| In electronic warfare,electronic intelligence provides real-time information about enemy electronic systems for decision-makers,while radar specific emitter identification provides an efficient and reliable means for electronic intelligence.Radar specific emitter identification is a process of distinguishing different individuals of enemy radar emitters by comparing intercepted radar signals’ intra-pulse characteristics.By identifying radar emitters,electronic intelligence can more accurately assess enemy electromagnetic threats and develop targeted electronic warfare strategies.This paper focuses primarily on individual modeling of radar emitters,feature extraction,and classifier design.Based on different individual modeling methods of radar emitters,the paper selects classification features and designs classifiers reasonably to accomplish the task of radar specific emitter identification.The main research contents of this paper are as follows:A precise description of unintentional phase modulation characteristic curves and an identification algorithm based on the Attention-based Gated Recurrent Unit(Attention-GRU)are proposed to model unintentional phase-modulated radar emitters and address the low recognition rate problem in traditional algorithms that utilize unintentional phase modulation characteristic curves.In the proposed algorithm,the unintentional phase modulation curves smoothed by uniform B-spline can more accurately represent individual radar emitters.In the Attention-GRU model,the gated neural network possesses powerful temporal modeling capabilities,and the attention mechanism can reasonably allocate attention resources based on different samples,resulting in a significant improvement in recognition accuracy.Simulation results demonstrate that the attention layer in the designed recognition model can reasonably allocate attention resources to different sample models,effectively improving recognition accuracy.The proposed algorithm achieves high recognition accuracy even with a small sample size and converges under medium-to-high signal-to-noise ratio conditions.A modeling method using phase noise to represent individual radar emitters is proposed.To address the problem of excessive redundant information in bispectra that prevents them from being excellent templates for identification,a bispectrum quadratic feature extraction and combination method is proposed.Based on this method,a radar specific emitter identification algorithm is formed using a 4-channel tensor and a residual neural network based on an efficient channel attention mechanism.Firstly,the causes of phase noise and its influence on the temporal waveform and spectrum of radar signals are analyzed.The models of radar signals with phase noise for different modulation types are derived based on the first-order Bessel function and the Jacobian-Anger expansion.Secondly,the bispectrum are introduced to expose individual differences caused by phase noise.Finally,four types of integrated bispectra are extracted,and a 4-channel tensor composed of these spectra is used as the classification feature.A one-dimensional residual neural network with efficient channel attention mechanism(Res Net1D-ECA)is employed as the classifier for identification.Simulation results and identification results on public radar datasets demonstrate that the method based on the 4-channel tensor and the residual neural network with an efficient channel attention mechanism can achieve high recognition accuracy and is applicable to real-world scenarios.Behavioral models of radio frequency power amplifiers are used to model radar emitters.The 1.5-dimensional spectrum and the 1/2-slice of the ambiguity function of the radiation sources are extracted,and a one-dimensional convolutional long short-term memory neural network(CNN1D-LSTM)is applied to accomplish radar specific emitter identification.Firstly,the Taylor series model and the memory polynomial model of the power amplifier are used to model the radar specific emitter identification.Secondly,the 1.5-dimensional spectrum and the1/2-slice of the ambiguity function of the radiation source signal are extracted as classification features.Finally,a dual-branch parallel model,CNN1D-LSTM,is constructed to achieve radar specific emitter identification. |
