Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.770305
Title: Learning and reasoning strategies for user association in ultra-dense small cell vehicular networks
Author: Kapoor, S.
Awarding Body: University of York
Current Institution: University of York
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
Recent vehicular ad hoc networks research has been focusing on providing intelligent transportation services by employing information and communication technologies on road transport. It has been understood that advanced demands such as reliable connectivity, high user throughput, and ultra-low latency required by these services cannot be met using traditional communication technologies. Consequently, this thesis reports on the application of artificial intelligence to user association as a technology enabler in ultra-dense small cell vehicular networks. In particular, the work focuses on mitigating mobility-related concerns and networking issues at different mobility levels by employing diverse heuristic as well as reinforcement learning (RL) methods. Firstly, driven by rapid fluctuations in the network topology and the radio environment, a conventional, three-step sequence user association policy is designed to highlight and explore the impact of vehicle speed and different performance indicators on network quality of service (QoS) and user experience. Secondly, inspired by control-theoretic models and dynamic programming, a real-time controlled feedback user association approach is proposed. The algorithm adapts to the changing vehicular environment by employing derived network performance information as a heuristic, resulting in improved network performance. Thirdly, a sequence of novel RL based user association algorithms are developed that employ variable learning rate, variable rewards function and adaptation of the control feedback framework to improve the initial and steady-state learning performance. Furthermore, to accelerate the learning process and enhance the adaptability and robustness of the developed RL algorithms, heuristically accelerated RL and case-based transfer learning methods are employed. A comprehensive, two-tier, event-based, system level simulator which is an integration of a dynamic vehicular network, a highway, and an ultra-dense small cell network is developed. The model has enabled the analysis of user mobility effects on the network performance across different mobility levels as well as served as a firm foundation for the evaluation of the empirical properties of the investigated approaches.
Supervisor: Grace, D. ; Clarke, T. Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.770305  DOI: Not available
Share: