parameters of the ultrasonic signal, linked to the pulse spec-
trum and its variations with depth (instantaneous frequency,
bandwidth, cepstral and correlation analysis, ¼). The
extracted parameters corresponded either to a whole regis-
ter, or to different segments of a register obtained through a
sliding window. This allowed us to observe feasible varia-
tions on one register. To illustrate this, Fig. 3 shows the
average dependency between the ®rst value of the cepstrum
sequence and the water/cement percentage. Cepstral analy-
sis is one standard procedure for extracting pulse informa-
tion from material or tissue structural noise [1]. The
cepstrum is de®ned as the inverse Fourier transform of the
spectrum logarithm. The convolution of the ultrasonic pulse
with the material re¯ectivity (time-domain) implies two
multiplicative factors in the spectrum (frequency domain):
a smooth variation factor (pulse spectrum) and a multipli-
cative noise factor (re¯ectivity spectrum). By taking the
logarithm, we convert the multiplicative noise in an additive
one, so the ®rst cepstrum coef®cients have information