routing protocols design for autonomous vehicles in mobile … · 2019. 10. 30. · autonomous...

57
Routing Protocols Design for Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755: Individual project Supervised by: Prof. Weifa Liang The Australian National University October 2019 c Shangru Liu 2019

Upload: others

Post on 26-Aug-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Routing Protocols Design forAutonomous Vehicles in Mobile

Edge Computing

Shangru Liu

A report submitted for the courseCOMP8755: Individual project

Supervised by: Prof. Weifa LiangThe Australian National University

October 2019c© Shangru Liu 2019

Page 2: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Except where otherwise indicated, this report is my own original work.

Shangru Liu25 October 2019

Page 3: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Acknowledgments

I would like to thank my supervisor Prof. Weifa Liang for his patient guid-ance throughout the project. I really appreciate his advice not only for theacademic field, but also for life and career. He always care so much aboutthe project, and also respond promptly to my questions and queries. Thanksto his encouragement, I came out of my darkest hour. Moreover, this reportimproves a lot to his careful editing.

I would also like to thank my parents. I am continually benefit from theirsupport and encouragement throughout my life.

iii

Page 4: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:
Page 5: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Abstract

Autonomous vehicles can significantly ease traffic congestion and improvetraffic safety, which will be an important part of Intelligent TransportationSystem (ITS) in future. The research on autonomous vehicles is mainly to meettheir needs for computing and communication. Mobile data offloading is analternative solution to fulfill autonomous vehicles’ needs in communicationby offloading mobile data from cellular networks to vehicular networks.

In this report, We propose two routing protocols named expired-timebased routing (ETBR) and expired-time update protocol (ETUP) to find V2Vpaths and dynamically update link information with the help of mobile edgecomputing (MEC). These two routing protocols aim to achieve data offload-ing from the cellular network to the vehicular network. We also conductexperiments to evaluate the performance of the proposed protocols. The mea-surements of performance evaluation are packet delivery rate, average linklifetime and offloading fraction which represents how much mobile data isrouted from the cellular network to the vehicular network. Experimentalresults show that the proposed routing protocols deliver around 97% of pack-ets successfully which perform much better than a existing routing protocol.Moreover, the results show that the average lifetime and offloading fractionincrease with the increase on the density of vehicles.

v

Page 6: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

vi

Page 7: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Contents

Acknowledgments iii

Abstract v

1 Introduction 11.1 Motivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.3 Report Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Related Work 7

3 Preliminaries 113.1 System model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113.2 Problem definition . . . . . . . . . . . . . . . . . . . . . . . . . . 12

4 Protocols design for AVNET 134.1 AVNET offloading . . . . . . . . . . . . . . . . . . . . . . . . . . 134.2 Expired-time-based Routing . . . . . . . . . . . . . . . . . . . . . 154.3 Expired-time update protocol . . . . . . . . . . . . . . . . . . . . 224.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

5 Evaluation 255.1 Experiment setting . . . . . . . . . . . . . . . . . . . . . . . . . . 255.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

6 Conclusion and Future Work 316.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

Bibliography 33

Appendix 1 37

Appendix 2 39

vii

Page 8: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

viii Contents

Appendix 3 43

Appendix 4 45

Page 9: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

List of Figures

1.1 An example of V2X communication . . . . . . . . . . . . . . . . 2

3.1 An MEC-based AVNET model . . . . . . . . . . . . . . . . . . . 12

4.1 The execution procedure . . . . . . . . . . . . . . . . . . . . . . . 144.2 Two cases of link lifetime between neighbor vehicles . . . . . . 164.3 Positions of neighbor vehicles when they are out of radio range 174.4 An example of running the algorithm . . . . . . . . . . . . . . . 194.5 Path setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214.6 Three cases of link lifetime between neighbor vehicles . . . . . 24

5.1 Delivery ratio of packets . . . . . . . . . . . . . . . . . . . . . . . 265.2 Average offloading fraction for different reporting interval . . . 275.3 Offloading fraction . . . . . . . . . . . . . . . . . . . . . . . . . . 285.4 Average lifetime . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

ix

Page 10: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

x LIST OF FIGURES

Page 11: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

List of Tables

4.1 Table of notations used in the report . . . . . . . . . . . . . . . . 23

xi

Page 12: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

xii LIST OF TABLES

Page 13: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Chapter 1

Introduction

Intelligent transportation system (ITS) [Chaturvedi and Srivastava, 2017] is aspecific application of Internet of things (IoT) technology in vehicles, whichaims to reduce traffic congestion and accidents. In order to reduce trafficaccidents caused by drivers’ errors, Autonomous Vehicle (AV) is introduced,which enables control of the vehicle from the driver to the automated drivingsystem [Fagnant and Kockelman, 2015]. Autonomous vehicles can signifi-cantly ease traffic congestion and improve traffic safety, which will be animportant part of ITS in future. Optimistically, autonomous vehicles can re-duce traffic accidents by 90%. Because more than 90% of traffic accidentsare caused by human drivers’ driving errors [Papadoulis et al., 2019]. Publichealth benefits also will be greatly optimized by reducing emissions, increas-ing cyclist safety once all vehicles operate in autonomous mode [Pettigrewet al., 2018]. Due to advances in sensor technology and deep learning [Li et al.,2018a], autonomous vehicles are expected to be commercialized and even com-pletely replace conventional vehicles in coming years. To support autonomousvehicle applications, salient challenges in computing and communication needto be addressed [Peng et al., 2019].

1.1 Motivations

From computing perspective, numerous computing tasks need to be com-pleted for sensing the real-time environment and making driving decisions [Suet al., 2018]. With the current battery life and computing power of autonomousvehicles, necessary urgent, delay-sensitive and computation-intensive taskscannot be fully supported within strict response time [Dai et al., 2019]. Mo-bile Edge computing (MEC) has been regarded as an effective technologyto overcome challenges for autonomous vehicles via offloading computingtasks to edge servers which are deployed near end users (usually at BaseStations)[Feng et al., 2017]. Therefore, it is important for the MEC server toexecute as many of these time-sensitive driving-decision tasks as possible.

1

Page 14: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

2 Introduction

Figure 1.1: An example of V2X communication

From communication perspective, Vehicle-to-everything (V2X) communica-tions [Peng et al., 2017] are important to autonomous vehicles for sharingcontext information include velocity, position, cruise control information andtraffic information. V2X communications can be divided into three categories:vehicle to infrastructure (V2I) communications, vehicle to vehicle (V2V) com-munications and vehicle to pedestrian (V2P) communications. As depictedin Figure 1.1, V2I communications mean that the vehicle uses IEEE 802.11pinterface to communicate with the preset Road Side Unit (RSU) when thevehicle is in the signal coverage of RSU. V2V communications are happenedbetween two peered vehicles when they are using IEEE 802.11p interface tocommunicate. Due to the mobility of vehicles, V2V communications are themost challenge part of autonomous vehicles for exchanging road informationand high definition maps to ensure vehicular safety and make route planning[Nguyen-Minh et al., 2016]. Current inter-connections of vehicles are mainlysupported by sub-6 GHz cellular network [Do-Duy and Kim, 2015] which isconsidered insufficient to provide high level quality of service (QoS) becauseof radio interference and limited resources. IEEE 802.11p is a standard whichis designed for vehicle communications with a upwards of bit rate of 27 Mb/sin the band of 5.9 GHz [Wu et al., 2019]. Millimeter-wave (mmWave) com-munications can achieve up to 2.5 GB/s with unlicensed 60 GHz spectrumthrough IEEE 802.11p interface which could support V2V communicationsrequirements of autonomous vehicles. There has been lots of research onusing IEEE 802.11p in vehicular networks efficiently to reduce the traffic inthe cellular network by delivering data through IEEE 802.11p network ratherthan cellular network [Li et al., 2018b]. These research are mainly using theinfrastructure for data offloading, i.e., V2I offloading. However, V2I commu-nications are not sufficient for data offloading. V2I communications need the

Page 15: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

§1.1 Motivations 3

deployment of more RSUs which cost more. V2V offloading communicationparadigm also was evaluated in some research. But the offloading fraction ofV2V communications is much lower than it of V2I communications due to thechallenge in finding V2V routing path.

