A robust method for wideband signal separation

A novel approach to solving the problem of signal separation under model uncertainties and unknown source signal characteristics is proposed. The approach features the incorporation of blind identification with clustering techniques. The approach is capable of estimating source locations and source signals under various uncertain conditions including unknown sensor gains, unknown combinations of near-field and far-field sources, unknown combinations of wideband and narrowband sources, unknown source spectral characteristics (their spectra may be overlapping or non-overlapping), and unknown number of signals.<<ETX>>