A Survey on Resource Allocation in VehicularNetworks

Md. Noor-A-Rahim, Zilong Liu, Haeyoung Lee, G. G. Md. Nawaz Ali, Dirk Pesch, Pei Xiao

Abstract—Vehicular networks, an enabling technology forIntelligent Transportation System (ITS), smart cities, andautonomous driving, can deliver numerous on-board dataservices, e.g., road-safety, easy navigation, traffic efficiency,comfort driving, infotainment, etc. Providing satisfactoryquality of service (QoS) in vehicular networks, however,is a challenging task due to a number of limiting factorssuch as hostile wireless channels (e.g., high mobility orasynchronous transmissions), increasingly fragmented andcongested spectrum, hardware imperfections, and explosivegrowth of vehicular communication devices. Therefore, it ishighly desirable to allocate and utilize the available wirelessnetwork resources in an ultra-efficient manner. In this paper,we present a comprehensive survey on resource allocation(RA) schemes for a range of vehicular network technologiesincluding dedicated short range communications (DSRC) andcellular based vehicular networks. We discuss the challengesand opportunities for resource allocations in modern vehic-ular networks and outline a number of promising futureresearch directions.

Index Terms—Intelligent Transportation System, Vehicularnetwork, Autonomous Driving, DSRC V2X, Cellular V2X,Resource Allocation, Network Slicing, Machine Learning.

I. INTRODUCTION

The prevalent vision is that vehicles (e.g., cars, trucks,trains, etc.) will in the future be highly connected withthe aid of ubiquitous wireless networks, anytime andanywhere, to provide unprecedented travel experiencesand offer a series of far-reaching benefits such as signifi-cantly improved road safety, enhanced situational aware-ness, less traffic congestion, reduced pollution emission,and lower capital expenditure. Central to this visionis a scalable and intelligent vehicular network whichis responsible for efficient information exchange amongvehicles and/or between vehicles and infrastructure.

Md. Noor-A-Rahim and Dirk Pesch are with the School ofComputer Science & IT, University College Cork, Ireland (E-mail:m.rahim@cs.ucc.ie and d.pesch@cs.ucc.ie).

Zilong Liu, Haeyoung Lee, and Pei Xiao are with Institutefor Communication Systems, 5G Innovation Centre, University ofSurrey, United Kingdom (E-mail: zilong.liu@surrey.ac.uk,haeyoung.lee@surrey.ac.uk, p.xiao@surrey.ac.uk).

G. G. Md. Nawaz Ali is with the Department of Automotive Engi-neering, Clemson University, USA (E-mail: gga@clemson.edu).

This publication has emanated from research conducted with thefinancial support of Science Foundation Ireland (SFI) and is co-fundedunder the European Regional Development Fund under Grant Number13/RC/2077. It has also received funding from the European UnionsHorizon 2020 Research and Innovation Programme under the EDGECO-FUND Marie Skodowska Curie grant agreement No. 713567. Thework of Z. Liu and P. Xiao was supported by the U.K. Engineeringand Physical Sciences Research Council under Grant EP/P03456X/1.

As an instrumental enabler for ITS, smart cities, andautonomous driving, vehicular networks have attractedsignificant research interests in recent years both fromthe academic and industrial communities [1–5]. In par-ticular, the concept of connected vehicles, also known asvehicle-to-everything (V2X) communications, has gainedsubstantial momentum by bringing in increased datathroughput and enhanced road safety along with novelonboard computing and sensing technologies. So far,there are two major approaches for V2X communica-tions: DSRC and cellular based vehicular communication[6], [7]. DSRC is supported by a family of standardsincluding the IEEE 802.11p amendment for WirelessAccess in Vehicular Environments (WAVE), the IEEE1609.1∼0.4 standards for resource management, security,network service, and multi-channel operation [8]. On theother hand, cellular based vehicular communication, alsocalled C-V2X, designed over cellular networks such asLong-Term Evolution (LTE) and 5G new radio (5G NR),allows every vehicle to communicate with different typesof communication entities, such as pedestrians, roadsideunits (RSU), satellites, internet/cloud, and other vehi-cles. Both V2X techniques have their respective advan-tages and limitations when they are adopted in vehic-ular environments. As a result, an integration of suchheterogeneous vehicular networks has been suggestedto exploit their unique benefits, while addressing theirindividual drawbacks.

Wireless networks suffer from a wide range of im-pairments like shadowing, path loss, time- and/orfrequency- selectivity of wireless channels, jammingand/or multi-user interference, etc. To deal with theseimpairments, radio resources (such as time slots, fre-quency bands, transmit power levels, etc.) should beallocated in an optimized manner to cater for instanta-neous channels and network conditions. Dynamic Re-source Allocation (RA) schemes are preferred as theygive rise to significantly improved performance (com-pared to the Static RA schemes) by efficiently exploitingdiversities from various dimensions [9–11]. For instance,authors in [12–16] studied RA schemes for device-to-device (D2D) V2X networks by taking into accountfast vehicular channel variations. Nevertheless, RA invehicular networks are far more challenging due to thefollowing reasons:

1) Highly dynamic mobility from low-speed vehicles

arX

iv:1

909.

1358

7v1

[cs

.IT

] 3

0 Se

p 20

19

(e.g., less than 60 km/h) to high-speed cars/trains(e.g., 500 km/h or higher) [17], [18]. The air in-terface design for high mobility communication,for instance, may require more time-frequency re-sources in order to combat the impairments in-curred by Doppler spread/shifts and multi-pathchannels.

2) Vast range of data services (e.g., in-car multi-media entertaining, video gaming/conferencing,ultra-reliable and low-latency delivery of safetymessages, high-precision map downloading, etc)with different QoS requirements in terms of reli-ability, latency, and data rates. In particular, somerequirements (e.g., high data throughput againstultra-reliability) may be conflicting and hence itmay be difficult to support them simultaneously.

3) Explosive growth of vehicular communication de-vices in the midst of increasingly fragmented andcongested spectrum. Moreover, these devices usu-ally have different hardware parameters and there-fore may display a wide variation in their commu-nication capabilities under different channel andnetwork conditions. For example, a vehicular sen-sor device aiming for long battery life (e.g., morethan 10 years) is unlikely to use sophisticated signalprocessing algorithms for power saving purposeswhereas more system resources and more signalprocessing capabilities may be required for ultra-reliable transmission of safety messages.

Driven by these challenges, over the past decade,numerous disruptive ideas and techniques have beenemerging aiming for optimizing/addressing various as-pects/challenges of vehicular networks. In the existingliterature, however, a survey with an extensive high-level overview as well as detailed up-to-date advanceson RA in vehicular networks is still lacking to the bestof our knowledge. To fill this gap and to stimulatemore innovations in this area, we provide a compre-hensive survey on the state-of-the-art of RA in vehicularnetworks and suggest a number of promising researchdirections.

This article is organized as follows. We start our dis-course in Section II by a high-level overview of vehicularnetworks which include DSRC network, C-V2X networkand heterogeneous network. Detailed literature surveyson these three types of vehicular networks are presentedin Sections III-V, respectively. As machine learning isgaining ever-increasing research attention in numerousareas such as data-driven decision making, we providea dedicated survey in Section VI on applications of ma-chine learning for RA in vehicular networks. In SectionVII, we summarize three important future directions ofthe RA research by taking advantage of network slicing,machine learning, and context awareness. Finally, thisarticle is concluded in Section VIII.

II. OVERVIEW OF VEHICULAR NETWORKS

A. DSRC Vehicular Network

DSRC is a wide-consensus wireless technology thatis designed to support ITS applications in vehicularnetworks. The underlying standard for DSRC is 802.11p,which is derived from IEEE 802.11e with small modifi-cations in the QoS aspects. DSRC supports communica-tions between vehicles and RSUs. The US Departmentof Transportation estimates that vehicle-to-vehicle (V2V)communications based on DSRC can eliminate up to 82%of all crashes involving unimpaired drivers in the US,and about 40% of all crashes occurred at intersections [8].These statistics imply a significant potential for DSRCtechnology to reduce crashes and to improve road safety.

DSRC technology supports two classes of devices [19],[20]: on-board unit (OBU) and road side unit (RSU),which are equivalent to the mobile station (MS) and basestation (BS) in traditional cellular systems, respectively.An overview of a typical DSRC vehicular network inshown in Fig. 1a. The Federal Communications Commis-sion in the United States has allocated 75 MHz licensedspectrum for DSRC communications in the 5.9 GHzfrequency band [21]. Out of the 75 MHz spectrum, 5MHz is reserved as the guard band and seven 10-MHzchannels are defined for DSRC communications. Theavailable spectrum is configured into one control channel(CCH) and six service channels (SCHs). The CCH isreserved for carrying high-priority short messages orcontrol data, while other data are transmitted over theSCHs. Several modulation and coding schemes (MCS)are supported by DSRC with the transmitter (TX) powerranging from 0 dBm to 28.8 dBm. Based on the commu-nication environments, the coverage distance may rangefrom 10m to 1km.

A fundamental mechanism for medium/channel ac-cess in DSRC technology is known as distributed coor-dination function (DCF). With this DCF, vehicles con-tend for the transmission channels using a carrier-sensemultiple access (CSMA) with collision avoidance (CA)technique. To transmit a packet from a vehicle, the chan-nel must be sensed idle for a guard period. This guardperiod is known as the distributed interframe space(DIFS). If the channel is sensed busy, the vehicle initiatesa slotted backoff process and vehicles are only permittedto start transmissions at the beginning of slots. Vehiclesrandomly choose their individual backoff time from therange [0, CW− 1], where CW is known as the contentionwindow. The backoff time counter is decreased by 1,when the channel is sensed idle for a time slot. Thecounter is frozen when the channel is sensed occupiedand reactivated after the channel is sensed idle again fora DIFS time interval period. A vehicle transmits when itsbackoff counter reaches zero. A packet collision occurswhen two or more vehicles choose the same time slotfor transmission. Along with the above channel access

(a) DSRC Vehicular Network. (b) Cellular Vehicular Network.

