基于深度残差网络的特定协议信号识别
|
摘要
针对短波信道下信号截获质量差,信道环境复杂以及单一特征识别率低等问题,提出了基于深度残差网络的信号特征自动提取算法,设计了一种具有自适应学习能力的短波特定通信协议识别模型.通过对具有特殊结构的协议信号的时频视觉差异进行理论推导,将信号的时频能量转换成灰度图像,并用于对所构建的深度残差网络进行训练.该方法克服了传统方法对信号质量要求高、先验信息需求多等缺陷,可直接对中频接收信号进行处理,适合实际工程应用.实验表明,当深度残差网络达到稳态时,识别准确率高,在低信噪比、多径衰落、多普勒频偏以及信号被强干扰所遮挡的情况下,依旧能准确识别协议类别.
To correctly classify the specific protocol signal,a signal recognition model with adaptive learning and automatic feature extraction ability is proposed.This model is based on the deep residual network,which can solve the drawbacks,such as,the poor quality of the intercepted communication signal,the complex condition of the short wave channel,and the low recognition rate of the single feature.After analyzing the visual characteristic of communication protocol signal with special structure in time frequency domain,the time frequency gray-images are obtained and utilized to train the deep residual network.This method does not need much prior knowledge and is insensitive to signal quality.Moreover,it can process the intermediate-frequency signal directly.Due to these advantages,the algorithm is suitable for engineering application.Simulation results show that,when the deep residual network reaches its steady status,the proposed model can accurately identify the protocol.And it is also proved effective even at complex circumstance where the multipath fading and the Doppler shift exist,the signal-to-noise ratio is low,and the interference is strong.
To correctly classify the specific protocol signal,a signal recognition model with adaptive learning and automatic feature extraction ability is proposed.This model is based on the deep residual network,which can solve the drawbacks,such as,the poor quality of the intercepted communication signal,the complex condition of the short wave channel,and the low recognition rate of the single feature.After analyzing the visual characteristic of communication protocol signal with special structure in time frequency domain,the time frequency gray-images are obtained and utilized to train the deep residual network.This method does not need much prior knowledge and is insensitive to signal quality.Moreover,it can process the intermediate-frequency signal directly.Due to these advantages,the algorithm is suitable for engineering application.Simulation results show that,when the deep residual network reaches its steady status,the proposed model can accurately identify the protocol.And it is also proved effective even at complex circumstance where the multipath fading and the Doppler shift exist,the signal-to-noise ratio is low,and the interference is strong.
引文
[1] 张俊林,王彬,汪洋,刘明骞.一种α稳定分布噪声下OFDM信号调制识别与参数估计算法[J].电子学报,2018,46(6):1390-1396.ZHANG Jun-lin,WANG Bin,WANG Yang,LIU Ming-qian.Analgorithm for recognition and parameters estimation of OFDM in Alpha stable distribution noise[J].Acta Electronica Sinica,2018,46(6):1390-1396.(in Chinese)
[2] 林祎,彭华,赵振华.基于小波降噪的短波通信信号协议识别特征提取算法[J].信息工程大学学报,2012,13(4):438-442.LIN Yi,PENG Hua,ZHAO Zhen-hua.Protocol recognition feature extraction algorithm of high frequency communication signals based on wavelet de-noising[J].Journal of Information Engineering University,2012,13(4):438-442.(in Chinese)
[3] 林肖辉,张润生.基于模板匹配的2G-ALG信号识别技术[J].无线电通信技术,2016,42(3):46-48,69.LIN Xiao-hui,ZHANG Run-sheng.Technology of 2G-ALG signal recognition based on spectrum template matching[J].Radio Communications Technology,2016,42(3):46-48,69.(in Chinese)
[4] 聂东举,叶进,闫坤,车俐.基于SVM算法的短波通信协议识别技术[J].系统工程与电子技术,2013,35(6):1307-1311.NIE Dong-ju,YE Jin,YAN Kun,CHE Li.Recognition technology for high frequency communication protocol based on SVM algorithm[J].Journal of Systems Engineering and Electronics,2013,35(6):1307-1311.(in Chinese)
[5] KIM N,KEHTARNAVAZ N,YEARY M B,et al.Dsp-based hierarchical neural network modulation signal classification[J].IEEE Transactions on Neural Networks,2003,14(9):1065-1071.
[6] He K,Zhang X,Ren S,et al.Deepresidual learning for image recognition[A].Computer Vision and Pattern Recognition[C].US:IEEE,2016.770-778.
