Deep Denoising Network Based Channel Estimation for RIS-Assisted ISAC Systems

Integrated sensing and communication (ISAC) technology enables simultaneous data transmission and environmental sensing by sharing spectral and hardware resources, significantly improving spectrum efficiency and overall system performance. Reconfigurable intelligent surface (RIS) is regarded as a transformative technology for future wireless communication networks. In ISAC systems, RIS can dynamically adjust the phase and amplitude of wireless signals to optimize channel conditions and enhance sensing accuracy. However, complex channel environments and noise interference pose significant challenges to achieving accurate channel estimation. To address the channel estimation problem in RIS-assisted ISAC systems, this paper proposes a multi-scale denoising network (MSDNet), comprising a pre-estimation network and a denoising network. The pre-estimation network performs initial estimation through preliminary feature extraction to obtain coarse estimation results, while the denoising network extracts features using multi-scale convolutional paths with varying dilation rates and employ a channel attention mechanism to weight critical features. Using normalized mean square error (NMSE) as the evaluation metric, simulations are conducted under various signal-to-noise ratio (SNR) conditions. In addition, the model performance is evaluated by varying key parameters such as the number of base station (BS) antennas and RIS elements. Simulation results demonstrate that MSDNet consistently outperforms conventional methods and existing denoising networks across a wide range of SNR scenarios, exhibiting strong estimation accuracy and robustness.