Research On Decoding Algorithm Of Polar Code Based On Deep Learning

Based on the channel polarization,Arikan proposed a new channel coding,called Polar code.It is theoretically proved that Polar code can achieve the channel capacity of binary discrete memory-less channel,and has low coding and decoding complexity.Deep learning is a machine learning method with multi-level nonlinear transformation.As a common network structure,the convolutional neural network(CNN)has the characteristics of weight sharing,local connection and so on,which effectively reduces the complexity of the network,and has strong robustness and fault tolerance ability,and is easy to be trained and optimized.This thesis studies the decoding algorithm of Polar code based on deep learning.The main works are as follows:First,to overcome the poor bit error rate(BER)performance of the successive cancellation(SC)decoding algorithm of polar codes,and the high delay of the improved successive cancellation list(SCL)decoding algorithm,a disturbed Polar code decoding algorithm is proposed based on convolutional neural network.In the proposed algorithm,more candidate codewords are generated by adding the secondary independent random noise,which is produced from the designed CNN,when the received signal is failed to decode.Numerical simulation results show that the proposed Polar code decoding algorithm has improved the BER performance without increasing the decoding complexity,in compared with the original SC decoding algorithm.At the same time,the proposed Polar code decoding algorithm has effectively reduced the decoding delay with a little BER performance loss,in compared with the SCL decoding algorithm.Secondly,to overcome the serious degradation of the performance of Polar code in correlated noise channel,a decoding algorithm based on PD-CNN(prior decision)is proposed.The algorithm introduces CNN into the decoding process to estimate the correlated noise of the channel.At the same time,the algorithm optimizes the input of the neural network by using the prior prediction,and makes the residual noise of the channel obey the Gaussian distribution by setting an effective loss function,so as to whiten the color noise of the channel,and the proposed algorihtm gives a new calculation method of likelihood information to improve the performance of Polar code to a certain extent.Numerical simulation results show that the proposed Polar code decoding algorithm has have better BER performance,in compared with the traditional Polar decoding algorithm under the same conditions;At the same time,the proposed Polar code decoding algorithm has effectively reduced the decoding delay under the same BER performance,in compared with the BP-CNNdecoding algorithm.