Showing 1 results for Reinforcement Learning (rl).
Mousa Abdollahvand, Sima Sobhi-Givi,
Volume 21, Issue 1 (3-2025)
Abstract
This paper introduces a new method for improving wireless communication systems by employing beyond diagonal reconfigurable intelligent surfaces (BD-RIS) and unmanned aerial vehicle (UAV) alongside deep reinforcement learning (DRL) techniques. BD-RIS represents a departure from traditional RIS designs, providing advanced capabilities for manipulating electromagnetic waves to optimize the performance of communication. We propose a DRL-based framework for optimizing the UAV and configuration of BD-RIS elements, including hybrid beamforming, phase shift adjustments, and transmit power coefficients for non-orthogonal multiple access (NOMA) transmission by considering max-min fairness. Through extensive simulations and performance evaluations, we demonstrate that BD-RIS outperforms conventional RIS architectures. Additionally, we analyze the convergence speed and performance trade-offs of different DRL algorithms, emphasizing the importance of selecting the appropriate algorithm and hyper-parameters for specific applications. Our findings underscore the transformative potential of BD-RIS and DRL in enhancing wireless communication systems, laying the groundwork for next-generation network optimization and deployment.
