Beruflich Dokumente
Kultur Dokumente
ca on 14 April 2010
These include:
Abstract
Index Terms
I. I
Cooperative driving can improve safety and efficiency by enabling vehicles to exchange
emergency messages to each other in the neighborhood. In vehicular ad hoc network (VANET),
vehicles transmit traffic and safety related information including traffic congestion avoidance
message, accident warning, and accident report, etc., which assists drivers to make proper
communications, inter-vehicle communications is more flexible for deployment with low cost [2],
and research on IVC systems has attracted great attention from academia, industry, and govern-
ments recently. The U.S. Federal Communications Commission (FCC) has approved 75 MHz
(5.850 − 5.925 GHz) bandwidth for Dedicated Short-Range Communications (DSRC) systems
have launched several projects to study cooperative driving in an IVC, such as Advanced
Driver Assistance Systems (ADASE2) [4], Crash Avoidance Metrics Partnership (CAMP) [5],
CarTALK2000 [6], FleetNet [7], Partners for Advanced Transit and Highways (PATH) [8], etc.
is extremely challenging especially due to the high mobility and hostile wireless environment.
First, as no acknowledgment (ACK) mechanism is applied for broadcast messages in the medium
access control (MAC) layer, message loss due to packet collisions or poor channel conditions
cannot be easily detected. Since life critical emergency messages have to be delivered to other
vehicles as fast and reliable as possible [9], the traditional broadcasting scheme without an ACK
mechanism is not suitable for emergency message delivery in an IVC system. Second, due to
the limited transmission range, message relaying from intermediate nodes1 is required to reach
remote vehicles. However, without an effective broadcast control mechanism, multiple copies of
the broadcast messages may be delivered among nodes, which could result in broadcast storm
problem [10] and degrade the network resource utilization. Some research works propose to
reduce message redundancy and prevent broadcast storm by selecting a subset of neighboring
nodes to forward the broadcast message. However, it is a non-trivial task to determine a proper
subset of nodes that can guarantee the message reliability and achieve efficient resource utilization
simultaneously.
To address the aforementioned issues, several broadcasting protocols have been proposed in the
1
the terms “node” and “vehicle” are used interchangeably throughout the paper.
literature. Some protocols use network layer broadcast control algorithms to reduce the message
redundancy [12]–[16], but they cannot guarantee the MAC layer reliability. Other protocols aim at
improving the the transmission reliability by repeatedly broadcasting messages [17] or selecting
the farthest node to relay messages [18]–[20]. However, repeated broadcast cannot completely
guarantee the transmission reliability but degrade the resource utilization. The farthest node may
suffer from high packet error rate (PER) and is not an ideal relay candidate, especially in high
speed vehicle networks. In this paper, we propose a cross layer broadcast protocol (CLBP) for
emergency message dissemination in a multihop IVC system, aiming to improve the transmission
reliability and minimize the message redundancy. Considering the particular characteristics of
VANETs, we design a novel relaying metric which is composed of geographical locations, phys-
ical layer channel conditions, and moving velocities of vehicles. Based on the metric, we apply
distributedly. In specific, after receiving the broadcast RTS (BRTS), each relay candidate starts
its backoff timer to reply a broadcast CTS (BCTS) based on the calculated relaying metric in a
distributed manner. After a successful BRTS/BCTS handshake, one node is successfully selected
as the next hop relay to forward the broadcast message in the desired propagation direction.
Furthermore, to support different services with various quality of service (QoS) requirements in
the IVC system, we adopt the priority based enhanced distributed coordination access (EDCA)
of IEEE 802.11e MAC to support safety services. The emergency messages are served with the
highest priority and thus the minimum channel access delay can be achieved.
The main contributions of this paper are three-fold. First, we design a novel metric for selecting
a proper relaying node to forward the emergency message. Second, based on the derived metric,
we propose a cross layer approach to efficiently broadcast emergency messages in the desired
propagation direction. MAC layer service differentiation is applied for emergency messages.
Third, we analytically study the network performance in terms of the packet error rate (PER) of
the emergency message, relay selection delay, and emergency message access delay. Analytical
and simulation results with NS-2 demonstrate that the proposed cross layer approach can quickly
and reliably deliver emergency messages while minimizing the broadcast message redundancy.
The remainder of this paper is organized as follows. We briefly review the related work
in Sec. II. The proposed CLBP is described in Sec. III. An analytical model is developed to
study the performance of CLBP in Sec. IV. The simulation results are given to demonstrate the
Broadcast protocols in mobile ad hoc networks (MANET) can be classified into four cate-
gories [11]: i) simple flooding [12], [13], in which a node rebroadcasts a new message until it
reaches all connected nodes in the network; ii) probability based methods [14], in which protocols
can be further divided into two sub-classes: (a) a node rebroadcasts a message according to a
predefined probability, and this scheme becomes simple flooding one if the probability is set
to 1; and (b) a node decides whether to rebroadcast a message based on the number of the
received copies during a certain interval; iii) area based method [14], in which a node that
can cover more additional area is selected to forward the received message; and iv) neighbor
knowledge method [15], [16], in which a node makes a forwarding decision according to the
knowledge of its one-hop or two-hop neighbors. All these aforementioned broadcast protocols
aim to reduce the number of redundant messages at the network layer, without considering
the MAC layer hidden terminal problem, collisions, and link reliability, etc. It is well known
that broadcast transmission is not reliable due to the lack of ACK scheme in the MAC layer.
However, some emergency messages are life critical, the delivery of which should be guaranteed.
Therefore, previous works on network layer broadcast protocol design can not be directly applied
to IVC. Recently, some protocols have been proposed for emergency message delivery in IVC.
In [25], a distributed MAC scheme for emergency message dissemination is presented. A node
reserves the data channel for emergency message broadcast by sending a pulse signal through
the control channel. A MAC designed for emergency message broadcasting is studied in [17],
where a node broadcasts emergency messages for several times to increase the transmission
reliability. However, repeatedly broadcasting messages cannot guarantee the successful reception
of broadcast messages but may increase the contention level in a distributed vehicle network and
waste the scarce wireless network resources. A black burst based ad hoc multi-hop broadcast
(AMB) protocol is proposed for emergency message dissemination in [18]. A neighboring node
sends channel jamming signal (black-burst) with the time duration proportional to its distance.
