摘要
作为通信系统中基本的组成部分,信道编码使信息在传输的过程中能够抵抗信道的干扰。随着人工智能的发展,深度学习(Deep Learning,DL)越来越多地被用到通信领域解决实际问题。近年来,有学者提出将DL应用于端到端的信道编译码系统,并展现出了其良好的性能。现有基于DL的信道编译码方法在特定信噪比(Signal-to-Noise ratio,SNR)下训练网络模型,然而,在部署实际通信系统时,并不能保证信道条件和训练时是一致的,导致对于不同的SNR,需要存储大量模型。基于此,提出了自适应信道SNR的Turbo自编码器信道编译码系统,通过引入注意力机制感知信道变化,生成与信道条件相匹配的编码码字。仿真结果表明,该方法能够有效应对信道条件的变化,大幅降低设备端神经网络参数的存储开销。
Channel coding is a fundamental component in communication systems,which allows reliable transmission under noisy conditions.With the development of artificial Intelligence,deep learning(DL)based methods are exploited to solve practical problems in wireless communications.Recently,DL based channel coding and decoding systems show good performance.Existing DL-based channel coding and decoding systems train models at a specific signal-to-noise ratio(SNR).However,it is not guaranteed that the channel conditions are consistent with the trained models in real communication systems,which could lead to storing a lot of models for different SNRs.In this paper,we propose a Turbo autoencoder adapting to different channel SNRs for channel coding and decoding.The proposed method can sense channel changes by introducing attention mechanism and generate codewords that match channel conditions.Simulation results show that the proposed method can effectively adapt to the changes of channel conditions and significantly reduce the storage overhead of neural network parameters on device.
作者
胡启蕾
许佳龙
李伦
钟章队
艾渤
陈为
HU Qilei;XU Jialong;LI Lun;ZHONG Zhangdui;AI Bo;CHEN Wei(State Key Laboratory of Rail Traffic Control and Safety,Beijing Jiaotong University,Beijing 100044,China;ZTE Corporation,Shenzhen 518057,China;Key Laboratory of Railway Industry of Broadband Mobile Information Communications,Beijing 100044,China;Frontiers Science Center for Smart High-speed Railway System,Beijing 100044,China;Beijing Engineering Research Center of High-speed Railway Broadband Mobile Communications,Beijing 100044,China)
出处
《无线电通信技术》
2022年第4期680-688,共9页
Radio Communications Technology
基金
国家重点研发计划(2018YFE0207600)
国家自然科学基金(62122012)
北京市自然科学基金(L211012,L202019)。
关键词
信道编译码
深度学习
注意力机制
channel coding and decoding
deep learning
attention mechanism