IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 68, NO.

5, MAY 2019 5005

Ultra-Reliable Low-Latency Communications in

Autonomous Vehicular Networks
Xiaohu Ge , Senior Member, IEEE

Abstract—Autonomous vehicles are expected to emerge as a main a solution for improving both reliability and latency in 5G
trends in vehicle development over the next decade. To support autonomous vehicular networks, i.e., implementing URLLC is
autonomous vehicles, ultra-reliable low-latency communications a considerable challenge.
(URLLC) is required between autonomous vehicles and infras-
tructure networks, e.g., a fifth-generation (5G) cellular networks. Studies have investigated methods of improving the latency or
Hence, reliability and latency must be jointly investigated in 5G reliability of vehicular networks [8]–[17]. A mobile-edge com-
autonomous vehicular networks. In this paper, utilizing the Eu- puting (MEC) architecture was proposed for cellular vehicular
clidean norm theory, we first propose a reliability and latency joint networks in which an inter-cell handover mechanism was devel-
function to evaluate the joint impact of reliability and latency in oped for vehicles to enhance the latency performance of mobility
5G autonomous vehicular networks. The interactions between re-
liability and latency are illustrated via Monte Carlo simulations management [8]. Based on software-defined networks (SDNs)
of 5G autonomous vehicular networks. To improve both the reli- and OpenFlow technologies, a SDN-enabled network architec-
ability and latency performance and implement URLLC, a new ture assisted by MEC was proposed to reduce the latency in
network slicing solution that extends from resource slicing to ser- vehicular networks [9], [10]. A multicast transmit beamforming
vice and function slicing is presented for 5G autonomous vehicular technology was developed for vehicle to everything (V2X) com-
networks. The simulation results indicate that the proposed net-
work slicing solution can improve both the reliability and latency munications that employs long-term evolution (LTE) multime-
performance and ensure URLLC in 5G autonomous vehicular dia broadcast single-frequency network (MBSFN) capabilities
networks. [11]. To reduce the latency in multi-hop vehicular networks, a
Index Terms—Reliability, latency, autonomous vehicle, vehicu-
new scheme was proposed to optimize the one hop transmis-
lar network, network slicing. sion range based on a genetic algorithm [12]. Considering a
limited number of roadside units (RSUs) along roads, a routing
scheme was proposed for broadcast-based safety applications
with ultra-reliability in vehicular ad hoc networks (VANETs)
I. INTRODUCTION [13]. The connectivity probability was analyzed for platoon-
EHICULAR networks are emerging as a key application based VANETs in which vehicles in the network have a Pois-
V scenario for fifth generation (5G) mobile communication
systems [1], and in the next decade, autonomous vehicles
son distribution considering different traffic densities [14]. The
simulation results showed that the connectivity probabilities in
will represent one of the main transmitters/receivers of 5G VANETs based on platoons are larger than those in VANETs
vehicular networks [2]. Compared with traditionally manned without platoons in the V2X communication scenarios. To im-
vehicles, autonomous vehicles are extremely dependent on prove the network capacity and system computing capability, a
ultra-reliable low-latency communication (URLLC) in 5G matrix game approach was developed to manage the cloudlet
vehicular networks. Previous studies have indicated that con- resources of vehicular networks [15]. A three-stage radio re-
flicts can occur between reliability and latency performance in source management algorithm was developed to optimize re-
vehicular networks [3], [4]. Hence, many studies have focused source sharing among vehicular users and cellular users in V2X
on optimizing either the reliability or latency performance in ve- applications [16]. By formulating the virtual resource alloca-
hicular networks [5]–[7]. Considering the URLLC requirement, tion and caching strategies as a joint optimization problem, a
a joint model of reliability and latency must be investigated new framework with information-centric wireless virtualization
for 5G autonomous vehicular networks. Moreover, developing and device-to-device communications was proposed to enable
content caching not only in the air but also in mobile devices
[17]. However, enhancing both the reliability and latency of
Manuscript received November 25, 2018; revised January 31, 2019; accepted
March 4, 2019. Date of publication March 8, 2019; date of current version autonomous vehicular networks remains a key issue. To re-
May 28, 2019. This work was supported by the National Key Research and solve this key issue, the network slicing technology is emerg-
Development Program of China under Grant 2017YFE0121600. The review of ing as a potential solution for URLLC in autonomous vehicular
this paper was coordinated by Prof. X. Fang.
The author is with the School of Electronic Information and Communica- networks.
tions, Huazhong University of Science and Technology, Wuhan 430074, China Numerous network slicing concepts and methods have been
(e-mail:,xhge@mail.hust.edu.cn). proposed in the literatures on wireless networks [18]–[28]. The
This paper has supplementary downloadable material available at http://
ieeexplore.ieee.org, provided by the author. basic principles of network slicing that underly the mapping of
Digital Object Identifier 10.1109/TVT.2019.2903793 dedicated and shared slices were discussed in [18] considering

0018-9545 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: BMS College of Engineering. Downloaded on May 15,2024 at 09:11:37 UTC from IEEE Xplore. Restrictions apply.
5006 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 68, NO. 5, MAY 2019

the necessary flexibility and scalability associated with 5G net- networks, and a maximum of reliability and latency joint
work implementation. A common framework was presented to function with respect to the vehicle density is validated
integrate and discussing the latest developments in 5G network via Monto-Carlo (MC) simulations.
slicing, and the identified gaps were evaluated [19]. Network 2) To improve both the reliability and latency performance
slicing was introduced as an integral approach to wireless net- of 5G autonomous vehicular networks, i.e., to implement
work virtualization and shown to promote the programmable URLLC, a new network slicing solution that extends
and configurable characteristics of network services [20]. A from resource slicing to service and function slicing is
mobile-oriented Open-Flow protocol (MOFP) was proposed for presented. Moreover, a new network slicing algorithm
the sharing of resources at adjacent base stations (BSs) and the is developed to improve both the reliability and latency
implementation of service slicing for 5G cellular networks [21]. performance of 5G autonomous vehicular networks.
In [22], network slicing technology was proposed to satisfy dif- 3) The simulation results indicate that the proposed network
ferent network slicing requirements and content by dynamically slicing solution can improve both of the reliability and la-
scheduling the edges and central clouds in the studied wireless tency performance of 5G autonomous vehicular networks,
network. Based on network slicing technology, the virtual re- i.e., by implementing URLLC. Moreover, the optimization
source allocation of full-duplex relaying (FDR) networks was of the proposed network slicing solution depends on the
formulated as an optimization problem, and an efficient alternat- vehicle density in the 5G autonomous vehicular networks.
ing direction method of multipliers (ADMM)-based distributed The remainder of this paper is outlined as follows. Section II
virtual resource allocation algorithm was developed to solve this describes the system model of autonomous vehicular networks.
problem [23]. The virtual network functions (VNFs) placement Section III investigates the coupled relationship between the re-
problem was formalized for radio access networks, and a slice liability and latency in 5G autonomous vehicular networks based
scheduling mechanism was proposed to ensure resource and per- on the proposed reliability and latency joint function. Addition-
formance isolation among different slices [24]. The concept of a ally, the reliability and latency performance are analyzed via MC
5G network slice broker was introduced for 5G networks, and it simulations for 5G autonomous vehicular networks. Section IV
enables mobile virtual network operators, over-the-top providers presents the proposed network slicing solution for improving the
and industry vertical market players dynamically to request and reliability and latency in 5G autonomous vehicular networks, in
lease resources from infrastructure providers via signaling [25]. which network slicing is extended from network resource slic-
In [26], the bandwidth of wireless networks was formed as a ing to service and function slicing. Section V compares and ana-
type of virtual slice by wireless network virtualization functions. lyzes the reliability and latency performance of 5G autonomous
Furthermore, [27] presented a framework that enables wireless vehicular networks with and without network slicing. Finally,
virtualization and discussed a number of challenges that must Section VI presents the conclusions.
be addressed for the deployment of wireless virtualization in
the next generation of mobile cellular networks. Network slic-
ing technology has been introduced into vehicular networks by
utilizing resource sharing schemes. Based on SDN and fog com- II. SYSTEM MODEL
puting technologies, a new vehicular network architecture was Without loss of generality, an urban outdoor environment is
proposed to improve the coverage probability of the network projected in a plane R2 , in which the roads are modeled by the
slicing technology in 5G vehicular networks [28]. However, in- Manhattan Poisson line process (MPLP) [29], [30]. In Fig. 1,
vestigations of URLLC in autonomous vehicular networks are the road distribution model features two unit-density homoge-
surprisingly rare in the open literatures. Additionally, the im- neous Poisson point processes (PPPs) Ψx , Ψy ⊂ R2 along the
pact of network slicing technology on the reliability and latency x-axis and y-axis, respectively. At each point in the processes,
performance of vehicular networks has been limited to simple an avenue (west-east direction) and a street (north-south direc-
scenarios, and methods of improving both the reliability and la- tion) grows infinitely along the x-axis and y-axis, respectively.
tency based on network slicing have not been investigated in 5G The resulting line process is denoted by , which divides the
autonomous vehicular networks. plane into an infinite number of blocks. Based on the MPLP
Motivated by the above research gaps, a reliability and latency road model, the RSUs are assumed to be uniformly distributed
joint function is proposed in this paper to evaluate the joint im- along the roads in Fig. 1.
pact of reliability and latency on 5G autonomous vehicular net- The infrastructure providers (InPs) I = {Inpa , Inpb , ...,
works. Moreover, a new network slicing solution with service, Inpi } in the plane R2 are assumed to be governed by an in-
function and resource slicing is proposed to improve both the dependent PPP distribution. The plane R2 is split into irregular
reliability and latency of 5G autonomous vehicular networks. polygons that correspond to different cell coverage areas. The
The contributions and novelties of this paper are summarized as split method is based on the Delaunay Triangulation method
follows. where the perpendicular bisector lines are connected by each
1) Based on the Euclidean norm theory, a reliability and la- pair of InPs [31]. This stochastic and irregular topology forms a
tency joint function is proposed to evaluate the joint impact so-called Poisson-Voronoi tessellation (PVT) [32]. In each cell
of the reliability and latency on 5G autonomous vehicular coverage area, an InP is located in the center of the cell and
networks. Moreover, the interactions between reliability connects with all RSUs inside this cell coverage area. The InP
and latency are analyzed for 5G autonomous vehicular gathers information from the RSUs and provides a virtualized

