Research And Application On Decoding Algorithms Of Polar Codes With Deep Learning

Polar Codes are the new type of channel coding scheme that can be proved to reach channel capacity,which has attracted widespread attention and research from the academic community.The excellent performance of Polar Codes enables it to be a strong candidate for the 5G communication standard.However,the decoding algorithm of the Polar Codes still needs to be further improved.In recent years,deep learning technology has become increasingly mature.Deep neural network is a powerful tool for classification problems and has achieved great success in various area.However,the channel decoding problem of Polar Codes is also essentially a classification problem,which can be tackled by Deep neural network.The deep neural network decoder(NND)is proposed with the advantages of low-latency high-throughput and one-shot property,but it can only be applied for ultra-short code length.Based on this background,the paper makes research on the optimization of the network structure and system framework of the NND.The relevant research contents are summarized as follows.The conventional deep Neural Network Decoder(NND)usually suffers the computational mismatch and the lack of generalization capability.In this paper,we propose a novel Graph-Neural-Network Decoder for the popular Polar Codes,which is directly constructed based on the regular decoding graph of Polar Codes and by replacing each of its basic 2-by-2 polarization elements with a simple Multi-Layer-Perceptron(MLP)based processing cell.The resultant decoder,namely PC-GNND,is thus endowed with the ability to infer over the skeleton of the decoding graph and has a greatly improved generalization capability.Simulation results show that,the proposed PC-GNND is capable of learning the exact code structure as well as the channel noise very efficiently with only a tiny fraction of the entire codebook and achieving better performance than that of conventional NNDs with far less parameters and training epochs.Moreover,the PC-GNND trained in a particular block length can be scaled to another one for different block lengths after fine tuning,which significantly reduces the computational cost of training.In this paper,a novel LSTM neural network decoder based on codeword's learning is proposed.The conventional neural network decoder system framework is based on information end,which requires the neural network to learn both the noise characteristics and the code structure mapping.However,the code structure is essentially a N-bit parity problem,which is difficult for a neural network to learn effectively.By changing the goals of the neural network decoder,a neural network decoder system framework based on codeword's learning is proposed to avoid the neural network passively learning the code structure mapping,and the LSTM neural network is used for implementation due to its strong generalization ability.The results of simulation show that the generalization capability of the proposed NND has been greatly improved.Only a small ratio of the codebook is required for the training of the proposed NND to learn the distribution of the codewords,and the decoding performance is close to MAP decoding algorithm.In addition,we also find that the codebook ratio boundary of the training set and the existence of a "chaos area",which can be the guidance of the generation of the training set.