In signal processing data are traditionally sampled according to the Shannon-Nyquist theorem in order to prevent aliasing effects. Here we focus on parametric methods and introduce a procedure that allows these methods to work with sub-sampled data. We actually make use of the aliasing effect to regularize the problem statement rather than that we avoid it.
The new approach adds a number of features to a standard exponential analysis, among which output validation, the automatic detection of the exponential model order, robustness against outliers, and the possibilty to parallellize the analysis.