Spectral Estimation and Iterated Whitening
Date:
Abstract
The power spectrum of a stationary stochastic process characterizes the amount of “energy” in different frequencies, and power spectra are a fundamental tool in data analysis, signal processing, and linear prediction and control. Standard methods for estimating power spectra from data can be highly inaccurate when the dynamic range of the spectrum is large. In this talk, I present a novel method for accurately estimating the power spectra of signals from data. The method, based on an iterated “whitening” procedure, is designed to work well for spectra with high dynamic range. I compare the iterated whitening method with two standard methods, and illustrate its use on a prototypical model of spatiotemporal chaos.