Instantaneous frequency based spectral analysis of nuclear magnetic spectroscopy data for metabolomics

Nuclear magnetic resonance spectroscopy (NMRS) signals are modeled as a sum of decaying complex exponentials. The spectral analysis of these signals in order to detect their components and estimate their parameters is crucial to the biochemical analysis of the samples under examination. This paper presents a novel time frequency representation based on a Gabor filterbank/notch filtering instantaneous frequency estimator, in order to enable the detection of weaker and shorter lived exponentials. Building on prior work involving filterank-based instantaneous frequency (IF) estimation, this new approach is an iterative procedure where a Gabor filterbank is first employed in order to obtain a reliable estimate of the IF of the strongest component present. This component is then notch filtered in order to un-mask weaker components and the procedure repreated. The performance of this method was evaluated using an artificial signal and compared to the short time Fourier transform and the original Gabor filterbank approach. The results clearly demonstrate the superiority of the new method in uncovering weaker signals and resolving components that are very close to one another in frequency.