Novel Spectral Processing Methods in NMR – Wavelet Transform and Pattern Based Analysis

Abstract

NMR spectroscopy provides a variety of information leading to an understanding of the properties of different materials. To extract this information from the experiments, one needs to have noise and artifacts free spectra and part of the present thesis examines ways of obtaining artifacts free spectra using wavelet transforms. The thesis addresses the inherent problem of low sensitivity of spin noise spectra and examines the utility of wavelet transform to mitigate this problem by distinguishing real peaks from the circuit-noise contaminated data. Suppression of the random circuit noise and the consequent enhancement of the correlated nuclear spin noise signal have been demonstrated with discrete wavelet transform. Spectra of both 1H and 13C nuclear spins have been considered and significant signal enhancements in both the cases have been observed. A detailed analysis of several possible wavelet, thresholding and decomposition solutions have been made to obtain the optimum conditions for signal enhancement. It is observed that the application of wavelet transform leaves the spin noise signal line shape essentially unchanged, which is an advantage for several applications involving spin noise spectra. Next, baseline distortions encountered in 1D NMR is considered and an algorithm is proposed for simultaneously denoising and base-line correction. Application of wavelet transform filters signals into high and low frequency components. The high frequency part contains mainly the signals of relevance as well as noise. Repeated application of wavelet transform leaves a low frequency component which has essentially the base-line information, that can be corrected in several ways. The high frequency component can also be subjected to noise-reduction algorithms. Combining these two parts back again in a reverse process, gives a spectrum that is essentially free of base-line distortions and with reduced noise. Another source of artifacts in modern NMR spectroscopy is the t1-noise. Though mitigated to a significant extent, t1-noise is still a problem in several experiments. Particularly in NOESY experiments where the cross-peak intensities can be quite small due to sample related reasons, t1-noise can make data interpretation difficult. In this thesis, the application of wavelet transform to eliminate t1-noise in 2D spectra is considered. Application of noise-reduction algorithm uniformly, as has been done in other cases, does not eliminate t1-noise. Instead, a variable noise thresholding helps in eliminating t1-noise significantly. This is demonstrated for two peptides and useful cross-peak information which is otherwise difficult to obtain has been extracted. The second part of the thesis deals with ssNMR methods pertaining to extraction of chemical shift anisotropy (CSA) parameters leading to improved methods of estimation of order and dynamics in liquid crystals. The method is based on the observation of similar patterns in the 2D SLF spectra of different liquid crystalline materials containing identical molecular moiety. The patterns are indicative of identical CSA parameters and molecular orientation, but different order parameters. These order parameters obtained from the dipolar couplings are used to estimate the CSA of the carbons of the molecular moiety using a least square fitting programme. A different approach using temperature as a means of generating spectra with different ordering is also considered and its application to the liquid crystal EBBA is made. Finally, ssNMR methods have been applied to the study of fillers for biodegradable polymers and their crystalline/amorphous nature has been studied by using a variety of solid state NMR methods.