Frequency-Domain Identification of Continuous-Time ARMA Models from Sampled Data

Authors:	Gillberg Jonas, Linköping University, Sweden Ljung Lennart, Linköping University, Sweden
Topic:	1.1 Modelling, Identification & Signal Processing
Session:	Continuous Time System Identification
Keywords:	Continuous-time systems; Parameter estimation; Continuous-timeARMA;noise model; Whittle likelihood estimator

Abstract

This paper treats direct identification of continuous-timeautoregressive moving average (CARMA) noise models. The approachhas its point of origin in the frequency domain Whittle likelihoodestimator. The discrete- or continuous-time spectral densities areestimated from equidistant samples of the output. For low samplingrates the discrete-time spectral density is modelled directly byits continuous-time spectral density using the Poisson summationformula. In the case of rapid sampling the continuous-timespectral density is estimated directly by modifying itsdiscrete-time counterpart.