Signal Detection Of Massive-MIMO System Based On Deep Learning

Due to its significant advantages in improving the communication capacity of the channel,massive multi-input multi-output(MIMO)technology has become one of the key technologies of the 5th generation telecommunication system.The signal processing dimension is greatly increased due to the scale of hundreds of antennas,and the pilot overhead for channel estimation is also unacceptable.Therefore,the signal detection design of massive MIMO system is faced with great challenges.Artificial intelligence technology represented by deep learning has been successfully applied to image processing,natural language processing and other fields,showing the advantages of processing complex data with high dimensions.This paper takes advantage of this to carry out research on massive MIMO signal detection technology based on deep learning.Deep learning has been preliminarily applied in communication signal detection,and its main idea is the direct migration of traditional network architecture.Since the sparsity and asymptotic orthogonality of massive MIMO are not considered,direct migration will lead to complex data feature extraction process,slow training convergence.It is difficult for signal detection performance to meet system requirements.Aiming at this key issue,this paper conducts the following research:The traditional design method of massive MIMO signal detection is to divide the functional modules according to the processing flow,and carry out modeling and optimization module by module.In this paper,the module independent design(MID)method is studied in depth,and it is found that the module processing errors will be amplified step by step,and the design methods of joint channel estimation and signal detection and optimized channel estimation are given,so as to improve the performance of massive MIMO signal detection with different user sizes.Under the scenario of small number of users,the classification algorithm of deep neural network has been used to solve the defects of massive MIMO algorithm with high complexity based on the joint channel estimation and signal detection idea.There are still some problems such as poor fault tolerance in categories,multiple optimization parameters,and the related features being drowned by noise.Aiming at this key problem,this paper studies a massive MIMO signal detection algorithm based on the classification of convolutional depth neural network by combining Gray coding,convolutional neural network and channel sparsity,which can achieve algorithm performance gain close to the optimal detection performance.The convergence speed of network training is improved.Under the scenario of large number of users,with the increase of the number of users,the above categories of signal detection classifiers grow exponentially.Aiming at this key problem,using the sparsity of massive MIMO channels,based on the design idea of optimized channel estimation,a sparse angular domain channel estimation algorithm based on the denoising convolutional neural network is studied,which can realize the channel estimation with low pilot overhead and high accuracy,and improve the overall signal detection performance.The study shows that the convolutional depth neural network improves the convergence speed by 1.5 times and the signal detection performance by 0.6 dB compared with the traditional deep learning algorithm.The sparse Angle domain channel estimation algorithm based on denoising convolutional neural network improves 16 dB in channel estimation performance,2.5 dB in overall signal detection performance compared with the semi-blind algorithm,and reduces the pilot cost by 99.2% compared with the least square channel estimation algorithm.