Description
This article adresses aspects of the a-priori computation of the Fisher information matrix. The aim of the type of Fisher matrix discussed is the experiment design for parameter estimation of nonlinear state space models. A definition for the sampling of experiments is stated, which is convenient for software implementation. Both, measurement samples that belong to a coherent sequence emph{and} samples from independent experiments, may be stored together in one single data structure. Since the covariance matrix of the prediction error is required in the Fisher matrix, a method for its a-priori prediction is proposed. A servo-hydraulic positioning system serves as a validation example. The comparison of the a-priori and a-posteriori Fisher matrices of this example show that the method proposed is suitable to provide an approximation of an optimality criterion for experiment design.
