An approximate min-max solution to the identification problem

A technique based on the min-max error criterion is proposed for identifying a dynamic system in terms of a discrete-time model from the system response to a deterministic input. A linear, single-input/single-output, lumped-parameter, time-invariant (at least during the measurement interval) system is assumed. Although the resulting problem is non-linear, it is shown that linear programming techniques are applicable if a realistic approximation is made. The technique is implemented on the digital computer and evaluated by considering a number of typical discrete-time systems. The effects of additive noise are also considered --Abstract, page ii