SEARCH: An SDN-enabled Approach for Vehicle

Path-Planning
Omar Sami Oubbati, Mohammed Atiquzzaman, Pascal Lorenz, Abdullah Baz,
Hosam Alhakami

To cite this version:

Omar Sami Oubbati, Mohammed Atiquzzaman, Pascal Lorenz, Abdullah Baz, Hosam Alhakami.
SEARCH: An SDN-enabled Approach for Vehicle Path-Planning. IEEE Transactions on Vehicular
Technology, In press, �10.1109/TVT.2020.3043306�. �hal-03110411�

HAL Id: hal-03110411

[Link]
Submitted on 14 Jan 2021

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXX 2020 1

SEARCH: An SDN-enabled Approach for Vehicle

Path-Planning
Omar Sami Oubbati, Member, IEEE, Mohammed Atiquzzaman, Senior Member, IEEE,
Pascal Lorenz, Senior Member, IEEE, Abdullah Baz, Senior Member, IEEE,
and Hosam Alhakami, Senior Member, IEEE

Abstract—With increasing vehicle density and the growth I. I NTRODUCTION

of accidents in urban areas, navigation management becomes
a serious problem. Even though there is a multitude of The last decade has witnessed a severe increase in
navigation systems, ambulances, taxis, or even ordinary vehicles, congestion on our roads. Indeed, different statistics across
sometimes find it challenging to reach their destinations on the web show that traffic congestion can negatively affect
time. There are two main reasons for this difficulty: (i) the economic growth of cities by losing travel time, wasted
lack of local knowledge of the area of navigation solutions
and (ii) their inflexibility against unforeseeable situations that money due to loss of productivity, and waste of fuel [1].
may occur on the roads. Indeed, the majority of navigation Even worse, the number of vehicles continues to increase at a
solutions are based only on the distance, the journey time, rate of four million vehicles a year, especially in metropolitan
or even statistics related to the density of vehicles to plan areas where the majority of the world’s population lives [2].
the full paths, while neglecting the dynamic nature of the As a result, a considerable effort has been undertaken by
vehicle traffic. Also, their respective centralized architectures
are unable to monitor both the traffic and unexpected events researchers to add new infrastructures (e.g., dedicated lanes
continuously and in real-time without not being overloaded for buses) to minimize the congestion [3]. Nevertheless, the
by the flow of message exchanges between the road entities impact of this classical solution is restricted because the speed
and the central processing entity. To address these issues, we of construction of new roads is much slower than the increase
propose in this paper a novel three-tier architecture, called in the number of new vehicles. To adapt to this situation
SDN-enabled Approach for Vehicle Path-Planning (SEARCH), to
enhance the situation awareness on urban roads, efficiently collect and traveling without wasting time on roads, scientists try to
traffic information in real-time, and decide the best navigation design path planning mechanisms to allow road-users to avoid
strategy. The proposed architecture exploits Unmanned Aerial traffic congestion whenever possible. However, most current
Vehicles (UAVs), Vehicular Ad hoc Networks (VANETs), 5G path planning solutions are based exclusively on the shortest
based cellular systems, and Software-Defined Networking (SDN) distance or shortest time path algorithms, without considering
to provide better and faster communication to changing road
conditions. Based on these technologies, some parameters related the impact of future traffic conditions during the duration
to vehicles and driving environments, such as speed, distance, of the trip. Consequently, scientists have looked heavily into
traffic jams, incidents, and travel flow, are efficiently collected the possibility of using dynamic path planning based on
and dynamically exploited to achieve faster paths between real-time traffic information, which gives rise to the following
any existing pairs of locations. Furthermore, the deployed challenges:
architecture of SEARCH can provide sufficient bandwidth to
support all data traffic needs to update vehicles during their • How to efficiently collect traffic information from the
journey efficiently and in real-time. To evaluate the performance roads in real-time?
of our architecture, we conduct a series of simulations and • How to detect any incidents or temporary events (e.g.,
perform a set of comparisons with relevant route planning
road repair) that can occur at any time on the roads?
algorithms. We found that the proposed architecture works
effectively in terms of saving on driving time to reach any target • When is the right time to make path re-planning
destinations. decisions?
Index Terms—5G; SDN; Traffic management; VANET; Path • How to accurately estimate the journey time of a given
planning. vehicle?
• What is the best network architecture that provides
Copyright (c) 2020 IEEE. Personal use of this material is permitted. optimal coverage, saves bandwidth, reduces latency, and
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending a request to pubs-permissions@[Link]. maximizes throughput?
O.S. Oubbati is with the Computer Science and mathematics Laboratory, To overcome all these challenges, a set of emerging
University of Laghouat, Algeria. E-mail: [Link]@[Link].
M. Atiquzzaman is with the University of Oklahoma, Norman, OK USA. technologies, such as 5G based cellular systems, Unmanned
E-mail: atiq@[Link] Aerial Vehicles (UAVs), Vehicular Ad hoc Networks
P. Lorenz is with the University of Haute Alsace, IUT Colmar, France. (VANETs), and Software-Defined Networking (SDN), are
E-mail: [Link]@[Link]
A. Baz is with the Department of Computer Engineering, College of expected to answer all those questions and significantly
Computer and Information Systems, Umm Al-Qura University, Makkah, Saudi mitigate urban traffic congestion. Indeed, 5G technology
Arabia. Email: aobaz01@[Link] offers enhanced scalability and higher bandwidths to support
H. Alhakami is with the Department of Computer Science, College of
Computer and Information Systems, Umm Al-Qura University, Makkah, Saudi different applications dealing with the traffic congestion issues
Arabia. Email: hhhakam@[Link] in the roads [4], [5]. VANETs have been heavily involved
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXX 2020 2

in nearly all safety and comfort applications, such as crash performance evaluation of our proposed algorithm. Finally, in
avoidance, autonomous driving, cooperative driving, collective Section V, we present the conclusion and future perspectives.
situational awareness, and even natural language processing
[6]. Therefore, VANETs are already approaching deployment,
II. R ELATED W ORKS
which would make the roads safer, the travel experience more
pleasant, and the transportation system more efficient [7], [8]. 5G and beyond 5G (B5G) are (at the time of writing
To assist 5G cellular networks and VANETs, UAVs are well this paper) the latest cellular networking technologies being
equipped with the required devices to ensure coverage, collect developed, which are specifically designed to both provide
traffic information, and explore different geographical areas, high data rates and promise a latency of 1 ms for
for example, through aerial photography [9]. To top it all, real-time applications [13]. These technologies support many
SDN technology can effectively manage all these networks technologies, including Heterogeneous Networks and SDN, to
by bringing flexibility, scalability, and programmability, and achieve massive device connectivity and ultra-low end-to-end
thus exploit the available network resources more efficiently latency [14]. Motivated by the features of 5G and B5G, we
[10], [11]. The combination of these technologies can satisfy expect that network virtualization in a VANET environment
the requirements of various applications, such as vehicle exploited by SDN technology can provide multiple advantages
path-planning schemes that have and continue to receive and better performances to make commuting a pleasant
significant attention from the research community [12]. These experience [15]. Moreover, the integration of SDN technology
schemes play a crucial role in reducing traffic congestion, into VANETs both helps to fulfill their management necessities
thereby enhancing the overall economy in terms of energy and enhances their performances in terms of flexibility and
consumption, time, and productivity. Indeed, they consist of programmability [16]. Even more, the SDN technology is
computing the shortest-time path between a pair of locations expected to be a keystone for future intelligentization in a
within a road network while considering several dynamic 6G [17]. In the following, we describe some background
parameters, such as the real-time traffic congestion level, information and terminologies on 5G and its different
unexpected incidents, and road layout. Therefore, designing technologies in the context of vehicular networking and their
a path planning algorithm is regarded as a challenging task assistance by UAVs. Also, we discuss some route planning
because it should consider the specific issues that characterize algorithms to set up the context of the problem.
any urban areas. By looking into the related research works, we only
Motivated by the challenges mentioned above and found a limited number of recent contributions in traffic
considering the advantages of employing such technologies management based on 5G, SDN, or VANET technologies. For
in urban traffic management, we propose an SDN-based instance, Ge et al. [18] proposed a hierarchical SDN-enabled
architecture for vehicle path planning in urban areas. vehicular architecture to address the issue of connectivity
The emerging SDN technology significantly enhances the loss between the controller and switches while achieving the
propagation of messages in VANETs, where a remote server minimum transmission delay between vehicles. Liu et al. [19]
and vehicles play the roles of a central controller (i.e., leveraged the technologies of SDN, Mobile Edge Computing
SDN controller) and switches, respectively. The proposed (MEC), and 5G to create a distributed architecture for urban
architecture consists of a path planning algorithm to allow all traffic management. To quickly assign appropriate channels
kinds of vehicles to dynamically estimate shorter time paths in an SDN-based IoT network, Tang et al. [20] considered
to reach any given destination based on the traffic density, a deep learning method and Partially Overlapping Channel
obstructions and accidents on the roads, and average speed Assignment (POCA) technique to efficiently avoid congestion
of vehicles in a given time-frame. UAVs are leveraged to and predict network traffic. In [21], a 5G-enabled SDN
make several tasks, such as detecting accidents and congestion, architecture is proposed, which provides larger bandwidth and
collecting traffic information, and communicating with 5G BSs higher data rates to enhance the capabilities of VANETs. Garg
to reduce their communication burden (i.e., preventing vehicles et al. [22] aimed to both maintain a certain level of quality of
in range to communicate directly with 5G BSs). All this service (QoS) and ensure secure communication to VANETs
collected information is reported to the SDN controller to build by developing an SDN-based framework for 5G vehicular
an accurate view of traffic in real-time in each road segment. network. In [23], a novel SDN-based architecture for vehicle
We assume that the SDN controller updates each vehicle with communication. It aims to minimize the communication cost
the current road conditions and generates re-planning to look between vehicles.
for alternative shorter time paths in case of congestion or On the side of UAVs, we have witnessed an unexpected
unexpected events blocking the roads. The various experiments spread of UAV-based applications, spanning from monitoring,
conducted to evaluate the performance of this architecture goods delivery, and public services, to communications,
show that it is possible to effectively adjust the travel path, network assistance, and security [29]. Due to shorter network
and thus address the problem of static path planning. coverage in 5G, UAVs can play the role of aerial BSs. For
The remaining of this paper is structured as follows. Section instance, in [30], UAVs are integrated to operate alongside
II provides an overview of the existing SDN-based traffic 5G cellular networks. The work in [31] discussed the
management solutions and route-planning algorithms in urban assistance of UAVs to 5G network supporting lightweight
areas. Section III describes the proposed architecture and its virtualization and network slicing. Sharma et al. [32]
major components. In Section IV, we discuss the experimental designed a novel approach to enhance 5G communication
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXX 2020 3

