基于多尺度特征提取的轻量化大规模MIMO系统CSI反馈  

Lightweight CSI Feedback Based on Multi-scale Feature Extraction for Massive MIMO Systems

作　　者：刘受清 朱正发 申滔 LIU Shouqing;ZHU Zhengfa;SHEN Tao(School of Electrical&Information Engineering,Changsha University of Science&Technology,Changsha 410114,China)

机构地区：[1]长沙理工大学电气与信息工程学院,湖南长沙410114

出　　处：《无线电工程》2025年第1期175-183,共9页Radio Engineering

基　　金：湖南省教育厅一般项目(19C0037);长沙理工大学科研创新项目(CLSJCX23067)。

摘　　要：在频分双工(Frequency Division Duplex,FDD)模式的大规模多输入多输出(Multiple Input Multiple Output,MIMO)系统中,针对资源有限的用户设备(User Equipment,UE)向基站(Base Station,BS)反馈信道状态信息(Channel State Information,CSI)反馈开销太大、反馈精度不足以及网络计算复杂度高的问题,提出一种基于深度可分离卷积和多尺度特征提取的轻量化CSI反馈方案。采用轻量的深度可分离卷积处理CSI,以降低压缩信息的损失,通过多尺度特征提取和残差学习进行恢复重建CSI。仿真结果表明,所提方案相对其他轻量化网络表现出较好的反馈精度。In Frequency Division Duplex(FDD)massive Multiple Input Multiple Output(MIMO)systems,User Equipment(UE)with limited resources has to feed Channel State Information(CSI)back to the Base Station(BS),resulting in high feedback overhead and low feedback accuracy.Taking high network computational complexity into account,a lightweight CSI feedback scheme based on depth-wise separable convolution and multi-scale feature extraction is proposed.In the scheme,CSI is processed by lightweight depth-wise separable convolution to reduce loss of compressed information,then CSI is recovered and reconstructed by multi-scale feature extraction and residual learning.Simulation results show that the proposed feedback scheme outperforms other lightweight networks in terms of feedback accuracy.

关 键 词：频分双工 大规模多输入多输出 深度可分离卷积 多尺度特征提取网络 轻量化 信道状态信息反馈 

分 类 号：TN929.5[电子电信—通信与信息系统]