(c) Heterogeneous Vehicular Network.

Fig. 1: Overview of Vehicular Networks.

mechanism, IEEE 802.11p adopts Enhanced DistributedChannel Access (EDCA) mechanism, which allows fouraccess categories in a vehicle with different priorities.

B. Cellular based Vehicular Network (C-V2X)Despite of the fact that DSRC is generally considered

as the de facto technique for vehicular networks, cellu-lar/LTE based vehicular communications (also known asC-V2X) have recently attracted significant attention dueto its large coverage, high capacity, superior quality ofservices, and multicast/broadcast support. An overviewof cellular based vehicular network in shown in Fig. 1b.LTE-V2V communication exploits LTE uplink resourceswhile utilizing single carrier frequency division multiple

access (SC-FDMA) at the PHY and MAC layers [22].According to the LTE specifications, the available band-width is subdivided into equally-spaced (spacing of 15kHz) orthogonal subcarriers. A resource block (RB) inLTE is formed by 12 consecutive subcarriers (i.e., 180kHz) and one time slot (i.e., 0.5 ms). The number of databits carried by each RB depends on specific Modulationand Coding Schemes (MCS).

To utilize the available radio resources, two side-link modes are defined by 3GPP standard release 14:Mode 3 and Mode 4. In Mode 3, it is assumed thatthe vehicles are fully covered by one or more evolvedNodeBs (eNBs) who dynamically assign the resources

being used for V2V communications through controlsignalling. This type of resource assignment is calleddynamic scheduling. An eNodeB may also reserve a setof resources for a vehicle for its periodic transmissions.In this case, the eNodeB defines how long resources willbe reserved for the vehicle. In Sidelink Mode 4, vehiclesare assumed to be in areas without cellular coverage andhence, resources are allocated in a distributed manner. Asensing based semi-persistent transmission mechanismis introduced in Sidelink Mode 4 to enable distributedresource allocation.

The distributed algorithm is implemented among ve-hicles, which optimizes the use of the available channelsby increasing the resource reuse distance between vehi-cles that are using the same resources. A distributed con-gestion control mechanism is also applied which calcu-lates the channel busy ratio and the channel occupancyratio. Then, a vehicle reserves resource for a randominterval and sends a reservation messages using Sidelink Control Information (SCI). The reservation messageis also called Scheduling Assignment (SA). Using SA,other vehicles which sense and listen to medium findout the list of busy resources and avoid selection of thoseresources. To increase the reliability, a vehicle may senda data message more than once in this mode. In Release14, 3GPP mentioned that D2D communications includedin Releases 12 and 13 can also be applied to vehicularnetworks as the localization characteristics of vehicularnetworks are similar to D2D networks [14], [23].

C. Heterogeneous Vehicular Network

Despite the potential of DSRC vehicular networks,DSRC technology suffers from several drawbacks [6],[24], [25] such as limited coverage, low data rate, andlimited QoS guarantee, and unbounded channel accessdelay. As a matter of fact, the PHY and MAC lay-ers of DSRC are inherited from IEEE 802.11 standardswhich have been originally optimized for wireless localarea networks with low mobility. As concluded in [25],although the current DSRC technology is shown tobe effective in supporting vehicular safety applicationsin many field trials, significant challenges remain foremploying DSRC technology in some hostile vehicularenvironments.

While cellular based vehicular networks can providewide coverage and high data rate services, they maynot be able to support decentralized communication asthe networks may become easily overloaded in situationwith very high vehicle density, e.g. traffic jams, etc. Thus,both DSRC and cellular based vehicular networks havetheir respective advantages and limitations when used invehicular environments. A depiction of a heterogeneousvehicular network in shown in Fig. 1c. A range of ef-forts [26–35] have been made towards the integration ofboth DSRC and cellular based vehicular networks (e.g.,LTE) for enhanced vehicular communications. Besides

the integration of DSRC and cellular based vehicularnetworks, emerging V2X applications require efficientutilization of heterogeneous access technologies, such asWi-Fi and TV broadcasting networks.

III. RESOURCE ALLOCATION IN DSRC NETWORKS

In this section, we review a number of resource allo-cation approaches for DSRC based vehicular networks.Previous works on the resource allocation strategies forDSRC network are mainly focused on MAC param-eter allocation, channel allocation and rate allocationtechniques. Hence, in the following, we classify all theresource allocation approaches for DSRC network inthose three categories.

0 2 0 4 0 6 0 8 0 1 0 0 1 2 00 . 00 . 10 . 20 . 30 . 40 . 50 . 60 . 70 . 80 . 91 . 01 . 11 . 2

Data

transf

er rat

io (be

t. fast

and s

low ve

hicle)

M e a n v e l o c i t y o f s l o w v e h i c l e

D e f a u l t S c h e m e H a r i g o v i n d a n ' s S c h e m e

Fig. 2: Comparison between default DSRC and thescheme proposed in [36] in terms of data transfer ratio(for fast and slow vehicles) versus mean velocity of slowvehicles.

A. MAC Parameter AllocationIn a traditional DSRC network, all vehicles adopt iden-

tical MAC parameters by default and hence have equalopportunity to access the network resources. However,this setting may be unfair for high-mobility vehicleswhich in turn could significantly degrade the networkperformance. For example, the data throughput of a ve-hicle with high velocity may degrade severely comparedto that of a slowly moving vehicle because the latter isexpected to have a better chance to communicate withits RSU (due to its long residence time in the coveragearea of the RSU). Several studies have been carriedout on MAC parameter allocation in DSRC networksto enhance reliability, throughput, and fairness. [36]presented a contention window allocation strategy toresolve the aforementioned unfairness problem. Specif-ically, an optimal selection on the minimum contentionwindow (required for any vehicle) has been derived bytaking into consideration the mean speed of vehicles in

1 0 2 0 3 0 4 02 0

3 0

4 0

5 0

6 0

7 0

8 0

9 0

1 0 0Av

erage

delay

(msec

)

A v e r a g e n u m b e r o f v e h i c l e s i n i n t e r f e r e c e r a n g e

D e f a u l t R o s s i ' s a l g o r i t h m

(a) Average transmission delay.

1 0 2 0 3 0 4 00 . 0

0 . 1

0 . 2

0 . 3

0 . 4

0 . 5

0 . 6

0 . 7

0 . 8

Throu

ghpu

t

A v e r a g e n u m b e r o f v e h i c l e s i n i n t e r f e r e c e r a n g e

D e f a u l t R o s s i ' s a l g o r i t h m

(b) Throughput.

Fig. 3: Network throughput with optimal contention window size [37].

the network. Fig. 2 compares the DSRC default schemeand the scheme proposed in [36] in terms of the datatransfer ratio (for fast and slow vehicles) versus meanvelocity of slow vehicles. It is observed that for theDSRC default scheme, the data transfer ratio increasesas the mean velocity of slow vehicles increases. In fact,in this case, the residence time of slowly moving vehiclesdecreases within RSUs coverage and hence the datatransfer decreases correspondingly. On the other hand,a relatively flat data transfer ratio is maintained withtheir proposed contention window allocation schemewhich ensures equal chances of communication with theRSU for both slow and fast vehicles1 A modified MACscheme was proposed in [36] to dynamically adapt theMAC parameters based on the residence time of vehicles.

To maximize the throughput among neighboring ve-hicles, a stochastic model was proposed in [37], [39] tofind the optimal maximum contention window usingthe surrounding vehicle density. Fig. 3 shows that theproposed protocol in [37], [39] offers much lower averagetransmission delay as well as significantly improvedthroughput (compared to the standard DSRC protocols)due to reduced packet collision with optimized con-tention window size.

In [40], two dynamic contention window allocationschemes have been proposed to improve the networkperformance in high mobility environments. The firstscheme is the p-persistent based approach [41] whichdynamically assign the contention window based onthe number of neighboring vehicles, while the second

1A contention window allocation approach similar to that in [36] canbe found in [38].

scheme performs contention window adaptation basedon the vehicle’s relative velocity. Fig. 4 compares theirproposed schemes in terms of the packet delivery ra-tios and network throughput. It is observed that bothschemes provide enhanced performance (compared tothe default DSRC one) as they give rise to reduced packetcollisions. Moreover, each scheme provides better perfor-mance than the other in certain scenarios. For example,the first scheme exhibits better packet delivery ratiowhen the number of vehicles in the network is large.In terms of network throughput, the second schemeoutperforms the first when the number of vehicles ishigher than 80.

B. Channel Allocation for Emergency MessagesDSRC/WAVE uses orthogonal frequency bands to

support multi-channel operation while consideringequal share of available channels to all messages. Emer-gency messages (e.g., mission critical messages that carrysafety-related information) in vehicular networks needto be processed with high priority, ultra reliability, andlow latency. Ryu et al. [42] proposed a multi-channelallocation strategy called DSRC-based Multi-channel Al-location for Emergency message dissemination (DMAE)by first identifying the available bandwidth of chan-nels and then allocating the channel with the largestbandwidth to the emergency message while maintainingQoS between RSU and OBU through periodic channelswitching. Fig. 5 compares the packet delivery ratio(PDR) and end-to-end delay between DMAE and thetraditional allocation scheme adopted by WAVE. It isobserved that the emergency PDR of DMAE is higherthan the PDR of WAVE as DMAE assigns available SCH

.

0 5 0 1 0 0 1 5 0 2 0 0 2 5 0

0 . 7 0

0 . 7 5

0 . 8 0

0 . 8 5

0 . 9 0

0 . 9 5

1 . 0 0Pac

ket de

livery

ratio

N u m b e r o f v e h i c l e s

C W m i n = 3 C W m i n = 7 C W m i n = 1 5 p - p e r s i s t e n t v e l o c i t y a d a p t i v e

(a) Packet delivery ratio.

