Model-based parameters estimation of non-stationary signals using time warping and a measure of spectral concentration

This paper proposes a parameters estimation algorithm for signals composed of multiple non-stationary components having the same basis modulation function which is described by an a priori known model and depends on a few unknown parameters. The procedure is based on time warping the signal in turn with every basis function resulted from different model parameters combinations and evaluating the concentration of the warped signal spectrum. The estimated parameters of the model are the ones which provide the best spectral concentration. Onwards, the amplitude, phase and modulation rate for each component are determined from the signal warped with the optimal basis function. The algorithm is tested with simulations and real data consisting of de-chirped radar signals and acoustic signals with harmonic components from underwater mammals.