Beamforming Design for Reconfigurable Intelligent Surface via Bayesian Optimization

Abstract

This letter investigates the transmission design in the multi-user downlink system using the reconfigurable intelligent surface (RIS). The conventional transmission scheme requires the channel estimation by transmitting pilots and feeds back the channel state information (CSI) to the base station, which causes lots of overhead and substantial radio-frequency chains at the RIS. In this letter, we propose a cost-efficient downlink transmission scheme via Bayesian optimization that dispenses with CSI. The beamformer at the BS and the phase rotation at the RIS are jointly designed by minimizing the sum mean square error (MSE). Since the proposed scheme dispenses with CSI, the channel estimation and the CSI feedback are not required, which saves the number of training slots and dispenses with RF chains at the RIS. Moreover, the proposed scheme can be extended to the minimax MSE problem and the energy harvesting problem. Simulation results show that the proposed transmission scheme outperforms the particle swarm optimization scheme.