DOA estimation in the presence of modeling errors, the Global Matched Filter approach

Taking into account the modeling errors on the array manifold in direction of arrival (DOA) estimation is a crucial issue in practice. It is regularly considered in the literature but no satisfactory solution is known and only sensitivity or performance analyses of existing methods (usually subspace-based) are proposed mostly in the exact data case. We show that projecting the observations on the nominal array manifold is a solution to this problem that greatly reduces the degradations in performance induced by modeling errors. Except for some very specific array geometries, this projection is unfeasible in practice, hence the interest of the Global Matched Filter (GMF) which, when using as inputs a finite number of beamformer-outputs, benefits from the projection that is implicitly performed by the conventional beamformer. Its performance is quite remarkable.