Radar Emitter Identification Based On Multi-domain Feature Extraction

In the current war environment,the proportion of electronic warfare in actual combat has gradually increased,and interception of radar signals often affects the war situation.However,the complication of the electrom agnetic environment makes it difficult to identify the intercepted radar signals.Therefore,it's necessary to correct identification of the radiation source,and additional interference is an important research direction in electronic countermeasures.Radar signal' single feature cannot determine the correct type of radation source,but adopting a more fields and more comprehensive feature extraction of signals can effectively improve the recognition rate.This paper focuses on extracting the multi-domain features of radiation sources and improves the recognition accuracy through multi-core learning..Firstly,the common radar radiation sources and the principle of radar unintentional modulation are summarized,and the radar signal source is further constructed to analyze the radar characteristics.For this subtle feature generated by the transmitter,feature extraction methods based on fractal theory and high-order statistics are used to extract features of the additional phase noise and envelope feature signals,and the obtained high-dimensional features are analyzed and simply selected.The multi-features are spliced.The features include the time domain,frequency,and phase characteristics of the signal,which can fully express the characteristics of t he signal.Then,the obtained multi-domain features are optimized.Because the extracted feature dimensions are high,features may not be obvious for some of the domain features.At the same time,there may be some noise in the high-dimensional features to make them fit.In this case,we need to reduce the dimensions of existing features,adopt feature extraction and feature selection methods in feature reduction,and compare the advantages and disadvantages of several methods to choose method to preserve the most useful features.Finally,the design of classifiers is completed for existing features.For multi-domain features,there may be heterogeneity.The recognition effect of single-core support vector machines for multi-domain features may be limited by the ability of a single kernel function to map features.Supporting vector machine design based on multi-core learning can solve this heterogeneous situation.The splicing-optimized features are classified by multi-kernel SVM,and the relationship between recognition rate and SNR is analyzed,and the measured data of radar emitters are processed.In this thesis,a complete radar emitter identification system is constructed by feature extraction,feature optimization and classifier design.In the case of a signal to noise ratio of-6d B,the system can increase the recognition rate by about 7%,and it raise up 5% with-4d B during two types of signals with different additional envelope characteristics and phase noise.