On the metric complexity of continuous-time systems

In this paper, metric complexities of certain classes of continuous-time systems are studied, using the time-domain sampling approach and the concepts of Kolmogorov, Gel'fand and sampling n-widths for certain classes of Sobolev space. A sampling theorem is obtained which extends Shannon's sampling theorem to systems with possibly non-band-limited spectra, The theorem demonstrates that continuous-time systems in certain Sobolev spaces can be approximately reconstructed causally from their sampled systems. The Kolmogorov, Gel'fand and sampling n-widths of various uncertainty sets in the Sobolev spaces are derived. The results show that the sampling approach is in fact asymptotically optimal, when the sampling interval is selected to minimize the loss of information in the sampling process, for the modelling of systems in such Sobolev spaces.