Efficient removal of unwanted signals in NMR spectra using the filter diagonalization method

It is often desirable to selectively remove corrupting or uninteresting signals from complex NMR spectra without disturbing overlapping or nearby signals. For biofluids in particular, removal of solvent and urea signals is important for retaining quantitative accuracy in NMR-based metabonomics. This article presents a novel algorithm for efficient filtering of unwanted signals using the filter diagonalization method (FDM). Unwanted signals are modeled in the time domain using FDM. This modeled signal is subtracted from the original free induction decay. The resulting corrected signal is then processed using established workflow. The algorithm is found to be reliable and fast. By eliminating large, broad, uninteresting signals, many spectra can be subjected to fully automated absolute value processing, allowing objective preparation of spectra for pattern recognition analysis.