Robust online signal extraction from multivariate time series

Robust regression-based online filters for multivariate time series are proposed and their performance in real time signal extraction settings is discussed. The focus is on methods that can deal with time series exhibiting trends, level changes, outliers and a high level of noise as well as periods of a comparatively steady state. The new filter is based on a robust two-step online procedure, and it recognises that the data are often measured on a discrete scale. The relevant properties and the performance of this new filter are discussed and investigated by means of simulations and by a medical application.