Individual Recognition Of Radar Emitters Based On Deep Neural Network

How to effectively identify radar radiation sources,thereby obtaining military intelligence information,identifying friend or foe in contemporary electronic warfare,and specifying combat plans is of great significance.However,with the continuous complexity of the electromagnetic environment and the continuous diversification of various new radar systems,processing and analyzing the intercepted radar radiation source signals to achieve radar radiation source identification has put forward higher requirements.How to accurately and quickly identify individual radar emitters is a key subject of current research in the field of electronic countermeasures.In order to better realize the individual recognition of radar emitters,this paper analyzes the causes of fingerprint features on the one hand,selects effective fingerprint features,extracts individual signal features to form a multi-dimensional feature matrix,and proposes improved random forest classification to achieve individual recognition of radar emitters.On the other hand,a method combining one-dimensional convolutional neural network and random forest classifier is proposed to identify individual signals from the original intermediate frequency pulse signal,and apply it to the motion scene,so as to better realize the individual recognition of radar emitters.The main work of the thesis is as follows:1.Model and simulate the transmitting signal and receiving signal of the radar radiation source.By exploring the radar radiation source transmitter,analyzing the fingerprint characteristics caused by the components,and studying the phase noise generated by the component oscillator,parasitic it on the chirp modulation signal,and realize the modeling and simulation of the radar radiation signal.2.The fingerprint characteristics of the radar emitters are studied,and amulti-dimensional feature matrix is obtained.The envelope,bispectrum,and variational modal decomposition of the signal are analyzed respectively,and different fingerprint features are extracted,and a multi-dimensional feature matrix is obtained.3.Carry out sub-forest optimization on random forest classification,and propose an improved random forest classification(IRFC)algorithm.Through comparative analysis,it is concluded that the IRFC algorithm has good performance in identifying individual radar emitters.When the signal-to-noise ratio is 12 d B and above,the recognition rate reaches90%,which effectively improves the recognition rate of individual radiation sources.It is applied to the recognition of radar emitters under the condition of differential signal-to-noise ratio,and the signal-to-noise ratio reaches more than 16 d B,and the result of individual identification reaches more than 87%,which realizes the differential signal-to-noise ratio radar under the condition of high signal-to-noise ratio.Individual identification of the radiation source.4.Propose a 1D-CNN-IRFC algorithm for individual identification of radar emitters,which combines one-dimensional convolutional neural network and random forest classification,and directly uses the original intermediate frequency pulse signal of the emitter to learn the characteristics that can better characterize the individual.,Effectively completed the individual identification of the radar radiation source.Through experiments,it is concluded that the recognition rate under the same signal-to-noise ratio condition reaches 95% at a signal-to-noise ratio of 0d B and above;the recognition rate under a differential signal-to-noise ratio condition reaches 90% at 0d B and above.Individual identification of radar emitters with differential signal-to-noise ratio under low signal-to-noise ratio conditions.Finally,it is applied to sports scenes to verify the applicability and effectiveness of the algorithm.