Required To Be IEEE Format…Please Follow Instructions Below As Indicated. Topic: 5G Vehicular Network Resource Management for Improving Radio Access Thro

Required To Be IEEE Format…Please Follow Instructions Below As Indicated. Topic: 5G Vehicular Network Resource Management for Improving Radio Access Through Machine Learning
Instructions:  It should be of professional quality and be in the format of an IEEE transaction style. Figures must be clear and drawn by you.  Proper citation of references must be embedded in the term paper. All term report should be printed in 8.5x11in format, 10 size, time new roman font, two-column, about 8 pages, left, right, top, bottom margin 1 inch, and contain the following:

Title page.
Abstract (summary of the paper).
Introduction (problem motivation, background materials, related work, summary of objectives and methods).
(i) Description of existing algorithms/protocols for survey papers; (ii) system model, assumptions, and/or formal problem formulation for research-oriented papers.
(i) Comparison among existing algorithms/protocols and discussion on possible improvements/enhancements;      (ii) project results (derivation, proof, justification, or detailed design/simulation) in one or more numbered sections for research-oriented      papers.
Conclusions (evaluation of results, suggestions for improvements, or future work).
References must follow IEEE Transactions format (at least 10 references).Proper citation of references must be embedded in the term paper.
Appendices (if appropriate).
In particular, references used must be clearly cited in your written report at proper locations  

Sample document is attached…. SPECIAL SECTION ON SECURE COMMUNICATION FOR THE NEXT
GENERATION 5G AND IOT NETWORKS

Received December 18, 2019, accepted January 2, 2020, date of publication January 7, 2020, date of current version January 14, 2020.

Digital Object Identifier 10.1109/ACCESS.2020.2964697

5G Vehicular Network Resource
Management for Improving Radio
Access Through Machine Learning
SAHRISH KHAN TAYYABA 1, HASAN ALI KHATTAK 1, (Senior Member, IEEE),
AHMAD ALMOGREN 2, (Senior Member, IEEE), MUNAM ALI SHAH 1,
IKRAM UD DIN 3, (Senior Member, IEEE), IBRAHIM ALKHALIFA 2,
AND MOHSEN GUIZANI 4, (Fellow, IEEE)
1Department of Computer Science, COMSATS University Islamabad, Islamabad 44550, Pakistan
2Chair of Cyber Security, Department of Computer Science, College of Computer and Information Sciences, King Saud University, Riyadh 11633, Saudi Arabia
3Department of Information Technology, University of Haripur, Haripur 22620, Pakistan
4Computer Science and Engineering Department, Qatar University, Doha 2713, Qatar

Corresponding author: Ahmad Almogren (ahalmogren@ksu.edu.sa)

The authors are grateful to the Deanship of Scientific Research, King Saud University for funding through Vice Deanship of Scientific
Research Chairs.

ABSTRACT The current cellular technology and vehicular networks cannot satisfy the mighty strides
of vehicular network demands. Resource management has become a complex and challenging objective
to gain expected outcomes in a vehicular environment. The 5G cellular network promises to provide
ultra-high-speed, reduced delay, and reliable communications. The development of new technologies such
as the network function virtualization (NFV) and software defined networking (SDN) are critical enabling
technologies leveraging 5G. The SDN-based 5G network can provide an excellent platform for autonomous
vehicles because SDN offers open programmability and flexibility for new services incorporation. This
separation of control and data planes enables centralized and efficient management of resources in a very
optimized and secure manner by having a global overview of the whole network. The SDN also provides
flexibility in communication administration and resource management, which are of critical importance
when considering the ad-hoc nature of vehicular network infrastructures, in terms of safety, privacy, and
security, in vehicular network environments. In addition, it promises the overall improved performance.
In this paper, we propose a flow-based policy framework on the basis of two tiers virtualization for
vehicular networks using SDNs. The vehicle to vehicle (V2V) communication is quite possible with
wireless virtualization where different radio resources are allocated to V2V communications based on
the flow classification, i.e., safety-related flow or non-safety flows, and the controller is responsible for
managing the overall vehicular environment and V2X communications. The motivation behind this study
is to implement a machine learning-enabled architecture to cater the sophisticated demands of modern
vehicular Internet infrastructures. The inclination towards robust communications in 5G-enabled networks
has made it somewhat tricky to manage network slicing efficiently. This paper also presents a proof of concept
for leveraging machine learning-enabled resource classification and management through experimental
evaluation of special-purpose testbed established in custom mininet setup. Furthermore, the results have
been evaluated using Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and Deep
Neural Network (DNN). While concluding the paper, it is shown that the LSTM has outperformed the rest
of classification techniques with promising results.

