Research On Polar Decoding Algorithms Based On Deep Learning

Polar code is a constructive code whose capacity can reach the Shannon limit.It has now become a coding scheme for 5G mobile communication systems and is a research direction in the frontier field of channel coding.However,the traditional decoding algorithms of polar codes still need to be improved in terms of error correction performance and decoding delay.The current research on the theoretical framework of the polar code deep learning decoding algorithm and the low-complexity and high-reliability neural network decoder is far from enough.In view of this,this paper studies the basic theory and key technologies of deep learning in polar code decoding on the basis of polar transmission theory.Under the guidance of this theoretical framework,this paper innovates and improves the theory and algorithm of the cyclic redundancy list SC algorithm,and conducts a comprehensive study on the improved CRC-SCL decoding algorithm and neural network coupling decoding,which ensures the polarization Encoding efficient,reliable transmission and accurate decoding.The work of this paper mainly includes the following aspects:(1)This paper studies the relevant theoretical framework of polar codes,and conducts simulation experiments on the polar code decoding algorithm.Using SC decoding algorithm,CRC-SCL decoding algorithm,the code lengths are 128,256,512,1024,2048,the code rate is 0.5,and the list length is 4 and 32.The experimental results show that when the code length of the polar code is longer,the error performance of the CRC-SCL decoding algorithm is better.(2)The traditional CRC-SCL decoding algorithm is improved in terms of theoretical algorithm.It is found that there are two problems in the decoding process of the traditional CRC-SCL decoding algorithm.One is in the decoding process,because SCL The decoding algorithm will generate multiple candidate list values,which makes the decoding delay and decoding complexity high.Secondly,in the process of cyclic redundancy,the metric value during decoding of some codewords will be too high due to operation errors.Therefore,for the above two problems,this paper improves the adaptive search path and path pruning and then performs cyclic redundancy,and theoretically verifies the rationality of the algorithm.Numerical simulations verify that the algorithm can significantly reduce the decoding complexity with little loss of bit error performance.(3)This paper designs and trains decoders based on MLP,CNN,RNN,DNN,LSTM,etc.on the deep learning machine platform in the laboratory,and studies the training parameters and network models required by each network.Model decoding performance,by comparing the experimental results,it is known that the CNN-based decoder can achieve the best performance with fewer training parameters.In view of this,this paper selects the CNN network as the decoding network.(4)Since this paper finds that when training the neural network,the training set will increase the difficulty of network training exponentially because the codeword information transmitted in the channel is too long,which makes the decoding delay longer.Therefore,this paper designs a polar code segmentation neural network decoding algorithm.According to the coding characteristics of polar codes,the long code is divided into modules,and then the neural network decoding and decoding are selected for coupling decoding.The CRC-SCL decoding algorithm couples each sub-block to complete the decoding.In this paper,a block neural network decoding algorithm for polar codes with code lengths of 32,64 and 128,candidate paths of 32,and code rate of 0.5 is implemented,and the bit error of the received code word at the long code is studied.The study found that when the signal-to-noise ratio is2.0d B,the probability of 3-bit and 4-bit codeword errors at the 4096 codeword of the long code is 0,and the probability of 2-bit codeword errors is 4.1%.The simulation experimental data shows that,The improved algorithm in this paper can reduce the decoding complexity,reduce the decoding delay,and improve the overall decoding performance at the expense of a small bit error rate.