THEORETISCHE ARBEIT at 4/2000

A New LMI Approach to Analysis of

Linear Systems Depending on Scheduling
Parameter in Polynomial Forms
Takehito Azuma, Ryo Watanabe, Kenko Uchida, Masayuki Fujita

Dedicated to Prof. Dr.-Ing. Dr.-Ing. E. h. Gçnther Schmidt on the occasion of his 65th birthday

This paper proposes a new LMI approach to analysis of linear systems depending on schedul-
ing parameter in polynomial forms: we first propose a method to reduce the parameter de-
pendent LMI condition, which characterizes internal stability and L2 gain, to the finite number
of LMI conditions by introducing a convex polyhedron which includes a polynomial curve
parameterized by scheduling parameter; next we propose a systematic procedure to construct
the convex polyhedron. Our approach enable us to analyze L2 gain of linear systems with
scheduling parameter in polynomial forms through computation of the finite number of LMIs.
To show efficacy of our approach, we finally make a numerical experiment of L2 gain analysis
for a gasturbine engine model which is described as a linear system with a scheduling para-
meter in polynomial form of two degree.

1 Introduction The work [3] develops the method to obtain the para-
meter independent solution to the affinely parameter de-
Recently, many results on analysis and design of scheduled pendent LMI. In [5], the authors propose the method to
control have been presented for linear systems with sche- solve the quadratically parameter dependent LMI in the
duling parameters [1; 3; 5; 6; 7; 10; 11; 12; 13]. In these affinely parameter dependent set of solutions. In [13], a
results, analysis and design are characterized by solutions convex covering technique is applied to quadratically
to algebraic/differential Riccati inequalities which depends parameter dependent LMI with the affinely parameter de-
on the scheduling parameter [1; 3; 5; 10; 11; 12; 13]. Using pendent solutions.(The same case is discussed in [5].)
Schur complement [4], the parameter dependent algebraic/ However, it seems difficult to extend these method to
differential Riccati inequality can be described as a para- more general cases such as general polynomial depen-
meter dependent Linear Matrix Inequality (LMI). Except dence cases discussed in this paper.
for the particular case (e.g. dependence on the scheduling In this paper, we focus on the parameter dependent LMI,
parameter is affine [3] or quadratic [5; 13]), the infinite which characterize internal stability and L2 gain for linear
number of computation is required to solve this parameter systems depending on the scheduling parameter in polyno-
dependent LMI directly. mial forms, and propose a method to reduce this condition
An adhoc method (gridding method) to reduce this para- to the finite number of LMIs by introducing a convex poly-
meter dependent LMI to the finite number of LMIs is hedron. Generally volume of the convex polyhedron must
discussed in [12]. Although this method can be applied be small to get less conservative result. We also propose a
to the general parameter dependent LMI, it provides, in systematic procedure to construct such small convex poly-
general, an approximate solution which satisfies the LMI hedron.Thus by taking our approach, the polynomial para-
only at the finite number of gridding points in the para- meter dependent LMI is solved through the finite number
meter space, and requires the very large number of com- of computations, and the L2 gain analysis can be performed
putation to guarantee the feasibility of the approximate with CAD like LMILAB [8]. It is noted however that we
solution. In the case of linear systems depending affinely must solve the larger number of LMIs to obtain less con-
on the scheduling parameter, the specific convexifying servative solutions, because in the procedure of construct-
methods are proposed which reduce the parameter depen- ing convex polyhedron the smaller volume leads to solving
dent LMI to the fixed finite number of LMIs [3; 5; 13]. the larger number of LMIs.

at ± Automatisierungstechnik 48 (2000) 4 # Oldenbourg Verlag 199

Brought to you by | New York University Bobst Library Technical Services
Authenticated
Download Date | 7/28/15 4:10 AM
 at 4/2000 THEORETISCHE ARBEIT