Authorized licensed use limited to: BMS College of Engineering. Downloaded on May 15,2024 at 09:11:37 UTC from IEEE Xplore. Restrictions apply.
GE: ULTRA-RELIABLE LOW-LATENCY COMMUNICATIONS IN AUTONOMOUS VEHICULAR NETWORKS 5007

Fig. 1. System model. Coverage areas of different hot spots are shown as circles with different radius, sij is the total number of AEs in circles marked as the
purple regions.

network platform (VNP) to share network resources with the A. Reliability and Latency Joint Function
RSUs in the PVT cell.
Based on the Euclidean norm theory, the reliability and la-
Based on the Palm theory [32], [33], the analytical results tency joint function of 5G autonomous vehicular networks
for a typical PVT cell, e.g., Areae in Fig. 1, can be extended
S (Lt , Lp ) is defined as follows:
to the entire PVT network. Hence, Areae is magnified in Fig. 1
to show the distribution details in a typical PVT cell. In Areae ,  1/2
S (Lt , Lp ) = |Lt |2 + |ω · Lp |2
the RSUs have network resources that can be used to communi-
cate with autonomous vehicles on the road by millimeter wave  1/2
wireless transmissions. Moreover, the beamforming technology = |It (Treq )|2 + |ω · Ip (Preq )|2 (1)
is adopted for wireless links between the RSUs and vehicles to
avoid the interference from adjacent RSUs. For each RSU, a where Lp and Lt are the reliability and latency utilities in
RSU to infrastructure provider (R2I) link is configured to pro- autonomous vehicular networks, respectively; ω is the weight
vide communications between the RSU and InP. factor that balances the effects of reliability and latency on
S (Lt , Lp ); Ip (Preq ) is the reliability utility function with the
reliability constraint Preq ; and It (Treq ) the latency utility func-
III. RELIABILITY AND LATENCY JOINT FUNCTION IN 5G tion with the latency constraint Treq , |·| is the norm operation.
AUTONOMOUS VEHICULAR NETWORKS Based on the reliability and latency joint function, both the re-
Most studies have separately modeled and analyzed the re- liability and latency performance can be evaluated by the same
liability and latency performances of vehicular networks [34]; metric, i.e., the norm length in Euclidean space. Therefore, the
however, these parameters are coupled in vehicular networks. URLLC can be investigated based on a uniform model of the re-
When an algorithm is applied to optimize the latency or reliabil- liability and latency joint function in 5G autonomous vehicular
ity performance of vehicular networks, the other performance networks.
parameters in the vehicular network will be affected, and per- To evaluate how reliability and latency are satisfied in au-
formance deterioration can occur [35]. Considering the URLLC tonomous vehicular networks, the reliability and latency utility
requirement in 5G autonomous vehicular networks, the inter- functions are extended as follows:
 
P −Preq
actions between reliability and latency must be modeled and aP
Ip (Preq ) = e Preq
, (2)
analyzed. In this section, a new reliability and latency joint  
Treq −T
function based on the Euclidean norm theory is proposed for It (Treq ) = e
aT Treq
, (3)
5G autonomous vehicular networks. Furthermore, the coupled
relationship between reliability and latency is analyzed for 5G where P and T are the reliability and latency values in 5G au-
autonomous vehicular networks. tonomous vehicular networks, respectively; Preq and Treq are

Authorized licensed use limited to: BMS College of Engineering. Downloaded on May 15,2024 at 09:11:37 UTC from IEEE Xplore. Restrictions apply.
5008 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 68, NO. 5, MAY 2019

