I. Introduction
It is often very difficult to manually derive time-domain, physics-based models for passive microwave structures. However, an accurate frequency-domain representation in the form of frequency-domain S- or Y-parameters can be obtained via numerical full-wave simulation. In such cases, there are three main classes of methods for performing time-domain transient simulation. The first is based on convolution techniques on the tabulated band-limited S- or Y-parameter data [1], [2]. Such an approach is general, flexible, and can be found in a number of commercial simulators; however, it suffers from a number of drawbacks. An accurate convolution would require a large number of frequency samples that are CPU expensive to obtain using full-wave simulation [3]. Furthermore, the convolution integral itself is expensive to compute at each time step of the simulation and has a high memory requirement [4]. Another class of methods for macromodeling microwave structures is based on tools, such as model order reduction [5], [6] and interconnect macromodels [7]. Such approaches are practical for cases where a large or distributed model can be found. More recently, a new class of more direct approaches has been proposed in the literature, which rely on the numerical generation of an SPICE-compatible time-domain macromodel directly from frequency-domain S- or Y-parameter data that can be computed using full-wave simulation [8]–[23]. This approach is general for any passive structure that can be simulated using full-wave techniques, and the resulting macromodel is compatible with a wide variety of simulators.