A popular technique for time delay estimation is to use an FIR filter to mo
del the time difference and the filter weights are interpolated with a sine
function to obtain the delay estimate. However, the sine interpolator requ
ires a sufficiently long filter length for accurate delay estimation. In th
is paper, we propose to process the filter weights via a least-squares-base
d method in order to acquire optimum delay estimates even with short filter
lengths. (C) 2001 Elsevier Science B.V. All rights reserved.