Multivariate analysis of high resolution nuclear magnetic resonance spectra in discrete Fourier transform domain

Abstract

This study presents the multivariable analysis of high resolution nuclear magnetic resonance (NMR) spectra in discrete Fourier transform (DFT) domain. Influences of external environment usually cause random shift that jeopardizes the predictive accuracy of NMR spectra to a certain extent. This study, firstly, theoretically proves that the random shift of NMR spectra could be restrained effectively in the DFT domain. Then, simulated and experimental data were employed to verify the restraining ability of the proposed method. Recognition analysis of NMR spectra dataset for the classification of rapeseed oil and regression analyses of NMR spectra dataset for predicting three critical properties of fish oil was considered in the experimental data analyses. Finally, according to the distribution characteristic of DFT coefficients, the compression of NMR spectra in DFT domain was discussed. Compared with original NMR spectra obtained from DFT of free induction decay (FID) and traditional equal interval integral (EII) method which is usually used to restrain random shift of NMR spectra, the proposed method was proved to be more powerful not only in improving the accuracy of the multivariable analysis of NMR spectra, but also in compressing the high dimensionality of NMR spectra.