status.
The selected protocols are analyzed using both deterministic (Two-Ray-Ground) and probabilistic (Na-
kagami) radio propagation models in order to evaluate the impact of more realistic radio channel conditions.
With the results obtained in our scenarios we rst present the good performance in terms of packet delivery
demonstrated by position-based approaches in all simulated scenarios. In more detail, we can observe that
although a realistic propagation model increases the number of collisions in the medium it can also benet
in terms of average hop distance when using a contention-based strategy. Finally, we indicate how to adjust
CBF, by computing the analytical estimation of its hop distance over probabilistic propagation models, to
reduce the number of collisions while keeping an acceptable round trip time.
The remainder of this technical report is organized as follows. In Sec. 2, we introduce the routing
strategies and different models utilized in this study pointing to the most relevant research related to our
work. In Sec. 3, the results of the simulation study comparing the different routing approaches and settings
are explained. The CBFs analytical hop distance computation and potential improvement are developed in
Sec. 4. Finally, in Sec. 5 we provide our concluding remarks.
2 Background and Related Work
Here, we briey explain necessary background information with respect to routing/forwarding protocols,
radio propagation models and the realistic highway movement traces utilized. At the end of the section we
point to additional related work.
2.1 Routing Protocols
Position Based Forwarding (PBF). (Greedy) Position-Based Forwarding is a long-known [6] method for
nding a route through a network utilizing node positions. In this protocol called Position Based For-
warding (PBF) here nodes pro-actively send beacon messages containing their node ID and their current
location. On the reception of such a message, the receiving node stores triples of the ID, the location and a
time stamp in a so-called neighbor table. The removal of neighbor table entries is done by a time-out.
When a node wants to send a packet, it rst queries the location of the destination node using a so-called
location service unless the application provides the location by itself. A location service is a distributed
algorithm that resolves the location of other nodes in the same network. The probably most simple one is
RLS [7], which oods a location request, i.e., the location request is propagated over the whole network.
When the node whose location is searched receives the request, it answers with a reply already using the
corresponding routing/forwarding mechanism(PBF in this case, since the request packet brings the position
of the requesting node along).
After the location resolution is done, PBF selects a neighbor from the neighbor table that offers most
progress towards destination (the greedy-constraint). In our case, a conservative approach to select the
next hop has been taken in order to increase the probability that the selected node is able to receive the mes-
sage, i.e., the furthest nodes inside the neighbor table are not selected to alleviate the effect of interferences,
mobility and fading. This means that all nodes in the neighbor table that are possibly out of communication
range under the assumption that they move with a dened highest speed v
max
, set to 234 km/h here, are not
taken into account as next forwarders. Additionally, we implemented the local-link callback feature: when
a node has not received any acknowledgment after the maximum MAC retries it will select another node
from the neighbor table (if any) and try it again. While this method does not need the notion of a route,
there are network constellations where an existing route to a destination cannot be found. In Highway Sce-
2
narios, however, these constellations do not occur due to reasons of geometry [8], meaning that whenever a
forwarding path exists and the neighbor tables are sufciently up-to-date, the path will be found.
Contention-Based Forwarding (CBF). Like PBF, Contention-Based Forwarding [8, 9] does not main-
tain routes and also assumes that the destination nodes position is provided by the application or by a
location service. Contrary to the previous scheme though, CBF does not make use of beacons. The sender
of a packet will broadcast the message to all its neighbors and these neighbors will nd out among them-
selves the one that will forward the packet. The forwarder is selected by the use of a contention period
where all nodes will select a waiting time depending on their distance to the nal destination (see Sec. 4).
Therefore, the node that offers a maximum progress will select the smallest waiting time, forwarding the
message at the end of this period, letting the other nodes know that they should not forward the packet. Note
that the main difference with respect to PBF is that CBF does not make use of a unicast ow to forward the
packet, i.e., the next forwarder selection is done after the actual transmission of the packet in every hop.
Ad hoc On-Demand Distance Vector (AODV). AODV [10] is a well-established ad hoc routing pro-
tocol and fundamentally works as follows: As a reactive protocol it will only be activated when a route is
needed. In that case, the originator of the message will ood a route request message. All nodes receiving
this request will record the number of hops to the originator and the last hop of this packet as a distance
vector to this destination. Then the packet is rebroadcasted after incrementing the hop count. This pro-
cess is repeated until the nal destination is reached, then a route reply will be sent backwards using the
just-established list of distance vectors. This creates a bi-directional route from originator to destination.
In order not to ood the whole network, AODV broadcasts its route request packets with a low limited
number of hops, time to live (TTL). If no reply is received a new request with a higher TTL is sent again.
Due to the linear geometry of our scenarios, we congured AODV to use the local repair mechanism, i.e.,
when one node in the chain can not reach its next hop, it tries to nd a new route to destination itself de-
pending on the distance to the message originator: the originator of the packet is informed if the route break
is nearer to sender than to destination; otherwise the node where the route broke tries to nd a new route to
destination without informing the originator.
Note that several improvements have been proposed for AODVin unidirectional links scenarios, e.g., [11].
However, due to our main goal of studying geo-addressing schemes the default version of the simulator
ns-2.28 [12] was utilized.
2.2 Propagation Models
A propagation model tries to approximate the received signal strength (RSS) of a radio transmission, typi-
cally using parameters like transmission power, distance between sender and receiver and antenna congu-
rations. This RSS is then used to determine success or failure of a packets reception.
Two-Ray-Ground (TRG) is a well known deterministic radio propagation model, it always determines
the same RSS for a xed distance between sender and receiver. In consequence, and in absence of inter-
ferences, a well dened communication range is experienced, that will be called intended communication
range in the following. This model is often used in the evaluation and/or design process of wireless com-
munication protocols, its advantage lying in its interpretability.
Several studies, e.g., [13], indicate that the probabilistic two-parameter Nakagami distribution [14]
matches the amplitude of a radio signal in a (mobile environment) fading channel at a given distance be-
tween sender and receiver. In our studies we make use of the Nakagami propagation model based on the
real highways measurements and analysis performed in [15]. See our previous work [16] and Sec. 4.1 for
a more detailed description.
Note in Fig. 1 how the assumptions of i) all nodes inside the intended communication range (in the
absence of interferences) will receive a specic message and ii) all nodes outside will not receive it are only
valid for the Two-Ray-Ground model, but not for Nakagami. For example, with a 500m communication
range, Nakagamis probability of reception with respect to the distance is a decreasing curve with a long
tail that is as low as 0.4 at the edge of the intended communication range. Since we cannot claim any
communication range for the Nakagami model, the term intended communication range always refers to
the range that is reached with the Two-Ray-Ground model.
2.3 Highway Traces
The nodes geographic position at a given time is the major determinant of both channel condition and
network topology. Thus when modeling a multi-hop highway scenario the mobility model is of high
signicance. The ones used for this technical report are real-world validated microscopic movement pat-
terns created by DaimlerChrysler for the FleetNet [18] project. The result of this effort was a set of highway
3
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
0 200 400 600 800 1000 1200 1400
R
e
c
e
p
t
i
o
n

