This article has been accepted for publication in IEEE Internet of Things Journal.

This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2022.3195655

IEEE INTERNET OF THINGS JOURNAL, VOL. 14, NO. 8, AUGUST 1, 2022 1

A Novel V2V-based Temporary Warning Network

for Safety Message Dissemination in Urban
Environments
Bingyi Liu, Weizhen Han, Wei Jiang, Dongyao Jia, Enshu Wang, Jianping Wang,
Chunming Qiao, Fellow, IEEE

Abstract—Vehicular communication networks (VCNs) have

been widely recognized as promising solutions to support safety-
related applications in urban transportation systems. However,
T HE rapid development of emerging technologies such
as advanced sensing and vehicular communication net-
works (VCNs) has dramatically enabled the modern intel-
constructing and maintaining such networks is quite challenging
due to the complex traffic and communication environment. ligent transport systems with traffic safety, efficiency, and
Substantial studies have focused on the design of the networking sustainability [1]. With the help of vehicular communication,
schemes and message dissemination protocols. Nonetheless, most Intelligent Vehicles (IVs) on the road can acquire timely
existing designs only consider the connectivity and rapid end- traffic information such as accident warnings, road situations,
to-end transmission, regardless of the network coverage and and route suggestions, and accordingly, coordinate to form a
duration. In this paper, we propose a novel temporary warning
network (TWN) for safety message dissemination in the urban swarm intelligent networking system to avoid potential traffic
traffic environment, in which both the spatial distribution and accidents [2]–[5]. Compared to passive sensing (e.g., using
temporal duration of the networking scheme are taken into LiDAR), the vehicular network can expand vehicles’ situation
account. Specifically, TWN is constructed by the selection of awareness of the potential risks beyond the visual range [6]. In
relay vehicles based on the spatio-temporal correlation of vehicle such a network system, it is critical to design networking and
trajectory so that the safety message can be quickly disseminated
within the regions of interest (RoI). To maintain TWN during communication schemes that ensure the freshness of informa-
an accident, a re-selection mechanism is also proposed, which tion and the coverage and timeliness of data dissemination.
enables newly-come vehicles in the RoI to receive the messages VCNs can be mainly categorized into infrastructure-based
in time. Finally, we conduct extensive numerical experiments and infrastructure-free schemes. As a typical infrastructure-
to validate the effectiveness of our method in various traffic based pattern, edge computing-enabled vehicular networks can
scenarios.
take advantage of cloud computing and IT services at the
Index Terms—vehicular communication networks, temporary edge of cellular networks, significantly improving network
warning network, safety message dissemination, vehicle trajec- performance and providing vehicles with more computing and
tory, broadcast coverage and duration, region of interest
storage resources. For example, 5G can achieve an ultra-low
latency of 1 ms for the air interface with the guarantee of
I. I NTRODUCTION 5 ms end-to-end latency for infrastructure mode [7]. Another
typical infrastructure-based network pattern is the Cooperative
Vehicle-Infrastructure System, a key technology for realizing
Manuscript received May 26, 2022; revised June 24, 2022; accepted July environmental awareness, information exchange, and collab-
25, 2022. This work was in part supported by National Natural Science Foun- orative control in vehicular networking [8]. Although many
dation of China (No. 61802288, 62176194), Key Research and Development
Program of Hainan (No. ZDYF2021GXJS014), Hong Kong Research Grant benefits can be obtained from infrastructure-based models,
Council under NSFC/RGC joint project N CityU140/20, Science Founda- some limitations cannot be neglected in terms of system de-
tion of Chongqing of China (cstc2021jcyj-msxm4262), Research Project of ployment. For example, infrastructure deployment and mainte-
Chongqing Research Institute, Wuhan University of Technology (ZD2021-04,
ZL2021-05), and Research Development Fund of XJTLU (RDF-21-02-082). nance are quite time-consuming and even impossible in some
(Corresponding author: Dongyao Jia) not actively growing markets. In particular, for a typical urban
Bingyi Liu is with the School of Computer Science and Artificial In- environment, there probably exist a large number of blind spots
telligence, Wuhan University of Technology, Wuhan, China, Sanya Science
and Education Innovation Park of Wuhan University of Technology, Sanya, where the vehicles are unable to connect to infrastructures.
China, and Chongqing Research Institute, Wuhan University of Technology, The V2V-based networking mode, on the other hand, makes
Chongqing, China. it possible to independently construct VCNs and realize IVs
Weizhen Han and Wei Jiang are with the School of Computer Science and
Artificial Intelligence, Wuhan University of Technology, Wuhan, China. data storage and processing at three levels, i.e., cloud, fog,
Dongyao Jia is with the School of Advanced Technology, Xi’an Jiaotong- and onboard [9]. With V2V-based communication, VCNs can
Liverpool University, China. flexibly arrange their storage and processing and better exploit
Enshu Wang and Chunming Qiao are with the University at Buffalo, The
State University of New York, Buffalo, NY, USA. the availability of multi-IVs data to facilitate various learn-
Jianping Wang is with the Department of Computer Science, City University ing in different vehicular applications more effectively and
of Hong Kong, Kowloon, Hong Kong. efficiently [7]. It is widely recognized that VCNs can extend
Copyright (c) 20xx IEEE. Personal use of this material is permitted.
However, permission to use this material for any other purposes must be the perception range of autonomous drivings, which depends
obtained from the IEEE by sending a request to [email protected]. on the content of the information itself obtained from inter-

© 2022 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: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on October 26,2022 at 09:08:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Internet of Things Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2022.3195655

IEEE INTERNET OF THINGS JOURNAL, VOL. 14, NO. 8, AUGUST 1, 2022 2

prediction of each vehicle. The proposed scheme can guarantee

5HJLRQ of ,QWHUHVW (RRI)
the efficiency of message dissemination and improve the
6FKHGXOHG 3ath
message coverage and duration in the RoI. To the best of our
KRVSLWDO
2UGLQDU\ 9ehicles 29V 
about to (nter the RRI
knowledge, this is the first attempt that the networking and
29V in the RRI
safety message dissemination scheme are integrally designed
DPEXODQFH
with the consideration of both coverage and duration of safety
V2V &RPPXQLFDWLRQ
messages in RoI. The main contributions of this paper are as
6afety 0essage
follows:
• We propose a novel temporary warning network (TWN)
Fig. 1. Examples of safety message dissemination. for safety message dissemination in urban traffic environ-
ments, in which both the spatial distribution and temporal
duration of the safety message are taken into account.
vehicle communications such as real-time maps and no longer • The TWN is constructed and maintained by the selection
pays attention to their sources. Thus the end-to-end mode of relay vehicles based on the spatio-temporal correlation
in traditional Vehicular Ad-hoc Networks (VANETs) will not of vehicle trajectory so that the safety message can be
be able to meet the needs of the information dissemination quickly and continuously disseminated within the RoI.
[10]. Accordingly, Information-Centric Networking (ICN) ar- • The performance of the proposed method is evaluated
chitecture in VANETs addresses data directly by changing the under extensive traffic scenarios with different vehicular
addressing scheme, rather than using the location of the data networking settings (e.g., vehicle density, RoI size, the
[11]. proportion of different types of vehicles) and compared
Although there are considerable V2V-based networking with existing methods. The numerical results confirm
schemes for message dissemination in VCNs, most of which the efficiency of our proposed design on safety message
mainly focus on the connectivity and rapid end-to-end trans- dissemination in RoI.
mission, disregarding the significance of the spatial coverage
and temporal duration of the networking scheme. For instance, II. R ELATED W ORK
as shown in Fig. 1, in order to avoid the congestion caused
by the accident or the secondary accident due to insufficient Efficient vehicular networking and message dissemination
observation of the traffic accident, it is expected that the in VCNs are critical for the provision of safety-related ser-
network for accident warning message dissemination should vices, and extensive studies have been devoted to devising
not only cover the vehicles in all involved road segments networking schemes and message dissemination solutions to
(i.e., the red part shown in Fig. 1) as soon as possible, but improve the delivery ratio and minimize the delivery overhead
also be active before the accident clearance. Another example and delay. They can be divided into two categories according
is when an emergency vehicle (e.g., an ambulance in the to whether additional communication devices are needed: V2I-
yellow road in Fig. 1) is on its urgent mission, it would based networking schemes that rely on RSU, base stations
inform the vehicles on its planned path in advance in order (BSs), or other devices, and V2V-based networking schemes
to reserve the way. This message dissemination is supposed that only use on-board communication devices.
to cover all the relevant road segments and vehicles until
the emergence vehicle reaches the destination. Therefore, it A. V2I-based Vehicular Networking and Message Dissemina-
is important to take the networking coverage and duration tion
metrics into account simultaneously. Traditional V2I-based networking schemes and message
Such a task implies several challenges to the vehicular dissemination methods usually utilize RSU for forwarder
networking scheme. Firstly, given the complexity of road selection and message dissemination. [12] proposes a multi-
and traffic conditions in urban environments, it is extremely hop clustering approach based on V2I communication where
tough to design an adaptive networking scheme for the safety each vehicle can capture and convey the state information
message dissemination to all vehicles and road segments in the about its multi-hop neighbors for the forwarder selection and
RoI. Unreasonable multi-hop transmission links significantly networking process. [13] designs a message coverage maxi-
increase the end-to-end delay. Besides, the complex dynamics mization algorithm (MCMA) that leverages both vehicles and
(e.g., high dynamics of vehicle mobility and network topology) carefully deployed RSUs for message dissemination in urban
in real-world urban environments always cause link interrup- environments to achieve the maximum message coverage.
tions. Hence, the proposed networking scheme needs to detect Some V2I-based communication schemes also consider
the network status and adjust the dissemination strategies to the spatio-temporal features of message dissemination. For
ensure the duration of communication coverage. instance, [14] proposes a spatio-temporal coordination-based
In this paper, we investigate the vehicular networking media access control (STMAC) protocol that adopts a
scheme for message dissemination in urban environments, in contention-free channel access scheme to share safety mes-
which the coverage and duration of networking are fully taken sages efficiently. In addition, [15] formulates a temporal data
into account. To this end, we propose a novel temporary transmission problem in the vehicular network. Specifically,
warning network (TWN) that selects appropriate vehicles with the help of V2I communication, vehicles can subscribe
as core nodes by taking advantage of the future trajectory to specific services when entering the RSU’s service range.

