Research On Channel Information Acquisition And Signal Detection Technology Of MIMO System Based On Deep Learning

Multiple-input multiple-output(MIMO)technology has potential performance in terms of improving spectral efficiency and improving communication link reliability,and has been widely used in wireless communication systems.In the MIMO communication system,the base station and the users configure multiple antennas to fully exploit and utilize the space diversity and space division multiplexing characteristics between the antennas,and increase the system capacity and transmission without additional spectrum bandwidth effectiveness.However,as the number of antennas increases,a series of system complication problems have emerged.For example,the optimal detection and estimation of signals,accurate channel estimation and feedback,etc.This paper proposes a solution based on deep learning for signal detection and channel feedback.The algorithm has the characteristics of good performance and low complexity.Aiming at the time-varying communication system,a MIMO soft decision signal detection method is proposed.The system channel state information and the received data training set are input into the deep neural network,and the weight and bias of the neural network are optimized by using the cross-entropy loss function and the forward root mean square gradient(RMSProp)descent algorithm.The output layer of the neural network uses the Sigmoid activation function.The inverse of the input value of the Sigmoid function is the log-likelihood ratio(LLR)value,which significantly reduces the complexity of calculating the LLR value.,Improve the signal detection performance.Simulation results show that the proposed algorithm has better performance than the minimum mean square error detection algorithm,and the performance of the proposed algorithm is close to the performance of the theoretically optimal maximum likelihood signal detection algorithm.Based on the framework of Long-short Term Memory(LSTM)network and attention mechanism,a feedback mechanism of Channel State Information(CSI)for massive MIMO system is proposed,which has high recovery accuracy and feedback pilot frequency.The advantage of low overhead.The scheme includes two parts of CSI compression and reconstruction based on Convolutional Neural Networks(CNN).The LSTM unit is used to learn the time correlation of the MIMO channel,and the correlation is used to establish theattention of the local information and automatic weighted feature information Mechanism to improve the accuracy of CSI recovery.Simulation experiments verify the effectiveness of the algorithm.The trained CNN can obtain higher feedback accuracy and better system performance in massive MIMO CSI online feedback reconstruction.Based on the long-short time-attention channel feedback mechanism,a lightweight massive MIMO system CSI feedback network is proposed.In this network,the fully connected network compresses the channel feature vector to a low-dimensional vector and inputs it to the LSTM-Attention network for time feature training.This mechanism effectively reduces the number of neural network training parameters and the computational complexity of the feedback mechanism while ensuring the accuracy of channel recovery,thereby speeding up the channel feedback rate.Based on various channel information training data sets,through offline iterative training and learning,an optimized CSI feedback network can be obtained.Compared with traditional neural network algorithms,lightweight neural networks also have higher performance and lower computational complexity.Related research results provide important ideas and solutions for solving channel information acquisition and signal detection problems in MIMO systems,and lay the foundation for the practical application of deep learning in 5G.