Structural Recognition Of Radar Emitter Based On Deep Learning

With the application of various new radar systems in the complex electromagnetic environment,it is of great significance to realize the recognition of non-cooperative radar in electronic warfare.The internal noise of radar radiation source components and the nonlinear distortion of power amplifier will lead to the non-ideal factors carried in radar signals.Such non-ideal factors are unique and cannot be replicated.This difference in radar signal carrying due to the structure of radar radiation source is called unintentional modulation information.The intentional modulation information mainly comes from the change of modulation mode and modulation parameters of radar radiation source.Both unintentional and intentional modulation information are taken into account in the inversion of radar radiation sources.The research of radar radiation source inversion includes several levels of tasks,among which the first level is structure identification,that is,the inversion of radar radiation source structure.The structure identification of radar radiation source is mainly divided into two parts.The first part obtains radar beam signal through forward modeling according to the structure;the second part determines the structure of radar radiation source by structure identification of radar beam signal and completes the inversion of the structure of radar radiation source.In the face of the problems encountered in electronic warfare today,and on the basis of the research fields of scholars,this paper uses the method of deep learning to carry out radar radiation source inversion.Firstly,two structures of phased array radar radiation sources,active and passive,were modeled through forward modeling.High power radar signals were generated using a combination of Direct Digital Synthesizer(DDS)and RF amplification links.Then,in order to solve the problem that only extracting the time-domain parameters of radar signals cannot achieve accurate recognition of radar radiation sources,a radar radiation source structure recognition method based on stack convolutional autoencoder and deep neural network(SCAE-DNN)is proposed to complete two processes: forward modeling and structure recognition.SCAE is used to replace the tedious calculations in traditional algorithms to extract RF structural features.There is a mapping relationship between RF structural features and the structure of radar radiation sources,which is not affected by differences in time,frequency,and spatial domains.Experimental results show that the method can accurately identify the structure of radar emitter,and has strong generalization ability for multiple modulated radar signals with different SNRs of additive Gaussian white noise.Finally,in order to solve the adverse effects caused by different types of radar emitters with the same structure in the inversion process,a radar emitter structure identification method combining compressed sensing theory and neural network is proposed.Compressed sensing can be used to compress signals,project high-dimensional signals into low-dimensional space,and combine the representation learning ability of SCAE network to build a new TCS loss function,complete the mining of deeper information of input signals,strengthen the correlation between RF structural features and the structure of laser radiation sources,and extract RF structural features through joint constraints of time domain and sparse domain.The experimental results show that this method can solve the adverse effects caused by different types of radar radiation sources under the same structure during the inversion process,and achieve more accurate recognition of radar radiation source structure.