0 5 0 1 0 0 1 5 0 2 0 0 2 5 00123456789

1 0

Overa

ll thro

ughp

ut (M

bps)

N u m b e r o f v e h i c l e s

C W m i n = 3 C W m i n = 7 C W m i n = 1 5 p - p e r s i s t e n t v e l o c i t y a d a p t i v e

(b) Delay time.

Fig. 4: Network throughput for different contention window sizes.

0 2 0 4 0 6 0 8 0 1 0 00 . 00 . 10 . 20 . 30 . 40 . 50 . 60 . 70 . 80 . 91 . 0

Packet

deliv

ery ra

tio

N u m b e r o f v e h i c l e s

D A M E W A V E

(a) Emergency packet delivery ratio.

0 2 0 4 0 6 0 8 0 1 0 00

5

1 0

1 5

2 0

2 5

3 0De

lay (m

sec)

N u m b e r o f v e h i c l e s

D A M E W A V E

(b) Delay time.

Fig. 5: Performance comparison of DMAE and WAVE [42].

with maximum bandwidth to the emergency messages.Moreover, DMAE outperforms WAVE in terms of delayperformance as it can assign emergency messages toreserved channels in the event of heavy traffic scenario.

C. Rate Allocation

IEEE 802.11p based communication supports multipleMCS to allow a wide range of data transmission ratesranging from 3 Mbps to 27 Mbps. The data rates andtransmission ranges for different MCS are shown inTable I. For the sake of simplicity, a constant MCS is often

assumed in previous works on vehicular communica-tions. This strategy may deteriorate the communicationperformance as constant MCS may not be suitable fordiverse traffic environments in different roadway sce-narios. As a solution, [43] proposed a new vehicularchannel access scheme (VCAS) to maintain a trade-offbetween overall throughput and fairness. In this scheme,a number of vehicles with similar transmission ratesare grouped into one channel to achieve the overallthroughput requirement, while the fairness requirementis achieved by controlling the group sizes. By adopting a

TABLE I: Different MCS and their corresponding data rates.

MCS Index Modulation Code rate Data rate (Mbps) Communication range (m)

1 BPSK 12 3 1000

2 BPSK 34 4.5 900

3 QPSK 12 6 800

4 QPSK 34 9 700

5 16-QAM 12 12 600

6 16-QAM 34 18 500

7 64-QAM 23 24 400

8 64-QAM 34 27 300

marginal utility model to allocate appropriate transmis-sion rate per SCH (determined by predefined transmis-sion distance thresholds), it is shown in [43] that theirproposed scheme can simultaneously achieve enhancedfairness and overall system throughput over the existingscheme adopted in DSRC system. More recently, [44],[45] presented allocation of variable MCS (i.e., variabledata rates) in network coding-assisted heterogeneous on-demand data access, in which the MCS for disseminatingdata items were assigned based on the distance of therequested vehicles from the RSU. Simulation resultsshow that the schemes proposed in [44], [45] are capableof improving the on-demand requests serving capabilityand reducing the system response time.

IV. RESOURCE ALLOCATION IN C-V2X

The capability of supporting diverse vertical indus-tries/applications is a major feature of 5G communi-cation systems and beyond. Examples of vertical in-dustries include smart homes/cites, e-health, factoriesof the future, intelligent refineries and chemical plants,and Cellular V2X (C-V2X). A strong catalyst for deeperand wider integration of wireless communications intoour lives, C-V2X has been advocated by many mobileoperators under the evolution of 3GPP’s LTE and 5GNR [46]. Compared to DSRC, C-V2X acts as a “long-range sensor” (aided by sophisticated cameras, radar,lidar, RSUs, cellular infrastructure and network) to allowvehicles to see/predict various traffic situations, roadconditions, and emergent hazards several miles away.

From a network point of view, there are three ma-jor 5G use cases to be supported: enhanced mobilebroadband (eMBB) communications, massive machine-type communications (mMTC), ultra-reliable and low-latency communications (URLLC). As far as C-V2X isconcerned with, eMBB, aiming to provide data rates ofat least 10 Gbps for the uplink and 20 Gbps for thedownlink channels, plays a pivotal role for in-car videoconferencing/gaming, various multimedia services, orhigh-precision map downloading, etc; mMTC will allowfuture driverless vehicles to constantly sense and learnthe instantaneous driving environments using massivenumber of connected sensors deployed in-car or attachedto the infrastructure; URLLC, targeting to achieve 1 msover-the-air round-trip time for a single transmission

with reliability of at least 99.999% is instrumental forautonomous emergency braking and hazard prevention.

That being said, C-V2X has to share and compete withother vertical applications for system resources (e.g.,spectrum/network bandwidth, storage and computing,etc) under a common physical infrastructure. RA for C-V2X therefore is a trade-off with a variety of data re-quirements from different vertical applications. A centralquestion is how to design an efficient network to provideguaranteed quality of service (QoS) for C-V2X whilebalancing the data services to other vertical applications.

A. RA for Traditional Cellular System

1 0 1 5 2 0 2 5 3 0 3 5 4 0 4 5 5 0 5 5 6 0

1 8 0 0 0 0 0

2 0 0 0 0 0 0

2 2 0 0 0 0 0

2 4 0 0 0 0 0

2 6 0 0 0 0 0

2 8 0 0 0 0 0

3 0 0 0 0 0 0

Netwo

rk sum

rate (

bps)

N u m b e r o f c o m m u n i c a t i o n l i n k s

T r a d i t i o n a l g r e e d y g r a p h a p p r o a c h A l g o r i t h m p r o p o s e d b y Z h a n g O p t i m a l a l l o c a t i o n s c h e m e

Fig. 6: Sum-rate comparison with traditional and optimalschemes [47].

Graph based interference aware RA strategies havebeen proposed in [47], [48], where the weights of theedges are assigned according to the interference termsbetween the related vertices. The scheme in [47] for-mulates an optimization problem with the objective ofmaximizing the network sum rate with low compu-tational complexity. It is shown in Fig. 6 that theirproposed scheme exhibits higher network sum rate thantraditional orthogonal communication mode. In contrast,the work in [48] aims at improving the connectivity ofvehicular communications by introducing a metric calledconnectivity index, which is obtained from the percentageof vehicles in the network being assigned with resourceswhile satisfying the interference constraints. With the aid

3 4 5 6 70 . 6 0

0 . 6 5

0 . 7 0

0 . 7 5

0 . 8 0

0 . 8 5

0 . 9 0

0 . 9 5

1 . 0 0Co

nnect

ivity

index

N u m b e r o f v e h i c l e s

O p t i m a l s o l u t i o n S u b - o p t i m a l s o l u t i o n o f M e n g ' s s c h e m e

(a) The gap between the optimal solution and the proposed sub-optimal solution.

0 . 1 0 0 . 1 5 0 . 2 0 0 . 2 5 0 . 3 0 0 . 3 5 0 . 4 0 0 . 4 5 0 . 5 00 . 0

0 . 1

0 . 2

0 . 3

0 . 4

0 . 5

0 . 6

0 . 7

0 . 8

Ratio

of fu

ll con

nectiv

ity co

mpon

ent

V e h i c l e ' s a r r i v a l r a t e

G r e e d y c o l o r i n g a l g o r i t h m ( K = 3 ) M e n g ' s a l g o r i t h m ( K = 3 ) G r e e d y c o l o r i n g a l g o r i t h m ( K = 6 ) M e n g ' s a l g o r i t h m ( K = 6 )

(b) The percent of the full connectivity (FC) components fordifferent vehicle arrival rates.

Fig. 7: Performance comparison of the RA scheme proposed in [48].

of the minimum spanning tree approach [49], Meng etal. [48] proposed a RA algorithm to improve the connec-tivity of the network. Fig. 7 shows the performance ofthe RA scheme proposed in [48]. The connectivity indexperformance is presented in Fig.7a with varying numberof vehicles, whilst the performance of brute force searchalgorithm is shown as a benchmark. We observe thatthe connectivity index of [48]’s algorithm is only 17.1%away from the optimum solution obtained from thebrute force search algorithm. In Fig. 7b, we present thefull connectivity performance of the algorithm proposedin [48] and compare with the greedy graph coloringalgorithm [50]. We observe a similar full connectivityperformance for both algorithms, while graph coloringalgorithm exhibits high computational complexity. Asexpected, the full connectivity percentage decays withthe increase of vehicle arrival rate (i.e., denser vehicularnetwork).

By exploiting geographical information, [52] proposeda joint RA and power control scheme for reliable D2D-enabled vehicular communications by considering slowfading channel information. Queuing dynamics was alsoconsidered in [52] in order to meet the requirements ofdifferent QoS in vehicular networks. [14] developed aheuristic algorithm, named Separate resOurce bLockandpowEr allocatioN (SOLEN), under large-scale vehicularfading channels to maximize the sum rate of cellularusers while satisfying the vehicular users’ requirementson latency and reliability. Similar to [14], [51] incorpo-rated dynamic MCS in the process of RBs and transmitpower allocation for guaranteed reliability and latency.A latency performance comparison between the works

of [14] and [51] is shown in Fig. 8. By adopting dynamicMCS in the allocation algorithm, the algorithm proposedin [51] outperforms that of [14] in terms of averageoutage probability and packet latency. To support D2D-based safety-critical vehicular communication, a cluster-based RA scheme was proposed in [15] by maximiz-ing the cellular users’ sum rate. This is achieved bya three-step heuristic algorithm with the knowledge ofthe slowly varying channel state information of uplinkchannel.

