Research On Signal Subtle Feature Extraction And Recognition Technologies

Individual transmitter identification technique is a very important research subject in the field of communication and electronic countermeasure, which is mainly baesd on the fact that the received electromagnetic signal contains the information resulted from the difference of transmitter hardware. Through advanced signal subtle features extraction and classification method to determine which signal comes from which radiation source and in combination with actural requirement in specific application background, we can acquire very valuable information. Unlike the research on the individual transmitter identification of different classes, this article mainly focuses on the signal subtle feature extraction and identification techniques of the transmitters which have the same types, and the same batches and work under the same way.Nowadays, the transient feature extraction using the turn-on signal is mostly researched for individual transmitter identification, mainly because the turn-on signal, containing no modulation information, is the impulse response of various components inside the transmitter and can best reflect the variation between different individuals. In this article the wavelet transform and fractal theory is studied and used to extract the signal subtle features from four different aspects of the boot signal; However, as a result of the short duration of the boot signal, timely and exact separation of the boot signal is extremely important, therefore the threshold method based on the variance fractal dimension together with the Bayes detection method are researched to locate the start of transient signal. An improved method based on Bayes detection is proposed, which still works well under low SNR case.Besides the feature extraction of transient signal, this article analyses the subtle features of steady-state signal as well. The extraction methods of pulse signal envelope feature using the complex wavelet transform and Hilbert transform are discussed in the article, by which we extract the rising edge, falling edge, envelope top drop and pulse width as the final feature values. What's more, the high order features of the envelope, J value and R value, are studied for further analyzing the envelope feature. Finally, the three widely used classifiers in individual identification system, namely nearest neighbor classifier, back propagation neural network classifier and support vector machine classifier, are studied and the identification performance of the above extracted feature sets, including transient feature sets and steady-state feature sets, is simulated by employing the three classifiers in Matlab. What's more, we present six different schemes of individual transmitter identification, which are totally fit for this article.The research work related in this article, to some extent, solves the problems of identifying individual transmitter which have the same types and the same batches and work under the same way, has theoretical significance and practical application prospects.