Unfalsified model parametrization based on frequency domain noise information (Q1973998)

The author studies the extension of results concerning the unfalsified model parameterization to the case where identification noises are magnitude bounded in the frequency domain. A plant model is called unfalsified if it is consistent with both the a priori assumptions on the plant and the noises, as well as the available identification experimental data. He proves that every unfalsified model can be represented by a linear fractional transform of a block diagonal norm bounded function matrix on the basis of the widely known AAK extrapolation theory if a certain parameterization condition is satisfied and that the dimensions of the involved matrix valued transfer functions are of the order of the identification experiment data length. Moreover, he answers how to verify the parameterization condition, that is, by computing the maximal singular values of two constant matrices. In order to achieve the above conclusions the relation between a lower triangular Toeplitz matrix and the Fourier transform of its first column elements plays an important role. However, an analysis shows that the parametrization condition can be satisfied only when the noise level is very low.