Intelligent physical layer architectures for 6G
Data
2026-06-19
Autores
Martins, Rafael Marasca
Título da Revista
ISSN da Revista
Título de Volume
Editor
Resumo
Abstract: The physical layer of next-generation wireless networks faces challenges that extend beyond classical theoretical limits and are frequently constrained by practical issues, such as restricted computational budgets and hardware non-idealities. This dissertation explores how incorporating intelligence into the physical layer can overcome these bottlenecks across three scenarios relevant to 6G networks. First, in Reconfigurable Intelligent Surface (RIS)-aided communications, the mismatch between the low update rate of GNSS positioning and the need for rapid surface reconfiguration compromises beam alignment. To address this limitation, a predictive control strategy based on the Kalman filter is proposed, employing constant-velocity and constantacceleration motion models. This predictive approach reduces received power degradation by a factor of up to five compared to a conventional reactive strategy, with the constant-acceleration model proving more robust for trajectories involving curves and abrupt direction changes. The second application scenario occurs in Integrated Sensing and Communication (ISAC) systems, where joint beamforming optimization for sensing and communication leads to non-convex problems whose conventional iterative solutions often fail to converge within realistic computational budgets. To overcome this bottleneck, a deep unfolding architecture is developed that reformulates the Successive Convex Approximation algorithm as a computational graph with trainable parameters. The resulting optimizer increases the feasibility rate under the same iteration count while also providing gains in sensing capabilities. The third intelligenceaided application at the physical layer pertains to optical wireless communications: the use of off-the-shelf optical components in IEEE 802.15.7 Color Shift Keying (CSK) modulation introduces distortions in the received constellation that render the conventional geometric detector inadequate. An end-to-end system model is developed, and three machine-learning-based detectors are evaluated, demonstrating that geometric constellation warping, rather than additive noise, constitutes the dominant performance limiter. The proposed detectors reduce the classification error by approximately 51%. Across all investigated scenarios, the intelligence incorporated into the physical layer—whether through model-based prediction, algorithm unfolding, or datadriven classification—yields substantial and practically realizable gains, highlighting the potential of these techniques for future 6G networks
Descrição
Palavras-chave
Machine learning, 6G, Intelligent reflecting surfaces, Integrated sensing and communication, Visible light communication, Physical layer, Transmission schemes, 6G networks, Beamforming, Reconfigurable intelligent surfaces (RIS)