You are on page 1of 16

Received: 22 May 2018 Revised: 17 August 2018 Accepted: 26 August 2018

DOI: 10.1002/dac.3817

SPECIAL ISSUE ARTICLE

Swarm intelligence‐based radio resource management for


D2D‐based V2V communication

Souhir Feki1 | Ahlem Masmoudi1 | Aymen Belghith2 | Faouzi Zarai1 |


3,4
Mohammad S. Obaidat

1
NTS'COM Research Unit, National
School of Electronics and
Summary
Telecommunications of Sfax, Tunisia Internet of Things is a promising paradigm that provides the future network of
2
Computer Science Department, Saudi interconnected devices. Device‐to‐Device (D2D) communication, which is con-
Electronic University (SEU), Saudi Arabia
sidered as an enabler for vehicle‐to‐everything applications, has become an
3
ECE Department, Nazarbayev
emerging technology to optimize network performance. In this paper, we study
University, Astana, Kazakhstan
4
KASIT, University of Jordan, Amman,
the Radio Resource Management (RRM) issue for D2D‐based Vehicle‐to‐
Jordan Vehicle communication. The RRM key role is to assure the proficient exploita-
tion of available resources while serving users according to their quality of
Correspondence
Souhir Feki, NTS'COM Research Unit, service parameters. An Ant Colony Optimization (ACO)‐based Resource
National School of Electronics and Allocation (ACORA) scheme is proposed in this paper. Swarm intelligence
Telecommunications of Sfax, Tunisia. algorithm ACO is adopted to reduce the computational complexity while
Email: souhir.feki@gmail.com
realizing satisfactory performance. Simulation results show promising perfor-
mance of our proposed ACORA scheme.

KEYWORDS
D2D communication, Internet of Things, radio resource management, software defined networks

1 | INTRODUCTION

Internet of Things (IoT) is an integrated element of future Internet containing existing and evolving networks as well as
Internet progress. Conceptually, it could be defined as a global dynamic network infrastructure with self‐management
and self‐configuration capabilities based on interoperable and standard communication protocols.1 Device‐to‐Device
(D2D) communication is an important part of the IoT, in which devices autonomously communicate and collaborate
with each other, share, and forward information without centralized control in a multi‐hop way. The ability to collect
pertinent information in real time is the key to support the value of IoT, expediting the creation of an intelligent envi-
ronment. Vehicle‐to‐Everything (V2X) is based on D2D communication. V2X is an emerging technology that enhances
and leverages existing Long‐Term Evolution (LTE) network elements and features to assure the communication
between vehicles and the infrastructure as well as between vehicles. The attractiveness of this emergent technology is
due to its ability to support safety use cases with stringent bandwidth, delay, and reliability requirements.2 In release
123 and release 134 of LTE, V2X is introduced as part of the Proximity Services (ProSe).5 It includes many applications,
such as travel information, navigation and driver assistance, payment transactions, fleet management, congestion avoid-
ance, safety, and traffic control.
V2X communication consist of different kinds of communications: Vehicle‐to‐Vehicle (V2V), Vehicle‐to‐
Network (V2N), Vehicle‐to‐Pedestrian (V2P), and Vehicle‐to‐Infrastructure (V2I).6 As V2V communication has

Int J Commun Syst. 2018;e3817. wileyonlinelibrary.com/journal/dac © 2018 John Wiley & Sons, Ltd. 1 of 16
https://doi.org/10.1002/dac.3817
2 of 16 FEKI ET AL.

