Worst-case identification in l1 for f.i.r linear systems

Investigates the sample complexity of a time-domain worst-case system identification problem for finite impulse response linear systems. The sample complexity of identification is the duration of the minimum length identification experiment that must be run in order to identify the unknown system to within a specified worst-case error bound. The identification criterion the authors treat is worst-case with respect to both modeling uncertainty and noise. In this paper the authors derive bounds on the sample complexity under various noise models. The authors´ results demonstrate how the character of noise affects the complexity of identification. A consequence of these results is that the complexity of identification can be quite reasonable if the distribution of the energy of the noise signal satisfies mild constraints