Abstract
Reconfigurable intelligent surfaces (RIS) have emerged as a promising technology for enhancing spectral and energy efficiency in next-generation wireless networks. However, the joint optimization of base station (BS) precoding and RIS reflection coefficients in multi-user multiple-input single-output (MU-MISO) systems is a highly non-convex problem with prohibitive computational complexity when addressed using conventional optimization techniques. To tackle this challenge, we propose a data-driven reinforcement learning framework based on the Soft Actor-Critic (SAC) algorithm for adaptive RIS-assisted MU-MISO optimization. The proposed approach exploits stochastic policy learning with entropy regularization, twin Q-networks, and automatic temperature adjustment to enhance exploration capability and training stability. To rigorously benchmark performance, we conduct a comprehensive comparative analysis against two widely adopted deterministic reinforcement learning baselines, namely Deep Deterministic Policy Gradient (DDPG) and Twin Delayed Deep Deterministic Policy Gradient (TD3), under varying transmit power levels, RIS sizes, and learning hyperparameters. Simulation results demonstrate that SAC consistently achieves superior long-term sum-rate performance, particularly in low-to-moderate signal-to-noise ratio (SNR) regimes and for larger RIS dimensions, while exhibiting greater robustness to hyperparameter tuning. In contrast, DDPG and TD3 often converge faster in early training stages but are more sensitive to parameter settings and prone to suboptimal asymptotic solutions. These results establish SAC as a robust and scalable solution for RIS-assisted MU-MISO systems, offering improved reliability and adaptability for future intelligent wireless networks.
Funding Information
  • National Science and Technology Council, Taiwan (NSTC 114-2221-E-194-034)