Time-domain synthesis of optimal inputs for system identification

In this paper new results are derived for the computation of D-optimal randomized input designs to identify unknown parameters in linear systems with process noise and in non-linear and distributed parameter systems without process noise. The Kiefer-Wolfowitz experimental design approach that was extended earlier by the author to frequency-domain synthesis of optimal inputs is generalized here to the computation of optimal probability measures on the space of time-domain inputs. Both discrete-time and continuous-time systems are considered and algorithms converging to the global optima are given. Extensions to other criteria and bounds for suboptimal designs are also derived.