Model reduction of time-varying linear systems using approximate multipoint Krylov-subspace projectors

Abstract

In this paper a method is presented for model reduction of systems described by time-varying differential-algebraic equations. This method allows automated extraction of reduced models for nonlinear RF blocks, such as mixers and filters, that have a near-linear signal path but may contain strongly nonlinear time-varying components. The models have the accuracy of a transistor-level nonlinear simulation but are very compact and so can be used in system-level simulation and design. The model reduction procedure is based on a multipoint rational approximation algorithm formed by orthogonal projection of the original time-varying linear system into an approximate Krylov subspace. The models obtained from the approximate Krylov-subspace projector can be obtained much more easily than the exact projectors but show negligible difference in accuracy.