1 - Filtrage-détection autoadaptatif de signaux non stationnaires

Show full item record

Files in this item

PDF 001.PDF TEXTE.pdf 753.6Kb

Pour citer ce document :
URI: http://hdl.handle.net/2042/1655
Title: 1 - Filtrage-détection autoadaptatif de signaux non stationnaires
Author: GOERIG (L.); DONCARLI (C.)
Abstract: Etude de l'estimation récursive d'un signal non stationnaire perturbé par un bruit additif indépendant. Les caractéristiques-statistiques des bruits sont estimées en temps réel et sont utilisées pour ajuster le gain du filtre de Kalman d'estimation d'état
Description: The aim of this paper is the recursive estimation of a non stationnary signal, disturbed by an independent additive noise . This kind of problem typically has to be solved when the signal is not correctly described by a statistical model, and particularly when the baseline has no regular and no predictible fluctuations, or when the additive measurement noise is non stationnary . In this case, the design of a state estimating filter (the signal and its derivatives) is quite difficult because it is necessary to perform the best tradeoff between the elimination of the additive noise, and the tracking of the high dynamics of the signal. We propose, first, a discrete-type "majorant" model (series of integrators)'which is a particular case of ARIMA models . Then, we show how to estimate in real time the statistical parameters of the noises in this model, which are used to adapt automatically the Kalman filter gain. However, risky high level dynamics of the signal may occur, and the model is no longer a "majorant" one . A bias appears, and to avoid this problem, we propose to join a high dynamics detector, based on the generalised likelihood ratio . This non linear part of the algorithm reduces the delay and finally leads to an efficient procedure .
Subject: Filtrage adaptatif; Identification; Détection; Filtre Kalman; Traitement signal; Intégrateur; Méthode récursive; Temps réel; Lissage; Filtrage
Publisher: GRETSI, Saint Martin d'Hères, France
Date: 1988

This item appears in the following Collection(s)

Show full item record

Advanced Search