Improving Prony´s estimator for multiple frequency estimation by a general method of parametric filtering

Prony´s estimator is known to be inconsistent for frequency estimation. In an attempt to diminish this problem, Dragosevic and Stankovic (1989) coupled the least squares method of autoregressive (AR) estimation with an iterative filtering scheme using an all-pole filter. But the inconsistency still persists. It is shown that the method becomes consistent if an appropriately parametrized filter is employed. The all-pole filter in particular can be reparametrized so that it too satisfies the parametrization requirement and thus leads to consistent estimates. Experimental results show that the modified method has a higher resolution than the discrete Fourier transform (DFT) and that its overall performance is quite remarkable.<>