Orthonormal basis functions for modelling continuous-time systems

Abstract

This paper studies continuous-time system model sets that are spanned by fixed pole orthonormal bases. The nature of these bases is such as to generalise the well-known Laguerre and two-parameter Kautz bases. The contribution of the paper is to establish that the obtained model sets are complete in all of the Hardy spaces , and the right half plane algebra A(Π) provided that a mild condition on the choice of basis poles is satisfied. A characterisation of how modelling accuracy is affected by pole choice, as well as an application example of flexible structure modelling are also provided.Zusammenfassungn diesem Artikel werden Modellmengen für Systeme in stetiger Zeit betrachtet, wobei diese Mengen von orthonormalen Basen mit fixierten Polen erzeugt werden. Diese Basen verallgemeinern die wohlbekannten Laguerre-Basen und die zweiparametrigen Kautz-Basen. In dieser Arbeit wird gezeigt, dass die erhaltenen Modellmengen in allen Hardy-Räumen , und in der Algebra A(Π) der rechten Halbebene vollständig sind, vorausgesetzt, dass eine schwache Bedingung an die Pole der Basis erfüllt ist. Eine Charakterisierung des Einflusses der Polvorgabe auf die Modellgenauigkeit, sowie ein Anwendungsbeispiel der Modellierung einer flexiblen Struktur werden gegeben.Résuméet article étudie des ensembles de modèles de systèmes à temps continu qui sont engendrés par des bases orthonormales à pôles fixes. La nature de ces bases est telle qu'elles généralisent les bases bien connues de Laguerre et de Keutz à deux paramètres. La contribution de cet article est d’établir que les ensembles de modèles obtenus sont complets dans tous les espaces de Hardy , ainsi que l'algèbre du demi-plan de droite A(Π), pourvu qu'une condition douce sur le choix des pôles des bases soit satisfaite. Nous fournissons également une caractérisation de la façon dont la précision du modèle est affectée par le choix des pôles, ainsi qu'un exemple d'application de modélisation de structures flexibles.

Full text

Page 1

Signal Processing 77 (1999) 261}274
Orthonormal basis functions for modelling
continuous-time systems
Hu K seyin Akc 7 ay??*, Brett Ninness???
?Institute for Dynamical Systems, Bremen University, Postfach 330440, 28334 Bremen, Germany
?Centre for Integrated Dynamics and Control (CIDAC), University of Newcastle, Callaghan, NSW 2308, Australia
?Department of Electrical and Computer Engineering, University of Newcastle, Callaghan, NSW 2308, Australia
Received 14 May 1998; received in revised form 24 February 1999
Abstract
This paper studies continuous-timesystem model sets that are spanned by "xed pole orthonormalbases. The nature of
these bases is such as to generalise the well-known Laguerre and two-parameter Kautz bases. The contribution of the
paper is to establish that the obtained model sets are complete in all of the Hardy spaces H?(?), 1(p(R, and the
right half plane algebra A(?) provided that a mild condition on the choice of basis poles is satis"ed. A characterisationof
how modelling accuracy is a!ected by pole choice, as well as an application example of #exible structure modelling are
also provided. ? 1999 Elsevier Science B.V. All rights reserved.
Zusammenfassung
In diesem Artikel werden Modellmengen fu K r Systeme in stetiger Zeit betrachtet, wobei diese Mengen von ortho-
normalen Basen mit "xierten Polen erzeugt werden. Diese Basen verallgemeinern die wohlbekannten Laguerre-Basen
und die zweiparametrigen Kautz-Basen. In dieser Arbeit wird gezeigt, dass die erhaltenen Modellmengen in allen
Hardy-Ra K umen H?(?), 1(p(R, und in der Algebra A(?)der rechten Halbebenevollsta K ndig sind, vorausgesetzt,dass
eine schwache Bedingung an die Pole der Basis erfu K llt ist. Eine Charakterisierung des Ein#usses der Polvorgabe auf die
Modellgenauigkeit, sowie ein Anwendungsbeispiel der Modellierung einer #exiblen Struktur werden gegeben. ? 1999
Elsevier Science B.V. All rights reserved.
Re 2 sume 2
Cet article e H tudie des ensembles de mode ` les de syste ` mes a ` temps continu qui sont engendre H s par des bases
orthonormales a ` po ( les "xes. La nature de ces bases est telle qu'elles ge H ne H ralisent les bases bien connues de Laguerre et de
Keutz a ` deux parame ` tres. La contribution de cet article est d'e H tablir que les ensembles de mode ` les obtenus sont complets
dans tous les espaces de Hardy H?(?), 1(p(R, ainsi que l'alge ` bre du demi-plan de droite A(?), pourvu qu'une
condition douce sur le choix des po ( les des bases soit satisfaite. Nous fournissons e H galement une caracte H risation de la
fac 7 on dont la pre H cisiondu mode ` le est a!ecte H e par le choix des po ( les, ainsi qu'un exemple d'applicationde mode H lisation de
structures #exibles. ? 1999 Elsevier Science B.V. All rights reserved.
Keywords: Rational basis functions; Orthonormal; Completeness; Continuous-time systems
*Corresponding author. Tel.: #49-421-218-9394; fax: #49-421-218-4235.
E-mail address: akcay@math.uni-bremen.de (H. Akc 7 ay)
0165-1684/99/$-see front matter ? 1999 Elsevier Science B.V. All rights reserved.
PII: S0 1 6 5 -1 68 4 (9 9 ) 00 0 3 9- 0