© 2022 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: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on October 26,2022 at 09:08:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Internet of Things Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2022.3195655

IEEE INTERNET OF THINGS JOURNAL, VOL. 14, NO. 8, AUGUST 1, 2022 3

In particular, the rapid development of 5G, mobile edge and [28], respectively. And [29] proposed a distance-based
computing (MEC) technologies has brought new solutions to relay node selection method at intersections under LTE-V2V
networking, and message dissemination [16]. 5G cellular net- communication mode.
works provide high-capacity and low-latency communications Besides, some studies have combined mathematical theory
for vehicles in highly mobile environments. [17] presents a to achieve good performance. [30] presents a distributed algo-
cluster-based broadcast algorithm in a 5G-VANET system, rithm in which the broadcast network is a clustered structure.
where the control messages are disseminated by cluster head in Nevertheless, the recognition and transition of vehicles (such
a cluster with dedicated short-range communications (DSRC), as cluster heads or gateways) need massive beacons, leading
and vehicular networking is formed by the base stations to channel congestion. In [31], and [32], non-cooperative
(BSs) and vehicles based on the software-defined 5G-VANET forwarding schemes are proposed with game theory through
architecture. This method utilizes historical data to model the the volunteer dilemma approach. [33] studies the temporal
socialites of vehicles and reduces the communication overhead topology of the vehicular network and reveals an invariant law
caused by frequent changes in network topology. However, binding the performance of store-carry-and-forward to proper-
since the cover range of a 5G BS is even smaller than 4G BS ties of the road traffic and road layout. It can provide valuable
or RSU, the blind area is much more and larger, which will guidelines for assessing and predicting the performance of
bring a loss to the timeliness of messages. An edge computing vehicular networks.
architecture is proposed in [18], which is composed of two Although the aforementioned methods are essential to sup-
layers—BS and autonomous vehicles. Each BS is equipped port reliable and efficient message dissemination, most of
with an edge server to cache contents (e.g., HD map, traffic them focus on the end-to-end mode and network connectivity,
condition, news) for nearby vehicles. [19] presents a message which cannot exactly meet the requirement of some typical
dissemination scheduling architecture that supports both the vehicular applications that are mainly associated with the
logically centralized control via the cloud node in the core information within specific areas, such as road conditions and
network and the distributed data dissemination via the fog real-time traffic path recommendations. Besides, most of the
nodes at the network edge. However, updating the contents existing safety message dissemination schemes only consider
on the edge nodes requires much communication overhead. spatial or temporal correlation among vehicles to select relays,
disregarding the safety message coverage and the forwarding
path maintenance at the same time. Moreover, Some recently
B. V2V-based Vehicular Networking and Message Dissemina- influential work has contributed to the construction and main-
tion tenance phase for improving clustering lifetime and reducing
Instead of depending on additional roadside infrastructures, communication overhead. However, many of them ignore the
V2V-based networking schemes select vehicles on roads to performance from a network construction perspective in a
establish a network for message transmission [20]. However, target area that is of great importance for safety message
ensuring network connectivity and stability is a challenge due dissemination in urban environments, especially in limited
to the high mobility of vehicles. To address this issue, [21] infrastructure deployment scenarios.
uses interest compatibility as a metric to select cluster heads
and proposes a location-based emergency message distribution III. S YSTEM M ODEL
method to reduce communication delay. A mobility-based This section demonstrates in detail our system model with
clustering algorithm is proposed in [22] with effective mul- main assumptions and specifications in terms of the traffic
timedia broadcasts in a fog computing network, which selects scenario and communication scheme, respectively. To facili-
cluster heads with similar average relative speeds to ensure tate further discussions, we first summarize some important
the stability of clusters. [23] constructs a stable hierarchical symbols and notations in Table I.
network to achieve effective information dissemination via
a backbone network. Based on the backbone network, [24] TABLE I
further utilizes mobility metrics such as movement direction, N OTATIONS
vehicle density, relative speed, and distance to implement dy-
Notation Description
namic mobility and stability-based clustering scheme, in which
the link lifetime estimation is performed by the movement TWN Temporary warning network
RoI Region of interest
direction and relative position of vehicles. [25] uses realistic AA Active area
vehicle trajectories to investigate the effect of node mobility IBV Initial broadcast vehicle
on the connectivity of clustering networks and proposes a CV Core vehicle
OV Ordinary vehicle
dynamic evolutionary network model to maintain the connec- BSM Broadcast starting message
tivity of clustering networks. ELT Expected leaving time
Some studies also consider more complex urban scenarios. vi Vehicle No. i
ri Road segment No. i
In [26], the authors propose a protocol named urban multi-hop (t)
pi Position probability vector
broadcast protocol (UMBP), which adopts flexible broadcast v
RAA AA of vehicle v
strategies according to the positions of the forwarding nodes. δv Selection factor of vehicle v
To reduce the delay of message dissemination, dynamics δth Selection factor threshold
prediction, and concurrent transmission are adopted in [27],

© 2022 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: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on October 26,2022 at 09:08:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Internet of Things Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2022.3195655

IEEE INTERNET OF THINGS JOURNAL, VOL. 14, NO. 8, AUGUST 1, 2022 4

A. Traffic Scenario and Definitions

We model the urban environment with the Manhattan grid,
a typical urban road topology that has been widely adopted in
the literature. It should be pointed out that this Manhattan grid
can be easily extended to other complex urban environments.
Our proposed scheme aims at safety message dissemination
in urban traffic scenarios. A general message dissemination
process can be described as follows. When a vehicle detects
an accident or an emergency vehicle (e.g., ambulance, fire
engines, etc.) requests a road reservation, the vehicle first Initial Broadcast
Core Vehicle (CV)
broadcasts a safety message which includes the information of Vehicle (IBV)

the emergency event, such as the time and place of occurrence, Ordinary Vehicle (OV, partial)

the estimated duration, and the road segments affected by Region of Interest
Active Area (AA, partial)
(RoI)
the event. According to this information, the relay vehicles
are then selected to construct a dynamic temporary warning Inter-AA Links Intra-AA Links (partial)

network for the safety message dissemination in the RoI.