TABLE I: Feature comparison of path planning schemes.

Features Ref. [24] Ref. [25] Ref. [26] Ref. [12] Ref. [27] Ref. [28] Our application
V2X and
A balance Route planning Leveraging both Dynamic path
Big data analysis driverless
function between with traffic current and Comfort-based planning with
Basic ideology for route technologies for
distance and congestion and future traffic route planning congestion
planning valet parking
time oil consumption status awareness
technology
Traffic view Global Global Global Global Critical areas Local Global
Route planning Deterministic Deterministic Deterministic Deterministic Deterministic
Non-deterministic
√ Non-deterministic
√
Re-routing × × × × × √
Traffic incident awareness × × × × × ×
Promoting Establishing the
Optimizing Demonstrating Selecting the
information Decreasing oil Reducing shortest time
Major advantage parking lot path comfortable fastest route for
sharing between consumption commuting time path in urban
planning routes parking space
Vehicles areas
Does not Neglecting Neglecting The cost of
Neglecting the Space and time
consider Unrealistic drivers’ feedback dynamic factors, aerial and
Major Limitation future congestion complexity of
unexpected gathered data among the such as moving terrestrial BSs’
of selected paths the algorithm
events considered costs obstacles deployment

networks by accurately deploying multiple UAVs that play enhancing the exploitation of existing network infrastructures.
the role of base stations. The authors in [33] proposed a To summarize, Table I provides a comparative study
UAV-assisted backhaul link considering the dynamic blockage between our proposed scheme and the previously proposed
of mmWave links. Different researches have involved vehicles route planning approaches based on several features.
to communicate with different entities based on the concept of
Vehicle-to-Everything (V2X), which paves the path to UAVs III. A N SDN- ENABLED A PPROACH FOR V EHICLE
to be fully integrated with VANETs on the ground [34]. As an PATH -P LANNING (SEARCH)
example, Oubbati et al. in [35] proposed UAV-assisted VANET
Nowadays, driving a vehicle in a big city become almost
for traffic management and emergency vehicle guidance.
unbearable due to the enormous traffic jam. As discussed
There are several route-planning strategies and algorithms
above, various emerging technologies are seen as promising
proposed in the literature. The work of Xiaoxue et al. [24]
solutions in reducing traffic congestion on roads and providing
proposed a path planning strategy based on a tradeoff between
better monitoring of traffic conditions. Therefore, we need
the path length and time balance to carry out optimal path
to design a new architecture based on these technologies
planning. In [25], the authors proposed an edge-based big
to improve emergency responsiveness and save time in
data analysis architecture to gather vehicle information and
transporting people, while considering the access, connectivity,
traffic status to provide the optimal driving path to vehicles.
and analysis features. In this section, we present this
Zhang et al. [26] designed a route-planning algorithm based
architecture and its main components. Moreover, we describe
on real-time data, such as fuel consumption, road patterns,
operation details of this framework, including packet exchange
and feedback of drivers. In [12], a double rewarded value
policy and vehicle traffic calculation. Finally, we discuss a
iteration network (VIN) for route planning and to predict how
specific use case when a vehicle wants to reach a given target
traffic will look like in the future by learning the routing
destination and how our path planning algorithm deals with it.
decisions of experienced drivers. Yu et al. [28] proposed a
route planning strategy to reach any parking spaces with the
shortest time. This strategy can significantly help to minimize A. Network Model
the congestion of the automatic valet parking lot. Based on As depicted in Fig. 1, the overall network architecture is a
the Dijkstra’s algorithm, Li et al. [27] designed a cloud-aided combination of different nodes, such as vehicles, UAVs, and
route planner to detect anomalies on roads, and then consider 5G base stations (5G BSs). The organization of the network
the most comfortable route. For this purpose, three metrics is based on SDN technology. SDN controllers are mainly
are considered during the path planning process: (i) road responsible for managing this technology by collecting data
roughness, (ii) travel time, and (iii) road remove. from all network nodes for building a global traffic information
Most of route planning schemes do not take into account graph of all vehicles on the roads. For this purpose, all
unexpected events occurred on roads, and they just use the vehicles periodically send Hello packets to the SDN controllers
shortest distance or least time cost as the main functionality. through BSs and UAVs that are on the front line of this data
Therefore, vehicles might be blocked at any time, and their collection. The SDN controllers are primarily in charge of
paths cannot be dynamically adjusted when the predetermined updating its neighboring list of each vehicle in the light of
paths become obstructed. In this work, we propose to enhance the latest information received from periodically Hello packets
commuting in urban areas through an SDN-based route sent by vehicles to estimate the number of vehicles and their
planning scheme applied over road networks that are weighted fluidity in each road segment. In particular, the location, speed,
by accurate travel times of road segments. This scheme aims and direction of vehicles are essential traffic information on
to dynamically identify shorter time paths while avoiding roads, which can be estimated by several wireless location
congestion and blocked roads. In addition, our second aim technologies other than GPS in large urban cities where
is to meet a certain level of networking requirements while skyscrapers and other obstructions may block GPS signals.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXX 2020 4

Application plane upper planes through optical fiber links. Generally, the
Networking Systems Services controllers monitor the positions of all vehicles in the
area in order to allow vehicles to plan their shorter time
Connectivity Orchestration Management Path planning Policy Security paths to reach their target destinations quickly.
Control plane
• Application plane: includes several software modules
OpenFlow Switches (i.e., applications) related to different aspects, such
as security applications, traffic efficiency, networking,
OpenFlow Global SDN
Controllers and entertainment. The mode of how to provide
Internet open interfaces to such modules is considered as
Data plane an advantageous factor to make simpler the network
optimization of functions and be up-to-date with the new
technology developments.
• Data plane: consists of the physical devices and
5G BS
(Local SDN
infrastructures in the network. All these physical
UAV
controller) resources play the role of OpenFlow switches, and they
are mainly including vehicles, UAVs, 5G BSs. Each
vehicle can generate various data information, such as
speed, position, previous routes, etc. Equipped with
VANET – UAV required communication modules, vehicles can transmit
communication all these generated data to the controller either through
5G communication
cellular BSs or UAVs.
Wire optical communication
To further clarify the functioning of our SDN-based
Fig. 1: The three-tier architecture of SEARCH.
architecture, we show the key functionalities of each
considered component in Fig. 2.
All this collected information allows vehicles to make smart
and efficient decisions regarding the path planning process. SDN controllers
1) Modes of communications: Our network architecture 2 3
is based on three modes of communications, namely SDN Path planning
Data collection
Model algorithm
Vehicle-to-BS (V2BS), Vehicle-to-UAV (V2U), and
Vehicle-to-Vehicle (V2V). In V2U communication, UAVs can
1 4
communicate directly with vehicles based on the assumption Vehiclei location (xi ,yi) Shorter time paths
Speed in a sub-link Si (m/s)
that they will not be affected by obstructions when they Vehiclei direction (θi , φi) 5
are in range of each other. Also, they can act as collector
nodes gathering periodically all data related to vehicles OpenFlow OpenFlow
wherever it is possible. V2BS are used for control message
dissemination, including also the periodical gathering of
vehicles’ information (i.e., position, speed, and direction)
included in their Hello packets, either from vehicles or from
UAVs. It should be stressed that UAVs communicate with
vehicles within the UAVs’ coverage area, while the BSs
communicate with vehicles that are out of the coverage area
of any UAVs. In other words, UAVs have top priority to
collect traffic information on roads when they are hovering
any areas in order to reduce the network load on 5G BSs.
Furthermore, each BS monitors the traffic information in a
real-time manner and follows each vehicle requesting a path
planning by sending essential updates about the status of the
traffic to all vehicles in its vicinity. In V2V communication,
vehicles can communicate with each other or serve as relays, Fig. 2: Key functionalities and components of SEARCH.
which does not enter the scope of the paper.
2) SDN planes: The network architecture is divided 3) SDN components: The working of the SDN model
into three SDN planes that are assigned specific SDN components is as follows:
responsibilities and rules. We describe each SDN plane below: • SDN Controllers: they belong to the control plane,
• Control plane: is managed by the SDN controllers that and they have control of the overall wired and wireless
define a set of rules with OpenFlow protocol. These networks, including 5G BSs, UAVs, and vehicles.
rules determine how the interconnection units (e.g., 5G Moreover, they are considered as intermediate nodes
BSs, UAVs, and vehicles) transmit the data frames to the between the data and application planes.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXX 2020 5

