Energy-weighted linear predictive spectral estimation: A new method combining robustness and high resolution

A new method for estimating the AR process coefficients for spectral estimation is introduced. The M selected coefficients achieve the minimum square error in fitting a recursion among the estimated covariance elements of the data which would be satisfied exactly if the statistics were known exactly and the data process fit the model assumptions (Mth-order AR). This minimization is shown to be identical to minimizing the average one-step prediction error with adaptive weights determined by the energy of the measured data. As in the Burg algorithm, forward and backward sweeps are averaged and the Levinson recursion is employed. Spectra computed from short, deterministic, and noisy data are compared with computed Burg spectra and show improvement in bias, resolution, and robustness of peak detection.