See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3267558

A Discussion of “Rational Approximation of Frequency Domain Responses by

Vector Fitting”

Article in Power Systems, IEEE Transactions on · March 2006

DOI: 10.1109/TPWRS.2005.860905 · Source: IEEE Xplore

CITATIONS READS
95 991

2 authors:

Wouter Hendrickx Tom Dhaene

University of Antwerp Ghent University
16 PUBLICATIONS 284 CITATIONS 515 PUBLICATIONS 10,286 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

pySUMO View project

FP6 OMOORENICE! - Operational MOdel Order REduction for Nanoscale IC Electronics View project

All content following this page was uploaded by Tom Dhaene on 02 January 2013.

The user has requested enhancement of the downloaded file.

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 21, NO. 1, FEBRUARY 2006 441

A Discussion of “Rational Approximation of Frequency

Domain Responses by Vector Fitting”
Wouter Hendrickx and Tom Dhaene, Member, IEEE

Abstract—Vector fitting (VF) is a popular iterative rational ap- where . If we form linear combinations of these basis
proximation technique for sampled data in the frequency domain. functions and add a polynomial of degree , we always
VF is nowadays widely investigated and used in the Power Systems get fractions with the same denominator
and Microwave Engineering communities. The VF methodology is
recognized as an elegant version of the Sanathanan–Koerner iter-
ation with a well-chosen basis.
Index Terms—Rational interpolation, system identification,
vector fitting.

(2)
I. INTRODUCTION

B ROADBAND rational approximations of the admittance

matrix are of paramount importance for accurate tran-
sient simulation of the frequency-dependent behavior of linear
where we have introduced the abbreviations
merator (which is of degree
) and
; we take
for the denominator.
for the nu-
when

power systems, such as transmission lines and transformers. The On the other hand, if we have a polynomial of degree
compact rational pole-zero models can easily be used in EMTP- , we can always write
type programs, and they can replace the tedious and error-prone
numerical transient simulations by fast and efficient iterative
convolutions. (3)
In [1], a new rational interpolation technique, vector fitting
(VF), was introduced. References [2]–[4] provide more infor-
mation on the subject. In this letter, we elaborate on the internals for some and some polynomial of degree . This is
of this technique. A thorough mathematical analysis provides nothing more than the partial fraction expansion.
insight into the inner working of the VF algorithm and teaches
how it can be improved, e.g., by using an orthonormal rational III. PRELIMINARIES: THE ITERATIVE LEAST-SQUARES SCHEME
basis [5]. Throughout this letter, we have adopted the same no- Suppose we want to approximate the values
tation as in [1]. We assume that the reader is familiar with the at sample points by a model of the form
VF methodology.
In Section II, we discuss the pole-based basis that is used
in VF. Section III describes a general rational iterative least- (4)
squares framework. Section IV finally provides the link with the
VF methodology. Section V contains our conclusions.
for some fixed basis functions and . An optimal solution (in
the least-squares sense) to this problem would be to find and
II. PRELIMINARIES: THE POLE-BASED BASIS
such that
We fix degree and fix (the starting poles)
and look at the basis functions
(5)

for (1)
is minimized. Common practice solving such problems is to lin-
earize the problem to get the following expression:

Manuscript received December 1, 2004; revised July 15, 2005. This work (6)
was supported by the FWO—Flemish Fund for Scientific Research. Paper no.
PESL-00111-2004.
The authors are with the Department of Mathematics and Computer Science,
University of Antwerp, Antwerp 2020, Belgium. Equation (6) is linear in its unknowns and and can be
Digital Object Identifier 10.1109/TPWRS.2005.860905 solved using standard least-squares techniques. Unfortunately,
0885-8950/$20.00 © 2006 IEEE
442 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 21, NO. 1, FEBRUARY 2006

the solutions of this linear problem are not the same as those of Multiplying the second equation in (9) by , evaluating in
the original problem. Note that the weighting factor of and equating both equations (as in [1]), and using (10), we
get the system of equations
(7)

was dropped in (6).

