
2020 年 9 月 Journal on Communications September 2020
第 41 卷第 9 期 通 信 学 报 Vol.41
No.9
分组马尔可夫叠加传输的神经网络译码
王千帆
1
,毕胜
2,3
,陈曾喆
4
,陈立
4
,马啸
2,3
(1. 中山大学电子与通信工程学院,广东 广州 510006;2. 中山大学数据科学与计算机学院,广东 广州 510006;
3. 中山大学广东省信息安全重点实验室,广东 广州 510006;4. 中山大学电子与信息工程学院,广东 广州 5100063)
摘 要:研究了分组马尔可夫叠加传输的神经网络(NN)译码方案。利用 NN,实现了不同网络结构、数据表征
形式的基本码译码器。在此基础上,将所实现的基本码译码器嵌入迭代译码机制中,提出了基于 NN 的分组马尔
可夫叠加传输的滑窗译码算法,并分析了其对应的性能下界。所提出的译码算法提供了一种将 NN 运用到长码译
码的解决思路,即用 NN 替代译码中的部分模块。仿真结果表明,利用 NN 实现的基本码译码器可以达到最大似
然译码性能。基于 NN 的分组马尔可夫叠加传输的滑窗译码算法性能在中高信噪比区域与对应精灵辅助下界贴合,
获得了额外的编码增益。
关键词:分组马尔可夫叠加传输;精灵辅助下界;神经网络;滑窗译码
中图分类号:TN92
文献标识码:A
doi: 10.11959/j.issn.1000−436x.2020158
Neural network decoding of the block Markov
superposition transmission
WANG Qianfan
1
, BI Sheng
2,3
, CHEN Zengzhe
4
, CHEN Li
4
, MA Xiao
2,3
1. School of Electronics and Communication Engineering, Sun Yat-sen University, Guangzhou 510006, China
2. School of Data and Computer Science, Sun Yat-sen University, Guangzhou 510006, China
3. Guangdong Key Laboratory of Information Security Technology, Sun Yat-sen University, Guangzhou 510006, China
4. School of Electronics and Information Technology, Sun Yat-sen University, Guangzhou 510006, China
Abstract: A neural network (NN)-based decoding algorithm of block Markov superposition transmission (BMST) was
researched. The decoders of the basic code with different network structures and representations of training data were im-
plemented using NN. Integrating the NN-based decoder of the basic code in an iterative manner, a sliding window de-
coding algorithm was presented. To analyze the bit error rate (BER) performance, the genie-aided (GA) lower bounds
were presented. The NN-based decoding algorithm of the BMST provides a possible way to apply NN to decode long
codes. That means the part of the conventional decoder could be replaced by the NN. Numerical results show that the
NN-based decoder of basic code can achieve the BER performance of the maximum likelihood (ML) decoder. For the
BMST codes, BER performance of the NN-based decoding algorithm matches well with the GA lower bound and exhi-
bits an extra coding gain.
Key words: BMST, GA lower bound, NN, sliding window decoding
收稿日期:2020−03−24;修回日期:2020−06−30
通信作者:马啸,maxiao@mail.sysu.edu.cn
基金项目:国家自然科学基金资助项目(No.61971454,No.61671486);广东省基础与应用基础研究基金资助项目(No.2020A1515010687);
广东省自然科学基金资助项目(No.2016A030308008)
Foundation Items: The National Natural Science Foundation of China (No.61971454, No.61671486), The Basic and Applied Basic
Research Foundation of Guangdong (No.2020A1515010687), The Natural Science Foundation of Guangdong Province
(No.2016A030308008)