• 5G BS: it can act as a router and at the same time as an

SDN switch, which is responsible for ensuring the control 𝐣𝟐
𝐞𝟏𝟐 𝟏𝟐
𝐙𝟏
UAV
of different network devices, such as UAVs and vehicles
in a particular area. Besides, it can estimate the traffic 𝟏𝟐
𝐙𝟐
fluidity on road segments that are not covered by UAVs.
• UAVs: they act also as SDN switches and they are in 𝟏𝟐
𝐙𝟑
charge of estimating the traffic fluidity information on L(2,3)
roads and forwarding this information to the closest 5G 𝐙𝟒
𝟏𝟐 5G BS
BSs. N12=8
𝐣𝟏
• Vehicles: they are equipped with two on-board network L(2,3) =30m V2U Communication
devices to support the communication with both UAVs V2BS Communication
and 5G BSs. A sub-link edge

It is worthy to summarize the role of the major components Fig. 3: Travel time estimation.
exploited by SEARCH in its processes in TABLE II.

TABLE II: The role and advantages of 5G, SDN, and UAVs the localization service of vehicles is based on the direction
over SEARCH. of arrival (DoA)/time of arrival (ToA) [36]. This type of
5G BSs UAVs SDN localization aims to realize GPS-free localization based on
Estimating
Path Updating
Traffic gathering shorter time
device-to-device (D2D) communications or 5G antenna. The
planning vehicles accident detection process, the vehicle localization service, and
paths
Warning vehicles the energy consumption of UAVs are our future research works
Accidents & Updating path
before or at Detection
congestion planning and are out of the scope of the present study.
intersections
Estimating
density of
Traffic status Collecting traffic information
vehicles at road C. Functionality of SEARCH
segments
Managing In any path planning strategy, the SDN technology can be
Reduce network considered as the best solution to cover any areas efficiently,
Providing
communication infrastructures
Networking bandwidths and collect, analyze, and transmit crucial information related to
burden over 5G and fully
scalability
BSs exploiting their any events that occurred on roads. Our framework is designed
resources
to allow 5G BSs and UAVs to sense the variation of traffic
status on the surrounding road segments. Also, it enables 5G
BSs to collect all possible data from vehicles and UAVs. All
B. Assumptions gathered data are transmitted to the SDN controllers to allow
The SEARCH framework assumes that the SDN controllers them to both build a global view of the traffic status on the
have full access to the digital map to locate neighboring path roads and quickly establish shorter time paths between any
junctions, as well as the required road information. Each road pairs of locations.
segment is partitioned into equal size zones of 30 m each
(c.f., Fig. 3). This partition is based on the assumption that the Initiate
route
traffic state is considered to be homogeneous, where strict lane planning
Calculate shortest time path
(from current to destination)
discipline is followed by all vehicles. UAVs and 5G BSs are Yes
considered to cover at least one zone where they collect data
Yes Yes Does this
from vehicles and maintain a local view of the traffic state of Reaching
Intersection
Received
impact the
update
their controlled zones. Each zone is supposed to be controlled path

by at least one BS, a UAV, or both. A global view of the overall Vehicle No No No
route Follow
traffic state can be obtained by exchanging the sectional view the same path
of adjacent BSs or UAVs with the SDN controllers.
The connectivity between vehicles and UAVs is ensured Yes Reaching
Destination
via IEEE 802.11p wireless interfaces. Moreover, vehicles and destination

UAVs can also communicate with 5G BSs through appropriate No

Destination
cellular interfaces. The connectivity between 5G BSs and
SDN controllers is ensured through high-speed optical fiber Fig. 4: Flowchart illustrating the functionality of SEARCH.
links. In the case when UAVs are hovering in clear weather,
they can detect incidents on roads using their image capture When a vehicle wants to reach a given destination, the
and processing capabilities. Otherwise, UAVs can exploit SDN controllers are already supposed to have the current
V2U communication to report such events from the involved global vision of all surrounding road segments. Initially, the
terrestrial nodes. It is worthy to note that these two solutions controllers calculate the shortest time path from the current
hold applicable for SEARCH. Since the satellite signals location to the desired destination of the vehicle, according
might be blocked by skyscrapers, SEARCH assumes that to the Dijkstra algorithm. Whenever the vehicle reaches
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXX 2020 6

an intersection, it receives an update related to the traffic Where Tij is the journey time of the whole link eij . tij
k is
conditions. If this update affects the earliest established path, the journey time of each sub-link constituting eij . To calculate
the Dijkstra algorithm is re-applied to calculate a new shortest tij ij ij
k , the journey time of the sub-link (Zk , Zk+1 ) can be
time path. Otherwise, the vehicle continues following the estimated using a combination between the traffic information
same path. This process is repeated until the vehicle reaches gathered by UAVs and 5G BSs and the Macroscopic Model
its desired destination. For more clarification, we summarize [37]. Therefore, we model the relationship between density
these steps in the flowchart depicted in Fig. 4. In the remainder and traffic flow of a sub-link as a continuous fluid model, as
of this section, we first describe the vehicle traffic calculation depicted in Fig. 5.
and then present the application of our SEARCH routing Based on this modeling, we can then calculate the average
planning framework on a concrete example. speed, and thus the journey time in the sub-link based on these
1) Journey time calculation: Vehicle driving time is two parameters. The Macroscopic Model is defined as follows:
considered as an essential criterion of information to define the
∂ ∂
minimum cost routes in the path planning process. Moreover, R(P, t) = g(P, t) (2)
it is also exploited as data to support driver decisions through ∂t ∂P
on-board guidance systems, editable message boards, and Where R is the vehicle density by distance. g is the traffic
radio dissemination. In our case, the road network is modeled flow per time, while P and t are defined as the location and the
as a cyclic graph G = (J, E) to provide flexibility to use current time, respectively. The average speed in the sub-link
well-known techniques in graph theories. J is defined by the can be calculated based on the equation (2) as follows:
set of N nodes or vertices (i.e., junctions), where J = {ji }Ni=1 . g
E is a set of edges between vertices (i.e., directional links S= (3)
R
between junctions), where E = {eij = (ji , jj ), ji , jj ∈ J} Based on the equation (3), the journey time of the sub-link
and eij 6= eji . Since the edges are directed, the traffic in (Zkij , Zk+1
ij
) can be estimated as follows:
both directions could be either dependent or independent,
according to the presence of traffic signals at intersections, L(k,k+1) L(k,k+1) L(k,k+1) × {R(k+1) + R(k) }
where SEARCH operates correctly in both situations (c.f., Fig. tij
k = = ×R =
S g {g(k+1) − g(k) }
3). In Table III, we define a set of parameters that are useful (4)
to describe the model clearly. Where L(k,k+1) is the length of each sub-link (i.e., 30 m)
depending on the length of the road segment where L(k,k+1)
TABLE III: Summary of parameters of SEARCH.
can be sometimes less or more than 30 m. g(k) is the gathered
Parameter Definition traffic at the location P(k) at time t. R(k) is the density of node
at the location P(k) at time t. Therefore, we can calculate the
eij A link from ji to jj . ij
traffic throughout the sub-link (i.e., gathered traffic in Zk+1 )
Dij Distance between ji and jj . based on the following equation (5):
Sij Default limited speed of eij .
Dij ij (nij ij
k+1 − nk ) + X
Tij Journey time of eij , where Tij = Sij
. gk+1 (t) = (5)
t
Zkij Beginning of the kth sub-link in eij . Where X is the total number of vehicles residing in the
Nij Number of zones in eij . sub-link between Zkij and Zk+1 ij
before the traffic gathering.
ij ij
As for nk and nk+1 , they represent the number of vehicles
jk kth junction.
entering and leaving the sub-link between Zkij and Zk+1 ij
,
L(k,k+1) Length between Zkij and Zk+1
ij
(by default 30 m). respectively. Similarly, the density of vehicles in the sub-link
can be computed as follows:

