Integrating channel knowledge map and deep reinforcement learning for optimizing RIS-assisted MU-MISO systems

A method that combines technical channel knowledge mapping (CKM) with deep reinforcement learning (DRL) to jointly optimize the phase offsets for transmission beamforming and reconfigurable intelligent surfaces (RIS) is developed in multi-user multiple-input single-output (MU-MISO) systems. The objective of the scheme is to enhance the overall downlink capacity under phase-sensitive reflection amplitude modeling conditions. The initial phase of the research involves pre-training using CKM to construct a model capable of accounting for positioning errors. Subsequently, the model will be migrated to a real scenario and formally trained based on the channel information that are obtained from channel estimation with channel estimation error. The proposed approach can efficiently exploit the environmental information and thus improve the performance and robustness of wireless communication systems. The results show that our approach has the potential to address the challenges of channel knowledge acquisition in hardware-constrained RIS-assisted wireless communication systems.