Twin-Timescale Beamforming for IRS-Assisted Millimeter Wave Massive MIMO-OFDM Systems

We investigate a twin-timescale joint beamforming problem for multiple intelligent reflecting surfaces (IRSs)-assisted multi-user mm Wave orthogonal frequency division multiplexing (OFDM) systems, where the base station (BS) employs a hybrid analog and digital precoder. To alleviate the burden of frequent channel state information (CSI) acquisition and reduce design complexity, we devise the passive beamforming vector and the analog precoder based on statistical CSI, while the digital precoder is designed based on low-dimensional instantaneous CSI. Specifically, the former long-term optimization can be formulated as a stochastic optimization problem. To address this problem, we propose two different solutions. The first method devises the passive beamforming vector and the analog precoder by maximizing the ergodic channel gain. We also propose a deep unrolling-based method to provide a unified framework for the stochastic optimization problem. Our simulation results demonstrate the effectiveness and computational efficiency of the proposed methods.