Please, cite this online document as:
Vasini E.M., Cobas C., Sykora S.,
PcBc  A Novel Automatic Phase and Baseline Correction Algorithm,
Presentation at XLVI GIDRM, Fisciano (Italy), September 2729, 2017.
DOI: 10.3247/SL6Nmr17.005.
Abstract
Manually, phase (PC) and baseline (BC) corrections in 1D NMR spectra are done sequentially one after another. In addition, the Bc is always done only on the 'real' part of the spectrum where the spectral peaks are much narrower than in the imaginary part and therefore interfere less with the visual assessment of the final quality. However, even when the computer is used just as a display tool, one notices that there is a mutual interference between the two corrections, especially close to the final solution. One often hesitates between a range of possible 'results', uncertain which is the best one (or the most correct one). Moreover, the fact that baseline correction is not done on the imaginary part means that if one tries to iterate the whole process, any change in phase parameters brings back into the displayed real part the uncorrected baseline artifacts present in the imaginary part, thus making iterations of the whole procedure problematic.
In the past, much work was done on automatic (i.e., objective) phase and baseline corrections of NMR spectra. As a result, there exists a number of algorithms that work quite well, but sometimes exhibit with residual problems. So far, such algorithms were emulating the manual procedure. Basically, they all 'fit' the parameters which describe the phase (ph0, ph1) and the baseline (various parametrized models) so as to maximize some 'final quality' assessment of the corrected spectrum. Historically, the 'quality functions' included peak heights, negative peak lobes, DISPA pattern symmetry, selected baseline points, peak ablation, etc. Clearly, the quality function Q(spectrum) is the heart of the problem because its maximum should match our subjective (and therefore ill defined) expectations about how a corrected spectrum should look. A typical case of fuzzy mathematical modelling!
Here we present a new type of the quality function Q, one based on the characteristics of a spectrum's histogram, which are very 'sensitive' to both phase and baseline distortions. On this basis we have devised a new algorithm which has also other strong points: first of all, it handles BOTH corrections simultaneously (PcBc rather than Pc + Bc), and it applies the Bc to BOTH the real and the imaginary parts of the spectrum.
Here we describe the fully automatic algorithm we have developed starting from these premises, and illustrate the results achieved on a number of 1D spectra of various kinds.
References
A full collection of references would take up a couple of pages. We think that the audience of this presentation is well aware of the practical problem, and also of some of its current solutions. On the other hand, the idea of using histograms for a joint PcBc adjustment did not so far appear anywhere in the literature. We will make up for the lack of references as soon as possible, probably in a regular paper.