Several papers introduced the design of routing protocols on finding V2Vrouting path. These studies use topology-based approach, position-basedapproach and opportunistic forwarding approach to design forwarding algo-rithms. These protocols are mainly using the decentralized way to find theV2V path. However, these protocols are not suitable for applying to the vehic-ular network offloading. The reason is that pure V2V communications usingthese routing protocols can not guarantee the delivery of packets due to themobility of vehicles especially in the scenario of low vehicle density. Moreover,the decentralized routing protocol is hard to switch vehicle communicationsbetween cellular network and vehicular network.

For example, topology-based routing protocols use network topology infor-mation for forwarding packets. DSDV(destination-sequenced distance vector)is a proactive routing protocol [Perkins and Bhagwat, 1994] which uses rout-ing tables to maintain routing information of nodes, regardless of whetherdata is sent between these two nodes. DSDV achieves link state and routinginformation between nodes through broadcast and use Bellman-ford algo-rithm to avoid a loop. As the topology of vehicular network changes rapidly,the frequent sending of control packets will waste a large amount of networkbandwidth. Therefore, DSDV is not suitable for vehicular networks. Ad-hocOn-Demand Distance Vector Routing (AODV) is a pure on-demand route ac-quisition system [Toor et al., 2008]. By AODV, nodes do not participate in anyperiodic routing information exchanges or maintain any routing table. A nodeonly discovers and maintains a route to another node until two nodes needto communicate. AODV performs routing discovery by broadcasting routerequest packets(RREQ). After receiving the request message, the destinationnode sends the route reply message(RREP) to the source node. However,AODV wastes a lot of network resources in rebroadcasting RREQ on interme-diate nodes. Moreover, AODV sets the same timeout for each routing table, itdoes not accurately reflect topology changes in the network. There are alsosome problems in the practical application of location-based routing proto-cols. GPSR(greedy perimeter stateless routing) is a routing protocol based onpositions of nodes [Karp and Kung, 2000], which can be obtained throughpositioning devices such as GPS on vehicles. Nodes exchange position infor-mation through broadcasting, and select next hop to forward packets based onpositions of neighbor nodes. GPSR includes two types of packet forwardingmethods: greedy forwarding and perimeter forwarding. In greedy forwardingmode, the protocol chooses the nearest neighbour node from the destination

Page 16: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

4 Introduction

node as the next hop for forwarding. Since the normal forwarding mode canonly obtain the locally optimal next hop, there may be no neighbour nodecloser to the destination node around the current node. Then GPSR switchesto perimeter forwarding mode to find routes. Perimeter forwarding modeuses Gabriel diagram or Relative Neighbour diagram to generate polygonplan, then traverses neighbour nodes on the polygon with the right handrule in order to find the forwarding node. The main problem with GPSR isthat it can get stuck in local maximums and overhead of repair mode is toolarge. GPCR(greedy perimeter coordinator routing), GSR (geographic sourcerouting) are improvements to the GPSR, but rely on additional map informa-tion. Such position-based protocols suffer from inaccuracy of acquired mapinformation and delay of location information.

For autonomous vehicular networks in MEC, the traffic of cellular networkis more of a concern. As density of autonomous vehicles on the road increase,large amounts of collected data need to be processed by the MEC server,which could cause congestion in cellular network. Some studies divide com-puting tasks need to be offloaded into delay-sensitive tasks and delay-toleranttasks [Samanta and Chang, 2019]. In autonomous vehicular networks, we canconsider V2V communications in both delay-sensitive communications anddelay-tolerant communications. It is important to provide low-latency servicesby MEC for computation of delay-sensitive tasks to enable safety applicationssuch as object detection and environment perception. Therefore, it is neces-sary to offload other MEC-irrelevant communications to V2V communicationsusing IEEE 802.11p based vehicular network.

By comparing similarities and differences between autonomous vehicularnetworks and conventional vehicle networks, we believe that these routingprotocols designed for conventional vehicle networks are not applicable to au-tonomous networks. In general, the differnece between autonomous vehicularnetworks and conventional vehicular networks are the following:

• The movement of autonomous vehicles is depend on rules rather thanjudgements of driver which are for conventional vehicles.

• The movement route of the autonomous vehicles is planned and deter-mined. Conventional vehicles have no information about their destina-tion or route, which is known only to the driver.

• The information that may be useful for autonomous vehicular networks,such as velocity, direction, planned route and so on, is the instruction ofvehicle movement. This information requires additional computation inconventional vehicle networks and is based on predictions that are notaccurate.

Page 17: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

§1.2 Contributions 5

Therefore, it is necessary to design routing protocol which could better sup-port communications of autonomous vehicles through autonomous vehicularnetworks in MEC. In this project, we focus on the discovery and maintenanceof V2V path, our proposed routing protocol should be able to:

• find the V2V path between vehicle Vs and vehicle Vd which are not inthe radio range of each other if a path actually exists in MEC.

• find the best V2V path if more than one path exists between vehicle Vsand vehicle Vd.

• know the path found between Vs and Vd is not available due to themobility of vehicles.

1.2 Contributions

In this report, we first design routing protocols which are suitable for au-tonomous vehicular networks. We then simulate the proposed routing proto-col in autonomous vehicular networks by Java language. Finally, we evaluatethe performance of proposed routing protocol compare to the existing routingprotocol in simulation model.

1.3 Report Outline

The rest of this report is arranged as follows: In Chapter 2, we summarizerelated work in vehicular network offloading. In Chapter 3, we introducethe system model, notations used in the report. In Chapter 4, the proposedrouting protocols are presented. In Chapter 5, we present and discuss theexperimental results. In Chapter 6, a conclusion of what we achieved in thisproject and future work are presented.

Page 18: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

6 Introduction

Page 19: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Chapter 2

Related Work

The surge of vehicle infotainment and traffic assistance applications in ITShave promoted the extensive research on the design of vehicular contentoffloading schemes to efficiently provide mobile data offloading from thecellular network to the vehicular network.

Bazzi et al. [Bazzi et al., 2015] discussed how to use IEEE 802.11p devicesequipped vehicles and road side units (RSU) to achieve the cellular resourcesaving in urban scenarios. They evaluated the impact of the density of OBU-equipped vehicles, the amount of deployed RSU on the road, and the adaptedrouting protocol. The results demonstrated that a remarkable reduction of thecellular network resource occupation can be obtained through the deploymentof few RSU and the use of simply routing protocols in an integrated simulationenvironment which consider both realistic vehicular movement model and thecommunication of wireless network. The results show that it can even achieve100% offloading in a scenario of high vehicle density. However, it requiresthe more deployment of RSUs when under the conditions that the networkdemand is high.