Such a message dissemination process can be modeled as Fig. 2. Tree-like structure of the TWN.
follows. First, we abstract the traffic scenario into two sets,
a set of all Road segments as R = {r1 , r2 , . . . , rM } and
a set of all Vehicles as V = {v1 , v2 , . . . , vN }, where M the IBV. To avoid serious packets collision caused by simulta-
and N represent the number of roads and that of vehicles, neous transmissions from different neighbors, the packets are
respectively. The communication links between the vehicles sent with different delays according to the similarity between
can be represented as an undirected graph G = (V, E), where the AAs and the RoI. Finally, notification of selection results
E is the set of edges that represent single-hop links between is transmitted to the selected CVs. The vehicles not selected
the vehicles. The active area (AA) of a vehicle is defined as CVs become ordinary vehicles (OVs). By the three-step
as the set of roads where the vehicle has a high probability connection procedure, a CV selects one or several other CVs
of passing or staying in the future. The vehicles involved in its AA to establish a tree-like TWN until the TWN covers
in message dissemination are divided into two categories the RoI, as shown in Fig. 2.
according to whether they have planned paths or not: the path- However, the intermittent interruption within the network
fixed vehicles with exact routes or destinations (e.g., buses) should be considered due to the complex traffic and road
and the cruise vehicles without precise destinations (e.g., taxis conditions. Moreover, the safety message is not available for
without passengers). The planned routes are regarded as their the vehicles that are not in the RoI at present but are about to
AAs for the path-fixed vehicles, while for cruise vehicles, enter. Therefore, the maintenance of the network is necessary
AAs are determined based on continuously-updated vehicle for the coverage and duration of safety message dissemination.
trajectory data. The vehicles selected as relay nodes are called In our proposed scheme, each CV should maintain a list
core vehicles (CVs). The set of road segments that may be of expected leaving time (ELT) for its subordinates. If a
affected by the emergency is called the region of interest (RoI). subordinate leaves the AA of the superior, the superior selects
RoI is determined by the safety message types and contents, an alternative CV. The re-selection of CVs will be elaborated
which can be redefined in different scenarios. As shown in in Sec. V. Meanwhile, the selected CVs continue to broadcast
Fig. 1, the RoI for the ambulance (the yellow region) only the safety message in their AAs within the lifetime of the
includes the path of the ambulance to the hospital, while the message to ensure the duration of message coverage for new-
RoI for the accident vehicles (the red region) includes road comers.
segments adjacent to the place of the accident.
IV. P RE -C LUSTERING -BASED V EHICLE P OSITION
B. Communication Model P REDICTION M ETHOD
The general communication process under the proposed Since this paper utilizes a distributed message dissemina-
scheme is as follows. In the beginning, all vehicles are tion scheme, obtaining the real-time status from the whole
aware of nearby potential emergency events, and a vehicle vehicular network is an extremely tough task for a vehicle in
becomes the initial broadcast vehicle (IBV) once it detects the urban scenario, and many communication resources are
an emergency. The safety message dissemination process is consumed. Therefore, it is necessary to analyze the network
then triggered by the IBV, and CVs are selected by IBV to topology and link stability characteristics to take pre-clustering
establish a TWN for safety message dissemination and update operations. Meanwhile, the trajectory’s spatio-temporal and
in the RoI. the AA of the vehicle can effectively improve the timeliness
The CV selection process is achieved via a three-step con- and coverage of safety message dissemination and, at the
nection procedure. At first, the IBV broadcasts a CV selection same time, significantly improve the stability of the network
request to its one-hop neighbors. A vehicle receiving the BSM topology, enhancing the duration of message coverage, and
will transmit kinetic information (e.g., position and velocity) to reduce overhead.

© 2022 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: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on October 26,2022 at 09:08:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Internet of Things Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2022.3195655

IEEE INTERNET OF THINGS JOURNAL, VOL. 14, NO. 8, AUGUST 1, 2022 5

A. Pre-clustering for Vehicular Network Characteristics Ex- 2) Let K be the number of zeros among these eigenvalues,
traction K is exactly the number of connected components in
In a V2V-based distributed network, it is difficult for a single the graph.
vehicle to obtain the global state of the whole traffic network. 3) For any f = (f1 , f2 , . . . , fN )′ , we have f ′ Lf =
1
P N 2
To select proper parameters for the TWN establishment in real 2 i,j=1 wij (fi − fj ) .
traffic situations, in this paper, we take advantage of historical Employ eigen decomposition to the Laplacian matrix L,
vehicular trajectory data and analyze the topology and link and let λ1 , λ2 , . . . , λN denote the eigenvalues. According to
quality between vehicles in a macroscopic view. the property 2) of the Laplacian matrix, K is the number of
The links between vehicles in our framework are divided connected components in graph G. Two nodes within the same
into two categories: the inter-AA links between CVs and the connected component can be connect via a multi-hop link.
intra-AA links between a CV and an OV. The inter-AA links To get the result of pre-clustering, we define
are used to establish the TWN in which the safety messages K
are transmitted and updated, while the intra-AA links are used 1 X W(Ci , Ci )
NCut(C1 , C2 , . . . , CK ) = (5)
for the periodic broadcast of safety messages within an AA. 2 i=1 vol(Ci )
Hence the inter-AA links need to be more stable to reduce P
the maintenance overhead of the TWN structure. In order to where Ci = G − Ci , W(Ci , Ci ) = j∈Ci ,k∈Ci wjk , vol(Ci )
increase the inter-AA link stability, we cluster the network of denotes the number of edges in cluster Ci . To get the mini-
vehicles at different time based on historical data in advance mum sum of weights of all edges between clusters (i.e., the
to obtain the optimal cluster size from a global perspective, maximum similarity within each cluster) and proper cluster
and the average size of the area covered by the clusters is size, our goal is to minimize the NCut.
used to determine the size of the AAs. In addition, the cluster For a subset Cj ⊂ V , we define f j = (f1j , f2j , . . . , fN j )′ ,
size can also reflect the link stability of the entire network, where
and the impact of parameters (e.g., the area similarity and ( p
1/ vol(Ci ), vi ∈ Cj ,
position) in networking can be determined based on the results fij = (6)
of clustering in different situations. 0, vi ∈ Cj .
As mentioned in the previous section, (i, j) ∈ E if the two And f j (j = 1, 2, . . . , K) form a matrix F . As the aforemen-
vehicles vi and vj are within the direct transmission range. tioned property 3), we have
We define the link stability of two adjacent vehicle nodes as
N
min (υi , υj ) 1 X
LS(i, j) = · cos (⟨υ i , υ j ⟩/2) (1) f ′j Lf j = wmn (fmj − fnj )2
max (υi , υj ) 2 m,n=1
 q 2
where υi , υj represent the speed of the two vehicles, and 1 X
⟨υ i , υ j ⟩ denotes the angle between the velocity of them. The = wmn 1/ vol(Cj ) − 0
2
definition of link stability takes into account both the speed m∈Cj ,n∈Cj
 2 (7)
and direction of vehicles on the link stability, namely, the link 1 X q
between the two vehicles with similar speeds and directions is + wmn 0 − 1/ vol(Cj )
2
more stable. Then we get the weighted adjacency matrix W m∈Cj ,n∈Cj

of the network with elements as 1 W(Cj , Cj )

= ·
( 2 NCut(Cj , Cj )
LS(i, j), (i, j) ∈ E,
wij = (2) Compare Eq. (5) with Eq. (7), we get
0, (i, j) ∈
/ E.
K
The degree matrix D represents the degree of the nodes, and X
it has the elements with NCut(C1 , C2 , . . . , CK ) = f ′j Lf j = tr(F ′ LF ) (8)
j=1
N
 X And it is subject to

 wik , i = j,
dij = (3) N
k=1 X 1 X

 k̸=i f ′j Df j = 2
fij dii = di,i = 1 (9)

0, i ̸= j. i=1
vol(Cj )
i∈Cj

Then the Laplacian matrix is calculated by Let F = D −1/2 H, we can get

L=D−W (4) F ′ LF = H ′ D −1/2 LD −1/2 F (10)

′ ′
The Laplacian matrix has the following properties F DF = H H = I (11)
1) It has N non-negative eigenvalues, and the minimum of Accordingly, we only need to find out the smallest non-
them is zero; zero eigenvalues of the matrix D −1/2 LD −1/2 , then employ

© 2022 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: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on October 26,2022 at 09:08:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Internet of Things Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2022.3195655

IEEE INTERNET OF THINGS JOURNAL, VOL. 14, NO. 8, AUGUST 1, 2022 6

normalization and K-means clustering to the matrix formed System (GPS), electronic map and on-board sensors. The
by the eigenvectors to get the result of clustering. The road vehicles collect their own position data at intervals and store
segments covered by vehicles within the same cluster are in the form of {time, road id}. Its state transition probability
labeled with the same cluster number, and the clustering results and distribution function of sojourn time are updated regularly
are stored in the electronic map in the form of {time, road id, according to Eq. (12) and Eq. (13).
cluster number}. 2) Position Probability Matrix: The vehicle either transfers
It shall be noted that the clustering results will not be its state or sojourns in a period in the future, thus we discuss
updated frequently once it is calculated by the cloud. This is the probability of the vehicle’s position in the future from these
because the trajectory characteristics of vehicles hardly change two aspects respectively. We define the position probability at
significantly within a short period of time. The pre-clustering a time slot t in the future as
results are distributed to the vehicles by the cloud.
(t)
pij = Pr (Xt = j|X0 = i) (14)
B. Semi-Markov Model Based Vehicle Position Prediction
Method There may be different paths for the state transition of a
vehicle (e.g., if a vehicle transfers from road section 1 to
Achieving fast propagation in the RoI is a major purpose in 4, there may exist paths as 1→2→4, 1→3→4, 1→2→1→4,
safety message dissemination. According to [34], the trajectory etc.), so the state transition probability in Eq. (12) cannot be
of most vehicles is rather stationary with time. Namely, a directly used as the position probability. We adopt an iterative
vehicle only visits a few particular road segments at a certain approach to calculate the position probability of the vehicle
time, which is called spatio-temporal correlation. As described considering different transition paths. Assume that the vehicle
in Sec. III, the set of these particular road segments is the is located on road segment ri , then the initial condition of the
AA of a vehicle. Therefore, if several vehicles are selected as position probability is
relays so that their AAs cover the RoI, the performance of
broadcast propagation will be significantly improved. (0)
pij = pij , j = 1, 2, . . . , M (15)
In this section, a vehicle trajectory prediction method is
(t)
proposed. The motion of each vehicle is modeled as a semi- pij in Eq. (14) can be calculated as follows:
Markov process, in which the position transition probability (1) If the vehicle has been sojourning on ri until t, we have
and sojourn time distribution function are evaluated from
M M
historical trajectory data to calculate the position probability. (t)
X X
The positions where a vehicle will appear with the highest pii = Pr (τij > t) = 1 − Fij (t) (16)
j=1 j=1
probability in the future are the vehicle’s trajectory prediction j̸=i j̸=i
of the vehicle.
(2) If the vehicle transfers from ri to rj before t, we have
1) Processing of Historical Trajectory Data: We denote the
position of a vehicle as its state. Assume that a vehicle is t
M X
(t) (t−z)
X
located on the road segment ri at time slot t, thus its state pij = Fik (z) · pik · pkj (17)
Xt = i. For the situation of Xt = i and Xt+1 = j, we can k=1 z=0
infer that the vehicle transfers from ri to rj at time slot t + 1, The term indicates that the state transition process can be
which is called state transition; on the contrary, Xt+1 = Xt = divided into three phases: i) sojourning on ri before time slot
i means there is no state transition happens to the vehicle, z; ii) transferring to rk at time slot z; iii) transferring to rj
which is called sojourn. The time slot is divided small enough after time slot z. The transition process from rk to rj can also
that the vehicle performs at most one state transition in one be divided into the three phases. The vehicle may not transfer
single time slot. directly from rk to rj between time slots z to t, so an iterative
Based on the vehicle trajectory data, the probability of a (t−z)
calculation is also performed to pkj . The time complexity
state transition from ri to rj can be calculated. We first define of Algorithm. 1 depends on the number of road segments on
the state transition probability as the map. Thus, it may be slightly high in a complex and large-
size road network. However, the computation of this algorithm
pij = Pr (Xt+1 = j|Xt = i) = lij /li (12) does not affect the performance of our networking scheme
where lij indicates the number of times in the historical since the process is completed in the cloud before our network
data that the vehicle transferred from ri to rj ; li indicates construction.
the number of times transfers from ri to all possible road Hence, the position probability of a vehicle can be obtained
segments; t is a random time slot. Then the distribution based on the state transition probability and the sojourn time
function of sojourn time on segment ri before the vehicle distribution function from the historical trajectory data. Let
transferred from ri to rj is defined as X0 = i, and the position probability vector can be calculated
according to Eq. (16) and Eq. (17) as
t−1
X
Fij (t) = Pr (τij < t) = Pr (τij = u) (13)
 
