Research And Analysis Of Fountain Codes Encoding And Decoding Performance

In face of the rapid increase of data and complex and changeable transmission environment,the conventional fixed rate channel coding and error control transmission scheme have gradually shown limitations due to the poor adaptability,large delay and feedback information congestion.The fountain codes make up for the shortcomings of the conventional transmission schemes by its rateless characteristic and perfect adaptability and have shown excellent performance in many fields.Fountain Codes,represented by LT Codes and Raptor Codes constructed by concatenating LT Codes with LDPC Codes,have been sophisticated in erasure channels and adopted by multiple international standards.Due to its rateless characteristic,the research on fountain codes based on LT Codes for noise channels has also attracted more and more attention,but low decoding efficiency and high error floor limit the application of fountain codes in noise channels.The conventional Belief Propagation(BP)decoding method and the random encoding method at the encoder are the two key factors for the poor decoding performance and the high error floor of the fountain codes in noise channels.This article studies this problem,and the main research contents are as follows,(1)Residual errors of BP are key factors of poor decoding performance and high error floor.A Cyclic Redundancy Check Aided Enhanced BP(CA-EBP)decoding method is proposed to find and correct the residual errors of BP.The characteristic of error patterns in decoding error frames and the relationship between error bits and their Log-likelihood Ratio(LLR)values generated by BP decoder are investigated.The bits with extremely high error probability are localized based on their LLR values.With the help of CRC detector,the residual error bits of BP are identified and corrected by bit flip.(2)The random encoding method can cause the low-degree variable nodes.Because it is difficult to get enough LLR values in the iterative decoding process,so the low-degree variable nodes have large error probability,which is another key factor for the high error floor.In order to solve this problem,the Regularized Variable-Node(RVN)encoding method is adopted at the encoder to avoid some low-degree variable nodes.At the decoder,the CA-EBP is used to correct the residual errors of BP and further reduce the error floor.