Research On Massive MIMO Channel Estimation Algorithm Based On Machine Learning

Massive Multiple Input Multiple Output(MIMO)technology has become a key technology of5 G and even future wireless communication due to its numerous advantages.In Massive MIMO systems,accurate channel state information is crucial for a series of operations such as signal detection,signal preprocessing,beamforming,resource allocation,etc.In recent years,machine learning,especially deep learning,has shone brightly in various fields.The further integration of wireless communication and machine learning in the future has become a new trendency.which provides new methods for channel estimation.This thesis focuses on the use of deep learning to solve channel estimation problems in massive MIMO systems.The main work of the thesis is as follows:(1)Aiming at the problems of low accuracy and high complexity of traditional channel estimation methods,a channel estimation algorithm based on super-resolution convolutional neural networks is proposed.The algorithm uses a super-resolution convolutional network to refine the initial low-precision channel input to reconstruct a high-precision channel.Analysis shows that this algorithm has lower complexity compared to the same type algorithms,and is more conducive to deployment in terminals with limited resources.Simulation results show that the proposed channel estimation algorithm has good channel estimation accuracy and generalization ability.(2)A convolutional network channel estimation algorithm based on Transfer Learning strategy is proposed for scenarios where the training and usage environments of deep networks do not match.The algorithm learns the process of recovering high-precision channels by using an upsampling network and a feature extraction network.By using Transfer Learning to adjust the entire upsampling network and parts of the feature network,the purpose of multiplexing channel prior knowledge and reducing training overhead is achieved.Simulation results show that the proposed channel estimation algorithm can achieve good channel estimation performance at a low cost in the case of channel mismatch.(3)A channel estimation algorithm based on attention assisted generation adversarial networks is proposed to address issues such as noise interference and high pilot overhead.This algorithm uses a denoising network to denoise the received signal,and uses a Generative Adversarial Network(GAN)to capture the channel distribution.In order to fully utilize the sparse structural characteristics of the channel itself,attention mechanisms are used to assign different weights to different features.Simulation shows that the channel estimation algorithm still has good channel estimation accuracy under conditions such as low pilot overhead and high noise interference.