摘要
针对信号调制识别对复杂通信环境缺乏适应性与精度不足的问题,提出一种基于深度学习的多特征复合神经网络框架.该框架首先使用前端卷积神经网络检测信号载波特征,再对前端初筛选信号执行预处理将其转换为信号时频图,最后设计了后端轻量化卷积神经网络,检测信号时频特征.基于TensorFlow平台的复合神经网络对机场真实信号检测精度达到99.23%,实验表明该方法可有效应用于实时机场信号检测.
In order to increase the generality and accuracy of radio modulation recognition in complex radio propagation environment,a multiple feature combined convolutional network system based on deep learning is proposed.Carrier features were detected with front convolutional network in the first stage.Then,the signal filtered by the front CNN was converted into spectrograms with the proposed pre-process method.Finally,the lightweight backend convolutional network was designed to extract the time-frequency features of spectrograms.The networks,which run on TensorFlow,achieved 99.23%accuracy with real airport communication signals.The experiment indicates that the proposed networks could be applied in real-time airport radio detection.
作者
侯进
吕志良
徐茂
吴佩军
刘雨灵
张笑语
陈曾
HOU Jin;Lü Zhiliang;XU Mao;WU Peijun;LIU Yuling;ZHANG Xiaoyu;CHENG Zeng(College of Information Science & Technology,Southwest Jiaotong University Chengdu 610031,China;Chengdu Hua Ri Communication Technology Co.Ltd.,Chengdu 610045,China)
出处
《西南交通大学学报》
EI
CSCD
北大核心
2019年第4期863-869,878,共8页
Journal of Southwest Jiaotong University
基金
浙江大学CAD&CG国家重点实验室开放课题(A1923)
成都市科技项目(2015-HM01-00050-SF)
关键词
调制识别
卷积神经网络
深度学习
载波特征
时频特征
modulation recognition
convolutional neural network
deep learning
carrier features
timefrequency features