Zhioua et al. [el Mouna Zhioua et al., 2014] proposed a vehicular offload-ing model named Vehicular Network Offloading Potential Prediction(VOPP).The conducted an analytical approach on the VOPP. The goal is to select themaximum amount of data streams that could be offloaded to downloader viathe V2I path in the Vehicular ad hoc Network (VANET) to investigate the po-tential of offloading cellular traffic in intelligent transportation systems(ITS).The downloader can request content through I2V link directly from the roadside unit(RSU) or a V2V link to some vehicle which can achieve I2V linkfrom RSU or from cellular network. VOPP considers link availability, gatewaycapacity, media contention, interference, data volume, and channel aggrega-tion between the downloader and the infrastructure to evaluate the capacity ofVANET for supporting additional traffics. The experimental results show thatoffloading fraction at 1 or 2 hops away from infrastructure can achieve similarfraction to the offloading at vehicle which is directly in the coverage of the

7

Page 20: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

8 Related Work

infrastructure. But the traffic offloading fraction is decreased sharply at 3 ormore hops from the infrastructure.In addition, they found a higher offloadingfraction in asymmetric typologies than in symmetric typologies. They alsofound that the distribution of traffic demand had no effect on the unloadingrate of the VOPP formulation.

Huang et al. [Huang et al., 2019] proposed a vehicle to infrastructure (V2I)offloading method through vehicles which are multi-hop away from IEEE802.11p based RSU in mobile edge computing (MEC). To make it possible, letthe vehicle X regularly communicate to its peer vehicle without the help ofinfrastructure by using the cellular network, i.e., the Internet. V2I offloadingcan be available if a multi-hop V2V path exists for connection between Xand the ahead RSU. In addition, if X is out of the radio coverage of thecorresponding RSU, V2I offloading can continue when there exists a multi-hop path for connection between the rear RSU and vehicle X. They found thatthe multi-hop V2I VANET offloading scheme could achieve 1.29 to 4.07 timesof the amount of data traffic offloading volume than the vehicle could onlyhave data offloading within the RSU signal coverage for those vehicles withina single MEC server’s domain. Meanwhile, data loss increased by about 0.7percent.

Zhioua et al. [Zhioua et al., 2014] evaluated VANET’s potential for theVANET networks capacity to offload a partial of the cellular traffic by con-sidering the quality of service(QoS) level of data flow in the offloading de-cision. Various constraints are related to the availability of vehicle to infras-tructure(V2I) link, contention of channel and media, capacity and quality ofvehicle to vehicle(V2V) link, capacity of data flow, and duration of the linkbetween vehicle and road side units(RSU) are considered in making offloadingdecisions. They found the amount of traffic data and the density of VANETnetwork have a great influence on the data offloading fraction. In addition,best effort and background traffic have a higher priority in offloading deci-sions than video traffic. They also found that the offloading fraction of I2Vlinks was higher than that of multi-hop V2V links.

Huang et al. [Huang et al., 2018] proposed a scheme of controller in Soft-ware Defined Network (SDN) based MEC for mobile data traffic offloadingfrom cellular networks to V2V paths of VANET. They adopted the MEC serverbased on the SDN architecture to address the complicated issues exists in V2Voffloading. Thus, it used a centralized scheme in management for calculatingV2V path and notifying nodes. They also proposed a repair method whichtry to recovery the broken links between vehicles. The performance of experi-mental results shows it performance better in both the cellular network andthe vehicular network in medium vehicle density.

Zhang et al. [Zhang et al., 2017] proposed a cloud based framework of

Page 21: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

9

mobile edge computing offloading in vehicular network, which is mainlyused to complete computing tasks between remote cloud and local on-boardterminal. The MEC server receives a task input message from the vehicle andthen calculate expected time of processing time to complete the offloadingtask based on consumption of the mobility of the vehicles and the executionof computation tasks. Therefore, the proposed cloud-based MEC frameworkhelps to decrease the delay time of offloading and the cost of offloadingtransmission.

Hou et al. [Hou et al., 2016] proposed a vehicular fog computing (VFC)architecture. In this VFC architecture, a collaboration of near end user edgedevices and vehicles are utilized as a infrastructure to conduct communica-tion and computation. The quality of services for applications can be greatlyimproved by aggregating abundant computing resources from individual ve-hicles.

Wang et al. [Wang et al., 2017] evaluated the travelling bus’s potential forproviding content offloading service. They maximized the offloading contentvolume from bus to sedans by proposing a bus-based content offloading(BBCO) algorithm. They consider the fairness of offloading opportunity foreach sedan. The service of content offloading for sedans scheduled slot by slotin BBCO algorithm to achieve maximum amount volume of the offloadingcontent and guarantee of the fairness of offloading opportunity by predictingestimating rate of the transmission and the number of buses that the sedanwill encounter on the path in future.

Wu et al. [Wu et al., 2019] proposed a context-awared approach to integratelicensed and unlicensed spectrum for V2V communications . They also pro-posed a decentralized scheme to joint caching, aggregation on different levelof edges to combine data caching, route aggregation and decentralized com-puting efficiently to make a compensation for the limited resources in wirelesssensor. Moreover, they introduced the asynchronous multi-hop broadcast andasynchronous multi-hop unicast to improve the performance of routing inmulti-hop unicast and multi-hop broadcast communications, respectively.

The most study conduct data offloading by V2I communications or limitedhops from relay vehicles to RSU. The performance of V2V data offloading isvery poor due to the difficulty in finding V2V path in a decentralized vehicularnetwork. Therefore, we propose a V2V path discovery strategy in a centralizedvehicular network with the help of MEC.

Page 22: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

10 Related Work

Page 23: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Chapter 3

Preliminaries

In this chapter, the system model and problem definition are introduced.

3.1 System model

The MEC-based autonomous vehicular network (AVNET) is a hybrid networkarchitecture with two components: a base station and a AVNET network. Thetopology of MEC-based autonomous vehicular network has a one-directionroad with intersections Ii, a fixed base station and a MEC server. The MECserver is deployed at the BS as depicted in Figure 3.1. Each autonomousvehicle is equipped with On Board Units (OBU) which can provide vehicle-to-vehicle(V2V) communications. Each autonomous vehicle also has bothcellular network interface and IEEE 802.11p network interface. Moreover, theradio range of each vehicle is same. We further assume there is no road siteunit (RSU) deployed besides the road. Thus, the autonomous vehicles canget no help in transmission and caching from RSUs and also V2I communica-tions are unavailable in our system. In our system, each autonomous vehiclecould communicate with a peered vehicle: 1) through cellular network byusing the cellular network interface 2) through V2V links by using the IEEE802.11p network interface with some routing strategy for V2V path discoveryapplied. Moreover, each autonomous vehicle can communicate with MECserver through BSs at anytime. For any autonomous vehicle Vi, it knows itsown position (donated as PVi), velocity (donated as VelVi), plan (donated asPlVi = Ij). The plan of Vi is represented as the autonomous vehicle Vi willleave this one-direction road at the intersection Ij which is extracted fromthe planned route of the autonomous vehicle. Each autonomous vehicle ismoved as planned route and velocity until an instruction make these attributeschanged.

We assume that each autonomous vehicle Vi can communicate with MECserver all the time. Vi can send its context information such as Pti

Vi, Velti

Vi,

PltiVi

and information of neighbor vehicles it sensed at time point ti to MEC

11

Page 24: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

12 Preliminaries

Figure 3.1: An MEC-based AVNET model

server. Therefore, for MEC server, the system model can be represented asan undirected graph Gt = {Nt, Et}. However, for each Vi ∈ Nt, Vi sends itscontext information to MEC server at different time point t. Therefore, theMEC server knows the AVNET as Gti = {Nti , Eti} at time point ti and P

tjVi

,

VeltjVi

, PltjVi

for each Vi ∈ Nti with tj ∈ {tj | j ≤ i}. The vehicle is a node inthis abstract system model. In this network system, vehicles and nodes areequivalent. We might use word "node" to represent the vehicle in the report.

3.2 Problem definition

Given the autonomous vehicular network Gti = {Nti , Eti} at time ti and contextinformation of Vi for each Vi ∈ Nti including but not limited to P

tjVi

, VeltjVi

, PltjVi

at some time point tj with tj ∈ {tj | j ≤ i}. The first problem is to find a pathwhich actually exists from Vs to Vd with the context information of all nodesin AVNET at different time point t for different V. The second problem is toknow the path found between Vs and Vd is not available due to the mobilityof vehicles.

