Research On Fingerprint Recognition Of Shipborne Navigation Radar

Radar emitter fingerprint recognition is one of the key technologies in modern electronic warfare.Its purpose is to quickly and effectively extract the "fingerprint" information that can represent the individual characteristics of radar emitter in the signal,and then use the extracted"fingerprint" to identify the individual radar emitter.For different radar emitter individuals of the same type of radar emitter,the differences between the signals transmitted by the individuals are very small,and the requirements for the extraction of radar emitter fingerprints are more stringent.The emitter characteristics extracted by traditional methods can no longer meet the requirements.Only if the extracted radar fingerprint information is sufficiently representative can different radar emitter individuals be effectively distinguished.In view of the above requirements,this paper completes the following work in combination with the shipborne navigation radar signals collected in the project and the structural parameters of the target shipborne navigation radar radiation source:First,this paper constructs the shipborne navigation radar signal simulation data set and the real data set.Starting from the mechanism of fingerprint generation of radar transmitter,this paper analyzes the structure of shipborne navigation radar transmitter,and concludes that the fingerprint generates mainly from high-frequency oscillator and power amplifier,and the generated fingerprint usually appears as the fluctuation of time domain envelope and the phase noise carried by pulse signal;By modeling the two types of noise,combining them into different types of fingerprints and adding them to the radar pulse data generated by simulation,the simulation data set is constructed;Through signal processing and data annotation of the shipborne navigation radar signals collected in the field,the construction of the real data set is completed.Secondly,this paper proposes RPB-DNA multi-domain fingerprint feature extraction algorithm.First of all,this paper studies the fingerprint feature extraction algorithm of radar emitter,proposes the pulse spurious relative position feature in frequency domain which is named RPF,and completes the dimensionality reduction of radio-frequency distinct native attribute(RF-DNA)fingerprint features,and combines the rectangular integral bispectral entropy,HHT marginal spectral entropy and RPF to propose the RPB-DNA multi-domain fingerprint feature extraction algorithm.Then this paper designed experiments to explore the influence of parameters NR and m in RPB-DNA on the recognition accuracy,and obtained the parameter settings under the optimal recognition accuracy in the experimental group;In order to explore the impact of different classifiers on the recognition accuracy,this paper carried out a comparative experiment on three types of statistical learning classifiers,K-nearest neighbor,decision tree,support vector machine(SVM),and two types of integrated learning classifiers,random forest and XGBoost.The experimental results show that SVM classifier has the best comprehensive performance in recognition accuracy and time consumption of training.Finally,the recognition accuracy of several feature extraction algorithms is compared on the real data set,and the recognition accuracy of the RPBDNA algorithm proposed in this paper is 82.5%,which is higher than that of other feature extraction algorithms.Thirdly,this paper studies the radar emitter fingerprint identification method based on deep learning,and proposes a shipborne navigation radar fingerprint identification method based on wavelet scattering convolution network and convolution neural network(CNN).The influence of the convolution kernel parameters and the number of CNN layers on the recognition accuracy of the proposed model is explored,and the ideal model parameter settings under the research scenario are obtained through a large number of experiments;Finally,the model is trained and tested with real data sets,and the recognition accuracy of the model on the real data set is 95.36%,which is higher than that of traditional machine learning based radar emitter fingerprint recognition methods and bispectral and CNN based radar emitter fingerprint recognition methods,which shows the effectiveness of the method proposed in this paper.