Page 2

Nomenclature
C
R
?
? M
D
T
H?(?)
"eld of complex numbers
"eld of real numbers
open right half-plane ?s3C: Re?s?'0?
closed right half-plane ?s3C: Re?s?*0?
open unit disk ?z3C: ?z?(1?
unit circle ?z3C: ?z?"1?
Hardy spaces of functions f(s) analytic
on
?
and such
(1/2?)sup????????f(x#jy)??dy (R, 0
(p(R and ??f???"sup????f(s)?(R
right half-plane algebra ?f: f3H?(?) and
continuous on ? M ?
disk algebra ?f: f analytic on D and con-
tinuous on D M ?
linear span of A
complex conjugate of a
that
??f????"
A(?)
A(D)
spA
a ?
1. Introduction
The use of orthonormal bases for the purposes of
approximation and analysis is fundamental to
many areas of applied mathematics. In particular,
in the areas of control theory, signal processing and
system identi"cation, there has long been interest
in the use of the trigonometric (FIR), &Laguerre',
and
&two-parameterKautz'
More recently, in a discrete-time setting, this inter-
est has been revived in a string of works
[9}11,30,32,41,42,44] and this has led workers to
consider orthonormal constructionswhich general-
ise the Laguerre and two-parameter Kautz cases
[7,8,12,18,34]. One of these e!orts presented in
[3,31] considers the orthonormal basis functions
de"ned on D?T by a choice of numbers ??3D,
n"1,2,2, as
bases[17,20,21].
B?(z)O?1!?????
1!??z
z!??
1!??z,
????(z),
(1)
??(z)O?
?
???
??(z)O1.
In the special case of ??"?3R this becomes the
discrete-timeLaguerre basis, and in the special case
of ??"?3C this provides the discrete-time two-
parameter Kautz basis; see [31] for more details on
the generalising aspects of de"nition (1).
In the case of considering continuous-time
model descriptions, several important works have
also recently appeared that employ continuous-
time Laguerre and two-parameter Kautz bases
[24,25,43] and rational wavelet bases [13]. The
purpose of this paper is to make some further
contributionin this area by consideringsome issues
related to a natural generalisation of continuous-
time Laguerre and Kautz bases.
The generalisation to be considered is analogous
to the extension presented in (1). Namely, a set of
basis functions ?B?(s)? are treated which are de"ned
by a choice of numbers a?3?, ∀n as
B?(s)O?2Re?a??
s#a?
s!a?
s#a?
????(s),
(2)
??(s)O?
?
???
,
??(s)O1.
We set B?(s),1. The rational basis functions
?B?????are orthonormal in H?(?) with respect to
the inner-product (see Section 3)
?B?,B??O1
2??
?
??
B?(j?)B?(j?)d?
"?
1; m"n,
0; mOn.
Analogous to the discrete-time case, the continu-
ous-time Laguerre basis (studied, for example, in
[23,33,43]) is obtained as a special case of (2) by the
choice a?"a3R and the continuous time two-
parameter Kautz basis (studied in [43]) by the
choice a?"a3C.
It should be acknowledged that both the con-
tinuous and discrete-time generalisations shown in
(2) and (1) enjoy a long history in both the pure
mathematics [14,26,40] and engineering literature
[19,28,36].
2. Main result
As mentioned in the introduction, an important
motivation for the consideration of orthonormal
262
H. Akc 7 ay, B. Ninness / Signal Processing 77 (1999) 261}274

Page 3

parameterisations is for approximation purposes.
In this setting, a dominantquestionmust arise as to
thequalityof the approximation.Pertainingto this,
one of the most fundamental properties that might
be required is that linear combinations of the basis
elements be capable of arbitrarily good approxima-
tion.
Put more precisely in the context of systems
theory, this involves considering an element f(s)
living in a normed linear function space (X,??)???),
and for arbitrary ?'0 and for su$ciently large
n being able to "nd an element g(s)3sp?B?(s)?????
such that ??f!g???)?. Here X is a complex vector
space and the linear span is with respect to the "eld
of complex numbers.
If this is in fact possible for arbitrarily small ?,
then sp?B?(s)????is said to be &complete' in X. The
choice of the function space depends on the
application of the approximate model, but for
quadraticoptimalcontrol purposesor meansquare
optimal prediction purposes, the choice H?(?)
would be appropriate, while for robust control or
estimation purposes, the choices A(?) or H?(?) (for
large p) would be suitable.
With regard to the discrete-time basis (1), the
approximation issues have been addressed in [3]
where the following result was obtained.
Theorem 1 (Akc 7 ay and Ninness [3, Theorem 6 and
Corollary 7]). Consider the set of functions ?B?(z)?
dexned by (1). Then the set X"sp?B?(z)????is
complete in A(D) and H?(T)for all 1)p(Rif and
only if
?
?
???
(1!????)"R. (3)
The main result of this paper is to establish an
analogous result for the continuous-time basis (2)
as follows.
Theorem 2. The model set spanned by the basis func-
tions ?B?(s)????is complete in all of the spaces
H?(?), 1(p(R, and A(?) if and only if
?
?
???
Re?a??
1#?a???"R. (4)
Condition (4) is satis"ed by the Laguerre and
two-parameter Kautz bases and the rational
wavelets in [13]. In fact, it is a very mild condition.
For example, when a sequence of poles with "xed
realpart is chosen,the only way it could be violated
is that the imaginaryparts ofthe poles must diverge
to in"nity faster than the linear rate. If a sequence
of magnitude bounded poles are chosen, then in
order to violate (4) the real parts of the chosen basis
poles must approach to the imaginary axis very
rapidly.
In contrast to the Laguerre and two-parameter
Kautzbases, whereallthe polesare"xed at thesame
value, the generalbasis(2) enjoysincreased #exibility
of pole location. For example, slow and fast modes
may coexist in the model structure. As a result,
a fewer number of basis functions (and hence, for
systemidenti"cation applications, afewer number of
data) may be used without sacri"cing modelling
accuracy. This feature is quanti"ed in Section 4.
Note that the conclusion of Theorem 2 may be
extended to H?(?). Moreover, it is possible to con-
struct orthonormal model sets that are norm dense
in H?(?), 1)p(R, and have a prescribed
asymptotic order [4]. In [4], it is also shown that
the Fourier series formed by the general basis func-
tions converge in all spaces H?(?), 1(p(R.
3. Proof of Theorem 2
Before proceeding to the proof of Theorem 2, we
shall demonstrate that the basis functions in (2) are
indeedorthonormal.Whenn'm, we have by Cau-
chy's Integral Theorem [37]
?B?,B??
"!1
2??
?
??
?4Re?a??Re?a??
(j?#a?)(j?#a?)
????(j?)
??(j?)
d?
"!1
2?j??R
?4Re?a??Re?a??
(s#a?)(s#a?)
????(s)
??(s)
ds
"!lim
????
1
2?j?
??
???
?4Re?a??Re?a??
(s#a?)(s#a?)
#O(r??)?
????(s)
??(s)
ds
H. Akc 7 ay, B. Ninness / Signal Processing 77 (1999) 261}274
263