ij (nij ij
k+1 − nk ) + X
P(k+1) P(k) Rk+1 (t) = (6)
L(k+1,k)
g(k+1)(t) S g(k)(t) According to the equation (3), we can deduct the average
R(k+1)(t) R(k)(t) speed in the sub-link between Zkij and Zk+1
ij
as follows:

ij
ij gk+1 (nij ij
k+1 − nk ) + X L(k+1,k)
Fig. 5: Journey time in a sub-link. Sk+1 = ij
= × ij
Rk+1 t (nk+1 − nij
k)+X
(7)
To estimate the full journey time of a given link, we need
Finally, the journey time in the sub-link between Zkij and
to calculate the journey time in each sub-link belonging to the ij
Zk+1 in the link eij can be easily extracted based on the
link eij , such as:
formula (4) as follows:
N
X L(k,k−1)
Tij = tij
k (1) tij
k = (8)
k=1 Skij
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXX 2020 7

2) Case Study: SEARCH route planning: Arriving on time traffic status of vehicles is permanently collected by 5G BSs
is considered as a crucial factor in most path planning and UAVs and forwarded to the SDN controllers. Besides,
strategies. This punctuality raises two key questions: ”When the hovering UAVs in the sky can make a real-time video
should I start the car?” and while on the way ”Which path diagnosis of any events on the roads that can hinder or block
should I follow?”. Indeed, the majority of path planning the traffic. All these events’ information is also gathered by
strategies are deterministic, which can be used to solve SDN controllers for analysis.
the problem. Nevertheless, practically, the majority of these
As illustrated in Fig. 6(a), since there are no UAVs in
strategies focus on minimizing the journey distance and
range, the red vehicle requests the closest 5G BS to get the
time well before starting the travel without considering
shortest time path towards the junction j4 (i.e., the target
unexpected events or congestion that would be happened on
destination). Once the request is fulfilled, the red vehicle starts
the road. Consequently, the proposed framework is based on a
its journey through the path j1 → j2 → j3 → j4 (c.f., Fig.
non-deterministic approach to change this impasse, as shown
6(b)). However, the journey will not last long, until a crucial
in Fig. 6. With the assistance of a low-latency SDN-based
update impacting the original path is intercepted at junction
heterogeneous network, the novel path planning scheme would
j2 . This update informs the red vehicle that the segment
be enhanced by two significant measures: (i) Congestion and
between j2 and j3 becomes congested. This requires initiating
accident detection and (ii) shorter time paths calculation.
the path planning calculation from the current junction j2 .
Congestion and incident detection: In the process of This calculation results in obtaining an alternative shorter
traditional path planning, when a vehicle requests to reach a time journey through the path j2 → j6 → j7 → j8 → j4 .
target destination, the centralized architecture tries to compute At the junction j7 , the red vehicle receives another crucial
the full path based on one or several features, such as the update from the closest BS, which represents an accident
distance, the journey time, the oil consumption, etc. However, that occurred between j7 and j8 (see Fig. 6(d)). Therefore,
the major limitation of such architectures is that there is no re-initiating the calculation of the shortest time path from j7
continual and real-time assessment of the variable nature of is unavoidable. Another journey is obtained from the current
traffic congestion on the roads and the unexpected incidents junction j7 through the path j7 → j3 → j4 , until reaching its
that may occur during the journey of the vehicle to avoid an target destination. This process is summarized in TABLE IV
indefinite duration of the trip. In the SEARCH framework, the in which there are the estimated journey times for each path.

j8 j7 j6 j5 j8 j7 j6 j5

Turn left
j2→j3→ j4
Dest. Dest.
j3 j2 j1 j3 j2 j1
j4 j4
Fastest route
to → j4

(a) Phase 1. (b) Phase 2.

Turn left
j3→j4

Accident
j8 j7 j6 j5 j8 J7 j7→j8 j6 j5
Accident

Turn right
j6→j7→j8→j4

Dest. Dest.
j3 j2 j1 j3 j2 j1
j4 Congestion j4
j2→j3

(c) Phase 3. (d) Phase 4.

Fig. 6: A map processing test of SEARCH.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXX 2020 8

TABLE IV: Processing and Functionality of SEARCH.

Estimated journey time (Dijkstra Update impacting the
Source Destination Journey path(s) Cause of the update
cost) from the current junction path
j1 → j2 → j3 → j4 115 s Occurred at j2 Congestion
. . . → j2 → j6 → j7 → j8 → j4 100 s Occurred at j7 Accident
j1 j4
. . . → j7 → j3 → j4 60 s Not occurred —
The full path: j1 → j2 → j6 → j7 → j3 → j4 Total journey time=160 s

Dynamic Shorter time paths calculation: The SDN estimated at 60 seconds to finally reach the target destination.
controllers permanently calculate the journey times in each In the end, the red vehicle has crossed the full path towards j4
lane of road segments. After that, the journey times are shared in 160 seconds, which is the shortest time while considering
with requesting vehicles. The Dijkstra algorithm is applied to the variation of traffic and unexpected events occurred on the
calculate shorter time paths. For this purpose, this algorithm is roads. Algorithm 1 shows the process of congestion avoidance
enhanced to consider only the influence of journey time in each using SEARCH. Moreover, it also illustrates the procedure of
link and to find the optimal path. Consequently, the optimal calculating new routes when a crucial update is received. This
path costs the least time, where all weights (i.e., journey times) process is permanently repeated to update vehicles within each
must be non-negative. intersection through UAVs or BSs.

30 s 30 s 30 s Algorithm 1: Congestion avoidance using SEARCH

j8 j7 j6 j5
30 s 30 s 30 s Input:
Phase 2 30 s 30 s 40 s 30 s 30 s 70 s 30 s 30 s C = {ci }N
i=1 ; // A set of vehicles
40 s 45 s 30 s H: Hello packet between UAVs, 5G BSs, and vehicles.
j4 j3 j2 j1
40 s 40 s 30 s K = {eij }Ni,j=1 ; // Roads with attributes
30 s 30 s 30 s M =∅; // A set of congested roads
j8 j7 j6 j5
30 s 40 s 30 s
Output:
Phase 3 30 s 30 s 40 s 40 s 30 s 40 s 30 s 30 s
R: Shortest time route.
1 At each UAVs and 5G BSs
50 s 100 s 30 s
j4 j3 j2 j1 foreach eij ∈ G do
40 s 30 s 30 s
+∞ 30 s 30 s Collect(ci .location) ; // Vehicles’
j8 j7 j6 j5
+∞ 40 s 30 s
positions
30 s 30 s
Collect({Snij }N
n=1 ) ; // speed in each
Phase 4 30 s 30 s 40 s 30 s 30 s 30 s
sub-link
30 s 60 s 50 s
j4 j3 j2 j1 Update(eij ,{Snij }N N
n=1 ,{ci }i=1 ∈ eij )
30 s 30 s 30 s
2 Congestion roads detection
Fig. 7: Weighted directed graphs of Fig. 6. foreach eij ∈ K do
Calculate(S ij ) ; // Average speed in eij
In our case, our framework transforms the map shown in if S ij ≤ 0.2 then
Fig. 6 into a directed weighted graph consisting of nodes and M ∪ {eij } ; // Add eij into M
edges, as illustrated in Fig. 7. Given that we have considered
a map with the same length of road segments, we distinguish 3 At each intersection
that all empty lanes take at most 30 seconds to be crossed. foreach ci ∈ C do
The other lanes can be crossed, depending on the circulation of route ← Extract(ci .B) ; // Extract the
vehicles inside them. At each update impacting the established vehicle route from Hello packets
path, a new calculation of the shortest time path is made. foreach eij ∈ route do
For instance, the red vehicle has carried out two different if eij ∈ M then
updates at the junctions j2 and j7 to avoid congested or K − {eij } ; // Exclude the
blocked road segments. Indeed, at the first step, the red vehicle congested road from K
computes the journey time at j1 , which is estimated at 115 s. Extract(ci .location) ; // Get the
However, after traveling 30 seconds, the vehicle reaches the current location of ci
junction j2 , in which it receives a crucial update impacting the foreach UAV and BS do
current path. Therefore, the vehicle re-calculates the shortest Update(K, {eij , Tij }N
i,j=1 ) ; // Update
journey time at j2 to find a new shorter time path through road attributes
j2 → j6 → j7 → j8 → j4 , which is estimated at 100 route ← Shortest time(K) ; // Shortest
seconds instead of taking the previous path with an estimation time path
of 150 seconds. After running 100 seconds, the red vehicle Send(route,ci ) ; // Send the newly
reaches the junction j7 to find the road blocked in front of calculated route to the vehicle
it (i.e., between j7 and j8 ). Therefore, a new shortest time
alternative path is found through j7 → j3 → j4 , which is
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXX 2020 9