INDEX TERMS Future internet architectures, machine learning, network reliability, privacy,
resource management, security, software defined networks, vehicular networks.

The associate editor coordinating the review of this manuscript and

approving it for publication was Ilsun You

I. INTRODUCTION
Substantial technological improvements lead to a pervasive
growth of smart things enabling communications among

6792 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/ VOLUME 8, 2020

https://orcid.org/0000-0002-0741-5861

https://orcid.org/0000-0002-8198-9265

https://orcid.org/0000-0002-8253-9709

https://orcid.org/0000-0002-4037-3405

https://orcid.org/0000-0001-8896-547X

https://orcid.org/0000-0001-7814-7983

https://orcid.org/0000-0002-8972-8094

https://orcid.org/0000-0002-0604-3445

S. K. Tayaba et al.: 5G Vehicular Network Resource Management for Improving Radio Access Through Machine Learning

FIGURE 1. SDN architecture depicting various components and their
respective roles.

entities from mobile devices to fast speed vehicles envisioned
in smart cities [1], [2]. The intercommunication demands for
quality of services (QoS) and quality of experience (QoE)
to realize the associated benefits such as fast communica-
tion, low latency, high reliability, maximized throughout, etc.
These requirements made networks more challenging in a
dynamic vehicular environment [3].

In recent years, the proliferation of intelligent vehicular
networks has captured the attention of industry and academia.
The substantial benefits associated with intelligent vehicles
can benefit an urban society in terms of traffic regulation,
resource management, and accident reduction among others.
A fast stride is observed in the development and implementa-
tion of Autonomous Driving Vehicles (ADVs), e.g., Waymo
by Google, Tesla self-driving cars, Aptiv, Audi A8, Ericsson’s
5GCAR project, etc. ADVs are more complex in nature due
to multiple integrated sensors, automotive control, and ultra-
fast communication capabilities. ADVs are anticipated to be
in the market till 2020 [4], and 25% ADVs are expected on
the roads by 2035.

Traditionally, wireless technologies provide infrastruc-
ture for vehicular communications, which can play a sig-
nificant role in efficient resource management as well as
transportation for the provisioning of QoS and QoE [5]. The
Third Generation Partnership Project (3GPP) is among the
incorporating standards for reliable communications among
Vehicles-to-Pedestrian (V2P), Vehicle-to-Vehicle (V2V),
Vehicle-to-Infrastructure (V2I), and futuristic Vehicle-to-
Everything (V2X) [2]. The enhancement in V2X services
enables multiplexing resources across vehicular networks.
However, the communication of wireless cellular networks is
a bit expensive in terms of latency for time-critical scenarios

in a vehicular network. 3GPP is a de-facto standard for
LTE-V2V communication standard protocol that allows com-
munications by directly exchanging messages with the LTE
infrastructure involvement [6].

