IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 21, NO.

10, OCTOBER 2020 4269

A Novel Hybrid MAC Protocol for Basic Safety

Message Broadcasting in Vehicular Networks
Minglong Zhang , G. G. Md. Nawaz Ali , Member, IEEE, Peter Han Joo Chong , Senior Member, IEEE,
Boon-Chong Seet , Senior Member, IEEE, and Arun Kumar

Abstract— Basic Safety Messaging plays a crucial role to vehicles, while the latter is for disseminating emergency mes-
provide road safety in vehicular ad-hoc networks (VANETs). sages (e.g., traffic accidents and hazardous weather condition)
To avoid potential accidents, vehicles periodically broadcast to vehicles in a specific geographical area. A BSM generated
safety information to neighboring vehicles. However, due to trans-
mission collisions, fading channels and other factors, vehicular by each vehicle usually contains a vehicle’s instant status
networks usually suffer a low packet delivery ratio (PDR) and information, such as velocity, direction, acceleration, and posi-
a large delay, which are intolerant of many safety applications. tion. A periodic exchange of BSM packets can create mutual
To tackle these issues, this paper proposes a hybrid medium awareness in surrounding environment, thus avoiding potential
access control (MAC) protocol for basic safety message (BSM) accidents. Noticeable, safety beacon are massively generated
dissemination based on the framework of Dedicated Short-Range
Communication (DSRC). Its partially centralized and partially at all times, but emergency information only originates from
distributed characteristic not only can effectively suppress the a few sources sometimes.
collisions, but keep compatibility with IEEE 802.11p. In addition, In DSRC, vehicles access channels and broadcast BSMs
the integration of Physical-Layer Network Coding (PNC) and via carrier-sense multiple access with collision avoidance
Random Linear Network Coding (RLNC) further strengthens (CSMA/CA). However, this mechanism only works effectively
the reliability and efficiency for BSM dissemination. Both the
theoretical analysis and comprehensive simulations indicate that, in very sparse networks (e.g., less than 10 vehicles in a
compared with existing schemes, the proposed protocol can vehicular network) [2]. With the increase of traffic density,
significantly improve the PDR by a range of 20% to 300%. transmission collisions arise and result in a poor service for
Meanwhile, in terms of normalized throughput, it increases by safety message delivery. Unlike other unicast communications,
varying percent between 20% and 160% in different scenarios. there is no retransmission on the occasion of BSM loss.
Index Terms— Vehicular communications, BSM dissemination, Thus, such unreliability is unacceptable for many periodic
MAC protocol, PNC, RLNC. applications. Meanwhile, BSM packets delivery ratio (PDR)
is worsened by 1) highly dynamic topology; and 2) fading
I. I NTRODUCTION channels caused by buildings, bridges and other large objects
on road [3]. Moreover, another challenge comes from the

O VER the last decade, the advent of vehicular ad-hoc

networks (VANETs) fulfilled various functionalities in
intelligent transportation systems (ITS). One typical and fore-
stringent requirement on rather low delivery latency. Being
distinct from emergency messages, BSMs need to be rapidly
disseminated to all vehicles in the vicinity within its lifetime
most application in ITS is to reduce the risk of road acci-
(e.g., 100 ms in DSRC).
dents by providing a series of safety-related services, such as
Towards solving those issues, a few studies firstly attempted
collision prevention and hazard warning, etc. The dedicated
repetition schemes [4], [5], in which each safety message
short-range communication (DSRC) [1] aimed to realize such
is broadcast repeatedly in several times. Although such
safety applications, which can be divided into two categories:
approaches are able to improve the PDR to some extent,
periodic and event-trigger application. The former is used
they still do not perform well for BSM broadcast in mod-
to prevent accident by periodically broadcasting basic safety
erate and dense networks as collisions caused by repetitive
message (BSM, also known as safety beacon) to neighboring
transmissions rise exponentially. In [6], transmission time is
Manuscript received March 6, 2018; revised July 25, 2018, November 19, dynamically controlled based on the observation of the time
2018, February 24, 2019, May 23, 2019, and August 18, 2019; accepted slots occupied by other vehicles to mitigate collisions, but it
August 28, 2019. Date of publication September 13, 2019; date of current still faces high packet loss rate in dense networks because
version October 2, 2020. The Associate Editor for this article was D. Wu.
(Corresponding author: Peter Han Joo Chong.) some vehicles may be deprived of transmission opportunities.
M. Zhang, P. H. J. Chong, and B.-C. Seet are with the Department Later, researchers tried to incorporate network coding (NC) in
of Electrical and Electronic Engineering, Auckland University of Tech- their protocols [7]–[10] because NC can effectively improve
nology, Auckland 1010, New Zealand (e-mail: [email protected];
[email protected]; [email protected]). the reliability in wireless networks. However, they cannot cope
G. G. Md. N. Ali is with the Department of Automotive Engineering, with the quickly deteriorating performance with the increas-
Clemson University, Greenville, SC 29607 USA (e-mail: [email protected]). ing vehicle volume in networks. The research [11] applied
A. Kumar is with the Department of Computer Science and Engineering,
NIT Rourkela, Rourkela 769008, India (e-mail: [email protected]). NC to address issues in urban hybrid vehicular networks
Digital Object Identifier 10.1109/TITS.2019.2939378 as well, but it focused on a routing scheme for multi-hop
1524-9050 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on August 27,2021 at 08:20:46 UTC from IEEE Xplore. Restrictions apply.
 4270 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 21, NO. 10, OCTOBER 2020

communications among vehicles and/or infrastructure. The nodes to transmit their BSMs via CSMA/CA. 2) It integrates
authors in [12] investigated how to get rid of certain BSMs PNC and random linear network coding (RLNC) [26] to
based on the observation of redundant and derivable infor- exchange BSMs efficiently and reliably. It can be widely
mation among consecutive messages. Although the proposed adopted because both roadway and intersection scenarios are
method is fully compatible with IEEE 802.11p, its applicable taken into account. 3) We provide an insight into the com-
scenarios are very limited. For instance, it cannot reduce munication complexity and theoretically analyze the proposed
BSMs in high-mobility networks. The utilization of protocol protocol. 4) A system model is built and comprehensive
sequence to access channels can reduce collisions, but it targets simulations are conducted to evaluate the performance of the
on optimizing the delivery delay, and the evaluation of PDR proposed protocol and validate the analytical results.
is missed [13]. In fact, all of the mentioned protocols work The rest of this paper is organized as follows. Section II
in a distributed manner and vehicles contend for accessing overviews the related works. Section III presents the system
channels. Consequently, transmission collisions are inevitable. model and defines performance metrics. Section IV describes
Works [14] and [15] try to make use of collisions by means the proposed MAC protocol in detail. Theoretical analysis and
of adopting physical-layer network coding (PNC), but both simulation results are included in Section V and Section VI,
schemes are only suitable for unicast communication rather respectively. Section VII indicates a multi-relay strategy to
than broadcast. Another alternative method [16], [17] that also further enhance the network performance. Finally, Section VIII
employs PNC for message broadcasting works in a partly dis- concludes the whole paper.
tributed way to restrain collisions. However, the performance
is volatile because BSM exchanging in the scheme mostly
II. R ELATED W ORKS
relies on a relay vehicle that may be highly dynamic and has
limited coverage. Diverse strategies proposed to tackle the aforementioned
To obviate contention for channel access, centralized MAC issues for BSM dissemination largely focus on innovation
protocols that employ time-division multiple access (TDMA) at MAC layer. They can be categorized into distributed,
have been proposed [18]–[20]. In [18], a road is divided into centralized and hybrid schemes.
multiple segments and a vehicle is selected in each segment A repetitive transmission approach was proposed in [4].
to assign time slots to other vehicles. Though the scheduling In this mechanism, a control channel (CCH) interval is slotted
can eliminate conflicts, the area of each segment is small and into L length-fixed time slots. Vehicles repeatedly broadcast
the inter-segment BSM dissemination is impossible due to each safety beacon several times to enlarge the likelihood of
simultaneous transmissions in adjacent segments. Paper [19] reception. This idea develops into two basic schemes: syn-
and [20] has the similar idea but the coordinator turns to be chronized fixed repetition (SFR) and synchronized persistent
a roadside unit (RSU). The delay is much larger than the repetition (SPR). In SFR, w (w < L) time slots are randomly
lifetime of the BSM under dense networks and it occupies an selected to broadcast a packet repeatedly, while in SPR a
extra service channel for time slot reservation. Liu et al. [21] packet is sent with a predetermined probability p at each
presented a network coding assisted scheduling algorithm for time slot. Although both schemes can diminish the reception
cooperative data dissemination via V2X communication, but failure, at the same time they increase the chance of concurrent
it is more suitable for non-safety information broadcasting. transmissions, especially in a network with a high density
More recently, in order to alleviate the broadcast storm of vehicles. To limit the number of collisions, a positive
and enhance the message delivery rate, some complicated orthogonal code is introduced in [5], while protocol sequence
broadcast protocols have been put forward. Protocols in [22] is adopted in [13]. A collision-recording based scheme is intro-
and [23] use fuzzy logic to determine the next hop relay duced to control the congestions in [27]. When the number
nodes. Comparably, a forwarding node selection scheme that of detected collisions increases, vehicles begin to decrease
integrates directional broadcast was presented in [24]. The the frequency of BSM transmission. By contrast, in [12],
authors in [25] even tried to use a service channel to relay the authors tried to reduce the number of packets by omitting
safety messages. Unfortunately, all of them are only suitable certain packets containing redundant or derivable informa-
for multi-hop dissemination of emergency messages rather tion in between consecutive sending packets. Y. Park et. al
than BSMs. proposed an application-level control scheme [6], in which
This study is dedicated to improving the reliability and the a CCH interval is divided into a number of asynchronous
efficiency for BSM dissemination in VANETs. We propose a epoches. To minimize the chance of transmission collisions,
hybrid MAC protocol that contains both distributed session the application layer hands over a BSM to the MAC layer in
and centralized session. It not only effectively suppresses the an epoch that is unlikely used by other vehicles.
transmission collisions, but shows compatibility with IEEE Additionally, some researchers also explored applying
802.11p and good robustness in dense networks. The main RLNC in distributed schemes to enhance the reliability. Based
contributions of this study include: 1) A novel MAC protocol on DSRC, paper [7] presented an application of RLNC for
is proposed for BSM broadcast. It consists of two centralized safety message dissemination among a cluster of nodes. Each
sessions and one distributed session. The BSM broadcasting, node has multiple opportunities to rebroadcast network coded
which is coordinated by an RSU mainly takes place in a packets that is produced by encoding its own packet with
centralized contention-free session, while a distributed session the ones overheard from its counterparts. A loss probability
is reserved for legacy vehicles using IEEE 802.11p and other upper bound was derived by using discrete phase type model.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on August 27,2021 at 08:20:46 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: NOVEL HYBRID MAC PROTOCOL FOR BASIC SAFETY MESSAGE BROADCASTING IN VEHICULAR NETWORKS 4271