p
r
o
b
a
b
i
l
i
t
y
Distance [m]
Two-Ray-Ground Nakagami
Figure 1: Probability of reception of a message with respect to the distance in absence of other nodes
interferences for Two-Ray-Ground and Nakagami models with an intended communication range of 500m.
Figure 2: Utilized highway scenario with 3 lanes per direction and 11 cars per km per lane (snapshot of the
HWGui [17] software of the FleetNet project).
4
scenarios with different number of lanes per direction and trafc densities (see [19] for a complete descrip-
tion). Fig. 2 shows a snapshot of such a typical scenario with 3 lanes per direction. The chosen set for
our evaluation consists in a 12km long highway scenario where vehicles travel at speeds from 50km/h
to 220km/h. We have chosen a medium/high trafc density scenario (2 lanes per direction with 12 cars
per km traveling in one direction and 22 cars per km in the opposite one) to study in detail the different
behavior of the analyzed protocols. Additionally, we have also used a low trafc density scenario (2 lanes
per direction with 4 cars per km in one direction and 12 in the other one) and one with high trafc density (3
lanes per direction and 33 cars per km in both directions). Using the three scenarios we cover a wide range
of trafc densities what ensures that our results are valid under a variety of trafc conditions.
2.4 Additional Related Work
This work is a continuation of [20], now using a probabilistic channel model and also evaluating CBF for
usage on highways. [21] discusses improvements of PBF on similar movement scenarios that we have used
for this work.
Also, our technical report is in line with the trend followed by some other studies in the eld of ad
hoc networking (e.g., opportunistic routing [22]) where the characteristics of a wireless channel are not
neglected, but instead, are taken advantage of.
3 Performance Comparison
After pointing out the main differences between the analyzed protocols we will compare their performance
under two different settings, i.e., deterministic and probabilistic radio channel models.
3.1 Simulation Set Up
The utilized simulation tool is the network simulator ns-2.28 [12]. However, its MAC/PHY implementation
has been adapted to IEEE 802.11p [23], a variant of 802.11a still not standardized, which is the technology
the above-mentioned projects have agreed upon. We refer the reader again to a previous work [16] for a
more detailed description. Moreover, bugs in the MAC and PHY modules were xed [24], and the PBF and
CBF forwarding modules implemented.
Our intention is to analyze how different distances between sender and receiver, and a different radio
propagation model (deterministic and probabilistic) affect the performance of the routing algorithms in both
directions of a communication. For this purpose, we simulated the highway scenarios described in Sec. 2.3,
where among all possible nodes we selected two specic vehicles (one communication pair) to exchange 10
Ping packets (request/reply). We performed several simulations where we increased the distance between
the two nodes forming a communication pair, up to 4500m. A larger distance results in an increased number
of hops since the intended communication range of all nodes is constant during the whole simulation. We
selected a 500m intended communication range as reasonable 1 hop maximum distance in ideal conditions
and absence of interferences (IEEE species a range up to 1000m [23] for this technology).
In each simulation only one communication pair was selected, while all other vehicles on the road
would only be potential intermediate nodes. The communication partners were picked such that they were
in theoretical multi-hop range, meaning that when applying a unit disk graph model, the resulting graph
contained routes between them during the whole communication time. In addition, they remain within the
same distance range (500m wide) during the whole packet exchange. For example, if the studied distance
was 3500m we can be sure that during the simulation time the two nodes are between 3000m and 3500m
apart and there are always enough vehicles in between to connect them via multi-hop.
In order to have statistical signicance, we selected 10 different scenarios (with the same number of
lanes and density) from the whole set of trafc patterns. In each scenario we select 10 different commu-
nication pairs (originator/destination) and run independent simulations with each one of them. Finally, for
each conguration setting, we compute the average and the condence interval (with 95% condence level)
of the studied metrics, see Sec. 3.2. The main conguration parameters are reported in Table 1. While we
have simulated many other settings, we will stick to these to describe the effects found.
3.2 Results
To compare the performance of PBF, CBF and AODV under both types of radio channel models we have
plotted different gures representing their behavior when increasing the communication distance with re-
spect to the selected metrics.
5
Table 1: Conguration parameters
Studied protocols PBF, CBF, AODV
Radio propagation models Two-Ray-Ground, Nakagami
Distance between comm. pair 500m to 4500m
Intended comm. range 500m
Ping packets generation rate 4 packets/s
Packet size 64 bytes
Number of Ping packets 10
PBF beaconing interval 2s
CBF max. contention time (T) 20ms
Vehicle density Medium: 12 cars/km up, 22 cars/km down, 2 lanes/dir.
- 3 bi-directional scenarios Low: 4 cars/km up, 12 cars/km down, 2 lanes/dir.
- up and down directions High: 33 cars/km up, 33 cars/km down, 3 lanes/dir.
0
0.2
0.4
0.6
0.8
1
0 500 1000 1500 2000 2500 3000 3500 4000 4500
P
i
n
g

