The paper deals with performance eval uation of some methods for sequential detection of changes in non-stationary digital signals. Methods are "exactly" valid under the assumption that the signal under study is an autoregressive process. However, they appear to be robust to this assumption and can, therefore, be applied to other signals as well. The detection algorithms herein considered are based on the quadratic forms of Gaussian random variable, which are x2 distributed under the null hypothesis (no change); random variables used to construct the quadratic forms include AR parameters, estimated residual variance and sample and partial residual correlations. The presented methods combine sequential and sliding block analyses. The considered methods performances are evaluated by simulation. Also, the methods' robustness as to the assumption of autoregressive data and to the model structure, is investigated.
Hypothesis testing, detection of changes, autoregressive modelling, non-stationary time series, quadratic forms, simulation.
All articles published by Studies in Informatics and Control (SIC) are published in accordance with the terms and conditions of the Creative Commons Attribution License (CC BY-NC-SA 4.0).