Fig. 1. Network model of BSM broadcasting in vehicular networks for two typical scenarios: (a) a roadway scenario; (b) an intersection scenario.

The dissemination of BSM packets at an intersection of two its system model and all vehicles being aware of each other’s
roads was investigated in [9] and [10]. The former only instant location in advance, etc) that it is far from feasible.
allows an intersection vehicle to carry out encoding buffered In addition, its PDR performance is prone to the location of
packets before broadcasting, while the latter integrates SPR [4] the relay.
with RLNC at each vehicle to repetitively transmit encoded
packets. Compared with DSRC, both schemes have improved III. S YSTEM M ODEL AND P ERFORMANCE M ETRICS
the packet delivery rate, but the packet reception rate plummets
when the number of vehicles in the network exceeds 40. A. System Model
Since the distributed schemes unlikely eliminate transmis- We are interested in one-hop BSM packets dissemination
sion collisions for the safety messages in a dense vehicular among vehicles in a vehicular network. In particular, each
networks, a few centralized protocols based on TDMA have vehicle is obliged to receive all BSM packets from neighboring
been designed. In [18], a road is divided into a number of vehicles within a time interval (also known as the lifetime of
segments and each segment has a fixed transmission period. a BSM packet).
Different vehicles transmit safety beacons in different time Fig. 1 shows the network model for a road segment scenario
slots assigned by a coordinator vehicle. In [19] and [20], and an intersection scenario. For each scenario, there are
an RSU measures the traffic in its coverage and manages the a number of moving vehicles but one RSU. Each vehicle
time slots for different nodes to ensure every vehicle have its is equipped with an onboard unit (OBU), which is able to
exclusive time slot to broadcast BSMs without contention. communicate with other OBUs and the RSU. Assume all
There are also hybrid MAC protocols for VANETs. The OBUs have the same communication range. Let R and r
authors in [21] developed a NC-assisted scheduling algorithm represent the communication range of the RSU and the OBU,
that fully exploits the joint effect of V2V and V2I com- respectively, where R > r > 0. The communication range
munication. Nodes in the network work in both centralized of the RSU is defined as a region of interest (ROI). In both
and distributed way. However, the algorithm only suits the scenarios, every vehicle disseminates its BSM to all other
non-safety data dissemination due to the large delay. Two vehicles in the same RoI via V2X communication. Some of the
papers [28] and [29] firstly proved the feasibility of imple- vehicles are within each other’s communication range (e.g., v 1
menting PNC at physical layer in vehicular environment. and v 6 in Fig. 1 (a)), but some of them are not (e.g., v 1 and v 3
The former examined the impact of motion-induced carrier in Fig. 1 (a)). In addition, all vehicles in the ROI can receive
frequency offset (CFO) and inter-carrier interference (ICI) by packets sent by the RSU. However, the RSU may not receive
applying conventional equalization methods, while the latter any information from certain vehicles (say v 2 in Fig. 1 (a))
delineated a belief propagation based algorithm to mitigate because it is out of the communication range of those vehicles.
the CFO/ICI effect on signal decoding. Ndih et. al designed A corresponding communication graph is explained in the
a PNC-based MAC protocol named VPNC-MAC [16], [17], next paragraph. To avoid confusion, throughout the whole
which consists of both centralized and distributed periods. paper, OBU, vehicle and node are interchangeable terms in
In VPNC-MAC, a relay vehicle is firstly determined, and then our discussion.
vehicles apply PNC to exchange their BSM packets with the We let S = {v 1 , v 2 , . . . , v N } and  denote the set of all
coordination of the relay. However, the protocol is based on N vehicles in the ROI and the RSU, respectively. We further
so many unrealistic assumptions (e.g., one-dimension road in assume that a vehicle’s sensing range is 2r , and N vehicles

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on August 27,2021 at 08:20:46 UTC from IEEE Xplore. Restrictions apply.
 4272 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 21, NO. 10, OCTOBER 2020

in the ROI follow a Poisson point process distribution and the delivery is usually intolerable. Let Td denote the time required
density of vehicles is α vehicles per kilometre. In addition, to get the BSMs from vehicles in a ROI for making a
d(x, y) is the Euclidean distance between any two points x neighbourhood map of a vehicle. Different number of vehicles
and y on the road, where x, y ∈ S ∪ . We define a hybrid may have different delivery latency.
communication graph HCG to characterize the topology of the 3) Throughput: Throughput is usually defined as how many
vehicular network, as follows: information bits are received by a node per time unit. With
Definition: H C G = G(V, E) is a communication graph regarding to a network, the network throughput is the sum of
with node set V = S ∪  and edge set E = E 1 ∪ E 2 ∪ E 3 ∪ per-node throughput over all nodes. This research investigates
E 4 . For any two vehicles v i, v j ∈ S, E 1 and E 2 contains
 one-hop BSM packet dissemination in a vehicular network.
bidirected edge v i , v j if d v i , v j ≤ r and d v i , v j > r , Therefore, we evaluate the network throughput that exclusively
respectively. For ∀v x , v y ∈ S, E 3 includes bidirected edge involves BSM packet delivery at all vehicles. Let χ(i ) denote
 x , ) if d (v x , ) ≤ r , while
(v  E 4 comprises directed edge total number of BSM packets received at vehicle v i , since
v y ,  if and only if r < d v y ,  ≤ R. there are N vehicles in the network, the network throughput
In the HCG, pairwise vehicles in E 1 are able to communi- can be defined as
cate with each other directly, vehicles connected by E 2 are able 
N
to receive/send signals from/to the RSU. By contrast, vehicles 8l χ(i )
i=1
involved in E 3 can only obtain signals from the RSU, but in TH = , (1)
the opposite direction, their signals cannot reach the RSU. T
We assume all OBUs and the RSU can access GPS signal where T is the overall time spent for all vehicles to complete
and have been synchronized by a mechanism shown in [30]. transmitting their BSM packets, and l is the predefined length
Besides, each vehicle generates a BSM packet at the beginning of a BSM packet in byte (one byte is equal to eight bits).
of every control channel (CCH) interval [1], and then the In DSRC, safety messages are disseminated during the CCH
dissemination process is conducted during the whole interval. interval (i.e., TCC H ). Further, we define a normalized through-
It is noteworthy that although vehicles may move at a high put to indicate the relative efficiency of a protocol while
speed, the network topology can be deemed unchanged during considering safety message broadcasting among all nodes in
a CCH interval. For example, for a vehicle is travelling a vehicular network, as follows:
at 100 km/h, its position change during one CCH interval 
N