IV. P ERFORMANCE E VALUATION to traditional distance-based and time-based path planning

To evaluate the effectiveness of our SEARCH framework, mechanisms that calculate the full path to be traveled before
we conduct a set of experiments. We considered OMNeT++ the beginning of the journey. We also evaluate the performance
4.3 [38], which is complemented by SUMO [39] and MobiSim of our SDN-based architecture under different metrics and
[40] as two mobility generators producing the random motions compare it with both the architecture presented in [18] and
of vehicles and UAVs, respectively. At a first step, we import a Non-SDN-based architecture. Table V summarizes the main
test urban area from OpenStreetMap [41], which is known for simulation parameters used in our experiments.
its busy streets of vehicles (see Fig. 8). Secondly, we extract
TABLE V: Simulation parameters to evaluate SEARCH.
the road layout in the form of directed edges and nodes (i.e.,
blue lines and red circles in Fig. 8(b), respectively), we define Parameters Values
the number of vehicles and UAVs for each scenario, and we Simulation time 900 (s)
select a number of accidents in each scenario of simulation. Area size 2 × 2 km2
To summarize, in this section, we first present the simulation Normal traffic density ≈ 80 vehicles/km2
model and the different considered parameters. Finally, we Rush-hour traffic density ≈ 480 vehicles/km2
Number of incidents [5, 30]
provide a clear interpretation of the obtained results. Number of UAVs [16, 144]
Number of 5G BSs 1 BS/km2
Vehicle speed smax 14 m/s
A. Parameter Settings Channel frequency 5.890e9 Hz
Transmit power 21.5 dBm
To evaluate the performances of our framework, we place Sensitivity -81.5 dBm
a sufficient number of UAVs and 5G BSs such that we Path loss model Free-space
ensure a full coverage along the roads. This full-coverage PHY/MAC model IEEE 802.11p
Bit rate 18 Mbit/s
allows us to get a global view of vehicular network topology Coverage of UAVs and vehicles ≈ 300 m
by collecting the movement information of all vehicles. For Hello interval 0.1 (s)
the vehicular network, we consider IEEE 802.11p wireless Packet size 1 KB
interfaces for each vehicle and UAV using the framework
Veins [42]. Also, we deploy up to 400 vehicles/km2 and
144 UAVs, which are randomly distributed to simulate traffic
and to capture any incidents on the roads. We randomly B. Results Analysis
select a pair of starting and ending points located at different To evaluate the performance of our framework and the
intersections (i.e., red circles), as shown in Fig. 8(b). It is other path planning algorithms, we first calculate the total
worthy to note that each point in the results represents the journey time of the vehicle to reach the target destination
mean of 10 simulation runs with a 95% confidence interval. for each applied path planning algorithm (c.f., Fig. 9).
The performance of our path planning algorithm is compared As depicted in Fig. 9(a), our framework portrays an

ne
rg 18 70
n
ngass

as
Bah

19 s
24 e Sternwa
Jelmoli
e

Urania
e

58
ss

26
tra

St 4 Oeten
ias

e in Rennweg
m 6 15a
Sihlstrasse
an

üh
e
ss

1
Ur

16 le ga
26 1 15
tz
tra

Bahnhofstrasse

ss
e S t einmühlepl a
ns

9 3 Arriving
we

28
20
Lö

53
point
Re
St.

42
32
nn

31
Ann

18a
we

34 62
se 63 3
aga

39
g

33 26 21
60
sse

34 Rennweg
Hiltl Club
6 31
e
as s Füsslistrasse
lstr
38 S ih
31 1
hof 33 Rennweg e
Coop City lg ass 23
tte
St. Annahof Ku 1
St

7
ss e
.

Glockenhof Food
An

e 15
ass St. Anna- Zum
n

hlstr
ag

Kapelle ss Trottba
a

e
35 44a 36
4
48 50 30a
0 29
30 strasse 48 42
Pelikan
N

34
üs

31 6 Augu
ch

Pavillon-
st inerg
e

2
le

ass
rs

Skulptur e
ck

46
tr

e
er

ss 21 17
Bahnenhof
as

City

Petra Gut
Lind
se

42 Münzp
ra

24
st

hofst

6
an

42 15 10 44
lik

Sihlstrasse 22 UBS
Pe

UBS
rass

11 5
Kaufleuten
e

42 Augustiner
tz
Ta

se zpla
M ün
l
ac

s
ra
ke

Pelikanplatz t 10
r

a ns 39 40 3
lik
N

Bahn
üs

Pe 41
ch

37
el

hofst
er

15 9 6
s tr
as

rasse
se
Ta

7 43 a
2 terstr
l ac

14
30 St. Pe
ke

asse
se

16
terstr
r
as

Pelikanplatz St. Pe
tr
s
an

Starting
lik

29a se 17
as
Pe

29
24 str
er Credit point
et
.P Suisse
Ta 64 St
Bahn

l
Ta

st
ra
la
se

ss
(b) Road layout.
ck

se
s

62 21
hofst
ra

e as
e

Peterho
r
st

ng 27
16 re
rass

Ba 61 Bä
st 58 23
ei
e

p la

(a) Generated map.

Fig. 8: Map of the simulation area in Zurich, Switzerland (47°22’24.7”N 8°32’19.1”E).
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXX 2020 10

320 1000 1200

Distance-based path planning 900

Distance-based path planning 1100 Distance-based path planning
300
Time-based path planning Time-based path planning Time-based path planning
1000
800
SEARCH Framework SEARCH Framework SEARCH Framework
280
900
Total time (s)

Total time (s)

Total time (s)

700

260 800
600
700
240 500
600

220 400
500

300
200 400

200
300
180
100 200

160
50 100 150 200 250 300 350 400 450 500 0 100
0 50 100 150 5 10 15 20 25 30
2
Number of vehicles/km Number of UAVs Number of incidents
(a) Total time vs. Density of vehicles (b) Total time vs. Density of UAVs (c) Total time vs. Number of incidents
(UAV density=16, Number of incidents=5). (Vehicle density=280, Number of incidents=5). (Vehicle density=280, UAV density=16).

Fig. 9: Simulation results of journey time of SEARCH.

outperforming performance under different vehicle densities. incidents. As a result, we obtain three graphs represented in
Indeed, compared to other algorithms, SEARCH can decrease Fig. 10. As shown in Fig. 10(a), SEARCH achieves a low
the journey time by more than 20%. This minimization is traveled distance compared to time-based path planning. This
due to its dynamic nature to calculate shorter time paths is because SEARCH establishes congestion aware paths, thus
while being aware of both incidents and traffic congestion avoiding traffic jams whenever it is possible and occasionally
on the roads. However, for the other algorithms, the average reducing distances. However, it is not the case for time-based
time journey tends to be increased as the density of vehicles path planning that initially selects the full path to the target
increases. This can be explained by the continuous selection of destination, which should avoid all congested and blocked
the initial path, thus exposed to traffic congestion and incidents roads, and thus occasionally results in some peaks of traveled
during the vehicle journey. In Fig. 9(b), we distinguish that distances during the vehicle journey. It is worthy to note that
SEARCH achieves the least journey time as the density even if we expect a proportional behavior of traveled distances
of UAVs increases. This is because many road segments according to the density of vehicles, but due to the time-based
are controlled by the increasing number of UAVs and criteria of selecting paths in both SEARCH and time-based
generate alerts to the SDN controllers at each detected approach, Fig. 10(a) shows unstable behavior of traveled
incident. Nevertheless, the results of the other algorithms distances. As for Fig. 10(b), decrease of traveled distance
are not affected by the variation of UAV density since their is remarkable, which is explained by the quick detection of
functioning is based only on the density of vehicles and incidents as the density of UAVs increases. However, the
the length of paths. Fig. 9(c) reveals that with only 16 distance-based path planning algorithm remains stable since
UAVs, SEARCH can enhance the vehicle journey time by it establishes static paths (i.e., shortest paths) beforehand
detecting most of the incidents and effectively regulate the while neglecting unexpected incidents on the roads. Fig. 10(c)
vehicle journey. In essence, compared to the other algorithms, shows that the distance traveled in SEARCH is less than the
SEARCH allows the vehicle to avoid the traffic jam that is time-based path planning algorithm. This is caused by the
formed around or on the way to the accident. continuous collecting of information related to the roads by
UAVs and 5G BSs, which can often find shorter distance paths
To compare the distance traveled for all the evaluated when incidents suddenly happen.
algorithms, we perform different experiments under different
densities of vehicles and UAVs while varying the number of To test the performance of SEARCH in terms of path