This paper is organized as follows. In section 2, we present Remark 1 Using the parameter dependent Lyapunov
a sufficient condition on internal stability and L2 gain for function V ˆ x0 Q  t††x, the following inequalities are de-
linear systems with scheduling parameter. We formulate rived for the result of the L2 gain analysis.
our problem in section 3, and show the main result of this
Q † > 0; 2†
note in section 4. In section 5, we demonstrate the proposed
method by a numerical example.
dQ
A0 †Q † ‡ Q †A † ‡ _ †
d
‡ C 0 †C † ‡ 2
Q †B †B0 †Q † < 0; 3†
2 Analysis of L2 Gain
8
 2 ‰0; 1Š; 8 _ 2 ‰ max ; max Š:
We first summarize results of analysis of the following lin-
ear systems with a scalar scheduling parameter. The condition (1) is a sufficient condition for (2) and (3). In
[12], another sufficient condition which is less conservative
x_ t† ˆ A  t††x t† ‡ B  t††w t†; x 0† ˆ 0;
: is derived, but the dimension of the condition is larger than
z t† ˆ C  t††x t†;
that of the condition (1).
where x t† 2 Rn is the state, z t† 2 Rl is the observed out-
put, w t† 2 Rm is the disturbance, and  t† 2 R is the sche-
duling parameter of  which is a continuously differenti-
able function of t. We assume the following properties on 3 Problem Formulation
 t†, We assume that A †, B † and C † are described as the
i†  t† 2 ‰0; 1Š; 8 t 2 ‰0; 1†; polynomial function of  (See Remark 2),
ii† j_ t†j  max ; max > 0; 8 t 2 ‰0; 1†: X
La X
Lb X
Lc
A † ˆ i Ai ; B † ˆ i Bi ; C † ˆ  i Ci :
Each of A †, B † and C † is the continuous matrix func- iˆ0 iˆ0 iˆ0
tion of .
Internally stability and L2 gain of  are defined as follows. And we restrict Q † such that
Definition 1 The system  is said to be internally stable Lq
X
if the trivial solution x  0 of the following ordinary differ- Q † ˆ i Qi ; Q0i ˆ Qi :
ential equation is exponentially stable. iˆ0

x_ t† ˆ A  t††x t†: The parameter dependent LMI condition (1) is described as

follows,
Definition 2 Let  be an internally stable linear system
with scheduling parameter. Then the L2 gain of  is de- F † ˆ F0 Qs † ‡ F1 Qs † ‡


‡ r Fr Qs † < 0; 4†
fined by
where Qs denotes Q0 , Q1 ,


, QLq † and r ˆ Lq ‡ max
kzkL2
G † ˆ sup ; La , Lb , Lc †. Fi is a symmetric matrix function and depends
2
w2L ;w6ˆ0 kwkL2 affinely on the unknown matrix Qs .
where k
kL2 denotes L2 norm. Our problem is to describe the sufficient condition for ex-
Then we have the next theorem for L2 gain analysis of  by istence of solutions to the following polynomial parameter
applying Schur Complement [4] to the result in [10]. dependent LMI condition,
Theorem 1 The system  is internally stable and G † is F0 Qs † ‡ F1 Qs † ‡


‡ r Fr Qs † < 0; 8  2 ‰0; 1Š;
less than if there exists a matrix function Q † defined on
[0, 1] such that as the finite number of parameter independent LMI condi-
2 tions.
Q † 0 0
6 0  dQ † 0 Remark 2 The assumption on system matrices of the
6 d  0  system  is only continuity on  2 [0, 1], so there exist
6 A †Q † ‡ Q †A †
6 0 0 finite matrices Ea , Fa , Eb , Fb , Ec and Fc such that the fol-
6 max dQ
6 d † lowing system includes the system  for any La, Lb and Lc .
4 0 0 C † " #
0 0 B0 †Q † X
La
k
x_ t† ˆ  t†Ak ‡ Ea
a t†Fa x t†
3
0 0 kˆ0
" #
0 0 7 X
Lb
7 ‡ k t†Bk ‡ Eb
b t†Fb w t†;
C 0 † Q †B † 7
7 < 0; 1†
I 0 5 "
kˆ0
#
0 2
I X
Lc
k
z t† ˆ  t†Ck ‡ Ec
c t†Fc x t†;
for all  2 [0, 1], where  ˆ 1 or 1. kˆ0

200
Brought to you by | New York University Bobst Library Technical Services
Authenticated
Download Date | 7/28/15 4:10 AM
 T. Azuma et al.: A New LMI Approach to Analysis of Linear Systems Depending on Scheduling Parameter . . . at 4/2000

where
a t†,
b t†, and
c t† satisfies Now we consider the construction of a small size convex

0a t†
a t†  I; polyhedron H and propose the following procedure (See
Figure 1):

