Research On Polarization Code Decoding Based On Temporal Convolutional Network

Polar code is a linear block code based on the channel polarization phenomenon proposed by Arikan.In the channel coding theories,the coding and decoding complexity is low,and it has the ability to approach the Shannon limit.The traditional polar code decoding method is Successive Cancellation(SC)decoding,which has low complexity but high delay.Deep Learning Technology can learn data characteristics in a complex network environment,and has powerful flexibility and computational advantages.This dissertation uses Temporal Convolutional Network(TCN)to handle the problem of polar code decoding.The main work is as follows:1.This dissertation has introduced the coding principle and algorithm of polar codes,analyzed the process of channel combination and classification in channel polarization,discussed the concept of Bhattacharyya parameters and the construction of generating matrix,and explained the butterfly algorithm and complexity analysis of SC decoding in detail.In addition,the improved decoding scheme based on SC decoding and its decoding ideas and the construction model of artificial neural network are also introduced.2.A TCN-based polar code decoding algorithm is proposed.This dissertation has analyzed the theories of TCN,discussed the expansion of the Causal Convolution and Residual Structure of the core algorithm of TCN,and then researched and built a deep learning model that uses the TCN model to decode polar codes.The proposed algorithm is simulated under the number,input SNR,training batch,residual block number,convolution kernel lengths,convolution kernels and training sequence number,and the factors that can further improve TCN decoding performance under AWGN channel and Rayleigh fading channel are also explored.3.The proposed algorithm has been put into operation in actual scenarios.By using ASCII code as a tool carrier and different TCN parameter training to build different deep learning models can verify the effectiveness of appropriate parameters to reduce the bit error rate in the text information transmission system based on the AWGN channel.