Page 25: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Chapter 4

Protocols design for AVNET

In this chapter, we introduce the execution procedure of AVNET data offload-ing scheme and propose routing protocols.

4.1 AVNET offloading

From the MEC-based AVNET model we introduced in 3.1, each autonomousvehicle is equipped with cellular network interface and IEEE 802.11p networkinterface. Thus, vehicle Vs can communicate with a peered vehicle Vd throughcellular network or through a V2V path Vs ↔ V0 ↔ V1 ↔ ... ↔ Vd if thereexists. To offload mobile data from cellular network to AVNET as more aspossible, Vs first check whether the peered vehicle Vd is in its V2V routingtable. If Vd is not in Vs ’s routing table, then Vs ask MEC server to discoverya V2V path between Vs and Vd. If MEC server either cannot find a V2V path,Vs communicate with Vd through cellular network. As depicted in Figure 4.1,the detailed procedure is:

1. Vehicle Vs want to initialize a communication with vehicle Vd. Vs firstchecks its AVNET routing table.

2. If Vd is in Vs’s routing table and the route is not expired, there exists aV2V path Vs ↔ ...↔ Vd. Vs can communicate with Vd through this V2Vpath until it is expired.

3. If Vd is not in Vs’s routing table or the V2V path Vs ↔ ...↔ Vd is alreadyexpired, Vs sends a path discovery request (PDREQ) to MEC server forasking to discovery a V2V path from Vs to Vd

4. If MEC server find a V2V path Vs ↔ ...↔ Vd that is available from Vs toVd, MEC server sends path discovery reply (PDREP) to Vs which informsVs to communicate with Vd through the V2V path it found.

13

Page 26: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

14 Protocols design for AVNET

Figure 4.1: The execution procedure

5. If MEC server cannot find an available V2V path from Vs to Vd, then itnotify Vs to communicate with Vd through cellular network.

In our proposed V2V AVNET data offloading architecture, the discoveryof V2V path discovery, the switch between the AVNET network and the cellu-lar network for peered vehicles communication are determined by the MECserver. Thus, the routing strategy for V2V path discovery under our pro-posed V2V AVNET data offloading architecture uses a centralized way tocompute the V2V path between source node Vs and destination node Vd. Itis difficult to apply decentralized routing protocols to build V2V path in ourproposed V2V AVNET data offloading architecture because the decision ofswitch between the AVNET network and the cellular network for communi-cation is made by each node itself. Thus, each node needs to set a timeoutfor undelivered packets through the AVNET network then switches to cellu-lar network for retransmitting undelivered packets. It significantly increasesthe time delay and causes network resource to be wasted on retransmission.In this report, Expired-time-based Routing (ETBR) and Expired-time UpdateProtocol (ETUP) are proposed in next two sections to discover V2V path andmaintain the routing information as change in topology of AVNET due to themobility of autonomous vehicles.

Page 27: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

§4.2 Expired-time-based Routing 15

4.2 Expired-time-based Routing

In our system model, the MEC server receives the context information fromvehicles at different time points which can make the calculated path do notexist. For example, the MEC server stores the context information includePt1

Vi, Velt1

Vi, Plt1

Viof node Vs at time point t1 and Pt2

Vj, Velt2

Vj, Plt2

Vjof node Vj at

time point t2. Assume that the current time is t3. Thus, any path calculatedbased on this information may not available because Vi and Vj are no longerat the position they communicated with the MEC server. In order to eliminatethe influence of data the MEC server received at different time points onthe calculated results, we introduce a way of using expired-time of links tocompute V2V paths. By using expired-time of links between neighbor nodes tocalculate V2V path, the MEC server do not need to know the current positionof each node which can eliminate the influence of expired information aboutPV of each node which can lead to a error in path calculation.

To calculate the expired-time between links of each neighbor nodes, eachautonomous vehicle Vi broadcasts a beacon contains information about <addrVi ,PVi , VelVi , PlVi> periodically. To make sure each vehicle has the newest PlVof neighbour vehicles, whenever an autonomous vehicle changes its plan, itbroadcasts a beacon. Thus, by broadcasting beacons, each autonomous vehiclecan know the context information includes <addrVj , PVj , VelVj , PlVj> of eachneighbour vehicle Vj it sensed. Whenever an autonomous vehicle receives abeacon from its neighbour, it computes the expired-time Et(Vi, Vj) of the linkbetween itself and the neighbour node the beacon comes from based on infor-mation about PlVi , VelVi , PVi in the beacon it received. Each vehicle maintains atable neighbour_information NIVi{addrVj 7→ Et(Vi, Vj)} for each neighbour Vj itsensed. NIVi contains the information about expired-time of links between Viand each neighbour Vj it sensed. Each vehicle also sends its NI periodically toMEC server through BSs. the MEC server stores NI of each vehicle it receivedin its database. Whenever a vehicle Vs sends PDREQ to the MEC server forrequesting a V2V path discovery, the server can calculate the V2V path for Vsbased on expired-time in NI it received from all vehicles.

In this system model, each vehicle only can move in one direction. Thelifetime of the link between two neighbor nodes Vi and Vj has two cases asdepicted in Figure 4.2. The first case is Vi and Vj are always in each others’radio range until one of them leaves the road at some intersection as PlV . Thesecond case is that Vi is out of the radio range of Vj before any one of themleaves the road.

For the case 1, the expired-time of the link between Vi and Vj is depend onthe time which of them will leave the road. That is, the first time one of them

Page 28: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

16 Protocols design for AVNET

(a) Case 1 (b) Case 2

Figure 4.2: Two cases of link lifetime between neighbor vehicles

leaves the road. We have:

Et(Vi, Vj) = Min(PlVi

VelVi

,PlVj

VelVj

) (4.1)

For the case 2, the expired-time of the link between Vi and Vj is dependon the time they are out of each other’s radio range as depicted in Figure4.3. If VelVi faster than VelVj , Vi will move ahead of Vj eventually. Thus, theexpired-time of the link between Vi and Vj is:

Et(Vi, Vj) =Robu − PVi + PVj

VelVi −VelVj

(4.2)

Otherwise, Vj will move ahead of Vi eventually. The expired-time of the linkbetween Vi and Vj is:

Et(Vi, Vj) =Robu + PVi − PVj

VelVj −VelVi

(4.3)

Thus, the expired-time of the link between Vi and Vj is depend on relativevelocity between Vi and Vj, if VelVi > VelVj , we have:

Et(Vi, Vj) = Min(PlVi

VelVi

,PlVj

VelVj

,Robu − PVi + PVj

VelVi −VelVj

) (4.4)

if VelVi < VelVj , we have:

Et(Vi, Vj) = Min(PlVi

VelVi

,PlVj

VelVj

,Robu + PVi − PVj

VelVj −VelVi

) (4.5)

Page 29: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

§4.2 Expired-time-based Routing 17

Figure 4.3: Positions of neighbor vehicles when they are out of radio range

Whenever an autonomous vehicle receives a beacon from the neighborvehicle, it calculates the expired-time of the link between itself and the nodethe beacon comes from. The autonomous vehicle then updates its neighborinformation NI and sends a context information report (CIR) message tothe MEC server. Thus, the MEC server can know the context informationof all vehicles in the AVNET. The MEC server can construct a undirectedgraph G = {V, E} to represent the topology of AVNET. If vehicle Vs wants tocommunicate with vehicle Vd, it first checks whether Vd in its routing table.If Vd does not in Vs’s routing table or the route is expired, Vs then sendsPDREQ to the MEC server for asking V2V path discovery between Vs andVd. When the MEC server receives PDREQ from a vehicle Vs, it is responsiblefor calculating a V2V path between Vs and Vd. Assume that there is a V2Vpath between Vs and Vd donate as Path(Vs, Vd) which consists of a sequence ofautonomous vehicles Vs, V0, V1, ..., Vn, Vd. The expired-time of the link betweenneighbor vehicles is Et(Vs, V0), Et(V0, V1), ..., Et(Vn, Vd) respectively. Thus, thepath expired-time of the link between Vs and Vd, is depend on the expired-time of the link between neighbor nodes through Vs, V0, V1, ..., Vn, Vd, which iscomputed as follow:

