Offline Processing of Bruker Digitally Filtered NMR Data

January 16^th 2000

Dear Barry,

The benefits of the oversampling procedure applied to NMR data is well known and certainly one of the most valuable aspects is the increase of the effective dynamic range of a spectrum. However, this requires collecting a FID with a great number of points to provide sufficient digital resolution. The large FID is digitally filtered to remove the signal outside the spectral region of interest and decimated to the desired spectral width, the whole procedure taking place on-the-fly by means of high-speed digital signal processors [1]. Unfortunately the procedure may have its own drawbacks. Commercial digital filters are proprietary, so one does not really know what has been done to the data, and some resultant processed spectra can assume a strange form after digital filtering. This seems to be the case with digital filtering employed on Bruker Avance spectrometers (Figure 1).

Figure 1. Beginning of typical digitally processed FID obtained on a Bruker DMX spectrometer.

Shown on the Figure 1 is the beginning of a typical digitally processed FID obtained on a Bruker DMX spectrometer. A rising signal starting from zero at the first data point and achieving a maximum at approximately the 60-th point is undoubtedly an artifact of digital filtering. The FID is correctly processed on XWINNMR, a component of the Bruker software suite but standard Fourier transformation leads to a spectrum with wiggles as shown below (Figure 2).

Figure 2. ¹H NMR spectrum of phenylbutyric acid in DMSO (DMX-400) obtained by straightforward Fourier transformation of digitally processed FID.

The observation that the spectrum shown in Figure 2. can be made to resemble a normal spectrum by applying several thousand degrees of linear phase correction gave a key to correct processing of such data. As was shown by W. M. Westler and F. Abildgaard [2] an appropriate number of initial points of the FID should be circularly left shifted before Fourier transformation. The resulting spectrum is almost identical to that processed by XWINNMR (Figure 3A). It is important to perform the circular left shifting after zero-filling and/or apodization to keep the flat baseline. Shown in Figure 3B is the spectrum obtained when exponential multiplication (with LB = 1 Hz) was applied after the circular left shift. Retaining the non-zero points at the end of FID is however a very cumbersome business. Besides this can interact in an unexpected way with some offline processing steps. For example the spectrum shown in Figure 3B would be resulted from the spectrum shown in Figure 3A by inverse Fourier transformation followed by exponential multiplication (with LB = 1 Hz).

Figure 3. ¹H NMR spectrum of phenylbutyric acid in DMSO (DMX-400). A - Fourier transformed by XWINMR, B - exponential multiplication (with LB = 1 Hz) was done after circular left shift of initial points, C- exponential multiplication was done after ‘digital to analog' conversion by XWINNMR, D - exponential multiplication was done after digital FID conversion by our technique

Our challenge was to think of a conversion of some kind way that would allow us to treat afterwards digitally filtered data like 'normal' data. after conversion of some kind. Bruker's own conversion routine "convdta" is applicable only to acquisition data [3] and sometimes significantly distorts baseline (Figure 3C) and appears to be unsuitable.

We would like to describe our empirical treatment that was found to eliminate all resulting nuisances generated by digital filtering. The basic steps for the digitally filtered FID are the following:

1. zero-filling, optional zero-filling and/or apodization,
3. circular left shift of appropriate number of points,
4. Fourier transformation,
5. phasing
6. extracting real part and its inverse Fourier transformation,
7. truncating the obtained FID

The resultant FID can be processed further without any complications. The first step of the procedure is necessary to keep the number of spectral points while steps 2-4 are is thatose proposed by W. M. Westler and F. Abildgaard [2].

Shown in Figure 3D is the Fourier transformed result of "converted" data after the same exponential multiplication (with LB = 1 Hz) of such "converted" data. Please note that a flat baseline is retained as displayed in Figure 3A. The quality of the baseline after this conversion is dependent on phasing quality. The best baseline is achieved when the purely adsorptive part is taken as real (Figure 3D) and the worst one - when the purely dispersive part is taken as real (result is very similar to Figure 3B). The protocol for conversion of Bruker-processed 1R files is very similar and has the advantage that the result does not depend on the effectiveness of the phasing routine.

More and more institutions and laboratories are utilizing third party tools for access and processing of NMR data. We believe that a thorough examination of the quality of data generated away from the spectrometer using these tools is necessary to take into account different approaches taken by the vendors.

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. W. M. Westler and F. Abildgaard, DMX Digital Filters and Non-Bruker Offline Processing II, private communication, 1995.
3. XWINNMR Software Manual, Bruker Analytische Messetechnik GmbH, p131, 1995