p
a
c
k
e
t

d
e
l
i
v
e
r
y

r
a
t
i
o
Communication distance [m]
PBF Two-Ray-Ground
CBF Two-Ray-Ground
AODV Two-Ray-Ground
PBF Nakagami
CBF Nakagami
AODV Nakagami
Figure 3: Packet Delivery Ratio of PBF, CBF and AODV when increasing the distance to destination for
Two-Ray-Ground and Nakagami. Note that the curves of PBF and CBF for both propagation models and
all communication distances all cover each other.
We can observe in Fig. 3 the performance of the different protocols under Two-Ray-Ground and Naka-
gami. It reports the packet delivery ratio for different distances between sender and receiver. Note that since
the intended communication range of all nodes is xed to 500m selecting a destination node 500m further
from the sender is equivalent to add, at least, one hop to the resulting communication path. As expected,
AODV achieves the lowest packet delivery ratio, further decreased under non-deterministic radio propaga-
tion. In more detail, we observe that communication fails mainly due to two reasons: i) mobility, i.e., some
chosen nodes drove far from their previous/next hop signicantly decreasing the probability to forward a
packet successfully, and ii) the random behavior of Nakagami made a too optimistic route choice [25], i.e.,
some intermediate nodes were quite far from each other so the data ow had low probability to reach its
destination. When not only mobility but also received signal strength uctuations are considered the search
and use of a xed route turns to be the worst choice.
Position based routing protocols are robust against both, node mobility and fading. Both schemes show
average bidirectional delivery rates higher than 99.7% for all simulated distances and propagation models
due to the linear geometry of the studied scenarios.
To better understand the effect of using different propagation models we can take a look at the total
load sent into the channel resulting from the different routing algorithms, Fig. 4. First, we observe a good
performance of all strategies when dealing with Two-Ray-Ground, only AODV results in a signicant in-
crease of load with distance; note a constant higher channel load for PBF due to the utilization of beacons.
Second, note the high increment of the experienced channel load of PBF and AODV under Nakagami when
6
0
200
400
600
800
1000
1200
1400
1600
0 500 1000 1500 2000 2500 3000 3500 4000 4500
R
e
s
u
l
t
i
n
g

l
o
a
d

[
k
B
y
t
e
]
Communication distance [m]
PBF Two-Ray-Ground
CBF Two-Ray-Ground
AODV Two-Ray-Ground
PBF Nakagami
CBF Nakagami
AODV Nakagami
Figure 4: Resulting load in the mediumof PBF, CBF and AODV when increasing the distance to destination
for Two-Ray-Ground and Nakagami.
increasing the number of hops between the sender and the receiver. The difference between PBF and CBF
responsible of their different performance is the strategy to select the next forwarding node. A node using
CBF broadcasts a message and just expects that one node, which is closer to destination than itself, receives
the packet and forwards it. PBF, on the other hand, selects a specic node from the neighbor table and
tries to communicate with it. The use of a non-deterministic propagation model notably increases the risk
that a successful data exchange between an intermediate node and its next hop needs more than one MAC
retry, or more than one neighbor in a worse case. That explains the high increase of transmitted load w.r.t.
the number of hops of PBF with Nakagami. Similarly, AODV increases its resulting load. When AODV
routes hold, they tend to need many retries since the neighbors are chosen poorly. On the other hand, when
all possible routes break, no more packets are transmitted from the source due to route requests time-outs
(distances further than 2500m at Fig. 4). This effect limits the number of packets that are sent at a total
resulting load of 500kByte.
Finally we plotted the round trip time experienced by the different protocols in Fig. 5. We can see
how the results are in line with the former gures. The worst performance, i.e., the longest round trip
time, corresponds to AODV, specially under the probabilistic propagation model. The reason for this is a
combined inuence of both mobility and the propagation model. Discovered routes in the request phase
may include hops over high distances when the Nakagami model is used. These hops though, may have
low probability of successful data transmission and lead to a high number of broken routes and packet
losses. Additionally, one single route failure can lead to several packet losses, if these packets already wait
in the interface queue and the local repair mechanism does not succeed. If we take a look at the zoom (the
square inside Fig. 5), we can see how the performance of PBF under Nakagami is affected when increasing
the distance between the communication pairs. Also in the zoom of Fig. 5, we can observe an interesting
phenomenon, CBF shows shorter round trip times when considering a non-deterministic propagation model
and PBF shows longer ones.
To explain the resulting CBFs round trip time with Nakagami, we also plotted the average number
of hops for both protocols for the different communication distances, Fig. 6. Again, we see the benet
of not pre-selecting the next forwarding node in the process of routing a packet when considering a non-
deterministic radio model. As mentioned before, PBF selects a node inside its intended communication
range and tries to communicate with it. It is reasonable to think that the unreliability of the link results
in a longer round trip time, i.e., it will use several MAC layer retries (or even select a new node) before
being acknowledged. On the other hand, a node using CBF does not pre-select a node inside its intended
communication range as a next forwarder. That way, CBF benets when a node outside this range receives
the packet, what is a possible situation only when considering a non-deterministic propagation model. That
explains that, e.g., the average number of hops could be smaller than 8 when the destination is further than
4000mand having all nodes an intended communication range of 500m.
7
0
1
2
3
4
5
6
7
8
9
0 500 1000 1500 2000 2500 3000 3500 4000 4500
P
i
n
g