N
8l χ(i)
of 50ms will be no more than 1.4 meters, which is not i=1
χ(i )
l
T TCC H i=1
considered as significant. T Hnorm = = . (2)
In order to obtain a better performance, we consider adopt-
8N(N−1) T N(N − 1)
TCC H
ing three decoders at the physical layer of the RSU: one
PNC decoder, one multi-user detection (MUD) decoder and IV. P ROTOCOL D ESIGN AND P ERFORMANCE
one single user successive interference cancellation (SU-SIC) The proposed MAC protocol (called NC-PNC MAC) will
based decoder [31], [32]. The PNC decoder attempts to decode be explained in detail in this section. It consists of three
a packet like Pv a ⊕ Pv b based upon the overlapped signals from consecutive sessions, including MAC setup session, CSMA
two nodes, while the MUD decoder attempts to decode two session and PNC session, as shown in Fig. 2, in which the
individual packets, Pv a and Pv b , which are sent by two vehicles life time of BSM packets can be predefined, according to the
v a and v b , respectively. The SU-SIC decoder is for single-user number of vehicles in a ROI.
case. At the MAC layer, both the RSU and OBUs are capable
of encoding and decoding RLNC packets, as well as uncoded
original packets. A. MAC Setup Session
At the very beginning of each CCH interval, the RSU
broadcasts a polling message to invite vehicles to join in PNC
B. Performance Metrics session. Vehicles that successfully receive the polling message
1) Packet Delivery Ratio: We denote P D R(i ) the packet reply to the RSU after a short inter-frame space (SIFS) interval
delivery ratio for a vehicle i in the ROI. If the vehicle receives through an orthogonal frequency division multiple access
n packets during a CCH interval, we define P D R(i ) = (n + (OFDMA)-based scheme [33]. Each vehicle randomly and
1)/N, where N is the total number of generated BSM packets independently selects one subcarrier out of Nsub subcarriers.
in the interval. If a vehicle collects all N − 1 packets, it is In order to reduce the collision in subcarrier selection, each
able to construct a complete neighbourhood map based on the vehicle randomly chooses one out of Nb transmission batches
information contained in those packets. Since each node has to transmit its join packet, the length of which is very short and
had its own BSM packet, the upper bound and lower bound only includes its ID and current location. Apparently, if two
of the PDR for a vehicle should be 1 and 1/N, respectively. or more vehicles pick the same subcarrier and the same batch,
Different nodes may have different PDRs due to its location collisions will be inevitable.
and channels, so we take the average PDR to evaluate the Since each subcarrier is orthogonal in frequency to each
performance of the entire network. other, the RSU is able to know which node can take part in
2) Delivery Latency: Many safety applications in VANETs the PNC session if no transmission collision takes place. Then
need to update information timely, so a large delay in BSM the RSU broadcasts a coordination packet to notify all vehicles

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on August 27,2021 at 08:20:46 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: NOVEL HYBRID MAC PROTOCOL FOR BASIC SAFETY MESSAGE BROADCASTING IN VEHICULAR NETWORKS 4273

Fig. 2. Packet dissemination process in NC-PNC MAC protocol.

about which two vehicles form a PNC pair as well as the B. CSMA Session
total number of PNC pairs, according to their instant locations. Some nodes may not be able to participate the PNC session
Here, one PNC pair consists of two vehicles that exchange after the MAC setup session because the RSU fails to receive
their safety messages through the PNC scheme [34] during their join packets. The failure is casued by various of factors,
the PNC session. To form a PNC pair, the RSU prioritizes two including fading channels, long distance from the RSU, dis-
vehicles at different sides of the RSU. For instance, the left- ability of sending a join packet and so on. Therefore, the RSU
side node v 1 and the right-side node v 3 in Fig. 1 (a) can be a cannot arrange an opportunity for those vehicles in the PNC
PNC pair, but v 2 and v 3 is not a wise choice because both v 2 session. Among all factors, the dominant one is the subcarrier
and v 3 are at the right side of the RSU. collision, which means some nodes select the identical data
Assume all PNC pairs form a set Q, so Q ⊆ S. For the subcarrier at the same transmission batch during the OFDMA
roadway scenario, Q can be further divided into two subsets uplink stage. The probability of a certain subcarrier and a batch
Q L and Q R , which stands for vehicles on the left and right chosen by n (n ≥ 2) nodes is:
side of the RSU, respectively. Among all PNC pairs in Q,     n
the RSU designates one pair as relay nodes. The relay nodes N Nsub Nb − 1 N−n 1
Pcoll (n) = . (5)
help those nodes, which lose the opportunity to take part in n Nsub Nb Nsub Nb
the PNC session, to forward their safety messages. Their role
Noticeably, it might be that more than one subcarrier and
will be elaborated in subsection IV-C. In order to have the
one batch are repeatedly selected (i.e., it may have multiple
best coverage of other vehicles, a pair of relay nodes v Relay
L
collisions). The overall collision probability will be considered
and v Relay
R must satisfy the following conditions: in Sec. V-A. For all unqualified nodes, the CSMA session
provides them with an opportunity to broadcast their BSM
v Relay
L
∈ Q L and v Relay
L
= arg max (d (v x , )) , packets, if T AI F S > TD I F S > TS I F S .
v x ∈Q L
R R R
   During the CSMA session, all participants contend for
v Relay ∈ Q and v Relay = arg max d v y ,  . (3)
v y ∈Q R accessing channels and transmitting their safety messages by
using the CSMA/CA. Vehicles in the vicinity including the
The conditions above are proved in Appendix A. For the relay nodes can receive these BSM packets and buffer them
intersection scenario, four relays can be selected likewise. for further purpose of decoding the PNC packets or RLNC
Since the length of the poll packet, join packet and coor- packets, which will be illustrated in the next subsection. The
dination packet are very short (i.e., the length of a poll or time TC S M A spent on this session can be calculated by:
join packet is no more than twenty bytes, and the length of a n C
coordination packet is less than two hundred bytes), the total
SM A
 
TC S M A = n C S M A TB S M + tb f [i ] + TD I F S , (6)
time spent on the MAC setup session is also very short. The
i=1
duration can be estimated as follows:
where n C S M A is the number of vehicles participating in the
Tset up = T poll + Nb T j oin + Tcoord + Nb TS I F S + TD I F S , (4) CSMA session, TB S M is a BSM packet transmission time,
tb f [i ] denotes a random back off time in the i t h transmission.
where T poll , T j oin , and Tcoord are the transmission time for one
poll packet, join packet and coordination packet, respectively,
C. PNC Session
Nb is the total number of batches, and TS I F S (or TD I F S ) is
the waiting time before sending a join packet (or the waiting The RSU launches the PNC session after the CSMA session.
time before the next session). It waits for an arbitrary inter-frame space (AIFS) before

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on August 27,2021 at 08:20:46 UTC from IEEE Xplore. Restrictions apply.
 4274 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 21, NO. 10, OCTOBER 2020

broadcasting a beacon packet, which requests the first PNC selected as relay nodes in Fig. 1, the k t h (1 ≤ k ≤ M) two
pair to transmit their BSM packets. After an interval of SIFS, compound uplink packets generated at two relay nodes are:
the first PNC pair send their BSM packets simultaneously. The
RSU tries to decode the overlapped signals by utilizing the 
ML 
MR
L
PRelay (k) = R
dki Pv i and PRelay (k) = ckj Pv j , (7)
three physical-layer decoders aforementioned in Section III. i=1 j =1
After decoding, the RSU will broadcast a downlink packet,
which also notifies the next PNC pair by piggybacking their respectively, where dk i (or ckj ) are random coefficients at
ID information. Upon receiving, the next PNC pair start node v 1 (or v 3 ), and Pv i (or Pv j ) are its own packet as well
the next round of BSM packet exchange, and this process as packets stored during the CSMA session. Based on the two
repeats for all PNC pairs until the last one. For the last RLNC uplink packets shown by (7), the PNC decoder of the
PNC pair, the process is slightly different from those before. RSU may generate a downlink packet also called a compound
The last pair comprises of two relays (or four relays at downlink packet as follows:
each direction for the intersection case). Each relay adopts comp
PRSU (k) = PRelay
L
(k) ⊕ PRelay
R
(k)
the RLNC to encode its own BSM packet with other BSM
packets buffered during the CSMA session. Being different 
ML 
MR

from other non-relay PNC pairs, the relay nodes work in a = ( dki Pv i ) ⊕ ( ckj Pv j ). (8)
i=1 j =1
burst mode: They consecutively conduct M rounds of exchange
but with different encoding coefficients for each round. Coding After receiving a downlink packet, each PNC pair is able
coefficients are randomly generated from Galois field (GF). to obtain their desired information by XORing the downlink
Here M = max(M L , M R ) + 1, where M L and M R is the packet with their own packets. Notably, nodes that have
number of BSM packets in the buffer of the left and right collected sufficient number of compound downlink packets can
relay, respectively, taking into account the roadway case. also obtain all individual packets Pv i ∀i (or Pv j ∀ j ) by carrying
In order to distinguish different signals, the notation Pv i , out the Gaussian Elimination.
PRelay and PR SU means a packet sent by a vehicle v i , a relay An interesting and remarkable phenomenon is that some
and the RSU, respectively. Considering the compatibility, other vehicles also possibly restore original packets when
we utilize the same MAC header of IEEE 802.11p [35] receiving downlink packets that belong to other PNC pairs if
but assign different values in Frame Control (FC) field for they can overhear their uplink packets. Moreover, for a relay
different types. More details about MAC header and FC values node, since it has the best coverage, most of the surrounding
assignment are in Appendix B. vehicles get benefits from decoding compound downlink pack-
According to three physical-layer decoders in the RSU, ets, since those nodes can overhear the compound uplink from
the decoding results based upon the superimposed uplink the relay. This characteristic is helpful to disseminate BSM
signals would be one of the following: packets. On the other hand, some vehicles may not overhear
1) One lone packet: e.g., only packet Pv 4 from the vehicle all uplink packets, so they cannot restore the original packets
v 4 is decoded by the MUD decoder or the SU-SIC decoder; even though they have received downlink packets from the
2) A bit-wise coded packet: e.g., Pv 4 ⊕ Pv 5 through the PNC RSU. This issue will be tackled by our extended multiple relay
decoder; scheme, which will be discussed in Section VII.
3) Two lone packets: e.g., both Pv 4 and Pv 5 through the The total time spent on PNC session is:
MUD decoder;
4) No packet is obtained. 
M
 
