Deep Reinforcement Learning Based UAV for Securing mmWave Communications

This paper focuses on the unmanned aerial vehicle (UAV) enabled millimeter (mmWave) communications from physical layer security perspective. A UAV is arranged as an aerial base station to provide ubiquitous connectivity for terrestrial users in the presence of multiple eavesdroppers. With statistical channel state information (CSI) of eavesdroppers, the beamforming vector and trajectory of UAV as well as user scheduling are jointly optimized to minimize the weighted sum of UAV flight period and secrecy outage duration. The considered problem is a combinatorial optimization problem with complicated objective function, and thus difficult to be solved by convex optimization-based methods. To this end, we formulate this problem as a Markov decision process (MDP) and develop a deep reinforcement learning (DRL) based method to optimize all variables simultaneously. Simulation results validate superiority of the proposed method over benchmarks and demonstrate its ability to obtain a compromise between secure transmission and energy efficiency.