Research On Polar Codes Decoding Algorithm Base On Deep Learning Over Correlated Noise And Fading Channels

In the channel coding scheme,the coding complexity of polar codes is low,which is of great research value.In the actual communication scenario,the channel will be fading and superimposed noise interference,but the performance of traditional polar code decoding is poor under its influence.Deep learning has powerful computational advantages.The convolutional neural network with lower complexity can reconstruct the associated noise and gain on the actual channel,thus eliminating their negative impact on the performance of traditional decoder to a certain extent.In this paper,the polar code decoding algorithm based on deep learning is studied under the channel with fading and noise correlation.The specific work is summarized as follows:(1)There is a polar code decoder based on the deep learning model of multilayer perceptron,CNN,recurrent neural network and short and long time memory network is designed over the flat fading channel.The effects of different network designs and parameter settings on the performance of the neural network decoder are compared with that of the successive cancelation decoder.The single neural network decoder can not improve the performance of the traditional polar code decoder in flat fading channel,but can reduce its decoding delay.(2)A polar code decoder with convolutional neural network cascading belief propagation decoder is proposed in the correlative noise channel.The CNN is responsible for predicting the original correlation noise,so as to reduce the influence of correlation noise on the performance of the traditional polar code decoder.The experiment focuses on the different correlation coefficients of noise in order to find out the influence of the coefficients on the performance of decoder.Experimental results show that show that the traditional successive cancelation decoding algorithm is not resistant to the strong correlation of channel noise,however,the polar code decoder based on deep learning can improve the performance.When the noise correlation coefficient is 0.8,the maximum performance gain of the proposed decoder is about 2d B.(3)A polar code decoder with double convolutional neural network cascading belief propagation algorithm under fast fading correlative noise channels is proposed.Two convolutional neural networks predict correlation noise and correlation channel gain,respectively,in order to reduce their influence on the performance of traditional polar code decoder.Experimental results show that the larger the correlation coefficient is,the greater the performance gain of the proposed decoder will be compared with the traditional confidence propagation decoder,when the correlation coefficient between noise and channel gain is 0.9,the proposed decoder can achieve a performance gain of 5-6 dB.