Multi-Agent Deep Reinforcement Learning for Secure UAV Communications

Abstract

In this paper, we investigate a multi-unmanned aerial vehicle (UAV) cooperation mechanism for secure communications, where the UAV transmitter moves around to serve the multiple ground users (GUs) while the UAV jammers send the 3D jamming signals to the ground eavesdroppers (GEs) to protect the UAV transmitter from being wiretapped. The 3D jamming guarantees the GEs not being interfered by the jamming signals. It is challenging to make a joint trajectory design and power control for a UAV team without central control. To this end, we propose a multi-agent deep reinforcement learning approach to achieve the maximum sum secure rate by designing the dynamic trajectory of each UAV. The proposed multi-agent deep deterministic policy gradient (MADDPG) technique is centralized training at high altitude platforms (HAPs) and distributed execution at each UAV, which enables the fully distributed cooperation among UAVs. Finally, the simulation results show the proposed method can efficiently solve the multi-UAV cooperation trajectory design problem in secure communication scenarios.