elevated density and speed, it introduces several technical challenges especially the latency and reliability require-
ments. Guaranteeing these Quality of Service (QoS) requirements for V2V services is still a challenge that must
be tackled.
IEEE 802.11p7 was the first standard for V2V communication until 2016. The Third Generation Partnership Project
(3GPP), introduced in release 14 of LTE, features addressing direct V2V communications.8,9 The main advantage of
using LTE for V2V communications, as analyzed in Bazzi et al,10 is that the adopted technology is the same used for
cellular communications.11 Therefore, the same hardware and protocols can be exploited. This technical aspect repre-
sents an important advantage as vehicles already embed cellular interfaces. Furthermore, resources in LTE are orthog-
onal, which allow higher multiplexing as well as capacity and reliability increase. However, the main problem with LTE
solution is the higher complexity of protocols.12,13
LTE Radio Resource Management (RRM) introduced in 3GPP TS 36.30014 includes a large variety of procedures
and techniques, such as packet scheduling and resource allocation.15-18 Resource allocation, in LTE‐V2V, is one of
the more openly debated subjects. Several resource allocation mechanisms for V2V have been proposed in the lit-
erature. Their common goal is to use efficiently radio resources in order to guarantee the QoS requirements and to
achieve the system performance targets.19 Most of these works aim to prioritize cellular links and maximize the
sum rate. Othmen et al20 investigated the resource allocation for many D2D links and Cellular Users Equipment
(C‐UEs). They proposed a greedy heuristic approach that can reduce interference from D2D to cellular network
using channel gain information. In this algorithm, the resources of cellular and D2D users are synchronized based
on the interference link gain from D2D transmitter to the base station. This algorithm succeeds in reducing the
interference between C‐UEs and Device Users Equipment (D‐UEs), but the resource allocation problem in the
D2D communication is formulated as a mixed integer nonlinear programming that is hard to solve in a short
scheduling period. In Algedir and Refai,21 the resource allocation problem of D2D communication under a two‐tier
cellular network was explored. A three step sub‐optimal algorithm is proposed to optimize overall system through-
put with minimum interference among D2D, cellular, and small cells. First, maximum interference permitted at
each Resource Block (RB) is calculated based on rate requirements of C‐UEs. Then, for each D2D pair, reuse
RB candidates are determined. Finally, best RB for each D2D pair is selected by applying exhaustive searching
method. This algorithm optimizes overall system throughput with minimum interference between D‐UEs and C‐
UEs, but the optimal allocation solution is determined using a high complexity algorithm. In Zhang et al,22 an
underlying resource sharing algorithm was proposed where vehicle‐to‐infrastructure and V2V communication links
are allowed, for their data transmission, to share the same radio resources. A resource allocation optimization
problem was then elaborated using the interference between diverse communication links. This underlying
resource sharing algorithm improves the spectrum efficiency of vehicular networks with low computational com-
plexity but does not consider the users' QoS requirements. The work presented in Feng et al23 aims to improve
the overall network sum rate and to guarantee the QoS requirements of C‐UEs and D2D users. For this purpose,
a sophisticated three stage power control and resource allocation algorithm was proposed. First, it executes the
admission control. Then, it assigns powers for every D2D pair with its C‐UE partners. Finally, to improve the over-
all network throughput, a suitable C‐UE partner is selected for each admissible D2D using a maximum‐weight
bipartite matching‐based scheme. This algorithm maximizes the system throughput and maintains the QoS of C‐
UEs and D‐UEs, but it focuses basically on resource sharing without considering an adaptive power allocation
approach. In Esmat et al,24 the problem of resource allocation for D2D communications in multi‐cell networks
was studied while considering the intra‐cell and inter‐cell interferences. Authors aim to maximize the network
sum rate while ensuring the QoS for C‐UEs and D‐UEs. They proposed a two‐level algorithm: firstly, the admission
control is executed to find the set of possible D2D pair with their C‐UEs partners that attain the lowest QoS
demands. Then, the suitable power is assigned to each D‐UE and its partners to improve the network sum rate.
This algorithm improves the access rate and throughput gain in the network, but it does not consider the fairness
metric between users, and services with high QoS requirements are penalized.
The resource allocation, in some works, should be implemented jointly with the power control to achieve the
entire potential of D2D communication. In Meshgi et al,25 a framework for resource allocations and power control
for multicast D2D communications was proposed. It aims at maximizing the throughput of C‐EUs and feasible
D2D multicast groups, while meeting a certain Signal to Interference and Noise Ratio (SINR) constraint for both
the C‐EUs and the D2D groups. The problem of channel and power allocation is formulated as a mixed integer non-
linear programming problem, which is impossible to solve in a short scheduling period, where one D2D group can
reuse the channels of multiple C‐EUs and the channel of each C‐EU can be reused by many D2D groups. This
FEKI ET AL. 3 of 16

algorithm provides an effective and flexible means to utilize and share radio resources in cellular networks without
causing harmful interference. The joint spectrum and power allocation with spectrum sharing for V2X communica-
tion was investigated in Wei et al.26 The authors proposed a matching‐based resource allocation and power control
algorithm. It aims to maximize the sum rate of C‐UEs and Vehicular Users Equipment (V‐UEs) with LTE‐Unlicensed
(LTE‐U) technology, under the constraints of fairness coexistence. V‐UEs are classified into safety and non‐safety
users based on the corresponding services. Despite that this algorithm proposes a low complexity solution for LTE‐
U‐based V2X communication, it increases the latency in the network and does not consider the fairness metric
between users. The multi‐cell scenario was studied in Ramezani‐Kebrya et al,27 where the authors optimized the
transmission power of C‐UEs and D‐UEs to maximize their sum rate. The approach is optimal when supposing only
two communications per cell and restricting the inter‐cell interference to only one neighboring cell. Nevertheless, the
authors discussed how to improve the solution to a multi‐user scenario and gave an upper limit on the performance
loss for the case of multiple neighboring cells.
High mobility in vehicular communications causes a rapid modification of the wireless channels over time, render-
ing traditional D2D schemes inapplicable for V2V communications. To fix this issue, in Sun et al,28 a heuristic algo-
rithm was proposed where reliability and latency requirements of V2V links have been formulated into optimization
issue using large‐scale fading information of vehicular channels. Likewise, authors in Sun et al29 proposed a two‐step
power control and resource allocation algorithm for V2V communication based on QoS requirements of C‐UEs and V‐
UEs. In Sun et al,30 multiple RBs are permitted to be shared between one vehicular and one cellular user and also
between several vehicular users. The algorithms presented in Sun et al,28-30 consider the QoS requirements for both
CUEs and V‐UEs and maximize the sum rate of C‐UE with certain fairness consideration; however, only slow fading
information is considered and the system capacity calculation ignores fast fading effects and thus will not reflect the
network real performance. Some works allocate resources based on users' positions. Kim et al31 proposed a position‐
based scheduling algorithm for V2V communication allocating different time and frequency resources based on the
position, direction, speed, and density of vehicles. The algorithm includes two resource allocation strategies according
to vehicle location: the urban and the freeway cases. In the urban case, the proposed algorithm allocates a special flex-
ible sub‐resource‐pool in the intersection area, due to the high density and the significant interference between vehi-
cles. For the freeway case, resources are allocated based on position and direction of vehicles. Freeway is divided into
different zones. Each zone has a dedicated sub‐resource‐pool. Vehicles, which penetrate different zones, have to allo-
cate resources from the new sub‐resource‐pool. The lack of available resources, in this case, may enhance the latency
and generate additional collisions. This algorithm can avoid the interference from neighboring vehicles, but its effi-
ciency may be limited for high speed vehicles, as they more frequently enter diverse zones. As well it does not con-
sider the user's QoS requirements. Another work has adopted the strategy of location‐based resource allocation for
V2V communication in Botsov et al.32 The authors designed a centralized resource allocation strategy based on the
vehicles' positions inside a single cell. It aims to guarantee the opportunities of continuous and periodic transmission
for V‐UEs, while keeping the control overhead at a minimum in the system. Firstly, the proposed approach defines a
spatial resource reuse strategy, to reuse the available cellular uplink and downlink resources, by separating between
C‐UEs and D2D users that use the same resources. Secondly, authors proposed two resource allocation schemes for
both C‐UEs and V‐UEs. Each terrestrial zone has two dedicated sets of RBs: a RB set for D2D communication and
a set of resources restricted for cellular communication. For V‐UEs, each V‐UE signals its location within the zone
topology to the evolved Node B (eNodeB) which then allocates, to the V‐UE, a subset of RBs from the appropriate
set of RBs dedicated for each zone. If available resources in a zone are not sufficient to simultaneously supply all
V‐UEs, concurrent terminals are instructed to share specific RBs. This algorithm reduces the signaling overhead
and interference in the network and satisfies the requirements of Cooperative Intelligent Traffic Systems (C‐ITS)
safety services, but it does not consider the user's QoS requirements and poses the problem of high packets delay
in the congested zones.
In this context, we propose an Ant Colony Optimization (ACO)‐based algorithm for radio resource allocation to
improve the overall network sum rate while satisfying the QoS requirement of cellular and vehicular users. In the
proposed algorithm, the V‐UEs utilize orthogonal RBs between each other, and to communicate with the eNodeB,
the C‐UEs utilize orthogonal RBs. We assess the performance of our approach and compare it with three well used
proposed resource allocation algorithms from the literature: position‐based resource allocation,31 location‐dependent
resource allocation,32 and random resource allocation where resource units are chosen randomly.33
This paper is structured as follows. Section 2 presents the proposed algorithm. Section 3 describes the simulation
results and performance analysis. Finally, Section 4 concludes this paper.
4 of 16 FEKI ET AL.