Thus, the farthest neighboring node sending the longest jamming signal wins the contention and
becomes the next hop relaying node. Nevertheless, the largest jamming duration used by the
relay candidate causes a long delay for emergency messages. In the position based multi-hop
broadcast protocol (PMBP) [19], the farthest neighboring node waits the shortest time duration
to reply the broadcast node. However, the farthest node usually suffers from a large path loss
and a high packet error rate (PER) which may cause MAC layer retransmissions and thus a
longer link delay. In [20], each node maintains a list of neighboring nodes and always selects
the farthest neighboring node as the next hop relay. However, network topology of IVC changes
dynamically due to the high mobility of vehicles, to effectively update the list of neighbors is
In order to reduce link delay and improve throughput, many channel condition based relaying
and routing metrics have been proposed in cooperative relaying schemes and routing protocols.
In [21], the expected transmission count (ETX) is proposed to measure the expected number of
transmissions that a node attempts until a packet is successfully delivered to the next relaying
node. The routing scheme based on ETX assures that the selected path achieves the minimum link
delay. Similar routing metrics such as expected transmission time (ETT) and weighted cumulative
ETT (WCETT) [22] also consider channel conditions and link reliability in the metric design.
In [23], the enhanced interior gateway routing protocol (EIGRP) adopts a composite metric,
which uses weight factors to decide the impacts of minimum link bandwidth, traffic load, link
delay, and reliability on path selection. The cooperative MAC (CoopMAC) is proposed in [24],
in which each node maintains a table of relays that can improve the link throughput, and selects
the relay with a better channel condition and higher data rate. All these schemes select paths or
relays based on channel conditions. However, they do not consider the specific characteristics
of VANET, i.e., high mobility. In this paper, we propose to jointly consider the geographical
locations, the channel conditions, and the relative velocities of vehicles to make a relay decision
in IVC.
Consider a highway with M lanes. Half of the lanes are for vehicles driving to one direction,
while the other half for vehicles driving to the opposite direction. A vehicle’s velocity is randomly
distributed among a discrete set V = {Vi | Vi−1 < Vi , i ∈ (1, P]}, and the velocity is directional
since vehicles may move to two different directions. Each vehicle is equipped with a half-
duplex transceiver and a Globe Positioning System (GPS) by which it can acquire its position
information, moving velocity, and moving direction. The transmission range of a vehicle Rt is
divided into a number of blocks, and the length of each block is φ which should be at least
the minimum safety distance for two adjacent moving vehicles. Therefore, there are Q = bRt /φc
blocks within Rt , and their distances to the broadcast vehicle are represented as {Bi | Bi = i·φ, i ∈
[1, Q]}. We use carrier sense multiple access with collision avoidance (CSMA/CA) based 802.11e
MAC for channel access and service differentiation among multiple nodes. To provide reliable
(BCTS) frames are exchanged before emergency messages. In addition, in the proposed CLBP,
one relaying node is selected to forward the emergency message in the desired propagation
direction, based on a novel relaying metric designed for the IVC system.
A. BRTS/BCTS handshake
The structure of a broadcast RTS (BRTS) frame is shown in Fig. 1. Compared to the traditional
RTS frame, five fields are added in BRTS: em in f o, t direction, t velocity, r x, r y. The field
em in f o takes the information of the source node which initially transmits the emergency
message, and it contains: i) the source node address init addr; ii) the position information of the
source node init x, and init y; iii) the sequence number of the emergency message em seq; and
iv) the weight factors α1 , α2 , α3 that are used for relaying metric calculation and relaying node
selection. t direction is the message propagation direction, t velocity is the moving velocity of
the current broadcast node, and r x and r y indicate the position of the broadcast node.
frame_control duration r_addr t_addr em_info t_direction t_velocity t_x t_y fcs
When a node has an emergency message for transmission, it first broadcasts BRTS based on
the CSMA/CA mechanism and starts a timer, tbrts r = tbrts + tdi f s + tbcts , where tbrts and tbcts are the
transmission times of BRTS and BCTS, respectively, and tdi f s is the time duration of a DIFS.
If there is no BCTS response within tbrts r , the node contends for channel access to rebroadcast
BRTS immediately until BCTS is successfully received. The broadcast node sets its duration
field in BRTS such that any node that hears the BRTS but is not eligible for replying BCTS
will set its NAV and defer its own transmissions accordingly.
After receiving BRTS, a node decides whether it is eligible for replying BCTS based on
direction information or the position information of the received BRTS. If init addr in the
received BRTS is the same as the address of the current broadcast node, it implies that this is
the first hop emergency message dissemination, and the node will decide whether to reply BCTS
based on propagation direction t direction. Otherwise, if its own position is between the original
source node and the current broadcast node, it will not reply BCTS since there is no distance
gain along the propagation direction. In this case, the node updates its NAV according to the
duration field in the received BRTS. Otherwise, it starts a backoff timer for replying BCTS and
keep sensing the channel in the mean time. As shown in Fig. 2, A is the source node that initiates
an emergency message, B is the current broadcast node. Node C will not reply BCTS since it
locates between A and B, while D is eligible for relaying the message and starts a backoff timer
upon receiving BRTS. This guarantees that the emergency message will be efficiently forwarded
Rt A block
B
C
Each eligible relaying node which locates at (x, y) and moves at velocity v will start a timer
for replying BCTS according to the following metrics: i) the distance from the current broadcast
node to itself; ii) the received SNR and PER which can be estimated from the received BRTS;
and iii) the velocity difference from the current broadcast node. Based on the three metrics, the
relay candidate evaluates the composite relaying metric F used for relay selection, which is
given by
4d e 4v
F = α1 · (1 − ) + α2 · + α3 · , (1)
BQ Emax 2VP
where
p
(x − r x)2 + (y − r y)2
4d = b c · φ,
φ
and
−v − −
4v = | →
−−−−−−−→
t velocity |,
4d is the transmission distance, e is the PER of the emergency message that is calculated based
on the measured SNR, 4v is the relative velocity, BQ is the distance of the farthest block in
the transmission range, Emax is maximum tolerable PER defined in [26], VP is the maximum
velocity, α1 , α2 , α3 (α1 > 0, α2 > 0, α3 > 0) are weight factors and usually configured by users.
