Kernel-based continuous-time system identification: A parametric approximation

In this paper, we discuss the non-parametric estimate problem using kernel-based LTI system identification techniques by constructing a Loewner-based interpolant of the estimated model. Through this framework, we have been able to retrieve a finite-dimensional approximation of the infinite-dimensional estimate obtained using the classical kernel-based methodologies. The employment of the Loewner framework constitutes an enhancement of recent results which propose to use a Pade approximant to obtain a rational transfer function from an irrational transfer function corresponding to the identified impulse response. The enhancement has been illustrated for the identification of the Rao-Garnier benchmark.