Research On Specific Emitter Identification Based On Primary Signal Suppression

As a key component of electronic support system,specific emitter identification,which provides intelligence support for threat classification,battlespace awareness,command decision-making,is an important approach to obtain information dominance on the battlefield.However,specific emitter identification also faces some limitations,which is manifested by low stability of individual features and the decline of recognition correct rate due to the influence of the primary signal with high energy on the features.How to further highlight the individual features from the radar emitter signals and improve the recognition rate becomes an urgent problem in specific emitter identification.Against this backdrop,specific emitter identification based on primary signal suppression is studied in this dissertation.Firstly,the unintentional modulation mechanism of emitter singals is analyzed in combination with oscillators and power amplifiers.Experiments show that unintentional modulation by the emitter hardware expands the spectrum of the ideal signal and produces sideband components.Using the addition method of phase noise and the common used Gaussian white noise channel,the modeling process of the linear frequency modulation(LFM)signal and sinusoidal frequency modulation(SFM)signal is completed.Secondly,the process of primary signal suppression based on synchrosqueezed wavelet transform is studied.The denoising method based on stationary wavelet transform,the time-frequency distribution under synchrosqueezed wavelet transform and the complete algorithm for primary signal suppression are analyzed in detail.Through simulation experiments,according to the intuitive time domain waveform comparison,and using the root mean square error and Pearson correlation coefficient as numerical indicators,the effectiveness of the primary signal suppression algorithm based on synchrosqueezed wavelet transform is demonstrated,and under the condition of lower signal to noise ratio,it can still maintain a basic suppression ability.Then,on the basis of primary signal suppression,the dynamic wavelet fingerprint,the double-spectrum encirclement integral and the fractal box-counting dimension are studied in turn.The joint simulation of the feature extraction before and after the primary signal suppression is carried out,and the individual recognition rate is evaluated by the support vector machine with only a single kernel.The simulation results show that the process of primary signal suppression improves the recognition rate of the three features by more than 10%compared with the case of retaining the primary signal,which verifies the positive effect of the primary signal suppression on specific emitter identification.Finally,in allusion to the heterogeneous problems that may exist in multi-domain features space,multiple kernel learning algorithms are studied.Support vector machines(SVM)based on Simple Multiple Kernel Learning(SimpleMKL)and Generalized Multiple Kernel Learning(GMKL)are constructed respectively,the classification performance is compared under the joint multi-domain feature.The experimental results show that compared with the traditional single kernel support vector machine,the SimpleMKL-SVM improves the individual recognition rate between 3% and 6%.The GMKL-SVM also achieves nearly the same performance as SimpleMKL-SVM in the case of binary classification.