PathEt(Vs, Vd) = Min{Et(Vs, V0), Et(V0, V1), ..., Et(Vn, Vd)} (4.6)

Usually, there are several paths between two vehicles. To find the optimalpath between two vehicles, we use the following proposed algorithm. TheMEC server use following proposed algorithm to compute the optimal V2Vpath between Vs and its destination node Vd based on the undirected graphG = {N, E} it constructed. The pseudo code is as follow:

An example of running the algorithm is depicted in Figure 4.4. At theinitial stage, an undirected graph based on all vehicles in the AVNET is con-structed. The graph has 8 vertices which are {Vs, V1, V2, V3, V4, V5, V6, Vd}. Thevalue of each edge is the expired-time of the link between neighbor nodes.The value of expired-time is a monotonically increasing number. The D(Vs)

Page 30: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

18 Protocols design for AVNET

Algorithm 1 V2V path discovery (Vs, Vd)

1: D(Vs) := ∞2: path = {}3: Q = {}4: for each node Vi in AVNET do5: if Vi 6= Vs then6: D(Vi) := −17: end if8: prev(Vi) = null9: Q = Q ∪ {Vi}

10: end for11: while Q is not empty do12: find Vm in Q s.t. D(Vm) is maximum among all D(Vm)13: if Vm = Vd then14: while prev(Vm) 6= null do15: add Vm to path16: Vm = prev(Vm)17: end while18: return path19: end if20: for each neighbour node Vn of Vm do21: alt := min(D(Vm), Et(Vm, Vn))22: if alt > D(Vn) then23: D(Vn) = alt24: prev(Vn) = Vm25: end if26: end for27: end while28: return path

Page 31: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

§4.2 Expired-time-based Routing 19

(a) Initial stage (b) Iteration1

(c) Iteration2 (d) Iteration3

(e) Iteration4 (f) Iteration5

(g) Iteration6

Figure 4.4: An example of running the algorithm

Page 32: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

20 Protocols design for AVNET

is set to ∞ which represents the link between Vs to itself is never expired.And D(V) of other nodes are set to -1 which represent that the node is un-reachable at this stage. In the 1st iteration, as depicted in Figure 4.4(b), thealgorithm chooses the node Vs which has the biggest D(V) = ∞ and updatesthe D(V) of each neighbor node of Vs. D(V1) is set to 11, D(V1) is set to 7and D(V3) is set to 9 in this iteration. As depicted in Figure 4.4(c), in the 2nditeration, the algorithm chooses the node V1 which has the biggest D(V) = 11and updates the D(V) of each neighbor node of V1. D(V4) is equal to theminimum value between D(V) of the node V1 which the edge comes fromand Et(V1, V4) which is the value of the edge. Thus, D(V4) is updated to4 in this iteration. In the 3rd iteration, the algorithm chooses the node V3which has the biggest D(V) = 9 and updates the D(V) of each neighbor nodeof V3. D(V6) is set to 9 in this iteration. In the 4th iteration, the algorithmchooses the node V6 which has the biggest D(V) = 9 and updates the D(V)of each neighbor node of V6. D(V5) is set to 9 in this iteration. As for the5th iteration, as depicted in Figure 4.4(f), the algorithm chooses the node V5which has the biggest D(V) = 9 and updates the D(V) of each neighbor nodeof V5. D(Vd) is set to 9 and D(V4) is updated to 8 which means a better pathVs ↔ V3 ↔ V6 ↔ V5 ↔ V4 will be expired later than original found pathVs ↔ V1 ↔ V4 in this iteration. In the 6th iteration, as depicted in Figure4.4(g), the destination node Vd is chosen. Thus, the algorithm end up andreturns the found V2V path Vs ↔ V3 ↔ V6 ↔ V5 ↔ Vd.

After path discovery process, if the MEC server finds a V2V path ex-ist between Vs and Vd, it sends a PDREP which contains the path it foundback to Vs. Then Vs tries to build a link with Vd by sending a route re-quest (RREQ) to Vd. The process of sending RREQ has three purpose: oneis to confirm the path found by the MEC server is actually exist. The sec-ond one is to exchange expired_time of each one hop link along the path.The last one is to store the reverse route information in each node’s routingtable. In ETBR, the path_expired_time of a V2V path is determined by theearliest expired_time of one hop link along the path. A RREQ should containsthe following information <src_address, dest_address, path, path_expired_time>.As the forwarding of RREQ, nodes along the path can check expired_timeof the link between itself and next hop. If the expired_time is earlier thanpath_expired_time in RREQ, it updates path_expired_time in RREQ as its ex-pired_time of next hop and updates earliest_expired to True which representsthe path_expired_time is determined by the link between these two nodes. Asdepicted in Figure 4.5, Vs initiate a RREQ to Vd, this RREQ contains <addrVs ,addrVd , Path(Vs, Vd) = {Vs, V1, V2, V3, V4, Vd}, PathEt(Vs, Vs) = ∞>. SinceVs is the source node, the RREQ has not been transmitted along the path,PathEt(Vs, Vd) is equal to the expired_time of the link between Vs and next

Page 33: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

§4.2 Expired-time-based Routing 21

Figure 4.5: Path setup

node Vs, that is Et(Vs, Vs). When V1 receives the RREQ from Vs, V1 checksPathEt(Vs, Vd) in the RREQ. This PathEt(Vs, Vd) represents the expired_timeof the path from Vs to V1. By comparing PathEt(Vs, Vd) and Et(Vs, V1), ifEt(Vs, V1) is earlier than PathEt(Vs, Vd), it means the path_expired_time is de-termined by the expired_time of the link between Vs and Vs. Thus, V1 storesthe routing information from itself to the source node Vs in its routing tablewith PathEt(Vs, Vd) in the RREQ and a boolean value earliest_expired as Truewhich represents that the link between Vs and V1 will be expired earliest alongthe path from Vs to V1. Then V1 updates the information in RREQ with newPathEt(Vs, Vd) = 5. For V2, Et(V1, V2) is not earlier than PathEt(Vs, Vd) in theRREQ. Therefore, V2 does no change of information in RREQ. As RREQ trav-elling from the source node to the destination node, each node along the pathcan have the routing information from itself to source node.

When Vd receives the RREQ, it replies a RREP to Vs. RREP packets shouldcontain the following information <src_address, dest_address, path_expired_time>.Different to RREQ, RREP does not contain the information of path. Becauseeach node along the path already has the routing information through thetravelling of RREQ. Similar to the process of forward path setup, PathEt(Vd, Vs)

Page 34: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

22 Protocols design for AVNET

also is changed in the process of reverse path setup. Recall that in the forwardpath setup phrase, V2 can only know the path_expired_time of the path betweenVs and V2, it does not know the path_expired_time of the path between V2 andVd. As depicted in Figure 4.5, V2 receives the RREP from V3. Then V2 checksPathEt(Vs, Vd) in the RREP and store the routing information from V2 to Vdwith the path_expired_time. Moreover, if Et(V2, V1) is earlier than PathEt(Vd, Vs),V2 updates the path_expired_time in RREP by PathEt(Vd, Vs) = 6. After doneboth forward path setup and reverse path setup, each node Vi along the pathcan have the routing information about path to both Vs and Vd stored in itsrouting table.

