Distributed RISs-Aided Secure Wireless Network Under Practical Phase Shift Model: A Deep Reinforcement Learning Approach

Reconfigurable intelligent surface (RIS) is a disruptive technology for enhancing the capacity and the coverage of the wireless network. In this paper, we focus on a distributed RISs-aided secure communication, where multiple distributed RISs are deployed to assist the communications of multiple source-destination pairs. At the same time, multiple eavesdroppers wiretap the transmitted signals. Additionally, the practical phase shift (PPS) model of RIS is taken into consideration, where the reflection amplitude and the phase shift are coupled, rather than the ideal phase shift (IPS) model adopted in most existing contributions. The sum secrecy rate (SSR) is maximized by jointly designing the power control of the sources as well as the phase shifts of all distributed RISs. To deal with the non-convex and highly intractable problem, an efficient deep reinforcement learning (DRL)-based method is developed. Numerical results demonstrate that the designed DRL-based framework outperforms other benchmarks in terms of SSR, even the performance loss is inevitable under the PPS model. Meanwhile, the effectiveness of the distributed RISs in improving SSR is verified.