STATISTICS AND LIMITS OF LINEAR-PREDICTION QUANTIFICATION OF MAGNETIC-RESONANCE SPECTRAL PARAMETERS

Linear prediction has attracted considerable interest as an alternativ e approach to fast Fourier transform for quantification of NMR signals . Based on a Monte Carlo method, we report a statistical investigation on two forms of linear prediction to examine the accuracy and precisi on with which the positions, linewidths, intensities, and phases of NM R lines in complex, crowded spectra may be determined. The two forms o f linear prediction differ in the methods used to obtain the predictio n coefficients, chosen to be least squares in one and total least squa res in the other. Both methods are shown to be very sensitive to the l evel of noise in the experimental spectrum, as well as to the predicti on order chosen for the analysis. Linear-prediction estimates of the f requencies are usually very reliable, while estimates of the other par ameters such as linewidths and intensities are usually poorer. (C) 199 4 Academic Press,Inc.