The work in [53] proposed a centralized RA algo-rithm by utilizing the spectral radius estimation the-ory. Their proposed algorithm maximizes the numberof concurrent reuses of resources by multiple vehiclesinstead of maximizing the sum rate (a method oftenused in traditional allocation algorithms). With eNodeBcentrally deciding the resource reuse for the vehiclesin the network, the scheme proposed in [53] exhibitssignificant improvement in the spectrum efficiency anddemonstrates the capability of maintaining the requiredQoS when the vehicle density is high.

[54] proposed a RA scheme to support V2X com-munications in a D2D-enabled cellular system, wherethe V2I communication is supported by a traditionalcellular uplink strategy and the V2V communication isenabled by the D2D communications in the reuse mode.[54] formulated an optimization problem to maximizethe sum ergodic capacity of the vehicle-to-infrastructure(V2I) links while satisfying the delay requirements ofV2V links. The optimization problem was solved bycombining the bipartite matching algorithm and theeffective capacity theory.

0 . 0 1 0 0 . 0 1 5 0 . 0 2 0 0 . 0 2 5 0 . 0 3 0 0 . 0 3 5 0 . 0 4 00

2 0

4 0

6 0

8 0

1 0 0

1 2 0

1 4 0

1 6 0Av

erage

packet

latenc

y (ms

ec)

P a c k e t a r r i v a l r a t e

S u n ' s S c h e m e M e i ' s S c h e m e

(a) Packet latency.

0 . 0 1 0 0 . 0 1 5 0 . 0 2 0 0 . 0 2 5 0 . 0 3 0 0 . 0 3 5 0 . 0 4 00 . 0

0 . 1

0 . 2

0 . 3

0 . 4

0 . 5

0 . 6

0 . 7

Avera

ge ou

tage p

robabl

ity

P a c k e t a r r i v a l r a t e

S u n ' s S c h e m e M e i ' s S c h e m e

(b) Outage probability.

Fig. 8: Performance comparison between RA algorithms in [51] and [14] with varying packet arrival rate per vehicleuser.

B. RA for Vehicular Computing System

In recent years, integration of vehicular network withcloud computing, also known as vehicular computingsystem, has attracted increasing attention for its capabil-ity of providing real-time services to on-board users [55],[56]. RA for vehicular computing systems has been in-vestigated in [57], [58]. A semi-Markov decision processbased RA scheme was proposed in [57] for a vehicularcloud computing system while considering heteroge-neous vehicles, i.e., vehicles with different amounts ofcomputing resources. In particular, [57] integrated thecomputational resources of vehicles and RSUs in thevehicular cloud computing system to provide optimumservices. [58] aimed to reduce the serving time by op-timally allocating the available bandwidth in a vehic-ular fog computing system. The optimization problemof [58], formulated based on the requirements of theserving methods, was solved in the following two steps:1) finding the sub-optimal solutions by applying theLagrangian algorithm; 2) performing selection processto obtain the optimum solution.

C. RA for Secured Vehicular Network

RA may also be exploited to enhance the secrecy ofcellular vehicular networks. By observing that LTE-basedV2X communication cannot properly preserve the pri-vacy, [59] evaluated the message delivery with specifiedsecurity. A joint channel and security key assignmentpolicy was presented in [59] to enable a robust andsecure V2X message dissemination. In [60], a RA schemewas proposed to enhance the physical layer security incellular vehicular communication. A max-min secrecy

rate based problem was formulated to allocate powerand sub-carrier while taking into account the outdatedchannel state information (CSI) due to the high mobility.The problem was solved in two stages: (i) with fixed sub-carrier assignment, allocating the power level by using abisection method allocation problem; (ii) finding subop-timal sub-carrier allocation by using greedy algorithm.

D. RA for Vehicle Platooning

In recent years, vehicle platooning networks have beengaining growing research interest as they can lead tosignificant road capacity increase. In [61], the authorsproposed a RA scheme for D2D based vehicle platooningto share control information efficiently and timely. Atime-division based intra-platoon and minimum rateguaranteed inter-platoon RA scheme was proposed toallocate the resources within the platoon, while ensuringoptimized cellular users’ rate. Moreover, to obtain astable platoon, a formation algorithm was proposed in[61] based on a leader evaluation method. Authors in[62] presented a RA strategy to reduce the re-allocationrate that enhances the number of guaranteed services invehicle platooning network. A time dynamic optimiza-tion problem was formulated in [62] under the constraintof a network re-allocation rate. To further reduce thecomputational complexity, their proposed optimizationproblem was converted into a deterministic optimiza-tion problem using the Lyapunov optimization theory[9]. Joint optimization of communication and control invehicle platooning was proposed in [63]. An improvedplatooning system model was developed by taking intoaccount both control and communication factors in a

vehicle platooning. A safety message dissemination sce-nario was considered under an LTE based vehicularnetwork, where the platoon leader vehicle coordinatesthe allocation of available communication and controlresources. A joint optimization problem of RB allocationand control parameter assignment was formulated withthe constraints of communication reliability and platoonstability. Through simulation results, it was shown thattheir proposed RA algorithm reduces the tracking errorwhile maintaining the stability of the platoon.

E. RA for Out-of-Coverage Scenario

A two-step distributed RA scheme was proposed in[64] for out-of-coverage (i.e., out of eNodeB coverage)LTE V2V communication. In the first step, RBs areassigned based on the heading directions of vehicles. Inother words, the same set of RBs are assigned to thevehicles moving in the same direction. In the secondstep, a channel sensing based strategy is utilized to avoidthe packet collision between the vehicles which travel inparallel on the road. Recently, authors in [65] studied RAscheme for delimited out-of-coverage scenario, wherethe network infrastructure assigns the resources to ve-hicles based on the estimated location of vehicles. Morerecently, authors in [66] analyzed and evaluated thesafety message broadcasting performance of LTE-V2Vout-of-coverage mode in an urban intersection scenario,where two resource allocation strategies were presentedto improve the broadcasting performance through vehi-cle assisted relaying.

V. RA FOR HETEROGENEOUS VEHICULAR NETWORKS

A graph based resource scheduling approach was pro-posed in [67] for cooperative relaying in heterogeneousvehicular networks. In LTE, vehicles close to the basestation usually enjoy high data rates due to favourableradio links, while vehicles far away from the base sta-tion suffer from lower data rates due to poor channelconditions. To tackle this problem, cooperative relayingmay be adopted to establish V2V communications fordistant vehicles through DSRC. [67] proposed a bipar-tite graph based scheduling scheme to determine thetransmission strategy for each vehicle user from basestation (i.e., cooperative or non-cooperative) and theselection of relaying vehicles. The scheme proposed in[67] consists of the following three steps: 1) constructa weighted bipartite graph, where the weight of eachedge is determined based on the capacity of the cor-responding V2V link, 2) solve the maximum weightedmatching problem using the KuhnMunkres algorithm,and 3) optimize the number of messages that need tobe relayed, where binary search was utilized to findthe optimal solution. The proposed approach guaranteesfairness among vehicle users and can improve the datarates for the vehicles far away from the base station.

1 E - 3 0 . 0 1 0 . 14 0

4 5

5 0

5 5

6 0

6 5

7 0

7 5

Overa

ll thro

ughp

ut of

V2I li

nk (b

ps/Hz

)

p v a n d p f

F a n g ' s M o d e l G u o ' s M o d e l

P m a x = 2 3 d B m

P m a x = 4 6 d B m

Fig. 9: Comparison of overall throughput of V2I linksfor Guo’s [54] and Fangs [68] methods with respect toreliability of the V2V link (pv) and cellular user link (p f )[54].

Fig. 10: Software defined network (SDN) based hetero-geneous vehicular network.

Very recently, a cascaded Hungarian channel alloca-tion algorithm was presented in [54] for non-orthogonalmultiple access (NOMA) based heterogeneous vehicularnetworks. [54] addressed the channel assignment prob-lem in high-mobility environments with different userQoS requirements and imperfect CSI by formulating achance constrained throughput optimization problem. InFig. 9, the overall throughput is compared with that ofthe RA method reported in [68]. Enhanced performanceis observed for the allocation scheme of [54], thanks to anefficient user scheduling algorithm which fully utilizesthe transmit power to maximize the throughput. It isalso observed that the method proposed in [54] provides

5 0 6 0 7 0 8 0 9 0 1 0 0 1 1 00

2

4

6

8

1 0

1 2

1 4

1 6To

tal uti

lity

B a n d w i d t h

N o n - S D N b a s e d S D N b a s e d

(a) Total utility vs. bandwidth

1 2 3 4 5 6 7- 202468

1 01 21 41 6

Total

utility

T i m e

N o n - S D N b a s e d S D N b a s e d

(b) Total utility vs. time

Fig. 11: Performance comparison between SDN and non-SDN based scenarios [69].

more benefits with increasing transmit powers.Xiao et al. [70] investigated the spectrum sharing for

vehicle users in heterogeneous vehicular networks byexploiting available white space spectrum such as TVwhite space spectrum. A non-cooperative game theoreticapproach was proposed with correlated equilibrium.Their proposed approach allows macrocell base stationsto share the available spectrum with the vehicle usersand improves the spectrum utilization by reusing thewhite space spectrum without degrading the macrocellperformance. By sharing the available spectrum from theLTE and Wi-Fi networks, [69] presented a quality of ex-perience (QoE) based RA scheme for a software definedheterogeneous vehicular network. The system modelconsidered in [69] is shown in Fig. 10. To maximizethe QoE of all vehicles, the proposed scheme exploitsthe CSI of vehicles to extract transmission qualities ofthe vehicles with different access points. A heuristicsolution was proposed to allocate the available resources(in LTE and Wi-Fi networks), which can be used inboth centralized and hybrid software defined networksystems. Fig. 11 presents the performance comparisonbetween the proposed SDN based scenario and non-SDN based scenario. In the non-SDN based scenario, theoptimization for the allocation of LTE and Wi-Fi resourceis carried out separately. Due to the joint optimizationof RA, the proposed method effectively allocate the re-sources and hence outperforms its non-SDN counterpart.An allocation approach for joint LTE and DSRC networkwas proposed in [71]. The proposed approach allocatesthe LTE resources to minimize the number of vehiclesthat compete for channel access in DSRC based commu-nication. The LTE resources are optimally allocated from

the eNodeB, which jointly pairs one vehicle to anotherand allocates the resources to the pair considering aguaranteed signal strength for all communication links.