2 | SY STEM MOD EL A N D ALG O R I T HM DE SC R I P T I O N

In this section, an ACO‐based system model for the resource allocation is proposed to resolve the optimization problem
of improving the overall network sum rate under the constraint of satisfying the QoS requirement of cellular and
vehicular users.

2.1 | Ant colony optimization


ACO is an algorithm of swarm intelligence that becomes a novel research hotspot. The swarm intelligence is a
metaheuristic approach to solve a diversity of issues. Its methods, used in combinatorial optimization issues, are popu-
lation‐based stochastic techniques, where the collective performance of individuals emerges from their exchanges with
their environment to create functional global models.34 ACO is an optimization technique. Its main idea is inspired
from the behavior of efficient seeking out food by colonies of ants.35 Ants use a volatile chemical substance, left behind
them, known as pheromone. The pheromone attracts other ants, and they follow the way with the maximum concen-
tration of pheromone. Ants communicate with each other through pheromones. Their typical task is to look for the
shortest way. A preliminary pheromone quantity is submitted in every edge of the graph, indicating the desirability
for traversing this edge. Each ant constructs a path using the following equation:

 α h

τ i; j
ηi; j
pm
i; j ¼  α  β : (1)
∑l∈M mi τ i;l ηi;l

Equation 1 indicates the probability for selecting edge ij when beginning from node i. τi,j is the pheromone value, ŋi,j
is the heuristic information indicating the attractiveness of the move, and M refers to the neighborhood of node i. The
maximum the rate for heuristic information and pheromone, the maximum is the probability for selecting this edge.
The weighting parameters α and β are used to adjust the influence of heuristic information and pheromone. ACO is
based on three phases: the pheromone evaporation, then the update of local pheromone for all found paths, and finally
the update of global pheromone on the best path. Over time, the trail of pheromone starts to evaporate in all traversed
paths, thus reducing its attractive strength. The evaporation effect is important as other ants will be less likely to pursue
the traversed way and the probability for discovering novel areas is elevated. The evaporation of the local pheromone is
given as follows:

τ i; j ¼ ð1−ρÞτ i; j (2)

Where 0 < ρ < 1 is a parameter indicating the local pheromone evaporation. The ρ role is to make the system converge
to a good solution by forgetting the bad routes. After building a complete path, pheromone values must be updated.
Each ant deposits pheromone on the path it went through, so the local pheromone concentration will be increased
on all the paths between the food and ants' nest as follows:

τ i; j ¼ τ i; j þ τ 0 (3)

Where τ0 is the local pheromone initial value. The local updating role is to change dynamically the desirability of nodes;
without local updating, the search area of ants will be restricted to a narrow area of the major prior solution. When all
the ants have traversed all the edges, only the ant that found the best path from the beginning is allowed to update the
concentration of pheromone on the edges, and the global pheromone is updated according to

best
τ i; j ¼ τ i; j þ Δτ (4)
i; j

Where Δτi,jbest indicates the pheromone on the best path, which is proportional to the quality of the solution.
FEKI ET AL. 5 of 16

2.2 | QoS requirements of C‐UEs and V‐UEs


Uses and kinds of interests of traditional C‐UEs are different to that of V‐UEs; thus, they must have different QoS
requirements. The Outage Probability (OP) and average symbol error probability are considered the main metrics of
wireless communication systems.36 In this paper, we consider the OP of users in order to meet the communication qual-
ity requirements and improve the network QoS.

2.2.1 | QoS requirements of V‐UEs


Vehicular communications have very stringent QoS requirements on the latency and reliability of the communication
links. Whenever RB is allocated to V‐UE, QoS requirements should be maintained using a resource allocation method.
In this paper, the QoS requirements of V‐UEs are presented by the latency and the OP, which is the probability that the
immediate channel mutual information is under the transmitted code rate.37 Thus, for Nk RBs assigned to V‐UE k, the
OP of user k is defined as:

N 
out k  
p ¼ Pr ∑ δ log2 1 þ γ k;n < M k (5)
k n¼1

Where δ is the number of symbols per RB, Mk is the required number of bits of user k, and γk,n is the instantaneous
SINR of user k on RB n. γk,n is calculated as follows:

Pk;n H k;n
γ k;n ¼ 0 (6)
N 0 þ ∑ k ∈ K=kPk0 ;n H k0 ;n

Where Pk,n and Hk,n are the transmission power and the channel gain, of user k over RB n, respectively. N0 represents
the noise spectral density. The expression of interference in the denominator defines the interference at user k produced
by the transmissions of other users k'∈K \ {k} on same RB n.

