MODEL-REDUCTION VIA LEAST-SQUARES PADE SIMPLIFICATION OF SQUARED-MAGNITUDE FUNCTIONS

Simplification of squared-magnitude functions using the least-squares Pade (LS-Pade) method is proposed as a new procedure for model order-reduction. One of the advantages of this method is that it overcomes the jomega-axis problem encountered in model simplification by means of squared-magnitude functions (SMF). This problem arises whenever poles of the simplified SMF are on the imaginary axis since in such cases spectral factorization cannot be applied to obtain the reduced-order model. It is shown that the standard LS-Pade algorithm can be directly applied to SMF in the simplification of transfer functions. Besides solving the jomega-axis problem from a practical point of view, the proposed procedure usually yields better models even in cases when no poles are on the imaginary axis. Two numerical examples are included to illustrate the main points of the new procedure.