For instance, if a user wants the messages to be delivered over a fewer number of hops or with
a reduced PER, he can set a larger α1 or α2 accordingly. Moreover, if the topology is relatively
The main objective of the proposed CLBP is to deliver the emergency message to other
vehicles as fast and reliable as possible. 4d is a metric to determine the number of hops, i.e.,
the message will be forwarded over a fewer number of hops with a larger 4d. In addition,
MAC layer delay of the message highly depends on the PER e. A higher PER may result in
retransmissions that lead to a longer link delay. Finally, a small relative speed 4v is usually
desirable in high speed vehicle networks to guarantee the channel between two moving nodes
is relatively stationary. The proof in [27] verifies that if two routing metrics are bounded, their
additive composite metric is also bounded. As 4d ∈ [B1 , BQ ], 4v ∈ [0, 2 · VP ], and e ∈ [0, 1],
the composite metric F is consequently bounded. The maximum and minimum values of F are
In order to avoid interruptions to BRTS/BCTS handshake from other flows, CLBP requires
the selected relaying node replies BCTS within DIFS interval. Applying the concept of mini-slot
in [28], [29], we further divide a DIFS interval into a number of mini-slots. The length of a
τ = 2 · ρ + t swith , (2)
where ρ is the maximum channel propagation delay within the transmission range Rt , and t swith
is the time duration that a transceiver switches between the receiving mode and transmitting
mode. In order to map the relaying metric F to a number of mini-slots, we further partition
the value between Fmin and Fmax into Wn segments, and each segment is 0 = (Fmax − Fmin )/Wn .
After evaluating the relaying metric F , an eligible relay candidate sets its timer to i mini-slots
if its F is within [Fmin + (i − 1) · 0 , Fmin + i · 0 ), where i ∈ [1, Wn ]. The relay candidate with the
minimum F value will reply BCTS first and thus be selected as a relaying node accordingly.
In other words, a node with a longer distance, better channel condition, and smaller velocity
After the transmission of BCTS which also takes fields init addr and em seq, if another
relay candidate overhears BCTS replying the same BRTS before its own timer expires, the node
will stop its own backoff timer and update its NAV according to the value of duration field
included in the received BCTS. Note that the duration fields in BRTS and BCTS are set to
L
be tbrts d = tdi f s + tbcts + t si f s + rb
+ t si f s + tack and tbcts d = tbrts d − tdi f s − tbcts , where t si f s is the
time duration of a SIFS, tack is the transmission time of an ACK frame, L is the payload size
of the emergency message, and rb is the basic rate. tbrts d is conservative because the receiver
waits at most one DIFS to reply BCTS in CLBP. Whenever a node receives or overhears other
It is possible that multiple relay candidates may choose the same mini-slot to reply BCTS,
which causes collisions. When a collision occurs, the relay candidates that have started their
backoff timers but have not replied BCTS will sense the channel busy, and they will stop their
own timers accordingly. If a relay candidate which has replied BCTS receives a rebroadcast
BRTS, it will enter the backoff stage again and divide 0 into Wn segments, each of which is
1 = 0 /Wn , i.e., the relay candidate will wait i (i ∈ [1, Wn ]) mini-slots to reply BCTS again if
The procedure continues until retransmissions due to BCTS collisions reach rmax times, which
implies some nodes have very close F values. Then from the rmax round, the relay candidates
that were collided in the last round will randomly select a mini-slot to reply BCTS until a relay is
successfully selected. In CLBP, the relaying metric consists of three variables and it is less likely
that two nodes have exactly the same F . Therefore, the proposed collision resolution scheme
is very efficient for selecting a unique relaying node. The psuedo code of the relay selection
After a successful BRTS/BCTS handshake, the current broadcast node that successfully re-
ceives BCTS will broadcast the emergency message after one SIFS interval. The selected relay
will acknowledge the reception of the emergency message if the transmission is successful.
To avoid message redundancy, each node in the system maintains a list to keep records of all
received emergency messages. Each entry in the list records the address of the source node and
the sequence number of the emergency message, and entries with out-of-date messages will be
deleted. A node which receives an emergency message will check the list and drop this message
if it has already been recorded. Otherwise, it will receive the message and update the list. After
successfully replying an ACK, the selected relay becomes the next relaying node and repeats
C. Priority
To provide safety related services with satisfactory delay guarantee in IVC system, we use
priority based IEEE802.11e enhanced distributed channel access (EDCA) for service differenti-
ation. We include the safety services and divide all services into five classes. Different classes
of services have different priorities to access the channel based on the access categories (AC)
as shown in Table I. The setting of arbitration inter-frame space (AIFS) and contention window
where σ is a time slot, PF is the persistence factor which is set to 1 for safety services and 2 for
other services. In other words, a node always selects a backoff counter from the minimum CW
for emergency message delivery while it doubles its CW after each collision for other services.