Page 4

" lim
????
1
2?j???
?????????????
?4Re?a??Re?a??
(s#a?)(s#a?)
????(s)
??(s)
ds
??
#O(r??)?"0,(5)
where the closed path ??consists of a segment of
the imaginary axis and an r radius semicircle in
? centred at the origin and is traversed counter
clockwise. When n"m, we have
?B?,B??"1
??
?
??
Re?a??
??#2?Im?a??#?a???d?"1,
where the second equality follows from the formula
3.3.16 in [1]
?
1
ax?#bx#cdx"
2
?4ac!b?tan??
2ax#b
?4ac!b?.
Now we return to the proof of Theorem 2. Consid-
eration is "rst given to the underlying space being
A(?).
Lemma 3. The linear span of the set ?B?(s)????with
B?(s) dexned in (2) is complete in A(?) if (4) holds.
Proof. This will be established by "rst addressing
the completeness of sp???(s)????in A(?). Notice
that since the bilinear map
s"1!z
1#z
preserves the supremum norms between A(?) and
A(D), the question of whether sp???(s)????is com-
plete in A(?) is equivalent to the question of the
completeness of
????
1!z
1#z?????
in A(D). Let
??O1!a?
1#a?
,
??O(!1)??
?
???
1#a?
1#a?
. (6)
Then provided Re?a??'0 for all n, ??3D for all n.
Also
???
1!z
1#z?"(!1)??
?
???
1#a?
1#a?
?
?
???
z!??
1!??z
"????(z).
Therefore, it is su$cient to establish the complete-
ness of sp???(z)????in A(D). To achieve this, con-
sider the functions ?B?(z)? de"ned in (1). Then
B?(z)"????(z)#????(z)
?1!?????
and hence sp?B?(z)?????Lsp???(z)?????. However
by Theorem 1, sp?B?(z)????is complete in A(D) if
and only if (3) is satis"ed.
Now, since
;
n*1
1#?a???*?1!a???"?1!a???1#a??
then by the de"nition in (6)
????"?
1!a?
1#a??*?1!a???
1#?a???
so that
?
?
???
(1!????))?
?
????1!?1!a???
"2?
?
???
1#?a????
Re?a??
1#?a???.
Conversely since
?
1!a?
1#a??"?
1!a?
1#a???
1#a?
1#a??"?1!a???
?1#a???)1#?a???
?1#a???
then
?
?
???
(1!????)*?
?
????1!1#?a???
"2?
?
???
?1#a????
Re?a??
?1#a???.
As well,
?1#a???)2(1#?a???)
264
H. Akc 7 ay, B. Ninness / Signal Processing 77 (1999) 261}274

Page 5

so that
?
?
???
Re?a??
1#?a???(?
?
???
(1!????))2?
?
???
Re?a??
1#?a???.
Therefore, (3) holds if and only if (4) holds implying
that under the de"nition (6), then (4) is necessary
and su$cient for sp?B?(z)????to be complete in
A(D) and hence for sp???(s)????to be complete in
A(?). Summing the identity
?2Re?a??B?(s)"????(s)!??(s);
k*1
over k"1,2,n then provides
?
?
???
Hence sp???(s)????Lsp?B?(s)????. This completes
the proof.
?
?2Re?a??B?(s)"1!??(s)"B?(s)!??(s).
Next, the question of the H?(?) su$ciency of (4)
is considered.
Lemma 4. Suppose that (4) is satisxed. Then
sp?B?????is complete in H?(?) for all 1(p(R.
Proof. Let m denote the multiplicity of a?. Suppose
"rst that m is "nite. Then reorder the basis poles so
that a?"a?"2"a?. We rede"ne the basis
functions in (2) by
B I?(s)O?
B?(s),
?
n(m,
s#a?
s!a??B???(s),
n*m.
(7)
This modi"cation of basis functions removes
a?from the pole parameter set ?a?????.
Let Q?denote the set containing all partial frac-
tion expansion terms of B I?. For example, when
m'1
Q?O?
1
s#a?
,2,
1
(s#a?)???,
1
s#a????,
and so on. Let
QO?
?
???
Q?.
Since
?
???
sp?B I?????is a complete set in A(?) by Lemma 3,
and so is sp?B??Q? since
sp?B I?????Lsp?B??Q?.
Let f3H?(?) and ?'0. Approximate f by a func-
tion g3A(?) that has the properties
Re?a??
1#?a???"R,
lim
?????
and
?s??g(s)?"0,
s3?,
?? f!g???(?.
This is possible since such functions form a dense
subset of H?(?) (see for example, Garnett [15, Co-
rollary 3.3 in Chapter II]). Let h(s)"(s#a?)g(s).
Then h3A(?). Since sp?B??Q? is a complete set in
A(?), there exists a u3sp?B??Q? such that
??h!u???(?
or
?g(s)!
u(s)
s#a??(
?
?s#a??,
s3?.
Hence
??f!
u
s#a????
(?#??
1
s#a????
?.
We have shown that the set
PO?
is a dense subset of H?(?) for all 1(p(R. It
remains to show that PLsp?B?????. This will
imply that sp?B?????is a complete set in H?(?) for
all 1(p(R. To this end, let v3sp?B??Q?. Since
v3sp?B??Q?, it can be written uniquely as a linear
v(s)
s#a?
: v(s)3sp?B??Q??
(8)
H. Akc 7 ay, B. Ninness / Signal Processing 77 (1999) 261}274
265

