Research On Rate-Compatible Polar Code Construction And Decoding Based On Deep Learning

As an important part of modern digital communication,channel coding inevitably receives people's attention.With the proposed Shannon theory,various encoding innovation,we try to get close to the Shannon limit.As the only channel code that polar codes have been theoretically proven to achieve the Shannon limit,it has attracted the attention of researchers.Polar codes use channel polarization theory to divide channels into full-noise sub-channels and no-noise sub-channels.Polar codes transmit information bits and frozen bits into no-noise sub-channels and full-noise sub-channels,respectively.However,due to the shortcomings of the polar code encoding structure,the code length cannot be flexibly changed,and polar code are difficult to apply to real scenes.In addition,the existing belief propagation decoding algorithm is more suitable for systems with high parallelism and lower latency than common serial cancellation decoding algorithm.However,due to the high number of decoding iterations and high complexity,it is also difficult to apply in practice.Therefore,in response to these issues,the main work and innovations of this article are as follows:(1)This paper is dedicated to improving the existing punctured polar codes,improving performance,and reducing the loss of channel reliability.First of all,this article introduces the polarization principle,coding method,construction method and decoding algorithm of the polarization code.Simultaneously,the two punctured algorithms with lower complexity are simulated and analyzed,and their advantages and disadvantages are explained through comparison.Subsequently,this article introduces the traditional belief propagation decoding algorithm and the existing belief propagation algorithm based on deep neural network and the belief propagation algorithm based on recurrent neural network in detail,expounding its advantages and disadvantages.(2)This paper proposed a rate-compatible non-systematic polarization code based on feedback bits,which reduces the puncturing of channels with higher reliability by selecting puncturing bits from the top channel.At the same time,the feedback bit is selected by the algorithm to fill in the feedback value;in addition,a simple method is proposed to calculate the feedback value,so that the method can be applied in real scenes.Through the two-step encoding method,the encoding rate is changed to obtain better performance.(3)This paper verifies through simulation that the error rate performance of the system polarization code is better than that of the non-system polar code,so the non-system polar code shortening algorithm based on feedback bits is applied to the system polarization code.The rate-compatible system polar code based on feedback bits has better error rate performance than the rate-compatible non-systematic polar code based on feedback bits,and the complexity is only slightly increased compared with the system code.(4)In order to further improve the performance of polar code decoding,this paper uses deep learning to improve the traditional belief propagation decoding algorithm.Compared with the existing belief propagation algorithm based on deep learning,the belief propagation decoding algorithm based on residual neural network proposed in this paper uses the principle of residual,which greatly reduces the large amount of memory loss that needs to be used in the neural network.And compared with the traditional belief propagation decoding algorithm,its decoding performance has also been greatly improved.(5)In this work,the performance simulation and complexity analysis of the proposed rate-compatible polarization code schemes have been carried out.The results show that the improved algorithm proposed has a greater improvement compared with the existing puncturing algorithm,and the error rate and in terms of frame error rate,its performance is also better than that of unpunctured mother polar code.Subsequently,this paper conducts simulation analysis and complexity comparison of the belief propagation decoding algorithm based on residual neural network.The results show that the scheme proposed in this paper only needs to train two weights,which can effectively reduce the memory loss.