Feature Extraction And Recognition Of Radar Radiation Based On Characteristics Of Unintentional Modulation On Pulse

Viewed from the formal of the wars in recent years, radar electronic countermeasures have become the dominant force in the battle. Radar reconnaissance technology, as a vital part of radar confrontation, can even determine victory in a war. Radiation recognition acts as the most important part of radar reconnaissance. Because unstable personality of the radar transmitter, radiated sources will carry with "fingerprint feature", namely characteristics of unintentional modulation on pulse, which would characterize the transmitter. This thesis studies the modeling of signal source added with frequency drift and phase noise unintentional modulation characteristics, feature extraction method of frequency drift characteristics of unintentional modulation, feature extraction method of phase noise characteristics of unintentional modulation and classification methods to recognize the radiation.Firstly, study the forming principles of unintentional modulation characteristics of the signal emitted by the radar transmitter. Complete source modeling by adding unintentional modulation characteristics including frequency drift and phase noise to ideal signal source. Gaussian white noise channel is used in the whole thesis..Secondly, study principle of Adaptive Fractional Spectrum and investigate its strengths over traditional methods with respect to analysis of time-frequency distribution. Experimental results show that frequency estimation error s of intermediate time for single LFM signals and multicomponent LFM signals with AFS are less than 3%, and AFS has the advantage of separating multicomponent nonlinear FM signals. For unintentional modulation characteristics of frequency drift; scout the signal every once in a while and analyze time-frequency distribution by means of Adaptive Fractional Spectrogram for several times to get several characteristic frequencies, connect all the characteristic frequencies as a curve, and extract features of curve. The simulation results show the effectiveness of the employed curve feature extraction method.Then, study principle of Hilbert-Huang Transform for separation of phase noise and useful signal to signal sources added with simple phase noise and analyze the simulated results. As for signal sources added with complexity phase noise characteristics of unintentional modulation, extract features of positive and negative diagonal slice of the bispectrum and analyze its advantages over bispectrum method. Improve recognition efficiency with PCA method to reduce dimension of bispectrum slice characteristics.Finally, under the background of this research, compare the performance of Support Vector Machine and BP neural network recognition method. Complete the identification of radar radiation with characteristics of curve and positive and negative diagonal slice of bispectrum after dimension reduction obtained as above by Support Vector Machine method. When SNR is 15 d B, recognition rate of radiation added with frequency drift reaches 99.17% and radiation added with phase noise reaches 100%.