IRS-Assisted Proactive Eavesdropping Over Fading Channels Based on Deep Reinforcement Learning

In this letter, we study an intelligent reflecting surface (IRS)-assisted proactive eavesdropping system, where a legitimate monitor (LM) eavesdrops a point-to-point suspicious wireless communication over Rayleigh fading channel with the assistance of IRS. In order to improve the long-term eavesdropping performance of the system, the reflecting ability of IRS is fully exploited, where the IRS's reflecting optimization problem is established. As the proposed problem is non-convex and difficult to solve, a double deep Q-Learning network (DDQN)-based algorithm is proposed to achieve the optimal reflecting beamforming policy. To this end, the optimization problem is transformed into a Markov Decision Process (MDP) and a reward function which can reflect the eavesdropping performance is designed for agent learning. The simulation results show that the proposed DDQN-based approach achieves the average improvement of 11.27% compared with classical deep Q-network (DQN) algorithm, and with the assistance of IRS, the eavesdropping rate of LM increases by 22.61% and 34.92% compared to proactive eavesdropping via spoofing relay (PESR) and proactive eavesdropping via jamming (PEJ) respectively.