A Meta-Learning Approach for Energy-Efficient Resource Allocation and Antenna Selection in STAR-BD-RIS Aided Wireless Networks

This paper focuses on a wireless network that utilizes beyond diagonal reconfigurable intelligent surfaces (BD-RIS). In this network, multiple BD-RISs assist a multi-antenna base station (BS) with two sectors that simultaneously transmit and reflect signals to single-antenna users. The goal is to maximize energy efficiency by jointly optimizing beamforming at the BS, the BD-RISs’ matrix, and antenna selection under the maximum power budget at the BS, BD-RISs’ matrix, and antenna selection constraints. The formulated problem is non-convex and challenging to be solved optimally. To address this difficulty, we propose a meta-soft actor critic (Meta-SAC) algorithm, which enables the BS to adjust its beamforming capabilities and BD-RISs’ matrix and assign antennas to users. Simulation results demonstrate the superiority of Meta-SAC in comparison with other meta algorithms and a reasonable response compared to the convex optimization benchmark. We also study the influence of system model parameters on the objective function of the proposed optimization problem. In addition, the results show that the multi-BD-RIS system reaches a higher energy efficiency and data rate compared to the provided benchmarks.