Abstract: Abstract: A channel estimation scheme based on Channel Denoising Network (CDN) is proposed to address the issue of low channel estimation accuracy in reconfigurable intelligence surface assisted communication systems. The channel estimation problem is modeled as the problem of channel noise elimination. Firstly, traditional algorithms are used to preliminarily pre estimate the received pilot signal, and then it is input into the channel estimation network to learn noise information and perform denoising processing to recover accurate channel coefficients. In order to improve the denoising ability of the network, attention mechanism modules and dilated convolution modules were designed to enhance the network's extraction of noisy subject information. At the same time, a feature fusion module was designed to prevent the loss of shallow features. The simulation results show that compared with classical DNCNN and CDRN networks, the proposed method reduces the normalized mean square error by an average of 2-2.89dB at different signal-to-noise ratios.

Key words: deep learning, reconfigurable intelligence surface, channel estimation, attention mechanism, convolutional neural network

摘要： 针对智能反射面辅助通信系统中信道估计精度低的问题，提出了一种基于信道去噪网络（CDN）的信道估计方案，将信道估计问题建模为信道噪声消除的问题。首先使用传统算法对接收到的导频信号进行初步预估计，然后将其输入信道估计网络以学习噪声信息并进行去噪处理，从而恢复出精确的信道系数。为了提高网络的去噪能力，设计了注意力机制模块和膨胀卷积模块以增强网络对噪声主体信息的提取，同时设计特征融合模块以防止浅层特征的丢失。仿真结果表明，与经典的DNCNN和CDRN网络相比，所提方法的归一化均方误差在不同信噪比下平均降低2~2.89dB。

关键词: 深度学习, 智能反射面, 信道估计, 注意力机制, 卷积神经网络