VI. MACHINE LEARNING BASED RA FOR VEHICULARCOMMUNICATIONS

In vehicular networks, whilst vehicles are expectedto employ various facilities such as advanced on-boardsensors including radar and cameras and even high-performance computing and storage facilities, massiveamounts of data will be generated, processed and trans-mitted. Machine Learning (ML) is envisaged to be aneffective tool to analyse such a huge amount of data andto make more data-driven decisions to enhance vehicu-lar network performance [73]. For details on machinelearning, readers can refer to [74–76].

For resource allocation, the traditional approach isto formulate an optimisation problem and then obtainan optimal or suboptimal solution depending on thetrade-off between target performance and complexity.However, in vehicular networks where the channel qual-ity and network topology can vary continuously, theconventional optimization approach would need to bererun whenever a small change happens, thus incurringhuge overhead [77]. While the ML approach could bean alternative to the prevalent optimisation method,research on applying ML in vehicular networks is stillat an early stage [73].

In the existing literature [72], [78–81], machine learn-ing considering the dynamic characteristics of vehicularnetworks has been applied to channel and power allo-cation, user association and handoff for load balancing,

Fig. 12: The structure of reinforcement learning for V2V links in [72]

and virtual resource management for V2V and V2I com-munications.

In [72], for V2V communications in a cellular network,a distributed channel and power allocation algorithmemploying deep reinforcement learning (RL) [76] hasbeen proposed. With the assumption that an orthogonalresource is allocated for V2I links beforehand, the studyfocuses on resource allocation for V2V links under theconstraints of V2V link latency and minimized interfer-ence impact to V2I links.

The structure of reinforcement learning for V2V linksis shown in Fig. 12. While the agent corresponds toeach V2V link, it interacts with the environment whichincludes various components outside the V2V links. Thestate for characterising the environment is defined as aset of the instantaneous channel information of the V2Vlink and V2I link, the remaining amounts of traffic, theremaining time to meet the latency constraints, and theinterference level and selected channels of neighboursin the previous time slot. At time epoch t, each V2Vlink, as an agent, observes a state st from the state setS , and depending on its policy π, takes an action atamong the action set A. The action is to select the sub-band and transmission power. Following the action, theagent receives a reward rt calculated by the capacity ofV2I links and the V2V latency. The decision policy πis determined by deep learning. At the beginning, theagent tends to select actions randomly. However, with anexploration and exploitation strategy, the agent prefersto exploit the effective actions yielding good rewardsin the past and it also explorers new actions that mayproduce higher rewards in the future. Since the proposedapproach can adjust the power and channel dynamicallyconsidering the latency constraints, the proposed algo-rithm is shown to outperform reference schemes in termsof the probability to satisfy the latency constraint of V2Vlinks.

In [82], a contention-based MAC protocol for V2Vbroadcast transmission using IEEE 802.11p standard forDSRC is investigated. In a scenario with fewer than 50vehicles, IEEE 802.11p can exhibit better performancethan LTE in terms of lower latency and higher packet

delivery ratio than LTE. However, as vehicle density getshigh, the standard becomes unable to accommodate theincreased traffic. In [82], with the aim of overcomingthe scalability issue with the vehicular density, the MLbased approach is proposed to find the optimal con-tention window to enable efficient data packet exchangeswith strict reliability requirements. As a independentlearning agent, each vehicle employs learning to decidecontention window size. The result of each packet trans-mission, either success or fail, is feedback and utilizedfor the window size decision. Through simulation re-sults, it is shown that the proposed ML based approachachieves more reliable packet delivery and higher systemthroughput performance.

In [78], the ML approach is exploited to develop theuser association algorithm for load balancing in hetero-geneous vehicular networks. Considering the regularitycharacteristics of the data flow (generated from vehicularnetworks) in the spatial-temporal dimension, a two-stepassociation algorithm is proposed. The initial associationdecision is made by a single-step reinforcement learning(RL) [75]. Subsequently, base station (i.e., macro, pico andfemto cells) uses historical association patterns to makedecisions for association. In addition, a base station, asan agent of learning, keeps accumulating feedback in-formation and update the association results adaptively.While each base station runs the proposed algorithmin a distributed manner, in the long run, it is shownthat both the real-time feedback and the regular trafficassociation patterns help the algorithm deal with thenetwork changes.

In [79], a vertical handoff strategy has been devisedby using a fuzzy Q-learning approach [75] for heteroge-neous vehicular networks consisting of a globally cov-ered cellular network complemented by the V2I mode.From the OBU side, various information including aver-age received signal strength (RSS) level, vehicle velocityand the amount of data is sent to the RSU side. Then,the RSU side considers the delivered information as wellas the traffic load (i.e., the number of users associatedwith the target network) and makes handoff decisionsby using the fuzzy Q-learning method. With the simu-

20 40 60 80 100Network Density (Vehicles)

0.2

0.4

0.6

0.8

1

Pack

et D

eliv

ery

Rat

io

IEEE 802.11pProposed ML

(a) Packet delivery ratio vs. network density for 256 byte packets

64 128 256 384 512Packet size (bytes)

0.2

0.4

0.6

0.8

1

Pack

et D

eliv

ery

Rat

io

IEEE 802.11pProposed ML

(b) Packet delivery ratio vs. packet size for 60 vehicles

Fig. 13: Performance comparison between IEEE 802.11p standard and the proposed ML based approach for DSCR[82].

lation results, it is shown that the proposed algorithm,which has a real-time learning capability, can determinethe network connectivity to ensure seamless mobilitymanagement without prior knowledge about handoffbehaviour.

In [80], [81], a machine learning approach is exploitedto devise the virtual resource allocation in vehicular net-works. Vertical clouds [83] consisting of various OBUs,RSUs, and remote cloud servers can provide a poolof processing, sensing, storage, and communication re-sources that can be dynamically provisioned for ve-hicular services. The importance of resource allocationin the vehicular cloud is highlighted in [80]. Poorlydesigned resource allocation mechanisms could result inQoS violation or under-utilisation of resources, whereasdynamic resource provisioning techniques are crucialfor meeting the dynamically changing QoS demandsof vehicular services. Against this background, a rein-forcement learning framework has been proposed forresource provisioning to cater for dynamic demandsof resources with stringent QoS requirements. In [81],a two-stage delay-optimal dynamic virtualisation radioscheduling scheme has been developed. Based on thetime-scale, the proposed algorithm is divided into twostages, macro allocation for large time-scale variables(traffic density) and micro allocation with short time-scale variables (channel state and queue state). The dy-namic delay-optimal problem is formulated as a partiallyobserved Markov decision process (POMDP) [74] and isthen solved by an online distributed learning approach.

VII. FUTURE RESEARCH DIRECTIONS

A. Network Slicing based Resource Allocation for C-V2X

Network slicing (NS) is a new paradigm that hasarisen in recent years which helps to create multiplelogical networks tailored to different types of data ser-vices and business operators [84]. NS offers an effec-tive way to meet the requirements of all use casesand enables individual design, deployment, customiza-tion, and optimization of different network slices on acommon infrastructure [85]. In addition to providingvertical slices (for vertical industries), NS may be usedto generate horizontal slices which aim to improve theperformance of user equipment (UE) and enhance theuser experience [86]. Although initially proposed for thepartition of core networks (CN) using techniques suchas network function virtualization (NFV) and softwaredefined networking (SDN), the concept of NS has beenextended to provide efficient end-to-end data services byslicing radio resources in radio access networks (RANs).The slicing of radio resources has mainly involves dy-namic allocations of time and frequency resources basedon the characteristics of multiple data services. This isachieved by providing multiple numerologies, each ofwhich constitutes a set of data frame parameters such asmulti-carrier waveforms, sub-carrier spacings, samplingrates, and frame and symbol durations. For example,an mMTC slice in C-V2X is allocated with relativelysmall subcarrier spacing (i.e., for massive connectivity)and hence large symbol duration. In contrast, URLLCrequires large subcarrier spacing to meet the require-ments of ultra-low latency and stringent reliability. Fig.

NF1 NF2

NF1NF3

NF1 NF2

NF1

NF1 NF2NF3

NF1 NF2

CN Slice #1

CN Slice #2

CN Slice #3

CN Slice #4

CN Slice #5

CN Slice #6

NF3

Radio slice #1

Radio slice #2

Radio slice #3

RAN slice #1

RAN slice #2

RAN slice #3

RAN slice #4

Fixed Access slice #1

Fixed Access slice #2

RSU-1

RSU-2

High-speed

train

Railway station

Vehicles

Slice pairing function

between radio and RAN

Slice pairing

function between

RAN/fixed access

and CN

Frequency (Subcarriers)

Tim

e(O

FD

MS

ym

bo

ls)

Numerology 1

Numerology 2

Numerology3

MTC

eMBB

URLLC

Fig. 14: Network slicing for a C-V2X network consisting of RSUs, high-speed trains, railway stations and movingvehicles.

14 depicts the NS for a C-V2X network consisting ofRSUs, high-speed trains, railway stations and vehicles.