A node maintains a routing table entry for each interested destination. Arouting table entry contains the following information:

• dest_address

• next_hop_address

• path_expired_time

• earliest_expired

path_expired_time is used to identify whether the path from Vs to Vd is notavailable any more. The destination of a transmission is in the routing tableof Vs with path_expired_time earlier than current time means the destination isnot achievable through this path. This routing table entry should be removedfrom Vs. Another PDREQ should be sent to the MEC server for finding a newpath from Vs to Vd. earliest_expired is important for updating path_expired_timewith the protocol we will introduce in Section 4.3. earliest_expired meanswhether the expired_time of a path is determined by the expired_time of one-hop link between this node and next hop node. The path_expired_time shouldbe updated if the expired_time of this link become later due to the planchangeof the vehicle. It also should be updated if the expired_time of any one-hoplink become earlier than current one.

4.3 Expired-time update protocol

In Expired-time-based routing, the state of the link is determined by PathEt. IfPathEt is later than current time, the path is on longer available. expired_timeis computed based on PlV of two vehicles. However, vehicles could changetheir plan due to the change of destination or the emergency happened on theroad. Therefore, a protocol that can dynamically change the expired_time ofrouting table entries is needed.

Page 35: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

§4.3 Expired-time update protocol 23

Table 4.1: Table of notations used in the report

Notations Descriptions

Ii the intersection i

Vi the vehicle i (also is called node i)

PVi the position of vehicle Vi

VelVi the velocity of vehicle Vi

PlanVi the vehicle Vi will leave the road at which Ii

addrVi the address of Vi (could be IP, MAC address)

Et(Vi, Vj) the expired-time of the link from Vi to Vj

NIVi the neighbor information of the vehicle Vi

Path(Vs, Vd) the V2V path from Vs to Vd

PathEt(Vs, Vd) the expired-time of V2V path from Vs to Vd

Expired-time update protocol is a supplement to the Expired-time-basedrouting. It cannot be used without Expired-time-based routing. To update thePathEt of the path could be affected by change of plan in some vehicles, once avehicle Vi change its PlVi , it should broadcast a beacon which contains <addrVi ,PVi , VelVi , PlanVi> to notify its neighbour nodes. Each neighbour node thatreceives the beacon can compute and update expired_time of the link betweenitself and the source node where the beacon comes from in its NI. That means,The expired_time of any path that has Vi as relay node may have changed. Todynamically change the PathEt, each vehicle Vj that receives the beacon shouldrun the algorithm to determine whether the PathEt of routing information inits routing table should be updated due to the change of Et(Vi, Vj). If thePathEt of the routing entry should be updated, Vj sends a path expired-timeupdate message(PETU) to the destination node.

As depicted in Figure 4.6, assume that V3 receives a beacon from V2 dueto its change in plan. three cases need to be consider when we want to updatethe PathEt according to change in Et(V2, V3).

• Case1: Et(V2, V3) change from 5 to 4. The earliest expiry link along thepath still is the link between V1 and V2 and PathEt(V1, V6) do not change.Thus, V3 do not need to notify other nodes along the path.

• Case2: Et(V2, V3) change from 2 to 5. The earliest expiry link usedto be the link between V2 and V3. The lifetime increases in link be-tween V2 and V3 cause lifetime increases in path from V1 and V6, whichmeans PathEt(V1, V6) becomes later. Thus V3 should sends PETU to V6to notify other nodes along the path that change in Et(V2, V3) causes

Page 36: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

24 Protocols design for AVNET

(a) Case 1 (b) Case 2 (c) Case 3

Figure 4.6: Three cases of link lifetime between neighbor vehicles

PathEt(Vs, Vd) become later and the new PathEt(V1, V6) is 5 currently.Moreover, with the forwarding of PETU, if any relay node has a ear-lier expiry link than PathEt(V1, V6), it should updates PathEt(V1, V6) andmake earliest_expired True in its routing table.

• Case3: Et(V2, V3) change from 5 to 2. The earliest expiry link along thepath changed from the link between V1 and V2 to the link between V2and V3. Thus, V3 should sends PETU to V6 to notify other nodes alongthe path that PathEt(V1, V6) should be updated according the change ofEt(V2, V3).

4.4 Summary

In this chapter, we first introduced the execution procedure of AVNET data of-floading through V2V communications. Then, we proposed a routing strategynamed Expired-time based routing (ETBR) which is a centralized approachwith the help of MEC. We also presented the pseudo code of path discov-ery algorithm. Finally, we proposed a routing protocol named Expired-timeupdate protocol (ETUP) which can transmit and update the information ofexpired-time of along the path path when expired-time of the path changed.

Page 37: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Chapter 5

Evaluation

In this chapter, numerical results of the evaluation on the performance of theproposed routing protocols for V2V AVNET offloading are presented.

5.1 Experiment setting

The V2V AVNET offloading scenario is simulated in Java language. The simu-lation is conducted on a 5km long road with 10 intersections. We investigateimpacts of autonomous vehicle density, frequency of reporting neighbor in-formation to the MEC server. In our evaluation, the performance analysis isbased on the delivery ratio of packets, offloading fraction and average lifetime ofV2V path. The measures of the performance are follows:

• The rate of packet delivery in the autonomous vehicular network.

• The fraction of data offloading from cellular network to vehicular net-work.

• The average lifetime of V2V path

In this project, proposed routing protocols are compared with AODVwhich construct the V2V path with minimum hop count for evaluation. Thetimeout of route expiration in AODV is set to 3000 milliseconds. The factorscould have a impact on the performance of routing protocols we considerin the report are vehicle density and interval of reporting neighbor informa-tion to the MEC server. We will evaluate the impact of these factors on theperformance of routing protocols.

5.2 Results

Figure 5.1 depicts the delivery ratio of packets for the scenario of differentdensity of vehicles. We find the delivery ratio of packets is always higher

25

Page 38: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

26 Evaluation

(a) low density of vehicle (b) medium density of vehicle

(c) high density of vehicle (d) average delivery ratio

Figure 5.1: Delivery ratio of packets

when using ETBR as the routing strategy than when using AODV. Whenusing ETBR as a routing strategy, the packet delivery ratio is even close to100%. However, we acknowledge the limitations of this experiment. We donot consider the contention of link layer and receive power. Moreover, we donot consider the overflow in receive buffer that would result in a packet dropwhen the volume of sending data is too huge. Therefore, the packet deliveryratio in reality should be lower than the experimental results. We also findthe average packet delivery ratio is almost same in the scenario of differentvehicle density when using ETBR. Because ETBR can unable the expired linkto avoid transmitting packets on a path that is unavailable. The average packetdelivery ratio is higher with the increase in the density of vehicles when usingAODV. The average delivery ratio is around 70% in the scenario of low vehicledensity and increase to around 80% in the scenario. The velocity decreasesas the density of the vehicle increases resulting in the decrease in the relativemoving speed between vehicles. Thus, connections between vehicles becomeless easily broken which make the packet delivery ratio higher.

Page 39: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

§5.2 Results 27

Figure 5.2: Average offloading fraction for different reporting interval

We evaluate the effect of different reporting neighbor information frequen-cies on offloading fraction as depicted in Figure 5.2. The experiment fordifferent reporting interval is conducted in the scenario of medium vehicledensity. We find the offloading fraction increase with decrease in reportinginterval. This is in line with our expectations. The MEC server can have moreaccurate information to calculate the V2V path when the report interval isshorter. Therefore, when the reporting interval is shorter, the probability offinding the path is greater which represents a V2V offloading can be achieved.

