摘要
调制方式自动识别(AMR)基于深度学习(DL)的方法 AMR-DL是当前信号调制方式识别领域的研究热点。然而,低信噪比(SNR)条件下的调制信号以及深度学习网络参数量过大会显著降低AMR-DL的识别准确率和计算效率。针对这两个关键问题,提出一种新型AMR-DL算法。新算法利用多观测样本累积方法提高信噪比,对增强后的信号进行小波变换,得到时频图像,将调制识别问题转换为时频图像的分类问题;同时设计了一个轻量级神经网络,用于时频图像特征提取与分类。仿真结果显示,所提新型AMR-DL算法中轻量级神经网络参数量较少,在0 dB下准确率可达98.1%,且与其他算法相比在低信噪比条件下显著提高了调制信号的识别准确率。
Automatic modulation recognition based on deep learning(AMR-DL)is a research hotspot in the field of signal modulation recognition.The modulated signals with low signal-to-noise ratio(SNR)and too many parameters of deep learning network can significantly reduce the recognition accuracy and calculation efficiency of AMR-DL.A new AMR-DL algorithm is proposed to address these two key issues.In the new algorithm,the multi-observation sample accumulation method is used to increase SNR.The enhanced signal is transformed into a time-frequency image by means of wavelet transform,transforming the signal modulation recognition problem into the classification problem of time-frequency image.A lightweight neural network is designed to extract the features extraction and classification of wavelet time-frequency images.The simulation results show that the proposed new-type AMR-DL algorithm has fewer lightweight neural network parameters and can achieve an accuracy of 98.1%at 0 dB.In comparison with other algorithms,the recognition accuracy of modulated signals can be improved significantly under low SNR conditions.
作者
张丹华
冯冀宁
ZHANG Danhua;FENG Jining(School of Computer and Cyberspace Security,Hebei Normal University,Shijiazhuang 050024,China;Hebei Key Laboratory of Network and Information Security,Hebei Normal University,Shijiazhuang 050024,China;Supply Chain Big Data Analysis and Hebei Engineering Research Center for Data Security,Hebei Normal University,Shijiazhuang 050024,China)
出处
《现代电子技术》
北大核心
2025年第12期179-186,共8页
Modern Electronics Technique
关键词
深度学习
调制方式自动识别
多观测样本累积
轻量级神经网络
小波变换
低信噪比
deep learning
automatic modulation recognition
multiple observation sample accumulation
lightweight neural network
wavelet transform
low signal-to-noise ratio
