| 摘要: |
| 针对通信系统中由于路径损耗导致的信道估计难题以及传统可重构智能表面（Reconfigurable Intelligent Surface，RIS）信道估计方法存在的高导频开销问题，提出了一种基于通道注意力残差网络（Channel Attention Residual Network，CA-ResNet）的信道估计方法。该方法结合最小二乘（Least Squares，LS）估计与深度学习技术，将低分辨率信道矩阵重建为高分辨率信道矩阵。为了减少导频开销，对 RIS 反射单元进行分组，每组单元共享相同的反射系数。CA-ResNet通过残差模块提取关键特征，并利用双池化通道注意力模块（Dual-pooling Channel Attention，DPCA）优化通道权重，从而进一步提升信道估计的精度。仿真结果表明，所提方法在分组数为4时，相比增强型超分辨率（Enhanced Deep Super-resolution，EDSR）和全局注意力残差网络（Global Attention Residual Network，GARN）模型的归一化均方误差分别下降了2.28~2.83 dB和1.21~1.33 dB。 |
| 关键词: 信道估计 智能反射面 深度学习 通道注意力机制 |
| DOI：10.20079/j.issn.1001-893x.250110002 |
|
| 基金项目:重庆市自然科学基金创新发展联合基金（中国星网）（CSTB2023NSCQ-LZX0114）；重庆市自然科学基金面上项目(cstc2021jcyj-msxmX0454) |
|
| Channel Attention Residual Network for Channel Estimation in RIS-assisted Communication Systems |
| YANG Liming,GAO Xiaomin |
| (School of Communications and Information Engineering，Chongqing University of Posts and Telecommunications，Chongqing 400065，China) |
| Abstract: |
| A channel estimation method based on channel attention residual network(CA-ResNet) is proposed to address the challenges caused by path loss in communication systems and the high pilot overhead in traditional reconfigurable intelligent surface(RIS) channel estimation methods.The method combines least squares(LS) estimation with deep learning techniques to reconstruct low-resolution channel matrices into high-resolution ones.To reduce pilot overhead，a strategy of grouping RIS reflection elements is introduced，where each group of elements shares the same reflection coefficient.CA-ResNet extracts key features through residual modules and optimizes channel weights using a dual-pooling channel attention(DPCA) module.Simulation results show that with a grouping number of 4，the proposed method reduces normalized mean squared error by 2.28~2.83 dB and 1.21~1.33 dB compared with enhanced deep super-resolution(EDSR) and global attention residual network(GARN)，respectively. |
| Key words: channel estimation reconfigurable intelligent surface deep learning channel attention mechanism |
《电讯技术》编辑部 