A step-wise approach for designing and applyingfunction decomposition for NS in a 5G CN has beenproposed in [87]. Their main idea is to identify thosefunctions which could be merged in different networkelements as well as their corresponding implicationsfor procedure and information storage. [88] presenteda concrete NS example in the vehicular network domainon efficient distribution of unexpected road conditionsamong cars within a certain range. By properly configur-ing the SDN switch and controller, it is shown in [88] thata network slice for such inter-car communication can bereadily created. In [89], the impact of NS on a 5G RAN,such as the CN/RAN interface, the QoS framework, andthe management framework, has been discussed. It ispointed out in [89] that dynamic NS is preferred in orderto cater for rapid change of traffic patterns. A compre-hensive work on applications of NS to support a diverserange of C-V2X use cases has been proposed in [90].Major C-V2X slices identified in [90] are: autonomousdriving, tele-operated driving, vehicular infotainment,and vehicular remote diagnostics and management. Forexample, the slice for supporting tele-operated drivingenables URLLC and the slice for vehicular infotainmentmay use multiple random access technologies (RATs) tosupport higher throughput. Moreover, slicing may becarried out in different vehicular devices according totheir storage and computing capacities as well as thenature of the data services, a scenario similar to mobileedge computing [90].

It is noted that NS can be carried out not only

at higher levels of wireless networks, but also in thephysical layer (PHY). In 2017, a multi-service systemframework implemented in both time and frequencydomains has been proposed in [91], [92]. A major issuehere is how to select and design multicarrier wave-forms with good time-frequency localization, low out-of-band power emission, low inter-carrier interference (ICI)among different sub-bands using different numerologies,and capability to support multi-rate implementation.Multicarrier waveform design for PHY NS such as fil-tered orthogonal frequency-multiple access (F-OFDM),windowed-OFDM, and universal filtered multi-carrier(UFMC) have been studied in [91], [93], [94]. In the con-text of C-V2X, the design of multiple numerologies formodest and high mobility environments is an interestingand pressing research issue. In this case, one needs todeal with doubly selective fading channels which couldlead to severe ICI and inter-symbol interference. Anotherinteresting research is direction is how to design andoptimize network slices to provide guaranteed quality ofservices (QoS) for C-V2X while balancing the services ofother vertical applications under the constraint of limitedradio resources.

B. Machine Learning Perspective in Resource Allocation

Whilst the strong potentials of applying ML in vehicu-lar networks have been discussed with the initial effortsin Section VI, how to adapt and exploit ML to accountfor the peculiar characteristics of vehicular networksand services still remains as challenges and representsa promising research direction [77]. Vehicular networkssignificantly differ from the scenarios where machine

learning has been conventionally exploited in terms ofstrong dynamics in wireless networks, network topolo-gies, traffic flow, etc. How to efficiently learn and predictsuch dynamics based on historical data for the benefitor reliable communications is still an open issue [73]. Inaddition, data is supposed to be generated and storedacross various units in vehicular networks, e.g., OBUs,RSUs, and remote clouds. It could be interesting to inves-tigate whether traditional centralised ML approaches canbe exploited to work efficiently in a distributed manner.For collective intelligent decision making in learning-capable vehicular networks, the overhead for informa-tion sharing and complexity of learning algorithms needto be taken into account [95].

C. Context Aware Resource Allocation for Vehicular Commu-nications

Existing work on resource allocation for vehicular net-works mostly deals with efficient allocation of resourceblocks such as frequency carriers or time-slots. However,most of the prior work on resource allocation did notconsider context-aware/on-demand data transfer appli-cations in vehicular networks. Since on-demand datatransfer applications need to meet constraints such asdeadline of the requested data items or priority of dataitems, to ensure a reliable service, there is a need forresearch to consider those more thoroughly. Althoughthere is a lot of prior work [96–98] on performanceevaluation of on-demand data dissemination scenariosin terms of the above constraints, they do not deal withthe allocation of resource blocks, which is important for5G networks.

VIII. CONCLUSIONS

In this paper, we have surveyed radio resource al-location schemes in vehicular networks. We have cat-egorized these schemes into three categories based onthe types of vehicular networks, i.e., DSRC vehicularnetwork, cellular vehicular network, and heterogeneousvehicular network. For each category, the available lit-erature is reviewed and summarized while highlightingthe pros and cons of the resource allocation schemes.We have also discussed several open and challengingfuture research directions for radio resource allocationin vehicular networks. It is anticipated that this paperwill provide a quick and comprehensive understandingof the current state of the radio resource allocationstrategies in vehicular networks while attracting andmotivating more researchers into this challenging area.

REFERENCES

[1] K. Liu, L. Feng, P. Dai, V. C. S. Lee, S. H. Son, and J. Cao, “Coding-assisted broadcast scheduling via memetic computing in SDN-based vehicular networks,” IEEE Trans. Intell. Transp. Syst., vol. 8,no. 19, pp. 1–12, Aug. 2018.

[2] Z. Wang, J. Zheng, Y. Wu, and N. Mitton, “A centrality-based RSUdeployment approach for vehicular Ad-Hoc networks,” in 2017IEEE Int. Conference on Communications (ICC), May 2017, pp. 1–5.

[3] H. Nguyen, M. Noor-A-Rahim, Z. Liu, D. Jamaludin, Y. L. Guan,H. Nguyen, M. Noor-A-Rahim, Z. Liu, D. Jamaludin, and Y. L.Guan, “A Semi-Empirical Performance Study of Two-Hop DSRCMessage Relaying at Road Intersections,” Information, vol. 9, no. 6,p. 147, Jun. 2018.

[4] X. Cheng, L. Yang, and X. Shen, “D2D for intelligent transporta-tion systems: a feasibility study,” IEEE Trans. Intell. Transp. Syst.,vol. 16, no. 4, pp. 1784–1793, Aug. 2015.

[5] G. G. M. N. Ali, M. N. A. Rahim, P. H. J. Chong, and Y. L.Guan, “Analysis and improvement of reliability through codingfor safety message broadcasting in urban vehicular networks,”IEEE Transactions on Vehicular Technology, pp. 1–1, 2018.

[6] H. Seo, K.-D. Lee, S. Yasukawa, Y. Peng, and P. Sartori, “LTEevolution for vehicle-to-everything services,” IEEE Commun. Mag.,vol. 54, no. 6, pp. 22–28, Jun. 2016.

[7] A. Bazzi, B. M. Masini, A. Zanella, and I. Thibault, “On theperformance of IEEE 802.11p and LTE-V2V for the coopera-tive awareness of connected vehicles,” IEEE Trans. Veh. Technol.,vol. 66, no. 11, pp. 10 419–10 432, Nov. 2017.

[8] J. B. Kenney, “Dedicated short-range communications (DSRC)standards in the United States,” Proc. IEEE, vol. 99, no. 7, pp.1162–1182, Jul. 2011.

[9] L. Georgiadis, M. J. Neely, and L. Tassiulas, “Resource allocationand cross-layer control in wireless networks,” Found. Trends Netw.,vol. 1, no. 1, pp. 1–144, Apr. 2006.

[10] R. Zhang, Y.-C. Liang, and S. Cui, “Dynamic resource allocationin cognitive radio networks,” IEEE Signal Process. Mag., vol. 27,no. 3, pp. 102–114, May 2010.

[11] X. Wang and G. B. Giannakis, “Resource allocation for wirelessmultiuser OFDM networks,” IEEE Trans. Inf. Theory, vol. 57, no. 7,pp. 4359–4372, Jul. 2011.

[12] M. Botsov, M. Klugel, W. Kellerer, and P. Fertl, “Location de-pendent resource allocation for mobile device-to-device commu-nications,” in 2014 IEEE Wireless Communications and NetworkingConference (WCNC), Apr. 2014, pp. 1679–1684.

[13] Y. Ren, F. Liu, Z. Liu, C. Wang, and Y. Ji, “Power control inD2D-based vehicular communication networks,” IEEE Trans. Veh.Technol., vol. 64, no. 12, pp. 5547–5562, Dec. 2015.

[14] W. Sun, E. G. Strom, F. Brannstrom, K. C. Sou, and Y. Sui, “Radioresource management for D2D-based V2V communication,” IEEETrans. Veh. Technol., vol. 65, no. 8, pp. 6636–6650, Aug. 2016.

[15] W. Sun, D. Yuan, E. G. Strom, and F. Brannstrom, “Cluster-basedradio resource management for D2D-supported safety-criticalV2X communications,” IEEE Trans. Wireless Commun., vol. 15,no. 4, pp. 2756–2769, Apr. 2016.

[16] N. Cheng, H. Zhou, L. Lei, N. Zhang, Y. Zhou, X. Shen, and F. Bai,“Performance analysis of vehicular device-to-device underlaycommunication,” IEEE Trans. Veh. Technol., vol. 66, no. 6, pp. 5409–5421, Jun. 2017.

[17] Y. Zhang and G. Cao, “V-PADA: vehicle-platoon-aware dataaccess in VANETs,” IEEE Trans. Veh. Technol., vol. 60, no. 5, pp.2326–2339, Jun. 2011.

[18] Z. Zhao, X. Cheng, M. Wen, B. Jiao, and C.-X. Wang, “Channel es-timation schemes for IEEE 802.11p standard,” IEEE Intell. Transp.Syst. Mag., vol. 5, no. 4, pp. 38–49, 2013.

[19] Y. L. Morgan, “Notes on DSRC & WAVE standards suite: itsarchitecture, design, and characteristics,” IEEE Commun. SurveysTuts., vol. 12, no. 4, pp. 504–518, 2010.

[20] H. Hartenstein and K. Laberteaux, VANET : vehicular applicationsand inter-networking technologies. Wiley, 2010.

[21] M. Noor-A-Rahim, G. G. M. N. Ali, H. Nguyen, and Y. L. Guan,“Performance analysis of IEEE 802.11p safety message broadcastwith and without relaying at road intersection,” IEEE Access,vol. 6, pp. 23 786–23 799, 2018.

[22] G. Cecchini, A. Bazzi, B. M. Masini, and A. Zanella, “MAP-RP:Map-based resource reselection procedure for autonomous LTE-V2V,” in IEEE International Symposium on Personal, Indoor andMobile Radio Communications, PIMRC, 2018.

