Research On Radar Emitter Signal Identification Based On Deep Learning

Radar emitter signal identification,as an important part of electronic countermeasures reconnaissance,has become a research hotspot in contemporary information warfare.In the increasingly complex electromagnetic environment,the traditional identification methods based on feature extraction of Pulse Description Word(PDW)and pattern matching with radar database can no longer meet the requirements of more accurate and efficient identification.Therefore,the Deep Learning approach has been introduced into the radar emitter signal identification topic,which has made some achievements.However,the problem of low recognition accuracy still exists in complex electromagnetic environment under the conditions of low signal-to-noise ratio,high random error,many false pulses and missing pulses.This paper has studied the radar emitter signal identification method based on deep learning,starting from both intra-pulse modulation and inter-pulse modulation,then proposed an intra-pulse modulation identification method based on 2D time-frequency information and CH-CNN and an inter-pulse modulation identification method based on CLDNN.The main work and innovation points are summarized below.1.This paper studied the process of non-cooperative radar signal processing and identification as well as the theoretical basis and basic model of Deep Learning,and proposed the process framework of inter-pulse modulation and intra-pulse modulation identification of radar emitter signals based on Deep Learning,which laid the foundation for the subsequent research of this paper.2.An identification method based on 2D time-frequency information and CH-CNN is proposed for intra-pulse modulation.Through data enhancement and optimization of the Convolutional Neural Networks,the recognition effect under low signal-to-noise ratio conditions is improved.Compared with other Deep Learning methods,the recognition accuracy of this method is significantly improved at a signal-to-noise ratio of-20.3.A CLDNN-based identification method is proposed for inter-pulse modulation.The error conditions of ideal environment,general environment and extreme environment are set in the simulation of inter-pulse PRI sequences,and the error factors of PRI sequences are fully considered.The CLDNN method is analyzed in terms of random errors,false pulses and missing pulses,respectively,and it is verified that the method outperforms the other five methods in the extreme environment for identification.