This thesis describes a variety of methods developed to increase the sensitivity and resolution of liquid state nuclear magnetic resonance (NMR) experiments. NMR is known as one of the most versatile non-invasive analytical techniques yet often suffers from low sensitivity. The main contribution to this low sensitivity issue is a presence of noise and level of noise in the spectrum is expressed numerically as "signal-to-noise ratio". NMR signal processing involves sensitivity and resolution enhancement achieved by noise reduction using mathematical algorithms. A singular value decomposition based reduced rank matrix method, composite property mapping, in particular is studied extensively in this thesis to present its advantages, limitations, and applications. In theory, when the sum of k noiseless sinusoidal decays is formatted into a specific matrix form (i.e., Toeplitz), the matrix is known to possess k linearly independent columns. This information becomes apparent only after a singular value decomposition of the matrix. Singular value decomposition factorises the large matrix into three smaller submatrices: right and left singular vector matrices, and one diagonal matrix containing singular values. Were k noiseless sinusoidal decays involved, there would be only k nonzero singular values appearing in the diagonal matrix in descending order providing the information of the amplitude of each sinusoidal decay. The number of non-zero singular values or the number of linearly independent columns is known as the rank of the matrix. With real NMR data none of the singular values equals zero and the matrix has full rank. The reduction of the rank of the matrix and thus the noise in the reconstructed NMR data can be achieved by replacing all the singular values except the first k values with zeroes. This noise reduction process becomes difficult when biomolecular NMR data is to be processed due to the number of resonances being unknown and the presence of a large solvent peak.
Date of Award | 2018 |
---|
Original language | English |
---|
- signal processing
- singular value decomposition
- algebras
- linear
- noise
- nuclear magnetic resonance
Beyond the noise : high fidelity MR signal processing
Masuda, R. (Author). 2018
Western Sydney University thesis: Doctoral thesis