摘要
阐述端到端通信系统通过联合训练优化,能够提升分类任务准确率,降低传输带宽需求。研究中考虑了两种信道模型,加性高斯白噪声(AWGN)信道和慢衰落信道。在CIFAR-10数据集上对通信系统进行了仿真评估,结果表明该系统能够自适应选择最佳的编码和调制策略,保证传输质量的同时节约传输资源。同时,与传统数字通信系统相比,该通信系统避免了“悬崖效应”,在大范围SNR波动下表现良好,并具有强鲁棒性。在相同带宽条件下,联合优化的通信系统相比信源信道分离编码的传统通信系统准确率提高了30%。
This paper describes how end-to-end communication systems can improve classification task accuracy and reduce transmission bandwidth requirements through joint training optimization.Two channel models were considered in the study,the additive Gaussian white noise(AWGN)channel and the slow fading channel.The simulation evaluation of the communication system was conducted on the CIFAR-10 dataset,and the results showed that the system can adaptively select the best encoding and modulation strategies,ensuring transmission quality while saving transmission resources.Meanwhile,compared with traditional digital communication systems,this communication system avoids the"cliff effect",performs well under large-scale SNR fluctuations,and has strong robustness.Under the same bandwidth conditions,the jointly optimized communication system has improved accuracy by 30% compared to the traditional communication system using source channel separation coding.
作者
张汉卿
刘建林
汤珊
杜逢亮
ZHANG Hanqing;LIU Jianlin;TANG Shan;DU Fengliang(China Electronics Technology Group Corporation 54th Research Institute,Hebei 050021,China;AVIC Import and Export Co.,Ltd.,Beijing 100010,China)
出处
《电子技术(上海)》
2024年第6期10-12,共3页
Electronic Technology
关键词
端到端通信系统
深度学习
联合信源-信道编码
码率控制
图像分类任务
end-to-end communication system
deep learning
joint source channel coding
rate control
image classification task