[23] X. Lin, J. Andrews, A. Ghosh, and R. Ratasuk, “An overview of3GPP device-to-device proximity services,” IEEE Commun. Mag.,vol. 52, no. 4, pp. 40–48, Apr. 2014.

[24] Z. Hameed Mir and F. Filali, “LTE and IEEE 802.11p for vehicu-lar networking: a performance evaluation,” EURASIP J. WirelessCommun. Netw., vol. 2014, no. 1, p. 89, Dec. 2014.

[25] G. Araniti, C. Campolo, M. Condoluci, A. Iera, and A. Molinaro,“LTE for vehicular networking: a survey,” IEEE Commun. Mag.,vol. 51, no. 5, pp. 148–157, May 2013.

[26] K. Zheng, Q. Zheng, P. Chatzimisios, W. Xiang, and Y. Zhou,“Heterogeneous vehicular networking: a survey on architecture,challenges, and solutions,” IEEE Commun. Surveys Tuts., vol. 17,no. 4, pp. 2377–2396, 2015.

[27] F. Dressler, H. Hartenstein, O. Altintas, and O. Tonguz, “Inter-vehicle communication: Quo vadis,” IEEE Commun. Mag., vol. 52,no. 6, pp. 170–177, Jun. 2014.

[28] R. Atat, E. Yaacoub, M.-S. Alouini, and F. Filali, “Delay efficientcooperation in public safety vehicular networks using LTE andIEEE 802.11p,” in 2012 IEEE Consumer Communications and Net-working Conference (CCNC). IEEE, Jan. 2012, pp. 316–320.

[29] C.-L. Huang, Y. Fallah, R. Sengupta, and H. Krishnan, “Adaptiveintervehicle communication control for cooperative safety sys-tems,” IEEE Network, vol. 24, no. 1, pp. 6–13, Jan. 2010.

[30] Liu Fuqiang and Shan Lianhai, “Heterogeneous vehicular com-munication architecture and key technologies,” ZTE Commun.,vol. 8, no. 4, pp. 39–44, 2010.

[31] Q. Zheng, K. Zheng, L. Sun, and V. C. M. Leung, “Dynamicperformance analysis of uplink transmission in cluster-based het-erogeneous vehicular networks,” IEEE Trans. Veh. Technol., vol. 64,no. 12, pp. 5584–5595, Dec. 2015.

[32] P. Dai, K. Liu, X. Wu, Y. Liao, V. C. S. Lee, and S. H. Son,“Bandwidth efficiency and service adaptiveness oriented datadissemination in heterogeneous vehicular networks,” IEEE Trans.Veh. Technol., vol. 67, no. 7, pp. 6585–6598, Jul. 2018.

[33] S. Cespedes and X. S. Shen, “On achieving seamless IP communi-cations in heterogeneous vehicular networks,” IEEE Trans. Intell.Transp. Syst., vol. 16, no. 6, pp. 3223–3237, Dec. 2015.

[34] K. Shafiee, A. Attar, and V. C. M. Leung, “Optimal distributedvertical handoff strategies in vehicular heterogeneous networks,”IEEE J. Sel. Areas Commun., vol. 29, no. 3, pp. 534–544, Mar. 2011.

[35] H. He, H. Shan, A. Huang, and L. Sun, “Resource allocation forvideo streaming in heterogeneous cognitive vehicular networks,”IEEE Trans. Veh. Technol., vol. 65, no. 10, pp. 7917–7930, Oct. 2016.

[36] V. P. Harigovindan, A. V. Babu, and L. Jacob, “Ensuring fairaccess in IEEE 802.11p-based vehicle-to-infrastructure networks,”EURASIP J. Wireless Commun. Netw., vol. 2012, no. 1, p. 168, Dec.2012.

[37] G. V. Rossi and K. K. Leung, “Optimised CSMA/CA protocolfor safety messages in vehicular ad-hoc networks,” in 2017 IEEESymposium on Computers and Communications (ISCC). IEEE, Jul.2017, pp. 689–696.

[38] E. Karamad and F. Ashtiani, “A modified 802.11-based MACscheme to assure fair access for vehicle-to-roadside communica-tions,” Computer Commun., vol. 31, no. 12, pp. 2898–2906, Jul. 2008.

[39] G. V. Rossi, K. K. Leung, and A. Gkelias, “Density-based optimaltransmission for throughput enhancement in vehicular ad-hocnetworks,” in 2015 IEEE International Conference on Communica-tions (ICC). IEEE, Jun. 2015, pp. 6571–6576.

[40] W. Alasmary and W. Zhuang, “Mobility impact in IEEE 802.11pinfrastructureless vehicular networks,” Ad Hoc Networks, vol. 10,no. 2, pp. 222–230, Mar. 2012.

[41] F. Cali, M. Conti, and E. Gregori, “IEEE 802.11 protocol: designand performance evaluation of an adaptive backoff mechanism,”IEEE J. Sel. Areas Commun., vol. 18, no. 9, pp. 1774–1786, Sep. 2000.

[42] M.-W. Ryu, S.-H. Cha, and K.-H. Cho, “DSRC-based channelallocation algorithm for emergency message dissemination inVANETs.” Springer, Berlin, Heidelberg, 2011, pp. 105–112.

[43] S.-T. Sheu, Y.-C. Cheng, and J.-S. Wu, “A channel access schemeto compromise throughput and fairness in IEEE 802.11p multi-rate/multi-channel wireless vehicular networks,” in 2010 IEEE71st Vehicular Technology Conference, 2010, pp. 1–5.

[44] G. G. M. N. Ali, M. Noor-A-Rahim, M. A. Rahman, S. K. Saman-tha, P. H. J. Chong, and Y. L. Guan, “Efficient real-time coding-assisted heterogeneous data access in vehicular networks,” IEEEInternet Things J., vol. 5, no. 5, pp. 3499–3512, Oct. 2018.

[45] ——, “An efficient cross-layer coding-assisted heterogeneous dataaccess in vehicular networks,” in 2018 IEEE International Confer-ence on Communications (ICC). IEEE, May 2018, pp. 1–7.

[46] Huawei White Paper, “Connected cars onthe road to 5G,” 2017. [Online]. Avail-

able: https://www.huawei.com/uk/industry-insights/outlook/mobile-broadband/xlabs/insights-whitepapers/huawei-whitepaper-connected-car-on-the-road-to-5g

[47] R. Zhang, X. Cheng, Q. Yao, C.-X. Wang, Y. Yang, and B. Jiao,“Interference graph-based resource-sharing schemes for vehicularnetworks,” IEEE Trans. Veh. Technol., vol. 62, no. 8, pp. 4028–4039,Oct. 2013.

[48] Y. Meng, Y. Dong, X. Liu, and Y. Zhao, “An interference-awareresource allocation scheme for connectivity improvement in ve-hicular networks,” IEEE Access, vol. 6, pp. 51 319–51 328, 2018.

[49] D. B. West, Introduction to graph theory. Prentice Hall, 1996.[50] T. Etzion and P. Ostergard, “Greedy and heuristic algorithms for

codes and colorings,” IEEE Trans. Inf. Theory, vol. 44, no. 1, pp.382–388, 1998.

[51] J. Mei, K. Zheng, L. Zhao, Y. Teng, and X. Wang, “A latency andreliability guaranteed resource allocation scheme for LTE V2Vcommunication systems,” IEEE Trans. Wireless Commun., vol. 17,no. 6, pp. 3850–3860, Jun. 2018.

[52] L. Liang, G. Y. Li, and W. Xu, “Resource allocation forD2D-enabled vehicular communications,” IEEE Trans. Commun.,vol. 65, no. 7, pp. 3186–3197, Jul. 2017.

[53] S. Zhang, Y. Hou, X. Xu, and X. Tao, “Resource allocation in D2D-based V2V communication for maximizing the number of concur-rent transmissions,” in IEEE 27th Annual International Symposiumon Personal, Indoor, and Mobile Radio Communications (PIMRC), Sep.2016, pp. 1–6.

[54] S. Guo and X. Zhou, “Robust resource allocation with imperfectchannel estimation in NOMA-based heterogeneous vehicular net-works,” IEEE Trans. Commun., vol. 67, no. 3, pp. 2321–2332, Mar.2019.

[55] M. Gerla, “Vehicular cloud computing,” in Annual MediterraneanAd Hoc Networking Workshop (Med-Hoc-Net). IEEE, jun 2012, pp.152–155.

[56] S. Bitam, A. Mellouk, and S. Zeadally, “VANET-cloud: a genericcloud computing model for vehicular Ad Hoc networks,” IEEEWireless Commun., vol. 22, no. 1, pp. 96–102, Feb 2015.

[57] C.-C. Lin, D.-J. Deng, and C.-C. Yao, “Resource allocation invehicular cloud computing systems with heterogeneous vehiclesand roadside units,” IEEE Internet Things J., vol. 5, no. 5, pp. 3692–3700, Oct. 2018.

[58] F. Lin, Y. Zhou, G. Pau, and M. Collotta, “Optimization-orientedresource allocation management for vehicular fog computing,”IEEE Access, vol. 6, pp. 69 294–69 303, 2018.

[59] K. J. Ahmed and M. J. Lee, “Secure resource allocation for LTE-based V2X service,” IEEE Trans. Veh. Technol., vol. 67, no. 12, pp.11 324–11 331, Dec. 2018.

[60] W. Yang, R. Zhang, C. Chen, and X. Cheng, “Secrecy-basedresource allocation for vehicular communication networks withoutdated CSI,” in 2017 IEEE 86th Vehicular Technology Conference(VTC-Fall). IEEE, Sep. 2017, pp. 1–5.

[61] R. Wang, J. Wu, and J. Yan, “Resource Allocation for D2D-EnabledCommunications in Vehicle Platooning,” IEEE Access, vol. 6, pp.50 526–50 537, 2018.