[7] Krizhevsky A,Sutskever I,Hinton G E.ImageNet classification with deep convolutional neural networks[A].International Conference on Neural Information Processing Systems[C].Curran Associates Inc,2012.1097-1105.
[8] Ketterer H,Jondral F,Costa A H.Classification of modulation modes using time-frequency methods[A].IEEE International Conference on Acoustics,Speech,and Signal Processing 1999[C].US:IEEE Computer Society,1999.2471-2474.
[9] Folland G B,Sitaram A.The uncertainty principle:A mathematical survey[J].Journal of Fourier Analysis & Applications,1997,3(3):207-238.
[10] MIL-STD-118-203-1A.Interoperability and Performance Standards for Tactical Digital Information Link(TADIL)A[S].USA:[s.n.],1988.
[11] He K,Sun J.Convolutional neural networks at constrained time cost[A].Computer Vision and Pattern Recognition[C].US:IEEE,2014.5353-5360.
[12] Ioffe S,Szegedy C.Batchnormalization:Accelerating deep network training by reducing internal covariate shift[A].International Conference on Machine Learning[C].JMLR.org,2015.448-456.
[13] CLOVER2000 Waveform& Protocol[S].Urbana:HAL Communications Corp,1999.
[14] MIL-STD-188-141A.Interoperability and Performance Standards for Medium and High Frequency Radio Systems[S].USA:[s.n.],1988.
[15] Watterson C,Juroshek J,Bensema W.Experimentalconfirmation of an HF channel model[J].IEEE Transactions on Communication Technology,1970,18(6):792-803.
[2] 林祎,彭华,赵振华.基于小波降噪的短波通信信号协议识别特征提取算法[J].信息工程大学学报,2012,13(4):438-442.LIN Yi,PENG Hua,ZHAO Zhen-hua.Protocol recognition feature extraction algorithm of high frequency communication signals based on wavelet de-noising[J].Journal of Information Engineering University,2012,13(4):438-442.(in Chinese)
[3] 林肖辉,张润生.基于模板匹配的2G-ALG信号识别技术[J].无线电通信技术,2016,42(3):46-48,69.LIN Xiao-hui,ZHANG Run-sheng.Technology of 2G-ALG signal recognition based on spectrum template matching[J].Radio Communications Technology,2016,42(3):46-48,69.(in Chinese)
[4] 聂东举,叶进,闫坤,车俐.基于SVM算法的短波通信协议识别技术[J].系统工程与电子技术,2013,35(6):1307-1311.NIE Dong-ju,YE Jin,YAN Kun,CHE Li.Recognition technology for high frequency communication protocol based on SVM algorithm[J].Journal of Systems Engineering and Electronics,2013,35(6):1307-1311.(in Chinese)
[5] KIM N,KEHTARNAVAZ N,YEARY M B,et al.Dsp-based hierarchical neural network modulation signal classification[J].IEEE Transactions on Neural Networks,2003,14(9):1065-1071.
[6] He K,Zhang X,Ren S,et al.Deepresidual learning for image recognition[A].Computer Vision and Pattern Recognition[C].US:IEEE,2016.770-778.
[7] Krizhevsky A,Sutskever I,Hinton G E.ImageNet classification with deep convolutional neural networks[A].International Conference on Neural Information Processing Systems[C].Curran Associates Inc,2012.1097-1105.
[8] Ketterer H,Jondral F,Costa A H.Classification of modulation modes using time-frequency methods[A].IEEE International Conference on Acoustics,Speech,and Signal Processing 1999[C].US:IEEE Computer Society,1999.2471-2474.
[9] Folland G B,Sitaram A.The uncertainty principle:A mathematical survey[J].Journal of Fourier Analysis & Applications,1997,3(3):207-238.
[10] MIL-STD-118-203-1A.Interoperability and Performance Standards for Tactical Digital Information Link(TADIL)A[S].USA:[s.n.],1988.
[11] He K,Sun J.Convolutional neural networks at constrained time cost[A].Computer Vision and Pattern Recognition[C].US:IEEE,2014.5353-5360.
[12] Ioffe S,Szegedy C.Batchnormalization:Accelerating deep network training by reducing internal covariate shift[A].International Conference on Machine Learning[C].JMLR.org,2015.448-456.
[13] CLOVER2000 Waveform& Protocol[S].Urbana:HAL Communications Corp,1999.
[14] MIL-STD-188-141A.Interoperability and Performance Standards for Medium and High Frequency Radio Systems[S].USA:[s.n.],1988.
[15] Watterson C,Juroshek J,Bensema W.Experimentalconfirmation of an HF channel model[J].IEEE Transactions on Communication Technology,1970,18(6):792-803.