2.2.2 | QoS requirements of C‐UEs


The system, in traditional cellular communications, strives usually to provide high data rates for C‐UEs with some levels
of fairness. In this section, we consider the requirements of C‐UEs as the tradeoff between the OP and the fairness infor-
mation of the user. The fairness information of each user k, User Fairness Index (UFI),38 is defined as follows:

UFI k ¼ T k =T req
k (7)

where T k and T req


k are the instantaneous throughput and the throughput required by user k, respectively.

2.3 | The proposed ACORA algorithm


The principal of the proposed ACO‐based Resource Allocation (ACORA) algorithm is shown in Figure 1. Firstly, after
traffic differentiation to V‐UEs and C‐UEs, the scheduling metric is calculated for each user in the scheduling process.
Then, a set of RBs are adaptively assigned to each class of users. Finally, in each class, RBs are allocated to users based
on the ACO mechanism.
As shown in Figure 1, traffic is differentiated into two classes: V‐UEs and C‐UEs. Each type of traffic will be then
delivered in an independent queue. Kc and Kv represent the number of V‐UEs and C‐UEs, respectively. K is the users'
total number (K = Kc + Kv). A set of scheduling metrics is associated to each class. The throughput and fairness metrics
are used for C‐UEs, and the packet head of line delay (delay Hol) metric is used for V‐UEs. The amount of RBs assigned
to each traffic is calculated using the Packet Drop Rate (PDR) of C‐UEs and the OP of V‐UEs.
6 of 16 FEKI ET AL.

FIGURE 1 Illustration of the proposed


ACORA algorithm

2.3.1 | Scheduling process


For each class of users, scheduling metrics MC_UE and MV_UE are calculated for C‐UEs and V‐UEs, respectively. In each
Transmission Time Interval (TTI), the scheduler selects for packets transmission the C‐UE and the V‐UE that maximize
MC_UE and MV_UE using Equations 8 and 11, respectively.
Metric MC_UEis calculated using the proportional fair scheduler in order to maximize the throughput while main-
taining a certain degree of fairness39:

d c ðt Þ
MC UE ¼ (8)
Rc ðt − 1Þ

dc ðt Þ ¼ logð1 þ SNRc Þ (9)



1 1
Rc ðt − 1Þ ¼ 1 − Rc ðt − 1Þ þ dc ðt Þ (10)
τ τ
Where dc(t) is the transmission rate of the cth user, Rc(t − 1) is the average throughput for the cth user at time (t – 1), and
τ is a constant value for averaging the cth user data rate.
Metric MV_UE aims to avoid deadline expiration, using Earliest Deadline First (EDF) scheduler.40 It is calculated as
follows:
1
M V UE ¼ (11)
τ v − DHoL;v
where τv and DHoL, v define the target delay and the packet head of line delay of the vth user, respectively.

2.3.2 | Allocation process


Our proposed resources allocation algorithm is designed for the overlay mode where the allocation of available RBs to
users is done in orthogonal manner. First, we assign for each type of traffic a set of RBs. Then, to improve the QoS of cel-
lular and vehicular traffic, we update in adaptive way the number of RBs assigned to vehicular and cellular traffic every
TTI according to the average OP (AvgOP) of V‐UEs and the PDR of C‐UEs. Thus, AvgOPV‐UEs is calculated as follows:

K
∑v¼1
V
pout
AvgOPV −UEs ¼ v
(12)
KV
th
Where pout
v is the OP of the v V‐UE.
The PDR of C‐UEs (PDRC‐UEs) is calculated in each TTI as the ratio between the total number of dropped packets
(due to time out) and the total number of generated packets:

K
∑c¼1
C
Prejected
PDRC−UEs ¼ K
c
(13)
∑c¼1
C
Pgenerated
c

Where Prejected
c is the total number of dropped packets by the cth C‐UE and Pgenerated
c is the total number of generated
packets by the cth C‐UE.
FEKI ET AL. 7 of 16

FIGURE 2 Flowchart of RBs sharing


between V‐UEs and C‐UEs

We assume that the numbers of RBs assigned to V‐UEs and C‐UEs change adaptively every TTI. As shown in
Figure 2, if PDRC − UEs is higher than a defined PDR threshold (PDRth) and AvgOPV − UEs is lower than a defined OP
threshold (OPth), then the number of RBs allocated to V‐UEs (SetV‐UEs) will decrease by Δ, where Δ is a percentage of
the total number of RBs. Else, if the PDRC − UEs is lower than the PDRth and the AvgOPV − UEs is higher than the OPth,
then the number of RBs assigned to V‐UEs will increase by Δ.

2.3.3 | ACO‐based resource allocation


In this section, we introduce a resource allocation technique that can fulfill the QoS requirements for all users by pri-
oritizing users with the lowest rate of OP. In the proposed algorithm, each cell maintains a graph for the ACO algorithm
and ants select users, with a certain probability, for the current RB. The chosen routes are saved in a matrix (Figure 3).
The probability that RB n is assigned to user k is calculated according to Equation 1, where the pair of nodes (i,j) is
replaced by the pair of user and RB (k,n). This probability is calculated as follows:
 α  β
τ k;n ηk;n
pk;n ¼
m
 α  β (14)
∑l∈M mn τ l;n ηl;n

FIGURE 3 ACO for RBs allocation


8 of 16 FEKI ET AL.

