Time domain approach to wideband source location

Consideration is given to the multiple source location problem, using a linear array of equispaced sensors. The method works in the time domain, and uses as a starting point the matrix covariance sequence of the vector of sensor outputs. The spectra of the source signals are assumed to be unknown and rational, but the approach can be easily extended to spectra that are flat in a fixed and known frequency band. The procedure is based on the block Hankel matrix built on the estimated matrix covariance sequence. The properties of this matrix are investigated, and it is shown that its rank is deficient. As for the spectral density matrix in the narrowband case, one can thus define a signal subspace. The identification of the vectors that generate this subspace and the location of the sources by the maxima of the projections of these vectors on the signal subspace then remain.<>