(t) (t) (t) (t)
pi = pi1 , pi2 , . . . , piM (18)
u=0
where τij denotes the sojourn time. All vehicles are assumed to Based on the pre-clustering results in Sec. IV-A, we can get
be equipped with communication devices, Global Positioning the optimal clustering size nC (t) at time t. The road segments

© 2022 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: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on October 26,2022 at 09:08:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Internet of Things Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2022.3195655

IEEE INTERNET OF THINGS JOURNAL, VOL. 14, NO. 8, AUGUST 1, 2022 7

Algorithm 1: Calculation of position probability

Input: the state transition probability pij , the
distribution function of sojourn time Fij (t)
Output: the position probability p(t)
1 for i ← 1 to M do
2 for j ← 1 to M do
3 UpdateFlag ← false ;
4 end
5 end
6 do
7 for i ← 1 to M do
8 for j ← 1 to M do
9 if i == j then
(t)
10 calculate the pij by Equation.16 ;
(t)
11 if pij does not converge then
12 UpdateFlag ← true ; Fig. 3. The definition of average relative position (ARP).
13 end
14 else
v
15 for i ← 1 to M do where RAA represents the AA of the vehicle v, RRoI repre-
16 for j ← 1 to M do sents the RoI, and 0 ≤ d ≤ 1. Then we define the average
(t) relative position (ARP) of the AA as
17 calculate the pij by Equation.17
(t)
18 if pij does not converge then 1 X
ARP v = ℓ (20)
19 UpdateFlag ← true ; ∥LvAA ∥
ℓ∈Lv
AA
20 end
21 end where LvAA represents the displacement vector set of a
22 end candidate vehicle v. The start point of a displacement vector
23 end is the position of the superior CV, and the endpoint is the
24 end center of a road segment in the AA. As shown in Fig. 3, for
25 end instance, the AA of vi includes three road segments (in blue
26 while not UpdateFlag; shade) with related position li1 , li2 and li3 respectively (black
dotted arrow), thus the ARP of vi equals to 1/3(li1 + li2 + li3 )
2) Selection of the CVs: In our framework, some vehicles
(t)
are selected as CVs to establish a TWN for message dissem-
corresponding to the largest nC (t) elements in pi are the ination and updating. In addition, when a CV leaves the RoI,
positions where the vehicle will most likely appear at time another vehicle needs to be selected as an alternative CV. The
slot t, namely the AA of the vehicle. CV selection process is achieved with three steps.
1) The IBV becomes the first CV after it detects the emer-
V. TWN BASED S AFETY M ESSAGE D ISSEMINATION gency and sends BSM to all its neighboring vehicles.
The velocity, position and AA of the sender are included
In this section, we propose a TWN-based Safety Message in the BSM.
Dissemination scheme (TWNSMD) which takes advantage of 2) The neighboring vehicles calculate the similarity and
the characteristics of traffic networks and vehicle trajectory. ARP according to Eq. (19) and Eq. (20) and send a
Briefly, by analyzing the characteristics of vehicle trajectory, feedback message to the superior CV after an interval
CVs are selected to establish a TWN for message dissemina- of T v = 1/dv , then they set a waiting timer for receiving
tion, while the network structure of TWN is determined by the notification.
the traffic network characteristics. In addition, the structure 3) The first CV selects one or several vehicles in its one-
maintenance of the TWN is designed to ensure the duration hop neighbors as subordinate CVs and then sends a
of the safety messages within the RoI. notification along with the safety message to them.
The three-step connection procedure terminates until the
entire RoI is covered by the CVs. In order to reduce the
A. Selection of CVs and Establishment of TWN CVs with overlapping AAs, the AA of the superior CV is
1) Area Similarity and Average Relative Position: We de- removed from the RoI when calculating the similarity. The
fine the similarity between an active area and the region of parameters transmitted by the candidate subordinate vehicles
interest as: include similarity, position, velocity, ARP and the ELT from
the AA of superior CV. In consideration of coverage of the
v
∥RAA ∩ RRoI ∥ safety message, the superior CV decides how many and which
dv = (19) candidates are selected as subordinate CVs.
∥RRoI ∥

© 2022 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: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on October 26,2022 at 09:08:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Internet of Things Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2022.3195655

IEEE INTERNET OF THINGS JOURNAL, VOL. 14, NO. 8, AUGUST 1, 2022 8

The superior CV sets a timer to wait for parameter packets Algorithm 2: Subordinate CVs selection of a CV vi
from candidate subordinate vehicles. It receives the parameter Input: Basic Safety Message, relative information
packets from all subordinate vehicles within the duration, and (e.g., content, location, area of interest, validity
selects the vehicle whose parameter packet is received at period) and status information of the vehicle
the earliest time as the subordinate CV. In this way, safety (e.g., speed, direction, active area).
messages can be forwarded quickly among CVs. The packets Output: The collection of selected CVs Vselc .
out of the duration will be discarded. For other received pa- 1 Vselc = ∅;
rameters packets, the superior CV first calculates the selection 2 for any vehicle vj in the neighbor list of vi do
factor of the corresponding candidate vehicle as 3 send BSM to vj ;
4 end
δ v = α · dv + β · β ′ · LS(sup, v) 5 set a timer Tinfo ;
D E.  (21) 6 while Tinfo does not expire do
+ γ · sin ARP v , ARP selc 2 7 if vi receives an information message from vj then
where dv denotes the similarity calculated by Eq. (19); sup 8 calculate δcand by Eq. 21;
and v denote the superior CV and a candidate subordinate 9 if ∥Vselc ∥ = 0 or δcand > δth then
vehicle. β ′ is the cluster factor, which equals to 1 when the 10 remove AAj from RoI;
vsup and vcand are on the roads with the same cluster number, 11 send the notification and safety message to
and equals to 0.5 otherwise. ARP selc represents the average vj ;
relative position between all selected CVs and the superior 12 set a timer TACK ;
CV. α, β, γ are weight coefficients satisfying α + β + γ = 1. 13 while TACK does not expire do
The first term in Eq. (21) measures the similarity (i.e., 14 if vi receives ACK message from vj
the degree of overlapping) between the AA of the candidate then
vehicle and the RoI; the second term represents the link 15 terminate TACK ;
stability; and the third term indicates the direction between the 16 Vselc ← Vselc ∪ {vj };
AA of the candidate vehicle and the active area of existing 17 end
CVs, which enables the coverage of message dissemination 18 end
and avoids repetitive selection of CVs in the same region. 19 end
With regard to different average cluster size of pre- 20 end
21 end
clustering results in Sec. IV-A, these coefficients can be set to
different values. The smaller clustering size leads to a smaller
AA, which makes the similarity in a low level by Eq. (19).
On the other hand, the dissemination direction can be chosen period. In this section, the CV re-selection mechanism is
more flexibly. As a result, α needs to be increased accordingly, proposed to solve the problem of link interruption of TWN
while γ can be slightly reduced. to ensure the duration and freshness of the safety message.
If δ v is greater than the given threshold δth , the vehicle is In an established TWN, a CV sets several timers of ELT
selected as the subordinate CV and added to the selected CVs of all subordinate CVs. The CV transmits a query message
list of the superior CV to the corresponding vehicle after a timer expires, and a link
interruption is assumed to happen if the CV does not receive
Vselc = {v ∈ V |δ v > δth } (22) a response. As mentioned in Sec. V-A2, ELT is included in
the packets transmitted by the candidate subordinate vehicles.
3) Establishment of the TWN: If a vehicle is notified that ELT of a vehicle will be set to the expiration time of the
it is selected as a CV by the superior, its AA is removed safety message if the vehicle stays in the RoI until the message
from the RoI to reduce the CVs with overlapping AA so that expires; otherwise, ELT will be set to the time that the vehicle
the channel utilization can be improved. Then the three-step leaves the AA of its superior CV.
connection procedure is repeated to finish the TWN structure The re-selection mechanism is initiated by a superior CV
until all the road segments or vehicles in the initial RoI are once the link between it and its subordinate node interrupts.
covered. The TWN has a tree-like structure in a macroscopic The BSM with a new RoI from which the AAs of existing CVs
view, as shown in Fig. 2. If the vehicle has not received the are removed will be transmitted to the one-hop neighbors of
notification until the waiting timer expired, it is set as an OV. the superior CV, and the subordinate CV will be re-selected by
We set a timeout mechanism in both Algorithm 2 and 3. Thus, the three-step connection procedure. Other CVs without link
the computation time of the two algorithms mainly depends interruption will ignore the BSM.
on the transmission delay between the vehicles.
C. Safety Message Dissemination
B. Maintenance of TWN Structure There are two phases involved in the safety message dis-
An emergency event usually lasts for a period until it has semination. Firstly, the safety message is disseminated from
been resolved. Thus, the safety message is required to cover the IBV to all CVs, which is completed in the process of
the RoI and can be updated efficiently during the emergency TWN establishment as mentioned above. To guarantee the

