Evaluation of criteria for detection of changes in nonstationary signals

Abstract

A method to segment a nonstationary signal into non-overlapping successive near stationary segments using the Covariance method of Autoregressive (AR) Modelling is presented. An assumption of "local ergodicityn within the near stationary segments allows improved spectral estimation because of the adaptive nature of the windowing_ Performance is evaluated on simulated AR data and real oceanographic signals.