Research On Channel Decoding Algorithm Based On Deep Learning

In the digital communication,channel coding and decoding technology is used to resist the influence of noise and interference in the process of information transmission and reduce bit error rate.However,the advantages,disadvantages and complexity of various coding and decoding schemes are different.Polar code is a kind of linear block code based on channel polarization phenomenon proposed by professor Erdal Arikan in 2008.The polarization code has a fixed coding structure and has been theoretically proven to be able to approach the Shannon limit infinitely,but in the case of short code word,its decoding performance is not ideal,and the decoding delay of SC(Successive Cancellation)decoding algorithm is high.In recent years,deep learning technology has been widely used in various fields,which can learn the types and characteristics of data in a complex and changing environment,and the model has strong flexibility and universality.This paper mainly studies the application of deep learning in polarization code decoding,builds a decoding model,analyzes its decoding performance,and applies the model to decode random code,hamming code and LDPC code to verify the applicability of the decoding model.Firstly,this paper analyzes two techniques of channel coding in 5G mobile communication system: Polar Code and LDPC Code,and their coding and decoding algorithms.The process of channel combination and splitting in channel polarization is analyzed,and the capacity distribution of each subchannel after channel polarization under three different code lengths is simulated.At the same time,the Polar code construction method based on Bhattacharyya and the SC decoding algorithm are introduced.Besides,two construction methods based on the full lower triangle and the approximate lower triangle are introduced,and the decoding performance of LDPC codes based on belief propagation algorithm is simulated and analyzed.Secondly,the decoding model of polarizing code based on deep learning is studied and constructed.The basic structure and training process of the neural network are analyzed.The design process of the BP neural network structure and the working principle of the back propagation algorithm are studied.The effects of different activation functions,loss functions,iteration times and network structure on the performance of polarization decoding network are analyzed by simulation.The decoding model of polarization codes based on deep learning is obtained through experiments,and the decoding performance achieves the maximum posteriori probability performance.Finally,the application of this decoding model in random code,hamming code and LDPC code is verified,and the performance of the decoding algorithm based on deep learning is compared with that of the traditional decoding algorithm.The decoding performance based on deep learning is better than the traditional decoding algorithm in decoding accuracy and decoding duration.