基于前向多层神经网络的分组码译码器设计  

Design of Block Code Decoders Based onFeedforward Multilayer Neural Network

作　　者：马晓敏[1] 杨义先[1] 章照止[2] 

机构地区：[1]北京邮电大学 [2]中国科学院系统科学研究所

出　　处：《通信学报》1999年第6期1-7,共7页Journal on Communications

基　　金：国家自然科学基金

摘　　要：把最大相关译码与神经网络神经元的内积特性及吸引域有机地联系起来，连接权决定译码码字，阈值设定决定神经元的纠错范围，从而形成一种可用于硬判决及软判决译码的神经译码器，并在理论上证明了此译码器可在ＤＭＣ信道的纠错能力范围内实现零错误概率硬判决译码，也可实现与最小欧几里德距离译码相当的软判决译码，并能在检错范围内检错。A neural network decoder is constructed for both hard decision decoding and soft decision decoding of block codes,which relate the maximum correlation decoding with characteristics of inner product and atractor of a neuron.Code words of decoding are dependent on connection weight of the neural network,and correction number of error is dependent on threshold of the neural network.This article has proved theoretically that the proposed neural decoder can implement hard desicion decoding with zero error probability within error correction capability of the decoded code,and implement soft decision decoding corresponding to the minimum Euclidian distance decoding.

关 键 词：神经网络 译码 硬判决 软判决 纠错码技术 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TN911.22[自动化与计算机技术—控制科学与工程]