TABLE I
P
In this section, we develop an analytical model to analyze the performance of the proposed
CLBP. To make the proposed scheme tractable, we make the following assumptions: i) nodes
are randomly distributed, and the node density is λ per Rt ; ii) all nodes are saturated, i.e., the
nodes always have data packets in their buffers for transmissions, and data packets of the same
access category AC[i] have the same transmission probability pi and collision probability qi that
can be obtained by Eqs. (1), (2), (3), (4), and (7) in [30]; iii) all data packets of the same access
category AC[i] have the same payload size Di that is larger than rts threshold; iv) PERs of
BRTS, BCTS, and ACK are negligible due to small packet size; and v) the retransmission time
In our proposed protocol, a node starts a timer (in terms of mini-slots) and contends to send
BCTS based on the composite relaying metric F in Eq. (1). 4d and 4v can be evaluated from the
received BRTS, and PER e is dependent on the channel conditions. To the best of our knowledge,
there is no consensus on fading and shadowing models for VANET so far [31]. In our analytical
model, we adopt the Friis free-space model [32] to determine the received signal power. Over
an additive white Gaussian noise (AWGN) channel, the bit error rate (BER) of the emergency
q q R∞ 2
2εb 2Pr
message with BPSK modulation is Q N0
= Q rb N0
[33], where Q(x) = √12π x e−t /2 dt,
εb is the received energy per bit, N0 is noise power spectral density, Pr is the received power,
q L L
2Pr I
rb is the basic rate. Since e = 1 − 1 − Q rb N0
= 1 − 1 − Q 4d
[34], Eq. (1) can be
rewritten as
" I L #
4d α2 4v
F = α1 · (1 − )+ · 1− 1−Q + α3 · (8)
BQ Emax 4d 2VP
q
2Pt Gt Gr (c/ fc )2
where I = rb N0 (4π)2
, Pt is the transmitted power, Gt and Gr are the transmitter and receiver
antenna gains, respectively, c is the speed of light, and fc is the carrier frequency. F is a function
of Fmin , Fmax are given in Appendix A. Therefore, the selection of mini-slots is dependent on
Emergency message access delay T is defined as the time interval from an emergency
message arriving at the head of the queue until it is successfully acknowledged, which includes: i)
an AIFS; ii) T b consisting of the backoff time, the backoff frozen time due to other transmissions,
retransmission time due to BRTS collisions, and a successful BRTS transmission time; iii) T c
time, and iv) T d the sum of delay due to retransmissions caused by the emergency message
errors, a successful emergency message transmission and its acknowledgement. Thus, we have
Relay selection delay T rs is defined as the time duration from a broadcast node attempting
to transmit BRTS until a relay is successfully selected, and we have T rs = AIFS [4] + T b + T c .
Denote w as the average time that a backoff timer of a broadcast node reaches 0, and we have
X
∞
Tb = qm4 (1 − q4 ) [(w + tbrts ) + m(w + tbrts r )] , (10)
m=0
where qm4 (1 − q4 ) is the probability that the broadcast node successfully delivers BRTS at backoff
stage m, and (w + tbrts ) + m(w + tbrts r ) is the corresponding delay. Denote w| j ( j ∈ [0, CW[4]]) as
the value of w given that the j’th time slot is selected. Since the broadcast node selects a time
X
CW[4]
1
w= (w| j) · , (11)
j=0
CW[4] + 1
where
Pj
k=1 Yk ,
j ∈ [1, CW[4]]
w| j =
(12)
0, j = 0,
and Yk is the mean of Yk which denotes time delay in the k’th slot of CW[4]. Yk can be one
idle time slot, or the frozen time due to a successful data transmission or collisions. Let Eks ,
Ekc , and Eki denote the events that a node transmits a message successfully in slot k, a collision
Y
4
pr(Eki ) = (1 − pi )ni ·xi,k
i=0
X
4 ! Y
ni
pr(Eks ) = xi,k · · pi · (1 − pi )ni −1 · (1 − pi )ni
i=0
1 i∈[0,4],i,i
where
1, i f AIFS [i] 6 AIFS [4] + k,
xi,k =
(14)
0, otherwise,
ni is the number of contending neighboring nodes belonging to AC[i], and xi,k denotes whether
neighboring nodes of AC[i] will contend for channel access with the broadcast node in the k’th
slot of CW[4]. As shown in Fig. 3, CW[4] is divided into three sub-windows cw0 , cw1 , cw2 . If
the broadcast node selects time slot 0, it only contends with neighboring nodes of AC[3], AC[2],
and x3,0 = 1, x2,0 = 1, x1,0 = 0, x0,0 = 0. Similarly, if the node chooses a time slot k from cw1 , it
must contend with neighboring nodes of AC[3], AC[2], AC[1], and x3,k = 1, x2,k = 1, x1,k = 1,
x0,k = 0.
CW[4]
cw0 cw1 cw2
AIFS[4]
AIFS[3] …...
AIFS[2] …...
AIFS[1] …...
AIFS[0] …...
We denote S as the average frozen time the broadcast node experiences for one successful
packet transmission, and S i as one successful transmission time of AC[i]. For AC[i] (i ∈ [0, 3]),
S i = 3 · t si f s + trts + tcts + Di /rd + tack + AIFS [4], while for safety services, S 4 approximately
equals to 2 · t si f s + tdi f s + tbrts + tbcts + L/rb + tack + AIFS [4]. Since the successful transmission
Q P
probability of AC[i] is n1i · pi · (1 − pi )ni −1 · m∈[0,4],m,i (1 − pm )nm , we obtain S = 4i=0 n1i · pi ·
Q
(1 − pi )ni −1 · m∈[0,4],m,i (1 − pm )nm · S i . Let C represent the average frozen time the broadcast
node experiences owing to one packet collision, and C approximately equals to trts + AIFS [4].
Finally, we have
T c , which denotes the time interval from the successful reception of BRTS to the successful
reception of BCTS, is a variable and depends on how long the broadcast node can successfully
receive BCTS from its relay candidates. In CLBP, a relay candidate starts its backoff timer to
reply BCTS after receiving BRTS from the broadcast node. In order to capture the activities of
the backoff timer of a relay candidate, the backoff process is illustrated in a three dimension
diagram with the state space (m, n, l), as shown in Fig. 4, where m (m ∈ [0, rmax )) is the backoff
stage, n (n ∈ [1, Wn ]) is the initial value of the backoff timer, and l (l ∈ [0, Wn ]) is the number
The state transitions of a backoff timer are given in Appendix B-A, and therefore we have
m−1
X
rmax Y
T c = S 0 · t0 + Cm · S m · tm , (16)
m=1 j=0
and the average time taken for a relay candidate successfully replying a BCTS at backoff stage
m, respectively, the derivations of which are given in Appendix B-B. Finally, T d which denotes
X
∞
Td = em · (1 − e) · t si f s + L/rb + t si f s + tack + m · (T b + T c + t si f s + L/rb + t si f s + tack ) , (17)
m=0
0,0,0
0,1,1 0,1,0
...
...
...
