Learning-Based STARS-Aided Communication System with Movable Antenna

Given the half-space coverage limitation inherent in the cooperative architecture of reconfigurable intelligent surfaces (RIS) and movable antenna (MA), this paper introduces an innovative joint control scheme that integrates simultaneously transmitting and reflecting surface (STARS) with MA. The proposed scheme seeks to maximize the weighted sum rate of a multi-user communication system through the joint optimization of the MA coordinates, transmit beamforming, and the transmission/reflection coefficient matrix of STARS. In the optimization process, the scheme utilizes the twin delayed deep deterministic policy gradient (TD3) algorithm in conjunction with an enhanced experience replay strategy to efficiently optimize the parameters. Specifically, to meet the specified constraints, the scheme imposes strict limitations on action outputs and introduces a well-designed penalty mechanism to ensure the system operates within the prescribed boundaries. Numerical simulations confirm that the joint design of STARS and MA can significantly increase the overall system rate, thereby substantiating the substantial potential of this integrated scheme to enhance communication system performance.