Learning-based detection for reflection modulation in reconfigurable intelligent surface assisted systems

Abstract

Reconfigurable intelligent surfaces have emerged as a promising technology for enhancing wireless communication by intelligently shaping the propagation environment through tunable passive elements. Beyond traditional passive beamforming, reflection modulation enables RIS to convey information by dynamically altering its reflection patterns. This thesis investigates the use of quadrature reflection modulation, a group-based RM scheme that allows information embedding through selective phase alignment across RIS element clusters, without compromising aperture gain. A single user downlink communication system is considered, where the direct link between the access point and the user is blocked, and the RIS establishes an indirect link. To support information decoding at the user, a neural network-based receiver is designed to jointly estimate the transmitted symbol and the RIS reflection pattern. The model is trained using simulated channel realizations and symbol-pattern combinations to learn the nonlinear mapping from received signal observations to discrete modulation classes. This approach provides a practical alternative to traditional maximum likelihood detection for RIS-assisted systems employing QRM.