Where Mnm represents the set of available users to select for ant m on RB n. Recall that the local pheromone is
updated when an ant constructs a solution, see Equation 3. Heuristic information ŋk,n indicates the desirability of that
allocation; it may guide the ants to the most promising solutions. Pheromone τk,n indicates how efficient it has been in
the past to select that particular allocation. After acquiring all solutions, the system evaluates, under different judge
functions, the quality of the solutions and updates the corresponding pheromone. In the proposed algorithm, for both
C‐UEs and V‐UEs, the heuristic information of user k on RB n is calculated based on the number of transmitted bits.
The higher the transmission rate of user k on the RB n, the higher is the probability for the user to be allocated. This
information is calculated as follows:

TABLE 1 Parameters and symbols used in the ACORA algorithm

Symbol Definition

K Total number of users.


KC Total number of C‐UEs.
KV Total number of V‐UEs.
MC‐UE Scheduling metrics for C‐UEs.
MV‐UE Scheduling metrics for V‐UEs.
dc (t) Transmission rate of the cth user at time (t).
Rc (t − 1) Average throughput for the cth user at time (t − 1).
τ Constant value for averaging the cth user data rate.
τv Target delay of the vth user.
DHol,v Packet head of line delay of the vth user.
AvgOPV‐UEs Average outage probability of V‐UEs.
pout
v Outage probability of the vth V‐UE.
PDRC‐UEs PDR of C‐UEs.
Prejected
c nrejected Total number of dropped packets by the cth C‐UE. ngenerated

Pgenerated
c
Total number of generated packets by the cth C‐UE.
PDRth PDR threshold for C‐UEs.
OPth Outage probability threshold for V‐UEs.
SetV‐UEs Number of RBs allocated to V‐UEs.
Δ A percentage of the total number of RBs.
pm
k;n Probability that RB n is assigned to user k.
Mnm Set of available users to select for ant m on RB n.
ŋk,n Heuristic information of user k and RB n.
τk,n Pheromone value of user k and RB n.
τ0 Initial value of pheromone.
α Weighting parameter used for the heuristic information.
β Weighting parameter used for the pheromone value.
ρ Parameter controlling the evaporation of the local pheromone.
Δτk,nbest Pheromone on the best path.
δ Symbols' number per RB.
γk,n SINR of user k on RB n.
Nk Number of RBs allocated to user k.
fk Judge function.
μ Weighting parameter used in the judge function.
UFIk User fairness index of user k.
FEKI ET AL. 9 of 16

FIGURE 4 Flow chart of ACO‐based resource allocation algorithm

 
ηk;n ¼ δ log2 1 þ γ k;n (15)

Where δ is the symbols' number per RB and γk,n is the SINR of user k on RB n. γk,n is calculated according to Equation 6.
The judge function of V‐UEs is calculated in terms of OP calculated according to Equation 5. Our optimization issue
N
k
is to maximize the network sum rate, so we aim to maximize ∑ δ log2 ð1 þ γk; nÞ for all Nk RBs allocated to user k. For
n¼1
C‐UEs, we define the judge function of users as the tradeoff between the OP and the fairness information of the user,
calculated according to Equation 5 and Equation 7, respectively. The judge function is defined as follows (Table 1):
 
f k ¼ μ * 1 − Pout
k þ ð1 − μÞ * UFI k (16)

Where μ is a weighting parameter used to modulate the contribution of (1‐Pkout) and UFIk. Therefore, after acquiring all
solutions, the user with the best tradeoff between the OP and the fairness index has the best allocation. Note that the
global pheromone is updated according to Equation 4. Figure 4 gives the flow chart of the proposed allocation algorithm.
ACORA algorithm
{Inputs}
K ← Number of UEs;
KV ← Number of V-UEs;
KC ← Number of C-UEs;
N ← Number of RBs;
Δτk,nbest ← The pheromone on the best path
Set_ UE = {UE1, UE2, …, UEK}
Set_V-UE = {V-UE1, V-UE2, …, V-UEKv}
Set_C-UE = {C-UE1, C-UE2, …, C-UEKc}
Set_RB = {RB1, RB2, …, RBN}
τ0 = 1; % initial value of pheromone
α =1; % weighting parameter
β = 3; % weighting parameter
ρ = 0.1; % parameter controlling the evaporation of the local pheromone
μ = 0.5; % weighting parameter
{Main}
% Calculate the heuristic information of user k on RB n
10 of 16 FEKI ET AL.

Calculate ŋ (k,n), with (k,n) ∈ Set_UE * Set_RB


% Calculate the probability that RB n is assigned to user k
Calculate P (k,n), with (k,n) ∈ Set_UE * Set_RB
% Select and assign the (UE, RB) pair with the highest probability P.
Find (k,n) = max P (k,n), with (k,n) ∈ Set_UE * Set_RB
Assign RBn to UEk
Update Set of RBs: Set_RB = Set_RB \ {n}
% Update the local pheromone
Update τk,n: τk,n=τk,n + τ0, with n∈Nk
Update τk,n: τk,n=τk,n- ρτk,n, with n∈Nk
% Update the global pheromone
Calculate the judge function of UEk
Find (k) = max judge function (k),with k∈ Set_UE
Update τk,n: τk,n = τk,n + Δτk,nbest, with n∈Nk

3 | S I M U L A TI O N R E S U L T S A N D D I S C U S S IO N
In this section, we introduce the simulation results of our proposed resource allocation algorithm as well as the two
resource allocation algorithms: position‐based resource allocation,31 location‐dependent resource allocation,32 and