T P NC = (n P NC − 1)(Tup + Tdown ) + Tup + Tdown
For case 2) and 3), the RSU will transmit a downlink i=1
packet such as PRPSU NC = P ⊕ P . For case 1), it will
v4 v5 + 2TS I F S (M + n P NC − 1) + T AI F S
buffer a lone packet at first, and then transmit an encoded
RLNC packet PRRSU L NC =  h P , where P is a previously = 2(M + n P NC − 1)(TB S M + TS I F S ) + T AI F S . (9)
i vi vi
i
buffered lone packet and h i is a random coefficient generated where Tup = Tdown = TB S M are transmission time for one
in GF field. In case 4), the RSU will also send an encoded uplink packet or downlink packet, and n P NC is the total
packet by encoding the buffered lone packets but with different number of PNC pairs.
coefficients in case 1). To ensure vehicles can decode all lone To clearly show the methodologies applied in the PNC
packets, we conceive a lower triangular matrix for all h i . That session and the entire procedure of the protocol, a flow chart
is, when the RSU obtains a lone packet at the first time, is provided in Appendix C.
it just forwards it to all vehicles without encoding. Then,
V. T HEORETICAL A NALYSIS
at the second time, it encodes the two packets (one is the
packet stored at the first time) and broadcasts the encoded A. Collision Probability for Subcarrier Selection
packet. Similarly, at the t t h time when the RSU decodes a lone As aforementioned, when each vehicle randomly and inde-
packet, it encodes all t packets and broadcasts the encoded pendently chooses one batch and one subcarrier to transmit
packet for downlink. its join packet, it may lead to transmission collisions. In this
As pointed out above, relay nodes generate and transmit M section, we will firstly derive the proportion of vehicles that
distinct uplink packets. As a case in point, if v 1 and v 3 are involves in carrier collisions.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on August 27,2021 at 08:20:46 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: NOVEL HYBRID MAC PROTOCOL FOR BASIC SAFETY MESSAGE BROADCASTING IN VEHICULAR NETWORKS 4275

In order to figure out how many vehicles involved in the Therefore, we deem PP NC ≈ pe in the following analysis.
subcarrier collision, we firstly calculate the repeated times Additionally, if a receiver v i locates within the coverage of a
of selected subcarriers. The whole process can be equivalent sender v j (i.e., 0 < D = d(v i , v j ) ≤ r ), we take pe = 0 (i.e.,
with the case that N vehicles randomly pick up subcarriers a reception can be guaranteed in the coverage). Otherwise,
out of S = Nb Nsub subcarriers, where Nb is the batch 0 < pe < 1 and can be calculated by [37]:
volume and Nsub is the number of data carriers at the physical  m
−m 2 a Ns 1 kγt h
layer. Apparently, N subcarriers will be selected (including the pe = γ √ (t + , ), (14)
k 2(t + 1) π 2 2
repeated ones). Let  denote the set of selected subcarriers
and x i is the i t h subcarrier in this set, where x i ∈  and where γ denotes the average signal to noise ratio during one
1 ≤ i ≤ N. We define an indicator Ixi as follows: frame, m is the fading figure in Nakagami fading model, k is
 the number of bits persymbol,Ns 
is the number of symbols
1, if x i has appeared in , 2
Ixi = (10) in a packet, γt h = 2k er f c−1 N2s and a = m m / (m).
0, otherwise. According to (14), pe is determined by multiple variables.
The probability for Ixi = 0 and Ixi = 1 are In terms of a specific data rate, a fixed transmission power
and the channel model, the erasure probability is a function
N −1 i−1 N −1 i−1
P(Ixi =0) = ( ) and P(Ixi =1) = 1 − ( ) . (11) of distance. Therefore, vehicles at different locations probably
N N have different data rates even if signals are from the same
Therefore, the expectation of repeated times for all subcar- transmitter. For simplicity, if D > r , we average pe over the
riers in  is: distance:
 D

N 
n
1
E Nr p = E Ixi = P(Ixi =1) pe = pe (x)d x. (15)
D −r r
i=1 i=1
 
1 N C. Communication Complexity
= N −S 1− 1− . (12)
S The communication complexity of a MAC protocol is
In addition, the probability of choosing one usually defined by the total amount of packets sent by all
 certain sub-
N  S−1  N−1 1 nodes in a network. Since there are N vehicles in the ROI
carrier out of S subcarriers only once is S S, in our network model, let N T be the total number of packets
1
and 
 the probability of a certain subcarrier never chosen is sent by all vehicles and the RSU during one dissemination
N  S−1  N interval, so N T = N T1 + N T2 + N T3 , in which three
S . Therefore, the proportion of vehicles associ-
0 right-side items represent the amount of packets sent in three
ated with subcarrier collisions is sessions respectively. Without loss of generality, we take the
    
N 1  S−1  N−1 N  S−1  N roadway scenario to evaluate the communication complexity
S 1− − + E Nr p and PDR performance. Similar results can be acquired for the
1 S S 0 S
Pcoll = intersection scenario as it applies the same protocol.
N
1 During the first session, each vehicle sends a join packet
= 1 − (1 − ) N−1
. (13) following the polling packet. By adding the coordination
Nb Nsub
packet, we have N T1 = N + 2.
Fig. 3 (a) shows the trend of proportions of vehicles
In the second session, vehicles that are ineligible for the
involved in subcarrier collisions. According the PHY of
PNC session access the channel and broadcast BSMs by means
DSRC, the number of data subcarrier is 48. Thus, we take
of CSMA/CA. The ineligibility, on one hand, is the result of
Nsub = 48 in simulations and analysis. Apparently, more
subcarrier collisions caused by the random OFDMA scheme,
vehicles and fewer uplink batch Nb will cause more collisions,
and on the other hand, comes from the large distance between
which conforms to our expectation. In addition, it is evident
some vehicles and the RSU (i.e., vehicles associated with the
that the analytical model precisely predicts the collision pro-
edge set E 4 in H C G). According to the property of Poisson
portions as any two relevant curves are overlapped.
process, the probability of k vehicles involved in the E 3 is
(2αr )k −2αr
B. Channel Erasure Probability P (k, 2r ) = e . (16)
k!
It has been shown that Nakagami distribution with proper
The expected number of vehicles in this range is
parameters would be a realistic channel model [36]. To be ∞ 
more realistic, we assume that all channels in the network are 
Nakagami fading channels with an erasure probability pe for E[K in ] = k P (k, 2r ) = 2αr  . (17)
the point-to-point communication case. In addition, let PP NC k=0

coll  vehicles are

denotes the probability that a RSU node cannot successfully Among those vehicles, Nin con = 2αr P

decode a PNC packet from superposed signals sent by two unqualified to take part in the PNC session due to subcarrier
nodes simultaneously. Generally, under the same condition, collisions. Nevertheless, their counterparts involved in E 4 that
pe ≤ PP NC . However, by using some advanced algorithms in might have one or both issues mentioned above lose the
PHY [29], the gap between them can be greatly diminished. opportunity. Likewise, the total number of vehicles and the

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on August 27,2021 at 08:20:46 UTC from IEEE Xplore. Restrictions apply.
 4276 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 21, NO. 10, OCTOBER 2020

ineligible vehicles of this type is E[K out ] = 2α(R − r ) and

con = 2α(R − r ) (1 − (1 − P
Nout coll ) (1 − pe )), respectively.
Since all the ineligible vehicles would broadcast a BSM during
the second session. Thus, the total number of transmissions in
this session is
N T2 = Nin
con
+ Nout
con

= 2αr Pcoll  + 2α(R − r ) (1 − (1 − Pcoll ) (1 − pe )) .

