IMPROVEMENTS OF QUANTITATION BY USING THE CADZOW ENHANCEMENT PROCEDURE PRIOR TO ANY LINEAR-PREDICTION METHODS

Improvements of the Cadzow enhancement procedure (EP) are reported. A numerical study showed that once EP has properly converged, it results in a near-optimum signal fitting the noisy data in the least-squares sense. At convergence, the estimated signal involves purely exponentially damped sinusoids; then any linear-prediction methods can be used indifferently in the subsequent quantitation step. Moreover, this final step needs only a few samples. Amplitude estimates have negligible bias even for low SNR and their standard deviations are nearly equal to their Cramer-Rao lower bounds. This is illustrated in a Monte Carlo simulation and an in vitro study using three different SVD-based quantitation methods (LPSVD, HSVD, TLS). EP used prior to any linear-prediction methods leads to an important reduction of the threshold SNR and results in near-optimal automatic quantitation methods. (C) 1994 Academic Press, Inc.