多元LDPC编码调制系统CNN辅助迭代检测译码算法  

CNN-aided Iterative Detection and Decoding for Nonbinary LDPC Coded Modulation Systems

作　　者：万飞 白宝明[1] 朱敏 WAN Fei;BAI Baoming;ZHU Min(State Key Laboratory of Integrated Services Networks,Xidian University,Xi’an 710071,China)

机构地区：[1]西安电子科技大学综合业务网理论及关键技术国家重点实验室,陕西西安710071

出　　处：《无线电通信技术》2022年第4期673-679,共7页Radio Communications Technology

基　　金：国家自然科学基金面上项目(62171356);中国电子科技集团公司第五十四研究所合作课题(SKX212010038);陕西省重点研发计划(2020ZDLGY05-04)。

摘　　要：针对相关噪声信道,提出了一种适用于多元LDPC码的深度学习辅助译码算法,将卷积神经网络(Convolution Neural Network,CNN)引入到基于硬信息的迭代大数逻辑算法中,以对抗相关信道噪声影响。在CNN与硬判决译码器之间的迭代能够减弱噪声对编码调制系统的影响,使得译码器可以获得更为准确的估计。为了充分发挥CNN的能力,对于经过高阶调制的复数形式信号数据,通过预处理转化为实数形式。仿真结果表明,相较于已有研究中提出的联合迭代检测译码结构,GF(64)域上的(42,21)多元LDPC码经64-QAM调制传输,CNN方案能够获得最高1 dB的性能增益,验证了其在抵抗相关信道噪声的有效性。In this paper,we propose a deep learning assisted iterative detection and decoding method for nonbinary LDPC(NB-LDPC)coded modulation systems under correlated channel noise,in which the convolution neural network(CNN)is incorporated into iterative hard-reliability based majority logic algorithms.The CNN-based architecture is used to resist the correlated channel noise.And iterative processing between CNN and hard-decision decoder alleviates the effect of the noise,thus the decoder can obtain accurate estimation.To make full use of the CNN,we adjust it and transform complex-valued dataset into real-valued one.Simulation results show that the proposed scheme can achieve improvement up to 1 dB compared with the original iterative joint detection decoding algorithm for(42,21)NB-LDPC code over GF(64)transmitted with 64-QAM signaling.It indicates that our proposed CNN has the ability to recover the processed data.

关 键 词：多元LDPC码 卷积神经网络 迭代检测译码算法 相关噪声 

分 类 号：TN919[电子电信—通信与信息系统]