2800 3800
Total distance traveled (m)

Distance-based path planning Distance-based path planning

Total distance traveled (m)

Distance-based path planning

Total distance traveled (m)

3200 3600
Time-based path planning 2700
Time-based path planning Time-based path planning
SEARCH Framework SEARCH Framework 3400 SEARCH Framework
3000 2600

3200

2800 2500

3000

2400
2600 2800

2300
2600
2400

2200
2400
2200
2100 2200

2000
50 100 150 200 250 300 350 400 450 500 2000 2000
0 50 100 150 5 10 15 20 25 30

Number of vehicles/km2 Number of UAVs Number of incidents

(a) Total distance vs. Density of vehicles (b) Total distance vs. Density of UAVs (c) Total distance vs. Number of incidents
(UAV density=16, Number of incidents=5). (Vehicle density=280, Number of incidents=5). (Vehicle density=280, UAV density=16).
Fig. 10: Simulation results of traveled distance of SEARCH.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXX 2020 11

2700 190 195

2600
SEARCH Framework SEARCH Framework 190
185
Estimated distance (m)
185

Estimated time (s)

2500

Estimated time (s)

180
180
2400

175
175
2300
170
2200 170
165

2100
1st measurement
165 160
2nd measurement
2000
160
155 3rd measurement
1900 150
4th measurement
155 5th measurement
1800 145
6th measurement
1700 150 140
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180

Updating time (s) Updating time (s) Updating time (s)

(a) Estimated distance vs. Updating time (b) Estimated time vs. Updating time (c) Comparison of six time measurements
(Vehicle density=280, UAV density=16, Number (Vehicle density=280, UAV density=16, Number (Vehicle density=280, UAV density=16, Number
of incidents=5). of incidents=5). of incidents=5).
Fig. 11: Performance of path calculation dynamicity.

calculation dynamicity, we conduct different tests on the of vehicles with UAVs while avoiding as much as possible 5G
distance and vehicle travel time (c.f., Fig. 11). In fact, Fig. BSs. However, in SD-VANET, there are only specific vehicles
11(a) shows the estimated distance to be traveled from the that can establish a direct connection with BSs.
source to the destination at each crucial update. We distinguish
that the distance is variable according to the traffic in the As in Fig. 12(b), it is clearly shown that the average latency
path. For example, the path is updated at times 30 and 60 achieved by SEARCH is significantly less than those produced
seconds to be decreased until stabilization at time 170 and by the traditional approach and SD-VANET. The reason is, in
180 seconds. As illustrated in Fig. 11(b), we calculate the our proposed SEARCH architecture, the control functionality
time needed from the source to the destination at each update, is divided among BSs and UAVs, thus reducing latency.
impacting the already established path. We notice that the time Moreover, vehicles communicate directly with BSs and UAVs
to reach the destination at 0.00 seconds is estimated at 165 without V2V communications. Nevertheless, SD-VANET
seconds. The time varies according to the path changes until it produces high latency compared to SEARCH with the increase
becomes stable. The stability is explained by not receiving any of density of vehicles, which is explained by the complex
updates. Fig. 11(c) shows a comparison between the estimated handovers carried out between vehicles.
time of six path measurements. The estimated time of the first
measurement is not considered as the most suitable path since Fig. 12(c) shows that the control overhead has a proportional
other paths, including the shortest path in terms of distance, behavior according to the density of vehicles in all strategies.
are obstructed at that time. When the vehicle makes the second This is explained by the fact that control messages are
update at the second intersection, it turns out that the second periodically transmitted by vehicles to 5G BSs, which are
calculated path has the lowest estimated time. The vehicle automatically increased when the density of vehicles increases.
continues receiving updates at each intersection until reaching Nevertheless, the different mechanisms applied by these
the last intersection, in which the vehicle makes the sixth approaches differentiate the control overhead levels between
measurement to find out that the estimated time to destination them. For instance, the traditional approach allows all vehicles
is close to the second measurement (i.e., the lowest journey to transmit their periodical messages to 5G BSs, which
time). significantly deteriorates its performance. SD-VANET [18]
has enhanced the performance of the traditional approach by
The performance of the SEARCH framework architecture
selecting only the closest vehicles to make the transmission.
is analyzed and compared with two different architectures.
As for SEARCH, vehicles tend to transmit as much as possible
The first architecture is a traditional approach in which
their periodical messages to existing UAVs in the sky, which
every vehicle establishes communication directly with the
significantly decreases the number of nodes that communicate
BS without considering any SDN controllers. The second
with BSs, and thus reducing the overhead to their lowest level.
architecture is based on the method proposed in [18], named
SD-VANET, in which every vehicle has to send signaling In Fig. 12(d), to accurately define the network utilization
data to the closest vehicle to the BS. The performance rate, we calculate the controller processing rate and the ratio
of SEARCH architecture is investigated by analyzing the of the total data arrival rate. It can be seen clearly that when
throughput, latency, control overhead, and processing latency. the network load is low enough, processing latency tends
It should be stressed that UAVs and incidents are neglected to be close to zero for all the evaluated architectures. With
during the evaluation of the mentioned above metrics. the increase of the network utilization ratio, SEARCH still
As shown in Fig. 12(a), it is clear that SEARCH architecture keeps the processing latency under 1 ms most of the time.
significantly improves the throughput compared to other Consequently, we can conclude that SEARCH meets the 5G
architectures. This improvement is explained by the provided latency requirement, which is not the case of SD-VANET and
high-data rates and direct communications of a large number the traditional approach.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXX 2020 12

1000 25

Traditional approach Traditional approach

900
SD-VANET [18] SD-VANET [18]
SEARCH Framework SEARCH Framework
800 20

700
Throughput (Mb/s)

Latency (ms)
600 15

500

400 10

300

200 5

100

0 0
50 100 150 200 250 300 350 400 450 500 50 100 150 200 250 300 350 400 450 500

Number of vehicles/km 2
Number of vehicles/km 2

(a) Throughput vs. Density of vehicles. (b) Latency vs. Density of vehicles.
550 60

Traditional approach Traditional approach

500
SD-VANET [18] SD-VANET [18]
450 SEARCH Framework 50 SEARCH Framework
Control overhead (Nb. of packets)

400 Processing latency (ms)

40
350

300
30
250

200
20
150

100
10

50

0 0
50 100 150 200 250 300 350 400 450 500 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
2 Network Utilization
Number of vehicles/km

(c) Control overhead vs. Density of vehicles. (d) Processing latency vs. Network utilization.

Fig. 12: Simulation results of our path planning algorithm.

V. C ONCLUSION front of a vehicle moving in a road segment. In the future,

we will strongly seek to extend SEARCH with an enhanced
This paper proposes a framework for dynamic path planning network architecture by reducing the number of deployed
based on a low-latency SDN-based heterogeneous network infrastructures and replacing them with mobile entities, such
called SEARCH. Due to the variable nature of congestion as UAV-BSs. It will then be promising for us to enhance the
level and random incidents, SEARCH can dynamically adjust path planning strategy based on predictive models and artificial
the best path of a vehicle during its journey. This solution intelligence (AI) techniques.
can significantly improve the problem of static path planning
by continuously collecting traffic information using 5G BSs
ACKNOWLEDGMENT
and UAVs deployed along roads. We exploit the Macroscopic
model to calculate the journey time in each road segment and The authors would like to thank the Deanship of Scientific
we use the Dijkstra algorithm to look for shorter time paths Research at Umm Al-Qura University for supporting this work
towards target destinations. Experimental analysis shows the by grant code 18-COM-1-01-0001.
robustness of the proposed framework in terms of decreasing
journey times associated with the different road constraints R EFERENCES
in urban areas. However, we are aware of a certain number
of limitations of our scheme, such as the deployment cost [1] L. C. Bento, R. Parafita, H. A. Rakha, and U. J. Nunes, “A study of
the environmental impacts of intelligent automated vehicle control at
of aerial and terrestrial network equipment and the failure intersections via V2V and V2I communications,” Journal of Intelligent
of our path planning algorithm when an accident occurs in Transportation Systems, vol. 23, no. 1, pp. 41–59, 2019.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXX 2020 13

