FIR modelling for errors-in-variables/closed-loop systems by exploiting cyclo-stationarity

摘要

Finite impulse response (FIR) modelling of errors-in-variables/closed-loop systems by correlation analysis usually yields biased estimates due to the additive noises on inputs and outputs. A non-parametric approach, the cyclic correlation analysis (CCRA), provides asymptotically unbiased and consistent estimates. The main feature of the CCRA is to eliminate the adverse effects of stationary noises by exploiting cyclo-stationarity that may exist naturally or be induced artificially. A complete study of the CCRA is developed, including the statistical performance of the estimated FIR model. Frequency-domain expressions of the statistical performance provide guidelines in designing a class of cyclo-stationary signals for modelling. Effectiveness and properties of the CCRA are validated and illustrated by numerical examples.