0,Wn ,Wn 0,Wn ,Wn -1 0, Wn , Wn -2 0, Wn ,2 0,Wn ,1 0,Wn ,0
...
...
V. SIMULATION RESULTS
In this section, we evaluate the performance of the proposed CLBP, in terms of the packet error
rate (PER), relay selection delay, and the emergency message access delay, via NS-2 simulations.
Vehicles are randomly distributed over a two-lane highway, and one is selected as the broadcast
node. The velocity of a vehicle takes a value among {20, 25, 30, 35, 40, 45, 50} m/s. In the default
setting, five data flows are set up with the rate 100 packets/s. Other simulation parameters are
TABLE II
P
We first compare the PER performance of CLBP with that of AMB proposed in [18] under
various N0 . For a smaller N0 , both CLBP and AMB achieve a low PER. When N0 increases, the
PER of AMB increases while that of CLBP does not change much. In CLBP, the broadcast node
jointly considers the distance, channel condition, and the relative velocity to select the next hop
relaying node. Under an ideal channel condition, the farthest relay candidate has the lowest F ,
and is selected as the relaying node; while under a poor channel condition, the received SNR at
the farthest relay candidate decreases and accordingly the achieved PER increases, in which case
a closer relay candidate with a lower PER may be selected with CLBP. As show in Fig. 6, the
PER of CLBP decreases slightly when N0 increases from −174.6 dBw/Hz to −173.88 dBw/Hz.
Therefore, CLBP assures the PER performance of the emergency message and thus is more
0.06 CLB P( )
AMB( )
0.05
0.04
0.03
PER
0.02
0.01
0.00
-0.01
-176.0 -175.5 -175.0 -174.5 -174.0 -173.5
N0(dBw/Hz)
Relay selection delay is defined as the interval from the time the broadcast node attempts
to deliver BRTS to the time it successfully receives BCTS. In Fig. 7(a), we compare the relay
selection delays of CLBP and AMB. By applying service differentiation in CLBP, the emergency
messages are served with the highest priority. Therefore, AMB which adopts basic CSMA/CA
achieves longer access delay compared with CLBP. In addition, the node sending the longest
channel jamming signal becomes the relaying node in AMB, while a node waiting the shortest
time to reply a BCTS becomes the relaying node in CLBP. As shown in Fig. 7(a), the relay
selection delay of CLBP is much shorter than that of AMB. However, both relay selection delays
of CLBP and AMB increase with the increase of node density due to retransmissions caused by
collisions.
Finally, we show the emergency message access delay under various node densities and
background noise levels in Fig. 7(a)-7(d). It can be seen that the emergency message access
5
CLBP
AMB
1
5 10 15 20
(N0 = -175.81dBw/Hz)
delays of AMB are higher than those of CLBP, and their differences increase with the node
densities and background noise levels. This is because, first, CLBP gives the highest priority for
safety services by adjusting AIFSN, PF, CWmin, and CWmax, which results in a smaller access
delay, whereas in AMB, emergency messages have to contend with other services with the same
priority. Second, in CLBP, the selected relaying node waits the minimum number of mini-slots
to reply BCTS, while in AMB, the selected relaying node sends the longest black burst signal
to win the opportunity to reply clear-to-broadcast (CTB). Third, under a poor channel condition,
the broadcast node in CLBP chooses an appropriate node with a reasonable PER performance
to relay the emergency message. In AMB, the broadcast node always selects the farthest relay
VI. C
In this paper, we have developed a composite relaying metric to select an appropriate re-
laying node, considering the special characteristics of vehicle networks. Based on the relaying
metric, we have proposed a cross layer broadcast protocol to efficiently disseminate emergency
18 18
14 14
12 12
10 10
2 4 6 8 2 4 6 8
Number of contending flows Number of contending flows
(a) (b)
Emergency message access delay (ms)
18
AMB( N0 = -173.47) AMB( , N0 = -173.47)
18
16 16
14 14
12 12
10 10
2 4 6 8 2 4 6 8
Number of contending flows Number of contending flows
(c) (d)
Fig. 7. Comparisons of emergency message access delays between AMB and CLBP
messages in an IVC. Analytical and simulations results with NS-2 have shown that CLBP can
quickly disseminate emergency messages and achieve high resource utilization. In our future
work, we will further study reliable broadcasting with user cooperation in both urban and rural
environments, incorporating various mobility models and road traffic conditions. We will also
analyze the end to end QoS performance of the proposed broadcasting protocol.
A A
D Fmin , Fmax , 0
L
Consider continuous function z(x, y) = α1 · (1 − BxQ ) + Eαmax
2
· 1 − 1 − Q xI + α3 · 2Vy P , where
x ∈ [B1 , BQ ], y ∈ [0, 2VP ], and its partial differential coefficient z0 x (x) and z0 y (y) can be expressed
as h iL−1
I
α1 α2 · L · I · 1 − Q x
0
z x (x) = − + √ (A.1)
BQ 2π · Emax · x2 · eI 2 /2x2
α3
z0 y (y) = . (A.2)
2VP
function of variable x if z0 x (x) < 0, where x ∈ [B1 , BQ ]. Let X∗ = {xi | z0 x (xi ) = 0, z00 x (xi ) > 0, xi ∈
n oSn x
[B1 , BQ ]}, and X ∗ = {xi | z0 x (xi ) = 0, z00 x (xi ) < 0, xi ∈ [B1 , BQ ]}. Let Z∗ = z(b xφi c · φ, V1 ) | xi ∈ X∗ z(d φi e · φ, V1 )
n oS
and Z ∗ = z(b xφi c · φ, V1 ) | xi ∈ X ∗
n x o α3
z(d φi e · φ, V1 ) | xi ∈ X ∗ . Similarly, z is a monotonic increasing function of y since z0 y = 2V P
> 0.