Page 6

combinationof the elements in Q and B?as follows:
v(s)"c?#?
?
???
c?
(s#a?)????,
where N(k) denotes the multiplicity of a?in the
set ?a?,a?,2,a?? and only a "nite number of the
coe$cients c?are nonzero. Notice that c?"0.
Then
v(s)
s#a?
"
c?
s#a?
#2#
c???
(s#a?)?
#
?
?
?????
c?
(s#a?)(s#a?)????.
for k'm, the terms c?/(s#a?)
(s#a?)???? admit further expansions
Since a?Oa?
c?
(s#a?)(s#a?)????
"
d?
s#a?
#
d?
s#a?
#2#
d????
(s#a?)????.
Hence
v(s)
s#a?
3sp?
1
s#a?
,2,
1
(s#a?)?,
1
s#a???
,2?
"spF,
where
FOQ??
1
(s#a?)??. (9)
Thus PLspF. To complete the proof for m(R,
we need to show that spFLsp?B?????. Let n be an
arbitrary positive integer. Write the partial fraction
expansions of the basis elements B?,B?,2,B?in the
following linear equation form:
?
???
B?
B?
?
B??
"?
???
???
?
0
2
2
0
0
?
???
?
???
?
2
????
??
1
s#a?
1
(s#a?)????
?
1
(s#a?)?????
.
The degree of B?is k, which implies that ???O0 for
all k)n and thus the lower triangular matrix
above is invertible. Hence, for i"1,2,n
1
(s#a?)????3sp?B?,k"1,2,n?Lsp?B?????.
Consequently spFLsp?B?????.
Suppose now that m"R. Then for each n,
de"ne Q?as the set containing all partial fraction
expansion terms of B?. In this case, the proof above
still applies with great simpli"cations since P de-
"ned in (8) equals to spQ, and F de"ned in (9)
equals to Q for all m.
Proof of Theorem 2. It only remains to establish the
necessity of (4) for completeness in H?(?) spaces.
Suppose then that (4) fails to hold. Then in this case
the "nite Blaschke products ??(s) de"ned by
?1!a???
1!a??,
converge (as nPR) uniformly on ? to a nonzero
function ?(s)3H?(?) which has zeros precisely at
the points a?; see, for example, Garnett [15, Chap-
ter II]. Therefore, the linear functional F de"nedon
H?(?) for all 1)p(R and A(?) by
??(s)O????(s),
??O?
?
???
??(s)O1
F(h)O1
2??
?
??
h(j?)
?(j?)
(!j?#1)?d?
is nontrivial and bounded. However, by Cauchy's
Integral Theorem, it vanishes for any B?of the form
(2) since
F(B?)"1
2??
?
??
!?2Re?a??
(j?#1)?(j?!a?)
j?#a?
j?!a?
???
????
???
?
?
?
?1!a???
1!a??
j?!a?
j?#a?
d?
266
H. Akc 7 ay, B. Ninness / Signal Processing 77 (1999) 261}274

