Book Description

Although originally designed to treat neither spectroscopic nor multidimensional data, FDM was adapted to process NMR spectra of both large and small molecules in liquids in 1D, 2D, 3D, and 4D spectra, with the multidimensional cases requiring important modifications. FDM outperforms the discrete Fourier Transform (DFT) whenever sensitivity is good, the spectrum is sufficiently sparse, the spectral peaks are sharp and Lorentzian, and the extent of the measured time-domain data is short, so that the FT spectrum exhibits line shapes that are distorted only by the time-frequency uncertainty principle. A further study of FDM to perturbations under non-ideal conditions lead to further improvements and finally a multi-dimensional, phase-sensitive, hybrid FDM (HFDM) algorithm that is introduced for the first time. HFDM can enhance the resolution of peaks distorted by time-frequency uncertainty, while still producing a spectrum in which all the original imperfections are present, such that a true noise floor prevents inadvertent contouring of features that look like resonances but that would in fact be buried in noise in any decent spectral estimate. This opens the door to the application of HFDM on NMR data sets with more reliability. Mixture separation is another important area of NMR research and will be the focus of the third chapter. Separating a mixture of signals into its source components is an engaging but challenging problem relevant to research in a wide variety of fields from neurology and medical signal processing to speech recognition systems. The vast majority of NMR spectroscopic data is taken on isolated, purified compounds; yet mixtures are clearly important to tackle. The pharmaceutical industry is a key example of a field that would benefit tremendously from more robust NMR mixture separation techniques. Structural information about promising candidates from the drug design process must be extracted from samples that also contain the target molecule, side products, and impurities. NMR can identify individual samples with respect to their unique fingerprint spectra; however, it has been traditionally poor at handling mixtures. Diffusion ordered spectroscopy (DOSY) has been one of the few NMR techniques that can be used for mixture separation. It uses translational diffusion to effect a partial separation and can be quite powerful when the individual compounds have a distinguishable difference in their diffusion constants. However, the separation problem becomes more difficult when the component molecules have similar chemical functional group and overlap in their NMR spectra. Blind Source Separation (BSS) is a technique that can be used to extract an individual NMR subspectrum from spectra of a number of different mixtures; and--if the molecules do not interact strongly--these individual subspectra adhere closely to those of the corresponding pure compounds under the same conditions. Furthermore, the "mixture spectra" need not be different physical samples: any method that can modify the intensities of all the peaks from a given molecule, in concert, is applicable. Previously a mixture of two components was successfully separated. Recent efforts have uncovered a flaw in the original BSS algorithm that causes behaves erratically for samples with three or more components. This complication will be described in detail, and a newly developed approach-- dubbed "BISI" for Blind Iterative Source Identification--will be introduced that overcomes those previously unforeseen problems.