Research On Key Technology Of Massive MIMO Channel Feedback For Intelligent Communications

Massive multiple-input multiple-output(MIMO)communication systems have been recognized as one of the key technologies of the fifth generation(5G)mobile communication systems.The use of a large number of antennas provides the system with higher spectral efficiency and energy efficiency However,the difficulty of acquiring channel state information(CSI)has increased at the same time,especially for the widely used frequency division duplex(FDD)system since the uplink and downlink channels have no reciprocity.Novel and efficient CSI feedback schemes must be proposed to adapt to massive MIMO systems.In recent years,the artificial intelligence technology,typically deep learning(DL),has rapidly developed and been introduced into various industries,achieving remarkable results.The application of DL in the field of communication rises the idea of intelligent communications,and offers a new solution for the CSI feedback problem of massive MIMO systems.In this thesis,we focus on intelligent communications and investigate the key techniques of channel feedback for FDD massive MIMO systems based on DL methods.First,we study the intelligent communication technologies for the physical layer of wireless communication systems.We investigate the basic structure and characteristics of three classical neural network architectures,which are fully connected neural networks,convolutional neural networks and recurrent neural networks.We study the application of DL in each module of the conventional wireless communication system,including modulation identification,channel decoding,channel estimation and signal detection.A novel end-to-end communication system architecture based on DL autoencoder is further explored,including its application in MIMO and multi-user scenarios.Subsequently,we propose a DL-based CSI feedback algorithm for time-varying channels of an FDD massive MIMO communication system.We set up an orthogonal frequency-domain multiplexing(OFDM)system that a base station(BS)with uniform linear antenna array(ULA)communicates with a single-antenna user equipment(UE)with certain velocity as our system model.We analyse the characteristics of the timevarying channel,including the sparse nature of the matrix in the angular-delay domain and the correlation property within coherence time.Correspondingly we design a DL network called CsiNet-LSTM to realize channel compression,feedback and reconstruction.We also describe the dataset generation scheme,training scheme,performance index,and hyperparameter settings.We compare the normalized mean-squared error(NMSE),cosine similarity and time complexity of CsiNet-LSTM and the state-of-art compressed sensing(CS)and DL methods in different compression ratios and communication scenarios.We also analyse the the impact of UE mobility and the number of antannas to CsiNet-LSTM performance by simulation.Simulation results show that CsiNet-LSTM significantly improves the reconstruction accuracy and robustness to compression ratio with extremely low time complexity.Finally,we propose an uplink channel-aided method for time-varying CSI feedback for an FDD massive MIMO communication system.We choose a FDD full-duplex communication system with a multi-antenna BS and a single-antenna UE as our system model.Then we analyse the reason why the uplink and downlink channels are correlated,and illystrate the feasibility of taking advantage of uplink channel for downlink CSI feedback.We design a DL network UA-CsiConvLSTM accordingly,and propose the dataset generation scheme and hyperparameter settings.The simulation compares the NMSE,cosine similarity performance and parameter complexity of UA-CsiConvLSTM and CsiNet-LSTM.The results show that UA-CsiConvLSTM improves the reconstruction performance in all scenarios under the premise of greatly reducing the parameter quantity,which illustrates the feasibility and practicability of the uplink channel-aided CSI feedback method.