On the Acceleration of the Vector Fitting for Multiport Large-Scale Macromodeling

摘要

Vector fitting (VF) is a robust macromodeling method to construct rational models of a network based on tabulated frequency responses. When fitting a network with a large number of ports, the matrix equation in VF quickly grows into a formidable size. The conventional strategy is to break the equation into several small QR factorizations and then combine the results into a least-squares (LS) problem. In this letter, an alternative solution procedure is proposed, the complexity of which is 2.7 times smaller than the conventional approach in terms of floating-point operations (FLOPs). Test cases of 40–200-port data show that the actual computing time can be reduced by as much as 85%, which presents a considerable saving for large-scale problems.