Coupled Phase-Shift STAR-RIS for Secure MIMO Communication: A DRL-Based Beamforming Design

A simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) aided multiple-input multiple-output (MIMO) secure communication system is explored. A practical coupled phase-shift model of STAR-RIS is taken into account for beamforming design. Based on this model, a joint active and passive beamforming optimization problem is formulate to maximize the long-term sum secrecy rate of transmission and reflection users. An efficient deep reinforcement learning (DRL)-based algorithm is proposed to address the multivariate coupling issue in the joint beamforming design problem. Simulation results show that: 1) All STAR-RIS schemes can achieve better secrecy performance gain than the conventional RIS; 2) Compared to the optimal independent phase shifts, couple phase shifts of STAR-RIS only lead to a slight performance loss.