快时变FDD大规模MIMO系统智能CSI反馈方法

Intelligent CSI feedback method for fast time-varying FDD massive MIMO system

针对快时变频分双工(FDD)大规模多输入多输出(MIMO)系统中因无线信道干扰使信道状态信息(CSI)矩阵中存在噪声以及多普勒频移导致的时间相关性使系统无法保证高可靠和低时延通信的问题,提出一种智能CSI反馈方法。该方法利用卷积神经网络(CNN)和批标准化(BN)网络对CSI矩阵中的噪声进行提取并且学习信道的空间结构,通过注意力机制提取CSI矩阵间的时间相关性以提高CSI重构的精度。利用标准的快时变信道模型仿真产生的数据对网络进行离线训练。系统仿真与分析表明,所提方法能够有效地抑制噪声的影响以及对多普勒引起的时间相关性进行提取。与代表性CSI压缩反馈方法和CsiNet方法相比,所提方法拥有更好的归一化均方误差(NMSE)和余弦相似度性能。

In the frequency division duplexing(FDD)massive multiple-input multiple-output(MIMO)system,the chan-nel state information(CSI)matrix existed noise caused by the wireless channel interference and the time correlation caused by Doppler shift.Because of these effects,the communication system couldn’t guarantee the requirements of re-liability and low delay.An intelligent CSI feedback method was adopted.The convolutional neural network(CNN)and batch normalization(BN)network was used to extract the noise in the CSI matrix and learned the spatial structure of the channel.The time correlation between the CSI matrices through the attention mechanism was extracted to improve the accuracy of CSI reconstruction.The data was generated by the standard fast time-varying channel model simulation to train the network offline.System simulation and analysis show that the proposed method can effectively suppress the in-fluence of noise and extract the time correlation caused by Doppler.Compared with the traditional CSI compression feedback algorithm and CsiNet algorithm,the proposed method has better NMSE and cosine similarity performance.