Radar detection for non-stationary Doppler signal in one burst based on information geometry: Distance between paths on covariance matrices manifold

Classical Radar processing for non-stationary signal, corresponding to fast time variation of Doppler Spectrum in one burst, is no longer optimal. This phenomenon could be observed for high speed or abrupt Doppler variations of clutter or target signal but also in case of target migration during the burst duration due to high range resolution. We propose new Radar Doppler processing assuming that each non-stationary signal in one burst can be split into several short signals with less Doppler resolution but locally stationary, represented by time series of Toeplitz covariance matrices. In Information Geometry (IG) framework, these time series could be defined as a geodesic path (or geodesic polygon in discrete case) on covariance Toeplitz Hermitian Positive Definite matrix manifold. For this micro-Doppler analysis, we generalize the Fréchet distance between two curves in the plane to geodesic paths in abstract IG metric spaces of covariance matrix manifold. This approach is used for robust detection of target in case of non-stationary Time-Doppler spectrum (NS-OS-HDR-CFAR).