r
o
u
n
d
t
r
i
p

t
i
m
e

[
s
]
Communication distance [m]
PBF Two-Ray-Ground
CBF Two-Ray-Ground
AODV Two-Ray-Ground
PBF Nakagami
CBF Nakagami
AODV Nakagami
0
0.05
0.1
0.15
0.2
3000 3500 4000 4500
Figure 5: Round trip time of PBF, CBF and AODV when increasing the distance to destination for Two-
Ray-Ground and Nakagami.
As a conclusion, we can state that CBF presented the best performance among the routing protocols.
Although it presents a higher number of duplicates, CBFs strategy of not selecting a specic forwarder
before the actual message transmission is a robust scheme to ght against fading channels.
3.2.1 Low car density
In this subsection we repeat the same experiments presented above when assuming a highway with lower
trafc density. In this case, we chose a 4 and 12 cars per per km in each direction of the highway, i.e., a
scenario with approximately half the node density as the previous one.
Although most of the results are inline with the ones obtained with a medium/high density, we will also
observe few differences with respect to packet delivery ratio and resulting load.
In Fig. 7 we can observe that the packet delivery ratio presents some values below 100% (97% at min-
imum) for the position based forwarding approaches when using the Nakagami radio propagation model.
Remember that the communication partners are chosen so that there is always a possible path between them
in ideal conditions, i.e., no interferences and no fading phenomena. However, having a lower number of
potential forwarders combined with an unreliable channel due to fading can result in a slight decrease on
the packet delivery ratio. Note though, that the position based strategies still present the best choice in terms
of successful communication exchange.
Fig. 8 represents a signicantly lower resulting load than the previous scenario in all cases. For CBF,
the lower load corresponds mainly to the ooding location mechanism with a lower node density (see
Sec. 3.3). Also in case of AODV, the ooding mechanism used to nd the destination is the main reason
for a lower load on the medium. For PBF, not only the RLS affects the resulting load on the channel, but
mainly the constant exchange of beacons; note how the resulting load is approximately half as before, in
accordance with the total number of nodes. Note that although a lower node density has different impact on
the different protocols, the relation among them stays mainly unchanged.
Finally, we can observe in Fig. 9 how a lower density of nodes has a minimal impact on the round trip
time. Here, as in the case of the packet delivery ratio, having enough nodes in the path, ensures that position
based algorithms can provide a good performance in terms of bidirectional data exchange.
3.2.2 High car density
For completeness, we present in this section the results obtained with a high vehicular density scenario,
a highway with 3 lanes per direction and 33 cars per km. It can be observed how the results depicted in
Figs. 10, 11 and 12, are totally inline with the ones obtained with lower nodes densities. The only effect
8
0
2
4
6
8
10
12
0 500 1000 1500 2000 2500 3000 3500 4000 4500
N
u
m
b
e
r

o
f

h
o
p
s

t
o

d
e
s
t
i
n
a
t
i
o
n
Communication distance [m]
PBF Two-Ray-Ground
CBF Two-Ray-Ground
AODV Two-Ray-Ground
PBF Nakagami
CBF Nakagami
AODV Nakagami
Figure 6: Number of hops to reach destination of PBF, CBF and AODV when increasing the distance to
destination for Two-Ray-Ground and Nakagami.
0
0.2
0.4
0.6
0.8
1
0 500 1000 1500 2000 2500 3000 3500 4000 4500
P
i
n
g

p
a
c
k
e
t

d
e
l
i
v
e
r
y

r
a
t
i
o
Communication distance [m]
PBF Two-Ray-Ground
CBF Two-Ray-Ground
AODV Two-Ray-Ground
PBF Nakagami
CBF Nakagami
AODV Nakagami
Figure 7: Packet Delivery Ratio of PBF, CBF and AODV when increasing the distance to destination for
Two-Ray-Ground and Nakagami under low trafc conditions.
9
0
100
200
300
400
500
600
700
0 500 1000 1500 2000 2500 3000 3500 4000 4500
R
e
s
u
l
t
i
n
g

l
o
a
d

[
k
B
y
t
e
]
Communication distance [m]
PBF Two-Ray-Ground
CBF Two-Ray-Ground
AODV Two-Ray-Ground
PBF Nakagami
CBF Nakagami
AODV Nakagami
Figure 8: Resulting load in the mediumof PBF, CBF and AODV when increasing the distance to destination
for Two-Ray-Ground and Nakagami under low trafc conditions.
-1
0
1
2
3
4
5
6
7
8
9
10
0 500 1000 1500 2000 2500 3000 3500 4000 4500
P
i
n
g

r
o
u
n
d
t
r
i
p

t
i
m
e

[
s
]
Communication distance [m]
PBF Two-Ray-Ground
CBF Two-Ray-Ground
AODV Two-Ray-Ground
PBF Nakagami
CBF Nakagami
AODV Nakagami
0
0.05
0.1
0.15
0.2
3000 3500 4000 4500
Figure 9: Round trip time of PBF, CBF and AODV when increasing the distance to destination for Two-
Ray-Ground and Nakagami under low trafc conditions.
10
0
0.2
0.4
0.6
0.8
1
0 500 1000 1500 2000 2500 3000 3500 4000 4500
P
i
n
g

p
a
c
k
e
t

d
e
l
i
v
e
r
y

r
a
t
i
o
Communication distance [m]
PBF Two-Ray-Ground
CBF Two-Ray-Ground
AODV Two-Ray-Ground
PBF Nakagami
CBF Nakagami
AODV Nakagami
Figure 10: Packet Delivery Ratio of PBF, CBF and AODV when increasing the distance to destination for
Two-Ray-Ground and Nakagami under high trafc conditions.
worth commenting from our point of view is the decrease on the packet delivery ratio shown by PBF for
high communication distances and Nakagami due to the high resulting load experienced.
3.3 Impact of the Reactive Location Service
Motivated by the existence of geo-addressed applications in VANETs we also studied the performance of
the position-based schemes assuming that the application relaying on the routing layer already knows the
position of the targeted node, or area, at the moment it generates the rst Ping packet. The results obtained
in the medium/high trafc density present, as expected, a lower channel load, a shorter round trip time and
no change with respect to the packet delivery ratio when RLS is not required.
Figs. 13 and 14 show a constant load increase for all distances and both propagation models when using
RLS; except for a 1 hop communication in case of PBF since the source node knows the destinations
position due the use of beacons. This constant difference results from the ooding mechanism used by
the RLS, i.e., broadcasting location requests over the whole network. Notice also a lower increase of the
resulting load in case of PBF when using RLS if we compare it with the increase experienced by CBF. PBF
makes use of the position information exchanged by the ooding mechanism to update its neighbor table,
hence, saving some beacons to be transmitted to the medium.
Fig. 15 and 16 show the inuence of RLS on the experienced round trip time with the Two-Ray-Ground
and the Nakagami model. Only the rst Ping packet of each communication exchange is taken into account
since the RLS is used only before the transmission of this rst packet. As expected, we observe an increasing
amount of additional RTT for higher communication distances when using RLS in every case. It is also
noticeable again that the tendency of longer RTTs for PBF and shorter ones for CBF if the Nakagami model
is used instead of Two-Ray-Ground.
4 CBF Analysis and Adjustment
With the Nakagami channel model, CBF shows the best performance w.r.t. resulting load, round trip time
and number of hops to destination. In the following, we develop the analytical estimation of CBFs hop
distance depending on the vehicular trafc density to evaluate the positive impact that it can have on CBFs
performance. First though, we provide some necessary details on CBFs next hop selection procedure in
order to understand the later proposal and improvements.
11
0
500
1000
1500
2000
2500
3000
3500
4000
4500
0 500 1000 1500 2000 2500 3000 3500 4000 4500
R
e
s
u
l
t
i
n
g