(18)
All the eligible nodes exchange their safety messages apply-
ing physical layer network coding in the last session. They
consist of a number of non-relay nodes and a pair of relay
nodes. Obviously, the number of non-relay nodes is Fig. 3. (a) Proportion of vehicles involved in subcarrier collisions; (b) Total
number of transmissions N T in one CCH broadcast interval.
Nnode = 2 αr (1− Pcoll )−2+2 α(R−r )(1− Pcoll )(1− pe ) .
(19)
The relay nodes encode and transmit multiple packets that
are overheard and buffered during the CSMA session. Since the
CSMA/CA may also incur transmission collisions, according
con
to [38], the probability of a successful transmission for Nin
con
nodes and Nout nodes are (remember the sensing range is 2r )
  Pcoll min( 2R+3r α,4αr)−1
1 2
PtC−in
SM A
= 1− and
CW + 1
  Pcoll min( R+3r α,4αr)−1
1 2
Pt −out = 1 −
CSM A
, (20)
CW + 1
Fig. 4. Analytical results vs simulation results for the average PDR
respectively, where C W is the contention window ranging performance under the different data rates in: (a) the roadway scenario; (b) the
between C Wmin (15) and C Wmax (1023). Therefore, the num- intersection scenario.
ber of packets relays received (adding their own packets) is
 
Nrelay = Nin Pt −in + Nout
con C S M A
Pt −out (1 − pe ) + 2. (21)
con C S M A where RC V (i ) is the total number of received BSM packets
of v i . The BSM broadcast only occurs in the CSMA session
Considering for each pair of uplink packets either from and the PNC session. As a consequence, the average PDR is
the relays or the non-relay nodes, the RSU will broadcast the summation of the average PDR in two sessions, as below
a downlink PNC packet after decoding, we can derive the
total in the PNC session is N T3 = P D R = P D RC S M A + P D R P NC . (24)
 number of transmissions

3
2 N node + N Relay . During the CSMA session, only interference-free transmis-
Finally, by adding up three components, we can obtain sions may lead to successful receptions, as transmission col-
N T = N T1 + N T2 + N T3 . To easily see the communication lisions cannot be eliminated even if applying the CSMA/CA.
complexity, after performing some simplifications, we have: According to the analysis in the last subsection, the number
of vehicles involved in E 4 that would transmit BSMs is Nout con ,
N T < 3 [N(1 + pe + Pcoll ) − αr (1 + pe )] (22)
and the probability of a successful transmission from v i is
Therefore, the communication complexity of the NC-PNC S M A (v ). Therefore, taking into account all transmissions
PtC−out i
MAC is linear and can be expressed as O(N). The conclusion from vehicles in this set, the average PDR of the network is
can also be observed in Fig. 3 (b), in which the total number
P D R1csma
of transmissions rises up linearly as the number of vehicles con
increases, and the analytical results are consistent with the N
out  
PtC−out
SM A
(v i ) (R X 1 (v i )+(1− pe)(N − R X 1 (v i )))
simulation results. i=1
=
N(N − 1)
D. PDR Analysis con P C S M A (E[R X ] + (1 −
Nout pe )(N − E[R X 1 ]))
t −out 1
= ,
The average PDR of vehicles in the ROI can be expressed N(N − 1)
by (25)

N in which
RC V (i )   
i=1 α(R + r )
PDR = , (23) E[R X 1 ] = min − 1, 2αr − 1 (26)
N(N − 1) 2

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on August 27,2021 at 08:20:46 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: NOVEL HYBRID MAC PROTOCOL FOR BASIC SAFETY MESSAGE BROADCASTING IN VEHICULAR NETWORKS 4277

can overhear the uplink packets. Consequently, they can

recover the original packets after receiving the corresponding
downlink PNC packets from the RSU. By using the
expectation of R X P NC , the network PDR regarding the
non-relay nodes can be expressed by (29) and (30).
P D R node
P NC1
in   
Nnode r
= · min α(R + ) − 1, 2αr − 1
N(N − 1) 2
    
r
+ N − min α(R + ), 2αr (1 − pe ) (29)
2
P D R node
P NC2
out   
Nnode 1
= · min α(R + r ) − 1, 2αr − 1
N(N − 1) 2
    
1
+ N − min α(R + r ), 2αr (1 − pe ) (30)
2
Vehicles on the left side that have received the left relay’s
uplink packets are able to restore the packets encoded by the
right relay, and vice versa for the right-side vehicles. On either
side, vehicles sending BSMs to a relay can be divided into two
groups that belong to E 1 and E 2 respectively. The number
relay relay
of vehicles in two sets is denoted by Nt x1 and Nt x2
respectively, where
relay
Nt x1 = αr Pcoll +min (αr, α(R−r )) (1−(1− Pcoll )(1−pe )),
relay
Nt x2 = max (0, α(R−2r )) (1 − (1 − Pcoll )(1 − pe )) .
(31)
Among those nodes, the probability of successful transmis-
Fig. 5. Average time consumed of each session with different number of
sion to the relay is
vehicles in: (a) the roadway scenario; (b) the intersection scenario. 1
) Nin +min(2αr,α(R−r)) pout −1 , (32)
relay con
Pt x = (1 −
denotes the expectation of R X 1 (v i ) = CW + 1
min (α (r + R − d(v i , )) − 1, 2αr − 1), which stands where pout = [1 − (1 − Pcoll )(1 − pe )]. Since packets from
relay
for the number of vehicles falling into the communication Nt x2 vehicles might be received with the probability 1 − pe ,
range of v i . On the other hand, for vehicles involved we have
in E 3 , 0 < d(v i , ) ≤ r . Based upon (17) and (20), 1  
relay relay relay relay
the corresponding PDR can be acquired similarly: P D R P NC = · Pt x Nt x1 + Nt x2 (1 − pe )
N(N − 1)
 
Pt −in (E[R X 2 ]+(1− pe)(N − E[R X 2 ]))
con C S M A
Nin · Rx
relay relay
+ (1 − pe )(N − Rx ) , (33)
P D R2csma = ,
N(N −1) relay
(27) where Rx = min (2αr, Rα) is the number of nodes in
   a relay’s coverage. By adding the two parts of PDR from
where E[R X 2 ] = min 12 α(2R + r ) − 1, 2αr − 1 . The non-relay nodes (i.e., (29) and (30)) and the one from relays
summation of P D R1csma and P D R2csma would lead to the PDR (i.e., (33)), we can get the PDR for the PNC session
of the second session.
relay
For the PNC session, let’s consider the non-relay nodes at P D R P NC = P D R node
P NC1 + P D R P NC2 + P D R P NC .
node
(34)
first. The total number of those vehicles is Nnode , which can
in out By substituting (34) and P D RC S M A = P D R1csma +
be further divided into two parts, Nnode and Nnode . The former
P D R2csma with relevant variables in (24), the average PDR
comes from eligible vehicles involved in E 3 , while the latter
of the entire network can be derived.
is the number of their counterparts in E 4 . From (19), we have
On the basis of (24) to (34), we plot the PDR performance
 
in
Nnode = 2 αr (1 − Pveh_coll ) − 2 and against various vehicle volumes for both the roadway and the
out
  intersection scenario where r = 100 m and R = 300 m,
Nnode = 2 α(R − r )(1 − Pveh_coll )(1 − pe ) . (28)
as shown by Fig. 4. In addition, by changing Ns and k in (14),
Suppose a non-relay node vi , the amount the corresponding pe for different data rates can be calculated.
of neighbours in its coverage is R X P NC = The figure also demonstrates the PDRs for three typical data
min (α (r + R − d(v i , )) − 1, 2αr − 1). Those nodes rates of DSRC. Under the same condition, lower data rate

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on August 27,2021 at 08:20:46 UTC from IEEE Xplore. Restrictions apply.
 4278 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 21, NO. 10, OCTOBER 2020