the required reliability and latency thresholds for the desired In this paper, beamforming technology is adopted for mil-
services in 5G autonomous vehicular networks, respectively; limeter wave transmissions. Considering the directionality and
and aP and aT are the weighted factors for the reliability and fast fading features of millimeter wave transmissions, the inter-
latency utility functions, respectively. When the values of reli- ference among vehicles and RSUs is ignored in 5G autonomous
ability and latency are less than the required thresholds in au- vehicular networks. The threshold of the signal-to-noise ratio
tonomous vehicular networks, the reliability and latency utility (SNR) at the receivers is configured as θ in one hop in a 5G au-
function values are always larger than 0 and less than 1, i.e., tonomous vehicular network. When the SNR at the receivers is
0 < It (Treq ) < 1 and 0 < Ip (Preq ) < 1. When the reliability larger than or equal to the threshold θ, the signal can be success-
and latency requirements are satisfied in autonomous vehicu- fully accepted at the receivers [39], [40]. The one hop successful
lar networks, i.e., the reliability and latency values are larger transmission probability Phop that the message can be success-
than or equal to the required thresholds, the reliability and la- fully transmitted in a time slot is as follows [41]:
tency utility function values are always larger than or equal to 1,
Phop = P (L ≤ Ptx (dB) − θ(dB) − N0 WmmW ave (dB)) ,
i.e., It (Treq ) ≥ 1 and Ip (Preq ) ≥ 1. Moreover, the reliability
(6)
utility function value increases as the reliability increases, and
the latency utility function value increases as the latency in au- where Ptx is the transmission power, N0 is the power spec-
tonomous vehicular networks decreases. In this case, the value of trum density of AWGN, and WmmW ave is the bandwidth of
S (Lt , Lp ) increases as the reliability and latency improve in 5G the millimeter wave transmissions. By substituting (5) into (6),
autonomous vehicular networks. When the value of S (Lt , Lp ) and the one hop successful transmission probability Phop can be
is less than 1, either the reliability or latency requirement has expressed as follows:
not been satisfied in the 5G autonomous vehicular networks.
Phop = P (ξ  Ptx (dB) − θ(dB)
B. Reliability Model −N0 WmmW ave (dB) − 69.6 − 20.9log10 d) ,
The unreliability of messages in 5G autonomous vehicular  
1 ψ (d)
networks is mainly caused by transmission errors in wireless = 1 + erf √ (7)
2 2σ
links between the vehicles and RSUs. In this case, the reliability
of autonomous vehicular networks is denoted by the probability where erf () is the deviation function and ψ (d) = Ptx (dB) −
of the message being successfully transmitted between the vehi- θ(dB) − N0 WmmW ave (dB) − 69.6 − 20.9log10 d.
cle and the RSU. The wireless link between the vehicle and RSU Considering the multi-hop process in a wireless link, the re-
is denoted by L, and the number of hops is NL . The distance liability of 5G autonomous vehicular networks is calculated as
between two nodes in the wireless link L is denoted by Di,j x
, follows:
x ∈ {V, R}, where i and j are positive integers, Di,j denotes
V  x 
P= Phop Di,j . (8)
R
the distance between two vehicles and Di,j denotes the distance x ∈L
Di,j
between the vehicle and the RSU. If the vehicle node is denoted
as DiV and the RSU node is denoted as DjR , then the wireless C. Latency Model
link L between the vehicle DiV and RSU DjR is expressed as a
set as follows: The messages between vehicles and adjacent RSUs are trans-
 V  mitted by vehicle to infrastructure (V2I) links. The RSU normal-
L = Di,1 V
, D1,2 , · · · , Dn,j
R
, (4) izes all received messages and forms a message queue to manage
all vehicle messages in the coverage area. Without loss of gen-
where n is a positive integer.
erality, the total latency between the vehicles and RSUs can be
In this paper, millimeter wave transmissions are adopted for
divided into propagation latency and handling latency. Hence,
the wireless links in 5G autonomous vehicular networks, and the
the total latency of message T can be expressed as follows:
frequency of the millimeter wave transmissions is 75 GHz [36],
[37]. Without loss of generality, every RSU is assumed to cover T = T tpM + T tqM , (9)
a road with a length of L meters in an urban environment. The
density of RSUs in an urban area is denoted as ρRSU = L1 . The where T tpM is the propagation latency in the wireless links and
vehicle density on the road is configured as ρi . When a vehicle T tqM is the handling latency in the queues of RSUs.
Va is covered by RSUi , the distance d between the vehicle and When the wireless link between the vehicle and the RSU com-
the RSU is assumed to be governed by a uniform distribution, prises one wireless hop, the propagation latency is expressed as
i.e., d ∈ U (0, L), with the expectation that E (d) = L /2 . Based follows:
on the measured results in [37], [38], the path loss P L[dB](·) 2tslot
Thop (d) = tslot /Phop =  , (10)
between the vehicle Va and the RSU RSUi can be expressed as 1 + erf ψ(d)
√
follows: 2σ
  where tslot is a constant time slot used for transmitting messages.
P L[dB](d) = 69.6 + 20.9 log(d) + ξ, ξ ∼ 0, σ 2 , (5)
When the wireless link L between the vehicle and the RSU
where ξ is additive white Gaussian noise (AWGN) with a mean is composed of multiple hops, i.e., a message from the vehicle
of 0 and variance σ 2 . is relayed via multiple vehicles to reach the desired RSU, the

Authorized licensed use limited to: BMS College of Engineering. Downloaded on May 15,2024 at 09:11:37 UTC from IEEE Xplore. Restrictions apply.
GE: ULTRA-RELIABLE LOW-LATENCY COMMUNICATIONS IN AUTONOMOUS VEHICULAR NETWORKS 5009


propagation latency of multi-hop T tpM in the wireless link L is 1 k=0
Dk = k (15)
expressed as follows: εl
l=1 1−εl k = 1, 2, · · · , c
 ∞
T tpM = Ttrans +Tproc
εl = e−lμt dF (t) , l = 1, 2, · · · . (16)
 x  0
= Thop Di,j + (NL − 1) tproc (11)
x ∈L
Di,j
The proof for Theorem 2 is provided in Appendix B.
Furthermore, the average dwelling time of a message in the
where Ttrans is the transmission latency in the wireless link GI/M/1/∞ queuing system is derived as follows:
L and Tproc is the total relaying latency, which includes the 1 K∗
processing latency tproc of every relaying vehicle in the wireless T tqS = E [W (t)] = + . (17)
μ cμ(1 − δ)2
link L.
Every RSU is configured with the same number of local re- Considering the GI/M/1/∞ queuing system at the RSU, the
source blocks RBrsu . Considering the random mobility of ve- parameter c is set as a constant, i.e., c = 1. Hence, the average
hicles on the roads, the number of vehicles in the coverage area dwelling time of a message is derived as follows:
of every RSU is different, as is the arrival rate of messages 1
at every RSU. In this case, the message handling process at T tqS = E [W (t)] = . (18)
μ (1 − δ)
the RSU is assumed as a GI/M/1/∞ queuing system. In this
GI/M/1/∞ queuing system, it is assumed that τ0 = 0 is the When the value of δ is limited as δ ∈ (0, 1), δ can be solved
arrival epoch of the first task; the inter-arrival times {τi , i ≥ 1} based on the following relation:
are independent and identically distributed with a general dis- ∞  ∞  ∞
(μt)k
tribution function denoted by F (t) , t ≥ 0; and the mean is 0 < δk · e−μt dF (t) = e−μ(1−δ)t dF (t) = δ.
∞ k!
λ = 0 tdF (t), (λ > 0), where λ is the arrival rate. The ser-
1 0 0
k=0
vice times {χi , i ≥ 1} during a service period are exponentially (19)
distributed at the rate of μ (μ > 0); therefore, the distribution When the CDF of message interval time in (12) is substituted
function of service times is denoted by G (t) = 1 − e−μt , t ≥ 0. into (19), (19) can be extended as follows:
Considering the GI/M/1/∞ queuing system, the average ser-  ∞ −λs t
vice time of a message at RSUs is dependent on the number of λs ρi Le−ρi L+ρi Le −λs t
e−μ(1−δ)t dt = δ. (20)
resource blocks in the RSU [42]. To simplify the analysis, μ10 0 1 − e−ρi L
denotes the average service time when a message is serviced by a
resource block, and RBrsu1 (i)·μ0 is the average service time when D. Performance Analysis of Reliability and Latency
a message is serviced by RSUi with RBrsu (i) resource blocks. To analyze the reliability and latency performance of 5G au-
In the GI/M/1/∞ queuing system, the handling latency of a tonomous vehicular networks, the default simulation parameters
message at an RSU corresponds to the following theorems. are configured as follows: the weight factor ω is set to 10, the
Theorem 1: Without loss of generality, RSUi (i = 1, 2, road length L covered by the RSU is 100 meters [43], [44], the
..., m) covers a road with a length of L meters. The average den- vehicle density on the road ρi is 0.2 vehicle per meter (veh/m)
sity of vehicles on the road is denoted as ρi , and the transmission [45], the transmission rate of messages generated from a vehicle
rate of messages generated by a vehicle is λs . The cumulative λm is 50 messages per second, the average service time 1/μ0 is
distribution function (CDF) of the message interval arrival time 5 milliseconds [46], the transmission power of the vehicle Ptx is
at RSUi is expressed as follows: 30 dBm [47], the noise power density N0 is -174 dBm/Hz [47],
the duration of a slot tslot is 50 microseconds [48], the SNR
e−ρi L(1−e ) − e−ρi L
−λs t

