基于Deep SVDD的通信信号异常检测方法  

Deep SVDD-based anomaly detection method for communication signals

作　　者：康颖[1,2] 赵治华 吴灏[1,2] 李亚星 孟进 KANG Ying;ZHAO Zhihua;WU Hao;LI Yaxing;MENG Jin(Institute of Military Electrical Science and Technology, Naval University of Engineering,Wuhan 430033, China;National Key Laboratory of Science and Technology on Vessel Integrated Power System, Naval University of Engineering, Wuhan 430033, China)

机构地区：[1]海军工程大学军用电气科学与技术研究所,湖北武汉430033 [2]海军工程大学舰船综合电力技术国防科技重点实验室,湖北武汉430033

出　　处：《系统工程与电子技术》2022年第7期2319-2328,共10页Systems Engineering and Electronics

基　　金：国家杰出青年科学基金(52025072);国家自然科学基金(61801502,61801501,62001497);国防科技重点实验室稳定支持项目(6142217200105);湖北省自然科学基金(ZRMS202001331);海军工程大学发展基金(425317S126)资助课题。

摘　　要：针对复杂电子对抗场景中的非理想信道环境,该文提出了一种基于深度学习的异常检测(anomaly detection,AD)方法。首先,分析了利用时频同相/正交(in-phase/quadrature,I/Q)采样数据进行AD的可行性;然后,设计了深度学习网络架构,并提出基于深度支持向量描述(deep support vector data description,Deep SVDD)和调制识别的AD方法。仿真及实验结果表明:相比于经典的单分类检测算法,该方法检测性能和实时性明显提升,且在非理想信道环境下表现鲁棒。该方法已在某型号项目原理样机上得到验证,具有很高应用价值。To solve the problem of anomaly detection(AD)of the non-ideal channel in complex electronic countermeasures,a deep learning based method is presented.First,the feasibility of using time-frequency in-phase/quadrature(I/Q)sampling data for anomaly detection(AD)is analyzed.Then,a deep learning network architecture is designed and an AD method based on deep support vector data description(Deep SVDD)and modulation classification is proposed.Simulation and experimental results show that the detection performance and real-time performance of the method are significantly improved compared with the classical algorithms of one-class classification,and the performance is robust in non-ideal channel environment.The method is validated on a sample machine and is of high application value.

关 键 词：异常检测 Deep SVDD 调制识别 干扰预警 

分 类 号：TN911.7[电子电信—通信与信息系统]