Abstract
A numerical spectral-deconvolution method with an autoregressive (AR) model is proposed to overcome a noise problem that limits resolution-enhancement ability of a traditional Fourier-transform (FT) deconvolution. The optimal AR-model is determined by a new criterion that uses similarity between measured and reconstructed spectra. Additionally, unreliable peak-heights appearing in an AR power-spectrum are corrected by using its convolution with a Gaussian function with a small width. The proposed method was applied to noisy and low-resolution spectra of a bio-molecule, and found to be useful for determination of its monoisotopic mass.
