Movement and Communication Co-Design in Multi-UAV Enabled Wireless Systems via DRL

This paper studies downlink communications in multi-unmanned aerial vehicle (UAV) enabled wireless communication systems with co-channel interference, where multiple UAVs are employed as aerial base stations (BSs) to serve mobile users on the ground. Our goal is to maximize the average sum-rate throughput of all users over a finite time horizon via UAV dynamic movement and communication (including user association and power control) co-design. Different from conventional offline designs based on the ideal assumption that channel state information (CSI) is perfectly known ahead of the flight, here we consider more challenging scenarios that the UAV can only obtain the real-time CSI along its flight. To tackle the online movement and communication co-design problem, we propose a deep reinforcement learning (DRL) algorithm based on deep deterministic policy gradient (DDPG), which can handle the problem of our interest with high-dimensional action space. Numerical results show that our proposed algorithm outperforms other benchmark schemes and baseline algorithms significantly.