We also evaluate the offloading fraction of V2V offloading in AVNET usingdifferent routing protocols. The experiments are conducted in the scenario ofdifferent vehicle density as depicted in Figure 5.3. We find the V2V AVNEToffloading fraction using ETBR is a little bit higher than using AODV. Offload-ing fraction is depend on the topology of the network, this shows that bothprotocols can correctly discover possible paths. However, routing table ofAODV stores many paths that don’t actually exist, i.e., expired paths, result-ing in low packet delivery rates. The advantage of ETBR over AODV is thatETBR avoids using expired paths which increases packet delivery rate. Theoffloading fraction bounce more vigorously in the scenario of medium vehicledensity because there are more alternative V2V paths than in the scenario oflow density of vehicles. Thus, the offloading fraction change more frequentlydue the broken links and newly established links. As for the offloading frac-tion in the scenario which has a high density of vehicle, the average velocityis decreased so that the links between vehicles are not broken as frequentas in the scenario of medium vehicle density. Figure 5.3(d) shows the aver-age offloading fraction in the scenario of different vehicle density using tworouting protocols. As expected, the V2V AVNET offloading fraction increaseswith the increase in vehicle density. The reason is more V2V paths exist whenvehicle density increase. It also shows the difference of offloading fractionusing two protocols increase. Because ETBR has more choices to choose the

Page 40: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

28 Evaluation

(a) low density of vehicle (b) medium density of vehicle

(c) high density of vehicle (d) average offloading fraction

Figure 5.3: Offloading fraction

best V2V path with the increase in vehicle density. AODV always tries to finda path with minimum hop-counts which suffer more frequent link brokenthan ETBR.

As depicted in Figure 5.4, the average lifetime of links established by ETBRis longer than links established by AODV in all scenario of different vehicledensity. The average lifetime of links in the scenario of different vehicle densityis almost same for using AODV. However, the average lifetime of links forusing ETBR increases significantly when vehicle density is high. The reasonis, there more alternative V2V paths from the source node to the destinationnode. Thus, ETBR can choose the path will be expired latest which make theaverage lifetime of the V2V path found by ETBR much longer than the pathfound by AODV.

Page 41: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

§5.3 Summary 29

Figure 5.4: Average lifetime

5.3 Summary

In this chapter, we provided the numerical results of evaluation on the per-formance of the proposed routing protocol for V2V AVNET offloading. Theresults shows our proposed routing protocol named ETBR achieve a muchhigher packet delivery ratio than AODV which build V2V path with minimumhop count. We also evaluated impact of different interval of reporting neigh-bor and vehicle density on the performance. Higher V2V AVNET offloadingfraction can be achieved in higher vehicle density with shorter reporting inter-val.

Page 42: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

30 Evaluation

Page 43: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Chapter 6

Conclusion and Future Work

6.1 Conclusion

In this report, we proposed two routing protocols named ETBR and ETUPto find V2V path and dynamically update link information to achieve dataoffloading from the cellular network to the vehicular network. ETBR usesa centralized approach to find V2V paths with the help of the MEC server.In ETBR, each autonomous vehicle broadcasts a beacon which contains itscontext information include position, velocity and plan. Each autonomousvehicle can calculate and store the expired-time of the link between itself andthe neighbor vehicle which the beacon comes from based on the informationin the beacon it received. Each autonomous vehicle also reports the expired-time of the between itself and neighbor vehicle to the MEC server. When thevehicle wants to communicate with other vehicle through cellular network, itfirst checks the destination whether exists in routing table and the route is notexpired. If not, the vehicle sends a control message RREQ to the MEC serverto request a V2V path discovery. Then the MEC server calculates the V2Vpath between source vehicle and destination vehicle based on the informationabout link expired-time from vehicles. ETUP can dynamically update linkinformation caused by plan change of the autonomous vehicle. Whenever anautonomous changes its plan, it broadcasts a beacon to notify other vehicles.Each vehicle executes ETUP when receiving beacon from neighbor vehicle tocheck whether expired-time of any link is changed due to the change of theplan. We conducted the experiments to evaluate the impact of vehicle densityand reporting interval on delivery ratio, offloading fraction and link lifetime.Experimental results show that the proposed routing protocols can achievearound 97% packet delivery ratio which is much higher than the existingrouting protocol. We found that the average lifetime and offloading fractionincrease with increase in the density of vehicles. Moreover, the offloadingfraction increases as the reporting interval is shortened.

31

Page 44: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

32 Conclusion and Future Work

6.2 Future Work

As the project only focused on the discovery of V2V paths, some factors thatmay affect link quality are not considered. The MEC server can know thecomputing and storage resource of each node in the network by periodicallyreceiving information from all nodes. One future research direction is that thediscovery of the V2V path takes more factors into account. The performanceof the node is depend various factors like gateway capacity, media contention,interference, data volume, and channel aggregation. Another direction offuture research is to consider the impact of vehicle behavior patterns such asplatoon and convoy on data transmission. Platoon and convoy in vehicles en-able us to dynamically select the header vehicle in platooning and convoyingas the edge server to achieve aggregation and caching. The discovery andselection of the header vehicle also can be achieved with the help of the MECserver.

Page 45: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Bibliography

Bazzi, A.; Masini, B. M.; Zanella, A.; and Pasolini, G., 2015. Ieee 802.11pfor cellular offloading in vehicular sensor networks. Computer Communica-tions, 60 (2015), 97–108. doi:10.1016/j.comcom.2015.01.012. (cited on page7)

Chaturvedi, M. and Srivastava, S., 2017. Multi-modal design of an intel-ligent transportation system. IEEE Transactions on Intelligent TransportationSystems, 18, 8 (2017), 2017–2027. doi:10.1109/tits.2016.2631221. (cited onpage 1)

Dai, H.; Zeng, X.; Yu, Z.; and Wang, T., 2019. A scheduling algorithm forautonomous driving tasks on mobile edge computing servers. Journal ofSystems Architecture, 94 (2019), 14–23. doi:10.1016/j.sysarc.2019.02.004. (citedon page 1)

Do-Duy, T. and Kim, D.-S., 2015. Interference-aware relay assignment schemefor multi-hop wireless networks. Wireless Networks, 21, 7 (2015), 2195–2207.doi:10.1007/s11276-015-0904-5. (cited on page 2)

el Mouna Zhioua, G.; Zhang, J.; Labiod, H.; Tabbane, N.; and Tabbane,S., 2014. Vopp: A vanet offloading potential prediction model. 2014 IEEEWireless Communications and Networking Conference (WCNC), (2014). doi:10.1109/wcnc.2014.6952726. (cited on page 7)

Fagnant, D. J. and Kockelman, K., 2015. Preparing a nation for autonomousvehicles: opportunities, barriers and policy recommendations. TransportationResearch Part A: Policy and Practice, 77 (2015), 167–181. doi:10.1016/j.tra.2015.04.003. (cited on page 1)

Feng, J.; Liu, Z.; Wu, C.; and Ji, Y., 2017. Ave: Autonomous vehicular edgecomputing framework with aco-based scheduling. IEEE Transactions onVehicular Technology, 66, 12 (2017), 10660–10675. doi:10.1109/tvt.2017.2714704.(cited on page 1)

Hou, X.; Li, Y.; Chen, M.; Wu, D.; Jin, D.; and Chen, S., 2016. Vehicular fogcomputing: A viewpoint of vehicles as the infrastructures. IEEE Transactions

33

Page 46: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

34 Bibliography

on Vehicular Technology, 65, 6 (2016), 3860–3873. doi:10.1109/tvt.2016.2532863.(cited on page 9)

