LMI-based optimal design of test signals for control-relevant system identification

An optimal design of test signals for control-relevant system identification is formulated as a linear matrix inequalities (LMI) problem. The proposed LMI-based design involves two steps: the calculation of the optimal covariance function by solving a LMI problem and the realization of the covariance function as a test signal. In a case study involving a nonlinear evaporation process, this technique is shown to reduce the required computational effort significantly while yielding a similar level of performance when compared to the direct approach, in which a non-convex optimization is solved directly for the test input sequence.