l
o
a
d

[
k
B
y
t
e
]
Communication distance [m]
PBF Two-Ray-Ground
CBF Two-Ray-Ground
AODV Two-Ray-Ground
PBF Nakagami
CBF Nakagami
AODV Nakagami
Figure 11: Resulting load in the medium of PBF, CBF and AODV when increasing the distance to destina-
tion for Two-Ray-Ground and Nakagami under high trafc conditions.
0
2
4
6
8
10
12
0 500 1000 1500 2000 2500 3000 3500 4000 4500
P
i
n
g

r
o
u
n
d
t
r
i
p

t
i
m
e

[
s
]
Communication distance [m]
PBF Two-Ray-Ground
CBF Two-Ray-Ground
AODV Two-Ray-Ground
PBF Nakagami
CBF Nakagami
AODV Nakagami
0
0.05
0.1
0.15
0.2
3000 3500 4000 4500
Figure 12: Round trip time of PBF, CBF and AODV when increasing the distance to destination for Two-
Ray-Ground and Nakagami under high trafc conditions.
12
0
50
100
150
200
250
300
350
0 500 1000 1500 2000 2500 3000 3500 4000 4500
R
e
s
u
l
t
i
n
g

l
o
a
d

[
k
B
y
t
e
]
Communication distance [m]
PBF Two-Ray-Ground with RLS
PBF Two-Ray-Ground without RLS
CBF Two-Ray-Ground with RLS
CBF Two-Ray-Ground without RLS
Figure 13: Impact of RLS. Resulting load of PBF and CBF when increasing the distance to destination
under Two-Ray-Ground.
0
200
400
600
800
1000
1200
1400
1600
0 500 1000 1500 2000 2500 3000 3500 4000 4500
R
e
s
u
l
t
i
n
g

l
o
a
d

[
k
B
y
t
e
]
Communication distance [m]
PBF Nakagami with RLS
PBF Nakagami without RLS
CBF Nakagami with RLS
CBF Nakagami without RLS
Figure 14: Impact of RLS. Resulting load of PBF and CBF when increasing the distance to destination
under Nakagami.
13
0
0.02
0.04
0.06
0.08
0.1
0.12
0 500 1000 1500 2000 2500 3000 3500 4000 4500
P
i
n
g

r
o
u
n
d
t
r
i
p

t
i
m
e

[
s
]
Communication distance [m]
CBF Two-Ray-Ground with RLS
PBF Two-Ray-Ground with RLS
CBF Two-Ray-Ground without RLS
PBF Two-Ray-Ground without RLS
Figure 15: Impact of RLS. Round trip time of PBF and CBF when increasing the distance to destination
under Two-Ray-Ground.
0
0.02
0.04
0.06
0.08
0.1
0.12
0 500 1000 1500 2000 2500 3000 3500 4000 4500
P
i
n
g

r
o
u
n
d
t
r
i
p

t
i
m
e

[
s
]
Communication distance [m]
PBF Nakagami with RLS
CBF Nakagami with RLS
PBF Nakagami without RLS
CBF Nakagami without RLS
Figure 16: Impact of RLS. Round trip time of PBF and CBF when increasing the distance to destination
under Nakagami.
14
Figure 17: Calculation of CBFs waiting time: F denotes the last forwarder, D the nal destination node
and N a node receiving the packet. r
radio
denotes the maximum distance towards destination a message is
supposed to travel.
CBF uses contention to implicitly select the next hop in the communication path. Each potential for-
warder, i.e., each node receiving the message to forward, computes the time t it must wait before transmit-
ting the packet depending on its suitability, i.e., its progress towards destination:
P(F, D, N) = max
_
0,
dist (F, D) dist (N, D)
r
radio
_
(1)
t(P) =
_
max {0, T (1 P)} , P > 0
, else
(2)
where P is the progress function depending on the positions of the last forwarder F, the nal destination
node D of the packet and of the receiving node N. The Euclidean distance between two positions is
expressed as dist, r
radio
denotes the maximumdistance towards destination a message is supposed to travel
and T denes the maximum contention time. Fig. 17 illustrates the situation and the parameters used for
the calculation. In Two-Ray-Ground, the setting of r
radio
is as straightforward as using Fs communication
range, since nodes further than this distance can receive a message with probability 0 (see Sec 2.2). With
Nakagami, however, the selection is not trivial since there exists no such border where the probability of
reception of a message drops to 0. Note that a short r
radio
can result in multiple collisions since all nodes
located further than r
radio
that receive the message will forward the message at the same time (t(P) = 0
for all of them). On the other hand, a long r
radio
value results in a longer average forwarding delay.
In a linear network topology, the ideal r
radio
setting in each case is the distance from F to the location
of the furthest receiving node. In the following, we derive an estimation of this value or, in other words, the
estimation of the hop distance in a CBF multi-hop communication.
4.1 Analytical estimation of hop distance
The average distance a message can travel in every hop depends on the propagation model, the nodes
density/distribution and the wireless interfaces conguration.
In order to estimate the hop distance we must rst consider the probability that a message is received at
a specic distance. The Nakagami probability density function (pdf ) [26, p. 102] describes the distribution
of the power x of a received signal and is given by:
f
pow
(x; m, ) =
m
m
(m)
m
x
m1
exp
_

x
_
(3)
where is the Gamma function, m denotes the Nakagami m-parameter and the average received power.
m and depend on the distance d between sender and receiver. The pdf f
pow
describes a Gamma distri-
bution and as a special case (if m N) an Erlang distribution. Therefore, in case that the Nakagami-m
parameter is an integer, the Nakagami cumulative density function (cdf) can be expressed as:
F
pow
(x; m, ) = 1 exp
_

mx

_
m

i=1
_
m

_
i1
(i 1)!
(4)
15
In our scenarios, assuming that packets travel further than 150m, (4) simplies to (using the same Nakagami
conguration as in [16], where the m-parameter is set to 1.0 for d > 150m):
F
pow
(x; 1, ) = 1 exp
_