F (t) = 1 − . (12) threshold θ is 5 dB [47], and the number of resource blocks in a

1 − e−ρi L RSU RBrsu is 10. Without loss of generality, the wireless link
between the vehicle and RSU is composed of random multiple
The proof for Theorem 1 is provided in Appendix A.
hops, which are governed by a uniform distribution. Moreover,
Theorem 2: When the message handling process at an RSU
MC simulations are used for the performance analysis in this
is assumed to occur based on a GI/M/1/∞ queuing system, the
paper.
distribution of the message dwelling time at the RSU is governed
Fig. 2 describes the propagation latency with respect to
by the following equation:
the density of vehicles considering different RSU densities.
K∗  When the RSU density is fixed, the propagation latency in-
W (t) = 1 − e−μt + 1 − e−cμ(1−δ)t , (13) creases with the vehicle density. When the vehicle density is
1−δ
fixed, the propagation latency decreases as the RSU density
with increases.
⎡ ⎛  ⎞⎤−1 Fig. 3 shows the handling latency with respect to the density
c
of vehicle considering different RSU densities. When the RSU
⎢ 1 c
⎜ k c (1 − εk ) − k⎟⎥
K∗ = ⎢
⎣1 − δ +
⎜ ⎟⎥
⎝Dk (1 − εk ) · c (1 − δ) − k ⎠⎦ ,(14)
density is fixed, the handling latency increases as the vehicle
k=1 density increases. When the vehicle density is fixed, the handling
latency decreases as the RSU density increases.
Authorized licensed use limited to: BMS College of Engineering. Downloaded on May 15,2024 at 09:11:37 UTC from IEEE Xplore. Restrictions apply.
5010 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 68, NO. 5, MAY 2019

Fig. 2. Propagation latency with respect to the vehicle density based on Fig. 5. Reliability with respect to the vehicle density considering different
different RSU densities. RSU densities.

Fig. 3. Handling latency with respect to the vehicle density considering dif-
ferent RSU densities.

Fig. 6. Relationship between the reliability and latency utility functions in 5G

autonomous vehicular networks.

Fig. 4. Total latency with respect to the vehicle density considering different
RSU densities.

Fig. 4 illustrates the total latency with respect to the density

Fig. 7. Reliability and latency joint function with respect to the vehicle density
of vehicle considering different RSU densities. When the RSU considering different RSU densities.
density is fixed, the total latency increases as the vehicle density
increases. When the vehicle density is fixed, the total latency
decreases as the RSU density increases.
Fig. 5 depicts the reliability with respect to the density of vehi- When the RSU density is fixed, the reliability utility function in-
cle considering different RSU densities. When the RSU density creases and the latency utility function decreases as the vehicle
is fixed, the reliability increases as the vehicle density increases. density increases. Hence, a conflict occurs between the reliabil-
When the vehicle density is fixed, the reliability increases as the ity and latency performance with respect to the vehicle density
RSU density increases. in 5G autonomous vehicular networks. When the vehicle den-
Fig. 6 indicates the relationship between the reliability and sity is fixed, the reliability and latency utility functions increase
latency utility functions in 5G autonomous vehicular networks. as the RSU density increases.

Authorized licensed use limited to: BMS College of Engineering. Downloaded on May 15,2024 at 09:11:37 UTC from IEEE Xplore. Restrictions apply.
GE: ULTRA-RELIABLE LOW-LATENCY COMMUNICATIONS IN AUTONOMOUS VEHICULAR NETWORKS 5011

Fig. 7 analyzes the reliability and latency joint function with EASSs depends on the type of service applications. Dif-
respect to the vehicle density considering different RSU densi- ferent types of service applications have different latency
ties. When the RSU density is fixed and the vehicle density is less requirements. Hence, the EASS requirements change for
than 0.19 veh/m, the values of the reliability and latency joint different types of service applications in 5G autonomous
function initially increase as the vehicle density increases. When vehicular networks.
the RSU density is fixed and the vehicle density is larger than Based on the analysis of the three types of service slices,
or equal to 0.19 veh/m, the values of the reliability and latency different types of service slices can be distinguished by the ar-
joint function decrease as the vehicle density increases. Hence, rival rate, handling time and latency requirement. To simplify
a maximum of reliability and latency joint function value exists the analysis of service slices in 5G autonomous vehicular net-
with respect to the vehicle density. When the vehicle density works, the service slice of a 5G autonomous vehicular network
is fixed, the values of the reliability and latency joint function is denoted as S {λa , μs , Treq }, where λa is the arrival rate of the
value increase as the RSU density increases. service slice, μs is the handling time of service slice and Treq is
the latency requirement of the service slice.
IV. NETWORK SLICING SOLUTION
Based on the results in Fig. 6, a conflict is observed regarding B. Function Slicing and Resource Slicing
the reliability and latency performance in 5G autonomous ve-
To improve both the reliability and latency, the protocol func-
hicular networks. To implement URLLC, both the reliability and
tions at the RSUs and InPs must be sliced and re-structured by
latency performance in 5G autonomous vehicular networks must
SDN technologies to support service and resource slicing in 5G
be improved. Considering the reliability and latency gains as-
autonomous vehicular networks. Therefore, we propose a new
sociated with using network slicing technology, network slicing
5G autonomous vehicular network architecture to implement
has emerged as an attractive solution for 5G autonomous vehic-
service, resource and function slicing, as shown in Fig. 8. In this
ular networks. In conventional network slicing technology, only
paper, function slicing is implemented by dynamically schedul-
network resources are sliced to improve the resource utilization
ing and assembling function modules in the logical layers of the
in wireless networks. However, satisfying the URLLC require-
proposed autonomous vehicular network architecture. Based on
ment in 5G autonomous vehicular networks based on network
the proposed autonomous vehicular network architecture, the
resource slicing alone is difficult. In this case, we extend network
protocol structure of an InP is composed of two logical layers,
slicing from resource slicing to service and function slicing to
i.e., the InP control layer and InP shared resource layer. The
improve both the reliability and latency of 5G autonomous ve-
InP control layer includes two function modules: the monitor-
hicular networks. The relevant methods are described in detail
ing module and the allocation module. The monitoring module
in this section.
monitors the number of occupied resource blocks at the RSUs in
an InP coverage area. Based on the information from the moni-
A. Service Slicing
toring module, such as the number of occupied resource blocks
Different types of vehicular services are available in 5G au- that reach the available number of resource blocks and the num-
tonomous vehicular networks. To improve the service access ber of resource blocks that is still required by the service slices
efficiency in 5G autonomous vehicular networks, we normalize at the RSU, the allocation module will schedule the unoccupied
different types of vehicular services into three types of service resource blocks from adjacent RSUs to support the requirements
slices. of the specified RSUs. The InP shared resource layer is a type
r State-Report Service Slice of virtual network resource pool that includes all unoccupied
State-report service slices (SRSSs) are used to provide the resource blocks in the coverage of the InP. The resource slicing
state of the vehicle, such as the speed and location. SRSSs is performed by the InP shared resource layer to improve the
have the following features: the arrival rate is high; the resource utilization efficiency in 5G autonomous vehicular net-
holding time of the state is short; and the SRSSs have sim- works. In the real world, resource blocks are stored at the RSUs.
ilar contexts. SRSSs are used for vehicle collision avoid- According to the logic relationships in the proposed autonomous
ance, vehicle lane changes and vehicle deceleration. Hence, vehicular network architecture, all unoccupied resource blocks
SRSSs must be quickly accessed. Moreover, SRSSs should in the coverage of the InP can be regarded as stored in the virtual
be processed at RSU locations to reduce the response time. network resource pool in the InP shared resource layer. Further-
r Event-Driven Service Slice more, all resource blocks stored in the virtual network resource
Event-driven service slices (EDSSs) are generated by spe- pool can be sliced to effectively satisfy the resource requirements
cific events, such as the emergency broadcasts and the road of service slices at the RSUs.
information updates. Although the EDSS arrival rate is low, The protocol structure of an RSU is divided into three logical
EDSSs usually involve with security issues. Hence, EDSSs layers: the RSU interface layer, RSU virtualization layer and
must be accessed with a low latency and high reliability in RSU resource layer. The RSU interface layer provides different
5G autonomous vehicular networks. interfaces corresponding to different types of service slices. The
r Entertainment-Application Service Slice main function of the RSU interface layer is to separate RSU re-
Entertainment-application service slices (EASSs) are gen- source requirements with different types of service slices. Based
erated from vehicle users. The latency requirement of on the RSU interface layer, the relationships among different

