一种深度学习的波束空间信道估计算法  

A Deep Learning Algorithm for Beamspace Channel Estimation

作　　者：郑娟毅 张庆珏 董嘉豪 郭梦月 杨溥江 ZHENG Juanyi;ZHANG Qingjue;DONG Jiahao;GUO Mengyue;YANG Pujiang(School of Communication and Information Engineering,Xi'an University of Posts and Telecommunications,Xi'an 710121,Shaanxi,China)

机构地区：[1]西安邮电大学通信与信息工程学院,陕西西安710121

出　　处：《计算机工程》2024年第5期298-305,共8页Computer Engineering

基　　金：国家自然科学基金(61901367)。

摘　　要：在时分双工(TDD)毫米波大规模多输入多输出(MIMO)系统中,因为波束空间信道具有稀疏性,导致将低维测量数据重建为原始高维信道时会带来较高的复杂度。针对上行链路,在不考虑稀疏度的情况下,将传统优化算法和基于数据驱动的深度学习方法相结合,提出一种改进的基于深度学习的波束空间信道估计算法。从重建过程入手,通过交替建立梯度下降模块(GDM)和近端映射模块(PMM)来构建网络。首先根据SalehValenzuela信道模型进行理论公式推导并生成信道数据;其次构建一个由传统迭代收缩阈值算法(ISTA)的更新步骤所展开的多层网络,并将数据传输到该网络,每层对应于一次类似ISTA的迭代;最后对训练好的模型进行在线测试,恢复出待估计的信道。构建Py Torch环境,将该算法与正交匹配追踪(OMP)算法、近似消息传递(AMP)算法、可学习的近似消息传递(LAMP)算法、高斯混合LAMP(GM-LAMP)算法进行对比,结果表明:在估计精度方面,所提算法相对表现较好的深度学习算法LAMP、GM-LAMP分别提升约3.07和2.61 d B,较传统算法OMP、AMP分别提升约11.12和9.57 d B;在参数量方面,所提算法较LAMP、GM-LAMP分别减少约39%和69%。In a Time Division Duplex(TDD)millimeter-wave massive Multiple-Input Multiple-Output(MIMO)system,because of the sparsity of the beamspace channel,the original high-dimensional channel is effectively reconstructed from low-dimensional measurement data.For the uplink,without considering sparsity,this study combines the traditional optimization algorithm with a data-driven deep learning method and proposes an improved beam spatial channel estimation algorithm based on deep learning.Starting from the reconstruction process,the network is constructed by alternately establishing a Gradient Descent Module(GDM)and a Proximal Mapping Module(PMM).Specifically,a theoretical formula is deduced according to the Saleh-Valenzuela channel model,and channel data are generated.Second,the data are transferred to a network comprising a fixed number of layers using the update step of the traditional Iterative Shrinkage Thresholding Algorithm(ISTA),and each layer corresponds to an iteration similar to that of ISTA.Finally,the trained model is tested online to restore the channel to be estimated.Through the construction of the PyTorch environment,the proposed algorithm is compared with the Orthogonal Matching Pursuit(OMP),Approximate Message Passing(AMP),Learnable AMP(LAMP),and Gaussian Mixture LAMP(GM-LAMP)algorithms.The results demonstrate that the proposed algorithm improves the estimation accuracy by approximately 3.07 and 2.61 dB compared with better deep learning algorithms,LAMP and GM-LAMP,and by approximately 11.12 and 9.57 dB with the traditional OMP and AMP algorithms.The number of parameters is approximately 39%and 69%less than those of LAMP and GM-LAMP algorithms,respectively.

关 键 词：大规模多输入多输出系统 稀疏信道估计 压缩感知 深度学习 迭代收缩阈值算法 无线通信 

分 类 号：TN928[电子电信—通信与信息系统]