Abstract
Estimates of time difference of arrival (TDoA) can be used in audio (and radio) for localization, positioning and navigation. Such estimates are typically obtained from one or several microphone (or antenna) pairs. In this paper we demonstrate that better TDoA estimates can be obtained by using sub-sample correlation. The main idea is to first obtain an initial estimate of the TDoA estimate and then refining these using a local optimization over sub-sample shifts. For a sub-sample shift one calculates the interpolated signal and evaluates the correlation criterion and the derivatives of the correlation with respect to sub-sample shifts. These are then used in an efficient implementation of determining optimal TDoA estimates using a local optimization of the correlation criterion. The method has been implemented and tested on real data, where it is demonstrated that using the improved TDoA estimates in a system for simultaneously estimating microphone and sound source positions gives better precision and lower residuals in the results.
