DISTRIBUTED LEARNING FOR RECONFIGURABLE INTELLIGENT SURFACES

Abstract

Reconfigurable Intelligent Surfaces (RIS) have gained significant attention as a transformative technology to enhance wireless communication by intelligently manipulating the propagation environment. By dynamically adjusting the phase shifts of a large array of passive elements, RIS can actively influence signal propagation to improve link quality, spectral efficiency, and interference mitigation. One of the fundamental challenges in RIS-assisted wireless commu- nication is maximizing the Signal-to-Noise Ratio (SNR) at the receiver. Achieving optimal SNR necessitates determining the best RIS phase shift configuration, a task complicated by the high-dimensional search space, dynamic channel conditions, and practical constraints such as discrete phase shifts and limited feedback from the receiver. Existing solutions, including ex- haustive search and model-based optimization techniques, suffer from significant computational complexity and are often infeasible for real-time adaptation in practical systems. To address these challenges, this thesis proposes novel online-learning-based Multi-Armed Bandit (MAB) algorithms tailored for RIS optimization. The RIS configuration problem is for- mulated as a sequential decision-making process where the system continuously learns the op- timal phase shift arrangement while balancing exploration and exploitation. Unlike traditional reinforcement learning approaches, our MAB-based framework provides a lightweight, adap- tive, and sample-efficient solution, making it particularly suitable for practical deployments where acquiring full channel state information is costly or impractical. We introduce multi- ple algorithmic variants that incorporate advanced exploration-exploitation trade-offs, ensuring rapid convergence to near-optimal configurations without excessive computational overhead. The proposed algorithms are evaluated through extensive simulations under diverse chan- nel conditions and system constraints. The results demonstrate that our approach significantly outperforms traditional heuristic and non-learning-based methods, achieving faster convergence, higher achievable SNR, and improved robustness against environmental variations. Addition- ally, we analyze the computational efficiency of our methods, demonstrating their suitability for i real-time RIS control in dynamic wireless environments. By leveraging learning-based strate- gies, our approach enables RIS to autonomously adapt to changing conditions, unlocking its full potential for next-generation wireless networks. Beyond performance gains, this thesis explores the practical considerations for imple- menting MAB-based RIS optimization in real-world networks. We address key aspects such as feedback mechanisms and computational complexity, ensuring that our proposed methods align with practical hardware capabilities. Furthermore, we extend our analysis to multi-user scenarios, cooperative RIS control, and integration with emerging wireless technologies, such as millimeter-wave and terahertz communications. These findings highlight the potential of in- telligent, adaptive, and scalable RIS-based communication systems that dynamically optimize their behavior based on real-time observations. This work contributes to the growing body of research on RIS-aided wireless networks by introducing efficient learning-based strategies for optimizing RIS configurations. The find- ings presented in this thesis pave the way for future studies exploring distributed learning ap- proaches, joint RIS and beamforming optimization, and energy-efficient RIS deployment strate- gies. Although ISAC and ISAC-RIS are not the core focus of this work, future research could explore the application of MAB frameworks in ISAC-RIS systems, where joint optimization of sensing and communication objectives may benefit from efficient online learning strategies. Our work not only enhances theoretical understanding but also provides a practical foundation for deploying RIS in next-generation wireless networks, including 6G and beyond.