Spectral Estimation by Model Fitting

Abstract

The classic spectral analysis methods for random processes of finite length (e.g., the periodogram method or application of the Wiener-Khinchin theorem) can only provide estimated values—with a trade-off between bias and spectral solution. An example is the power spectrum of a time-limited segment of the process x(t) estimated by application of the Blackman-Tukey procedure using the Wiener-Khinchin theorem. The bias in the estimate obtained using this procedure can be reduced by weighting the autocovariance function, or by smoothing the estimated spectrum. The estimate is improved, but spectral resolution becomes worse.