Therefore, the minimum value Fmin and maximum value Fmax of the discrete function F can be
expressed as
z(BQ , V1 ), z0 x (x) 6 0, x ∈ [B1 , BQ ]
Fmin =
min(Z∗ ), Z∗ , ∅
z(B1 , V1 ), z0 x (x) > 0, x ∈ [B1 , BQ ],
z(B1 , VP ), z0 x (x) 6 0, x ∈ [B1 , BQ ]
Fmax =
max(Z ∗ ), Z∗ , ∅ (A.3)
z(BQ , VP ), z0 x (x) > 0, x ∈ [B1 , BQ ],
A B
D tc
Let random variables {F1 , F2 , · · · , FN } denote the relaying metrics of N relay candidates,
L
and Ft = α1 · (1 − 4d
BQ
t
) + α2
Emax
· 1 − 1 − Q I
4dt
4vt
+ α3 · 2V P
is the relaying metric of node t.
Notice that the distances between the broadcast vehicle and other vehicles are not independent
variables because two vehicles can not locate in the same position. As a result, as the functions
of distances, the routing metrics {F1 , F2 , · · · , FN } are not independently distributed either. In
a highway consisting of M lanes, at most M vehicles can choose the same block. Let events
4d M+1 = BQ − 1, · · · }. Denote vt,i (y) as the relative velocity between node t and the broadcasting
node, when Ft = y and event Ai occurs. For example, 4d1 = B1 in event A1 , and therefore
L
v1,1 (y) = 2V
α3
P
· y − α1 · (1 − B1
BQ
) − α2
Emax
· 1 − 1 − Q I
B1
. Let ψt,m represent the number of
where m = 0 /(Wn )m .
We denote Nm as the set of contending relay candidates at the backoff stage m, and initially
|N0 | = N. After receiving BRTS, relay candidate t starts its backoff timer and prepares to reply
BCTS. The transition probability Pr[(0, n, 0)|(0, 0, 0)] , which denotes that t starts a backoff timer
(XN )
M·Q
1
= M·Q · Pr [Fmin + (n − 1) · 0 6 Ft < Fmin + n · 0 | Ai ]
N i=1
(XN )
M·Q
1
= M·Q · Pr vt,i (Fmin + (n − 1) · 0 ) 6 4vt < vt,i (Fmin + n · 0 ) . (B.4)
N i=1
Because the velocities of the broadcast node and relay candidate t are directional and ran-
domly distributed, the relative velocity 4vt is also randomly distributed among the (2P − 1)!
relative velocities {4V1 , 4V2 , · · · , 4V(2P−1)! }. In addition, since 4vt does not depend on events
A1 , A2 , · · · , A(M·Q) , for specific values vt,i (Fmin + (n − 1) · 0 ) and vt,i (Fmin + n · 0 ), we can acquire
N
Pr[vt,i (Fmin + (n − 1) · 0 ) 6 4vt < vt,i (Fmin + n · 0 )], and therefore probability Pr[(0, n, 0)|(0, 0, 0)]
is obtained.
In the proposed scheme, when node t has started the backoff timer and successfully backoff
one more mini-slot, it means the initial values of all other nodes’ backoff timers are larger than
the mini-slots that node t has elapsed. Therefore, the transition probability Pr[(0, n, l+1)|(0, n, l)],
which represents l mini-slots has elapsed and t’s backoff timer can backoff one more mini-slot,
can be expressed as
1, l = 0,
Pr[(0, n, l + 1)|(0, n, l)] =
T (B.5)
Pr j∈N0 ψ j,0 > l + 1 , l ∈ [1, Wn − 1],
where
\
Pr ψ j,0 > l + 1
j∈N0
(X
N )
M·Q
\
= Pr F j > Fmin + l · 0 | Ai · Pr(Ai )
i=1 j∈N0
(X) \
M·Q
N
= Pr 4v j > v j,i (Fmin + l · 0 ) · Pr(Ai )
i=1 j∈N0
(X) Y
M·Q
N h i
1
= M·Q · Pr 4v j > v j,i (Fmin + l · 0 ) . (B.6)
N i=1 j∈N0
Node t stops its backoff timer and returns to the initial state (0, 0, 0) when it or any other
relay candidate successfully transmits BCTS. In the former case, the number of elapsed mini-
slots equals to the initial value of t’s backoff timer, and less than the initial value of any other
timer. Whereas, in the latter case, at least one other timer’s initial value equals to the mini-slots
elapsed, and the initial value of t’s timer is larger than the mini-slots elapsed. Therefore, the
where
M·Q
(X
\ \ N ) Y h i
1
Pr ψ j,0 > l + 1 ψt,0 = l = M·Q · Pr 4v j > v j,i (F min + l · 0
)
j∈N0 , j,t N i=1 j∈N0 , j,t
· Pr vt,i (Fmin + (l − 1) · 0 ) 6 4vt < vt,i (Fmin + l · 0 ) ,
(B.8)
and
\ \ \
Pr ψ j,0 > l ψt,0 > l + 1 − Pr ψ j,0 > l + 1
j∈N0 , j,t j∈N0
M·Q
(XN ) Y h i
1
= M·Q · Pr 4v j > v j,i (Fmin + (l − 1) · 0 ) · Pr 4vt > vt,i (Fmin + l · 0 )
N i=1 j∈N0 , j,t
(X) YM·Q
N h i
1
− M·Q · Pr 4v j > v j,i (Fmin + l · 0 ) , (B.9)
N i=1 j∈N0
If t’s backoff timer decreases to 0, it means the elapsed mini-slots equal to the initial value
of the timer. t can either successfully deliver BCTS and returns to the initial state (0, 0, 0) as
expressed by Eq. (B.8), or transmit BCTS simultaneously with other relay candidates. In the
latter case, given that t’s BCTS collides with those from other relay candidates, the probability
that t sets the its backoff timer to be n in the next backoff stage is given by
hT i
where Pr j∈Nm ψ j,m > l + 1 | |Nm | = xm can be obtained similarly as Eq. (B.6), and Pr(|Nm | =
xm | |Nm−1 | = xm−1 ) is given by Eq. (B.18). Conditioning on |Nm |, |Nm−1 |, · · · ,|N1 |, we can acquire
B. Calculation of T c
(XN )
M·Q
=1− Pr(4v1 > v1,i (y), · · · , 4vN > vN,i (y) | Ai ) · Pr(Ai )
i=1
(XN )Y
M·Q
N
1
= 1 − M·Q · Pr(4v j > v j,i (y) | Ai )
N i=1 j=1
(X) Y
M·Q
N N
1
= 1 − M·Q · Pr(4v j > v j,i (y)). (B.12)
N i=1 j=1
bility of BCTS at backoff stage m, respectively. Let ψ0 = b(Y − Fmin )/0 c and ψm = b(Y − Fmin −
Pm−1
i=0 ψi · i )/m c, where m ∈ [1, rmax ]. Without loss of generality, we consider Ft = Y as the
and K j,t,m (x, y, z), H j,t,m (x, y, z) denote the events K j,t,m and H j,t,m under the conditions 4d j =
(X) \
M·Q
N
=N· Pr K j,t,0 | Ai · Pr(Ai )
i=1 j∈N0 , j,t
(X) Y
M·Q
N
N
= M·Q · Pr(K j,t,0 | Ai )
N i=1 j∈N0 , j,t
(XN ) Y
M·Q
N
= M·Q · Pr(K j,t,0 | 4d j = d j,i , 4dt = dt,i )
N i=1 j∈N0 , j,t
(XN ) Y
M·Q
X
(2P−1)!