In this paper, we proposed a resource allocation frame-
work for autonomous vehicular networks that may provide
optimization of resource allocation in a vehicular network,
as shown in Figure 2. Vehicles may communicate with other
vehicles or infrastructure. The proposed policy optimizes
flow requests from vehicles and a priority is assigned based
on the criticality of the application demand. Flows grouped-
based on application scenarios, such as applications for
road safety, infotainment applications, and/or applications for
comfort indicators. The flow slices are classified using Traffic
Classifier (TC). The elements of privacy and security are very
crucial in affecting the performance and utilization of vehic-
ular networks. Thus, confidentiality, authenticity, encryption,
and other security features are all assumed to be parts of the
overall environment.

The rest of the paper is organized as follows: starting with
Section II, we present the background and related studies of
SDN-based autonomous vehicular networks. In Section IV,
a framework for resource allocation in SDN-based networks
is presented. Section V discusses the achieved throughput
and response time of the proposed framework in addition
to compare the efficiency of machine learning techniques.
Finally, Section VI concludes the study and gives future
directions.

II. BACKGROUND AND RELATED WORK
The widespread use of large networks over the Internet has
proved to be hindering in giving users an optimal qual-
ity of service. Traditional network architectures prove cum-
bersome in terms of energy efficiency, dynamic network
configuration, agile network measurement, and flexible net-
work deployment [7]. Due to the unchanged architecture of
legacy networks for the past few decades, software defined
networks (SDN) have envisioned as an emerging approach
providing programmability, adaptiveness, and flexibility.
As SDN architecture provides a global network overview,
logical centralization, and strengthening a network, it also
introduces problems of resource management, some of which
have been addressed and brought forward by Khelifi et.al [8].

The joint initiative between the European Commission and
European ICT industry for 5G Public Private Partnership
Group (5G PPP) advocates the use of multiple technologies
for pervasive computing that enables multiple radio access
technologies (RATs) [9]. The advocates of 5G claim for the
provision of mobility, high flexibility, low latency, high reli-
ability, security, privacy, and maximized data rate in a highly
dynamic environment [10], [11]. The 5G METIS project [12]
provides a flexible architecture and service management in
wireless communications between a V2X network where
a vehicle communicates with other vehicles or an external
environment leveraging the existing and futuristic wireless
networks.

VOLUME 8, 2020 6793

S. K. Tayaba et al.: 5G Vehicular Network Resource Management for Improving Radio Access Through Machine Learning

FIGURE 2. Machine learning-based optimization high-level view for resource allocation.

The SDN is a well structured and layered architecture
consisting application layer, control plane, and data plane,
as shown in Figure 1. As evident from the architecture, SDN
controller is responsible for resource management through
slice management. This enables SDN to orchestrate the 5G
network through network functions virtualization (NFV),
which is considered as the key enabling technology for 5G
networks while considering resource management [13]. The
control plane and data/forwarding plane are separated in
SDN, enabling centralized management, greater flexibility,
and network programmability with a global view of the
overall network. The centralized control provides network
intelligence for resource orchestration. The SDN can also
benefit vehicular networks in terms of efficient resource
allocation and network management by providing fine-grain
traffic steering [14].

The data plane comprises data forwarding entities, vehi-
cles, physical infrastructure, e.g., base station (BS) or access
points (APs), of different RAT technologies such as 5G LTE,
Wi-Fi, WiMAX, etc. The V2X communication is also held
on this layer of the SDN network. Similarly, the control
plane is the decoupled entity from the rest of the distributed
forwarding devices. It is logically centralized on a server that
controls the overall functionality of the network functions.
The control plane can program forwarding devices through
southbound API interfaces. In addition, control plane also
defines rules/instruction sets for forwarding devices. Hence,
it is known as the network brain where all control logic
resides in the applications and controllers, which form the
control plane.