_
(5)
The average received power depends on d, the distance between sender and receiver. Assuming the simple
Free Space model for , this dependency can be expressed as:
(d) =
P
t
G
t
G
r

2
(4d)
2
L
(6)
where P
t
is the transmission power, G
t
and G
r
the antenna gains, the wavelength and L the path-loss
factor (see [27]). Then, the cdf F
1,pow
(x; d) := F
pow
(x; 1, (d)) can be expressed as a function of d:
F
1,pow
(x; d) = 1 exp
_

x
C
d
2
_
with C =
P
t
G
t
G
r

2
(4)
2
L
(7)
In ns-2, a packet is received successfully if the received signal power is greater than the Receiving
Threshold R
xTh
. Thus the probability for successful reception at a certain distance d can be expressed as:
P
R
(d) =1 F
1,pow
(R
xTh
; d)
=exp
_

R
xTh
C
d
2
_
(8)
At this point we have an estimation of the probability of successfully receiving a message at a specic
distance from the sender (in the absence of other nodes interferences). Now, for simplicity reasons and
justied by our focus on linear scenarios, we model our road as a line where there is a node every [meters]
at the moment a message is sent. Thus node n
i
is positioned at x
i
= i [m], i N. In these conditions, the
probability that the furthest node receiving a packet sent by n
0
(with position x
0
= 0[m]) is node n
i
can be
expressed as:
P
F
(i; ) = P
R
(i )

j=i+1
(1 P
R
(j ))
= exp
_

R
xTh
C
(i )
2
_

j=i+1
_
1 exp
_

R
xTh
C
(j )
2
__
(9)
Then, the expected value of P
F
(), EP
F
(), denes the expected average hop distance of a multi-hop
communication:
EP
F
() =

i=0
(i P
F
(i; ))
=

i=0
_
_
i exp
_

R
xTh
C
(i )
2
_

j=i+1
_
1 exp
_

R
xTh
C
(j )
2
__
_
_
(10)
In order to corroborate our analytical estimation and our simulation tool we simulated the idealized
road described above, i.e., one car every [m]. We can see in Fig. 18 the perfect match of the results
obtained for the average hop distance by analysis and simulation for different distances between the cars.
The conguration values are summarized in Table 2.
Both results show that lower distances between cars can achieve higher average hop distances. The
higher average distance that a message can travel in one hop is the result of a higher number of nodes that
can probably receive, and therefore forward, the packet.
Furthermore, given the specic PHY settings (see Table 2) and together with the nodes distribution, it is
possible to interpolate the resulting curve. In our case we use a polynomial p() of degree 3 with a relative
error lower than 2% for distances between cars greater than 10 meters:
p() = 0.0005
3
+ 0.1191
2
11.1337 + 968.3044 (11)
16
500
600
700
800
900
1000
1100
1200
0 10 20 30 40 50 60 70 80 90 100
H
o
p

d
i
s
t
a
n
c
e

[
m
]
Distance between nodes [m]
Analytical estimation of average hop distance
Average hop distance (simulation, 10 000 samples)
Figure 18: CBFs expected and simulated value of the average hop distance EP
F
() when increasing the
distance [meters] between cars positioned in a line.
Table 2: Physical layer parameters
Transmission power (P
t
) 7.6543 10
4
W
Reception threshold (R
xTh
) 3.1632 10
13
W
Antenna gains (G
t
, G
r
) 4dB
Carrier frequency (f) 5.9GHz
Wavelength () 50.8mm
Path-loss factor (L) 1
4.2 CBFs Adjustment
In the simulated scenario of Sec. 3.2, CBF achieved (with high distances to destination, 4500m) an average
hop distance of 681m. However, if we would use that car density, i.e., 34 cars/km, on the simple model
described above, i.e., one car every 29.4m, we would estimate an average hop distance of 725m. The
difference between the estimated and the achieved hop distance is due to the non-uniform distribution of
cars in a realistic highway scenario and the suboptimal r
radio
setting, i.e., 500m.
As explained above, a shorter r
radio
can result in a higher resulting load in the medium due to collisions
among the forwarders. In Fig. 19 we can observe how a higher choice of r
radio
results in a lower resulting
load. Note also the remarkable benet from using r
radio
= 725m, i.e., the expected value using the model
from Sec. 4.1, and the little one of extending to r
radio
= 1000m in terms of resulting load.
On the other hand, if we take a look at Fig. 20 we become aware of the trade-off between decreasing
the channel load and increasing the round trip time. In other words, having a higher r
radio
achieves lower
resulting load (less collisions) but results in a higher round trip time due to longer contention periods of
all potential forwarders. If we take a look at the achieved average hop distance, Fig. 21, we observe that a
higher r
radio
leads, as expected, to higher average hop distances. Note that the distance traveled for the last
hop is not averaged since it is limited by the position of the destination node. As in Fig. 19, we can observe
that i) increasing r
radio
until the expected value offers a signicant improvement and ii) setting r
radio
to
even higher distances results in little benet.
Taking all observations together, we can state that the expected hop distance, i.e., 725m, presents a good
choice of r
radio
. It slightly increases the round trip time (by only 40ms for a bidirectional communication
over 4500m) but it presents a remarkable improvement with respect to coordination among the intermediate
nodes. Fromour point of view, any improvement in the channel load can result highly benecial in the future
when the technology is developed and there exist a high number of equipped vehicles on the roads.
For completeness, we present in the following the results obtained when adjusting r
radio
in the low and
high nodes density scenarios.
17
40
60
80
100
120
140
160
180
200
220
0 500 1000 1500 2000 2500 3000 3500 4000 4500
R
e
s
u
l
t
i
n
g

