Evaluation of algorithms for automated phase correction of NMR spectra

In our attempt to fully automate the data acquisition and processing of NMR analysis of dissolved synthetic polymers, phase correction was found to be the most challenging aspect. Several approaches in literature were evaluated but none of these was found to be capable of phasing NMR spectra with sufficient robustness and high enough accuracy to fully eliminate intervention by a human operator. Step by step, aspects from the process of manual/visual phase correction were translated into mathematical concepts and evaluated. This included area minimization, peak height maximization, negative peak minimization and baseline correction. It was found that not one single approach would lead to acceptable results but that a combination of aspects was required, in line again with the process of manual phase correction. The combination of baseline correction, area minimization and negative area penalization was found to give the desired results. The robustness was found to be 100% which means that the correct zeroth order and first order phasing parameters are returned independent of the position of the starting point of the search in this parameter space. When applied to high signal-to-noise proton spectra, the accuracy was such that the returned phasing parameters were within a distance of 0.1-0.4 degrees in the two dimensional parameter space which resulted in an average error of 0.1% in calculated properties such as copolymer composition and end groups.