脉冲信道下基于深度学习的BP译码方法  

BP decoding method based on deep learning in impulsive channels

作　　者：潘睿 袁磊[1] PAN Rui;YUAN Lei(School of Information Science & Engineering, Lanzhou University, Lanzhou 730000, China)

机构地区：[1]兰州大学信息科学与工程学院,甘肃兰州730000

出　　处：《系统工程与电子技术》2020年第9期2116-2122,共7页Systems Engineering and Electronics

基　　金：国家重点研发计划项目(2017YFE0118900);甘肃省自然科学基金(18JR3RA268)资助课题。

摘　　要：在脉冲信道下,针对置信传播(belief propagation,BP)算法对短码长的低密度奇偶校验(low-density parity-check,LDPC)码译码性能差的问题,提出了一种基于深度学习的BP译码方法。首先,根据Tanner图构建两种深度神经网络模型,通过对Tanner图中边的权重重新赋值来提升译码性能。然后,简化信道对数似然比(log-likelihood ratio,LLR)的计算方法,通过模型训练优化近似计算参数,得到对参数γ鲁棒的译码模型。最后,构造鲁棒训练集,训练得到对参数α和γ鲁棒的译码模型。仿真结果表明,在高码率时,该方法相对于传统BP译码算法性能显著提升,且在近似计算信道LLR值时,译码性能在不同参数的脉冲信道下均具有鲁棒性。Aiming at the poor decoding performance of the belief propagation(BP)algorithm for low-density parity-check(LDPC)codes with short block lengths,a BP decoding method based on deep learning is presented in impulsive channels.Firstly,two deep neural network models are constructed by the Tanner graph,and the weights of edges in the Tanner graph are reassigned to improve the decoding performance.Then,the calculation method of the log-likelihood ratio(LLR)of the channel is simplified,the model is trained to optimize approximate calculation parameters,and the decoding model robust to parameterγis obtained.Finally,a robust training set is constructed to obtain the decoding model robust to parameterαandγ.Simulation results show that,at a high code rate,the performance of the proposed method is significantly improved compared with traditional BP algorithms.Furthermore,when approximately calculating the LLR value of the channel,the decoding performance is robust in impulsive channels with different parameters.

关 键 词：深度学习 脉冲信道 信道译码 低密度奇偶校验码 

分 类 号：TN911.7[电子电信—通信与信息系统]