Authorized licensed use limited to: BMS College of Engineering. Downloaded on May 15,2024 at 09:11:37 UTC from IEEE Xplore. Restrictions apply.
5012 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 68, NO. 5, MAY 2019

Fig. 8. Autonomous vehicular network architecture.

types of service slices are normalized by separating the resource restrict the percentage of local RSU resource blocks used for
requirements. other RUSs, i.e., 0 ≤ akg ≤ aM kg
ax
< 1 when k = g. In the pro-
The function of the RSU virtualization layer is to allocate posed 5G autonomous vehicular network architecture, the RSU
locational RSU resource blocks to satisfy the requirements of virtualization layer of RSUk can schedule the resource blocks
the service slices from the RSU interface layer. Moreover, the from adjacent RSUs for local service slicing and reduce the
RSU virtualization layer can optimize the handling process to handling latency of the service slice. The handling latency of
promptly respond to the service slices with low latency con- a service slice at Inpk is denoted as T sqk (k = 1, 2, ..., NRSU )
straint. and the handling latency of service slice at the local RSU RSUk
The RSU resource layer manages the resource blocks at the is denoted by T sqM,k . When network slicing is adopted for 5G
RSU location. The RSU resource blocks were previously allo- autonomous vehicular networks, the handling latency T tqM in
cated for the location service slices. Thus, the RSU virtualiza- the queue of the RSU is calculated based on the total handling
tion layer can require resource blocks from the associated InP latency of a service slice at the RSU RSUk , i.e., T tqM = T sM,k .
control layer when the RSU resource layer cannot satisfy the Moreover, the total handling latency of a service slice at RSUk
requirements of the location service slices. When the resource is derived as follows:
blocks from the InP shared resource layer are scheduled in the 
RSU virtualization layer, the RSU resource layer will allocate akk g(g =k) akg
T sM,k = NRSU T sM,k + NRSU
q
T sqk . (21)
the additional resource blocks from the RSU virtualization layer g=1 a kg g=1 a kg
to meet the requirements of the location service slices.
Based on the proposed autonomous vehicular network archi-
tecture in Fig. 8, the resource blocks in a 5G autonomous ve-
C. Network Slicing Algorithm hicular network can be multiplexed by the following process.
Assuming that NRSU RSUs exist in a coverage area of an When a service slice arrives at RSUk , the associated Inpk de-
InP, the average number of NRSU is derived in Appendix C. A termines whether the local resource blocks of RSUk can satisfy
NRSU ∗ NRSU matrix ANRSU ∗NRSU is used to describe the mul- the latency requirement of service slice TSreq . Additional, when
tiplexing of resource blocks at an RSU when the network slicing the local resource blocks can satisfy the latency requirement
technology is adopted for 5G autonomous vehicular networks. of the service slice, i.e., T < TSreq , Inpk decreases the number
Notably, akg ∈ [0, 1) , 1 ≤ k ≤ NRSU and 1 ≤ g ≤ NRSU , is of local resource blocks that can be scheduled by RSUk until
the element of the matrix ANRSU ∗NRSU . When k = g, akg is T = TSreq . In this case, the number of occupied resource blocks
the percentage of resource blocks at the g − th RSU that is is RBrsu (k) , which satisfies T = TSreq . The remaining resource
scheduled for the k − th RSU. When k = g, akg is the per- blocks RBrem (k) = RBrsu (k) − RBrsu (k) at RSUk are
centage of resource blocks at the g − th RSU used for local added to the virtual network resource pool which can be sched-
service slices. To ensure that local service slice handling occurs uled by Inpk . When the local resource blocks can not satisfy
at the local RSU, an upper bound ratio aM kg
ax
is established to the latency requirement of a service slice, i.e., T > TSreq , Inpk

Authorized licensed use limited to: BMS College of Engineering. Downloaded on May 15,2024 at 09:11:37 UTC from IEEE Xplore. Restrictions apply.
GE: ULTRA-RELIABLE LOW-LATENCY COMMUNICATIONS IN AUTONOMOUS VEHICULAR NETWORKS 5013

increases the additional number of resource blocks scheduled by

RSUk until T = TSreq . In this case, Inpk schedules the number
of resource blocks RBreq (k) = RBrsu (k) − RBrsu (k) from
the virtual network resource pool to support the service slice
handling at RSUk . The algorithm is described in detail in the
Network Slicing Algorithm.

Algorithm 1: Network Slicing Algorithm.

Begin:
1: Initialize ANRSU ∗NRSU , NRSU , RBrsu (k), RBrem ,
RBreq , TSreq , aM ax
kg ;
2: for k = 1:1: NRSU do
3: T = T tpM + T tqM ;
4: while T < TSreq do
5: update RBrsu (k) by decreasing RBrsu (k); Fig. 9. Reliability and latency utility functions in 5G autonomous vehicular
6: calculate T with RBrsu (k) by (11); networks with and without network slicing.
7: RBrem (k) = RBrsu (k) − RBrsu (k) ;
8: RBrem = RBrem + RBrem (k);
9: while T > TSreq do
10: update RBrsu (k) by increasing RBrsu (k);
11: calculate T sM,k with RBrsu (k) by (21);
12: RBreq (k) = RBrsu (k) − RBrsu (k);
13: RBreq = RBreq + RBreq (k);
14: while RBrem & RBreq = 0 do
15: schedule resource blocks from the virtual network
resource pool to the RSU, which requires the
resource blocks to reduce the handling latency of the
service slice;
16: update the matrix ANRSU ∗NRSU ;
17: for k = 1:1: NRSU do
18: calculate T sM,k by (21);

Fig. 10. Reliability and latency joint function with respect to the density
V. SIMULATION RESULTS AND DISCUSSION of vehicle considering different upper bound ratios aM
kg
ax in 5G autonomous

vehicular networks with and without network slicing.

To analyze the performance of the network slicing algorithm
for 5G autonomous vehicular networks, the default simulation
parameters in Section III are used in the following simulations.
When service slicing is adopted in 5G autonomous vehicular net- slicing. When network slicing is adopted in 5G autonomous ve-
works, all messages are classified into three types: SRSS, EDSS hicular networks, the reliability utility function increases and
and EASS. Different types of service slices have different data the latency utility function decreases as the vehicle density in-
packet sizes. Different data packet sizes have different trans- creases. A conflict still occurs between the reliability and latency
mission times that correspond to different slot values tslot . To utility functions. Compared with the values of reliability and la-
simplify the simulation analysis, the values of tslot are set to 25, tency utility functions without network slicing, the reliability
50 and 100 milliseconds, which correspond to the SRSS, EDSS and latency utility function values are both improved by adopt-
and EASS in 5G autonomous vehicular networks with network ing the network slicing algorithm in 5G autonomous vehicular
slicing, respectively. Because SRSSs are regularly transmitted networks.
by vehicles, the number of SRSSs is obviously larger than the Fig. 10 analyzes the reliability and latency joint function
numbers of EDSSs and EASSs. Without loss of generality, the with respect to the density of vehicle considering different up-
proportion of the numbers of SRSS, EDSS and EASS is con- per bound ratios aM kg
ax
in 5G autonomous vehicular networks
figured as 8:1:1. The network slicing solution is adopted for 5G with and without network slicing. When the upper bound ra-
autonomous vehicular networks, and the reliability and latency tio is fixed, the values of the reliability and latency joint func-
are then analyzed and compared based on MC simulations in tion with network slicing are larger than those without network
this section. slicing. When the network slicing solution is adopted and the
Fig. 9 compares the reliability and latency utility functions in vehicle density is less than 0.12 veh/m, the reliability and la-
5G autonomous vehicular networks with and without network tency joint function values for aM kg
ax
= 0.8 are larger than those

