Design And Analysis Of Polar Code Decoder Based On Neural Network

In recent years,deep learning based on neural networks has made dramatic progress in many fields such as computer vision,games,and bioinformatics.This paper attempts to introduce the neural network into the polar code decoder and explore the engineering application of the neural network in decoding technology.1.A new type of decoder structure is designed,which is to introduce the general neural network structure in the project into a polar code decoder.This decoder can directly realize the functions of demodulation and decoding in traditional digital communication systems.Then based on the complex-valued Multilayer Perceptron(MLP)and Complexvalued Convolutional Neural Network(CV-CNN),two corresponding decoders are designed,and the performance of the decoder is analyzed in two different formats: BPSK and QAM.The results show that the complex-valued neural network as a decoder can achieve the decoding performance of the maximum likelihood method,but this decoder only has a short code(<128 bits),and the performance is better.2.We propose a neural network decoding algorithm based on Belief Propagation(BP).This neural network is essentially based on the traditional BP decoding algorithm,and its neurons are parameterized processing elements(PPE).The neural network decoding algorithm is named BP-PPE,and then we use deep learning algorithm to optimize the parameters of BP-PPE.The results show that the performance and complexity of BP-PPE are superior to traditional BP and SC algorithms.