NMR Newsletter, October 1999

Improved Baseline Correction of FT NMR Spectra

Sergey Golotvin and Antony Williams
September 10^th 1999

Dear Barry,

As we all know, baseline correction can be a very essential step to obtain a high quality NMR spectra in some cases. Rolling baselines can make it difficult to identify peaks, as well as introduce a significant error into any quantitative measurements. The most convenient way of correcting baseline distortion is to construct a model of the baseline in the frequency domain and subtract it from the spectrum [1]. To this end, a majority of NMR desktop software allow the user to manually set the points belonging to baseline and interpolate between them using analytical functions to model the baseline completely. While the results can often be good enough the method requires manual intervention and cannot be used for batch processing. On the other hand quality of automatic procedures is rarely sufficient when the baseline has severe distortions. The failures are generally due to both inadequate types of analytical functions used for modeling and poor recognition of the baseline.

We have recently developed a new method for baseline correction which is governed by two easily adjustable parameters. The essence of the method is attributed to two parts: baseline modeling and baseline recognition.

It is a natural desire for visual inspection to have the baseline be a flat straight line with no constant offset. Our idea is to use the smoothed spectrum to model the baseline. We have used an averaging of neighboring data points of the spectrum, an operation equivalent to convolution with a rectangular function. The number N of points to average or the width of the rectangle is the first parameter of our procedure. It is evident that subtraction of the smoothed spectrum from the original one can give a relatively flat straight line, but it makes sense only for the area that does not contain peaks. As a result we have a fragmented model, which needs interpolation over the area where peaks are present. Interpolation can be as simple as connection by straight lines, or as complicated as we care to make it. The second important point is know the recognition of the baseline separate from the peaks.

Since the ultimate definition of "baseline" is "not containing peaks" it is natural to employ for its recognition the same tool that is used for peak identification. We have developed the following procedure. To decide whether the i-th point belongs to the baseline it is placed in the center of a rectangle with a width of N spectral points.. Among these N points the minimal and maximal values are found. If their difference does not exceed the noise standard deviation multiplied by a definite factor (the second parameter of our technique) the i-th point is considered to belong to baseline. Finally the baseline model is then subtracted from the spectra.

Figures 1 and 2 below illustrate the results obtained with our approach. A characteristic rolling baseline can be observed in the spectrum in Figure 1 and can be accounted for by corruption of the first several points of FID. A well known algorithm was applied to this problem [2], and its performance is comparable with that of our method. However for the second case (Figure 2.) our approach gives a far superior result. Due to adequate baseline modeling and recognition our technique has been shown to be superior and therefore of inherent value for desktop post-processing.

Figure 1. Original experimental (top) and baseline corrected (bottom) ¹H NMR spectra of cyclopentanol in CDCl₃. The rolling baseline is due to corrupted first several points of FID.

Figure 2. Original (top) and baseline corrected (bottom) ¹H NMR spectra of 6-bromo-1,5-dinitrobicyclo[3.3.1]non-6-ene-3-carboxylic acid in DMSO-D₆. The hump in the original spectrum is due to non-optimal water suppression.

Yours sincerely,
Sergey Golotvin and Antony Williams,
Advanced Chemistry Development

References:
1. Hoch J.C. and Stern A.S.: NMR data processing, p. 196, New-York, 1996, Wiley-Liss.
2. Heuer A. and Haeberlen U., J.Magn.Reson. 85, 79 (1989)

Download the publication in RTF format as 122 Kb zip file.