| 摘要: |
| 为了在可重构智能反射面（Reconfigurable Intelligent Surface，RIS）辅助通信系统中精确估计信道，并解决信道估计开销过高的问题，提出了一种基于快速超分辨率卷积神经网络（Fast Super-Resolution Convolutional Neural Network，FSRCNN）的信道估计方案。在信道估计的初始阶段，选择关闭部分反射元件，并借助少量导频信号完成信道估计，将估计结果视为低精度与低分辨率（Low Resolution，LR）的图像,通过线性插值将其扩展为具有低精度的高分辨率（High Revolution，HR）图像。随后，利用FSRCNN提高估计结果的精度，并通过基于深度残差网络的噪声去除模型（CNN-based Deep Residual Network，CDRN）进一步提升信道估计的准确性。数值结果表明，相较于基准方案，所提的信道估计方案在保持低信道估计开销的同时，得到了更准确的信道估计结果。 |
| 关键词: 可重构智能反射面 信道估计 超分辨网络 深度残差网络 |
| DOI：10.20079/j.issn.1001-893x.231116005 |
|
| 基金项目:重庆市自然科学基金面上项目（cstc2021jcyj-msxmX0761）；广西重点研发计划（AB24010317） |
|
| An RIS Channel Estimation Method Based on Super-Resolution Networks |
| GAN Chenquan,b,GUO Yuhang |
| (a.School of Communication and Information Engineering;b.School of Cyber Security and Information Law,Chongqing University of Posts and Telecommunications,Chongqing 400065,China) |
| Abstract: |
| To precisely estimate the channel in reconfigurable intelligent surface(RIS)-assisted communication systems,and solve the issue of high channel estimation overhead,the authors propose a channel estimation scheme based on the fast super-resolution convolutional neural network(FSRCNN).In the initial phase of channel estimation,a subset of reflective elements is deactivated,and a limited number of pilot signals are utilized to conduct the channel estimation,and the estimation outcomes are treated as low-precision and low-resolution(LR) images.These LR images are subsequently upscaled to high-resolution(HR) images with low precision through linear interpolation.Subsequently,the precision of the estimation outcomes is enhanced using FSRCNN,and further elevated through the convolutional neural network-based deep residual network(CDRN).Numerical results demonstrate that,the proposed method achieves more accurate results while concurrently minimizing the associated computational overhead of low channel estimation compared with the baseline method. |
| Key words: reconfigurable intelligent surface(RIS) channel estimation super-resolution network deep residual network |
