Time delay estimation (TDE) is commonly performed in practice by cross
correlation of digitized echo signals. Since time delays are generally
not integral multiples of the sampling period, the location of the la
rgest sample of the crosscorrelation function (ccf) is an inexact esti
mator of the location of the peak. Therefore, one must interpolate bet
ween the samples of the ccf to improve the estimation precision. Using
theory and simulations, we review and compare the performance of seve
ral methods for interpolation of the ccf. The maximum likelihood appro
ach to interpolation is the application of a reconstruction filter to
the discrete ccf. However, this method can only be approximated in pra
ctice and can be computationally intensive. For these reasons, a simpl
e method is widely used that involves fitting a parabola (or other cur
ve) to samples of the ccf in the neighborhood of its peak. We describe
and compare two curve-fitting methods: parabolic and cosine interpola
tion. Curve-fitting interpolation can yield biased time-delay estimate
s, which may preclude the use of these methods in some applications. T
he artifactual effect of these bias errors on elasticity imaging by el
astography is discussed. We demonstrate that reconstructive interpolat
ion is unbiased. An iterative implementation of the reconstruction pro
cedure is proposed that can reduce the computation time significantly.
(C) 1995 Academic Press, Inc.