Suppression of Noise to Extract the Intrinsic Frequency Variation from an Ultrasonic Echo

Abstract

We suppress the effects of noise and extract the intrinsic frequency variation from an ultrasonic echo using a singular value decomposition of the Wigner distribution of the signal. This method performs a non-linear filtering on the echo signal to reduce greatly the interference effects between overlapping echoes - correlated noise, as well as uncorrelated electronic noise. We can then evaluate the local frequency both accurately (i.e. without bias by noise) and efliciently (i.e. with a small data set). This analysis is carried out on the Wigner time-frequency function calculated from a segment of echo data. With the help of the singular value decomposition of thc Wigner distribution, we reject echo segments corrupted by interference and noise. In this way, correlated interference is suppressed much more efficiently than by averaging. We illustrate this approach on the traditional problem of estimating the attenuation slope from the downward shift in mean frequency with depth. Our analysis of simulated echoes shows that accurate estimates can be obtained from a single A-line. Moreover, we obtain the correct answer with two orders of magnitude less data than the conventional Fourier approach. The resulting savings in data acquisition time and computation time are substantial.