Huang, C.-M.; Chiang, M.-S.; Dao, D.-T.; Su, W.-L.; Xu, S.; and Zhou, H.,2018. V2v data offloading for cellular network based on the software definednetwork (sdn) inside mobile edge computing (mec) architecture. IEEE Access,6 (2018), 17741–17755. doi:10.1109/access.2018.2820679. (cited on page 8)

Huang, C.-M.; Wu, Z.-Y.; and Lin, S.-Y., 2019. The mobile edge computing(mec)-based vanet data offloading using the staying-time-oriented k-hopaway offloading agent. 2019 International Conference on Information Network-ing (ICOIN), (2019). doi:10.1109/icoin.2019.8718188. (cited on page 8)

Karp, B. and Kung, H. T., 2000. Gpsr: greedy perimeter stateless routing forwireless networks. IEEE Computer Society, , 2000 (2000), 243–254. (cited onpage 3)

Li, L.; Ota, K.; and Dong, M., 2018a. Humanlike driving: Empirical decision-making system for autonomous vehicles. IEEE Transactions on VehicularTechnology, 67, 8 (2018), 6814–6823. doi:10.1109/tvt.2018.2822762. (cited onpage 1)

Li, N.; Martinez-Ortega, J.-F.; Diaz, V. H.; and Fernandez, J. A. S., 2018b.Probability prediction-based reliable and efficient opportunistic routing al-gorithm for vanets. IEEE/ACM Transactions on Networking, 26, 4 (2018),1933–1947. doi:10.1109/tnet.2018.2852220. (cited on page 2)

Nguyen-Minh, H.; Benslimane, A.; and Deng, D.-J., 2016. Reliable broadcast-ing using polling scheme based receiver for safety applications in vehicularnetworks. Vehicular Communications, 4 (2016), 1–14. doi:10.1016/j.vehcom.2016.01.003. (cited on page 2)

Papadoulis, A.; Quddus, M.; and Imprialou, M., 2019. Evaluating the safetyimpact of connected and autonomous vehicles on motorways. AccidentAnalysis Prevention, 124 (2019), 12–22. doi:10.1016/j.aap.2018.12.019. (citedon page 1)

Peng, H.; Li, D.; Abboud, K.; Zhou, H.; Zhao, H.; Zhuang, W.; and Shen,X., 2017. Performance analysis of ieee 802.11p dcf for multiplatooningcommunications with autonomous vehicles. IEEE Transactions on VehicularTechnology, 66, 3 (2017), 2485–2498. doi:10.1109/tvt.2016.2571696. (cited onpage 2)

Peng, H.; Ye, Q.; and Shen, X., 2019. Spectrum management for multi-accessedge computing in autonomous vehicular networks. IEEE Transactions on

Page 47: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Bibliography 35

Intelligent Transportation Systems, (2019), 1–12. doi:10.1109/tits.2019.2922656.(cited on page 1)

Perkins, C. E. and Bhagwat, P., 1994. Highly dynamic destination-sequenceddistance-vector routing (dsdv) for mobile computers. ACM SIGCOMMComputer Communication Review, 24, 4 (1994), 234–244. doi:10.1145/190809.190336. (cited on page 3)

Pettigrew, S.; Talati, Z.; and Norman, R., 2018. The health benefitsof autonomous vehicles: public awareness and receptivity in australia.Australian and New Zealand Journal of Public Health, 42, 5 (2018), 480–483.doi:10.1111/1753-6405.12805. (cited on page 1)

Samanta, A. and Chang, Z., 2019. Adaptive service offloading for revenuemaximization in mobile edge computing with delay-constraint. IEEE Inter-net of Things Journal, 6, 2 (2019), 3864–3872. doi:10.1109/jiot.2019.2892398.(cited on page 4)

Su, Z.; Hui, Y.; and Luan, T. H., 2018. Distributed task allocation to enablecollaborative autonomous driving with network softwarization. IEEE Journalon Selected Areas in Communications, 36, 10 (2018), 2175–2189. doi:10.1109/jsac.2018.2869948. (cited on page 1)

Toor, Y.; Muhlethaler, P.; Laouiti, A.; and La Fortelle, A., 2008. Vehiclead hoc networks: applications and related technical issues. IEEE Communi-cations Surveys Tutorials, 10, 3 (2008), 74–88. doi:10.1109/comst.2008.4625806.(cited on page 3)

Wang, Z.; Zhong, Z.; Ni, M.; Hu, M.; and Chang, C.-Y., 2017. Bus-basedcontent offloading for vehicular networks. Journal of Communications andNetworks, 19, 3 (2017), 250–258. doi:10.1109/jcn.2017.000042. (cited on page9)

Wu, C.; Chen, X.; Yoshinaga, T.; Ji, Y.; and Zhang, Y., 2019. Integratinglicensed and unlicensed spectrum in the internet of vehicles with mobileedge computing. IEEE Network, 33, 4 (2019), 48–53. doi:10.1109/mnet.2019.1800453. (cited on pages 2 and 9)

Zhang, K.; Mao, Y.; Leng, S.; He, Y.; and ZHANG, Y., 2017. Mobile-edgecomputing for vehicular networks: A promising network paradigm withpredictive off-loading. IEEE Vehicular Technology Magazine, 12, 2 (2017), 36–44. doi:10.1109/mvt.2017.2668838. (cited on page 8)

Zhioua, G. e. m.; Labiod, H.; Tabbane, N.; and Tabbane, S., 2014. A trafficqos aware approach for cellular infrastructure offloading using vanets. 2014

Page 48: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

36 Bibliography

IEEE 22nd International Symposium of Quality of Service (IWQoS), (2014). doi:10.1109/iwqos.2014.6914328. (cited on page 8)

Page 49: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Appendix 1

Conducting computing tasks and data caching near end users under MECcan dramatically decrease response latencies. MEC offloads time-critical anddata-intensive computational tasks of autonomous vehicles. Unlike normaluser equipment, autonomous vehicles may send requests and get responsesfrom server in different locations through different Access Points. We focuson the design/analysis of routing protocols for this scenario under the MEC.The main tasks should include:

• Literature review.

• Design/analysis of routing protocols in MEC.

• Simulation of the scenario with designed protocols.

37

Page 50: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

38 Bibliography

Page 51: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Appendix 2

39

Page 52: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:
Page 53: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:
Page 54: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

42 Bibliography

Page 55: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Appendix 3

Except where otherwise indicated, all codes are my own original work

The simulation experiments are conducted in Java language. Node.javasimulates the function of nodes and the network MECserver class simulatesthe functions of the edge server. Also, the algorithm of V2V path discoveryis implemented in this class. simulation.java contains arguments setting forexperiments. run_experiments.py is a script to run experiments, get resultsand produce figures for demonstration.

43

Page 56: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

44 Bibliography

Page 57: Routing Protocols Design for Autonomous Vehicles in Mobile … · 2019. 10. 30. · Autonomous Vehicles in Mobile Edge Computing Shangru Liu A report submitted for the course COMP8755:

Appendix 4

# COMP8755 Individual project

## Originqlity declaration Except where otherwise indicated, all codes aremy own original work

## modules Node.java simulates the function of nodes and the networkMECserver class simulates the functions of the edge server. Also, the algo-rithm of V2V path discovery is implemented in this class. simulation.javacontains arguments setting for experiments. run_experiments.py is a script torun experiments, get results and produce figures for demonstration.

## How to Run the CodeThe simulation is run with the command:python3 run_experiments -d Density -i Intervalwhere:-d Density specifies the vehicle density (default: medium) [options: low,

medium, high]-i Interval specifies the interval of reporting neighbor information to the

MEC server (default: medium) [options: short, medium, long]for example, the command:python3 run_experiments -d high -i shortwill simulate the senario of high vehicle density and short interval of

reporting neighbor information to the MEC server for vehicles.

45