Authorized licensed use limited to: BMS College of Engineering. Downloaded on May 15,2024 at 09:11:37 UTC from IEEE Xplore. Restrictions apply.
5014 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 68, NO. 5, MAY 2019

for aM kg
ax
= 0.9. If a high value of upper bound ratio, e.g., Assuming that there are n vehicles in the coverage of a RSU,
aM ax
= 0.9 is configured for low vehicle density scenarios, i.e., when the message arrival time interval at the RSU is denoted by
kg s s
the vehicle density is less than 0.12 veh/m, overmuch resource TRSU , the probability that TRSU is less than or equal to the time
blocks will be scheduled by adjacent RSUs and then the reliabil- duration t is derived as follows:
ity and latency joint function of local RSU has to be obviously P {TRSU
s
 t} = 1 − P {TRSU
s
> t}
reduced. As a consequence, the reliability and latency joint func-
tion of autonomous vehicular networks is decreased. When the = 1 − P {There are no messages generated
network slicing solution is adopted, and the vehicle density is in the covergae area of the
larger than or equal to 0.12 veh/m, the reliability and latency
joint function values for aM kg
ax
= 0.8 are less than or equal to RSU within the time slot (0, t]}
those for akg = 0.9. If a low value of upper bound ratio, e.g.,
M ax ∞  −λs t n (ρi L)n −ρi L
e n! e
aM ax
= 0.8 is configured for the high vehicle density scenarios, = 1 − n=1 −ρ
. (23)
kg 1−e i L
i.e., the vehicle density is larger than or equal to 0.12 veh/m, a
few of resource blocks can be scheduled by adjacent RSUs and Hence, the CDF of the message arrival time interval at the RSU
then the reliability and latency joint function of adjacent RSUs is derived as follows:
has to be suppressed. As a result, the reliability and latency joint e−ρi L(1−e ) − e−ρi L
−λs t

function of autonomous vehicular networks is repressed. F (t) = 1 − . (24)

1 − e−ρi L
Additional the PDF of the message arrival time interval at the
VI. CONCLUSION RSU is derived as follows:
−λs t
In this paper a reliability and latency joint function is pro- λs ρi Le−ρi L+ρi Le −λs t
posed to evaluate the joint impact of reliability and latency on f (t) = . (25)
1 − e−ρi L
5G autonomous vehicular networks. Moreover, the interactions
between reliability and latency are quantitatively analyzed Thus, Theorem 1 is proven.
based on MC simulations. The simulation results indicate that
a maximum reliability and latency joint function value occurs APPENDIX B
with respect to the vehicle density in 5G autonomous vehicular Assume that the message handling process at the RSU is a
networks. To improve both the reliability and latency in 5G GI/M/1/∞ queuing system. Thus, p− j is the probability that
autonomous vehicular networks, i.e., to implement URLLC, a there are j messages in the queue of the RSU. Based on the
 
network slicing solution is proposed in this paper. Moreover, results in [49], p−j = 0, j ≥ 0 is not a stable distribution in the
we propose a new vehicular network architecture in which  
GI/M/1/∞ queuing system when cμ λ
≥ 1. p− j = 0, j ≥ 0 is
network slicing is extended from network resource slicing
a stable distribution in the GI/M/1/∞ queuing system when
to service and function slicing. Furthermore, a new network λ λ −
slicing algorithm is developed to implement URLLC in 5G au- cμ < 1. Considering cμ < 1, the probability pj is expressed as
follows:
tonomous vehicular networks. The simulation results show that

∞
the proposed network slicing algorithm can improve both the p−
j = p−i · pij (1) , j = 0, 1, 2, · · · , (26)
reliability and latency of 5G autonomous vehicular networks. i=0
Therefore, our results indicate that network slicing technology where pij (1) is the one step transition probability when the
can support URLLC in 5G autonomous vehicular networks.
In future work, ultra-reliable low-latency performance must ∞ −of queuing messages changes from i to j. Considering
number
j=0 pj = 1 and j ≥ c, we obtain the following result:
be optimized in future autonomous vehicular networks, and 
the follow potential topics should be addressed: 1) optimize ∞ −cμt (cμt)i−j+1
pij (1) = 0 e (i−j+1)! dF (t) , i ≥ j − 1 . (27)
the service slicing scheme to match the type and number of
0, i<j−1
messages in future autonomous vehicular networks; 2) optimize
the network resource slicing scheme to account for the service When (27) is substituted into (26) and j ≥ c, the probability p−
j
slicing results; and 3) optimize the function slicing scheme to is derived as follows:
support the service and network resource slicing schemes in ∞  ∞
− − (cμt)i−j+1
future autonomous vehicular networks. pj = pi · e−cμt dF (t),
i=j−1 0 (i − j + 1)!

APPENDIX A j = c, c + 1, c + 2, · · · (28)
The message interval time generated from a vehicle is denoted Based on the results in [49], the following relation can be
as T0 . For the time duration t ≥ 0, s ≥ 0, we have the following derived:
result can be obtained: ∞  ∞
(cμt)k cμ
k· e−cμt dF (t) = > 0. (29)
P {T0 > t + s|T0 > s} = P {T0 > t} = e−λs t . (22) 0 k! λ
k=0

Authorized licensed use limited to: BMS College of Engineering. Downloaded on May 15,2024 at 09:11:37 UTC from IEEE Xplore. Restrictions apply.
GE: ULTRA-RELIABLE LOW-LATENCY COMMUNICATIONS IN AUTONOMOUS VEHICULAR NETWORKS 5015

∞ 
Additionally, based on (29), the following equation can be t
cμ(cμx)j−c
derived: = K∗ δ j−c · e−cμx dx
0 (j − c)!
∞  ∞ j=c
(cμt)k 
δk · e−cμt dF (t) = f (cμ (1 − δ)) = δ. (30) K∗
0 k! = 1 − e−cμ(1−δ)t . (37)
k=0 1−δ
Based on the results in [49], (30) can be solved, and the so- The service time of the m − th message is denoted by χm
lution is exclusively in the range of (0, 1). When the solution of and the distribution of χm is denoted by χm (t). Hence, the dis-
(30) is denoted as δ (0 < δ < 1), the probability p−
j is calculated tribution of the message dwelling time is expressed as W (t) =
as follows: lim (χm (t) + Wqm (t)). Because Wqm and χm are indepen-
m→∞
p− ∗ dent, the distribution of the message dwelling time is derived as
j =K ·δ , j = c − 1, c, c + 1, · · · ,
j−c
(31)
follows:
where K ∗ is a constant. When j = c − 1, K ∗ is calculated as K∗ 
follows: W (t) = 1 − e−μt + 1 − e−cμ(1−δ)t . (38)
1−δ
K ∗ = δ · p−
c−1 . (32) Thus, Theorem 2 is proven.

