Home > Research > Publications & Outputs > Stochastic unobserved component models for adap...
View graph of relations

Stochastic unobserved component models for adaptive signal extraction and forecasting

Research output: Contribution to conference - Without ISBN/ISSN Conference paperpeer-review

Published
Publication date1998
Number of pages10
Pages234-243
<mark>Original language</mark>English
EventProceedings of the 1998 8th IEEE Workshop on Neural Networks for Signal Processing VIII - Cambridge, Engl
Duration: 31/08/19982/09/1998

Conference

ConferenceProceedings of the 1998 8th IEEE Workshop on Neural Networks for Signal Processing VIII
CityCambridge, Engl
Period31/08/982/09/98

Abstract

The adaptive, off-line signal processing and forecasting of nonstationary signals described by an unobserved component (UC) model is presented. The off-line signal processing based on a dynamic harmonic regression (DHR) model of the UC type and its performance via a typical problem of audio signal restoration are discussed. The simplest possible method of detecting outliers is used but more sophisticated approaches based on either their statistical properties of the KF innovations or the difference between the KF and FIS estimates are possible.