Research On BCH Code Decoding Method Based On Deep Learning

With the development of the times,and the advent of 5G communication,the business requirements and quality assurance of communication transmission continue to advance,which put forward higher requirements on the quality and transmission speed of information in the process of communication data transmission.The BCH code,as an important part of the linear block code in the error correction code,has the advantages of strong error correction ability,convenient coding,simple structure,and excellent performance in the medium and short code length.However,the traditional decoding method has a high number of iterations,a long period extension,low throughput,and the inability to be parallel,which can no longer meet the needs of data transmission in today's era.Binary BCH codes are used as the standard in this paper,which mainly study the decoding method of BCH codes based on deep neural networks.The main work of this article includes:(1)A neural network multi-class decoding model and a neural network multi-label class decoding model are proposed in this paper based on the characteristics of the BCH code data set.These two models use the powerful function fitting ability of neural networks to classify BCH codes with noise,and process the decoding process of BCH as multi-classification tasks and multi-label tasks respectively.It is shown by simulation experiments that when the bit rate is low,the deep neural network multi-class decoding model is superior to the traditional BM iterative decoding algorithm.(2)When the code rate is high,the performance of the neural network classification decoding model does not reach the expected,which is not comparable to the performance of the traditional iterative decoding algorithm.In order to solve this problem,here combines a neural network and K-nearest neighbor algorithm,and proposes a new decoding algorithm DKNN,which is to give the neural network unpredictable samples to the K-nearest neighbor algorithm for processing.It is shown by simulation experiments that the DKNN algorithm completely surpasses the traditional iterative decoding algorithm when the bit rate is large,and has better performance.(3)The codeword's type of the information bit increases exponentially with the increase of the information bit.The increase in the calculation volume and storage space of the deep neural network leads to the difficulty of network learning,when the information bit of the BCH code increases.It is proposed by this paper to use integrated ideas to decode the BCH codes with long codewords to solve the BCH codes with long codewords.The purpose of integrated decoding is to train several sub-classifiers.In the decoding stage,the final decoding result,through collecting the output of the sub-classifier,is decided according to the voting statistics.It is shown by simulation experiments that when the BCH codeword is long,the integrated decoding algorithm can perform classification decoding at a relatively small cost,and has better decoding performance.