On Methods For Specific Radar Emitter Identification

Specific emitter identification(SEI) plays an important role in modern electronic reconnaissance systems and electronic support measurement systems. By analyzing the intercepted radar signal, it can recognize individual electronic emitters which may be of same type or same parameter. The provided reliable and timely intelligence information could be further used to locate and track the potential radar emitters, as well as the particular units operating them. This dissertation mainly devotes to the core methods in specific radar emitter recognition, with particular attention to the methodology involved in the study. The main research results are as follows:1. Two ambiguity function(AF) based feature extraction methods for unintentional modulation recognition of radar emitters are proposed, which utilize the Zero-Doppler cut and representative-cut of AF, respectively. The promising results demonstrate that the adoption of near-zero cuts as a feature set not only accords with the characteristics of real emitter signals, but also alleviates the computation burden in existing AF based methods. Further, Direct Discriminant Ratio criterion(DDR) is employed to rank the kernel points along the obtained cut, which leads to the preservation of the most discriminant features of radar emitter signals.2. By systematic modeling of the phase-noise in radar transmitter via AF, a "joint-action platform" is established, which connects both the information item of unintentional modulation and the actual recognition performance. Theoretical analysis and extensive experiments on real data not only confirm that unintentional modulation is closely related to the phase-noise, but also support our previously proposed methods with strong evidence. Instead of utilizing the modular value of AF as in traditional methods, we propose to use the real part of Zero-Doppler cut as a high-effective feature, which shows a consistent advantage over the corresponding modular feature and can serve as a good alternative to the representative-cut in practical application.3. Canonical Correlation Analysis(CCA) based feature fusion is elaborated for specific radar emitter recognition. Two kinds of cut-concatenation schemes are designed to construct two different pairs of complementary feature vectors respectively, which will facilitate the subsequent feature fusion via CCA or Discriminative CCA. The suggested framework not only offers stable performance even when time-domain feature is not valid, but also improves the recognition accuracy greatly, due to the successful information fusion and redundancy reduction conducted in the AF subset.4. Aiming at online recognition of radar emitters, a generalized framework for online fuzzy weighting is presented, which incrementally calculates the membership of each coming sample by taking into account the memberships of previous samples in a pairwise manner. Such pairwise-distance based scheme can not only identify possible outliers, but also show well adaptation to the sequentially received samples in online setting. We apply it to online passive-aggressive(PA) algorithm in a direct way. The resulting Fuzzy Passive-Aggressive(FPA) achieves comparable classification accuracy with benchmark incremental SVM, while still enjoying the time efficiency of simple PA, which is a Perceptron-like algorithm. Besides, FPA exhibits the best performance among PA family, which makes it a robust and efficient alternative to PA, in order to deal with unavoidable outliers in large-scale or high-dimensional real datasets.5. Two-dimensional Shifting Discriminant Analysis(2DSDA) and Transformative Discriminant Analysis(TransDA) are presented respectively. The former utilizes shift operations regarding adjacent rows or columns to preserve the local geometry structures in images in terms of the covariance information, while the latter extends the former from the unilateral form to the bilateral one through sequential projections. Moreover, by controlling the parameter of the shift-operation, our methods not only achieve promising performance in LDA family, but also unify LDA, 2DLDA and even the 2-order multi-linear LDA with a flexible mathematical expression.6. Unconstrained Clustering is discussed under the framework of possibilistic theory in order to resolve the limitations resulting from the probabilistic constraints. By introducing the robust estimators and information theoretic criteria, the robust gaussian clustering(RGC) and fully unsupervised possibilistic entropy clustering(FUPEC) methods are proposed respectively, which have clear physical meaning and overcome the sensitivity to the noise. Moreover, FUPEC can automatically determine the number of the clusters and provide effective estimation of the prototype, even when the clusters vary significantly in size and shape and the data set is contaminated by heavy noise.Inspired by Descartes’ “Discourse on the Method”, we evaluate the reasonableness of related problem decomposition and highlight our research thoughts, which indicate the establishment of the Methodology of Specific Radar Emitter Identification.