快时变信道下基于深度学习的OFDM系统信道估计  

Deep learning based channel estimation for OFDM systems in fast time-varying channel

作　　者：季策 宋博翰[1] 耿蓉 梁敏骏 JI Ce;SONG Bohan;GENG Rong;LIANG Minjun(School of Computer Science and Engineering,Northeastern University,Shenyang 110169,China;Key Laboratory of Intelligent Computing for Medical Imaging,Ministry of Education,Northeastern University,Shenyang 110169,China;College of Information Science and Engineering,Northeastern University,Shenyang 110819,China)

机构地区：[1]东北大学计算机科学与工程学院,辽宁沈阳110169 [2]东北大学医学影像智能计算教育部重点实验室,辽宁沈阳110169 [3]东北大学信息科学与工程学院,辽宁沈阳110819

出　　处：《系统工程与电子技术》2023年第11期3649-3655,共7页Systems Engineering and Electronics

基　　金：国防预研项目国防重大培育项目(N2116015)资助课题。

摘　　要：针对快时变信道的非平稳特性会造成信道估计性能变差的问题,在基扩展模型下提出了一种基于深度学习的信道估计算法,并将其应用于正交频分复用(orthogonal frequecy division multiplexing,OFDM)系统中。首先,根据快时变信道矩阵的局部相关特性,构建时频特征提取网络,利用卷积结构提取快时变信道在时域和频域的相关特征,并嵌入到下一级网络中进行特征的融合。其次,利用门控循环网络捕捉信道在不同符号处的变化相关性,在快时变信道环境下实现更准确的信道估计。仿真结果表明,与其他快时变环境下的信道估计算法相比,算法的估计性能提升明显;同时,网络的轻量化结构使算法的复杂度最低下降20%。In order to solve the problem of poor channel estimation performance caused by non-stationary characteristics of fast time-varying channels,a channel estimation algorithm based on deep learning under basis expansion model is proposed and applied to orthogonal frequency division multiplexing(OFDM)systems.Firstly,according to the local correlation characteristics of the fast time-varying channel matrix,a time-frequency feature extraction network is constructed to extract the relevant features of the channel in time domain and frequency domain by using the convolution structure,and is embedded in the next network for feature fusion.Secondly,the gated recurrent unit is used to capture the time correlation of channel changes at different symbols,so as to achieve more accurate channel estimation in the fast time-varying channel environment.Simulation results show that compared with other channel estimation algorithms in fast time-varying environments,the performance of the proposed algorithm is improved obviously.Meanwhile,the lightweight structure of the network reduces the complexity of the algorithm by at least 20%.

关 键 词：信道估计 正交频分复用系统 快时变 深度学习 基扩展模型 

分 类 号：TN911.72[电子电信—通信与信息系统]