Deinterleaving Models And Algorithms For Advanced Radar Emitter Signals

The deinterleaving of radar emitter signals is a crucial technique in signal processing of Electronic Intelligence, which directly determines the performance of electronic reconnaissance equipment. The classical methods of pulse trains deinterleaving are generally based on five conventional parameters, i.e., time of arrival (TOA), radio frequency (RF), pulse width (PW), pulse amplitude (PA) and direction of arrival (DOA). In the low dense environment composed of conventional radar signals, these methods are effective and can obtain satisfactory results. However, as the countermeasure activities in modern electronic warfare are becoming more and more drastic, the density of electromagnetism signals environment arrives at the degree of mega, and the new-style modern radars with complicated systems, which have the ability to change many parameters of signal in manner of sliding, hopping, even agility, are becoming mainstream in the equipped radars. In such a high dense, complex, and changeful signal environment, pulse loss and parameter overlapping are serious, and the signal regularities utilized by the classical deinterleaving methods are badly destroyed. As a result, the performances of these conventional methods descend rapidly. At present, signal sorting, especially the deinterleaving of advanced radar signals, has become the primary bottleneck of signal processing in electronic reconnaissance and confined farther improvement of the performance of electronic countermeasure equipment.This dissertation specializes in the issues concerned with the deinterleaving of advanced radar emitter signal. Especially, three important facets, feature extraction and analysis for radar emitter signal, novel model and algorithm of deinterleaving, and evaluating the confidence level of deinterleaving results are lucubrated. The research fruits are as follows.1. In order to reduce the overlapping degree of the signal space spanned by five classical parameters, the feature extraction algorithms of derived characters of instantaneous frequency (IF) and ambiguity function main ridge (AFMR) slice are proposed. Via these algorithms, some new parameters can be extracted, so the more effective deinterleaving characteristics vector (DCV) can be constructed.In view of the non-stationary characteristic of the majority of advanced radar emitter signals, the improved instantaneous autocorrelation algorithm is performed to extract IFs of those typical radar signals at the beginning of Chapter 2. Then, a cascade normalization based feature extraction method is put forward so that the characteristics derived from IF can be extracted. The results of classification experiments based on hierarchical decision and kernelized fuzzy clustering with validity assessment (KFCVA) both demonstrate that, the derived features of IF can reflect the differences of signals with different modulation types, and have a good ability to resist noise.Considering the specific effect of ambiguity function in characterizing the inherent structure of a signal, a novel fractional autocorrelation and moment description based approach to extract the features of AFMR slice is presented in Chapter 3. Via the proposed method, a 3-D characteristics vector, which describes the direction of AFMR, the centroid of the AFMR slice and the inertia radius relative to the centroid, is obtained. It shows that, the features of AFMR slice have strong compactness within clusters and good performance of anti-noise.2. Aiming at the issue of complicated characteristics distribution and undistinguishable boundary between clusters of advanced radar signals, two deinterleaving models which are both based upon kernel clustering are proposed. And the theoretical problems relative to these models are discussed in detail.The first model grounds on the kernelized hard clustering (KHC) algorithm. In which, we design an efficient similarity-based validity index to determine the optimal number of emitters in the deinterleaved pulses stream, and bring forward the grid-based outlier detection algorithm to detect the noise pulses. Also, we introduce the small-class cleaning technique to control the false alarm rate in the phase of post-processing, and discuss the possibility of parallel arithmetic of the proposed model laconically.In the second model, the fuzzy processing is adopted so that the complicated boundary between clusters can be disposed in a more effective manner. Firstly, we generalize six noted cluster validity indices for standard fuzzy c-means algorithm into high-dimensional kernel space for the purpose of acquiring their corresponding kernelized expressions. Then, the KFCVA algorithm is presented and is taken as the prototype of the deinterleaving model. In this model, the membership-based outlier detection method is proposed, which can be ulteriorly combined with the kernelized validity index, so that new assessment criteria can be constructed. In the phase of post-processing, besides the small-class cleaning, the concept of confidence level of fuzziness is used to evaluate the compactness of each obtained emitter.Considering multifarious influencing factors, such as the number of emitters, the count of intercepted pulses, pulse loss rate, the pulse number of each intercepted emitter, the number of outlier, signal-noise-ratio and the type of modulation etc., the performances of the deinterleaving models presented above and some different combinations of features are studied in depth. The results show that, the kernel method based models have superior performance in the deinterleaving of advanced radar emitter signals. They are less affected by the above influencing factors, and the lowest pulse number of each intercepted emitter is less than that of conventional deinterleaving method. Additionally, the simulation experiments give some useful conclusions concerning the performance of different DCV, which offer the corresponding guidance and basis to explore the new effective DCV.3. In order to analyse and evaluate the confidence level of deinterleaving results, this dissertation presents a separability test method refer to as Sorted-DOA Difference Test (SDDT) and introduces the concept of confidence level of DOA (CLDOA). It shows that, SDDT has the ability to test the separability of each pulse cluster obtained via the deinterleaving algorithm, and its resolution of test increases rapidly with the augment of the pulse quantities. In addition, CLDOA gives a good assessment to confidence degree of the considered pulse cluster in the DOA dimension. So the information about deinterleaving reliability can be achieved by SDDT and CLDOA.This work is supported by the National Natural Science Foundation of China (No. 60572143) and the National Electronic Warfare Laboratory Foundation.