Intelligent and bandwidth-efficient medium access control protocols for IEEE 802.11p-based Vehicular Ad hoc Networks

Vehicle-to-Vehicle (V2V) technology aims to enable safer and more sophisticated transportation via the spontaneous formation of Vehicular Ad hoc Networks (VANETs). This type of wireless networks allows the exchange of kinematic and other data among vehicles, for the primary purpose of safer and more efficient driving, as well as efficient traffic management and other third-party services. Their infrastructure-less, unbounded nature allows the formation of dense networks that present a channel sharing issue, which is harder to tackle than in conventional WLANs. This thesis focuses on optimising channel access strategies, which is important for the efficient usage of the available wireless bandwidth and the successful deployment of VANETs. To start with, the default channel access control method for V2V is evaluated hardware via modifying the appropriate wireless interface Linux driver to enable finer on-the-fly control of IEEE 802.11p access control layer parameters. More complex channel sharing scenarios are evaluated via simulations and findings on the behaviour of the access control mechanism are presented. A complete channel sharing efficiency assessment is conducted, including throughput, fairness and latency measurements. A new IEEE 802.11p-compatible Q-Learning-based access control approach that improves upon the studied protocol is presented. The stations feature algorithms that “learn” how to act optimally in VANETs in order to maximise their achieved packet delivery and minimise bandwidth wastage. The feasibility of Q-Learning to be used as the base of selflearning protocols for IEEE 802.11p-based V2V communication access control in dense environments is investigated in terms of parameter tuning, necessary time of exploration, achieving latency requirements, scaling, multi-hop and accommodation of simultaneous applications. Additionally, the novel Collection Contention Estimation (CCE) mechanism for Q-Learning-based access control is presented. By embedding it on the Q-Learning agents, faster convergence, higher throughput, better service separation and short-term fairness are achieved in simulated network deployments. The acquired new insights on the network performance of the proposed algorithms can provide precise guidelines for efficient designs of practical, reliable, fair and ultra-low latency V2V communication systems for dense topologies. These results can potentially have an impact across a range of related areas, including various types of wireless networks and resource allocation for these, network protocol and transceiver design as well as QLearning applicability and considerations for correct use.