[62] Y. Meng, Y. Dong, C. Wu, X. Liu, Y. Meng, Y. Dong, C. Wu, andX. Liu, “A low-cost resource re-allocation scheme for increasingthe number of guaranteed services in resource-limited vehicularnetworks,” Sensors, vol. 18, no. 11, p. 3846, Nov. 2018.

[63] J. Mei, K. Zheng, L. Zhao, L. Lei, and X. Wang, “Joint radioresource allocation and control for vehicle platooning in LTE-V2Vnetwork,” IEEE Trans. Veh. Technol., vol. 67, no. 12, pp. 12 218–12 230, Dec. 2018.

[64] J. Yang, B. Pelletier, and B. Champagne, “Enhanced autonomousresource selection for LTE-based V2V communication,” in 2016IEEE Vehicular Networking Conference (VNC), Dec. 2016, pp. 1–6.

[65] T. Sahin and M. Boban, “Radio resource allocation for reliableout-of-coverage V2V communications,” in IEEE 87th VehicularTechnology Conference (VTC Spring), Jun. 2018, pp. 1–5.

[66] M. N. A. Rahim, G. G. M. N. Ali, Y. L. Guan, A. Beshah,P. H. J. Chong, and P. Dirk, “Broadcast performance analysisand improvements of the LTE-V2V autonomous mode at roadintersection,” IEEE Trans. Veh. Technol. (Early Access), pp. 1–1, 2019.

[67] K. Zheng, F. Liu, Q. Zheng, W. Xiang, and W. Wang, “A graph-based cooperative scheduling scheme for vehicular networks,”IEEE Trans. Veh. Technol., vol. 62, no. 4, pp. 1450–1458, May 2013.

[68] F. Fang, H. Zhang, J. Cheng, S. Roy, and V. C. M. Leung, “Jointuser scheduling and power allocation optimization for energy-efficient NOMA systems with imperfect CSI,” IEEE J. Sel. AreasCommun., vol. 35, no. 12, pp. 2874–2885, Dec. 2017.

[69] W. Huang, L. Ding, D. Meng, J.-N. Hwang, Y. Xu, and W. Zhang,“QoE-based resource allocation for heterogeneous multi-radiocommunication in software-defined vehicle networks,” IEEE Ac-cess, vol. 6, pp. 3387–3399, 2018.

[70] Z. Xiao, X. Shen, F. Zeng, V. Havyarimana, D. Wang, W. Chen,and K. Li, “Spectrum resource sharing in heterogeneous vehic-ular networks: a noncooperative game-theoretic approach Withcorrelated equilibrium,” IEEE Trans. Veh. Technol., vol. 67, no. 10,pp. 9449–9458, Oct. 2018.

[71] X. Cao, L. Liu, Y. Cheng, L. X. Cai, and C. Sun, “On optimaldevice-to-device resource allocation for minimizing end-to-enddelay in VANETs,” IEEE Trans. Veh. Technol., vol. 65, no. 10, pp.7905–7916, Oct. 2016.

[72] H. Ye and G. Y. Li, “Deep reinforcement learning for resourceallocation in V2V communications,” in Proc. 2018 IEEE Int’l. Conf.Commun. (ICC). IEEE, 2018, pp. 1–6.

[73] L. Liang, H. Ye, and G. Y. Li, “Toward intelligent vehicularnetworks: a machine learning framework,” IEEE Internet ThingsJ., vol. 6, no. 1, pp. 124–135, Feb. 2019.

[74] C. Jiang, H. Zhang, Y. Ren, Z. Han, K. Chen, and L. Hanzo,“Machine Learning Paradigms for Next-Generation Wireless Net-works,” IEEE Wireless Commun., vol. 24, no. 2, pp. 98–105, Apr.2017.

[75] R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduc-tion. Cambridge, MA, USA: MIT Press, 1998.

[76] K. Arulkumaran, M. P. Deisenroth, M. Brundage, and A. A.Bharath, “Deep reinforcement learning: a brief survey,” IEEESignal Process. Mag., vol. 34, no. 6, pp. 26–38, Nov. 2017.

[77] H. Ye, L. Liang, G. Y. Li, J. Kim, L. Lu, and M. Wu, “Machinelearning for vehicular networks: recent advances and applicationexamples,” IEEE Signal Process. Mag., vol. 13, no. 2, pp. 94–101,Jun. 2018.

[78] W. Li, X. Ma, J. Wu, K. S. Trivedi, X. L. Huang, and Q. Liu, “Ana-lytical model and performance evaluation of long-term evolutionfor vehicle safety services,” IEEE Trans. Veh. Technol., vol. 66, no. 3,pp. 1926–1939, Mar. 2017.

[79] Y. Xu, L. Li, B. Soong, and C. Li, “Fuzzy Q-learning based verticalhandoff control for vehicular heterogeneous wireless network,” inProc. 2014 IEEE Int’l. Conf. Commun. (ICC). IEEE, 2014, pp. 5653–5658.

[80] M. A. Salahuddin, A. Al-Fuqaha, and M. Guizani, “Reinforcementlearning for resource provisioning in the vehicular cloud,” IEEEWireless Commun., vol. 23, no. 4, pp. 128–135, Aug. 2016.

[81] Q. Zheng, K. Zheng, H. Zhang, and V. C. M. Leung, “Delay-optimal virtualized radio resource scheduling in software-definedvehicular networks via stochastic learning,” IEEE Trans. Veh.Technol., vol. 65, no. 10, pp. 7857–7867, Oct. 2016.

[82] A. Pressas, Z. Sheng, F. Ali, D. Tian, and M. Nekovee,“Contention-based learning mac protocol for broadcast vehicle-to-vehicle communication,” in 2017 IEEE Vehicular NetworkingConference (VNC), Nov. 2017, pp. 263–270.

[83] M. G. E. Lee, E. Lee and S. Y. Oh, “Vehicular cloud networking:architecture and design principles,” IEEE Commun. Mag., vol. 52,no. 2, pp. 148–155, Feb. 2014.

[84] N. Alliance, “NGWN 5G White Paper,” Tech. Rep., Feb. 2015.[85] X. Foukas, G. Patounas, A. Elmokashfi, and M. K. Marina, “Net-

work slicing in 5G: survey and challenges,” IEEE Commun. Mag.,vol. 55, no. 5, pp. 94–100, May 2017.

[86] Q. Li, G. Wu, A. Papathanassiou, and U. Mukherjee, “An end-to-end network slicing framework for 5G wireless communicationsystems,” Aug. 2016. [Online]. Available: http://arxiv.org/abs/1608.00572

[87] M. R. Sama, X. An, Q. Wei, and S. Beker, “Reshaping the mobilecore network via function decomposition and network slicing forthe 5G Era,” in 2016 IEEE Wireless Communications and NetworkingConference, Apr. 2016, pp. 1–7.

[88] T. Soenen, R. Banerjee, W. Tavernier, D. Colle, and M. Pickavet,“Demystifying network slicing: from theory to practice,” in 2017IFIP/IEEE Symposium on Integrated Network and Service Management(IM), May 2017, pp. 1115–1120.

[89] I. da Silva, G. Mildh, A. Kaloxylos, P. Spapis, E. Buracchini,A. Trogolo, G. Zimmermann, and N. Bayer, “Impact of networkslicing on 5G radio access networks,” in 2016 European Conferenceon Networks and Communications (EuCNC), Jun. 2016, pp. 153–157.

[90] C. Campolo, A. Molinaro, A. Iera, and F. Menichella, “5G networkslicing for vehicle-to-everything services,” IEEE Wireless Commun.,vol. 24, no. 6, pp. 38–45, Dec. 2017.

[91] L. Zhang, A. Ijaz, P. Xiao, and R. Tafazolli, “Multi-service system:an enabler of flexible 5G air interface,” IEEE Commun. Mag.,vol. 55, no. 10, pp. 152–159, Oct. 2017.

[92] L. Zhang, A. Ijaz, J. Mao, P. Xiao, and R. Tafazolli, “Multi-servicesignal multiplexing and isolation for physical-layer network slic-ing (PNS),” in 2017 IEEE 86th Vehicular Technology Conference(VTC-Fall), Sep. 2017, pp. 1–5.

[93] L. Zhang, A. Ijaz, P. Xiao, M. M. Molu, and R. Tafazolli, “FilteredOFDM systems, algorithms, and performance analysis for 5G andbeyond,” IEEE Trans. Commun., vol. 66, no. 3, pp. 1205–1218, Mar.2018.

[94] X. Zhang, L. Zhang, P. Xiao, D. Ma, J. Wei, and Y. Xin, “Mixednumerologies interference analysis and inter-numerology interfer-ence cancellation for windowed OFDM systems,” IEEE Trans. Veh.Technol., vol. 67, no. 8, pp. 7047–7061, Aug. 2018.

[95] X. S. H. Peng, Le Liang and G. Y. Li, “Vehicular communications:a network layer perspective,” IEEE Trans. Veh. Technol., vol. 68,no. 2, pp. 1064–1078, Feb 2019.

[96] C. Zhan, V. C. S. Lee, J. Wang, and Y. Xu, “Coding-based databroadcast scheduling in on-demand broadcast,” IEEE Trans. Wire-less Commun., vol. 10, no. 11, pp. 3774–3783, Nov. 2011.

[97] X. Wang, C. Yuen, and Y. Xu, “Coding-based data broadcastingfor time-critical applications with rate adaptation,” IEEE Trans.Veh. Technol., vol. 63, no. 5, pp. 2429–2442, Jun. 2014.

[98] S. Chen, J. Hu, Y. Shi, Y. Peng, J. Fang, R. Zhao, and L. Zhao,“Vehicle-to-everything (V2X) services supported by LTE-based

systems and 5G,” IEEE Commun. Standards Mag., 2017.