Time delay estimation using a signal subspace model

Classical methods for estimating time difference of arrival (TDOA) are based on generalized cross-correlation. Data from the two channels are transformed to the frequency domain to form the conjugate product XY/sup */. After appropriate weighting, an inverse DFT is performed and the peak of the resulting generalized cross-correlation function is located in the time domain. We show that the data product XY/sup */ satisfies an appropriate signal subspace model and we replace the IDFT with a signal subspace technique such as MUSIC. The result is an enhanced ability to locate the peak. Further, application of methods such as root MUSIC or ESPRIT produce direct numerical estimates for the time delay without the need to search for a peak. Results on simulated data using a parabolic equation (PE) ocean acoustic propagation model are presented.