Joint Optimization of Access and Backhaul Links for UAVs Based on Reinforcement Learning

In this paper, we study the application of unmanned aerial vehicle (UAV) base stations (BSs) in order to improve the cellular network capacity. We consider flying BSs where BS equipments are mounted on UAVs, making it possible to move BSs freely in space. We study the optimization of UAVs' trajectory in a network with mobile users to improve the system throughput. We consider practical two-hop communications, i.e., the access link between a user and the UAV BS, and the backhaul link between the UAV BS and a macrocell BS plugged into the core network. We propose a reinforcement learning based algorithm to control the UAVs' mobility. Additionally, the proposed algorithm is subject to physical constraints of UAV mobility. Simulation results show that considering both the backhaul and access links in the UAV mobility optimization is highly effective in improving the system performance than only focusing on the access link.