Research On Channel Estimation And Equalization Methods For Massive MIMO-OFDM Systems Based On Neural Networks

Massive multi-input multi-output(MIMO)has been regarded as one of the key technologies for fifth generation(5G)mobile communication systems,as it can significantly enhance the system capacity with high spectrum and energy efficiency.For massive MIMO-OFDM system,accurate channel estimation and equalization at the receiver is an essential part,especially when the number of antennas increases,the use of conventional pilot-based method may cause serious pilot pollution problems and high computational complexity.In this paper,a novel channel framework that integrates the imperfect channel estimation and equalization of the massive MIMO-OFDM into the deep learning scheme is proposed.A proposed deep neural network divides the data filtered by the channel into data sets and test sets,used for training and learning,respectively.This type of method can break through the limitations of traditional communication systems,cope with more complex channel environments,and reduce the complexity of algorithms while achieving performance improvements.The performance of the channel estimation method introduced by DNN is better than that of conventional method,there is no requirement of prior statistics information about channel autocorrelation matrix,noise variance and the complex operation of matrix inversion.Regardless of the application scenarios in which the channel model is considered to achieve angular emission angle or correlation,neural networks can be used to ignore the one-by-one estimation of detailed channel parameter information,which is simpler and smarter.The channel equalization based on the DNN network is regarded as a classification problem,and the input signals are divided into different classes according to the constellation diagram of the modulation method.The received signals are restored to the corresponding classes by the DNN network equalizer,and the transmitted signals are recovered.Based on this,this paper combines the correspondence between the known pilot data sent by the massive MIMO-OFDM system and the network output value,and uses the idea of mathematical classification and weighting to propose an optimized DNN network cost function.Simulation experiments show that the improved DNN network equalizer can not only provide better equalization performance,but also greatly improve the network convergencespeed at the initial stage of training and reduce the time required for training.