Study Of Radar, Active Jamming Signal Feature Analysis And Recognition Algorithms

Active jamming plays a crucial role in modern electronic warfare. The feature extraction and identification of active jamming is the precondition and foundation of anti-jamming, which is very significant in the field of Electronic Counter Measures (ECM). The objective of identification is to decide the type of active jamming without any priori knowledge about the enemy, and then take the corresponding measures to suppress the jamming. The identification of active jamming is studied in this dissertation. The main work of this dissertation is summarized as follows:1. Study the principle of active jamming, and analyze the models of noise jamming and deception jamming. This work is very important because it is the foundation of feature extraction and identification.2. Based on the theory of wavelet transform, bispectrum, chirplet transform, analyze active jamming in different domains and extract the characters without the influence of the noise. At the same time, study the features quantificationally by modern signal processing when Jamming-to-Noise Rate (JNR) varies in a certain range. From different views, different feature extraction methods are used to explore the characteristics of active jamming.3. Based on the grey theory, the degree of incidence between the extracted features is calculated to select the best feature subset from the combined feature set. The simulation results show that the most important features are selected and that the computational complexity is reduced by using this method.4. After the features are selected adequately, three classifier design methods, including Artificial Neural Network (ANN), tree discriminator and Support Vector Machine (SVM) are studied to recognize the active jamming automatically. The validities and superiorities of three classifiers are shown in the comparative experiments, and several conclusions are obtained: The ANN algorithm can perform the identification of active jamming effectively, but has an excessive computational load. Tree discriminator is simple to implement and has a good performance. However, it is sensitive to the selection of threshold. SVM can identify the types automatically and the computational complexity is small. SVM is superior to ANN and tree discriminator both in classification capability and efficiency.