TABLE 2 Simulation parameters

Parameter Value

Number of eNodeB 1
Cell radius 1.5 km
System bandwidth 5 MHz
Number of RBs 25 RBs
Shadow fading deviation 9 dB
Carrier frequency 2 GHz
OFDM symbols per slot 6
Traffic model for CUEs VoIP, video, and FTP
FTP delay threshold 300 ms
Video delay threshold 150 ms
VoIP delay threshold 100 ms
C‐UEs speed Between 5 and 150 (km/h)
V‐UEs speed Between 30 and 150 (km/h)
Maximum UEs transmit power 23 dBm
Total number of V‐UEs/C‐UEs 100‐500
Simulation length 5000 TTI
TTI length 1 ms
Scheduling/allocation resource Per TTI
τ0 1
α (of C‐UEs) 3
β (of C‐UEs) 1
α (of V‐UEs) 1
β (of V‐UEs) 1
ρ 0.1
μ 0.5
FEKI ET AL. 11 of 16

TABLE 3 Use cases of V2V services41

V2V Service Scenarios Description Maximum Latency Message Size

Warning of emergency This service enables every vehicle to obtain the speed, location, 50 ms 50‐300 bytes
vehicle and direction information of a surrounding emergency vehicle
to assist safety operation.
Emergency stop This service describes V2V communication used in emergency 50 ms 50‐300 bytes
stop case to launch safer behavior for other vehicles near the
stationary vehicle.
Warning of pre‐crash This service gives warnings to vehicles in imminent and 20 ms 50‐300 bytes
sensing unavoidable collision.

random resource allocation.33 We consider a simulation model with a total number of C‐UEs and V‐UEs users varying
between 100 and 500 where the number of V‐UEs is 50% of the total number of users. C‐UEs have random distribution
and positions inside the cell and V‐UEs move in freeway scenario. The users' locations should be updated every 100 ms
during the simulation. The simulation parameters are presented in Table 2. The use cases implemented in vehicular
traffic are presented in Table 3.
In this section, the network sum rate, the fairness index, the resource utilization ratio, and the PDR in each scenario
are presented and analyzed.
Figures 5 and 6 depict the influence of ACO parameters α and β. The choice of these parameters will indeed affect
the performance of the proposed scheme. We note that when we increase β (ie, the heuristic information influence) on
the detriment of α, the network sum rate of C‐UEs increases while the fairness index of C‐UEs decreases. As shown in
Figures 5 and 6, when taking different values of α and β for C‐UEs, the margin between the two curves is more

FIGURE 5 Network sum rate in ACORA algorithm

FIGURE 6 Fairness index in ACORA algorithm


12 of 16 FEKI ET AL.

important in the fairness index comparing with that the sum rate. Thus, we deduce that selecting α = 3 and β = 1 pro-
vides the best tradeoff network sum rate/fairness index. For V‐UEs, we notice that the pheromone and heuristic infor-
mation have the same target, improving network sum rate. Thus, for V‐UEs, we select α = 1 and β = 1.
Figures 7 and 8 show the average user throughput of C‐UEs and V‐UEs inside the cell. Our algorithm reaches the best
sum rate as it considers the channel state of both C‐UEs and V‐UEs. In the scheduling phase of C‐UEs, it utilizes the
required data rate and the propagation channel condition to maximize system capacity and to keep balance between
throughput and equity. Like the resources allocation phase, it utilizes efficiently the radio resource since it calculates
the heuristic information based on the number of bits that can be transmitted. As for the judge function, user with the
lowest OP has the best allocation. With an increased number of users (up to 500 users), the location‐dependent resource
allocation becomes more efficient in terms of network sum rate, number of served packets (Figure 11), and PDR
(Figure 12). In fact, this algorithm implements the non‐orthogonal sharing mode (ie, both V2V and cellular traffics use
the same resources) where a RB is used, simultaneously, by a V‐UE and C‐UE. This mode of sharing causes the problem
of interference management. As expected, the random allocation and position‐based resource allocation provides the
worst sum rate. The position‐based resource allocation assigns statically a number of RBs to each set of UEs. Moreover,
it allocates resources independently of users' requirements and channel response.
ACORA considers the fairness metric in both scheduling and allocation phases of C‐UEs. The resources allocation
process is triggered according to the channel state of all users in the network. In order to avoid discrimination between
users with poor or strong channel quality, it considers all the variation in the channel state by using the global phero-
mone, which brings more fairness. This explains the results shown in Figures 9 and 10, where the proposed algorithm
gives the best fairness rates.
In D2D communication, assigning a static number of RBs to each type of traffic or spatial area yields a spec-
trum resource waste. The proposed algorithm allocates RBs in adaptive way. It calculates the number of RBs

FIGURE 7 Network sum rate of V‐UEs

FIGURE 8 Network sum rate of C‐UEs


FEKI ET AL. 13 of 16

FIGURE 9 Fairness index of V‐UEs

FIGURE 10 Fairness index of C‐UEs

assigned to cellular traffic according to the PDR rate of C‐UEs. In fact, when achieving a certain PDR threshold,
the scheduler assigns more RBs to cellular traffic. Thus, according to Figures 11 and 12, our proposed algorithm
achieves the best PDR rate. On the one hand, the smart management of the BRs decreases the PDR rate. On
the other hand, the high rate of OP may increase the PDR rate while our proposed algorithm assigns RBs to
UEs with lowest outage probability.
As shown in Figure 13, the queuing delay of vehicular traffic with the proposed approach is better. In ACORA, we
consider the waiting time in scheduling phase using the EDF scheduler, so we give priority to V‐UEs according to the
latency as it is a latency intolerant traffic.

FIGURE 11 Number of served packets


14 of 16 FEKI ET AL.

