Minimum Mean Square Distance Estimation of Subspaces in presence of Gaussian sources with application to STAP detection

Abstract

We consider the problem of signal subspace estimation from a given sample set. We rely on the Bayesian framework developped in [1] to obtain minimum mean square distance (MMSD) estimators, which minimize the expected distance between the true projection matrix UUT and its estimate ÛÛT. In this work, we extend the estimators of [1] to the context of linear model with Gaussian sources, with respectively Bingham and von Mises Fisher priors for the basis Ū. Numerical simulations are given in order to assess the performance of the proposed estimator. The interest of the considered approach is finally illustrated on a real data set, for a Space Time Adaptive Processing (STAP) application.