Convex passivity enforcement of linear macromodels via alternate subgradient iterations

This paper introduces a new algorithm for passivity enforcement of linear lumped macromodels in scattering form. As typical in most state of the art passivity enforcement methods, we start with an initial non-passive macromodel obtained by a Vector Fitting process, and we perturb its parameters to make it passive. The proposed scheme is based on a convex formulation of both passivity constraints and objective function for accuracy preservation, thus allowing a formal proof of convergence to the unique optimal passive macromodel. This is a distinctive feature that differentiates the new scheme with respect to most state of the art methods, which either do not guarantee convergence or are not able to provide the most accurate solution. The presented algorithm can thus be safely used for those cases for which existing techniques fail. We illustrate the advantages of proposed method on a few benchmarks.