| Wireless communication systems and related technologies develop rapidly,the 5th Generation(5G)mobile communication technology should improve the data transmission rate and security performance while ensuring the data transmission quality of communication users.Massive multiple-input multiple-output(MIMO)wireless communication system and related technologies have become the hot research content of people’s attention.Previously,deep learning brings a new solution to the channel feedback problem of massive MIMO systems.Some existing channel feedback research can ensure the realization of feedback to a certain extent by using traditional compressed sensing related theories and algorithms,but it has disadvantages such as high computational overhead and slow execution speed.The channel state information feedback method using deep learning is expected to be faster and more accurate in channel state information recovery.By fitting data rules,the mapping relationship between the input and output of the neural network can be constructed.Therefore,in the research on massive MIMO wireless communication,based on the inspiration of compressed sensing algorithm and the advantages of deep learning method,this thesis combines the field of wireless communication with the field of deep learning,from the perspective of intelligent communication,for downlink Channel feedback to expand learning and discussion.The main research contents of this thesis are as follows:1.In massive MIMO systems,CSI feedback based on deep learning provides some feedback accuracy and feedback efficiency with respect to dimensional compression of the CSI matrix,but does not consider the corresponding quantization feedback of the dimensionally compressed code words.In this thesis,a CSI feedback scheme based on the combination of dimensional compression and code word quantization is proposed,CQCsi Net,which enhances the feature extraction of the original CSI signal by extending the convolutional field of perception and adding multi-scale feature fusion,so that the CQCsi Net feedback scheme can guarantee a certain field of perception and obtain multi-scale feature information to achieve the purpose of CSI dimensional compression.To further improve the feedback performance,quantization and other related operations are added to make the CQ-Csi Net feedback method more suitable for the actual system storage and transmission requirements.When performing CSI reconstruction,the CSI feature information is further utilized by adding an attention mechanism in the decoder design to make the CSI matrix obtain different weights on different feature map channels.It is experimentally verified that the CQ-Csi Net feedback has excellent CSI feedback accuracy and robustness both indoors and outdoors,and ensures good CSI feedback performance even with different settings of feedback bits and quantization bits.2.The CSI feedback scheme based on deep learning has achieved improvement in reconstruction accuracy and feedback efficiency,but still suffers from the problems of being considered as a black box and lack of explainability due to its use.To address these problems,this thesis proposes the CSI feedback scheme ISTA-Csi Net based on the deep compressed sensing algorithm unfolding based on the advantages of interpretability of compressed sensing algorithms and data-driven tunable parameters of neural networks.ISTA-Csi Net helps to improve the interpretability of neural network structure design by corresponding the iterative solving algorithm to an equivalent deep learning network.ISTA-Csi Net helps to improve the efficiency of the encoder network in compressing the CSI matrix by using a trainable measurement matrix to mine and exploit CSI-related data features.ISTA-Csi Net also uses end-to-end training to update the parameters of the encoder and decoder networks.The comparative experimental results show that ISTA-Csi Net improves the interpretability of CSI feedback scheme,and also guarantees the reconstruction accuracy of CSI feedback. |
PDF Full Text Request