The SDN was initially designed for carrier infrastructure,
i.e., wired network, and it proved its success in data cen-
ters. However, its implications are demonstrated in other
network scenarios such as mesh networks, wmSDN, sensors
network management, SDN-based Wi-Fi networks, and wire-
less domain [15]. The SDN-based vehicular network imple-
mentation gets attention in recent years. In the same way,
SDN-based cellular network’s support for vehicular networks
is being investigated by academia, but mostly architectural
layout and theoretical work remain a focus. Despite the fact of

SDN associated benefits, there are certain overheads involved
in the dynamic network management in vehicular networks.
Resource allocation is critical due to the profoundly changing
network dynamics because of high mobility and less reliable
wireless link capabilities.

Xie et. al [16] have proposed a blockchain-enabled
infrastructure for Internet of Things (IoT) that provides secu-
rity and trust in SDN-enabled 5G VANETs. This definitely
puts additional burden on the overall network processing.
One optimal solution for the mitigation of resource manage-
ment is to adopt AI-based resource management techniques
[17]. This paper aims to implement a secure deep learning-
based API for the detection of diverse and sophisticated
resource demand in SDNs [18], [19].

Generally, VANETs are sub-forms of mobile ad-hoc net-
works wherein vehicles communicate using V2V or V2I
fashion with other vehicles and roadside units (RSUs) [8].
A VANET comprises highly mobile vehicles, supporting
infrastructure, and wireless communication. The communi-
cation between vehicles is done using Dedicated Short-Range
Communication (DSRC), which is considered as the de-
facto standard for vehicular communications. Beside DSRC,
different wireless technologies are also used for vehicular
communications such as Wi-Fi, WIMAX, and LTE. Wireless
technologies are advanced and sophisticated in the context of
providing a plethora of applications from road safety appli-
cations, cooperative driving to autonomous human-like driv-
ing, and provisioning of high bandwidth-demanding video
streaming access in a fast mobility environment. For sup-
porting a multitude of applications, various heterogeneous
RAT technologies are getting attention from the research
community [20], [21].

An SDN based architecture (SDVN) for vehicular net-
works is proposed in [22], which incorporates vehicles and
RSUs that act like SDN switches. The SDN-based archi-
tecture supports V2V, V2I, and vehicular-to-cloud (V2C)
communications. The data plane is constructed using an
overlay network where all forwarding devices behave like
SDN switches. The vehicle status information is collected and
monitored by the SDN controller. The status policy update

6794 VOLUME 8, 2020

S. K. Tayaba et al.: 5G Vehicular Network Resource Management for Improving Radio Access Through Machine Learning

is done using trajectory-prediction, which reduces the fre-
quency of status updates.

The SDN-based autonomous transportation system is
presented in [23] in which V2X communications of the ADV
network is presented. The RSU maintains local informa-
tion about the surrounding environment and communicates
with the ADV. This architecture is based on time division
multiple access (TDMA) for a four-lane infrastructure. The
SDN-enabled eNodeB (eNB) facilitates local vehicular net-
works. The RSU controller in the proposed SDN architec-
ture is responsible for the incorporation of data forwarding
by receiving it on the southbound interface and processes
it towards the northbound interface. This provides multiple
services to users, vehicles, and the complete infrastructure,
where the V2I communication is done using a wireless
link.

A software-defined in-vehicle networking (SDIVN)
architecture is proposed for the autonomous vehicle
resiliency, robustness, and timely message delivery in [24].
In [25], authors present a resource allocation mechanism
ADVs by combing mobile edge access computing (MEC) and
SDN in autonomous AVNETs. Resources are stored on the
MEC to get short response delay where the NFV is used to
efficiently manage resources among different MEC servers.
The MEC servers are placed on the edge, which help in
gaining scalability and efficient resource utilization.

Zheng et al. [26] proposed Soft-defined heterogeneous
Vehicular nEtwork (SERVICE) framework for radio resource
virtualization in LTE for VANETs. In this framework,
stochastic learning is used for the estimation of delay and
optimization of virtual resource scheduling. The framework
has two stages for resource virtualization, i.e., macro and
micro virtualization resource allocation, known as MaVRA
and MiVRA, respectively. The large and small time scale
variables are used to categorise services such as traffic den-
sity operated in MaVRA and network state, and queue state
operated in MiVRA time scale. The controller acts as a
proxy service fulfilling user requests in a hierarchical control
architecture. Two layers of the controller are the primary
controller– responsible for the global SERVICE network, and
the secondary controller–that acts as a regional controller.

