How to get high resolution results from sparse and coarsely sampled data (Q2300771)

Using uniform downsampling, the authors reconstruct the parameters of an exponential sum, where ESPRIT and generalized overdetermined eigenvalue solver are used. The aim of this study is to retrieve fine-scale information directly from coarse-scale measurements acquired at a slower information rate, in function of the sparsity and not the bandwidth of the signal. The technique presented allows to overcome the Shannon-Nyquist sampling rate limitation and may improve the conditioning of the numerical linear algebra problems involved. This technique maintains a regular sampling scheme. Since a sub-Nyquist sampling rate can cause terms to collide at the time of the sampling, the authors explain how to unravel collisions. Several numerical examples illustrate both the collision-free situation and the case in which the collision of terms happens.