© 2022 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: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on October 26,2022 at 09:08:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Internet of Things Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2022.3195655

IEEE INTERNET OF THINGS JOURNAL, VOL. 14, NO. 8, AUGUST 1, 2022 9

Algorithm 3: Response from the subordinate candidate

vehicles
Input: Basic Safety Message, relative information
(e.g., content, location, area of interest, validity
period) and status information of the vehicle
(e.g., speed, direction, active area).
Output: Vehicle’s role and its superior CV.
(a) (b)
1 received a BSM from superior CV vi ;
2 calculate d and ARP according to Eq. 19 and Eq. 20;
3 calculate ELT according to position prediction or
planned path;
4 set a timer Tsend ;
5 while Tsend does not expire do
6 wait;
7 end
(c) (d)
8 send d, ARP , position and velocity to vi ;
9 set a timer Tnotf ; Fig. 4. Parallel transmission of safety messages with overlapping RoI. (a)
10 while Tnotf does not expire do TWN establishment from Message A. (b) Generation of message B. (c) TWN
establishment from Message B. (d) Parallel transmission of overlapping RoI.
11 if receives the notification to be selected as CV
then
12 set myself as a CV;
13 set vi as my superior CV;
14 end
15 end
16 if does not receive the notification to be selected as
CV then
17 set myself as an OV;
18 set vi as my superior CV;
19 end

Fig. 5. Road topology and traffic condition.

reliability of message dissemination among CVs, we adopted a
retransmission operation to ensure the successful transmission
of safety messages. Specifically, when a packet is transmitted, safety message, part of the existing TWN can be used to
the sender needs to delay a backoff window to retransmit if reduce the time and communication overhead of the network
a channel collision is detected. Besides, when a superior CV establishment process (see Fig. 4(c)). Specifically, if a CV
does not receive the ACK message from a subordinate CV receives another safety message, the CV examines the AA
or the safety message needs to be updated, the retransmission of its superior and subordinate CVs and transmits the new
operation will also be adopted. Secondly, the CVs dissem- message to the CVs in the RoI of the new message. The CVs
inate the safety message to the OVs in AAs. The message in the overlapping area of the two RoIs merge the two safety
dissemination in the AA of a CV triggers once the three-step messages together and broadcast within the AA (see Fig. 4(d)).
connection procedure of the CV finishes and subordinate CVs
are determined. To improve the performance of safety message VI. S IMULATION AND P ERFORMANCE
dissemination when the connectivity among vehicles is poor, In this section, we conduct experiments using Veins sim-
the safety message will be periodically broadcast in the AA ulator, which combines OMNeT++ for event-driven network
of CVs until the CV leaves the RoI or the safety message simulation and SUMO for the generation of traffic environ-
expires. Also, the broadcast frequency varies according to the ment and vehicle movement. We evaluate the performance of
position of CV in the RoI, i.e., the CVs close to the border our scheme in terms of i) coverage and timeliness of message
of the RoI broadcast the message with a higher frequency so dissemination and ii) duration of message dissemination in
that the vehicles newly entering the RoI can quickly access the RoI. We also compare the performance with other popular
safety message, while the CVs near the center adopt a lower ones, including UMBP [26], MCMA [13] and the standard
frequency in order to reduce the data redundancy. flooding protocol.
The TWN also supports multiplexing to reduce the overhead The safety message dissemination coverage of all vehicles
when multiple safety messages are transmitted simultaneously. in the RoI is the primary target of our scheme. Nevertheless,
For instance, after a TWN is established from Message A since vehicles are not evenly distributed on the road, we also
(see Fig. 4(a)), if another safety message B is generated (see introduce the metrics of road coverage which reflects the
Fig. 4(b)), and its RoI overlaps with the RoI of the existing coverage of safety message from the geometric perspective.

© 2022 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: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on October 26,2022 at 09:08:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Internet of Things Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2022.3195655

IEEE INTERNET OF THINGS JOURNAL, VOL. 14, NO. 8, AUGUST 1, 2022 10

TABLE II B. Coverage of Networking and Message Dissemination

S IMULATION PARAMETERS
Due to the dynamics of road conditions, the coverage under
Parameter Value two different definitions varies with time. In this part, we
Physical protocol IEEE 802.11p
evaluate our scheme’s timeliness and message coverage. As
Ratio of cruise vehicles 25%–75% described at the beginning of this section, the coverage of
Transmission range 250 m safety message dissemination is defined in terms of both
Maximum vehicle velocity 20 m/s
Simulation time 960 s
vehicles and road segments, while the growth rate of coverage
Rebroadcast period 5–20 s indicates the timeliness of the message dissemination.
Safety message lifetime 20 min Fig. 6 shows the safety message coverage with the dif-
Transmission power 20 mW
Bit rate 6 Mbps
ferent density of vehicles and the structure of RoI. It can
α 0.3 be observed that TWNSMD achieves almost 100% coverage
β 0.3 in different structures of RoI. Particularly, the distribution
γ 0.4
of vehicles is relatively concentrated in a grid-like structure
and therefore takes less time than in a truck-like structure.
By contrast, UMBP, MCMA and flooding reach about 90%
Thus, the coverage of the safety message is defined from two
coverage. Besides, UMBP has a longer coverage delay because
following aspects, i) for road segments in the RoI, coverage is
it chooses the farthest one-hop neighbor as the next-hop relay,
defined as the ratio of road segments covered by the safety
which may fail to transmit the safety message. Flooding has a
message in the RoI; ii) for vehicles in the RoI, the ratio
faster coverage in the initial phase without the relay selection
of vehicles currently received the safety message in the RoI
process, but much channel congestion leads to the decrease
represent the coverage.
of coverage speed afterward. MCMA can also cover the area
at a faster rate in the initial process. The reason is that
A. Simulation Settings
MCMA adopts RSUs to forward messages, which achieve
We adopt IEEE 802.11p as the vehicular communication larger coverage than selecting vehicles to relay messages.
protocol in the simulation. Nonetheless, it should be noted But MCMA ignores the temporal correlation among vehicles,
that our solution can also support the latest LTE-V2X and NR- leading to disconnections among some vehicles, which makes
V2X protocols. We assume that in case of emergency, a source it difficult for MCMA to ultimately achieve the same coverage
vehicle broadcasts a BSM to its one-hop neighbors to trigger as our proposed TWNSMD. In TWNSMD, we take link
the broadcast process. Table II lists the detailed simulation stability into account to effectively reduce the retransmissions,
parameters. resulting in a faster transmission of safety messages in the RoI.
Since the structure of RoI may affect the performance
of our message dissemination scheme, we set two different
structure of RoI: a trunk-like RoI and a grid-like RoI. It shall C. Duration of Networking and Message Dissemination
be noted that several other parameters may have significant We measure the duration of networking and message dis-
impact on the performance of our proposed scheme, e.g., semination by the variation of the safety message coverage
the selection factor threshold, the size of AA and RoI, the over time, which is similarly defined in terms of both vehicles
broadcast frequency, etc. The optimization of such parameters and road segments. Fig. 7 shows coverage-time under different
should be jointly considered with the specific applications density of vehicles.
in vehicular networks, which will be further studied in our As for the duration of safety message dissemination, we
future work. In this study, we focus on the framework of evaluate the coverage trends of our method over a longer
TWN establishment and maintenance, so we determine the period. In Fig. 7(a) and Fig. 7(b), the coverage of safety
parameters and study the impact on the performance of the message reaches to around 70% after 200 seconds under grid-
proposed framework by simulations. like RoI. The remaining 30% vehicles that cannot be covered
The trajectory data are collected from DiDi Express and by safety message mainly exist in the region near the border of
DiDi Premier drivers. This dataset contains traffic information RoI or with lower vehicle density. In Fig. 7(c) and Fig. 7(d),
from a rectangular area (6 km ×10 km) which is divided into the vehicles in trunk-like RoI are less sensitive to the safety
8205 regions according to the hexagon grid data and includes message and therefore the coverage persists around 60% after
trajectories of 1826 vehicle trajectories in Chengdu, China. 200 seconds. Besides, we can observe from Fig. 7 that TWN
The measurement interval of the track points is approximately can maintain the coverage at a certain level over time. The
2-4 seconds. The track points are bound to physical roads reason is that we adopted a retransmission operation and a
to match the trajectory data and the actual road information. periodically broadcast mechanism to guarantee the successful
As illustrated in the sub-figure of Fig. 5, the map contains transmission of safety messages. In contrast, other existing
irregular intersections, curve roads, one-way streets, etc. Also, methods ignore the maintenance of networks, resulting in
there exists an unbalanced distribution of traffic flows in the a rapid decrease in coverage over time. In addition, there
dataset. For instance, we can observe that the traffic in one is no significant difference for the duration of the message
direction is much higher than in the other direction in some under different vehicle densities and structures of RoI, which
road segments, which usually happens in the morning/evening proves the feasibility of the proposed method in different
rush hour between the residential areas and office regions. scenarios. By comparison, most existing methods, such as

