Construction of Protograph LDPC Codes Based on the Convolution Neural Network

作     者：Zhiyuan Xiao Liguang Li Jin Xu Jin Sha Zhiyuan Xiao;Liguang Li;Jin Xu;Jin Sha

作者机构：School of Electronic Science and EngineeringNanjing UniversityNanjing 210023China Institute of Wireless ResearchZTE CorporationShenzhen 518055China 

出 版 物：《China Communications》 (中国通信（英文版）)

年 卷 期：2023年第20卷第5期

页      面：84-92页

核心收录：

学科分类：0810[工学-信息与通信工程] 07[理学] 08[工学] 081104[工学-模式识别与智能系统] 070104[理学-应用数学] 0811[工学-控制科学与工程] 0701[理学-数学] 

基　　金：supported in part with the Project on the Industry Key Technologies of Jiangsu Province(No.BE2017153) the Industry-University-Research Fund of ZTE Corporation. 

主　　题：LDPC codes protograph codes iterative decoding threshold neural network 

摘      要：This paper presents an intelligent protograph construction algorithm.Protograph LDPC codes have shown excellent error correction performance and play an important role in wireless communications.Random search or manual construction are often used to obtain a good protograph,but the efficiency is not high enough and many experience and skills are needed.In this paper,a fast searching algorithm is proposed using the convolution neural network to predict the iterative decoding thresholds of protograph LDPC codes effectively.A special input data transformation rule is applied to provide stronger generalization ability.The proposed algorithm converges faster than other algorithms.The iterative decoding threshold of the constructed protograph surpasses greedy algorithm and random search by about 0.53 dB and 0.93 dB respectively under 100 times of density evolution.Simulation results show that quasi-cyclic LDPC(QC-LDPC)codes constructed from the proposed algorithm have competitive performance compared to other papers.