Electronic Thesis and Dissertation Repository

Thesis Format

Integrated Article

Degree

Doctor of Philosophy

Program

Electrical and Computer Engineering

Supervisor

Dr. Serguei Primak

2nd Supervisor

Dr. Xianbin Wang

Co-Supervisor

Abstract

Vehicle ad hoc networks (VANETs) have become a popular topic in modern research. The main advantages of these networks include: improved security, traffic optimization, and infotainment. However, deploying such networks in practice requires extensive infrastructure. To estimate the network load, one needs to have information about the network, such as the number of clusters, cluster size, etc. Since VANETs are formed by vehicles that rapidly change their location, the network topology is constantly changing, making its analysis by deterministic methods impossible. Therefore, in this dissertation, we use probability theory methods to obtain probability distributions of such fundamental network properties, such as the number of clusters, cluster size, and the number of disconnected vehicles in the case in which the vehicles are located on a highway. In previous articles, some of these characteristics are obtained only in terms of average values, while the total distributions remained unknown. The distribution of the largest cluster size is an important characteristic of the network. It is derived in the dissertation for the first time. We also study the distribution of the number of clusters and the size of the average cluster in the case of a 2D map with an almost arbitrary road topology. To the best of our knowledge, these results are the first for such a general map case.

Studying these properties raises a number of new questions about how these network properties change over time. We obtain distributions of the network characteristics, such as the duration of communication between vehicles, and the duration of cluster existence. We also derive the probability that a cluster exists between two time moments, as well as other network properties. The obtained distributions are new in the case of the Markov channel model. The results regarding the distribution of cluster lifetime and the probability of cluster existence between two fixed time moments are obtained in the literature for the first time.

This dissertation also addresses the security aspect of VANET. We consider single and multichannel anti-jamming games in the case in which two communicating vehicles are being pursued by the jammer, which tries to disrupt the communication. The optimal strategies of the vehicles and the jammer are described as the Nash equilibrium of this game. We prove theorems that express Nash equilibrium through communication parameters. The considered model with quadratic power term is new as well as the results regarding the Nash equilibrium in the single and multichannel cases. We also first examine performance of such state-of-the-art machine learning algorithms as Dueling Q-learning and Double Q-learning, which by trial and error, successfully converge to the Nash equilibrium, deduced theoretically.

Summary for Lay Audience

The subject of this dissertation is vehicle networks. The wide deployment of the vehicle networks will happen in this decade, so this topic attracts increased scientific interest. It is assumed that the cars will share information about traffic accidents, weather conditions, their coordinates, speed, acceleration, etc. The vehicle networks will also provide Internet access, video and audio streaming, and other services. All this data will be used for traffic optimization, driving safety improvement, and passenger entertainment. However, the vehicle networks are unstable due to the high car speeds, so the vehicles are organized into small groups that share information. Such groups are called clusters and are the object of study in the first part of the dissertation. Statistical data related to such network characteristic as cluster size helps to estimate the network load. Network load prediction is important for installing antennas along the roads that support vehicle networks. We use the probability theory methods to study such characteristics as the number of clusters, cluster size, and the number of disconnected vehicles in the case in which the vehicles are located on a highway and 2D map.

The second part of the dissertation is devoted to the security aspect of the vehicle networks. We consider a game in which a jammer chases two vehicles in order to disrupt their communication. We consider both single-channel and multi-channel cases. In the multi-channel case, it is assumed that the vehicles change the communication channel in order to avoid channel attacks. We derive theorems that express the optimal vehicle and jammer strategies through the communication channel parameters. However, in practice, some of the communication parameters are unknown, which narrows the scope of applicability of the obtained theorems. Therefore, we also examine machine learning algorithms that by trial and error converge to the theoretically obtained optimal strategies.

Share

COinS