Now we propose an iterative scheme. Suppose we have al- (11)
ready found some and that minimize (6). This result
can be used to start the iterative algorithm that minimizes fol-
lowing weighted expressions (for ) if and only if

(12)
(8)
for all sample points .
The first set of (11) is the same as in VF for poles , except
Each iteration tries to correct the missing weighting factor in (6) for the weighting factor
by using the denominator of the previous iteration.
This approach is known as the Sanathanan–Koerner (SK) it-
eration [6] in the -domain for continuous time systems or the (13)
Steiglitz–McBride iteration [7] in the -domain for discrete time
systems. These iterative techniques are extensively used, and
their properties have been studied. For more information, we The second set of (12) is exactly the VF formulation for a set of
refer to [8] and [9]. The VF methodology also can be seen as an poles .
implementation of the iteration described in this section, as we Suppose now that we have solved the VF pole identification
will see in the next section. system with starting poles . Expanding in its poles and
zeros gives rise to an expression that matches exactly
IV. CLOSER LOOK AT VECTOR FITTING if is chosen to match the zeros of . In that case, solving
the weighted problem (11) obtains the same approximation
We now take a closer look at the VF iteration [1]. In each
as solving (12). The last is exactly the second iteration in the
iteration, both and are approximated by rational
VF methodology, while the first exactly matches the iterative
functions using a common set of fixed poles
scheme of Section III with equal to the basis
function specified in Section II and
and .

V. CONCLUSION
(9)
VF is identified as an elegant reformulation of the SK iter-
ation. The pole-based basis that is used in VF has numerical
with and . advantages over the power-series approach, which is still often
Based on (2) and (3), we can easily perform a basis transfor- used in other implementations of the SK iteration. Powers of
mation, or do a so-called “pole-relocation,” for each equation in (which lead to ill-conditioning) are avoided in the linear system
(9) by using the pole-based basis.
Furthermore, the explicit pole representation of it makes the
enforcement of stable poles easily achievable (by flipping the
unstable poles) as compared to other implementations of the SK
iteration.
Note that the SK iteration is known not to exactly minimize
a least-squares distance. However, it provides good models and
remains stable during successive iterations [9], [10]. Theoreti-
(10) cally, both the VF and SK methodologies should compute the
same solution to the problem. In practice, VF produces favor-
able linear systems that do not contain high powers of as in a
for some coefficients and a polynomial . direct implementation of SK.
IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 21, NO. 1, FEBRUARY 2006 443

REFERENCES [6] C. K. Sanathanan and J. Koerner, “Transfer function synthesis as a ratio

of two complex polynomials,” IEEE Trans. Autom. Control, vol. AC-9,
[1] B. Gustavsen and A. Semlyen, “Rational approximation of frequency
no. 1, pp. 56–58, 1963.
domain responses by vector fitting,” IEEE Trans. Power Del., vol. 14, [7] K. E. Steiglitz and L. E. McBride, “A technique for the identification
no. 3, pp. 1052–1061, Jul. 1999.
of linear systems,” IEEE Trans. Autom. Control, vol. AC-10, no. 4, pp.
[2] A. Semlyen and B. Gustavsen, “Vector fitting by pole relocation for the 461–464, 1965.
state equation approximation of nonrational transfer matrices,” Circuits [8] P. A. Regalia, Adaptive IIR Filtering in Signal Processing and Control.
Syst.: Analog Dig. Signal Process., vol. 19, no. 6, pp. 549–566, 2000.
New York: Marcel-Dekker, 1995.
[3] B. Gustavsen, “Computer code for rational approximation of frequency [9] , “Comments on ’A weighted least-squares method for the design
dependent admittance matrices,” IEEE Trans. Power Del., vol. 17, no.
of stable 1-D and 2-D IIR digital filters’,” IEEE Trans. Signal Process.,
4, pp. 1093–1098, Oct. 2002. vol. 47, no. 7, pp. 2063–2065, Jul. 1999.
[4] W. Hendrickx, D. Deschrijver, and T. Dhaene, “Some remarks on the
[10] W. S. Lu, S. C. Pei, and C. C. Tseng, “A weighted least-squares method
vector fitting iteration,” in Proc. Eur. Conf. Mathematics Industry. Lec- for the design of stable 1-D and 2-D IIR filters,” IEEE Trans. Signal
ture Notes in Computer Science, to be published.
Process., vol. 46, no. 1, pp. 1–10, Jan. 1998.
[5] D. Deschrijver and T. Dhaene, “Rational modeling of spectral data using
orthonormal vector fitting,” in Proc. IEEE Workshop Signal Propagation
Interconnects, 2005, pp. 111–114.

View publication stats