An Attention-Based Denoising Neural Network for mmWave RIS-Assisted SIMO Channel Estimation

Reconfigurable Intelligent Surface (RIS) is promising for future wireless communication. This letter introduces an attention-based denoising neural network for RIS-aided single-input multiple-output (SIMO) channel estimation (CE). The proposed method leverages the feature extraction ability of the encoder and the signal reconstruction ability of the decoder to achieve a more robust and accurate CE. During the offline training phase, an encoder-decoder based neural network is trained. The encoder extracts features related to the channel state and predicts the channel matrix, and the decoder restores the noise-free received signals. In the online estimation phase, only the well-trained encoder is involved. The multi-view attention mechanism is integrated for greater feature learning ability, further improving the CE performance. Simulation results show that the proposed method significantly improves CE accuracy and reduces online estimation complexity.