Deep Learning-Driven Channel Estimation Technique for Intelligent Reflecting Surface-Assisted Communication Systems

To address the issue of low channel estimation accuracy in intelligent reflecting surface (IRS)-assisted communication systems, a channel estimation scheme based on a Channel Denoising Network (CDN) is proposed, modeling the channel estimation problem as a channel noise elimination problem. First, traditional algorithms are used to perform an initial estimation of the received pilot signals, and then the estimates are fed into the channel estimation network to learn noise information and perform denoising, thereby recovering accurate channel coefficients. To enhance the network's denoising capability, an attention mechanism module and a dilated convolution module are designed to improve the extraction of primary noise information, while a feature fusion module is designed to prevent the loss of shallow features. Simulation results show that, compared to the classic DNCNN and CDRN networks, the proposed method reduces the normalized mean square error (NMSE) by an average of 2 to 2.89 dB across different signal-to-noise ratios.