摘 要：文章通过深度强化学习的方法来寻求二进制线性编码的有效解码策略。在加性高斯白噪声的条件下，将置信传播（BP）解码算法中软信息的迭代看作是对软信息的连续决策，并将其映射到马尔可夫决策过程，用深度强化学习网络代替传统译码器，扩大探索空间以提高译码性能，从而实现对数据驱动的最佳决策策略的学习。结果表明，相较于传统 BP 解码器，在误码率＝10-5时，学习型BP解码器在BCH码上取得大约0.75 dB的优势，这在一定程度上解决了以往研究中过于依赖数据的问题。
关键词：深度强化学习；置信传播译码；马尔可夫决策；最佳决策
DOI:10.19850/j.cnki.2096-4706.2021.21.025
中图分类号：TP18 文献标识码：A 文章编号：2096-4706（2021）21-0098-05
Belief Propagation Decoding Algorithm Based on Deep Reinforcement Learning
GAO Yuanhao1, LIU Naijin1, LU Yuanming2
(1.Qian Xuesen Laboratory of Space Technology, China Academy of Space Technology Beijing 100090, China; 2.63893 Troops of PLA, Luoyang 471000, China)
Abstracts: This paper uses a deep reinforcement learning approach to find an efficient decoding strategy for binary linear codes. Under the condition of additive Gaussian white noise, the iteration of soft information in the belief propagation (BP) decoding algorithm is regarded as a continuous decision-making of soft information, which is mapped to the Markov decision-making process. The deep reinforcement learning network is used to replace the traditional decoder, expand the exploration space to improve the decoding performance, so as to realize the learning of the best data-driven decision-making strategy. The results show that compared with the traditional BP decoder, when the bit error rate is 10-5, the learning BP decoder has an advantage of about 0.75 dB in BCH code, which solves the problem of relying too much on data in previous research to a certain extent.
Keywords: deep reinforcement learning; belief propagation decoding; Markov decision-making; best decision-making
参考文献：
[1] GRUBER T，CAMMERER S，HOYDIS J，et al. On deep learning-based channel decoding [C]//2017 51st Annual Conference on Information Sciences and Systems (CISS).Baltimore：IEEE，2017：1-6.
[2] O’SHEA T，HOYDIS J. An introduction to deep learning for the physical layer [J].IEEE Transactions on Cognitive Communications and Networking，2017，3（4）：563-575.
[3] NACHMANI E，BE’ERY Y，BURSHTEIN D. Learning to decode linear codes using deep learning [C]// 2016 54th Annual Allerton Conference on Communication, Control, and Computing (Allerton).Monticello：IEEE，2016：341-346.
[4] BENNATAN A，CHOUKROUN Y，KISILEV P. Deep learning for decoding of linear codes-a syndrome-based approach [C]//2018 IEEE International Symposium on Information Theory (ISIT).Vail：IEEE，2018：1595-1599.
[5] NACHMANI E，MARCIANO E，LUGOSCH L，et al. Deep learning methods for improved decoding of linear codes [J].IEEE Journal of Selected Topics in Signal Processing 2018，12（1）：119-131.
[6] NACHMANI E，BE’ERY Y，Burshtein D. Learning to decode linear codes using deep learning [C]//2016 54th Annual Allerton Conference on Communication, Control, and Computing （Allerton）.Monticello：IEEE，2016：341-346.
[7] LIANG F，SHEN C，WU F. An iterative BP-CNN architecture for channel decoding [J]. IEEE Journal of Selected Topics in Signal Processing，2018，12（1）：144-159.
[8] Wang X B，Zhang H Z，Li R，et al. Learning to flip successive cancellation decoding of polar codes with LSTM networks [C]//2019 IEEE 30th Annual International Symposium on Personal,Indoor and Mobile Radio Communications （PIMRC）. Istanbul：IEEE，2019：1-5.
[9] CARPI F，HÄGER C，MARTALÒ M，et al. Reinforcement learning for channel coding: Learned bit-flipping decoding [C]//2019 57th Annual Allerton Conference on Communication, Control, and Computing （Allerton）. Monticello：IEEE，2019：922-929.
[10] Ryan W，Lin S. Channel codes: classical and modern [M].Cambridge university press，2009.
[11] BOSSERT M，HERGERT F. Hard-and soft-decision decoding beyond the half minimum distance---An algorithm for linear codes （Corresp.） [J]. IEEE transactions on information theory，1986，32（5）：709-714.
[12] KOU Y，LIN S，FOSSORIER M P C. Low-density paritycheck codes based on finite geometries: a rediscovery and new results [J].IEEE Transactions on Information theory，2001，47（7）：2711-2736.
[13] Zhang J T，FOSSORIER M P C. A modified weighted bit-flipping decoding of low-density parity-check codes [J].IEEE Communications Letters，2004，8（3）：165-167.
[14] JIANG M，ZHAO C M，SHI Z H，et al. An improvement on the modified weighted bit flipping decoding algorithm for LDPC codes [J].IEEE Communications Letters，2005，9（9）：814-816.
作者简介：高源浩（1997—），男，汉族，重庆铜梁人，硕士研究生在读，研究方向：基于强化学习的线性分组码译码方法。