Filtering techniques for efficient inversion of two-dimensional Nuclear Magnetic Resonance data

The inversion of two-dimensional Nuclear Magnetic Resonance (NMR) data requires the solution of a rst kind Fredholm integral equation with a two-dimensional tensor product kernel and lower bound constraints. For the solution of this ill-posed inverse problem, the recently presented 2DUPEN algorithm [V. Bortolotti et al., Inverse Problems , 33(1), 2016] uses multi- parameter Tikhonov regularization with automatic choice of the regularization parameters. In this work, I2DUPEN, an improved version of 2DUPEN that implements Mean Windowing and Singular Value Decomposition filters, is deeply tested. The reconstruction problem with filtered data is formulated as a compressed weighted least squares problem with multi-parameter Tikhonov regularization. Results on synthetic and real 2D NMR data are presented with the main purpose to deeper analyze the separate and combined e effects of these filtering techniques on the reconstructed 2D distribution