| 摘要: |
| 在大规模多输入多输出（Multiple-Input Multiple-Output，MIMO）系统中，基站根据用户反馈的信道状态信息进行自适应编码调制以提高频谱效率，因此需要将用户侧估计到的信道状态信息反馈到基站。由于反馈过程存在的延迟会降低系统性能，因此在考虑延迟的情况下，对基于深度学习的信道状态信息自编码器CsiNet进行改进，使用下行信道的延迟状态信息作为信道状态信息自编码器的期望输出信号来对自编码器进行训练，减少了反馈延迟误差的影响。仿真结果表明，在延迟为1时隙时，所提方案的归一化均方误差(Normalized Mean Square Error,NMSE)仅为自回归-主成分分析方案、基于压缩感知的方案和基于卷积神经网络的CsiNet方案的1/8~1/7，并且随着时隙增加，NMSE性能提升越明显。 |
| 关键词: 大规模MIMO 信道状态信息 压缩反馈 深度学习 自编码器 |
| DOI： |
|
| 基金项目:国家科技重大专项(2018ZX03001026-002) |
|
| An Improved CSI Feedback Delay Algorithm Based on Convolutional Neural Network for Massive MIMO Systems |
| WANG Yue,DUAN Hongguang,ZHENG Xinglin |
| (School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China) |
| Abstract: |
| In massive multiple-input multiple-output(MIMO) systems,base station performs adaptive modulation and coding based on channel state information(CSI) fed back by user side to improve spectral efficiency.Therefore,the CSI estimated by the user side needs to be fed back to base station.Since the delay of the feedback process will reduce the system performance,this paper improves CSI auto-encoder based on deep learning by considering the feedback delay,the auto-encoder is trained by setting the delayed version of the downlink channel as the desired output to reduce the effects of feedback delay.The simulation results show that the normalized mean square error(NMSE) of the proposed scheme is about 7~8 times lower than that of the autoregressive-principal component analysis（AR-PCA）scheme,compressed sensing(CS) based scheme and convolutional neural network(CNN) based CsiNet scheme when the delay is 1 time slot.With the increase of time slots,the NMSE performance improvement becomes more and more obvious. |
| Key words: massive MIMO channel state information compress and feedback deep learning auto-encoder |
《电讯技术》编辑部 