© 2022 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: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on October 26,2022 at 09:08:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Internet of Things Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2022.3195655

IEEE INTERNET OF THINGS JOURNAL, VOL. 14, NO. 8, AUGUST 1, 2022 11

100 100 100 100

80 80 80 80
Coverage (%)

Coverage (%)

Coverage (%)

Coverage (%)
60 60 60 60

40 40 40 40
TWNSMD (dense) TWNSMD (dense) TWNSMD (dense) TWNSMD (dense)

TWNSMD (average) TWNSMD (average) TWNSMD (average) TWNSMD (average)

20 TWNSMD (sparse) 20 TWNSMD (sparse) 20 TWNSMD (sparse) 20 TWNSMD (sparse)

MCMA average MCMA average MCMA average MCMA average

UMBP (average) UMBP (average) UMBP (average) UMBP (average)

0 0 0 0
Flooding (average) Flooding (average) Flooding (average) Flooding (average)

0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8

Time (s) Time (s) Time (s) Time (s)

(a) (b) (c) (d)

Fig. 6. Coverage and timeliness under different density of vehicles and two road typologies. (a) Coverage for vehicles, a grid-like RoI. (b) Coverage for road
segments, a grid-like RoI. (c) Coverage for vehicles, a trunk-like RoI. (d) Coverage for road segments, a trunk-like RoI.

100 100
100 100

80 80 80 80
Coverage (%)

Coverage (%)

Coverage (%)

Coverage (%)
60 TWNSMD (dense) 60 TWNSMD (dense) 60 60
TWNSMD (dense)
TWNSMD (dense)
TWNSMD (average) TWNSMD (average) TWNSMD (average)
TWNSMD (average)
TWNSMD (sparse) TWNSMD (sparse) TWNSMD (sparse)
TWNSMD (sparse)
MCMA (average) MCMA (average) 40 MCMA (average) 40
40 40 MCMA (average)
UMBP (average) UMBP (average) UMBP (average)
UMBP (average)
Flooding (average) Flooding (average) Flooding (average)
Flooding (average)
20 20
20 20

0 0

0 0
0 200 400 600 800 1000 0 200 400 600 800 1000 0 200 400 600 800 1000 0 200 400 600 800 1000

Time (s) Time (s) Time (s) Time (s)

(a) (b) (c) (d)

Fig. 7. Duration of message dissemination with different density of vehicles and two road typologies. (a) Coverage for vehicles, a grid-like RoI. (b) Coverage
for road segments, a grid-like RoI. (c) Coverage for vehicles, a trunk-like RoI. (d) Coverage for road segments, a trunk-like RoI.

100 100 100 100

95 95 95 95

90 90 90 90 Flooding Flooding

grid-like RoI grid-like RoI

Redundancy packets
85 85 85 85
Coverage (%)

Retransmissions
Coverage (%)

80 80 80 80

75 75 75 75

70 70 70 70

65 65 65 65
10% cruise vehicle 10% cruise vehicle 10% cruise vehicle 10% cruise vehicle
25% cruise vehicle 25% cruise vehicle
60 25% cruise vehicle 60 25% cruise vehicle
60 60
50% cruise vehicle 50% cruise vehicle
50% cruise vehicle 50% cruise vehicle
Coverage for vehicle under grid-like RoI Coverage for road under grid-like RoI
55 55 55 Coverage for vehicle under trunk-like RoI 55 Coverage for road under trunk-like RoI

0 200 400 600 800 1000 0 200 400 600 800 1000 0 200 400 600 800 1000 0 200 400 600 800 1000

Time(s) Time (s) Time (s) Time (s)

Vehicle density Vehicle density

(a) (b)
(a) (b)
100 100 100 100

80 80 80 80
Fig. 9. Overhead of safety message dissemination. (a) Number of retrans-
60 60 60 60
missions with different density of vehicles in trunk-like and grid-like RoI. (b)
Coverage (%)

Coverage (%)

40 40 40 40
Redundancy with different density of vehicles in in trunk-like and grid-like
20 20 20 20
RoI.
0 0 0 0

Message A Message A
Message A Message A
Message B (100% RoI overlapping with A) Message B (100% RoI overlapping with A)
Message B (100% RoI overlapping with A) Message B (100% RoI overlapping with A)
-20 Message C (50% RoI overlapping with A)
-20 Message C (50% RoI overlapping with A)
-20
Message C (50% RoI overlapping with A)
-20
Message C (50% RoI overlapping with A)

Message D (0% RoI overlapping with A) Message D (0% RoI overlapping with A)
Message D (0% RoI overlapping with A) Message D (0% RoI overlapping with A)
Coverage for vehicle under trunk-like RoI Coverage for road under trunk-like RoI
Coverage for vehicle under grid-like RoI Coverage for road under grid-like RoI
-40 -40 -40 -40

0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10
Generally, the coverage of safety messages declines faster,
Time (s) Time (s) Time(s) Time (s)
and the duration of stable coverage will become slightly
(c) (d) lower in the situation of more cruise vehicles. In the case
Fig. 8. Duration of safety message dissemination. (a) Duration of message of two road typologies, the difference of region similarity
with different ratio of cruise vehicles under grid-like RoI. (b) Duration of becomes more noticeable, especially when the active area of
message with different ratio of cruise vehicles under trunk-like RoI. (c) the vehicle is highly similar to the RoI under grid-like RoI.
Dissemination of two safety messages under grid-like RoI. (d) Dissemination
of two safety messages under trunk-like RoI. This is because the predicted future positions of the cruise
vehicles may be different from the actual trajectory, which
leads to the deviation between the actual trajectory and the
UMBP, MCMA and the flooding method, do not consider calculated active area. Therefore, trajectory prediction with
the broadcast duration, which results in a rapid decline of higher accuracy over a longer period may improve the duration
coverage. of the safety message dissemination.
Furthermore, the proportion of cruise vehicles also has an We also evaluate the compatibility of TWN for the multiple
impact on dissemination performance. Fig. 8(a) and Fig. 8(b) safety messages dissemination by different IBVs. The existing
show the duration of the message dissemination with different TWN network structure can be reused in case other safety
cruise vehicles ratios under two road typologies, respectively. messages with overlapping RoI need to be transmitted. We

© 2022 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: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on October 26,2022 at 09:08:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Internet of Things Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2022.3195655

