融合深度学习和栈式自编码算法的异常无线电信号监测方法  

A Radio Anomaly Monitoring Method Based on Deep Learning and Stacked Autoencoder Algorithm

作　　者：魏小忠 申浩[2] 刘红杰 WEI Xiaozhong;SHEN Hao;LIU Hongjie(Jiangxi Radio Monitoring Station,Nanchang 330000,China;Yingtan Radio Administration Bureau of Jiangxi Province,Yingtan 335000,China;Borsche Technologies Co.,Ltd,Beijing 10009&China)

机构地区：[1]江西省无线电监测站,江西南昌330000 [2]江西省工业和信息化厅鹰潭市无线电管理局,江西鹰潭335000 [3]北京博识广联科技有限公司,北京100098

出　　处：《移动通信》2020年第12期80-85,共6页Mobile Communications

摘　　要：针对当前无线电复杂环境下频谱异常快速监测存在大量干扰信息导致精度不高的问题,本文在重构无线电信号时空数据的基础上,采用深度学习的方法提取无线电信号的时空特征;然后,采用栈式自编码网络对无线电信号的时空特征进行稀疏编码再重构;最后,采用聚类算法对重构特征进行聚类,自适应门限值技术获取动态的阈值,并实现异常无线电信号的快速识别。实验表明,融合深度学习和栈式自编码的无线电频谱异常检测算法能够增强异常信号检测模型的自主性,为智能通信系统的分析提供了新的思路。For solving the low-precision problem of the rapid monitoring of spectrum abnormalities caused by massive interference information in current complex radio environment, this paper uses a deep learning method to extract the spatial-temporal features based on the spatial-temporal data reconstruction of radio signals. Then, the stacked autoencoder network is used for the sparse coding and reconstruction of the spatial-temporal features. Finally, clustering algorithm is used to cluster the reconstructed features and the dynamic threshold is obtained by the adaptive threshold technology, and the rapid identification of abnormal radio information is realized. Experiments show that the radio spectrum anomaly detection algorithm based on the deep learning and stacked autoencoder can enhance the autonomy of anomaly detection models and provide a new road for the analysis of intelligent communication system.

关 键 词：时空特征 栈式自编码网络 聚类分析 自适应门限值技术 

分 类 号：TN98[电子电信—信息与通信工程]