CSI feedback network for massive MIMO based on Transformer
In order to solve the problem of low feedback accuracy of channel state information feedback method in frequency-division duplex massive multi-input multi-output system,a CSI feedback network named CDTransformer based on Transformer and convolutional decomposition(CD)is proposed.CDTransformer incorporates convolutional decomposition into an improved Transformer network architecture to improve network performance without adding computational complexity,and enables lightweight deployment by binarizing the full connectivity layer of the network.A MixedTransformer network model is proposed for the limited power of the client.The encoder adopts a single layer convolutional neural network with low computational cost and simple structure,while the decoder adopts the same structure as the CDTransformer model.The CDTransformer model incorporates the Transformer structure and convolutional decomposition to improve CSI feedback accuracy and enable lightweight deployment.In addition,the MixedTransformer model was introduced,combining the advantages of CDTransformer and convolutional neural networks to provide better performance in power limited situations.The results show that compared with the CsiTransformer network model,the normalized mean square error and cosine similarity of CDTransformer network model are improved by 37.7%and 0.2%,respectively.
frequency division duplexmassive MIMOTransformerchannel state informationconvolutional neural network
NETL
NSTL
万方数据
维普
