非完美反馈链路下可变压缩率CSI反馈方法  

Changeable compression ratio CSI feedback method underimperfect feedback links

作　　者：黄凤翔 段红光[1] 毛翔宇 HUANG Fengxiang;DUAN Hongguang;MAO Xiangyu(School of Communications and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China)

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065

出　　处：《重庆邮电大学学报(自然科学版)》2025年第2期232-240,共9页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：重庆市自然科学基金项目(CSTB2022NSCQ-MSX1125)。

摘　　要：在频分双工(frequency division duplex,FDD)大规模多输入多输出(multiple input multiple output,MIMO)系统中,针对目前基于深度学习的信道状态信息(channel state information,CSI)反馈方法,其压缩率固定及FDD反馈链路存在不对称问题,提出一种基于可变压缩率与去噪神经网络的CSI反馈方法。该方法采用自编码器结构,在用户端使用固定压缩率的编码器对CSI进行压缩,随后通过反馈约束来限制反馈码字长度,当基站端接收到反馈码字后通过反馈约束对码字进行零填充,使用去噪单元对其去噪,再使用解码器将去噪后的码字重建为CSI矩阵。在簇延时线(clustered delay line,CDL)信道环境中仿真。实验结果表明,所提方法较CsiNet可减少约48%的存储开销,且能有效抑制反馈误差对信道状态信息重建造成的影响。In frequency division duplex(FDD)massive multiple input multiple output(MIMO)systems,a channel state information(CSI)feedback method based on changeable compression rate and denoising neural network is proposed to address the fixed compression rate and asymmetric FDD feedback link of current deep learning-based CSI feedback methods.This method adopts an autoencoder structure,which compresses CSI using a fixed compression rate encoder at the user end.Then,feedback constraints are used to limit the length of the feedback codeword.When the base station receives the feedback codeword,it first fills the codeword with zero through feedback constraints,then uses a denoising unit to denoise it,and finally uses a decoder to reconstruct the denoised codeword into a CSI matrix.Simulation in a clustered delay line(CDL)channel environment shows that the proposed method can reduce storage overhead by about 48%compared to CsiNet,and effectively suppress the impact of feedback errors on channel state information reconstruction.

关 键 词：大规模多输入多输出 信道状态信息 压缩反馈 深度学习 

分 类 号：TN929.5[电子电信—通信与信息系统]