Considering c = 1, i.e., the GI/M/1/∞ queuing system is APPENDIX C

implemented, K ∗ can be simply denoted by K ∗ = δ · p−
0 . Fur-
thermore, the probability p− is derived as follows: Given that the density of InP is λInp , a typical PVT cell area
j
denoted by RInp , follows a Gamma distribution [47]. The PDF
p− −
j = p0 · δ , j = 0, 1, 2, · · ·.
j
(33) of the InP distribution is expressed as follows:
∞ − (bλInp )a a−1 −bxλInp
Considering j=0 pj = 1, the probability of an empty queue fRInp (x) = x e , (39)
Γ (a)
is calculated by p− −
0 = 1 − δ. By substituting the value of p0 into
− ∞
(33), the probability pj can be calculated as follows: where Γ (x) = 0 tx−1 e−t dt is a Gamma function, a is the
shape parameter and bλInp is the inverse scale parameter of
p−
j = (1 − δ) · δ , j = 0, 1, 2, · · · .
j
(34) the Gamma distribution. The default values of a and b are con-
When c is configured as c = 1, K ∗ is calculated as follows: figured as a = 3.61 and b = 3.57 in the simulation analysis,
respectively. The road length in a coverage area of an InP is
K ∗ = δ · (1 − δ) . (35) derived as follows:
 ∞
(bλInp )a a−1 −bxλInp
In this paper the message is assumed to be serviced by the first- LRInp (x) = ρroad x e dx, (40)
in-first-out (FIFO) scheme in the GI/M/c/∞ queuing system. 0 Γ (a)
The waiting time of the m − th message is denoted by Wqm in where ρroad is the road density in an urban environment. Con-
λ
the GI/M/c/∞ queuing system. When cμ < 1, the distribution sidering the uniform distribution of RSUs along the road, the
of Wqm is expressed as P {Wqm ≤ t} = Wqm (t). When the expected NRSU is derived as follows:
−
queuing length is Nm = j and j < c, the m − th message can ∞ (bλInp )a a−1 −bxλInp
ρroad Γ(a) x e dx
be directly handled without waiting. When j ≥ c, the m − th E (NRSU ) =
0
. (41)
message must be wait until the services for j − c + 1 messages ρRSU
have been completed in the queuing system. Based on the total
probability theorem, the waiting time of the m − th message is REFERENCES
expressed as follows: [1] NGMN.Alliance, “5G White Paper,” version 1, Feb. 2015.
[2] K. Yang, S. Ou, and H.-H. Chen, “On effective offloading services for
c−1
 −  resource-constrained mobile devices running heavier mobile internet ap-
Wqm (t) = P Nm =j ·0 plications,” IEEE Commun. Mag., vol. 46, no. 1, pp. 56–63, Jan. 2008.
[3] X. Ge, Z. Li, and S. Li, “5G software defined vehicular networks,” IEEE
j=0
Commun. Mag., vol. 55, no. 7, pp. 87–93, Jul. 2017.
∞  [4] K. Liu, J. K. Y. Ng, V. C. S. Lee, S. H. Son, and I. Stojmenovic, “Co-
 −  t
(cμx)j−c · cμ
+ P Nm =j · e−cμx dx. operative data scheduling in hybrid vehicular Ad Hoc networks: VANET
j=c 0 (j − c)! as a software defined network,” IEEE/ACM Trans. Netw., vol. 24, no. 3,
pp. 1759–1773, Jun. 2016.
(36) [5] Y. Zaki, “LTE mobile network virtualization,” Mobile Netw. Appl., vol. 16,
no. 4, pp. 424–432, Aug. 2011.
λ [6] X. Ge, H. Chen, G. Mao, Y. Yang, and S. Tu, “Vehicular communica-
Considering cμ < 1, the limitation of the waiting time of the
tions for 5G cooperative small cell networks,” IEEE Trans. Veh. Technol.,
m − th message is derived as follows: vol. 65, no. 10, pp. 7882–7894, Oct. 2016.
c−1
[7] W. Cheng, X. Zhang, and H. Zhang, “Full-duplex spectrum-sensing and
MAC-protocol for multichannel non-time-slotted cognitive radio net-
lim Wqm (t) = p−
j ·0 works,” IEEE J. Sel. Areas Commun., vol. 33, no. 5, pp. 820–831,
m→∞
j=0 May 2015.
[8] L. Li, Y. Li, and R. Hou, “A novel mobile edge computing-based archi-
∞  t
cμ(cμx)j−c tecture for future cellular vehicular networks,” in Proc. IEEE Wireless
+ p−
j · e−cμx dx Commun. Netw. Conf., San Francisco, CA, USA, 2017, pp. 1–6.
j=c 0 (j − c)!

Authorized licensed use limited to: BMS College of Engineering. Downloaded on May 15,2024 at 09:11:37 UTC from IEEE Xplore. Restrictions apply.
5016 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 68, NO. 5, MAY 2019