l
o
a
d

[
k
B
y
t
e
]
Communication distance [m]
Intended communication range 500m
Intended communication range 725m
Intended communication range 1000m
Figure 19: CBFs resulting load in the medium for different values of r
radio
when increasing the distance
to destination under Nakagami.
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0 500 1000 1500 2000 2500 3000 3500 4000 4500
P
i
n
g

r
o
u
n
d
t
r
i
p

t
i
m
e

[
s
]
Communication distance [m]
Intended communication range 1000m
Intended communication range 725m
Intended communication range 500m
Figure 20: CBFs round trip time for different values of r
radio
when increasing the distance to destination
under Nakagami.
18
560
580
600
620
640
660
680
700
720
740
1000 1500 2000 2500 3000 3500 4000 4500
A
v
e
r
a
g
e

d
i
s
t
a
n
c
e

p
e
r

h
o
p

(
w
/
o

l
a
s
t
)
Communication distance [m]
Intended communication range 1000m
Intended communication range 725m
Intended communication range 500m
Figure 21: CBFs average hop distance for different values of r
radio
when increasing the distance to desti-
nation under Nakagami.
4.2.1 Low car density
The same observations described before can also be noticed when the scenario with low trafc density is
used. In average, there are 24 cars per km, that results in an average car distance of 41.7m. The expected
average hop distance for this car density is 675m. As we can see in Fig. 22, the impact of adjusting the
r
radio
parameter to its expected value lowers the resulting load by up to 14%, while a higher value for r
radio
does not improve the situation anymore. Fig. 23 presents the price to pay in terms of RTT when reducing
the load on the medium.
4.2.2 High car density
In case of high trafc density the effect of the r
radio
adjustment can be seen in Figs. 24 and 25. In our case,
r
radio
has been set to 820m, the theoretical average hop distance.
Again, we can observe the high decrease in the resulting load with r
radio
= 820m and the little further
decrease for r
radio
= 1000m, Fig. 24. On the other hand, the resulting RTT for r
radio
= 1000m is
remarkably higher than for r
radio
= 820m, Fig. 25.
Summarizing, we can state that adjusting the r
radio
parameter in the way described is a good balance
between resulting load and observed Round Trip Time.
5 Conclusions
Some future VANETs applications will most probably require geo-addressing strategies. Motivated by
this fact, this paper studies the impact on packet forwarding strategies of multi-path fading, an important
characteristic in vehicular environments often neglected in wireless studies. To accomplish our purpose,
we have used a simulation setup not only using realistic vehicle movements, but also a probabilistic radio
propagation model (Nakagami) adjusted to VANET radio characteristics.
In such a context, we rst performed a simulative comparison of two position-based schemes, one with
beacons (PBF) and another with a contention-based forwarding strategy (CBF). Additionally, due to com-
pleteness we included a topology-based routing protocol (AODV). The three protocols have been analyzed
under the assumption of both deterministic (Two-Ray-Ground) and probabilistic (Nakagami) propagation
models. From the results obtained we can draw the following conclusions: i) the radio propagation model
utilized has a great impact on protocol performance, ii) probabilistic channel models can not only have a
negative impact but also enhance protocol performance in certain aspects iii) not selecting a specic node
as next hop before the transmission of a message is a robust strategy against both unreliable links and
dynamic topologies and iv) although the explicit pre-selection of a next-hop may lead to almost perfect
19
20
30
40
50
60
70
80
90
100
110
120
0 500 1000 1500 2000 2500 3000 3500 4000 4500
R
e
s
u
l
t
i
n
g

l
o
a
d

[
k
B
y
t
e
]
Communication distance [m]
Intended communication range 500m
Intended communication range 675m
Intended communication range 1000m
Figure 22: CBFs resulting load for different values of r
radio
when increasing the distance to destination
under low nodes density under Nakagami.
0
0.05
0.1
0.15
0.2
0.25
0 500 1000 1500 2000 2500 3000 3500 4000 4500
P
i
n
g

r
o
u
n
d
t
r
i
p

t
i
m
e

[
s
]
Communication distance [m]
Intended communication range 1000m
Intended communication range 675m
Intended communication range 500m
Figure 23: CBFs round trip time for different values of r
radio
when increasing the distance to destination
under low nodes density under Nakagami.
20
100
150
200
250
300
350
400
0 500 1000 1500 2000 2500 3000 3500 4000 4500
R
e
s
u
l
t
i
n
g

l
o
a
d

[
k
B
y
t
e
]
Communication distance [m]
Intended communication range 500m
Intended communication range 820m
Intended communication range 1000m
Figure 24: CBFs resulting load for different values of r
radio
when increasing the distance to destination
under high nodes density under Nakagami.
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
0 500 1000 1500 2000 2500 3000 3500 4000 4500
P
i
n
g

