ROBUSTNESS ISSUES IN CONTINUOUS-TIME SYSTEM IDENTIFICATION FROM SAMPLED DATA

This paper explores the robustness issues that arise in the identification of continuous-time systems from sampled data. A key observation is that, in practice, one cannot rely upon the fidelity of the model at high frequencies. This implies that any result which implicitly or explicitly depends upon the folding of high frequency components down to lower frequencies will be inherently non-robust. We illustrate this point by referring to the identification of continuous-time auto-regressive stochastic models from sampled data. We argue that traditional approaches to this problem are sensitive to high frequency modelling errors. We also propose an alternative maximum likelihood procedure in the frequency domain, which is robust to high frequency modelling errors.