Gradient flow bearing estimation with blind identification of multiple signals and interference

We present a technique for reliably estimating the 3D direction cosines of a broadband travelling wave impinging on an array of four sensors of dimensions smaller than the shortest wavelength in the source. Gradient flow converts observed mixtures of delayed source signals into instantaneous linear mixtures of their temporal derivatives through observation of spatial gradients of the field. This formulation is equivalent to independent component analysis (ICA), where the mixing matrix directly yields the direction cosines. Experiments with acoustic data from a microphone array show improved bearing accuracy through second order blind identification (SOBI) of band-limited Gaussian signal from interfering sources.