r
o
u
n
d
t
r
i
p

t
i
m
e

[
s
]
Communication distance [m]
Intended communication range 1000m
Intended communication range 820m
Intended communication range 500m
Figure 25: CBFs round trip time for different values of r
radio
when increasing the distance to destination
under high nodes density under Nakagami.
21
end-to-end packet delivery, the consequential MAC retries induce additional delay and load costs. For these
reasons we believe that CBF is a serious candidate to be used in future vehicle-to-vehicle communications.
Second, after understanding the behavior of CBF under a probabilistic radio model we conducted an
analytical study of its expected hop distance. Moreover, we corroborated this result with simulations and
showed how it can be used in order to improve CBFs performance. The result of this adjustment is a
reduction of the number of collisions, i.e., a better synchronization among neighbors, with the only trade-
off of a slight increase in delay.
Our current work comprises a broader study of the CBF strategy, including other scenarios, such as
sensor network type, and applications, such as robust information dissemination inside a geographical area.
Acknowledgment
Marc Torrent-Moreno and Holger Fler acknowledge the support of the German Ministry of Education and
Research (BMB+F) for the Network on Wheels project under contract number 01AK064F. The authors
thank Jon Letamendia-Murua for help on the improvement of the ns-2 wireless modules, Thomas King for
the routing modules, and Roland Krger for his work on the movement traces.
References
[1] DSRCDedicated Short Range Communications Project,
http://www.leearmstrong.com/DSRC/DSRCHomeset.htm.
[2] Car-to-Car Communication Consortium, http://www.car-to-car.org.
[3] Internet ITS Consortium, http://www.internetits.org.
[4] The Network on Wheels Project, http://www.network-on-wheels.de.
[5] A. Farag, Scalable Analysis and Design of Ad Hoc Networks Via Random Graph Theory, in
DIALM 02: Proceedings of the 6th international workshop on Discrete algorithms and methods for
mobile computing and communications. New York, NY, USA: ACM Press, September 2002, pp.
4350.
[6] G. G. Finn, Routing and addressing problems in large metropolitan-scale internetworks, ISI Re-
search Report, University of Southern California, Tech. Rep. ISI/RR-87-180, 1987.
[7] M. Ksemann, H. Fler, H. Hartenstein, and M. Mauve, A Reactive Location Service for Mobile Ad
Hoc Networks, Department of Computer Science, University of Mannheim, Tech. Rep. TR-02-014,
November 2002.
[8] H. Fler, H. Hartenstein, J. Widmer, M. Mauve, and W. Effelsberg, Contention-Based Forward-
ing for Street Scenarios, in WIT 04: Proceedings of the 1st International Workshop on Intelligent
Transportation, March 2004.
[9] H. Fler, J. Widmer, M. Ksemann, M. Mauve, and H. Hartenstein, Contention-Based Forwarding
for Mobile Ad-Hoc Networks, Elseviers Ad Hoc Networks, vol. 1, no. 4, pp. 351369, 2003.
[10] C. E. Perkins and E. M. Royer, Ad-Hoc On-Demand Distance Vector Routing, in WMCSA 99:
Proceedings of the 2nd IEEE Workshop on Mobile Computing Systems and Applications, February
1999, pp. 90100.
[11] M. K. Marina and S. R. Das, Routing performance in the presence of unidirectional links in multihop
wireless networks, in MobiHoc 02: Proceedings of the 3rd ACM international symposium on Mobile
ad hoc networking & computing. New York, NY, USA: ACM Press, June 2002, pp. 1223.
[12] The ns-2 network simulator, http://www.isi.edu/nsnam/ns/.
[13] W. Zhang and N. Moayeri, Classication of Statistical Channel Models for Local Multipoint Dis-
tribution Service Using Antenna Height and Directivity, IEEE 802.16 working group contribution
IEEE802.16.1pc-00/07, January 2000.
22
[14] M. Nakagami, The m-distribution, a General Formula of Intensity Distribution of the Rapid Fading,
in Statistical Methods in Radio Wave Propagation: Proceedings of a Symposiumheld at the University
of California. W.G. Hoffman, Ed. Oxford, England: Pergamon, 1960, pp. 336.
[15] V. Taliwal, D. Jiang, H. Mangold, C. Chen, and R. Sengupta, Empirical determination of channel
characteristics for dsrc vehicle-to-vehicle communication, in VANET 04: Proceedings of the 1st
ACM international workshop on Vehicular ad hoc networks. New York, NY, USA: ACM Press,
October 2004, pp. 8888.
[16] M. Torrent-Moreno, D. Jiang, and H. Hartenstein, Broadcast Reception Rates and Effects of Priority
Access in 802.11-Based Vehicular Ad-Hoc Networks, in VANET 04: Proceedings of the 1st ACM
international workshop on Vehicular ad hoc networks. New York, NY, USA: ACM Press, October
2004, pp. 1018.
[17] The HWGui Software, http://www.informatik.uni-mannheim.de/pi4/projects/HWGui/.
[18] The FleetNet Project, http://www.et2.tu-harburg.de/eetnet/.
[19] H. Fler, M. Torrent-Moreno, R. Krger, M. Transier, H. Hartenstein, and W. Effelsberg, Studying
Vehicle Movements on Highways and their Impact on Ad-Hoc Connectivity, Department of Mathe-
matics and Computer Science, University of Mannheim, Tech. Rep. TR-2005-003, June 2005.
[20] H. Fler, M. Mauve, H. Hartenstein, M. Ksemann, and D. Vollmer, MobiCom Poster: Location-
Based Routing for Vehicular Ad-hoc Networks, Mobile Computing and Communications Review
(MC2R), vol. 7, no. 1, pp. 4749, January 2003.
[21] T. King, H. Fler, M. Transier, and W. Effelsberg, On the Application of Dead-Reckoning to
Position-Based Routing for Vehicular Highway Scenarios, in CoNEXT 05: Proceedings of the 1st
Conference on Future Networking Technologies, Toulouse, France, October 2005, pp. 258259.
[22] S. Biswas and R. Morris, ExOR: Opportunistic Routing in Multi-Hop Wireless Networks, in SIG-
COMM 05: Proceedings of the 2005 conference on Applications, technologies, architectures, and
protocols for computer communications. New York, NY, USA: ACM Press, August 2005, pp. 133
144.
[23] IEEE 802.11p Task Group, http://grouper.ieee.org/groups/802/11/.
[24] F. Schmidt-Eisenlohr, J. Letamendia-Murua, M. Torrent-Moreno, and H. Hartenstein, Bug Fixes on
the IEEE 802.11 DCF module of the Network Simulator ns-2.28, University of Karlsruhe, Tech. Rep.
TR-2006-1, January 2006.
[25] H. Lundgren, E. Nordstrm, and C. Tschudin, Coping with communication gray zones in IEEE
802.11b based ad hoc networks, in WOWMOM 02: Proceedings of the 5th ACM international work-
shop on Wireless mobile multimedia. New York, NY, USA: ACM Press, September 2002, pp. 4955.
[26] M. K. Simon and M.-S. Alouini, Digital communication over fading channels, 2nd ed. Wiley-
Interscience, 2005.
[27] T. S. Rappaport, Wireless communications, 2nd ed. Prentice Hall, 2002.
23