0b t†
b t†  I;
(Step 1) We divide the domain of  into f1 , 2 ,


,

0c t†
c t†  I: d g. We call d Division Number.
This observation motivates us to focus the case of system   
i 1 i
matrices of polynomial function, A †, B †, C †. i ˆ  j  2 ; : 9†
d d
We define a curve Ti as
4 Main Results Ti ˆ f‰; 2 ;


; r Š0 j  2 i g:

We show main results of this paper. First we propose a (Step 2) For each Ti , we define a polyhedron Hi whose
method to reduce the polynomial parameter dependent vertices are f p0i , p1i ,


, pri g.
"     #0
LMI condition (4) to the finite number of LMI conditions. i 1 i 1 2 i 1 r
0
In this method, a convex polyhedron plays an important pi ˆ ; ;


; ;
d d d
role, which includes a curve characterized by the schedul-
ing parameter. Next we propose a procedure to construct pji ˆ pji 1
‡ i; j† ej j ˆ 1; 2;


; r 1†;
this convex polyhedron. 10†
pri ˆ pri 1
‡ i; r†er
The proposed method is given as follows. "    r # 0
i i 2 i
Theorem 2 Let H be a convex polyhedron which in- ˆ ; ;


; ;
cludes a curve T defined as d d d
 0 where
T ˆ  2


r j  2 ‰0; 1Š :  j  
i i 1 j
And let vertices of H be fp1 , p2 ,


, pq g, where pi 2 Rr . i; j† ˆ ;
d d
Assume that there exists Qs which satisfies the following
LMI condition for all pi i ˆ 1, 2,


, q†, and e1 , e2 ,


, er are basis vector on Rr . Note that Hi is a
convex polyhedron on Rr because f pji p0i g j ˆ 1, 2,


,
F0 Qs † ‡ pi1 F1 Qs † ‡


‡ pir Fr Qs † < 0; 5† r† are linearly independent.
where pij is the jth element of pi . Then Qs satisfies (4) for (Step 3) We define the convex hull H as
any  2 ‰0; 1Š.
H ˆ Co H1 [ H2 [


[ Hd †; 11†
Proof (5) is equivalent to
where Co S† denotes the convex hull of S.
Xr
0 F0 Qs † ‡ pij 0 Fj Qs † < 0; 6†
jˆ1

for all  2 Rn ;  6ˆ 0. We express (6) as

0 F0 Qs † ‡ h f ; Qs †; pi i < 0; 7†
where h; i denotes inner product and
 0
f ; Qs † ˆ 0 F1 Qs † 0 F2 Qs †


0 Fr Qs † :
(7) is satisfied for all vertices of H and for all  2 Rn , so the
following condition is obtained
X
q
0 > 0 F0 Qs † ‡ i hf ; Qs †; pi i
iˆ1
X
q
ˆ 0 F0 Qs † ‡ hf ; Qs †; i pi i; 8†
iˆ1
P
for all i  0, riˆ1 i ˆ 1. This implies that Qs satisfies
the parameter dependent LMI condition (4) for all  2 [0,
1], because any point on T can be described as linear com-
bination of fp1 , p2 ,


, pq g.
Remark 3 Theorem 2 shows that Qs in Theorem 2 satis-
fies (5) for any point in H. Thus volume of the convex
polyhedron H is closely related to the conservativeness of
Figure 1: Con-
the evaluation. It is necessary to construct small size H for struction of con-
sharp evaluation. vex hull.

201
Brought to you by | New York University Bobst Library Technical Services
Authenticated
Download Date | 7/28/15 4:10 AM
 at 4/2000 THEORETISCHE ARBEIT

Theorem 3 The convex hull H given by (11) is a convex Remark 4 As the division number increases, conserva-
polyhedron which includes the curve T defined in Theo- tiveness improves but the number of LMIs also increases.
rem 2. (If the division number is d, the number of LMIs is d r ‡ 1†
Proof It is obvious that H is a convex polyhedron includ- where r is the degree of polynomial.) Using our approach,
ing Hi i ˆ 1, 2,


, d†, because H is the convex hull of H1 we must consider such trade-off between improvement of
[ H2 [


[ Hd . So we show that Hi includes Ti . In this conservativeness and increase of the number of LMIs.
proof, we first describe T † as linear combination of fp0i ,
p1i ,


