A NOVEL DETECTION-ESTIMATION SCHEME FOR NOISY NMR SIGNALS - APPLICATIONS TO DELAYED ACQUISITION DATA

Many potentially interesting and useful classes of NMR experiments gen erate data for which conventional spectral estimation and quantificati on aa the Fourier transform are unsatisfactory. In particular, recentl y introduced solid-state NMR experiments which involve long delays bef ore data acquisition fall into this category, as the free induction de cays are heavily ''truncated'' and have low signal-to-noise ratios. A novel detection-estimation scheme is introduced in order to analyze da ta from such experiments and others where the sensitivity is low and/o r the data record is strongly damped or truncated, Based on the assump tion of exponential data modeling, the number of signals present is fi rst detected using criteria derived from information theory and the sp ectral parameters are then estimated using the matrix pencil method, M onte Carlo simulations and experimental. applications are carried out to demonstrate its superior statistical and computational performances and its general applicability to delayed acquisition data. Over the r ange of note levels investigated, it is found that this approach is es sentially near-optimal in the sense that the estimated spectral parame ters have biases almost equal to zero and variances very close to thei r theoretical Cramer-Rao lower bounds. Compared to the popular method of linear prediction with singular value decomposition, this method no t only improves the estimation accuracy (by a factor of 2-4) with a lo wer ''break-down'' signal-to-noise threshold (approximate to 1.5 dB), but also reduces the computational cost by about an order of magnitude , It also holds great promise in effectively reducing truncation artif acts. It is concluded that this approach not only facilitates the anal ysis of delayed acquisition data, but can also become a valuable tool in the routine quantification of general NMR spectra, ia listing of pr ograms is also included in the Appendix. (C) 1997 Academic Press.