Abstract: Based on the prony algorithm, Delsuc's approach for prediction of NMR spectra is more useful in practice. However, missing lines and spurious signals will be und time by time in the very noisy spectra and one must get the periori knowleage of the number of peaks in a spectrum.
In this paper, the recursive process has been traced. It is found that the criterion r_min in Delsuc's approach is very difficult to be chosen. Some peaks will be missed if r_min is too small. Otherwise, the spursious signals will appear if it becomes larger. Meanwhile, the coherency of sampling points in weak signal is less than that in strong signals due to the influence of noise which will increase the instability of data in recursive process. Here an additional criterion of peak area is adopted to remove the drawbacks of Delsuc's approach The success of this suggestion is proved by simulation and experiment.

Key words: Peak area criterion, NMR, LPSVD