Robust Mixed Control and LPV Control with Full Block Scalings

Carsten W. Scherer

For systems a ected by time-varying parametric uncertainties, we devise in this chapter a technique that allows to equivalently translate robust performance analysis speci cations characterized through a single Lyapunov function into the corresponding analysis test with multipliers. Out of the multitude of possible applications of this so-called full block S-procedure, we will concentrate on a discussion of robust mixed control and of designing linear parametrically-varying (LPV) controllers.

Abstract

In the past years mixed control problems have attracted considerable interest, and the LMI approach to controller design has shown bene cial to tackle design speci cations that have deemed untractable. For the design of output feedback controllers, it has been revealed quite recently how a suitable non-linear change of controller parameters allows to proceed in a straightforward fashion from an analysis speci cation for a controlled system formulated in terms of matrix inequalities to the corresponding synthesis inequalities for controller design 6, 12, 20]. One purpose of this chapter is to extend this general paradigm to robust performance speci cations if the underlying system is a ected by time-varying parametric uncertainties 13, 22]. We introduce a general technique that allows to equivalently translate robust performance objectives formulated in terms of a common Lyapunov function into the corresponding analysis test with multipliers. Similar approaches, as those in 13, 22], are usually based on the so-called S-procedure that introduces conservatism since the resulting multipliers have, in general, a block-diagonal structure. Instead, we propose to use full block multipliers that are only indirectly described by linear matrix inequalities such that the reformulation will not introduce conservatism. Hence we call the underlying abstract result for quadratic forms a full block S-procedure. For robust stability speci cations with a ne dependence on the uncertainties, such an equivalent reformulation has been provided before in 15].
Mechanical Engineering Systems and Control Group, Delft University of Technology, Mekelweg 2, 2628 CD Delft, The Netherlands

1 Introduction

Scherer

The main goal of this chapter is to generalize these ideas to robust performance speci cations if the matrices describing the system can be rational functions of the parameters, and to base the transformation on one abstract result that can be applied with ease to several performance speci cations. Moreover, we will provide a discussion of the (straightforward) consequence for robust controller design. Related results for the robust stabilization problem have appeared in the inspiring paper 10]. Finally, we will turn to the design of linear parametrically varying controllers that has been fully tackled for blockdiagonal multipliers 14, 9, 21]. Our purpose is to show how to overcome this structural restriction to arrive at a solution even if the multipliers are full block matrices that are only indirectly described by LMIs. Contrary to previous approaches one has to modify the structure of the scheduling function: one cannot just employ a copy of the parameters to schedule the controller but one has to use a quadratic function thereof. We concentrate on systems that are a ected by time-varying parametric uncertainties. Since we deal with linear fractional system representations, we assume that all these parameters are collected in the matrix . The admissible set of values that can be taken by the uncertainties is denoted as Rk l with 0 2 and assumed to be compact. Note that captures both the size of the uncertainties as well as their structure. Typically, for a rational dependence of the system description on the parameters, the matrix can be taken blockdiagonal, and the blocks take a real repeated structure for the results in this chapter, such a structure is not required. Given the value set , the actual time-varying parametric uncertainties consist of all continuous curves : 0 1) ! : Let us now look at (1) 0 x Bm B 1 0 x 1 _ 1 0 A B1 B2

2 System Description

u 0 a linear time-invariant system in which the uncertainties enter in a linear fractional fashion to de ne the time-varying uncertain system under investigation. Obviously, w1 ! z1 constitutes the uncertainty channel, whereas wj ! zj , j = 2 : : : m, are the performance channels used to describe the desired performance speci cations, and u ! y is the control channel. y C F1 F2 Fm

B z1 B B z2 B B .. B B . B @ zm

C B C1 D11 D12 D1m C B C B D2m C B C2 D21 D22 C =B .. .. .. . . . .. C B C B . . . . C A B @ Cm Dm1 Dm2 Dmm

E1 E2
.. .

Em

C B w1 C B C B w2 C B C B .. C B C B . C B A @ wm

C C C C C w1 = (t)z1 C C C A

Robust Mixed and LPV Control with Full Block Scalings

In fact, u is the control input and y is the measured output, and any linear time-invariant system that closes the loop as (2)

x _c u

c Bc = A Cc Dc

xc y

is said to be a controller. The resulting controlled system admits the description (3) 0 _ 1 0 A

B B B B B B @

z1 z2

zm

C B C B C B C B = C B .. C B A @ .

C1 C2
.. .

B1 B2 D11 D12 D21 D22

.. . .. .

Cm Dm1 Dm2

...

Bm D1m D2m

Dmm

10 C B C B C B C B C B .. C B A @ .

w1 w2 wm
.. .

