Refining accuracy of the spectral lines estimation by a sparsity based approach

The paper addresses a classical problem of spectral analysis, that of high accuracy estimation of line-spectra, based on zero-padded Discrete Fourier Transform (DFT), transposed in the modern approach of sparse vectors estimation. Based on the analysis of the specific dictionary, some conclusions are drawn concerning the possibility to obtain high accuracy and resolution, by using a limited temporal analysis window. The issue of uniqueness is discussed. Two algorithms are presented, based on Matching Pursuit (MP) and Orthogonal Matching Pursuit (OMP).