Application of minimum coherence criterion for blind source separation

The Blind Source Separation (BSS) is performed using multiple observations of mixtures of dependent or independent astrophysical sources of diffuse microwave radiation. Two separation methods are used and compared: (a)FastICA, which are already been used for similar astrophysical data, (b) the minimum coherence criterion. Results show that both methods give better results in the case of independence aming the sources that in the case of dependence. In particular, in the case of the data sets considered, method (b) (newly applied to similar astrophysical data) generally gives better source estimates than method (a), especially in the case of correlated sources, where FastICA is not optimal because of its assumption of orthogonality among the sources.