A mesh-free adaptive parametric macromodeling strategy with guaranteed stability

This paper proposes a fully automated procedure for the generation of behavioral time-domain macromodels of complex multiport electric, electronic or electromagnetic systems, whose response depends on several design parameters. The latter are embedded in closed form in the macromodel structure through a mesh-free radial basis function representation, which allows scalability to a possibly large number of parameters. A greedy process is proposed to iteratively select a reduced number of training frequency responses, so that the macromodel accuracy is enforced uniformly in the parameter space. Examples with up to ten independent parameters demonstrate the effectiveness of proposed algorithm.