N
= M·Q · Pr(K j,t,0 | 4d j = d j,i , 4dt = dt,i , 4vt = 4Vl ) · Pr(4vt = 4Vl )
N i=1 j∈N0 , j,t l=1
(XN ) Y
M·Q
X
(2P−1)! h i
N
= M·Q · Pr K j,t,m (d j,i , dt,i , 4Vl ) (B.15)
N
· (2P − 1)! i=1 j∈N0 , j,t l=1
where d j,i , dt,i denote the values of 4d j and 4dt in event Ai , respectively. For m ∈ [1, rmax − 1],
we have
X
N X
x1 X
xm−1 ! \
! !
xm
Sm = ··· Pr K j,t,m | |Nm | = xm · Pr(|Nm | = xm | |Nm−1 | = xm−1 ) · · ·
x1 =2 x2 =2 xm =2
1 j∈Nm , j,t
!
· Pr(|N2 | = x2 | |N1 | = x1 ) · Pr(|N1 | = x1 ), (B.16)
where
(XN )
M·Q
\ 1 Y (2P−1)!
X h i
Pr
K j,t,m | |Nm | = xm = M·Q · Pr K j,t,m (d j,i , dt,i , 4Vl ) ,
j∈Nm , j,t N
· (2P − 1)! i=1 j∈Nm , j,t l=1
(B.17)
and
X X X ! ! !
1 X
N x1 xm−1 Wn
xm 1 xm −1
Sm = ··· · ( ) · Pr(|Nm | = xm | |Nm−1 | = xm−1 ) · · ·
x1 =2 x2 =2 xm =2
1 Wn k=1 Wn − k
!
· Pr(|N2 | = x2 | |N1 | = x1 ) · Pr(|N1 | = x1 ) , (B.19)
In the proposed scheme, if a relay candidate transmits BCTS simultaneously with other relay
candidates and introduces BCTS collisions, it will reply BCTS again after receiving a rebroadcast
BRTS until the retransmission times reach rmax . Then, it will randomly select a mini-slot to reply
the BCTS, and any mini-slot has the same probability 1/Wn to be chosen. If a relay candidate
selects mini-slot k and successfully replies BCTS, other relay candidates should randomly select
mini-slots between k + 1 and Wn , and Eq. (B.19 is obtained. Finally, we can acquire S m and
Cm = 1 − S m . Let tm denote the average time taken for a relay candidate successfully replying
BCTS at backoff stage m, which contains the backoff time, delay of retransmissions caused by
BCTS collisions, and BCTS successful transmission time. Therefore, it can be represented as
ψm · τ + tbcts + m(T b + tdi f s + tbcts ), m ∈ [0, rmax − 1]
tm =
Wn · τ + tbcts + m(T b + tdi f s + tbcts ), m = rmax ,
2
Pm−1
where ψm = b(Y − Fmin − i=0 ψi · i )/m c is the mean of ψm . Given the PMF of Y in Eq. (B.12), Y
and ψm can be obtained. At the backoff stage rmax , a relay candidate uniformly selects a mini-slot
to reply BCTS, and the average number of mini-slots it backoffs is Wn /2. Given S m , Cm , and
R
[1] C. L. Robinson, L. Caminiti, D. Caveney, and K. Laberteaux, “Efficient Coordination and Transmission of Data for
Cooperative Vehicular Safety Applications,” In Proc. of VANET’06, pp. 10-19, Sep. 2006.
[2] J. Luo and J. P. Hubaux, “A Survey of Inter-Vehicular Communication,” Technical Report IC/2004/24, School of Computer
and Communication Sciences, EPFL, 2004.
[3] Dedicated Short Range Communications Working Group, [Online], Available: http://grouper.ieee.org/groups/scc32/dsrc/.
[4] M. Heddebaut, J. Rioult, J. P. Ghys, C. Gransart, and S. Ambellouis, “Broadband Vehicle-to-Vehicle Communication Using
an Extended Autonomous Cruise Control Sensor,” Meas. Sci. Technol., vol. 16, no. 6, pp. 1363-1373, Jun. 2005.
[5] M. Shulman and R. K. Deering, “Third Annual Report of the Crash Avoidance Metrics Partnership April 2003-March
2004,” Nat. Highw. Traffic Safety Admin., DOT HS 809 837, Jan. 2005.
[6] D. Reichardt, M. Miglietta, L. Moretti, P. Morsink, and W. Schulz, “CarTALK 2000: Safe and Comfortable Driving Based
upon Inter-Vehicle-Communication,” In Proc. of Intelligent Vehicle Symposium, PP. 545-550, Jun. 2002.
[7] R. Kruger, H. Fuler, M. Torrent-Moreno, M. Transier, H. Hartenstein, and W. Effelsberg, “Statistical Analysis of the
FleetNet Highway Movement Patterns,” University of Mannheim, Mannheim, Germany, Tech. Rep. TR-2005-004, Jul.
2005.
[8] M. Ergen, D. Lee, R. Sengupta, P. Varaiya, “WTRP-Wireless Token Ring Protocol,” IEEE Trans. Vehi. Technol., vol. 53,
no. 6, pp. 1863-1881, Nov. 2004.
