基于深度学习的CSI压缩反馈方案研究

Research of CSI compression feedback based on deep learning

针对基于深度学习的信道状态信息(channel status information,CSI)压缩反馈技术,本文提出了一种基于多层感知机(multi-layer perceptron,MLP)-Transformer深度自编码器的CSI压缩反馈方案。该方案从现有算法的不足出发,本着对编码器模块的轻量级设计准则,以及Transformer自注意力机制的应用,设计了一种高精度CSI压缩重构的网络模型。仿真结果表明,基于MLP-Transformer的深度自编码器CSI压缩反馈方案在频分双工(frequency division duplex,FDD)下行大规模多输入多输出(multiple input multiple output,MIMO)系统场景的不同压缩比下,均有很高的CSI重构精度,且其编码器模块的轻量级设计对比其他方案具有更低的编码器复杂度。

For compression feedback technology of channel status information(CSI)based on deep learning,this paper proposed a CSI compression feedback based on multi-layer perceptron(MLP)-Transformer deep autoencoder.Starting from the shortcomings of existing algorithms,this scheme realized a network model of high-precision CSI compression and reconstruction recovery,based on the lightweight design criteria of encoder module and the application of Transformer self-attention mechanism.The simulation results show that the CSI compression feedback scheme of deep self encoder based on MLP-Transformer has high CSI reconstruction accuracy under different compression ratios of frequency division duplex(FDD)downlink massive multiple input multiple output(MIMO)system scenarios,and that the lightweight design of its encoder module has lower encoder complexity than other schemes.