FIGURE 12 PDR

FIGURE 13 Average queuing delay of V‐UEs

4 | CONCLUSION

Maximizing the network sum rate is a target characteristic of resource allocation schemes. However, there are other cru-
cial QoS requirements that have to be considered. A trade‐off between QoS requirements and performance, when con-
ceiving a resource allocation algorithm in wireless networks, is a challenge. In this paper, an ACO‐based system model
for packet scheduling and radio resource allocation is proposed in overlay mode to maximize the overall network sum
rate while satisfying the QoS requirements of both C‐UEs and V‐UEs. In this algorithm, two sets of RBs are adaptively
assigned to each class of users, where the V‐UEs utilize orthogonal RBs between each other and to communicate with
the eNodeB, and the C‐UEs utilize orthogonal RBs. Then, a packet scheduling process is performed using the propor-
tional fair scheduler for C‐UEs and the EDF scheduler for V‐UEs. By using mathematical constraints, the resource allo-
cation process was elaborated as an optimization issue, taking into consideration the C‐UEs and V‐UEs requirements.
The proposed algorithm is compared with well‐known resource allocation algorithms from the literature. Simulation
results indicate that our approach has significant performance gain. It yields not only better cellular and vehicular sum
rates as well as fairness performance, but also improves PDR compared with the considered existing competing
schemes. When the network becomes more congested and the number of users exceeds 400 users, the location‐depen-
dent resource allocation algorithm outperforms the proposed ACORA and achieves better cellular sum rate, PDR, and
average queuing delay for vehicular traffic. As this algorithm adopts the underlay mode, resource can be used simulta-
neously by V‐UEs and C‐UEs. This mode improves the spectrum utilization but causes co‐channel interference that
needs to be well managed to avoid the performance degradation of either V‐EUs or C‐EUs.
Furthermore, when analyzing the runtime complexity of our algorithm as compared with other resource allocation
algorithms, we found that our runtime complexity is much lower and thus optimized due primarily to the pheromone
FEKI ET AL. 15 of 16

behaviour. In fact, the use of pheromone narrows considerably the search space for optimal solution by taking into
account the optimum made allocations in the previous TTIs. Thus, the expected runtime until reaching the optimal
solution is much lower, and the complexity of the proposed scheme is significantly lower than that of other schemes,
which make it an attractive and efficient resource allocation scheme suitable for D2D communication underlying
LTE/LTE‐A networks.
In the current work, the proposed resource allocation algorithm is designed for overlay mode. In the future work,
the proposed ACORA will be combined with a resource sharing algorithm, since resource sharing becomes a necessity
and crucial phase when designing resource allocation algorithms.

ORCID
Souhir Feki http://orcid.org/0000-0001-8177-0645
Faouzi Zarai http://orcid.org/0000-0001-9250-7885

R EF E RE N C E S
1. Dorsemaine B, Gaulier J‐P, Wary J‐P, Kheir N, Urien P. Internet of Things: a definition & taxonomy. In: Proceedings of the International
Conference on Next Generation Mobile Applications, Services and Technologies; 2015; Cambridge, UK.
2. Othmen S, Zarai F, Belghith A, Obaidat MS, Kamoun L. Secure and reliable multi‐path routing protocol for multi‐hop wireless networks.
J Ad Hoc Sensor Wireless Networks. 2017;36(1– 4):127‐147.
3. 3GPP TS V0.1.4. Overview of 3GPP Release 12; 2014.
4. Wireless technology evolution towards 5G: 3GPP release 13 to release 15 and beyond, white paper, 5G Americas; 2017.
5. 3GPP TR 36.885 V14.0.0. Study on LTE‐based V2X services; 2016.
6. 3GPP TR 22.885 V14.0.0. Study on LTE support for V2X services (Release 14); 2015.
7. Silva CM, Masini BM, Ferrari G, Thibault I. A survey on infrastructure‐based vehicular networks. Mob Inf Syst. 2017;2017:1‐28.
8. Seo H, Lee KD, Yasukawa S, Peng Y, Sartori P. LTE evolution for vehicle‐to‐everything services. IEEE Commun Mag. 2016;54(6):22‐28.
9. Sun SH, Hu JL, Peng Y, Pan XM, Zhao L, Fang JY. Support for vehicle‐to‐everything services based on LTE. IEEE Wirel Commun.
2016;23(3):4‐8.
10. Bazzi A, Masini BM, Zanella A, Thibault I. On the performance of IEEE 802.11p and LTE‐V2V for the cooperative awareness of con-
nected vehicles. IEEE Trans Veh Technol. 2017;66(11):10419‐10432.
11. Othmen S, Rekik M, Zarai F, Belghith A, Kamoun L. Shortest and secure routing protocol for multi‐hop cellular networks (SSRP‐MCN).
Sec Comm Networks. 2016;9(12):5346‐5362.
12. Feki S, Zarai F, Belghith A. A Q‐learning‐based scheduler technique for LTE and LTE‐advanced network. In: Proceedings of the Interna-
tional Conference on Wireless Networks and Mobile Systems; 2017; 27–35, Madrid, Spain.
13. Feki S, Zarai F. Cell performance‐optimization scheduling algorithm using reinforcement learning for LTE‐advanced network. In: Pro-
ceedings of the IEEE International Conference on Computer Systems and Applications; 2017; 1075–1081, Hammamet, Tunisia.
14. 3GPP TS 36.300. Evolved UTRA and evolved UTRAN, overall description; 2008.
15. Khdhir R, Mnif K, Belghith A, Kamoun L. Tabu approach for adaptive resource allocation and selection carrier aggregation in LTE‐
advanced network. In: Proceedings of the IEEE International Conference on Computer and Information Dent Technology; 2016;
347–353, Nadi, Fiji.
16. Khdhir R, Mnif K, Belghith A. New adaptive resource allocation scheme in LTE‐advanced. In: Proceedings of the Global Conference for
Wireless and Optical Communications; 2016; 749‐759, Málaga, Spain.
17. Belghith A, Belhaj Mohamed M, Obaidat MS. Efficient bandwidth call admission control in 3GPP LTE networks. In: Proceedings of the
IEEE Global Communication Conference; 2016; 1–6, Washington, DC, USA.
18. Belghith A, Turki N, Cousin B, Obaidat MS. Flexible call admission control with preemption in LTE networks. In: Proceedings of the
International Conference on Communications; sMay 2016; 1‐7, Kuala Lumpur, Malaysia.
19. Trabelsi S, Belghith A, Zarai F, Obaidat MS. Performance evaluation of a decoupled‐level with QoS aware downlink scheduling algorithm
for LTE networks. In: Proceedings of the IEEE International Conference on Internet of Things, Sydney; 2015; 696‐704, Australia.
20. Zulhasnine M, Huang C, Srinivasan A. Efficient resource allocation for device‐to‐device communication underlaying LTE network. In:
Proceedings of the IEEE International Conference on Wireless and Mobile Computing, Networking and Communications; 2010;
368–375, Niagara Falls, NU, Canada.
16 of 16 FEKI ET AL.

