REINFORCEMENT-LEARNING-BASED CROSS LAYER DESIGN IN MOBILE AD-HOC NETWORKS

Abstract

Mobile Ad-hoc networks (MANETs) are drawing increasing research interest because of the revolutionary development of mobile devices and wireless communication technology in recent years. One important feature of MANETs is that Nodes can move freely without any requirement of infrastructure. Therefore communications are purely based on peer-to-peer packet forwarding, which makes it an ideal option for fast network deployment such as in military or disaster areas. While mobility and lack of central controller are the most important features of MANETs, they are also the greatest challenges: constant changes in network topology and local information sharing make end-to-end Quality of Service (QoS) optimization difficult. In this dissertation, we propose a reinforcement-learning-based solution to address the cross-layer optimization on QoS performances in MANETs.