Optimizing Sum Rate in RIS-Aided Multi-User MISO Systems with Curiosity-Driven Learning

Reconfigurable Intelligent Surfaces (RIS) have emerged as a transformative paradigm for advancing next-generation wireless communications, characterized by their ability to control the electromagnetic environment, energy efficiency, and near-passive operation. In this paper, we formulate an optimizing problem to enhance communication in a RIS-assisted multi-user Multiple-input Single-output (MISO) system. The proposed approach leverages the inherent advantages of the MISO system, particularly its ability to provide spatial diversity and enhanced signal quality, which is crucial in complex multi-user environments. We choose the sum rate as a performance matrix, as it measures the long-term average capacity that a communication system can achieve. Our approach maximizes the sum rate by optimizing the RIS phase shift matrix and base station beamforming. Given the problem's non-convexity, we employ a deep reinforcement learning (DRL) framework augmented with curiosity-based learning. In the proposed algorithm, the agent receives intrinsic rewards based on its exploration by using the Random Network Distillation (RND) method of curiosity-based learning applied with the Deep Deterministic Policy Gradient (DDPG) algorithm. Compared to prior optimization methods, the proposed algorithm is entirely model-free and can accommodate wireless channel fluctuations. The proposed framework achieves a sum rate nearly 28% higher than the DRL algorithm without using curiosity-based learning as demonstrated in the simulation results.