A Möbius Transform Approach for Improved MIMO Frequency Domain Identification

Paper in proceeding, 2021

To improve the numerical conditioning of a frequency domain subspace algorithm (FSID) when identifying continuous time transfer functions a general Mobius transformation can be employed. The Mobius transformation is a more general class of transformations which include the bilinear transformation as a special case. For the state-space model class of MIMO transfer functions we derive the relations describing how the state-space realization of a transfer function is converted under a Mobius transformation. With an extensive numerical example we illustrate that a particular choice of the Mobius transformation leads to transfer function estimates with, in general, significantly improved accuracy. Copyright (C) 2021 The Authors.