Deep Reinforcement Learning-Based Trajectory Planning for Secure UAV Communication

Abstract

Preventing illegal eavesdropping and establishing reliable and secure wireless communication is an urgent and challenging task. This paper studies the ground-to-UAV (G2U) communication system with a malicious eavesdropper. The optimization goal of this paper is to maximize the accumulated secrecy rate of the system by optimizing the UAV trajectory and connection sequence, which is a mixed-integer nonlinear programming (MINLP) problem. We formulate this problem as a Markov decision process (MDP) with finite states and actions, and implement the double deep Q network (DDQN) algorithm to solve it. To verify the effectiveness of the scheme, experiments are carried out in a simulation environment. Simulation results and numerical analysis demonstrate that the DDQN-based scheme can optimize the trajectory of the legitimate UAV, has a faster convergence speed and better security than the nature deep Q network (DQN) algorithm, and significantly improves the communication performance of the system.