Page 7

"!1
2??
?
??
?(j?)
(j?#1)?(j?#a?)d?
?(s)
(s#1)?(s#a?)ds
1
2?j?
???
#O(r??)?
"!1
2?j??R
????
"!lim
??
?(s)
(s#1)?(s#a?)ds
"lim
????
1
2?j???
???????????
?(s)
(s#1)?(s#a?)ds#O(r??)?
??
"0,
where
?(s)O?2Re?a???
?
???
?1!a???
1!a??
?
?
?????
?1!a???
1!a??
s!a?
s#a?
is analytic on ? and the closed path ??is as in (5).
Similarly, F(B?)"0. (Note that the remainder term
above vanishes as O(r??) in this case). Hence by an
application of the Hahn}Banach Theorem (see, for
example [2, Section 30]), sp?B?????de"ned by (2)
is not dense in any of the spaces A(?) and
H?(?), 1)p(R. This concludes the proof.
?
It should be pointed out that the su$ciency part
of Lemma 3 together with Lemma 4 gives the half
of the su$ciency condition in Achieser [2, Section
A.4] for the completenessof the model sets spanned
by B?and the Cauchy kernels
1
s#a,
in the Lebesgue spaces ¸?, 1(p(R, and in the
space of complex functions continuous on the imagi-
nary axis including R. The linear span of the Cau-
chy kernels does not contain systems which have
repeated poles whereas with the bases (2), a greater
#exibility is utilised on the choice of basis poles.
B?(s)O
a3?a?,a?,2?
3.1. Ensuring real-valued impulse response
Up until this point, the basis (2) has been con-
sidered with complete generality of pole location
save for the condition (4). However, in any applica-
tion involving the modelling of a physical process,
it is necessary to ensure that the underlying
modelled impulse response is real valued. If com-
plex valued choices for ?a?? are made in order to
accommodate resonant characteristics, then this
realness of impulse response is lost unless some
restriction is placed on how linear combination of
the basis functions is taken.
The purpose of this section is to illustrate how to
use the basis formulation (2) in such a way that
imposing realness of the weightings in the linear
combination ensures realness of the underlying im-
pulse response. This is achieved by requiring that if
a set of numbers ?a?,a?,2,a?? used to de"ne bases
via (2) contains a complex valued element (say a?),
then it always also includes its conjugate a?.
To be more explicit on this point, suppose that
the numbers a?,2,a???are real so that the basis
functions B?,2,B???have real-valued impulse re-
sponses. We now wish to include a complex pole at
!a?. Then two new basis functions B??, B??with
real impulse responses should be formed as a linear
combinationof B?and B???generatedby (2). These
new functions then replace B?and B???in any
modelling applications that require a real-valued
impulse response.
The linear combination we are suggesting can be
expressed as
?
c??
B??
B???"?
c?
c?
c????
B?
B????,
c?, c??, c?, c??3C.
(10)
Therefore, considering only B??for the moment, if
we choose complex poles in conjugate pairs as
!a???"!a?then
?2Re?a??(?s#?)
s?#(a?#a?)s#?a???????(s),
B??(s)"
(11)
where????(s) has real-valuedimpulseresponse and
the real coe$cients ?, ? are related to the choice of
c?, c?by
c?"a??#?
2a?
,
c?"a??!?
2a?
.
H. Akc 7 ay, B. Ninness / Signal Processing 77 (1999) 261}274
267

Page 8

Therefore, to ensure a unit norm for B??we must
choose ? and ? according to the constraint that
?c???#?c???"1 which becomes
x?Mx"2?a???,
where
(12)
xO(?,?)?,
MO?
?a???
0
0
1?.
Now, suppose we make two pairs of choices:
x"(?,?)? giving a basis function B??and y"(??,??)?
giving another basis function B??. These two choices
correspond to two pairs of complex numbers
?c?,c?? and ?c??,c???. The requirement c?c??#
c?c??"0 ensuring orthogonality of B??and B??can
be expressed as needing
x?My"0(13)
to hold, and in fact many solutions x and y to (12)
and (13) will exist. To formulate them, suppose we
begin by choosing any x satisfying (12). All solu-
tions to (12) are given by ?"?2cos? and
?"?a???2sin?, 0)?(2?. Then for a "xed ?,
a unique y that satis"es (12) and (13) is found by
rotating x ninety degrees in the normalised eigen-
space of M:
!1
1
y"M?????
0
0 ?M???x
or, to be more explicit
y"?2(!sin?, ?a??cos?)?.
To summarise this discussion, if we want to include
complex modes in a model structure, then we ob-
tain two basis vectors B??and B??from two linear
combinations of B?and B???that come from the
unifying construction (2). Let 0)?(2?. Then the
basis functions B??and B??are found as
B??(s)"?4Re?a??(scos?#?a??sin?)
s?#(a?#a?)s#?a???
B??(s)"?4Re?a??(!ssin?#?a??cos?)
s?#(a?#a?)s#?a???
These real-valued impulse response basis vectors
B??and B??are then used for modelling instead of
(14)
????(s),
????(s).
B?and B???. If we require further basis functions
with complex modes then we repeat the process in
(10) by forming B????and B????from linear combi-
nations of B???and B???and so on, and in this
way arbitrary complex pole con"gurations may be
accommodated. For example, when a?"a???"
2"a????"a??????, with chosen ?"0 the
above basisconstruction
k"0,2,m
process yields for
B????(s)"
?2bs
s?#bs#c?
s?!bs#c
s?#bs#c?
?????(s),
B????(s)"
?2bc
s?#bs#c?
s?!bs#c
s?#bs#c?
?????(s)
where b"2Re?a?? and c"?a???. With n"1
plugged in, this is the de"ning formula for the
two-parameter Kautz functions in [43].
Having now illustrated how the constraint of
realness of impulse response may be easily accom-
modated via constraining realness of linear combi-
nation weights, it remains to establish that this
latter restriction does not destroy completeness
properties. For this purpose, note that the basis
functions of the form described above and the basis
functions generated by (2) are related by a unitary
block diagonal matrix of the form
;Odiag?1,2,1,?
Let ;?3C??? denote truncations of ;. For a given
G3H?(?) and ?'0, via Theorem 2 there exists
a G?3H?(?) given by
c?
c??
c?
c???,2?.
G?(s)"?
?
???
??B?(s),
??3C
such that ??G!G????)?/2. Then G?can be written
as
G?(s)"??;??
?
??(s)
"?
?
???
??B??(s),
??Oa?#jb?;
a?,b?3R,
where ??O??;??
to a set of basis functions that have real-valued
impulse responses.
?
and here ???O(B??,2,B??) refers
268
H. Akc 7 ay, B. Ninness / Signal Processing 77 (1999) 261}274