[9] Y. G. Bi, K. H. Liu, L. X. Cai, X. Shen, and H. Zhao, “A Multi-Channel Token Ring Protocol for QoS Provisioning in
Inter-Vehicle Communications,” IEEE Trans. Wirel. Commun., to appear.
[10] Y.-C. Tseng, S.-Y. Ni, Y.-S. Chen, and J.-P. Sheu, “The Broadcast Storm Problem in a Mobile Ad Hoc Network,” Wireless
Networks, vol. 8, no. 2/3, pp. 153-167, 2002.
[11] B. Williams and T. Camp, “Comparison of Broadcasting Techniques for Mobile Ad Hoc Networks,” In Proc. of
MobiHoc’02, pp. 194-205, Jun. 2002.
[12] C. Ho, K. Obraczka, G. Tsudik, and K. Viswanath, “Flooding for Reliable Multicast in Multi-hop Ad Hoc Networks,” In
Proc. of DIALM’99, pp, 64-71, Aug. 1999.
[13] J. Jetcheva, Y. Hu, D. Maltz, and D. Johnson, “A Simple Protocol for Multicast and Broadcast in Mobile Ad Hoc Networks,”
Internet Draft: draft-ietf-manetsimple- mbcast-01.txt, Jul. 2001.
[14] S. Ni, Y. Tseng, Y. Chen, and J. Sheu, “The broadcast Storm Problem in a Mobile Ad Hoc Network,” In Proc. of
MobiCom’99, pp. 151-162, Aug. 1999.
[15] H. Lim and C. Kim, “Multicast Tree Construction and Flooding in Wireless Ad Hoc Networks,” In Proc. of MSWIM’99,
pp. 61-68, Aug. 2000.
[16] W. Peng and X. Lu, “On the Reduction of Broadcast Redundancy in Mobile Ad Hoc Networks,” In Proc. of Mobihoc’00,
pp. 129-130, Aug. 2000.
[17] Q. Xu, T. Mak, J. Ko, and R. Sengupta, “Medium Access Control Protocol Design for VehicleCVehicle Safety Messages,”
IEEE Trans. Veh. Technol., vol. 56, no. 2, pp. 499-518, Mar. 2007.
[18] G. Korkmaz, E. Ekici, and F. Ozguner, “Black-Burst-Based Multihop Broadcast Protocols for Vehicular Networks,” IEEE
Trans. Veh. Technol., vol. 56, no. 5, pp. 3159-3167, Sept. 2007.
[19] Y. Bi, H. Zhao, and X. Shen, “A Directional Broadcast Protocol for Emergency Messages Exchange in Inter-Vehicle
Communications,” In Proc. of ICC’09, pp. 1-5, Jun. 2009.
[20] L. Briesemeister and G. Hommel, “Role-based Multicast in Highly Mobile but Sparsely Connected Ad Hoc Networks,”
In Proc. IEEE/ACM Workshop MobiHOC, pp. 45-50, Aug. 2000.
[21] D. S. J. D. Couto, D. Aguayo, J. Bicket, and R. Morris, “A High-Throughput Path Metric for Multi-hop Wireless Networks,”
In Proc. MobiCom’03, pp. 134-146, Sept. 2003.
[22] R. Draves, J. Padhye, and B. Zill, “Routing in Multi-Radio, Multi-Hop Wireless Mesh Networks,” In Proc. MobiCom’04,
pp. 114-128, Sept. 2004.
[23] A. Zinin, “Cisco IP Routing,” Boston, MA: Addison-Wesley, 2002.
[24] P. Liu, Z. Tao, S. Narayanan, T. Korakis, and S. S. Panwar, “CoopMAC: A Cooperative MAC for Wireless LANs,” IEEE
J. Selected Areas Commun., vol. 25, no. 2, pp. 340-354, Feb. 2007.
[25] J. Peng and L. Cheng,“A Distributed MAC Scheme for Emergency Message Dissemination in Vehicular Ad Hoc Networks,”
IEEE Trans. Veh. Technol., vol. 56, no. 6, pp. 3300-3308, Nov. 2007.
[26] IEEE Standard for Information Technology–Telecommunications and Information Exchange between Systems–Local and
Metropolitan Area NetworksłSpecific Requirements Part 11: Wireless LAN Medium Access Control (MAC) and Physical
Layer (PHY) Specifications.
[27] G. Mohammad and S. Marco, “Maximizable Routing Metrics,” IEEE/ACM Trans. Netw., vol. 11, no. 4, pp. 663-675, Aug.
2003
[28] A. Bletsas, A. Khisti, D.P. Reed, and A. Lippman, “A Simple Cooperative Diversity Method Based on Network Path
Selection,” IEEE J. Selected Areas Commun., vol. 24, no. 3, pp. 659-672, Mar. 2006.
[29] H. Shan, W. Zhuang, and Z. Wang, “Distributed Cooperative MAC for Multihop Wireless Networks,” IEEE Commun.
Mag., vol. 47, no. 2, pp. 126-133, Feb. 2009.
[30] D. X. Xu, T. Sakurai, H. L. Vu, “An Access Delay Model for IEEE 802.11e EDCA,” IEEE. Trans. Mob. Comput., vol. 8,
no. 2, pp. 261-275, Feb. 2009.
[31] T. K. Mak, K. P. Laberteaux, R. Sengupta, M. Ergen, “An Access Delay Model for IEEE 802.11e EDCA,” IEEE Trans.
Veh. Technol., vol. 58, no. 1, pp. 349-366, Jan. 2009.
[32] J. W. Mark and W. Zhuang, Wireless Communications and Networking, Prentice Hall, Inc., 2003.
[33] J. G. Proakis, Digital Communications: Fourth Edition, McGraw-Hill Companies, Inc., 2000.
[34] J. A. Roberts, “Packet Error Rates for DPSK and Differentially Encoded Coherent BPSK,” IEEE Trans. Veh. Technol., vol.
42, no. 2/3/4, pp. 1441-1444, Feb./Mar./Apr. 1994.