超大规模太赫兹系统深度学习信道估计算法  

Deep learning channel estimation algorithm for ultra-massive terahertz systems

作　　者：于舒娟[1] 赵阳[1] 魏玉尧 张昀 高贵 赵生妹[2] YU Shujuan;ZHAO Yang;WEI Yuyao;ZHANG Yun;GAO Gui;ZHAO Shengmei(College of Electronic and Optical Engineering&College of Flexible Electronics(Future Technology),Nanjing University of Posts and Telecommunications,Nanjing 210023,China;School of Communications and Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)

机构地区：[1]南京邮电大学电子与光学工程学院、柔性电子(未来技术)学院,江苏南京210023 [2]南京邮电大学通信与信息工程学院,江苏南京210003

出　　处：《通信学报》2025年第1期144-156,共13页Journal on Communications

基　　金：国家自然科学基金资助项目(No.62375140);江苏省研究生科研与实践创新计划基金资助项目(No.KYCX23_0994)。

摘　　要：为了进一步提升THz超大规模MIMO系统混合场信道估计性能,基于不动点网络(FPN)引入了一种基于跨通道信息交互的Transformer注意力机制模块与快速傅里叶变换卷积网络(FCN),提出了一种基于图像恢复网络的信道估计算法FPN-OTFN,将信道估计问题建模为图像恢复问题。在导频处采用最小二乘算法获得初始信道信息,并将其作为所提FPN-OTFN算法的输入,通过训练学习低精度信道图像和高精度图像间的映射关系,恢复出真实的信道状态信息。仿真实验结果表明,所提算法不仅继承了FPN框架的高效性、自适应性,同时对THz信道拥有较高的估计精度和良好的鲁棒性。In order to further improve the hybrid-field channel estimation performance in terahertz ultra-massive multiple-input multiple-output systems,an efficient cross channel Transformer module for image restoration and a fast Fourier transform convolutional network were introduced based on the fixed point network,and a scalable and efficient deep learning model FPN-OTFN was proposed,which models the channel estimation problem as an image restoration problem.Firstly,the least squares algorithm was used to obtain the channel information at the pilot location,and then the channel information was input into the proposed FPN-OTFN algorithm.By training and learning the mapping relationship between low precision channel images and high-precision images,the true channel state information was restored.The simulation results show that the proposed scheme not only inherits the high efficiency and adaptivity of the FPN framework,but also possesses high estimation accuracy and good robustness for THz channels.

关 键 词：信道估计 THz超大规模MIMO系统 深度学习 图像恢复 注意力机制 

分 类 号：TN92[电子电信—通信与信息系统]