Page 9

Now, assume that G(s) has a real-valued impulse
response so that G(j?)"G(!j?). Then, using the
fact that ??f??"??f??"??f(!j?)?? for any f3H?(?)
provides
??G!G????)?
N??G!?
?
???
??B?????
)?
N??G(!j?)!?
?
???
??B??(!j?)???
)?
N??G!?
?
???
??B?????
)?
and hence via the triangle inequality
??G!?
?
???
a?B?????
"??G!?
?
???
??#??
2
B?????
)?
2#?
2"?
so that indeed, arbitrarily accurate modelling of
real impulse G(s) is possible by taking real linear
combinations of real impulse versions of the basis
?B??.
4. Approximation of 5nite-dimensional systems
While the completeness result of Theorem 2 pro-
videsa theoreticalpedigreefor consideringbases(2)
for system approximation purposes, it leaves open
the question of the quality of approximation for
a "nite number of bases. Addressing this will be the
concern of this section, where a useful tool is to use
the so-called&reproducingkernel' K?(s,?) associated
with the linear space sp?B?(s)?????.
Lemma 5. Consider the basis functions ?B??????de-
xned by (2). Then
K?(s,?)O?
?
???
B?(?)B?(s)"1!??(?)??(s)
s#?
.
Proof. The proof will be by induction. First, when
n"1
a?#a?
(?#a?)(s#a?)
while
B?(?)B?(s)"
1!??(?)??(s)
s#?
"?1!(?!a?)(s!a?)
(?#a?)(s#a?)?
a?#a?
(?#a?)(s#a?)
1
s#?
"
so that the result holds for n"1. Now suppose it
holds for n'1. Then
K?(s,?)"K???(s,?)#B?(?)B?(s)
"1!????(?)????(s)
s#?
#
a?#a?
(?#a?)(s#a?)
????(?)????(s)
"
1
s#?
!?
(?#a?)(s#a?)!(a?#a?)(s#?)
(s#?)(?#a?)(s#a?)
?
?????(?)????(s)
"1!??(?)??(s)
s#?
.
Therefore, by induction the result holds for
all n.
?
The utility of this result becomes apparent in the
derivation of the following expression for the
"nite-order approximation error.
Lemma 6. Suppose f(s) is analytic on ? and has
a partial fraction expansion
f(s)"?
?
???
c?
s#??
.
H. Akc 7 ay, B. Ninness / Signal Processing 77 (1999) 261}274
269

Page 10

Dexne f?(s) as an approximation to f(s) obtained by
projection onto sp?B?(s)?????:
f?(s)O??
???
Then
?f,B??B?(s).
?f(j?)!f?(j?)?)?
?
????
c?
j?#???
?
?
????
??!a?
??#a??.(15)
Proof. By the de"nition of f?(s) and for any ?3?
????
"1
f?(?)"??
1
2?j??Rf(s)B?(s)ds?B?(?)
2?j??f(s)K?(s,?)ds,
where ? consists of the imaginary axis and an
in"nite radius semi-circle in the open right half-
plane;it is traversedclockwise.Using this de"nition
and Cauchy's integral formula gives, for ?3?
arbitrary
f(?)"1
2?j??
f(s)
?!sds.
Therefore, by Lemma 5 and using the fact that
s"!s for s3jR
?f(?)!f?(?)?
"?
1
2?j??
f(s)
?!s??(?)??(s)ds?
"?
??(?)
2?j
?
?
???
c???
1
(s#??)(?!s)
?
?
???
s#a?
s!a?
ds?
"???(?)??
?
?
???
c?
1
(?#??)
?
?
???
??!a?
??#a??,
where in moving to the last line Cauchy's residue
theorem was used to evaluate the integral after
performing the change of variable sC!s. Taking
the limit as Re???P0 then gives the result.
The resultexposes the dependence of the approx-
imation error on the choice of poles ?!a?? in the
base B?(s). Namely, the closer the poles ?!a?? are
chosento the poles ?!??? of the function f(s) being
approxiated then the more accurate the approxi-
mation of f(s) will be, and in such a way as to
decrease exponentially with increasing n.
Certainly the error bound (15) gives strong mo-
tivation for the consideration of the general basis
(2), since (in contrast to the Laguerre and Kautz
caseswhere all the poles are "xed at the same value)
the increased #exibility of pole location ?!a?? will
increasethe possibility of making???!a?? small(for
some l) for every k, and hence making the total
product ????????!a?????#a???? as small as pos-
sible.
5. Application example
We conclude the paper by presenting an applica-
tion example that illustrates the utility of the basis
(2) for modelling purposes. The example involves
measurements of the frequency response of a 58 cm
long, 5 mm wide cantilevered piezo-electric lami-
nate beam (for further details, see [29]). These
measurements are shown as dots in Fig. 1. For the
purposes of control (sti!ness compensation using
the piezo-electric actuators) a transfer function
model that explains this frequency response is re-
quired. There are many ways in which this may be
achieved [6,27,35,38,39], but for the purpose of
illustrating the e$cacy of the basis (2), the simple
least-squaresmethodintroducedby Levy[22] to "t
a model
G?(s)"N(s)
D(s)"b?s?#b???s???#2#b?s#b?
s?#d???s???#2#d?s#d?
to the frequencyresponse measurements?G??????at
the frequencies ????????by means of minimising the
cost
<?"?
?
???
?D(j ??)G?!N(j ??)??
will be studied. Our purpose is not to present best
possible results on this data set, but to illustrate by
an example that the model parameterised by the
orthonormal basis (2) should be preferred to poly-
nomial models when one is concerned with numer-
ical conditioning.
270
H. Akc 7 ay, B. Ninness / Signal Processing 77 (1999) 261}274

