Pulse Repetition Interval Modulation Recognition Of Advanced Radar Emitter Signals

Radar emitter signal recognition is an important task and a key issue of electronic reconnaissance system. The identification results provide intelligence information and real-time support for electronic countermeasures and electronic counter-countermeasures. The results affect or even decide the ending of the electronic warfare. With the development of radar technology and electronic jamming, more and more complex type radars put into service and the threat of guided weapons becomes stronger, which brings many difficulties to radar emitter signal identification and requires recognition system to have high recognition precision and strong stability.In the complex type radar emitter signal recognition, this paper concentrates on radar emitter pulse repetition interval (PRI) modulation recognition. Based on the introduced emitter identification model, this paper extracts features from radar emitter signals with different PRI modulations as feature vector. Then the feature vector is used to classify automatically radar emitter signal with different PRI types using intelligent classifiers. The main work and research fruits are as follows.1. The issue of radar emitter signal recognition is summarized. The state-of-the-art of radar emitter signal recognition is dealt with in detail and is analyzed systematically. The problem to solve is pointed out.2. The radar pulse signals with complex PRI modulations and its mathematical model are analyzed synthetically. The main reasons that lower signal recognition rate and the simulation software of radar emitter signals are given.3. After analyzing the existing radar emitter signal recognition models, this paper presents a radar emitter signal recognition model based on signal parameters extraction.4. Based on analysis of radar emitter signal PRI train, a feature extraction method is presented to extract a two-dimensional feature vector to distinguishradar emitter signals with different PRI modulation types in time domain. Experimental results show that the feature extraction method not only separates the different PRI modulation types, but also lowers greatly the dimension of feature vector so as to decrease the complexity of classification.5. Neural network is adopted to solve the problem of recognizing radar emitter PRI modulation types. Probability neural network (PNN) is used to design classifier to recognize the radar emitter PRI modulation types. Experimental results show that the feature vector extracted not only has good recognition ability for PRI modulation types, but also keeps the good recognition ability while noise exists. The experimental results also show that the PNN classifier has good recognition ability while the dimension of feature vector is low, but the recognition ability deteriorates as the dimension of feature vector increases.6. Support vector machine (SVM) is used to design classifiers to recognize radar emitter PRI modulation type. After analyzing the SVM kernel functions, the examples of linear and non-linear pattern classification problems are given respectively. A multi-class SVM classifier is designed to recognize radar emitter PRI modulation types. This paper also compares the classification performances of SVM classifiers and PNN classifies.Experimental results show: (i) kernel function selection is a key problem when SVM is used to design classifiers. Radial basis function is a good kernel function when a non-linear pattern classification problem is solved, (ii) SVM classifier can solve linear and complex non-linear pattern recognition problem perfectly. The performance of SVM classifier is decided by small Support Vectors, so the computing complexity is decreased to a large degree. The process of classification using SVM is a process to find an optimal plane which separate different classes, (iii) When adopting SVM classifier, the feature extraction method presented in this paper not only decreases the input vector of classifier (from n (ri?2) to 2) and has error-tolerance ability, but also achieves the same recognition precision as high feature vector, (iv) SVM classifier has stronger adaptability of feature vector dimensionality and noise than PNN classifier. SVM classifier keeps good recognition ability when the dimension of input vector increases.This work was supported by National EW Laboratory Foundation (NO. NEWL51435QT220401) and National Natural Science Foundation (NO. 60572143)...