[2] H. Teng, W. Liu, T. Wang, A. Liu, X. Liu, and S. Zhang, “A [23] Z. He, J. Cao, and X. Liu, “SDVN: Enabling rapid network innovation
cost-efficient greedy code dissemination scheme through vehicle to for heterogeneous vehicular communication,” IEEE network, vol. 30,
sensing devices (V2SD) communication in smart city,” IEEE Access, no. 4, pp. 10–15, 2016.
vol. 7, pp. 16 675–16 694, 2019. [24] G. Xiaoxue et al., “Optimal route planning of parking lot based on
[3] L. J. Basso, F. Feres, and H. E. Silva, “The efficiency of bus rapid dijkstra algorithm,” in roceedings of the International Conference on
transit (BRT) systems: A dynamic congestion approach,” Transportation Robots & Intelligent System (ICRIS). IEEE, 2017, pp. 221–224.
Research Part B: Methodological, vol. 127, pp. 47–71, 2019. [25] C. Zhao, M. Dong, K. Ota, J. Li, and J. Wu, “Edge-MapReduce-based
[4] I.-S. Comşa, S. Zhang, M. E. Aydin, P. Kuonen, Y. Lu, R. Trestian, and intelligent information-centric IoV: Cognitive route planning,” IEEE
G. Ghinea, “Towards 5G: A reinforcement learning-based scheduling Access, vol. 7, pp. 50 549–50 560, 2019.
solution for data traffic management,” IEEE Transactions on Network [26] J.-d. Zhang, Y.-j. Feng, F.-f. Shi, G. Wang, B. Ma, R.-s. Li, and X.-y.
and Service Management, vol. 15, no. 4, pp. 1661–1675, 2018. Jia, “Vehicle routing in urban areas based on the oil consumption
[5] T. Yamazato, N. Kawagita, H. Okada, T. Fujii, T. Yendo, S. Arai, weight-Dijkstra algorithm,” IET Intelligent Transport Systems, vol. 10,
and K. Kamakura, “The uplink visible light communication beacon no. 7, pp. 495–502, 2016.
system for universal traffic management,” IEEE Access, vol. 5, pp. [27] Z. Li, I. V. Kolmanovsky, E. M. Atkins, J. Lu, D. P. Filev, and
22 282–22 290, 2017. Y. Bai, “Road disturbance estimation and cloud-aided comfort-based
[6] I. Bisio, C. Garibotto, A. Grattarola, F. Lavagetto, and A. Sciarrone, route planning,” IEEE transactions on cybernetics, vol. 47, no. 11, pp.
“Smart and robust speaker recognition for context-aware in-vehicle 3879–3891, 2016.
applications,” IEEE Transactions on Vehicular Technology, vol. 67, [28] L. Yu, H. Jiang, and L. Hua, “Anti-Congestion Route Planning Scheme
no. 9, pp. 8808–8821, 2018. Based on Dijkstra Algorithm for Automatic Valet Parking System,”
[7] M. Lourenço, F. S. Souza, C. M. Silva, R. I. Meneguette, and Applied Sciences, vol. 9, no. 23, p. 5016, 2019.
D. L. Guidoni, “A Hybrid V2I and V2V Approach for Urban [29] I. Bisio, C. Garibotto, F. Lavagetto, A. Sciarrone, and S. Zappatore,
Traffic Management in Vehicular Networks,” in Proceedings of the “Blind detection: Advanced techniques for WiFi-based drone
Latin-American Conference on Communications (LATINCOM). IEEE, surveillance,” IEEE Transactions on Vehicular Technology, vol. 68,
2019, pp. 1–6. no. 1, pp. 938–946, 2018.
[8] O. S. Oubbati, N. Chaib, A. Lakas, P. Lorenz, and A. Rachedi, [30] W. Shi, H. Zhou, J. Li, W. Xu, N. Zhang, and X. Shen, “Drone assisted
“UAV-assisted supporting services connectivity in urban VANETs,” vehicular networks: Architecture, challenges and opportunities,” IEEE
IEEE Transactions on Vehicular Technology, vol. 68, no. 4, pp. Network, vol. 32, no. 3, pp. 130–137, 2018.
3944–3951, 2019. [31] G. K. Xilouris, M. C. Batistatos, G. E. Athanasiadou, G. Tsoulos, H. B.
[9] O. S. Oubbati, M. Atiquzzaman, P. Lorenz, M. H. Tareque, and M. S. Pervaiz, and C. C. Zarakovitis, “UAV-assisted 5G network architecture
Hossain, “Routing in flying Ad Hoc networks: Survey, constraints, and with slicing and virtualization,” in Proceedings of the IEEE Globecom
future challenge perspectives,” IEEE Access, vol. 7, pp. 81 057–81 105, Workshops (GC Wkshps). IEEE, 2018, pp. 1–7.
2019. [32] V. Sharma, K. Srinivasan, H.-C. Chao, K.-L. Hua, and W.-H. Cheng,
[10] O. S. Oubbati, M. Atiquzzaman, T. A. Ahanger, and A. Ibrahim, “Intelligent deployment of UAVs in 5G heterogeneous communication
“Softwarization of UAV Networks: A Survey of Applications and Future environment for improved coverage,” Journal of Network and Computer
Trends,” IEEE Access, vol. 8, pp. 98 073–98 125, 2020. Applications, vol. 85, pp. 94–105, 2017.
[33] M. Gapeyenko, V. Petrov, D. Moltchanov, S. Andreev, N. Himayat, and
[11] A. Rego, L. Garcia, S. Sendra, and J. Lloret, “Software Defined
Y. Koucheryavy, “Flexible and reliable UAV-assisted backhaul operation
Network-based control system for an efficient traffic management for
in 5G mmWave cellular networks,” IEEE Journal on Selected Areas in
emergency situations in smart cities,” Future Generation Computer
Communications, vol. 36, no. 11, pp. 2486–2496, 2018.
Systems, vol. 88, pp. 243–253, 2018.
[34] S. Chen, J. Hu, Y. Shi, Y. Peng, J. Fang, R. Zhao, and L. Zhao,
[12] J. Li, D. Fu, Q. Yuan, H. Zhang, K. Chen, S. Yang, and F. Yang, “A
“Vehicle-to-everything (V2X) services supported by LTE-based systems
traffic prediction enabled double rewarded value iteration network for
and 5G,” IEEE Communications Standards Magazine, vol. 1, no. 2, pp.
route planning,” IEEE Transactions on Vehicular Technology, vol. 68,
70–76, 2017.
no. 5, pp. 4170–4181, 2019.
[35] O. S. Oubbati, A. Lakas, P. Lorenz, M. Atiquzzaman, and A. Jamalipour,
[13] X. Foukas, G. Patounas, A. Elmokashfi, and M. K. Marina, “Network
“Leveraging Communicating UAVs for Emergency Vehicle Guidance in
slicing in 5G: Survey and challenges,” IEEE Communications Magazine,
Urban Areas,” IEEE Transactions on Emerging Topics in Computing,
vol. 55, no. 5, pp. 94–100, 2017.
2019.
[14] J. Bhatia, Y. Modi, S. Tanwar, and M. Bhavsar, “Software defined [36] R. Zhang, F. Yan, L. Shen, and Y. Wu, “A vehicle positioning method
vehicular networks: A comprehensive review,” International Journal of based on joint TOA and DOA estimation with V2R communications,”
Communication Systems, vol. 32, no. 12, p. e4005, 2019. in Proceedings of the IEEE 85th Vehicular Technology Conference (VTC
[15] R. Hussain, F. Hussain, and S. Zeadally, “Integration of VANET and Spring). IEEE, 2017, pp. 1–5.
5G Security: A review of design and implementation issues,” Future [37] M. Van den Berg, A. Hegyi, B. De Schutter, and J. Hellendoorn, “A
Generation Computer Systems, vol. 101, pp. 843–864, 2019. macroscopic traffic flow model for integrated control of freeway and
[16] I. Yaqoob, I. Ahmad, E. Ahmed, A. Gani, M. Imran, and urban traffic networks,” in Proceedings of the 42nd IEEE International
N. Guizani, “Overcoming the key challenges to establishing vehicular Conference on Decision and Control (IEEE Cat. No. 03CH37475),
communication: Is SDN the answer?” IEEE Communications Magazine, vol. 3. IEEE, 2003, pp. 2774–2779.
vol. 55, no. 7, pp. 128–134, 2017. [38] A. Varga and R. Hornig, “An overview of the OMNeT++ simulation
[17] F. Tang, Y. Kawamoto, N. Kato, and J. Liu, “Future intelligent and environment,” in Proceedings of the 1st international conference on
secure vehicular network toward 6G: Machine-learning approaches,” Simulation tools and techniques for communications, networks and
Proceedings of the IEEE, vol. 108, no. 2, pp. 292–307, 2019. systems & workshops. ICST (Institute for Computer Sciences,
[18] X. Ge, Z. Li, and S. Li, “5G software defined vehicular networks,” IEEE Social-Informatics and . . . , 2008, p. 60.
Communications Magazine, vol. 55, no. 7, pp. 87–93, 2017. [39] M. Behrisch, L. Bieker, J. Erdmann, and D. Krajzewicz,
[19] J. Liu, J. Wan, D. Jia, B. Zeng, D. Li, C.-H. Hsu, and H. Chen, “SUMO–simulation of urban mobility: an overview,” in Proceedings of
“High-efficiency urban-traffic management in context-aware computing the Third International Conference on Advances in System Simulation
and 5G communication,” IEEE Communications Magazine, vol. 55, (SIMUL 2011), 2011.
no. 1, pp. 34–40, 2017. [40] S. M. Mousavi, H. R. Rabiee, M. Moshref, and A. Dabirmoghaddam,
[20] F. Tang, Z. M. Fadlullah, B. Mao, and N. Kato, “An intelligent “Mobisim: A framework for simulation of mobility models in mobile
traffic load prediction-based adaptive channel assignment algorithm in ad-hoc networks,” in Proceedings of the third IEEE International
SDN-IoT: A deep learning approach,” IEEE Internet of Things Journal, Conference on Wireless and Mobile Computing, Networking and
vol. 5, no. 6, pp. 5141–5154, 2018. Communications (WiMOB). IEEE, 2007, pp. 82–82.
[21] S. Din, A. Paul, and A. Rehman, “5G-enabled Hierarchical architecture [41] M. Haklay and P. Weber, “Openstreetmap: User-generated street maps,”
for software-defined intelligent transportation system,” Computer Ieee Pervas Comput, vol. 7, no. 4, pp. 12–18, 2008.
Networks, vol. 150, pp. 81–89, 2019. [42] C. Sommer, R. German, and F. Dressler, “Bidirectionally coupled
[22] S. Garg, K. Kaur, G. Kaddoum, S. H. Ahmed, and D. N. K. network and road traffic simulation for improved IVC analysis,” IEEE
Jayakody, “SDN-based secure and privacy-preserving scheme for Transactions on mobile computing, vol. 10, no. 1, pp. 3–15, 2010.
vehicular networks: A 5G perspective,” IEEE Transactions on Vehicular
Technology, vol. 68, no. 9, pp. 8421–8434, 2019.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. XX, NO. XX, XXX 2020 14

