DRL‐Based Secure Communication for Reconfigurable Intelligent Surface‐Assisted MISO Systems

ABSTRACT In recent years, advancements in programmable materials have led to the widespread adoption of reconfigurable intelligent surface (RIS) technology in physical layer security. However, the complex channel environment in communication systems poses significant challenges to optimizing the phase shifts of RIS. Traditional mathematical approaches require multiple approximate optimizations and are computationally intensive. To address these issues, a joint active‐passive beam‐forming scheme using deep reinforcement learning is proposed for RIS‐assisted systems. In the context of a continuous action space, the secrecy rate between the legitimate receiver and the eavesdropper is utilized as the immediate reward for training the parameters of the network. Additionally, the deep deterministic policy gradient (DDPG) framework is employed to enhance the optimization of joint active‐passive beam‐forming, facilitating simulation learning within a continuous action space. Simulation results demonstrate that the proposed method is capable of deriving the maximum secrecy rate through real‐time observation of immediate rewards and continuous interaction with the environment.