TABLE I
K EY P HYSICAL AND MAC L AYER PARAMETERS

will lead to better PDR performance. Furthermore, to validate

the theoretical analysis presented above, simulations are con-
Fig. 6. PDR comparison among NC-PNC MAC and other existing schemes
ducted in NS-3 under the same configuration (the simulation for: (a) the roadway scenario; (b) the intersection scenario.
setup will be elaborated later in Section VI). The analytical
results follow tightly the simulation ones, which approve the
PDR performance of the proposed protocol. B. Batch Volume Selection in OFDMA Uplink
It is necessary to be aware of the probability of subcarrier
VI. S IMULATION R ESULTS collision while specifying an appropriate value for Nb in sim-
In this section, we first present the simulation setup for ulations. To keep the collision probability on a low level and
our proposed NC-PNC MAC protocol. The simulation tool according to Fig. 3, we need to keep Nb as large as possible.
we use is NS-3 [39]. Then we use simulation data to evaluate However, this strategy will expand the MAC setup session, thus
its performance and compared it with benchmarks. enlarging the delivery latency substantially. Therefore, to keep
balance between collision probability and time consumption,
A. Simulation Setup we set Nb = 4 if the number of vehicle is less than 80, else
Nb = 8 in the simulation. Fig. 5 reveals the effect of such
The simulation is set up according to the system model strategy. It shows that the time spent on the MAC setup session
in Section III. The key physical and MAC layer parameters almost doubles when the number of vehicle increases from
are listed in Table I. The dissemination of BSM packets below 80 to above 80.
experiences Nakagami fading. The probability density function
of the signal amplitude X is subject to
C. Simulation Results and Discussions
2m m x 2m−1 −m x 2 As mentioned before, BSM packets contain real-time safety
f X (x) = eω , (35)
(m)ωm information. Each BSM packet has a limited life time and it
where we assume fading figure m is 1.5 (for distance less should be disseminated within a certain period. Apparently,
than 80m) and 0.8 (for distance larger than 80m), according the proposed protocol consumes different amount of time as
to related empirical measurements [40]. In addition, a so- the number of vehicles varies. Fig. 5 illustrates the average
called log-distance path loss model is added. In the simulation, time consumed (e.g., Tset up + TC S M A + T P NC ) for two scenar-
vehicles distribute randomly on each lane and move with a ios with two typical data rates. The latency increases linearly
constant speed. We run the simulation 500 times. For each as the number of vehicles increases, which is substantiated
time, the whole process depicted by Fig. 2 is continuously by the communication complexity in Section V-C. It can be
conducted 100 rounds, and vehicles have different initial up to 95 ms (or 99 ms) in the roadway scenario (or the
positions and moving speed. intersection scenario) for 160 vehicles with a 6 Mbps data
In order to validate the effectiveness and efficiency of rate. Remarkably, under the same condition (i.e., the same
the proposed protocol, we also choose several other MAC data rate and the same number of vehicles), the latency of
protocols as benchmarks with the same configuration. These the intersection case is larger than that of the roadway case,
protocols are IEEE 802.11p, which utilizes CSMA/CA to because there are one more pair of relays in the intersection,
access channels, NC Relaying scheme [10], in which both which need more time to exchange compound packets shown
RSU and vehicles are able to encode and forward the message in (7) and (8).
they received, and RSU-NC [9], in which RSU carries out The PDR performance of the proposed NC-PNC MAC is
RLNC scheme to relay packets while vehicles still adhere evaluated in Fig. 6 with respect to the proportion r/R =
to CSMA/CA. The communication range can be changed by 1/3 and data rate = 6Mbps. In order to be fair, the three
adjusting transmission power. The definition of a communica- benchmarks spend the same time period of the NC-PNC
tion range is that vehicles can receive the transmitted packet MAC. Compared with three benchmarks, the NC-PNC MAC
more than 95% probability within this range. According to significantly improve the PDR performance in both scenarios
this, in our simulation, the communication range is 100 m, as well as different networks with varied vehicle density.
150 m, 200m and 300 m, corresponding to the transmission Unlike benchmarks, it can maintain a relative high PDR (no
power of 30 dBm, 35 dBm, 40 dBm and 45 dBm, respectively. less than 0.8) even when the number of vehicles reaches 80.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on August 27,2021 at 08:20:46 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: NOVEL HYBRID MAC PROTOCOL FOR BASIC SAFETY MESSAGE BROADCASTING IN VEHICULAR NETWORKS 4279

and distributed part(CSMA session). Therefore, after the PNC

session, channels enter an idle state until the next round
if the actual consumed time is less than the CCH interval.
On the other hand, although the proposed protocol has a much
better PDR performance than several benchmarks, the average
PDR still does not reach as close as 100%, based upon the
observation in Fig. 6. There are many factors accounting
for this phenomenon, such as decoding failure at the RSU
or OBUs because of fading channels, insufficient packets
collected to decode a compound packet and so on. However,
the dominant factor is that some remote nodes cannot obtain
desired packets even if they have received downlink PNC
Fig. 7. PDR performance for different proportions of two communication packets from RSU, because they fail to overhear the uplink
ranges in: (a) the roadway scenario; (b) the intersection scenario. packets. For instance, in Fig. 1 (a), the vehicle v 5 is out of the
transmission range of v 4 . If v 4 takes part in the PNC session,
v 5 possibly does not receive the uplink packet Pv4 . Thus, even
By contrast, the PDR curves of others plunge as the number of if the node v 5 can receive a downlink packet Pv 4 ⊕ Pv x from
vehicles increases. The benefit firstly stems from the property the RSU, it still is not able to retrieve the target packet Pv x ,
of the NC-PNC MAC, which effectively suppresses collisions, which is a safety message from a certain node v x .
because BSM transmissions mainly take place during the Based upon the discussion above, there are two strategies
collision-free PNC session and only a small part of vehicles to deal with the problem. The first one is to extend the PNC
involves in the CSMA session. In addition, by virtue of relays session by blindly reassigning new PNC pairs. Let assume
and the RSU, remote nodes and nodes with transmission col- node v 6 and node v x become a PNC pair, node v 5 can decode
lisions can also broadcast their safety messages by integrating Pv x with a high probability after receiving Pv 6 ⊕ Pv x from
RLNC and PNC. However, the collision rate in other schemes the RSU if it has overheard uplink packet Pv 6 previously.
is inevitably high, especially for dense vehicular networks. However, this strategy cannot guarantee that those distant
Thus, massive packets are lost and a high PDR cannot be nodes like v 5 can always decode the target information from
ensured. downlink packets. In addition, intentionally shuffling PNC
To see how the proportion of two communication ranges pairs to ensure a high decoding probability is not easy and
r and R affects the PDR performance, we compare the PDR needs massive calculations at RSU.
for different proportions, as shown in Fig. 7. The simulation By contrast, the second strategy proposed is able to solve
results show that with the increase of r/R, a better PDR the issue more effectively. In this strategy, extra relays are
performance can be reached. The reason is that when the assigned (called auxiliary relays) along with relays set previ-
proportion goes up, more edges are included in the set E 1 and ously (called primary relays). To make a distinction, we denote
E 3 in H C G. As a result, more vehicles can communicate with this strategy NC-PNC-MR MAC. Both primary and auxiliary
each other directly and may take part into the PNC session, relays apply RLNC to encode and forward BSM packet but
which in turn lead to a larger PDR. with distinguished purposes. Primary relays only serve to relay
Data rate is another key factor that can impose on PDR packets received during the CSMA session, while auxiliary
performance in vehicular networks. Generally, a vehicular relays intend to relay BSM packets from vehicles in the
network with dense vehicles may need more time to complete vicinity of RSU to remote vehicles (i.e., inside vehicles close
a round of dissemination. To avoid exceeding the life time of to the rim of the larger solid circle in Fig. 1).
the safety messages, higher data rate should be considered The RSU selects one auxiliary relay v aux Relay at each side,
for those networks. At physical layer, different data rates according to the location information contained in join packets
imply applying different coding rates and modulations, which sent during the first setup session. Therefore, for the roadway
have impacts on packet error rate at reception. The solid scenario, one at left side and the other at the right side of the
curves in Fig. 4 is the PDR performance for a network RSU are selected. Certainly, four auxiliary relays are selected
adopting various typical data rates specified in [1]. Lower data for the intersection case. We apply the following conditions to
rates result in a better PDR performance, which meets our determine a left-side auxiliary relay:
expectation. More importantly, despite with a high data rate
of 12 Mbps, compared with other schemes using a lower data
rate of 6 Mbps, the proposed MAC has a much better PDR v aux_L
Relay ∈ Q and
L

performance than IEEE 802.11p and even maintains the same v aux_L
Relay = arg  max  (d (v x , )) . (36)
level as the NC based scheme (see Fig. 6). v x ∈ Q L \v Relay
L

VII. NC-PNC MAC W ITH M ULTIPLE R ELAYS The remaining auxiliary relays can also be identified in a
Distinguishing from some distributed protocols like similar way. Once both the primary relays and the auxiliary
IEEE802.11p, the NC-PNC MAC is a hybrid protocol with relays are determined, the RSU also announces them in the
both centralized parts (MAC setup session and PNC session) coordination packet.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on August 27,2021 at 08:20:46 UTC from IEEE Xplore. Restrictions apply.
 4280 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 21, NO. 10, OCTOBER 2020

TABLE II
FC VALUES FOR D IFFERENT T YPES OF S IGNALS IN NC-PNC MAC

Fig. 8. Performance for NC-PNC-MR MAC with different number of

vehicles (r/R=1/3, data rate=6 Mbps) in: (a) PDR performance; (b)average
consumed time.

Fig. 9. Normalized throughput comparison among different schemes

(r/R=1/3, data rate=6 Mbps) in: (a) the roadway scenario; (b) intersection
scenario.