Page 11

Fig. 1. Estimation using the polynomial and the orthonormal bases. The dots are the measurements, the solid line is the estimate using
the basis (2) to parameterise the model, the dash}dot line is the estimate using the Tchebychev polynomials to parameterise the model.
The Tchebychev and the polynomial estimates are identical.
As is well known [35], "nding this estimate in-
volves solving the so-called &normal equations'
?
"?
??????????????????
??
(j??)?G?
?
(j??)?G??
!(j??)???G?,2, !G?, (j??)?,2, 1
??
!(j??)???G?,2, !G?, (j??)?,2, 1?
?
d???
?
d?
b?
?
b??
,
for which the numerical stability of the solution is
highly dependent [16], on the conditioning of the
matrix ???. However this can be altered via re-
parameterisations of the model G(s). For example,
in [5] the parameterisation
N(s)"b?#?
?
???
D(s)"s?#???
?
???
where each
p?(s)is
Tchebychev' polynomial (p?"1) is suggested as
a means of improving numerical conditioning.
In Fig. 1, the dash}dot line shows the results of
using the above Tchebychev parameterisation to "t
an n"18th-order model to the observed frequency
response. Note that the second resonance peak is
completely missed, and that the resonance frequen-
cies from the 3rd resonant mode onwards are sig-
ni"cantly shifted.
b?p?(s),
d?p?(s),
(16)
anorder
k
&modi"ed
H. Akc 7 ay, B. Ninness / Signal Processing 77 (1999) 261}274
271

Page 12

Fig. 2. The singular values of ? using the polynomial (natural and the Tchebychev) and the rational orthonormal basis (2).
However, if the model is parameterised using the
orthonormal basis (2) as
N(s)"b?#?
?
???
D(s)"s?#?
?
???
with the pole choice !a?"!a"!2??, then
the ensuing 18th-order least-squares estimate is the
solid line shown in Fig. 1, which now captures the
second resonance peak, and correctly matches the
resonance frequencies from the 3rd resonant mode
onwards.
Since the model structures (16) and (17) both
spanthe same manifold of rational models,the only
explanation for the di!erence in results is that of
di!erences in numerical conditioning. Fig. 2 shows
the singular values of ? for the three model para-
meterisation choices. (There are 36 singular values
of ? since the chosen model order is 18). Consider-
ing the log-scale employed, the parameterisation
usingthe basis (17) enjoys a two order of magnitude
b?B?(s),
d?B?(s),
(17)
better conditioning (the ratio of the largest to the
smallest singular value) than either a Tchebychev
polynomial, or conventional polynomial para-
meterisation.
As a consequence, for such applications of mod-
elling resonant structures over large bandwidths,
we suggest that the basis (2) should be employed in
the interests of the resultant frequency response
having no artifacts due to poor numerical condi-
tioning.
6. Conclusion
This paper has provided a preliminary study of
the approximation properties of a particular class
of rational orthonormal bases that are suitable for
continuous-timesystem modelling. The main result
was to establish that the bases were capable of
arbitrarily good approximation with respect to
a wide variety of norms employed in the system-
theoretic analysis of stable systems. The utility of
272
H. Akc 7 ay, B. Ninness / Signal Processing 77 (1999) 261}274