, pri g, where T † is a point on Ti and is defined as
T † ˆ ‰ 2


r Š0 ;  2 i :
5 Numerical Example
Next we show that each coefficient of its linear combina-
tion is not less than 0 and the sum of all coefficients is equal Consider the following linear system with scheduling para-
to 1. meter,
 
T † can be expressed as follows, x_ t† ˆ A0 ‡  t†A1 ‡ 2 t†A2 x t†
 
T † ˆ e1 ‡ 2 e2 ‡


‡ r er ‡ B0 ‡  t†B1 ‡ 2 t†B2 w t†; 13†
ˆ p0i p0i ‡ e1 ‡ 2 e2 ‡


‡ r er z t† ˆ Cx t†; 14†
  (  )
0 i 1 2 i 1 2 where ‰A0 j A1 j A2 Š, ‰B0 j B1 j B2 Š, C are given as fol-
ˆ pi ‡  e1 ‡  e2 lows,
d d 2
   4:3650 0:67230 0:33630
i 1 r   6
‡


‡ r er : A0 j A1 j A2 ˆ 4 7:0880 6:5570 4:6010
d
2:4100 7:5840 14:310
By using (10), T † is described as linear combination of pji
0:56081 0:85534 0:58923
j ˆ 0, 1,


, r†.
(  ) 0:66981 1:3750 0:99093
X r
i 1 j
T † ˆ pi ‡ 0
 j
i; j† 1 fpji pij 1 g 3:1917 1:7971 2:5887
d 3
jˆ1 0:66981 1:3750 0:99093
    7
i 1 2:8963 1:5292 10:516 5;
ˆ 1  i; 1† 1 p0i
d 3:5777 2:8389 1:9087
"(  )
Xr 1
i 1 j 2
‡  j
i; j† 1 2:3740 0:74850
jˆ1
d   6
(   ) # B0 j B1 j B2 ˆ 4 1:3660 3:4440
i 1 j‡1 0:94610 9:6190
 j‡1
i; j ‡ 1† 1
pji
d 0:16023 0:35209
  
i 1 r 0:11622 2:4839
‡  r
i; r† 1 pri : 12†
d 0:11058 4:6057
3
Let be 0:15623 0:13063
X
r 7
0:49582 4:0379 5;
T † ˆ j †pji ;
jˆ0 0:030616 0:89473
 
i 1  
0 † ˆ 1  i; 1† 1 0 1 0
d Cˆ ;
0 0 1
(  )
i 1 j  t† 2 ‰0; 1Š; 8 t 2 ‰0; 1†:
j 1
j † ˆ  i; j†
d This is the model of the gasturbine engine [9] described as
(   ) a linear system with a scalar scheduling parameter, where
i 1 j‡1 1
j‡1 i; j ‡ 1† x1 , x2 and x3 are the compressor speed, the fan speed and
d the outlet pressure respectively, and  is the normalized
j ˆ 1; 2;


; r 1† compressor speed. The scheduled system matrices, A †
   ˆ A0 ‡ A1 ‡ 2 A2 , B † ˆ B0 ‡ B1 ‡ 2 B2 , are given
i 1 r 1
r † ˆ r i; r† as interpolation of the three system matrices, which are
d
identified by using maximum likelihood method at three
It is obvious that the sum of j † j ˆ 0, 1,


, r† is equal to equilibrium points of the nonlinear simulation model, as
1. It is easy to show that j † j ˆ 0,


, r† is not less than 0 shown in Figure 2. In this figure,  denotes the equili-
for all  2 i . brium point.

202
Brought to you by | New York University Bobst Library Technical Services
Authenticated
Download Date | 7/28/15 4:10 AM
 T. Azuma et al.: A New LMI Approach to Analysis of Linear Systems Depending on Scheduling Parameter . . . at 4/2000

a11 (θ) a12(θ) a13(θ) 1.06

–4.2 –0.4 0
Case 1
–0.6 –0.2 Case 2
–4.3
1.04 Case 3

a12

13
a11

–0.8 –0.4

a
–4.4
–1 –0.6

–4.5 –1.2 –0.8 1.02

0 0.5 1 0 0.5 1 0 0.5 1
θ θ θ

a (θ) a (θ) a (θ)

L2 gain
21 22 23
8 –6 0 1

–7 –2
7.5
a21

22

23
–8 –4
a

a
7 0.98
–9 –6

6.5 –10 –8
0 0.5 1 0 0.5 1 0 0.5 1
θ θ θ 0.96

a31(θ) a32(θ) a33(θ)