An auxiliary relay node does not necessarily help encode

and forward all overheard uplink safety messages for the
purpose of keeping efficiency. As aforementioned, it only
relays the packets from vehicles close to the RSU. There-
fore, we define a forwarding range drelay aux
= R/3, which
means that for any vehicle v i ∈ Q, if d (v i , ) < drelayaux
,
their packets will be forwarded by a corresponding relay Fig. 10. Applied methodologies in the NC-PNC MAC.
node v aux
Relay .
In order to evaluate the performance of the NC-PNC-MR VIII. C ONCLUSION
MAC, simulations are conducted in NS-3, for both scenarios. This study is motivated by unsolved issues in basic safety
The PDR performance and average consumed time are shown messages dissemination in vehicular networks. We proposed
in Fig. 8. From the first figure, we can see that two PDR curves a novel hybrid MAC protocol to tackle the existing issues.
increase about 15%, compared with the NC-PNC MAC. The It consists of two centralized sessions and one distributed
PDR is no less than 0.9 when the number of vehicles is below session. The BSMs are mainly broadcast over the centralized
80 and can ensure a PDR of 0.8 even the number of vehicles PNC session while a few remnant vehicles broadcast BSMs
reaches 160. At the same time, the consumed time increases during the CSMA session, which is also reserved for legacy
approximately 10%∼15%. vehicles applying IEEE 802.11p. Such characteristic is able
Finally, we compare the normalized throughput among five to effectively suppress the transmission collisions and keep
MAC protocols for both the roadway and the intersection sce- the compatibility with DSRC. In addition, the integration of
nario in Fig. 9. Both NC-PNC MAC and NC-PNC-MR MAC PNC and RLNC greatly improves the efficiency and reliability
have higher throughput than several benchmarks. In another during the one-hop dissemination via the V2V and V2I com-
word, the proposed schemes have higher efficiency for safety munications. Moreover, we theoretically analyze the subcarrier
information dissemination in vehicular networks, for both collision probability, the communication complexity and the
sparse and dense networks. PDR performance. The analytical results are validated by

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on August 27,2021 at 08:20:46 UTC from IEEE Xplore. Restrictions apply.
 ZHANG et al.: NOVEL HYBRID MAC PROTOCOL FOR BASIC SAFETY MESSAGE BROADCASTING IN VEHICULAR NETWORKS 4281

[2] Y. P. Fallah, C. Huang, R. Sengupta, and H. Krishnan, “Congestion

control based on channel occupancy in vehicular broadcast networks,”
in Proc. IEEE Veh. Technol. Conf., Sep. 2010, pp. 1–5.
[3] F. Bai, D. D. Stancil, and H. Krishnan, “Toward understanding char-
Fig. 11. IEEE 802.11p MAC header. acteristics of dedicated short range communications (DSRC) from a
perspective of vehicular network engineers,” in Proc. ACM Int. Conf.
Mobile Comput. Netw., Sep. 2010, pp. 329–340.
simulations implemented in NS-3. Both the analytical results [4] Q. Xu, T. Mak, J. Ko, and R. Sengupta, “Medium access control protocol
and simulation results show the proposed MAC outperforms design for vehicle–vehicle safety messages,” IEEE Trans. Veh. Technol.,
its counterparts and has a better scalability regarding different vol. 56, no. 2, pp. 499–518, Mar. 2007.
[5] F. Farnoud, B. Hassanabadi, and S. Valaee, “Message broadcast using
traffic densities. Finally, by making using of the remaining idle optical orthogonal codes in vehicular communication systems,” in Proc.
time during a dissemination interval, a multi-relay scheme is 1st Int. Workshop Wireless Netw. Intell. Transp. Syst., 2007, pp. 1–4.
designed to boost the PDR performance further. [6] Y. Park and H. Kim, “Collision control of periodic safety messages with
strict messaging frequency requirements,” IEEE Trans. Veh. Technol.,
vol. 62, no. 2, pp. 843–852, Feb. 2013.
A PPENDIX A [7] B. Hassanabadi and S. Valaee, “Reliable periodic safety message
P ROOF OF R ELAY N ODES S ELECTION broadcasting in VANETs using network coding,” IEEE Trans. Wireless
Commun., vol. 13, no. 3, pp. 1284–1297, Mar. 2014.
This appendix proves two conditions for relay nodes selec- [8] J.-S. Park et al., “Delay analysis of car-to-car reliable data delivery
tion that ensure the best coverage, as follows: strategies based on data mulling with network coding,” IEICE Trans.
Inf. Syst., vol. E91-D, no. 10, pp. 2524–2527, Jan. 2008.
proof: Without loss of generality, take the left relay node at
[9] Y. Gao, G. M. N. Ali, P. H. J. Chong, and Y. L. Guan, “Network coding
first. On the left side of the RSU, for a vehicle v x ∈ Q L , since based BSM broadcasting at road intersection in V2V communication,”
it is communicable to the RSU, so 0 ≤ d(v x , ) ≤ r . Obvi- in Proc. IEEE Veh. Technol. Conf., Sep. 2016, pp. 1–5.
ously, any vehicle v i between v x and RSU is communicable [10] Y. Gao, G. G. M. N. Ali, P. H. J. Chong, and Y. L. Guan, “BSM dissemi-
nation with network coded relaying in VANETs at NLOS intersections,”
to v x due to d(v x , v i ) ≤ d(v x , ). Therefore, the most distant in Proc. IEEE Int. Conf. Commun. (ICC), May 2017, pp. 1–6.
vehicle to the RSU in Q L (called  v Ralay
L ) covers the most [11] D. Wu, Y. Zhang, L. Bao, and A. C. Regan, “Location-based crowd-
vehicles in between. On the other hand, assume Qout L denotes sourcing for vehicular communication in hybrid networks,” IEEE Trans.
Intell. Transp. Syst., vol. 14, no. 2, pp. 837–846, Jun. 2013.
the set of left-side vehicles of the RSU but the distance from [12] M. F. Tsai and C. H. Chang, “A reducing broadcast message method
the RSU is larger than r (i.e., not in Q L ). For an arbitrary in vehicular networks,” in Proc. 10th Int. Conf. Intell. Inf. Hiding
vehicle v j ∈ QoutL , d(v ,
j v Ralay ) ≤ d(v j , v x ) where v x ∈ Q
L L Multimedia Signal Process., Aug. 2014, pp. 744–747.
[13] Y. Wu, K. W. Shum, W. S. Wong, and L. Shen, “Safety-message
and v x = v Ralay
L . Therefore, the node  v Ralay
L covers the most broadcast in vehicular ad hoc networks based on protocol sequences,”
vehicles in v j ∈ Qout . Thus, v Relay = 
L L v Ralay is the determined
L IEEE Trans. Veh. Technol., vol. 63, no. 3, pp. 1467–1479, Mar. 2014.
left relay. The conditions for the relays on other sides can be [14] P. K. Sahu, A. Hafid, and S. Cherkaoui, “M-PNC: Multi-hop physical
layer network coding for shared paths in vehicular networks,” in Proc.
proved correspondingly. Wireless Commun. Mobile Comput. Conf., Oct. 2015, pp. 274–279.
[15] E. D. N. Ndih and S. Cherkaoui, “Toward neighborhood prediction
A PPENDIX B using physical-layer network coding,” in Proc. IEEE Int. Conf. Com-
mun. (ICC), Jun. 2012, pp. 676–680.
MAC H EADER D ESIGN [16] E. D. N. Ndih and S. Cherkaoui, “Reliable broadcasting in VANETs
To distinguish different types of packets and keep the using physical-layer network coding,” in Proc. Int. Conf. Commun. Inf.
Technol., Jun. 2012, pp. 363–368.
compatibility with the DSRC, we adopt the same format of
[17] E. D. N. Ndih and S. Cherkaoui, “MAC for physical-layer network
MAC header but assign different values in the FC field, which coding in VANETs,” Int. J. Bus. Data Commun. Netw., vol. 8, no. 4,
is shown by Fig. 11 and Table. II. pp. 84–106, Oct. 2012.
[18] J. Sahoo, E. H.-K. Wu, P. K. Sahu, and M. Gerla, “Congestion-
controlled-coordinator-based MAC for safety-critical message transmis-
A PPENDIX C sion in VANETs,” IEEE Trans. Intell. Transp. Syst., vol. 14, no. 3,
A F LOW C HART OF A PPLIED M ETHODOLOGIES pp. 1423–1437, Sep. 2013.
[19] Y. Kim, M. Lee, and T.-J. Lee, “Coordinated multichannel MAC protocol
The proposed MAC involves a series of methodologies. for vehicular ad hoc networks,” IEEE Trans. Veh. Technol., vol. 65, no. 8,
For the first session and the second session, it applies the pp. 6508–6517, Aug. 2016.
OFDMA [33] and the CSMA/CA [35], respectively. Both of [20] W. Guo, L. Huang, L. Chen, H. Xu, and J. Xie, “An adaptive collision-
free MAC protocol based on TDMA for inter-vehicular communica-
them are widely used in many existing networks, such as LTE tion,” in Proc. Int. Conf. Wireless Commun. Signal Process. (WCSP),
and WiFi. Therefore, we highlight the ones adopted in the PNC Oct. 2012, pp. 1–6.
session. As depicted by Fig. 10, during each uplink, a pair [21] K. Liu, J. K.-Y. Ng, J. Wang, V. C. S. Lee, W. Wu, and S. H. Son,
“Network-coding-assisted data dissemination via cooperative vehicle-
of nodes or relays transmit their BSMs independently and to-vehicle/-infrastructure communications,” IEEE Trans. Intell. Transp.
simultaneously, then the RSU broadcasts a downlink packet Syst., vol. 17, no. 6, pp. 1509–1520, Jun. 2016.
by employing PNC. The non-relay nodes send original BSMs [22] C. Wu, S. Ohzahata, Y. Ji, and T. Kato, “Joint fuzzy relays and network-
only once for uplink, while the relays transmit encoded packets coding-based forwarding for multihop broadcasting in VANETs,” IEEE
Trans. Intell. Transp. Syst., vol. 16, no. 3, pp. 1415–1427, Jun. 2015.
(by using RLNC) for uplink multiple times. [23] C. Wu, X. Chen, Y. Ji, S. Ohzahata, and T. Kato, “Efficient broadcasting
in VANETs using dynamic backbone and network coding,” IEEE Trans.
R EFERENCES Wireless Commun., vol. 14, no. 11, pp. 6057–6071, Nov. 2015.
[24] Y. Bi, H. Shan, X. S. Shen, N. Wang, and H. Zhao, “A multi-
[1] J. B. Kenney, “Dedicated short-range communications (DSRC) standards hop broadcast protocol for emergency message dissemination in urban
in the United States,” Proc. IEEE, vol. 99, no. 7, pp. 1162–1182, vehicular ad hoc networks,” IEEE Trans. Intell. Transp. Syst., vol. 17,
Jul. 2011. no. 3, pp. 736–750, Mar. 2016.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on August 27,2021 at 08:20:46 UTC from IEEE Xplore. Restrictions apply.
 4282 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 21, NO. 10, OCTOBER 2020

