Fast broadband modeling of frequency-domain responses by piecewise interpolation

Broadband macromodeling of frequency domain responses by Vector Fitting can be computationally demanding in CPU time and memory. This paper presents a practical solution to reduce the computational cost that is involved with the modeling of high-order frequency responses. It applies a piecewise fitting strategy that makes use of a fast rational interpolation scheme to identify a representative set of data samples and an appropriate model order. This information is exploited by the Vector Fitting algorithm to extract the poles of the broadband macromodel in a reduced amount of time. The calculation of the residues is then solved as a linear approximation problem, and standard model reduction techniques can be applied as an optional step to remove pole redundancies in the model. It is shown by an example that substantial savings are obtained in terms of computation time and memory requirements, when compared to the standard fitting procedure.