The integrated architecture of SDN and Fog Comput-
ing (FC) for VANETs, named SDFC-VeNET, is proposed
in [27]. The control layer and forwarding layer reside on
edge consisting of multiple controllers in the control layer.
The controller provides a control function and resource man-
agement. The FC provides a cache mechanism for reduc-
ing the access time of the services. Two types of resource
allocation policies are used in SDFC-VeNET architecture,
i.e., macro resource allocation policy and micro resource allo-
cation policy leveraging CSI and QSI. The proposed scheme
used sparse code multiple access (SCMA) and full-duplex
to support massive communications in VANETs and uses
OFDM and half-duplex for V2I communications. It provides
centralized scheduling for a mixed SCMA/OFDMA scheme
for virtual resource allocation.

In the last few years, SDN integration in wireless and
cellular networks gained much attention. Different studies are
proposed for the incorporation and implementation of SDN
in autonomous VANETs, as depicted in 3. Software-defined
vehicular ad-hoc network (VANET) is proposed that provides
V2V communications using central SDN controller. It lever-
ages cellular communications for control transmissions by
incorporating stationary SDN RSUs. Resources are allocated
statically by using different wireless radio technologies such
as WiMax, WiFi, etc.

III. RESOURCE ALLOCATION IN AUTONOMOUS VEHICLES
In vehicular networks, autonomous vehicles need to com-
municate with other vehicles, i.e., V2V communication, and
with the roadside infrastructure wherein reliable communica-
tion is stringent, needs to collect real-time information such
as speed, direction, and location. Moreover, the increasing
demand for data services and high bandwidth under increas-
ing vehicular density, dynamic traffic conditions, and dis-
tributed RSUs across the infrastructure become challenging
objectives in an autonomous vehicular network. Tradition-
ally, DSRC is used for V2V communications and RAT tech-
nologies. However, these technologies have their limitations,
such as radio access in LTE uses orthogonal frequency mul-
tiple access, which suffers from spectrum limitations and
intrinsic delay incur in communications between vehicles.
In the case of safety applications, this delay response can
cause disaster. Moreover, limited spectrum restrains from
massive connectivity as anticipated in 5G networks.

Autonomous driving needs to precisely monitor the traf-
fic conditions in both safety and non-safety scenarios.
Despite, ADV comprises sensitive sensors, artificial intelli-
gence algorithms, and high-speed communication gadgets;
several challenges still encounter efficient management.
• Communication capabilities with the neighboring
vehicle to monitor the exact traffic situation.

• Inefficient network utilization
• Frequent information updates due to high mobility
• Optimal resource allocation based on the intermittent
traffic condition

IV. PROPOSED SDN-BASED VEHICULAR
NETWORK FRAMEWORK
In this study, we propose a policy framework for
resource allocation in SDN-based 5G cellular networks for
autonomous vehicles and suggested architecture, as shown
in Figure 3. The overall network is composed of several key
components. The main module is the SDN controller, which
is responsible for providing network hypervisor services.
Moreover, the global scheduler is responsible for keeping
records of all requests in the queue and thus enables the
resource optimizer to use machine learning enabling resource
management on the control plane.

