We have recently shown(I, 2) that using one form of nonuniform selective data sampling during acquisition, which we call exponential sampling, one should be able to obtain increased resolution in two-dimensional NMR spectra. The use of exponential sampling was demonstrated using one-dimensional spectra acting as models for the second dimension in 2D NMR spectra. The data were acquired normally and the appropriate data sampling was produced before maximum entropy method (MEM) data processing on a separate computer. We have also demonstrated the first application of exponential sampling to a 2D NMR spectrum of a protein (3, 4). It is for such spectra of biological macromolecules that the method is intended, exponential sampling in t, followed by suitable data processing gives an improvement on conventional sampling of the same number of t, points even when the latter is also followed by MEM data processing. Exponential sampling is ideally suited to experiments where the signals of interest are in-phase; i.e., the free induction decay is cosine modulated in tl . This includes, for example, the NOESY experiment. Where the signal does not simply decay exponentially in tl (e.g., COSY) modified sampling schemes are neces=-u-Y (2). When processing very noisy data sets using MEM the relative intensities of peaks are distorted. This is a problem, particularly when processing 2D NOESY spectra where the relative intensities are used to determine ‘H-‘H distance constraints. We have, therefore, explored the possibility of using other data processing methods in conjunction with our methods of selective data sampling. In this note we show that exponential sampling can also be used in conjunction with the CLEAN method (3). CLEAN was introduced in the field of astronomy by Hogbom (5) and later applied by others to NMR data (6-8). A discussion of its application to radioastronomical data has recently been given (9) and we have recently made a comparison of the results obtainable after conventional sampling using CLEAN, MEM, and linear predictive singular value decomposition (3). When CLEAN is used to process conventionally sampled truncated data, a “mask” or “beam” is constructed by Fourier transforming a step function corresponding to the length of acquisition of the truncated signal followed by zeros (7). The mask has characteristic broadness and “sine wiggles.” This corrupted frequency-domain mask is then subtracted from the corrupted spectrum, produced by a Fourier transform of
Read full abstract
Mar 1, 1991
Peter A Gorry 
Open Access
