基于深度学习的电磁信号调制识别方法  

Modulation and Identification Method of Electromagnetic Signal Based on Deep Learning

作　　者：李鹏程 程俊平 叶畅 李耿 Li Pengcheng;Cheng Junping;Ye Chang;Li Geng

机构地区：[1]解放军信息安全测评认证中心 [2]中国电子技术标准化研究院

出　　处：《安全与电磁兼容》2023年第1期34-40,共7页Safety & EMC

基　　金：综合计划J共性技术研究(H2208030303)。

摘　　要：电磁信号调制识别是电磁信息安全领域的重要技术基础。该文针对无线衰落造成电磁信号调制识别准确率低的问题,研究比较了基于深度学习的无线衰落信道电磁信号的调制识别方法。通过Matlab仿真生成同向正交(IQ)电磁信号数据,比较分析了AlexNet、VGGNet、ResNet和DenseNet四类神经网络模型的信号调制识别准确率,得到适合应用于无线衰落信道电磁信号调制识别的模型。结果表明,DenseNet神经网络对信号调制识别的准确率最好,达到82.10%。本研究为电磁信号调制识别在电磁信息安全等领域的应用提供重要参考。The identification of electromagnetic signal modulation is an important technical basis for electromagnetic information security.Aiming at the low accuracy of electromagnetic signal modulation identification due to wireless fading,this paper compares the modulation identification methods based on deep learning.The data of IQ electromagnetic signals are generated by Matlab simulation,and the modulation recognition accuracy of AlexNet,VGGNet,ResNet and DenseNet are compared and analyzed.The results show that DenseNet neural network has the best recognition accuracy of signal modulation,reaching 82.10%.This paper provides an important reference for the application of electromagnetic signal modulation identification in the field of electromagnetic information security.

关 键 词：调制识别 无线衰落信道 深度学习 卷积神经网络 

分 类 号：TN911.3[电子电信—通信与信息系统] TP18[电子电信—信息与通信工程]