The SDN-based 5G cellular network consists of two main
parts. The SDN core network and the SDN-enabled wireless
data plane. A fine-grained decomposition of the front haul

VOLUME 8, 2020 6795

S. K. Tayaba et al.: 5G Vehicular Network Resource Management for Improving Radio Access Through Machine Learning

FIGURE 3. SDN-based 5G architecture for VANET service provisioning.

network of the wireless part in the SDN cellular network
is achieved by implementing a wireless slice manager. The
wireless slice manages allocation and schedule resources
on the instruction of global scheduler in the SDN con-
troller. Wireless slicing is a virtualization concept to enable
a single network to behave in a manner similar to multiple
logical virtual networks. Thus, enabling granular controller
and fine-grained management of resources. The backhaul
network of this architecture is sliced based on the flow
request from mobile virtual network operators (MVNO). The
network-level slicing is handled by the hypervisor that acts
as a network virtualized engine, which flexibly slices net-
work infrastructure and radio resources based on the MVNO
demands of resources and traffic types.

This policy can dynamically adjust its behavior in response
to changes and QoS requirements. We define an optimal pol-
icy for network bandwidth allocation and queue management
at the cellular BS. The optimized policy considers the queue-
length resource allocation in the wireless slice manager to
bridge the gap between resource demands at the BS and
bandwidth allocation at the controller.

The network hypervisor, i.e., FlowVisor, centrally manages
the SDN cellular network and is responsible for handling
virtual slice manager at the BS, i.e., wireless hypervisor
instance, and allocates bandwidth as per the requirement
to achieve fairness. The virtual slice manager at each BS
serves as the local controller. This local controller reduces
the communication overhead at the centralized controller.
The overhead is computed in terms of time computation and
processing for allocating resources dynamically. The virtual
manager logically partitions physical resources into slices
and assigns resources based on adaptive policies.

A. SDN-BASED VANET ARCHITECTURE
The basic architecture consists of the following modules in
the control plane.

1) ADAPTIVE POLICY GENERATOR
The adaptive policy generator is responsible for creating
network policies based on predefined rules agreed by InPs
and MNOs. The ADP module collects network statistics and
SLA from the NSM module and generates an adaptive policy
for resource allocation based on the resources available and
allocated capacity to each network slice.

2) RESOURCE MANAGEMENT MODULE (RMM)
The resource management module (RMM) calculates the
available resources in the network and analyzes real-time
network information in addition to generate global resource
management in the SDN controller. The RMM executes
resource allocation algorithm for optimization and allocation
of resources to different network slices. We used the cuckoo
search optimization algorithm for achieving the optimization
state of resource allocation.

3) TRAFFIC ADAPTIVE SCHEDULING (TAS)
The traffic adaptive scheduling (TAS) is a module in the
SDN controller, which is responsible for the collection of
network information. The TAS can envision a global view
because of centralization of the controller, which possesses
insight into the global network information. Network slices
span all resources of the whole network. The TAS clas-
sifies the QoS flow classes by using the traffic classifier
module.

6796 VOLUME 8, 2020

S. K. Tayaba et al.: 5G Vehicular Network Resource Management for Improving Radio Access Through Machine Learning

4) TRAFFIC CLASSIFIER
The QoS is an important parameter to get enhanced perfor-
mance. Network QoS enforced by applying different QoS
parameters thresholds and techniques. One method to ensure
QoS is the classification of traffic based on some prede-
fined criteria. This traffic classification is used to do effec-
tive scheduling in the network. Queuing mechanisms are
vital techniques to prioritize the overall traffic in the net-
work elements. The efficient use of these queuing techniques
ensure bandwidth for different traffic classes and control the
congestion.

In the proposed framework, we consider three differ-
ent queues in each port of the OpenFlow-enabled device:
(i) Priority Queue (PQ), (ii) Bandwidth sensitive Queue (BQ),
and (iii). No strict queue (NSQ). The most prioritized traffic
buffered in the PQ. The traffic prioritization is based on sen-
sitivity of the application and user requirement with intensive
delay boundaries. The threshold value is defined for high
priority traffic. If the traffic required delay is less than the
defined threshold, it is inserted in the PQ.

