Академический Документы
Профессиональный Документы
Культура Документы
2, APRIL 1982
tions. the zeros of p,,- 1 ( 5 ) lie close to or inside the convex hull of the
zeros of p n ( s ) . In particular. if all zeros of p , ) ( 5 )have negative real part,
thenthesame
is true forthe zeros of p,-,(s). This is shownbythe
5 ) negative real part, then the
follow-ing argument: if all zeros of ~ , ~ ( have
to thelemma.the
sameholds for its reciprocalpolynomial.According
zeros of the derivative of the reciprocal polynomial \vi11 also have negative
p , , - , ( 5 ) . the normalized
real part. Finally. thesame will betruefor
reciprocal polynomial of the derivative.
Similarly, it can be shown that if all zeros of p J s ) have positive real
part. then the same is true for the zeros of p,#- 1 ( 51. and if all zeros of
p,,( 5 ) are real (imaginar).). then all the zeros of p,#- ,( 5 ) are real (imaginary) and inside the convex hull.
AIgorirhm: Let the transfer function be
I. INTRODUCTION
It is with somehesitation that we mingle into whatseems to be an
internal scientific affair [I]-[3]. butit is our feeling that themodel
reduction problem is an important one which has received far too little
attention in the literature. All contributions to the field. small or great.
should therefore be equally welcome.
Recall that themethodsuggestedbyKrishnamurthyandSeshadri
consists inseparately reducing the degrees of thenumeratoranddenominator polynomials of a given transfer function. Lotver order polynomials are generated from the Routh stability array.
Seriouscriticism on thesuggestedmethodhasbeenexertedbyRao.
Lamba. and Rao in [2] and by Sin& in [3]. Singhs critique brings fonvard
the so-called nonuniqueness of the algorithm. whereby it is meant that
different transfer functions give rise to the same model. Since any model
reductiontechnique
necessarily containsanoninjectivemapping.this
point is somewhat hard to understand, however.
As to the discussion in [2] and the companion Authors Reply. it is not
our intention to judge on the theoretical justification or the applicability
of the suggested method. Instead. wewish to point out that there
is a
great variety of ways to reduce the order of a given polynomial. Of these
we have chosen to study differentiation. an operation well known to all
persons with even a superficial contact n-ith mathematical analysis.
11.
THE ALGORITHM
Pn(5)=U;
(1)
(s-:,)
,=I
then the zeros of p,:(s) d o not lie outside the convex hull of the zeros of
P( 5 1.
Proof: T h ~ result
s
is originally due to F. Gauss and F. Lucas. A proof
can be found in many elementary textbooks on analytical functions. e g .
[j.p. 841.
3
A drawback of straightfonvard differentiation is that zeros with a large
modulus tend to be better approximated thanthose with a small modulus.
Thisproblem is remedied. e.g.. by differentiatingthe reciprocalpoly-
0
111. ERROR ASALYSIS
P L E
8
494412s4+6681024s3+30708720s2+579556XO~+40840800
H,(s) = - .
5 18102~+284880s4+1648200s~+4499040~2+6064800s+3225600
nomial.reciprocatingback,
and normalizing. tiiventhepolynomial
the reduced order polynomial then becomes
(1).
and
H,(s) = 4 .
(2)
F~~ polynomials
Institute of Technol-
26994240~+14555j200~+19~536000~
The pole-zero locationsaretabulated
inTable I. It is apparenthow well
the poles and zeros of the reduced-order systems approximatethose of the
original system. Note especially that the number of complex poles and
Of the approsimants never exceed the number Of complex poles and
zeros of the original
transfer
function.
This isin
case
indeed
the
not
[ 11.
0018-9ZX6/82/0400-0454500.75
cl982 IEEE
AC-27.NO. 2.455
APRIL 1982
prohlern." Dep. Automar. Cmtr.. Lund ImtTechnd. Lund. Swcdcn. Tech. Rep.
CODEX: L U T F D ~ / [ T F R T - ? ~ I I I , ~ I - ~ I ~ , ; ~Dec
~ I YIYXO
X~).
j5] E. Hrlle. .Afluhrrwl F~uxrro~r
T h e m . bo1 I
Kew York: Bla~sddell.1959.
TABLE I
Order of
Approximant
Poles
r=8
(exact)
-5.
-I
Zeros
= /. - 1. -3. -4.
- 1.03C0.63li.-2.64.
-8. -10
r=l
-7.80,
-4.90.
-3.83,
-9.7ga
-1.12.-1.19~1.06r.
-3.78,
-4.41.
-6.24.
-4.15.
- 9.05
r=6
-1.20=0.668r.-2.93.
-6.06,-8.83
r=5
- 1.48. - 1.802 1.09;.
-6.19 -6.45 -4.21.
r=4
-2.15~0.6391, -4.90
- 1.76, -2.29-CO.948i.
1.71 kO.698i.
-3.89.
- 5.23
r=3
-2.18.-2.79,-3.22
-2.65,-3.02
r=2
-7.38,-3.01
r= I
- 2.66
- 2.82
I.
INTRODUCTION
Let R"""(s) be the set of rational HI X H matrices and let & be the ring
PE R"'""(s)
of proper and stable rational functions. Consider a plant
and let (!V. D ) and ( E , ,i') beany right coprimeandleftcoprime
Then there exist matrices X and Y over 5 of
factorizations of P over
appropriate dimension such that
s.
XX
+ YD= I .
It has been shown [ I ] that the set of controllers that stabilize the plant
P with unity feedback (Youla compensators) is given by
~ ^ ( P ) = { ( Y - R , ~ ) - ' ( x + R E ) : RES^'"^^}.
0 01
'
'
'
'
'
103
'
'
wlmd/sI
Y-R.i'=Li(I+VY)
Fig. I .
X+ R D = C V X .
(1)
Postmultiplyingbothsides of theequation by
[E ];
-
andrecalling
ACKNOWLEDGMENT
we have
REFERENCES
V. Krishnamurth? and V . Seshadn. "Model reduction using the Routh stability
criterion." I E E E Trulrs. A r o o n m . Comr rol. AC-23. pp 729-731. Aug 197X.
A. S Rao. S. S. Lamha. and S. V Rao. "Comment> on 'Model rcductlon using the
Routh stability criterion."' I E E E Trmo. .Auronwr Conrr.. bo1 A C - I 4 p 5 1 8 . June
="[D-V,-X].
1979.
0018-92S6/82/~~-0455$00.75'El982 IEEE