IEEE INTERNET OF THINGS JOURNAL, VOL. 14, NO. 8, AUGUST 1, 2022 12

consider a scenario wherein message A is first completely and better exploit the availability of onboard data to facilitate
transmitted in different structures of RoI. Then we verify the various vehicular applications more effectively and efficiently.
coverage of messages B, C, and D with varying settings of Communication efficiency: In the foreseeable future, the
the overlapping percentage of RoI related to message A. As data generated by intelligent vehicles will likely increase
shown in Fig. 8(c) and Fig. 8(d), for 100% RoI overlapping exponentially and be transmitted at a high frequency (similar
(message B), the transmission of the new message is around to the current big data trend). However, traditional V2V-based
21% faster in trunk-like RoI and around 25% faster in grid- or clustering-based networking schemes mostly focus on end-
like RoI than independent transmission (message D), and 15% to-end data transmission, ignoring the network coverage and
faster in trunk-like RoI and 20% faster in grid-like RoI for 50% duration. The data transmission efficiency may not be achieved
overlapping (message C). In our framework, safety messages due to the high overhead of frequent information exchange
with different overlapping RoI are transmitted by the same caused by link establishment. Moreover, in a traditional vehic-
part of TWN in the overlapping area to reduce the time for ular cloud network, data are normally uploaded and processed
TWN establishment. in the cloud or at the edge server, which may become a
bottleneck for the vehicular applications with the requirement
D. Overhead of large and frequent data exchange. The proposed TWN
consider the trajectory of vehicles and explore the spatio-
Message dissemination methods in VCN may cause channel
temporal features of safety message to establish a stable local
collision and packet loss. When a packet is transmitted, the
vehicular network, which makes it possible to realize IV data
sender needs to delay a backoff window to retransmit if a
storage and processing at three levels, i.e., onboard, edge, and
channel collision is detected. This mechanism also applies in
cloud. Thus, it can alleviate the aforementioned limitations
the process of TWN establishment when a superior CV does
by pre-processing the collected data so that the data can be
not receive the ACK message from a subordinate CV. Fig.
aggregated/filtered before uploading.
9(a) shows the total number of retransmissions when the three
Deployment and cost: The TWN proposed in this paper
methods achieve the maximum message coverage respectively
is built on top of V2V communication and does not rely
in different simulation scenarios. It can be observed our
on infrastructures. V2V communication is self-organizing and
method outperforms to UMBP and flooding, especially when
depends only on some on-board devices with no user costs. In
the CVs are located in grid-like RoI. This is because CVs are
other words, from the perspective of network construction, we
selected to minimize the overlapping area between the AAs,
do not need to provide incentives for relay vehicles to join the
which reduces the channel collisions in the same area.
TWN and forward safety messages. Therefore, our solution
The number of redundancy packets are also measured as
can be easily incorporated in traditional cloud-based VCN,
the metric of overhead. Due to the retransmissions, the same
and accordingly, further extend the coverage and connectivity
packets will be repeatedly received by some vehicles. Fig.
of vehicular network. More importantly, as the number of
9(b) shows the number of redundancy packets for different
smart cars with high computing and communication resources
structures of RoI when the three methods achieve 100%
increases, the proposed TWN can be deployed at a much lower
message coverage. It can be seen that the redundancy packets
cost.
are significantly reduced in the proposed method, especially
when the vehicle is at the grid-like RoI. This is because the
broadcast frequency can be adaptively adjusted according to VII. C ONCLUSIONS AND F UTURE W ORK
the position of CVs in RoI. Meanwhile, the grid-like RoI may
In this paper, we have proposed a novel TWN for safety
provide more suitable subordinate vehicles for the CVs.
message dissemination in the urban traffic scenario. Compared
to traditional networking methods in VANET, the spatio-
E. Discussion temporal correlation of the vehicle trajectory is utilized to
In this part, we discuss some potential benefits based on the establish and maintain a TWN for safety message dissemina-
design and simulation results of the proposed TWN, and also tion. Specifically, to achieve better performance of the TWN,
the advantages compared to some existing work. we also extract the connectivity characteristics of the vehicle
Response time: Real-time response is a critical evaluation network from the historical trajectory data and the results are
indicator to most vehicular applications, such as vehicle con- utilized for the establishment of the network. The optimal
trol and road safety. However, conventional vehicular cloud coverage, timeliness, duration, and redundancy of dissemina-
computing architecture is not specifically designed to meet tion are also considered when establishing and maintaining the
this low-latency requirement [9], since data collected from TWN structure. Extensive simulation results indicate that our
smart vehicles will be processed remotely instead of locally. scheme achieves a faster coverage and much longer continuity
Due to the transmission delay and potential connectivity issues of message dissemination in the RoI with less overhead.
(e.g., out of communication coverage), the average response Moreover, since our proposed scheme is infrastructure-free
time for cloud-based applications will likely be more than and only depends on some onboard devices with lower costs,
a second [10]. The proposed TWN, located in proximity to it will be easily deployed in practical traffic applications such
smart vehicles, can significantly reduce the response time as content sharing and information spreading. The simulation
for delay-sensitive applications. it can provide flexible and results in this paper can provide practical guidance for the
efficient communication capacity to the end nodes in a VCN, system implementation in real traffic scenarios.

© 2022 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: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on October 26,2022 at 09:08:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Internet of Things Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2022.3195655

