Unsupervised Learning-Based Joint Precoding and Phase Shift Design for RIS-Assisted mmWave Communication Systems

This paper investigates a reconfigurable intelligent surface (RIS) aided downlink millimeter-wave multiple-input-multiple-output (MIMO) system. We aim to maximize the system spectral efficiency by jointly design the hybrid precoder at the base station (BS), the RIS phase shifts and the digital combiner at the user equipment (UE). For practical application, we adopt discrete Fourier transform (DFT) codebook-based hybrid precoding, and discrete RIS phase shifts. In order to achieve high performance with small computational delay, we propose a novel two-stage unsupervised learning-based approach to jointly design the codebook-based hybrid precoder and the discrete RIS phase shifts. Corresponding to the two-stage network architecture, we also develop a efficient phased training strategy. Simulation results demonstrate that the proposed approach can achieve satisfactory performance and the computational delay is small.