Minmax and least squares multivariable transfer function curve fitting: error criteria, algorithms and comparisons

Two types of algorithms for minmax frequency response curve fitting for multivariable systems in matrix fraction description are discussed for different absolute error criteria. The first algorithm is a generalisation of the well-known SK iteration, whereas the second algorithm is based on solving a sequence of reweighted nonlinear least squares problems. The error criteria treated are the maximum spectral norm of the error, the maximum infinity norm of the error, and the maximum infinity norm taken over the real and imaginary parts of the error. The performance of the algorithms is illustrated on simulated experimental data.