Source localization in the presence of unknown colored noise

The problem of the localization of multiple sources in the presence of arbitrary noise of unknown spatial spectral density is considered. The arbitrary spectrum of the noise is identified by a class of models that covers the set of rational spectra. The array geometry has to be known but may be arbitrary. The sources are assumed to be uncorrelated. We propose a unsophisticated approach the algorithmic part of which consists of a standard linear programming algorithm. It is a non-parametric estimation approach where the observations are decomposed on the association of different bases of candidate models. The computational complexity of the method is reasonable, the performances are illustrated by simulations and appear to be quite good. Theoretical analysis of these performances are under investigations.