Empirical mode decomposition facilitating Bayesian Target Tracking for Cognitive Radar

A method based on the Empirical Mode Decomposition (EMD) is addressed for the facilitation of Bayesian Target Tracker (BTT), the central unit for a Cognitive Radar receiver. As in all Bayesian methods, the BTT heavily relies on the availability of accurate a priori information on the operating conditions. To this end, a Radar Scene Analyzer (RSA) is a crucial part of a Cognitive Radar in that it provides the a priori knowledge the BTT requires. Herein, a complementary RSA structure is developed building on the statistical properties of the EMD on fractal Gaussian processes. In particular, EMD is applied to the measured radar data, yielding the Intrinsic Mode Functions (IMF). At absence of a target, coherent sea clutter data exhibit fractal Gaussian character, which manifests itself in the second order statistical properties of IMFs. This is exploited to form a “Null Hypothesis”, which is used for grouping the IMFs into two subsets on the basis of an accept/reject procedure. Hence the refinements of the raw radar returns are constructed from the superpositions of accepted or rejected IMFs. Effectively, the likelihood function for the target+clutter case can be remarkably distinguished in statistical terms from that for the clutter-alone case, so as to facilitate the Bayesian target tracker. The performance of the method is visually illustrated using the live-recorded McMaster IPIX dataset, and numerically assessed by the Kullback-Leibler distance.