The generalization of narrowband localization methods to broadband environments via parametrization of the spatial correlation matrix

The need to localize a radiating signal source is necessitated in applications ranging from distant talker speech pick-up to automatic video camera steering. The majority of literature detailing source localization is presented in the narrowband signal context; however, acoustic signals are naturally very broadband. As a result, classical narrowband techniques do not apply in the strict sense, and as a result, the broadband localization literature is yet unrefined. This paper details the manner in which classical narrowband localization methods may be generalized to broadband signal environments - specifically, a nonlinear parametrization of the spatial correlation matrix allows one to apply steered beamforming, minimum variance, subspace, and linear predictive spectral estimation methods to the broadband localization problem.