[25] O. S. Eyobu, J. Joo, and D. S. Han, “Cooperative multi-channel G. G. Md. Nawaz Ali (S’11–M’16) received the
dissemination of safety messages in VANETs,” in Proc. IEEE Region B.Sc. degree in computer science and engineering
10 Conf. (TENCON), Nov. 2016, pp. 1867–1870. from the Khulna University of Engineering and
[26] T. Ho et al., “A random linear network coding approach to multicast,” Technology, Bangladesh, in 2006, and the Ph.D.
IEEE Trans. Inf. Theory, vol. 52, no. 10, pp. 4413–4430, Oct. 2006. degree in computer science from The City University
[27] N. Chaabouni, A. Hafid, and P. K. Sahu, “A collision-based beacon rate of Hong Kong, Hong Kong, in 2013, with the
adaptation scheme (CBA) for VANETs,” in Proc. IEEE Int. Conf. Adv. Outstanding Academic Performance Award. From
Netw. Telecommun. Syst. (ANTS), Dec. 2013, pp. 1–6. October 2015 to March 2018, he was a Postdoc-
[28] I. W.-H. Ho, S. C. Liew, and L. Lu, “Feasibility study of physical-layer toral Research Fellow with the School of Electrical
network coding in 802.11p VANETs,” in Proc. IEEE Int. Symp. Inf. and Electronic Engineering, Nanyang Technological
Theory, Jun. 2014, pp. 646–650. University (NTU), Singapore. He is currently a
[29] L. F. Xie, I. W.-H. Ho, S. C. Liew, L. Lu, and F. C. M. Lau, “Mitigating Post-Doctoral Fellow with The Clemson University International Center for
Doppler effects on physical-layer network coding in VANET,” in Proc. Automotive Research (CU-ICAR), Department of Automotive Engineering,
IEEE Int. Symp. Person Indoor Mobile Radio Commun. (PIMRC), Greenville, SC, USA. His current research interests include vehicular cyber
Aug. 2014, pp. 121–126. physical system (VCPS), wireless broadcasting, mobile computing, and net-
work coding. He is a reviewer of a number of international journals, including
[30] IEEE Standard for Wireless Access in Vehicular Environments (WAVE)
the IEEE T RANSACTIONS ON I NTELLIGENT T RANSPORTATION S YSTEMS ,
Multi-Channel Operation, IEEE Standard 1609.4, 2011. IEEE Intelligent Transportation Systems Magazine, IEEE T RANSACTIONS
[31] M. Zhang, L. Lu, and S. C. Liew, “An optimal decoding strategy for ON V EHICULAR T ECHNOLOGY , and Wireless Networks.
physical-layer network coding over multipath fading channel,” IEEE
Trans. Veh. Technol., vol. 64, no. 9, pp. 4365–4372, Sep. 2015.
[32] L. Lu, L. You, and S. C. Liew, “Network-coded multiple access,” IEEE
Trans. Mobile Comput., vol. 13, no. 12, pp. 2853–2869, Dec. 2014. Peter Han Joo Chong (M’01–SM’19) received
[33] S. Sen, R. R. Choudhury, and S. Nelakuditi, “No time to countdown: the Ph.D. degree from The University of British
Migrating backoff to the frequency domain,” in Proc. ACM Int. Conf. Columbia, Canada, in 2000. He was an Associate
Mobile Comput. Netw., Sep. 2011, pp. 241–252. Professor (tenured) with the School of Electrical
[34] S. Zhang, S. C. Liew, and P. P. Lam, “Hot topic: Physical-layer and Electronic Engineering, Nanyang Technological
network coding,” in Proc. ACM Int. Conf. Mobile Comput. Netw., 2006, University (NTU), Singapore. From 2011 to 2013,
pp. 358–365. he was an Assistant Head of the Division of Com-
[35] IEEE Standard for Information Technology—Local and Metropolitan munication Engineering. From 2013 to 2016, he was
Area Networks—Specific Requirements—Part 11: Wireless LAN Medium the Director of INFINITUS, Centre for Infocomm
Access Control (MAC) and Physical Layer (PHY) Specifications Amend- Technology. He is currently a Professor and the Head
of the Department of Electrical and Electronic Engi-
ment 6: Wireless Access in Vehicular Environments, IEEE Standard
802.11p-2010, 2011. neering, Auckland University of Technology, Auckland, New Zealand. He is
also an Adjunct Professor with the Department of Information Engineering,
[36] J. Yin, G. Holland, T. ElBatt, F. Bai, and H. Krishnan, “DSRC channel The Chinese University of Hong Kong, Hong Kong. His research interests are
fading analysis from empirical measurement,” in Proc. IEEE 1st Int. in the areas of mobile communications systems, including MANETs/VANETs,
Conf. Commun. Netw. China, Oct. 2006, pp. 1–5. multihop cellular networks, and the Internet of Things/Vehicles.
[37] P. Ferrand, J.-M. Gorce, and C. Goursaud, “Approximations of the packet
error rate under quasi-static fading in direct and relayed links,” EURASIP
J. Wireless Commun. Netw., vol. 1, no. 7, p. 12, Dec. 2015.
[38] Q. Yang, J. Zheng, and L. Shen, “Modeling and performance analysis of Boon-Chong Seet (M’03–SM’11) received the
periodic broadcast in vehicular ad hoc networks,” in Proc. IEEE Glob. Ph.D. degree in computer communication engi-
Telecommun. Conf. (GLOBECOM), Dec. 2011, pp. 1–5. neering from Nanyang Technological University,
[39] Network Simulator Tools NS3. Accessed: Oct. 1, 2017. [Online]. Avail- Singapore, in 2005. Upon graduation, he worked as a
able: https://www.nsnam.org/ Research Fellow under the Singapore–Massachusetts
[40] L. Cheng, B. E. Henty, D. D. Stancil, F. Bai, and P. Mudalige, Institute of Technology Alliance (SMA) Program at
“Mobile vehicle-to-vehicle narrow-band channel measurement and char- the National University of Singapore. Since Decem-
acterization of the 5.9 GHz dedicated short range communication ber 2007, he has been with the Department of
(DSRC) frequency band,” IEEE J. Sel. Areas Commun., vol. 25, no. 8, Electrical and Electronic Engineering, Auckland
pp. 1501–1516, Oct. 2007. University of Technology, New Zealand, where he is
currently an Associate Professor and a Leader of the
Wireless Innovations in Engineering Research Group. His research activities
span the fields of information and communication technologies, including
vehicular communications and networking.

Arun Kumar received the B.Tech. degree in com-

puter science and engineering from the Institute
of Engineering and Rural Technology, Allahabad,
in 2006, the M.Tech. degree in computer science
and engineering from the National Institute of Tech-
nology (NIT), Rourkela, in 2008, and the Ph.D.
degree from the School of Computer Engineering,
Nanyang Technological University (NTU), Singa-
pore, in 2014. He was a Post-Doctoral Research
Minglong Zhang received the B.E. and M.S. Fellow with the Institute of Information Science,
degrees from Jiangnan University and Peking Academia Sinica, Taipei, Taiwan, from 2014 to
University in 2005 and 2011, respectively. He is 2015. He was also a Research Associate with the Infocomm Centre of
currently pursuing the Ph.D. degree with the Depart- Excellence, School of Electrical and Electronics Engineering, NTU. He has
ment of Electrical and Electronic Engineering, been a Research Fellow with the Electrical Machines and Drives Laboratory,
Auckland University of Technology, Auckland, New Department of Electrical and Computer Engineering, National University of
Zealand. From 2011 to 2016, he was a Research Singapore, Singapore, since 2015. He has been an Assistant Professor with
Assistant with the Institute of Network Coding, The the Department of Computer Science and Engineering, NIT, since 2018.
Chinese University of Hong Kong, Hong Kong. His His research interests include wireless sensor networks, ad hoc and mobile
research interests include wireless communication, networks, the Internet of Things/Vehicles, communication algorithms, and
VANETs, and 5G V2X. computer network analysis.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY JALANDAR. Downloaded on August 27,2021 at 08:20:46 UTC from IEEE Xplore. Restrictions apply.