Page 13

the generalising nature of the particular bases con-
sidered here was also exposed by establishing that
signi"cantly improved "nite-order approximation
accuracy was possible by exploiting the #exibility
in allowed pole position. This is in contrast to the
more well known Laguerre and two-parameter
Kautz bases, which are obtained as special cases of
the bases considered here by choosing all the poles
"xed at one location.
Acknowledgements
The authors would like to thank Dr. Reza
Moheimani for providing the experimental data
used in the application example. Akc 7 ay gratefully
acknowledges the partial support of Alexander
von Humboldt Foundation. Ninness gratefully ac-
knowledges the support of CIDAC and the Austra-
lian Research Council.
References
[1] M. Abramowitz, I.A. Stegun, Handbook of Mathematical
Functions with Formulas, Graphs and Mathematical
Tables, Dover, New York, 1965.
[2] N. Achieser, Theory of Approximation, Dover, New York,
1992.
[3] H. Akc 7 ay, B. Ninness, Rational basis functions for robust
identi"cation from frequency and time domain measure-
ments, Automatica 34 (1998) 1101}1117.
[4] H. Akc 7 ay, B. Ninness, Orthonormal basis functions for
continuous-time systems and ¸?convergence, Math. Con-
trol, Signals Systems (1999), to appear.
[5] D.S. Bayard, Multivariable frequency domain identi-
"cation via 2-norm minimization, Proceedings of the
American Control Conference, Chicago, IL, 1992, pp.
1253}1257.
[6] D.S. Bayard, High-order multivariable transfer function
curve "tting: Algorithms, sparse matrix methods and ex-
perimental results, Automatica 30 (1994) 1439}1444.
[7] P. Bodin, T. Oliveira e Silva, B. Wahlberg, On the con-
struction of orthonormal basis functions for system identi-
"cation, Proceedings of 13th IFAC World Congress, San
Francisco, 1996, pp. 291}296.
[8] J. Bokor, L. Gianone, Z. Szabo, Construction of generalis-
ed orthonormal bases in H?, Tech. Report, Computer and
Automation Institute, Hungarian Academy of Sciences,
1995.
[9] W. Cluett, L. Wang, Frequency smoothing using Laguerre
model, in: Proc. IEE-D 139 (1992) 88}96.
[10] G. Davidson, D. Falconer, Reduced complexity echo can-
cellation using orthonormal functions, IEEE Trans. Cir-
cuits Systems 38 (1991) 20}28.
[11] A.C. den Brinker, Laguerre-domain adaptive "lters, IEEE
Trans. Signal Process. 42 (1994) 953}956.
[12] N.F. Dudley Ward, J.R. Partington, Robust identi"ca-
tion in the disc algebra using rational wavelets and or-
thonormal basis functions, Internat. J. Control 64 (1996)
409}423.
[13] N.F. Dudley Ward, J.R. Partington, A construction of
rational wavelets and frames in Hardy}Sobolev spaces
with applications to system modelling, SIAM J. Control
Optim. 36 (1998) 654}679.
[14] B. Epstein, Orthogonal Families of Analytic Functions,
Macmillan, New York, 1965.
[15] J.B. Garnett, Bounded Analytic Functions, Academic
Press, New York, 1981.
[16] J.Golub, C. Van Loan, Matrix Computations,Johns Hop-
kins University Press, Baltimore, MD, 1989.
[17] J. Head, Approximation to transient by means of Laguerre
series, Proc. Cambridge Philos. Soc. 52 (1956) 640}651.
[18] P. Heuberger, P.M.J. Van den Hof, O. Bosgra, A generaliz-
ed orthonormal basis for linear dynamical systems, IEEE
Trans. Automat. Control AC-40 (1995) 451}465.
[19] W.H. Kautz, Network synthesis for speci"ed transient
response, Tech. Report 209, Massachusetts Institute of
Technology, Research Laboratory of Electronics, 1952.
[20] W.H. Kautz, Transient synthesis in the time domain, IRE
Trans. Circuit Theory 1 (1954) 29}39.
[21] Y. Lee, Synthesis of electric networks by means of the
Fourier transforms of Laguerre's functions, J. Math. Phys.
XI (1933) 83}113.
[22] E.C. Levy, Complex curve "tting, IRE Trans. Automat.
Control AC-4 (1959) 37}44.
[23] P. Ma K kila, Approximation of stable systems by Laguerre
"lters, Automatica 26 (1990) 333}345.
[24] P. Ma K kila K , Laguerre series approximation of in"nite di-
mensional systems, Automatica 26 (1990) 985}995.
[25] P. Ma K kila K , Laguerre methods and H? identi"cation of
continuous-time systems, Internat. J. Control 53 (1991)
689}707.
[26] F. Malmquist, Sur la de H termination d'une classe de fonc-
tions analytiques par leurs valeurs dans un ensemble
donne H de points, in: Proceedings of Comptes Rendus du
Sixie ` me Congre ` s des mathe H maticiens scandinaves, Copen-
hagen, 1925, pp. 253}259.
[27] T. McKelvey, H. Akc 7 ay, L. Ljung, Subspace-based multi-
variable system identi"cation from frequency response
data, IEEE Trans. Automat. Control AC-41 (1996)
960}979.
[28] J. Mendel, A uni"ed approach to the synthesis of or-
thonormal exponential functions useful in systems analy-
sis, IEEE Trans. Systems Sci. Cybernet. 2 (1966) 54}62.
[29] S.O.R. Moheimani, H.R. Pota, I.R. Petersen, Spatial con-
trol for active vibration control of piezoelectric laminates,
Proceedings of 37th IEEE Conference on Decision and
Control, Tampa, Florida, 1998, pp. 4308}4313.
H. Akc 7 ay, B. Ninness / Signal Processing 77 (1999) 261}274
273

Page 14

[30] X. Nie, D. Raghuramireddy, R. Unbehauen, Orthogonal
expansion of stable rational transfer functions, Electron.
Lett. 27 (1991) 1492}1494.
[31] B. Ninness, F. Gustafsson, A unifying construction of
orthonormal bases for system identi"cation, IEEE Trans.
Automat. Control 42 (1997) 515}521.
[32] U G . Nurges, Laguerre models in problems of approxima-
tion and identi"cation, in: Adaptive Systems, Plenum,
1987, pp. 346}352. Translatedfrom Avtomaticai Telemek-
hanika 3 (March 1987) 88}96.
[33] J.R. Partington, Approximation of delay systems by
Fourier}Laguerre series, Automatica 27 (1991) 569}572.
[34] H. Perez, S. Tsujii, A system identi"cation algorithm using
orthogonal functions, IEEE Trans. Signal Process. 38
(1991) 752}755.
[35] R.Pintelon, P. Guillaume, Y. Rolain, J. Schoukens, H. Van
Hamme, Parametric identi"cation of transfer functions in
the frequency domain } a survey, IEEE Trans. Automat.
Control 39 (1994) 2245}2260.
[36] D. Ross, Orthonormal exponentials, IEEE Trans. Com-
mun. Electron. 71 (1964) 173}176.
[37] W. Rudin, Real and Complex Analysis, third ed.,
McGraw-Hill, New York, 1987.
[38] C.K. Sanathanan, J. Koerner, Transfer function synthesis
as a ratio of two complex polynomials, IEEE Trans. Auto-
mat. Control AC-8 (1963) 56}58.
[39] M.D. Sidman, F.E. DeAngelis, G.C. Verghese, Parametric
system identi"cation on logarithmic frequency response
data, IEEE Trans. Automat. Control AC-36 (1991)
1065}1070.
[40] S. Takenaka, On the orthogonal functions and a new
formula of interpolation, Jpn. J. Math. II (1925) 129}145.
[41] B. Wahlberg, System identi"cationusing Laguerremodels,
IEEE Trans. Automat. Control AC-36 (1991) 551}562.
[42] B. Wahlberg, System identi"cation using Kautz models,
IEEE Trans. Automat. Control AC-39 (1994) 1276}1282.
[43] B. Wahlberg,P. Ma K kila K , Onapproximationof stablelinear
dynamical systems using Laguerre and Kautz functions,
Automatica 32 (1996) 693}708.
[44] C. Zervos, P. Belanger, G. Dumont, Controller tuning
using orthonormal series identi"cation, Automatica 24
(1988) 165}175.
274
H. Akc 7 ay, B. Ninness / Signal Processing 77 (1999) 261}274