–1.5 14 –14
0.94
12 0 5 10 15 20 25
–2 –14.5
Division Number
a31

a32

33

10
a

–2.5 –15
Figure 3: Evaluation of L2 gain.
8

–3 6 –15.5
0 0.5 1 0 0.5 1 0 0.5 1
θ θ θ

tion is improved as the division number increases; conser-

vativeness is greatly improved while the division number is
b (θ)
11
b (θ)
12
small; conservativeness is improved very little at the divi-
0.9
sion number larger than some number (e.g. 20 in this exam-
ple). Thus it seems that we can obtain good evaluation of
0.8
2.35
b12
b11

L2 gain even if the division number is not enough large.

0.7

0.6

2.3
0 0.5 1
0.5
0 0.5 1
If the interdependence of the parameters, 1 t† ˆ  t†, 2 t†
θ θ
ˆ 2 t†, is neglected, the system (13) is reduced to the lin-
1.6
b
21
( θ)

5
b
22
( θ)
ear systems with two, independent, scheduling parameters,
1.4 4.5
x_ t† ˆ ‰A0 ‡ 1 t†A1 ‡ 2 t†A2 Šx t†
15†
21

22

‡ ‰B0 ‡ 1 t†B1 ‡ 2 t†B2 Šw t†;

1.2 4
b

1 3.5

0.8
0 0.5 1
3
0 0.5 1
where 1 t† and 2 t† satisfy
θ θ

b (θ)
31
b (θ)
32
1 t† 2 ‰0; 1Š; 2 t† 2 ‰0; 1Š; 8 t 2 ‰0; 1†;
_1 t† 2 ‰ 10; 10Š; _2 t† 2 ‰ 20; 20Š; 8 t 2 ‰0; 1†:
1 –8

0.95
–10

Since the system (15) depends affinely on scheduling para-

b31

b32

0.9
–12
0.85
meters, the technique of [5] can be applied. That is, taking
the solution to the vector form of (3) in the affine form,
0.8 –14
0 0.5 1 0 0.5 1
θ θ

Q 1 ; 2 † ˆ Q0 ‡ 1 Q1 ‡ 2 Q2 :
and using the technique of [5], we evaluate the L2 gain as
Figure 2: System parameters.
1 ˆ 1:5764:
We evaluate L2 gain of the above system for the next three The corresponding result of the L2 gain analysis using our
cases, technique is that of Case 2, and the value is evaluated from
Case 1: Q † ˆ Q0 , Figure 3 as
Case 2: Q † ˆ Q0 ‡ Q1 ‡ 2 Q2 and max ˆ 10:0, 2 ˆ 0:960:
Case 3: Q † ˆ Q0 ‡  Q1 ‡ 2 Q2 ‡ 3 Q3 and max ˆ
Thus we can see less conservative evaluation of the L2 gain
10.0.
by considering interdependence of scheduling parameters.
Note that, in these cases, we cannot evaluate L2 gain by
using directly the technique in [3] or [5], because the sys-
tem parameters A † and B † are quadratic function of .
6 Conclusions
The result is shown in Figure 3. In this figure, the vertical
axis denotes L2 gain evaluated by the proposed technique We proposed a new LMI approach to analysis of linear sys-
and the horizontal axis denotes the division number. ‡,  tems depending on scheduling parameter in polynomial
and  are L2 gain evaluated for Case 1, Case 2 and Case 3 forms. In the numerical case study, we evaluated actual L2
respectively. Results for Case 1 and Case 2 are almost iden- gain for a given linear system with scheduling parameter
tical. This result shows that conservativeness of the evalua- and verified its efficacy. Though we considered the polyno-

203
Brought to you by | New York University Bobst Library Technical Services
Authenticated
Download Date | 7/28/15 4:10 AM
 at 4/2000 THEORETISCHE ARBEIT

mial case in this paper, the proposed approach can be ex- Manuscript received: 16th August 1999.
tended to more general cases [2].

