Optimisation des surfaces intelligentes reconfigurables : méthodes basées sur des modèles et sur l'apprentissage automatique

Recently, reconfigurable intelligent surfaces (RIS) have garnered significant attention from both industry and academia. An RIS is a planar surface that consists of a large number of low-cost passive reflecting elements. By carefully adjusting the phase shifts of the reflecting elements, an RIS can reshape the wireless environment to improve communication performance. This thesis focuses on three key objectives: (i) Developing efficient modeling and optimization techniques for RIS-aided communication systems. (ii) Designing novel multi- beam optimization methods for RIS in both single-input single-output (SISO) and multiple- input multiple-output (MIMO) systems, incorporating specified mask reradiation constraints. (iii) Proposing an advanced optimization frame- work for RIS-aided multiantenna systems using the multiport network theory model. Chapter 1 introduces the evolution of wireless and mobile communication systems, high-lighting key application scenarios for next- generation communication systems. It also analyzes potential techniques to address these scenarios, with a particular focus on RIS. The chapter further explains the working principle of RIS and its potential application in 6G networks. Chapter 2 provides an overview of various channel models for RIS-aided systems, including the geometric channel model, Kronecker correlation channel model, Rician fading channel model, and EM-compliant channel model. It discusses the optimization challenges in RIS- aided systems and reviews general optimization techniques for RIS phase shift design, encompassing both traditional optimization methods and AI-driven approaches.Chapter 3 explores the design of an RIS- aided system and proposes a beamforming design for RIS optimization, which accounts for specified reradiation masks at the design stage. Novel beamforming methods for RIS optimization are proposed, leveraging semidefinite programming and model-aided neural networks to achieve improved efficiency and reduced computational complexity. Chapter 4 investigates three dual-beam design schemes for RIS-aided MIMO wireless communications. A Bayesian general linear model and alternating optimization techniques are utilized to address the complexities of dual-beam design. A model-based neural network is then introduced to enhance computational efficiency and execution speed. For practical real-world applications, the unit-module method is proposed to effectively handle discrete mask constraints. Simulation results demonstrate the effectiveness and robustness of the proposed schemes, highlighting their potential for improving RIS-aided communication systems. Chapter 5 introduces an advanced optimization framework for RIS-aided multi-antenna systems using the multiport network theory model. By optimizing the precoding matrix and RIS reflection coefficients through WMMSE and alternating optimization algorithms, significant performance improvements were achieved. The study highlights the impact of structural scattering on reradiation patterns, emphasizing the importance of accounting for these effects in practical RIS applications. Finally, Chapter 6 concludes the thesis and discusses possible future research directions.