B. OPTIMIZED RESOURCE ALLOCATION PROCEDURE
The SDN-based cellular network management and bandwidth
allocation policy use virtualization manager at the centralized
controller and the BS for core and wireless virtualization.
It provides a flow-based queue-length resource allocation in
the wireless domain. The overall flow management is per-
formed on the controller level. It can dynamically adjust its
behavior in response to changes and QoS requirements due to
the presence of a virtual slice manager at each BS and a local
controller in addition to allocate resources to the incoming
flow. The scheduling granularity is achieved by identifying
QoS flows in the network flow.

The flows are categorized as safety flow and non-safety
flows. The safety flows get more priority ratio based on the
criticality, and the non-safety flows get less priority. In the
proposed framework, bandwidth allocation is the main fair-
ness allocation criterion, i.e., fair allocation of bandwidth
so that to achieve rate maximization for different applica-
tion/services flow in the SDN. For the optimal solution,
we use the Cuckoo search algorithm.

V. PERFORMANCE EVALUATION AND DISCUSSION
For the simulation purpose, we use a light-weight fork of
mininet emulator, version 2.3.0d1, mininet-wifi running on
the Ubuntu virtual machine of version 18.04 LTS. Mininet
is considered as a standard emulator for experimenting the
SDN concepts. Mininet-wifi provides large functionalities of
the wireless networks from propagation models to mobility
scenarios, as shown in Figure 1. The car node, implemented
in [28], provides a vehicular architecture to offer the proof-
of-concept of a vehicular network in SDN scenarios. It is one
such example that we have leveraged to achieve the video
content request from the cloud.

In the proposed scenario, we have considered an info-
tainment application, which is deployed at the application
plane. The core mechanism of the application service is

TABLE 1. Training results for machine learning models adopted for the
experiment.

.

FIGURE 4. Training score (Accuracy and Loss) for the dataset generated
by the mininet-wifi for traffic generation and floodlight as the SDN
controller.

to receive data. We have conducted V2X simulations using
mininet-wifi where multiple RSUs can communicate with
each other and vehicles can perform V2V and V2I commu-
nications. The SDN default controller is used for simulating
the transmission of video from a vehicle to a client. The
resource allocation in vehicular networks is managed by the
SDN controller in an efficient manner.

The overall simulation was run multiple times using
stochastic process in order to mimic realistic scenarios. The
dataset that we gathered includes information about the
requests, which are generated from client nodes at the data
plane. We have used deep learning techniques to train the
system using historical data of the network. The historic
network data consists of both standard and tagged signatures.

We have used long short-term memory (LSTM),
convolutional neural network (CNN), and deep neural

VOLUME 8, 2020 6797

S. K. Tayaba et al.: 5G Vehicular Network Resource Management for Improving Radio Access Through Machine Learning

FIGURE 5. Classification accuracy of CNN, LSTM, and DNN compared for
resource management using machine learning model on experimental
data.

network (DNN) models for predicting results and have com-
pared them for the best suitable solution. The CNN essentially
is a neural network variation mainly used for computer vision
with hidden layers. Similarly, the DNNs are feed forwarding
neural networks, which basically leverage many layers to
successfully classify regions of interest in a data.

Moreover, the LSTM being a break through in machine
learning techniques ensures to avoid this layer’s Indepen-
dence issues. The LSTM by remembering things between
layers achieves a great feat of promising results specially in
text and speech analysis. In our case, we feel that all three
of these have promising results. The experimental data is
divided into train and test sets for training and testing the
system, respectively. The training data, which is 80% of the
dataset, is fed into learning algorithms. These algorithms train
themselves on the data and predict the test set, which is 20%
for results, as shown in Table 1.

In the experiments, we use the SDN Floodlight controller
as a development environment. We split data into 80:20 for
training and testing. For LSTM, we use Scikit-learn1 library
for the implementation using parameters relu/sigmoid, 64,
20 and Adam for activation, batch size, epochs, and …

Looking for this or a Similar Assignment? Click below to Place your Order