References
[1] P. Apkarian and R. J. Adams, Advanced Gain-Scheduling Dr. Takehito Azuma is Research Associate for De-
Techniques for Uncertain Systems, In Proc. of American Con- partment of Electrical, Electronics and Computer
trol Conference, pp. 3331±3335, 1997. Engineering at Waseda University. His current re-
[2] T. Azuma, R. Watanabe, and K. Uchida, An Approach to Sol- search interests are in the fields of gain scheduling
ving Parameter-Dependent LMI Conditions Based on Finite and robust control.
Number of LMI Conditions, In Proc. of American Control Address: Department of Electrical, Electronics and
Conference, pp. 510±514, 1997. Computer Engineering, Waseda University, Okubo
[3] G. Becker, A. Packard, D. Philbrick and G. Balas, Control of 3-4-1, Shinjuku, Tokyo 169-8555, Japan; E-Mail: ta-
Parametrically-Dependent Linear Systems : A Single Quadra- [email protected]
tic Lyapunov Approach, In Proc. of American Control Confer-
ence, pp. 2795±2799, 1993.
[4] S. Boyd, L. El Ghaoui, E. Feron and V. Balakrishnan, Linear Dr. Ryo Watanabe is Associate Professor for Depart-
Matrix Inequalities in System and Control Theory, SIAM, stu- ment of Computer-Controlled Mechanical Enginee-
dies in applied mathematics, vol. 15, 1994. ring at Osaka University, Japan. From 1989 to 1997,
he was with Ishikawajima Harima-Heavy Industry.
[5] P. Gahinet, P. Apkarian, M. Chilali, Affine Parameter-Depen- His research interests include robust control, nonli-
dent Lyapunov Functions and Real Parametric Uncertainty, near control and their engineering applications.
IEEE Trans. Automat. Contr., vol. 41, pp. 436±442, 1996. Adddress: Department of Computer-Controlled Me-
[6] A. Packard, Gain Scheduling via Linear Fractional Transfor- chanical Engineering, Osaka University, Yamada-
mations, Syst. & Control Letters, Vol. 22, No. 2, pp. 79±92, oka 2-1, Suita, Osaka 565-0871, Japan; E-Mail: wata-
1994. [email protected]
[7] J. S. Shamma and M. Athans, Gain Scheduling: Potential, Ha-
zards and Possible Remedies, IEEE Control Systems Maga- Prof.Dr. Kenko Uchida is Professor at Waseda Uni-
zine, Vol. 12, pp. 101±107, 1992. versity. From 1982 to 1983, he was a Research Fellow
[8] P. Gahinet, A. Nemirovski, A. J. Laub and M. Chilali, LMI of the AvH Foundation in the Institute of Automatic
Control Toolbox for Use with MATLAB, The Math Work- Control Engineering at the Technical University of
s,Inc, 1995. Munich. His research interests are in the areas of ro-
bust control and optimization.
[9] R. Watanabe, M. Kurosaki, K. Uchida, E. Shimemura and M.
Fujita, H1 Control of Gasturbine Engines for Helicopters, In Address: Department of Electrical, Electronics and
Proc. of American Control Conference, pp. 1123±1127, 1993. Computer Engineering, Waseda University, Okubo
3-4-1, Shinjuku, Tokyo 169-8555, Japan; E-Mail: ku-
[10] R. Watanabe, K. Uchida, M. Fujita, and E. Shimemura, L2 [email protected]
Gain and H 1 Control of Linear Systems with Scheduling Pa-
rameter, In Proc. of 33rd IEEE Conference on Decision and
Control, pp. 1412±1414, 1994. Prof.Dr. Masayuki Fujita is Professor at Kanazawa
University, Japan. From 1994 to 1995, he held a visi-
[11] R. Watanabe, K. Uchida, M. Fujita and E. Shimemura, H 1 ting position in the Institute of Automatic Control
Control of Linear Systems with Scheduling Parameter, Trans. Engineering at the Technical University of Munich.
SICE, Vol. 31, No. 4, 1995. (In Japanese) His research interests include robust and nonlinear
[12] F. Wu, A. Packard, G. Becker, Induced L2 -Norm Control for control.
LPV System with Bounded Parameter variation Rates, In Address: Department of Electrical and Computer
Proc. of American Control Conference, pp. 2379±2383, 1995. Engineering, Kanazawa University, 2-40-20 Kodatsu-
[13] J. Yu, A Sideris, H1 control with parametric Lyapunov func- no, Kanazawa 920-8667, Japan; E-Mail: [email protected]
tions, In Proc. of 34th IEEE Conference on Decision and nazawa-u.ac.jp
Control, pp. 2547±2552, 1995.

204
Brought to you by | New York University Bobst Library Technical Services
Authenticated
Download Date | 7/28/15 4:10 AM