IEEE INTERNET OF THINGS JOURNAL, VOL. 14, NO. 8, AUGUST 1, 2022 13

Although connected vehicles are not yet sufficiently [15] K. Liu, V. C. S. Lee, J. K.-Y. Ng, J. Chen, and S. H. Son, “Temporal
widespread currently and are inadequate to support the imple- data dissemination in vehicular cyber–physical systems,” in IEEE Trans.
Intell. Transp. Syst. (TITS), vol. 15, no. 6, pp. 2419–2431, 2014.
mentation of our proposed TWN, we believe the popularity of [16] G. Luo, H. Zhou, N. Cheng, Q. Yuan, J. Li, F. Yang, and X. Shen,
connected vehicles is the future trend and our TWN is able “Software-defined cooperative data sharing in edge computing assisted
to provide reliable and efficient safety message dissemination 5g-vanet,” in IEEE Trans. Mob. Comput. (TMC), vol. 20, no. 03, pp.
1212–1229, 2021.
service. In future work, some novel learning-based methods [17] G. Luo, H. Zhou, N. Cheng, Q. Yuan, J. Li, F. Yang, and X. S. Shen,
such as multi-agent reinforcement learning (MARL) will also “Software defined cooperative data sharing in edge computing assisted
be considered to further optimize the parameters settings (e.g., 5g-vanet,” in IEEE Trans. Mob. Comput. (TMC), 2019.
[18] Q. Yuan, H. Zhou, J. Li, Z. Liu, F. Yang, and X. S. Shen, “Toward
selection factor threshold, RoI size, and broadcast frequency) efficient content delivery for automated driving services: An edge
and improve the performance of TWN. Besides, it is very computing solution,” in IEEE Netw., vol. 32, no. 1, pp. 80–86, 2018.
difficult to achieve accurate of vehicle trajectory predictions [19] K. Liu, K. Xiao, P. Dai, V. C. Lee, S. Guo, and J. Cao, “Fog comput-
ing empowered data dissemination in software defined heterogeneous
in highly dynamic urban environments. In the future, we will vanets,” in IEEE Trans. Mob. Comput. (TMC), vol. 20, no. 11, pp. 3181–
consider more advanced deep learning frameworks to achieve 3193, 2020.
more accurate trajectory prediction for TWN construction and [20] K. Liu, X. Xu, M. Chen, B. Liu, L. Wu, and V. C. Lee, “A hierarchical
architecture for the future internet of vehicles,” in IEEE Commun. Mag.,
maintainance. vol. 57, no. 7, pp. 41–47, 2019.
[21] M. Ali, A. W. Malik, A. U. Rahman, S. Iqbal, and M. M. Hamayun,
“Position-based emergency message dissemination for internet of ve-
R EFERENCES hicles,” in Int. J. Distrib. Sens. Netw. (IJDSN), vol. 15, no. 7, p.
1550147719861585, 2019.
[1] Y. C. Hu, M. Patel, D. Sabella, N. Sprecher, and V. Young, “Mobile edge [22] M. Syfullah, J. M.-Y. Lim, and F. L. Siaw, “Mobility-based clustering
computing—a key technology towards 5g,” ETSI white paper, vol. 11, algorithm for multimedia broadcasting over ieee 802.11 p-lte-enabled
no. 11, pp. 1–16, 2015. vanet.” in IEEE Trans. Intell. Transp. Syst. (TITS), vol. 13, no. 3, pp.
[2] Z. Khan, P. Fan, S. Fang, and F. Abbas, “An unsupervised cluster-based 1213–1237, 2019.
vanet-oriented evolving graph (cvoeg) model and associated reliable [23] M. Ren, L. Khoukhi, H. Labiod, J. Zhang, and V. Veque, “A new
routing scheme,” in IEEE Trans. Intell. Transp. Syst. (TITS), vol. 20, mobility-based clustering algorithm for vehicular ad hoc networks
no. 10, pp. 3844–3859, 2019. (vanets),” in Proceedings of NOMS 2016-2016 IEEE/IFIP Network
[3] F. Lyu, H. Zhu, N. Cheng, H. Zhou, W. Xu, M. Li, and X. Shen, Operations and Management Symposium. IEEE, 2016, pp. 1203–1208.
“Characterizing urban vehicle-to-vehicle communications for reliable [24] M. Ren, L. Khoukhi, H. Labiod, J. Zhang, and V. Veque, “A mobility-
safety applications,” in IEEE Trans. Intell. Transp. Syst. (TITS), vol. 21, based scheme for dynamic clustering in vehicular ad-hoc networks
no. 6, pp. 2586–2602, 2019. (vanets),” in Veh. Commun., vol. 9, pp. 233–241, 2017.
[4] W. Chen, F. Lyu, F. Wu, P. Yang, G. Xue, and M. Li, “Sequential [25] X. Qu, E. Liu, R. Wang, and H. Ma, “Complex network analysis of vanet
message characterization for early classification of encrypted internet topology with realistic vehicular traces,” in IEEE Trans. Veh. Technol.
traffic,” in IEEE Trans. Veh. Technol. (TVT), vol. 70, no. 4, pp. 3746– (TVT), vol. 69, no. 4, pp. 4426–4438, 2020.
3760, 2021. [26] Y. Bi, H. Shan, X. S. Shen, N. Wang, and H. Zhao, “A multi-
[5] B. Liu, D. Jia, K. Lu, D. Ngoduy, J. Wang, and L. Wu, “A joint control– hop broadcast protocol for emergency message dissemination in urban
communication design for reliable vehicle platooning in hybrid traffic,” vehicular ad hoc networks,” in IEEE Trans. Intell. Transp. Syst. (TITS),
in IEEE Trans. Veh. Technol. (TVT), vol. 66, no. 10, pp. 9394–9409, vol. 17, no. 3, pp. 736–750, Mar. 2016.
2017. [27] X. M. Zhang, L. Yan, K. H. Chen, and D. K. Sung, “Fast, efficient
[6] F. Lyu, N. Cheng, H. Zhou, W. Xu, W. Shi, J. Chen, and M. Li, “Dbcc: broadcast schemes based on the prediction of dynamics in vehicular ad
Leveraging link perception for distributed beacon congestion control in hoc networks,” in IEEE Trans. Intell. Transp. Syst. (TITS), pp. 531–542,
vanets,” in IEEE Internet Things J. (IoT-J), vol. 5, no. 6, pp. 4237–4249, 2019.
2018. [28] X. M. Zhang, L. Yan, H. Zhang, and D. K. Sung, “A concurrent
[7] P. Wang, B. Di, H. Zhang, K. Bian, and L. Song, “Cellular v2x transmission based broadcast scheme for urban VANETs,” in IEEE
communications in unlicensed spectrum: Harmonious coexistence with Trans. Mob. Comput. (TMC), vol. 18, no. 1, pp. 1–12, Jan. 2019.
vanet in 5g systems,” in IEEE Trans. Wirel. Commun. (TWC), vol. 17, [29] M. Noor-A-Rahim, G. G. M. N. Ali, Y. L. Guan, B. Ayalew, P. H. J.
no. 8, pp. 5212–5224, 2018. Chong, and D. Pesch, “Broadcast performance analysis and improve-
[8] B. Liu, D. Jia, K. Lu, H. Chen, R. Yang, J. Wang, Y. Barnard, and ments of the lte-v2v autonomous mode at road intersection,” in IEEE
L. Wu, “Infrastructure-assisted message dissemination for supporting Trans. Veh. Technol. (TVT), vol. 68, no. 10, pp. 9359–9369, 2019.
heterogeneous driving patterns,” in IEEE Trans. Intell. Transp. Syst. [30] S. Kamakshi and V. S. Shankar Sriram, “Plummeting broadcast storm
(TITS), vol. 18, no. 10, pp. 2865–2876, 2017. problem in highways by clustering vehicles using dominating set and
[9] B. Liu, D. Jia, J. Wang, K. Lu, and L. Wu, “Cloud-assisted safety mes- set cover,” in Sensors, vol. 19, no. 9, p. 2191, May 2019.
sage dissemination in vanet–cellular heterogeneous wireless network,” [31] A. Naja, M. Boulmalf, M. Essaaidi, and O. Ait Oualhaj, “Adaptive alert
in IEEE Syst. J. (ISJ), vol. 11, no. 1, pp. 128–139, 2017. content dissemination protocol inspired from volunteer’s dilemma game
[10] J. Wang, J. Liu, and N. Kato, “Networking and communications in for Vehicular Ad-hoc Networks,” in Veh. Commun., vol. 19, p. 100180,
autonomous driving: A survey,” in IEEE Commun. Surv. Tutorials Oct. 2019.
(COMST), vol. 21, no. 2, pp. 1243–1274, 2018. [32] E. Limouchi and I. Mahgoub, “Volunteers dilemma game inspired
[11] D. Grewe, M. Wagner, and H. Frey, “A domain-specific comparison of broadcast scheme for vehicular ad hoc networks,” in IEEE Trans. Intell.
information-centric networking architectures for connected vehicles,” in Transp. Syst. (TITS), vol. 20, no. 12, pp. 4439–4449, Dec 2019.
IEEE Commun. Surv. Tutorials (COMST), vol. 20, no. 3, pp. 2372–2388, [33] L. Bedogni, M. Fiore, and C. Glacet, “Temporal reachability in vehicular
2018. networks,” in Proceedings of IEEE INFOCOM 2018 - IEEE Conference
[12] O. Senouci, Z. Aliouat, and S. Harous, “Mca-v2i: A multi-hop clustering on Computer Communications, 2018, pp. 81–89.
approach over vehicle-to-internet communication for improving vanets [34] B. Wu, H. Shen, and K. Chen, “Exploiting active subareas for multicopy
performances,” in Future Gener. Comput. Syst. (FGCS), vol. 96, pp. routing in VDTNs,” in IEEE Trans. Veh. Technol. (TVT), vol. 67, no. 5,
309–323, 2019. pp. 4374–4388, 2017.
[13] A. Jalooli, M. Song, and W. Wang, “Message coverage maximization
in infrastructure-based urban vehicular networks,” Vehicular Communi-
cations, vol. 16, pp. 1–14, 2019.
[14] J. Jeong, Y. Shen, S. Jeong, S. Lee, H. Jeong, T. Oh, T. Park, M. U.
Ilyas, S. H. Son, and D. H. Du, “Stmac: Spatio-temporal coordination-
based mac protocol for driving safety in urban vehicular networks,” in
IEEE Trans. Intell. Transp. Syst. (TITS), vol. 19, no. 5, pp. 1520–1536,
2017.

© 2022 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: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on October 26,2022 at 09:08:11 UTC from IEEE Xplore. Restrictions apply.
 This article has been accepted for publication in IEEE Internet of Things Journal. This is the author's version which has not been fully edited and
content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2022.3195655

IEEE INTERNET OF THINGS JOURNAL, VOL. 14, NO. 8, AUGUST 1, 2022 14

Bingyi Liu received the Ph.D. degree in the De- Jianping Wang received the B.S. and M.S. degrees
partment of Computer Science from City University in computer science from Nankai University, Tianjin,
of Hong Kong, Kowloon, Hong Kong in 2018, and China, in 1996 and 1999, respectively, and the Ph.D.
in the Department of Computer Science, Wuhan degree in computer science from the University of
University, Wuhan, China in 2017, respectively. He Texas at Dallas, Richardson, TX, USA, in 2003.
was a Visiting Ph.D. student in the Department of She is currently a Professor in the Department
Computer Science, University at Buffalo, The State of Computer Science, City University of Hong
University of New York, NY, USA from 2016 to Kong, Kowloon, Hong Kong. Her research interests
2017. He is currently an Associate Professor in the include dependable networking, optical networks,
School of Computer Science and Artificial Intel- cloud computing, service oriented networking, and
ligence, Wuhan University of Technology, Wuhan, data center networks.
China, Sanya Science and Education Innovation Park of Wuhan University
of Technology, Sanya, China, and Chongqing Research Institute, Wuhan
University of Technology, Chongqing, China. His research interests include
wireless networks, vehicular ad-hoc network, and Internet of Things.
Chunming Qiao (F’10) is currently the Director
of the Laboratory for Advanced Network Design,
Analysis, and Research, University at Buffalo. He
has pioneered research on optical Internet and in
particular, optical burst switching. He was the Chair
Weizhen Han received the B.E. degree in 2019 and
of the IEEE Technical Committee on High Speed
is currently pursuing the M.E. degree in Wuhan Uni-
Networks and the IEEE Subcommittee on Integrated
versity of Technology, Wuhan, China. His research
Fiber and Wireless Technologies, which he founded.
interests include vehicular ad-hoc network.
He was elected an IEEE Fellow for his contributions
to optical and wireless network architectures and
protocols.

Wei Jiang received the B.E. degree in 2018 and

is currently pursuing the M.E. degree in Wuhan
University of Technology, Wuhan, China. His re-
search interests include vehicular ad-hoc network
and mobile edge computing.

Dongyao Jia received the B.E. degree in automation

from Harbin Engineering University, Harbin, China,
in 1998, and the Ph.D. degree in computer science
from the City University of Hong Kong in 2014.
He is currently an Associate Professor at the School
of Advanced Technology, Xi’an Jiaotong-Liverpool
University, China. He worked as a Research Fellow
in the University of Queensland from 2018 to 2021,
and the Institute for Transport Studies, University of
Leeds from 2015 to 2018, respectively. He also holds
ten-year working experience in the telecom industry
in China. His research interests include connected and automated driving,
intelligent transport systems, traffic simulation, and the Internet of Things.

Enshu Wang is currently pursuing the Ph.D. degree

in computer science and engineering with the Uni-
versity at Buffalo, The State University of New York,
Buffalo, NY, USA. His research interests include
data mining and vehicular cyber-physical systems.

© 2022 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: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on October 26,2022 at 09:08:11 UTC from IEEE Xplore. Restrictions apply.