21. Algedir A, Refai HH. Adaptive D2D resources allocation underlaying (2‐tier) heterogeneous cellular networks. In: Proceeding of the IEEE
International Symposium on Personal, Indoor and Mobile Radio Communications; October 2017; 1‐6, Montreal, Canada.
22. Zhang R, Cheng X, Yao Q, Wang CX, Yang Y, Jiao B. Interference graph based resource sharing schemes for vehicular networks. IEEE
Trans Vehicul Technol. 2013;62(8):4028‐4039.
23. Feng D, Lu L, Yuan‐Wu Y, Li GY, Feng G, Li S. Device‐to‐device communications underlaying cellular networks. IEEE Trans Commun.
2013;61(8):3541‐3551.
24. Esmat HH, Elmesalawy MM, Ibrahim II. Uplink resource allocation and power control for D2D communications underlaying multi‐cell
mobile networks. Int J Electron Commun. September 2018;93:163‐171.
25. Meshgi H, Zhao D, Zheng R. Optimal resource allocation in multicast device‐to‐device communications underlaying LTE networks. IEEE
Trans Veh Technol. September 2017;66(9):8357‐8371.
26. Wei Q, Wang L, Feng Z, Ding Z. Cooperative coexistence and resource allocation for V2X communications in LTE‐unlicensed. In: Pro-
ceeding of the Consumer Communications & Networking Conference; January 2018; 1‐6, Las Vegas, USA.
27. Ramezani‐Kebrya A, Dong M, Liang B, Boudreau G, Seyedmehdi SH. Joint power optimization for device‐to‐device communication in
cellular networks with interference control. IEEE Trans Wirel Commun. 2017;16(8):5131‐5146.
28. Sun W, Strom EG, Brannstrom F, Sou KC, Sui Y. Radio resource management for D2D‐based V2V communication. IEEE Trans Veh
Technol. 2016;65(8):6636‐6650.
29. Sun W, et al. D2D‐based V2V Communications with Latency and Reliability Constraints. In: Proceeding of the IEEE Globecom Work-
shops; December 2014; 1414–1419, Austin, TX, US.
30. Sun W, Yuan D, Strom EG, Brannstrom F. Cluster‐based radio resource management for D2Dsupported safety‐critical V2X Communica-
tions. IEEE Trans Wireless Comm. 2016;15(4):2756‐2769.
31. Kim J, Lee J, Moon S, Hwang I. A position‐based resource allocation scheme for V2V communication. Wireless Personal Comm, Springer
US. 2017;1‐18.
32. Botsov M, Klugel M, Kellerer W, Fertl P. Location dependent resource allocation for mobile device‐to‐device communications. In: Pro-
ceedings of the IEEE Wireless Communications and Networking Conference; May 2014; 1679–1684, Istanbul, Turkey.
33. 3GPP R1‐135481. Discussion on resource allocation in D2D communications. San Francisco, USA; 2013.
34. Hinchey MG, Sterritt R, Rouff C. Swarm and swarm intelligence. Computer. 2007;40(4):111‐113.
35. Goss S, Aron S, Deneubourg J‐L, Pasteels JM. Self‐organized shortcuts in the argentine ant. Naturwissenschaften. 1989;76(12):579‐581.
36. Aleksić D et al. Outage probability comparison of MRC, EGC and SC receivers over short term fading channels. Int J Commun.
2016;1:104‐109.
37. Caire G, Taricco G, Biglieri E. Optimum power control over fading channels. IEEE Trans Inf Theory. July 1999;45(5):1468‐1489.
38. Rodrigo F, Cavalcanti P. Resource Allocation and MIMO for 4G and Beyond. New York, USA, ISBN: 978‐1‐4614‐8056‐3: Springer‐Verlag;
2014.
39. Kwan R, Leung C, Zhang J. Proportional fair multiuser scheduling in LTE. IEEE Signal Process. 2009;16(6):461‐464.
40. Liu D, Lee Y‐H. An efficient scheduling discipline for packet switching networks using earliest deadline first round robin. In: Proceedings
of the International Conference on Computer Communication and Networks; July 2003; 5–10, Dallas, USA.
41. 3GPP TR 23.785 V14.0.0. Study on architecture enhancements for LTE support of V2X services (Release 14); 2016.

How to cite this article: Feki S, Masmoudi A, Belghith A, Zarai F, Obaidat MS. Swarm intelligence‐based radio
resource management for D2D‐based V2V communications. Int J Commun Syst. 2018;e3817. https://doi.org/
10.1002/dac.3817

You might also like