Energy-aware deep-learning framework for joint denoising and channel state evolution in reconfigurable intelligent surface-enabled mmWave massive MIMO

This research presents a robust learning-based framework for the RIS-assisted MIMO-OTFS communication system by devising a new hybrid Channel Attention CNN (CaCNN) and Channel Spatial CNN (CsCNN) architecture with Lightweight Attention enhanced Transformer (AHCTNet) modeling. This aims at solving those challenging problems, including accurate channel estimation and effective noise suppression in highly dynamic wireless environments such as multipath fading, Doppler spread, and time–frequency selectivity, where conventional RIS-based methods always face serious performance degradation. For handling such impairments, the proposed hybrid deep-learning network processes the received signal in the delay–Doppler domain. Channel attention CNN adaptively emphasizes the informative features in channels, which efficiently suppresses interference and noise. Simultaneously, the long-range dependencies and global variations of the channel are captured by the Lightweight Transformer module; therefore, it is well-suited for tracking fast variations in the mmWave channel. Furthermore, the channel spatial CNN decoder learns spatial correlations that are introduced by RIS reflections and the geometry of the MIMO arrays, thereby enabling an appropriate reconstruction of the received signal. Comprehensive simulation results confirm the exceptional improvement of normalized mean square error (NMSE), achievable rate, and throughput to prove the applicability and efficiency of the proposed scheme in future 6G wireless communication systems.