Omar Sami Oubbati is an Associate Professor at Pascal Lorenz received his [Link]. (1990) and
the Electronics department, University of Laghouat, Ph.D. (1994) from the University of Nancy, France.
Algeria and a Research Assistant in the Computer Between 1990 and 1995 he was a research engineer
Science and Mathematics Lab (LIM) at the same at WorldFIP Europe and at Alcatel-Alsthom. He
university. He received his degree of Engineer is a professor at the University of Haute-Alsace,
(2010), [Link]. in Computer Engineering (2011), France, since 1995. His research interests include
[Link]. degree (2014), and a PhD in Computer QoS, wireless networks and high-speed networks.
Science (2018). From Oct. 2016 to Oct. 2017, he was He is the author/co-author of 3 books, 3 patents and
a Visiting Student with the Laboratory of Computer 200 international publications in refereed journals
Science, University of Avignon, France. His main and conferences. He was Technical Editor of the
research interests are in Flying and Vehicular ad hoc IEEE Communications Magazine Editorial Board
networks, Energy harvesting and Mobile Edge Computing, Energy efficiency (2000-2006), IEEE Networks Magazine since 2015, IEEE Transactions
and Internet of Things (IoT). He is the recipient of the 2019 Best Survey Paper on Vehicular Technology since 2017, Chair of IEEE ComSoc France
for Vehicular Communications (Elsevier). He has actively served as a reviewer (2014-2018), Financial chair of IEEE France (2017-2019), Chair of
for flagship IEEE Transactions journals and conferences, and participated Vertical Issues in Communication Systems Technical Committee Cluster
as a Technical Program Committee Member for a variety of international (2008-2009), Chair of the Communications Systems Integration and Modeling
conferences, such as IEEE ICC, IEEE CCNC, IEEE ICCCN, IEEE WCNC, Technical Committee (2003-2009), Chair of the Communications Software
IEEE ICAEE, and IEEE ICAIT. He serves on the editorial board of Vehicular Technical Committee (2008-2010) and Chair of the Technical Committee
Communications Journal of Elsevier and Communications Networks Journal on Information Infrastructure and Networking (2016-2017). He has served
of Frontiersin. He has also served as guest editor for a number of international as Co-Program Chair of IEEE WCNC’2012 and ICC’2004, Executive
journals. He is a member of the IEEE and IEEE Communications Society. Vice-Chair of ICC’2017, Panel sessions co-chair for Globecom’16, tutorial
chair of VTC’2013 Spring and WCNC’2010, track chair of PIMRC’2012
and WCNC’2014, symposium Co-Chair at Globecom 2007-2011, ICC
2008-2010, ICC’2014 and ’2016. He is associate Editor for International
Journal of Communication Systems (IJCS-Wiley), Journal on Security and
Communication Networks (SCN-Wiley) and International Journal of Business
Data Communications and Networking, Journal of Network and Computer
Applications (JNCA-Elsevier). He is senior member of the IEEE, IARIA
fellow and member of many international program committees. He was IEEE
ComSoc Distinguished Lecturer Tour during 2013-2014.

Mohammed Atiquzzaman received the M.S.

and Ph.D. degrees in electrical engineering and
electronics from the University of Manchester,
U.K., in 1984 and 1987, respectively. He currently Abdullah Baz received the [Link]. degree in electrical
holds the Edith J. Kinney Gaylord Presidential and computer engineering from UQU, in 2002, the
Professorship with the School of Computer Science, [Link]. degree in electrical and computer engineering
University of Oklahoma, USA. His research has from KAU, in 2007, and the [Link]. degree in
been funded by the National Science Foundation, communication and signal processing and the Ph.D.
National Aeronautics and Space Administration, degree in computer system design from Newcastle
U.S. Air Force, Cisco, and Honeywell. He University, in 2009 and 2014, respectively. He was
co-authored Performance of TCP/IP Over ATM a Vice-Dean, and then the Dean of the Deanship
Networks and has authored over 300 refereed publications. His current of Scientific Research with UQU, from 2014 to
research interests are in areas of transport protocols, wireless and mobile 2020. He is currently an Associate Professor with
networks, ad hoc networks, satellite networks, power-aware networking, the Computer Engineering Department, a Vice-Dean
and optical communications. He CoChaired the IEEE High Performance of DFMEA, the General Director of the Decision Support Center, and the
Switching and Routing Symposium (2003, 2011), IEEE GLOBECOM and Consultant of the University Vice Chancellor with UQU. His research interests
ICC (2014, 2012, 2010, 2009, 2007, and 2006), IEEE VTC (2013), and SPIE include data science, ML, AI, VLSI design, EDA/CAD tools, intelligent
Quality of Service Over Next Generation Data Networks conferences (2001, transportation, computer system and architecture, smart systems, smart health.
2002, and 2003). He was the Panels Co-Chair of INFOCOM’05, and has Since 2015, he has been served as a Review Committee Member of the IEEE
been on the program committee of many conferences, such as INFOCOM, International Symposium on Circuits and Systems (ISCAS) and a member
GLOBECOM, ICCCN, ICCIT, Local Computer Networks, and serves on of the Technical Committee of the IEEE VLSI Systems and Applications.
the review panels at the National Science Foundation. He was the Chair of In 2017, IEEE has elevated him to the grade of IEEE Senior Member. He
the IEEE Communication Society Technical Committee on Communications served as a Reviewer in a number of journals, including the IEEE Internet
Switching and Routing. He received the IEEE Communication Society’s Fred of Things, the IET Computer Vision, the Artificial Intelligence Review, IEEE
W. Ellersick Prize and the NASA Group Achievement Award for outstanding Access, and the IET Circuits, Devices and Systems.
work to further NASA Glenn Research Center’s efforts in the area of the
Advanced Communications/Air Traffic Management’s Fiber Optic Signal
Distribution for Aeronautical Communications project. He received from
IEEE the 2018 Satellite and Space Communications Technical Recognition
Award for valuable contributions to the Satellite and Space Communications Hosam Alhakami received the [Link]. degree in
scientific community. He also received the 2017 Distinguished Technical computer science from King Abdulaziz University,
Achievement Award from IEEE Communications Society in recognition of Saudi Arabia, in 2004, the [Link]. degree in internet
outstanding technical contributions and services in the area of communications software systems from Birmingham University,
switching and routing. He is the Editor in Chief of Journal of Networks Birmingham, U.K., in 2009, and the Ph.D.
and Computer Applications, the founding Editor in Chief of Vehicular degree in software engineering from De Montfort
Communications, and serves served on the editorial boards of many journals, University, in 2015. From 2004 to 2007, he worked
including IEEE Communications Magazine, IEEE Journal on Selected Areas with Software Development Industry, where he
in Communications, IEEE Transactions on Mobile Computing, Real Time implemented several systems and solutions for a
Imaging Journal, Journal of Sensor Networks, and International Journal of national academic institution. Dr. Alhakami was
Communication Systems. the Vice-Dean of the Deanship of Admission
and Registration for Academic affairs with UQU, from 2015 to 2020.
Currently, he is an associate professor of the computer science department
with UQU. His research interests include algorithms, semantic web, and
optimization techniques. He focuses on enhancing real-world matching
systems using machine learning and data analytics in a context of supporting
decision-making.