[9] J. Liu, J. Wan, B. Zeng, Q. Wang, H. Song, and M. Qiu, “A scalable [31] X. Ge, B. Yang, J. Ye, G. Mao, C. Wang, and T. Han, “Spatial spectrum and
and quick-response software defined vehicular network assisted by mo- energy efficiency of random cellular networks,” IEEE Trans. Commun.,
bile edge computing,” IEEE Commun. Mag., vol. 55, no. 7, pp. 94–100, vol. 63, no. 3, pp. 1019–1030, Mar. 2015.
Jul. 2017. [32] T. Mattfeldt, “Stochastic geometry and its applications,” J. Microsc.,
[10] T. Rodrigues, K. Suto, H. Nishiyama, N. Kato, and K. Temma, “Cloudlets vol. 183, no. 3, pp. 257–257, 1996.
activation scheme for scalable mobile edge computing with transmis- [33] Y. Zhong, T. Quek, and X. Ge, “Heterogeneous cellular networks with
sion power control and virtual machine migration,” IEEE Trans. Comput., spatio-temporal traffic: Delay analysis and scheduling,” IEEE J. Sel. Areas
vol. 67, no. 9, pp. 1287–1300, Sep. 2018. Commun., vol. 35, no. 6, pp. 1373–1386, Jun. 2017.
[11] I. Safiulin, S. Schwarz, and M. Rupp, “Multicast beamforming capabili- [34] K. Yang, S. Ou, K. Guild, and H.-H. Chen, “Convergence of ethernet
ties of LTE MBSFN for V2X communications,” in Proc. IEEE 86th Veh. PON and IEEE 802.16 broadband access networks and its QoS-aware dy-
Technol. Conf., Toronto, ON, Canada, 2017, pp. 1–5. namic bandwidth allocation scheme,” IEEE J. Sel. Areas Commun., vol. 27,
[12] X. Li, B. J. Hu, H. Chen, B. Li, H. Teng, and M. Cui, “Multi-hop delay no. 2, pp. 101–116, Feb. 2009.
reduction for safety-related message broadcasting in vehicle-to-vehicle [35] K. Yang, S. Ou, H.-H. Chen, and J. He, “A multihop peer-communication
communications,” IET Commun., vol. 9, no. 3, pp. 404–411, 2015. protocol with fairness guarantee for IEEE 802.16-based vehicular net-
[13] S. I. Sou and O. K. Tonguz, “Enhancing VANET connectivity through works,” IEEE Trans. Veh. Technol., vol. 56, no. 6, pp. 3358–3370,
roadside units on highways,” IEEE Trans. Veh. Technol., vol. 60, no. 8, Jun. 2007.
pp. 3586–3602, Oct. 2011. [36] X. Ge, Y. Sun, H. Gharavi, and J. Thompson, “Joint optimization of compu-
[14] C. Shao, S. Leng, Y. Zhang, A. Vinel, and M. Jonsson, “Performance tation and communication power in multi-user massive MIMO systems,”
analysis of connectivity probability and connectivity-aware MAC protocol IEEE Trans. Wireless Commun., vol. 17, no. 6, pp. 4051–4063, Jun. 2018.
design for platoon-based VANETs,” IEEE Trans. Veh. Technol., vol. 64, [37] A. Ghosh et al., “Millimeter-wave enhanced local area systems: A high-
no. 12, pp. 5596–5609, Dec. 2015. data-rate approach for future wireless networks,” IEEE J. Sel. Areas Com-
[15] R. Yu, J. Ding, X. Huang, M. T. Zhou, S. Gjessing, and Y. Zhang, “Op- mun., vol. 32, no. 6, pp. 1152–1163, Jun. 2014.
timal resource sharing in 5G-enabled vehicular networks: A matrix game [38] J. Liu, H. Nishiyama, N. Kato, and J. Guo, “On the outage probability of
approach,” IEEE Trans. Veh. Technol., vol. 65, no. 10, pp. 7844–7856, device-to-device communication enabled multi-channel cellular networks:
Oct. 2016. A RSS threshold-based perspective,” IEEE Trans. Commun., vol. 34,
[16] W. Sun, D. Yuan, E. G. Strom, and F. Brannstrom, “Cluster-based radio no. 1, pp. 163–175, Jan. 2016.
resource management for D2D-supported safety-critical V2X communi- [39] X. Ge, X. Tian, Y. Qiu, G. Mao, and T. Han, “Small-cell networks with
cations,” IEEE Trans. Wireless Commun., vol. 15, no. 4, pp. 2756–2769, fractal coverage characteristics,” IEEE Trans. Commun., vol. 66, no. 11,
Apr. 2016. pp. 5457–5469, Nov. 2018
[17] K. Wang, F. R. Yu, and H. Li, “Information-centric virtualized cellular net- [40] J. Chen, X. Ge, and Q. Ni, “Coverage and handoff analysis of 5G frac-
works with device-to-device communications,” IEEE Trans. Veh. Technol., tal small cell networks,” IEEE Trans. Wireless Commun., vol. 18, no. 2,
vol. 65, no. 11, pp. 9319–9329, Nov. 2016. pp. 1263–1276, Feb. 2019.
[18] P. Rost et al., “Network slicing to enable scalability and flexibility in [41] B. Yang, G. Mao, M. Ding, X. Ge, and X. Tao, “Dense small cell networks:
5G mobile networks,” IEEE Commun. Mag., vol. 55, no. 5, pp. 72–79, From noise-limited to dense interference-limited,” IEEE Trans. Veh. Tech-
May 2017. nol., vol. 67, no. 5, pp. 4262–4277, May 2018.
[19] X. Foukas, G. Patounas, A. Elmokashfi, and M. K. Marina, “Network [42] B. Yutaka, “Analysis of a GI/M/1 queue with multiple working vacations,”
slicing in 5G: Survey and challenges,” IEEE Commun. Mag., vol. 55, Oper. Res. Lett., vol. 33, no. 2, pp. 201–209, 2005.
no. 5, pp. 94–100, May 2017. [43] S. I. Sou, “A power-saving model for roadside unit deployment in vehicular
[20] M. Richart, J. Baliosian, J. Serrat, and J. L. Gorricho, “Resource slicing in networks,” IEEE Commun. Lett., vol. 14, no. 7, pp. 623–625, Jul. 2010.
virtual wireless networks: A survey,” IEEE Trans. Netw. Service Manage., [44] S. H. Sun, J. L. Hu, Y. Peng, X. M. Pan, L. Zhao, and J. Y. Fang, “Support for
vol. 13, no. 3, pp. 462–476, Sep. 2016. vehicle-to-everything services based on LTE,” IEEE Wireless Commun.,
[21] X. Xu, H. Zhang, X. Dai, Y. Hou, X. Tao, and P. Zhang, “SDN based vol. 23, no. 3, pp. 4–8, Jun. 2016.
next generation mobile network with service slicing and trials,” China [45] K. Liu, J. K. Y. Ng, V. C. S. Lee, S. H. Son, and I. Stojmenovic, “Co-
Commun., vol. 11, no. 2, pp. 65–77, Feb. 2014. operative data scheduling in hybrid vehicular ad hoc networks: VANET
[22] P. Rost et al., “Mobile network architecture evolution toward 5G,” IEEE as a software defined network,” IEEE/ACM Trans. Netw., vol. 24, no. 3,
Commun. Mag., vol. 54, no. 5, pp. 84–91, May 2016. pp. 1759–1773, Jun. 2016.
[23] G. Liu, F. R. Yu, H. Ji, and V. C. M. Leung, “Distributed resource allocation [46] J. Liu, Y. Shi, L. Zhao, Y. Cao, W. Sun, and N. Kato, “Joint placement of
in virtualized full-duplex relaying networks,” IEEE Trans. Veh. Technol., controllers and gateways in SDN-enabled 5G-satellite integrated network,”
vol. 65, no. 10, pp. 8444–8460, Oct. 2016. IEEE J. Sel. Areas Commun., vol. 36, no. 2, pp. 221–232, Feb. 2018.
[24] R. Riggio, A. Bradai, D. Harutyunyan, T. Rasheed, and T. Ahmed, [47] L. Xiang, X. Ge, C. X. Wang, F. Y. Li, and F. Reichert, “Energy efficiency
“Scheduling wireless virtual networks functions,” IEEE Trans. Netw. Ser- evaluation of cellular networks based on spatial distributions of traffic load
vice Manage., vol. 13, no. 2, pp. 240–252, Jun. 2016. and power consumption,” IEEE Trans. Wireless Commun., vol. 12, no. 3,
[25] K. Samdanis, X. Costa-Perez, and V. Sciancalepore, “From network shar- pp. 961–973, Mar. 2013.
ing to multi-tenancy: The 5G network slice broker,” IEEE Commun. Mag., [48] X. Li, B. J. Hu, H. Chen, B. Li, H. Teng, and M. Cui, “Multi-hop delay
vol. 54, no. 7, pp. 32–39, Jul. 2016. reduction for safety-related message broadcasting in vehicle-to-vehicle
[26] C. Liang and F. R. Yu, “Wireless network virtualization: A survey, some communications,” IET Commun., vol. 9, no. 3, pp. 404–411, 2015.
research issues and challenges,” IEEE Commun. Surv. Tut., vol. 17, no. 1, [49] J. Medhi, CHAPTER 7 - Queues With General Arrival Time and Service-
pp. 358–380, Jan.–Mar. 2015. Time Distributions, Stochastic Models in Queueing Theory, 2nd ed. Or-
[27] C. Liang and F. R. Yu, “Wireless virtualization for next generation mobile lando, FL, USA: Academic, pp. 339–373, 2003.
cellular networks,” IEEE Wireless Commun., vol. 22, no. 1, pp. 61–69,
Feb. 2015.
[28] N. Zhang, S. Zhang, P. Yang, O. Alhussein, W. Zhuang, and X. S. Shen, Xiaohu Ge (M’09–SM’11) received the Ph.D. degree
“Software defined space-air-ground integrated vehicular networks: Chal- in communication and information engineering from
the Huazhong University of Science and Technology
lenges and solutions,” IEEE Commun. Mag., vol. 55, no. 7, pp. 101–109,
(HUST), Wuhan, China, in 2003. Since 2005, he has
Jul. 2017.
been a Full Professor with the School of Electronic
[29] E. Valkeila, “An introduction to the theory of point processes, volume
II: General theory and structure, 2nd edition by Daryl J. Daley, David Information and Communications, HUST. He is an
Adjunct Professor with the Faculty of Engineering
Vere-Jones,” Int. Statistical Rev., vol. 76, no. 2, pp. 316–316, 2008.
and Information Technology, University of Technol-
[30] K. Maurice, M. Lebourges, and S. Zuyev, “Stochastic geometry and ar-
ogy Sydney, Ultimo, NSW, Australia. He is the Direc-
chitecture of communication networks.” Telecommun. Syst., vol. 7, no. 1,
pp. 209–227, 1997. tor of the China International Joint Research Center
of Green Communications and Networking.

Authorized licensed use limited to: BMS College of Engineering. Downloaded on May 15,2024 at 09:11:37 UTC from IEEE Xplore. Restrictions apply.