Analysis of Reinforcement Learning Schemes for Trajectory Optimization of an Aerial Radio Unit

Abstract: This paper introduces the deployment of unmanned aerial vehicles (UAVs) as lightweight wireless access points that leverage the fixed infrastructure in the context of the emerging open radio access network (O-RAN). More precisely, we propose an aerial radio unit that dynamically serves an under served area and connects to the distributed unit via a wireless fronthaul between the UAV and the closest tower. In this paper we analyze the UAV trajectory in terms of AI when it serves both UEs and central units (CUs) at the same time in multi input multi output (MIMO) fading channel. We first demonstrate the nonconvexity of the problem of maximizing the overall network throughput based on UAV location, and then we use two different machine learning approaches to solve it. We first assume that the environment is a gridworld and then let the UAV explore the environment by flying from point A to point B, using both the offline Q-learning and the online SARSA algorithm and the achieved path-loss as the reward. With the intention of maximizing the average payoff, the trajectory in the second scenario is described as a Markov decision process (MDP). According to simulations, MDP produces better results in a smaller setting and in less time. In contrast, SARSA performs better in larger environments at the expense of a longer flight duration.