Research On Radar Signal Sorting Technology Based On Multilevel Variational Mode Decomposition

Radar signal sorting is an important part of modern military.Interrupted sampling repeater jamming(ISRJ)is a new type of jamming,ISRJ can generate one or more false targets with high coherence in low transmission power,and ISRJ is highly deceptive for radar target tracking and detection.Variational mode decomposition(VMD)is a kind of signal decomposition algorithm,which is mainly used in seismic time-frequency analysis,gear fault diagnosis,signal denoising and other fields.In this thesis,we propose to apply VMD on the radar signal sorting in the presence of ISRJ,and an improved multilevel VMD sorting algorithm is proposed too,and we combine it with deep learning to improve the signal sorting effects in the presence of the jamming.The main research contents and contributions of this thesis are summarized as follows:1.This thesis firstly introduces the common methods of radar signal processes,which including the linear frequency modulation signal,the principle of pulse compression and the constant false alarm rate detection algorithm.Then the three types of ISRJ are discussed in detail,and the characteristic differences between the jamming and the target are analyzed by time-frequency analysis.2.We propose a radar signal sorting algorithm in the presence of ISRJ based on VMD,and the simulation experiments verify the effectiveness of the proposed scheme.ISRJ and target echo have different center frequencies after dechirping,which means that the jamming and the target in received echo can be separated by VMD.VMD-Net is the network implementation of the VMD algorithm.VMD-Net effectively reduces the number of iterations of VMD which can achieve better results with fewer iterations.However,the effect of VMD is greatly associated with noise,the signals with different center frequencies can’t be separated by VMD in the case of too much noise.Therefore,we propose three improvements in this thesis.The first is determining the number of sub-signals in VMD algorithm automatically by constant false alarm rate detection,the second is reconstructing target by singular spectrum analysis,and the third is multilevel VMD.The improved algorithm greatly improves the radar signal sorting effect in the presence of low signal-tonoise ratio noise,and can effectively separate the jamming and the target from the received echo.3.We propose a radar signal sorting algorithm in the presence of ISRJ based on VMD and deep learning.The normalized cross-correlation coefficient is used as the matching degree between the sub-signal and the transmitting signal,but the cross-correlation coefficient is sensitive to the noise.The cross-correlation coefficient is small in the case of too much noise,and it is large in the case of little noise.It is difficult to determine a threshold for selecting sub-signals for the multilevel VMD process.Deep neural networks are good at feature learning,the calculation of cross-correlation coefficient can be replaced by deep residual network classification.The signals decomposed by VMD are down-sampled to a uniform size which are used as the input of the neural networks,then the sub-signals are classified into the target class and the jamming or noise class.Simulation experiments show that the deep neural networks improve the radar signal sorting effect in the presence of ISRJ.