1 C C C C w1 = (t)z1 C C A

with the realization (4)

! 0 A + BDcC BCc Bj + BDcFj 1 A Bj = B C B C A B F @ A: c c c j Ci Dij Ci + Ei DcC Ei Cc Dij + Ei DcFj

The multi-objective robust controller design problem can be sketched as follows: Find a controller that exponentially stabilizes the closed-loop system, and that achieves a mixture of various performance speci cations on the di erent performance channels of the controlled system for all possible uncertainties a ecting the system. In the next section we extend well-known analysis tests for nominal performance as described in 20] to the corresponding robust performance tests against time-varying uncertainties in terms of a constant quadratic Lyapunov function. The essential new ingredient will be the ease how one can derive the corresponding multiplier test by just referring to the abstract full block S-procedure that is presented in Lemma A.1 in the appendix.

3 Robust Performance Analysis with Constant Lyapunov Matrices

In this section, we provide analysis tests that are based on nding a constant quadratic Lyapunov function in order to guarantee the following properties: Well-posedness of the linear fractional transformation used to describe the uncertain system. Uniform (in the uncertainty) exponential stability. Robust performance, speci ed as quadratic performance (such as bounding the L2 -gain or dissipativity) or as bounding the H2 norm, the generalized H2 norm, or the peak-to-peak gain.

Scherer

3.1 Well-Posedness and Robust Stability

We call the description (3) well-posed if (5)

alternative representation ! _ (6) z =

i

I ; D11 is non-singular for every 2 : Then the channel wj ! zi of the uncertain closed-loop system admits the

A( Ci(

(t)) (t))

Bj ( Dij (

(t)) (t))

!
wj

with the function ! A( ) Bj ( ) = Ci( ) Dij ( ) ! ;1 C ;1 D A + B ( I ; D ) B + B ( I ; D ) 1 11 1 j 1 11 1 j = C + D (I ; D );1 C D + D (I ; D );1 D i i1 11 1 ij i1 11 1j that is, due to (5), continuous (and even smooth) on . If the interconnection is well-posed, (3) or (6) are uniformly exponentially stable if there exist constants K and > 0 such that, for every uncertainty (:) and for every unforced (w2 = 0 : : : wm = 0) system trajectory (:),

k (t)k Ke;

(t;t0 ) k

(t0 )k for all t t0 0:

Under the hypothesis (5), it is well-known that uniform exponential stability is guaranteed by the existence of an X > 0 that satis es the Lyapunov inequality A( )T X + X A( ) < 0 for all in the admissible value set . In the following result the Lyapunov inequality is rewritten into a form that facilitates the application of the full block S-procedure. Theorem 3.1. Suppose the interconnection (3) is well-posed, and suppose there exists an X > 0 satisfying (7)

A(

!T

0 X X 0

A(
!

< 0 for all

2 :

Then the uncertain system (3) is uniformly exponentially stable. Proof. The very standard proof is only included for the convenience of the reader. By compactness of , there exists an > 0 such that

A(

!T

0 X X 0

A(

+ X <0

for all 2 . Now suppose (:) is any admissible uncertainty curve, and let (:) be an arbitrary unforced system trajectory. With v(t) := (t)T X (t), we obtain v _ (t) + v(t) 0 for all t 0

Robust Mixed and LPV Control with Full Block Scalings

by left-multiplying (t)T and right-multiplying (t). With the integrating factor e t , this implies

v(t) v(t0 )e; (t;t0 ) for all t t0 0:

Due to
min(X )k

(t)k2

k (t)k

v(t)

max (X )k

(t)k2 we nally obtain (t0 )k for all t t0 0

max(X ) e; 2 (t;t0 ) k min (X )

what nishes the proof with explicit formulas for the constants and K . The condition (7) is formulated in terms of the rational function A( ) on the whole set such that it is pretty hard to verify directly. The main purpose of the full block S-procedure proposed in this chapter is to equivalently reformulate this test into a more explicit condition that makes use of multipliers. Here, the relevant set of multipliers is de ned as (8)

8 < P := :P 2 R(k+l)

(k+l)

: P = PT

!T
I

!
I

> 0 for all 2

9 =

Whenever required we will tacitly assume that any such multiplier is partitioned as ! ! Q S P = S T R conformable to I : For robust stability, we will reveal in detail how to apply Lemma A.1 such that we can be very short in extending the technique to robust performance tests. Theorem 3.2. The interconnection (3) is well-posed and X satis es (7) if and only if there exists a multiplier P 2 P with (9)

1 0 0 I 0 T 0 B B A B1 C C BX B C B @0 @ 0 I A B C1 D11 0

10 1 X 0 0 CB I 0 C 0 0 0 C B A B1 C AB @ 0 I C A < 0: 0 Q S C 0 ST R C1 D11

Proof. To apply Lemma A.1, introduce

0 0 X 0 01 0 0 0 I 0 1 BX 0 0 0C B B B1 A B1 C C B C B N =B S = im S = im B C B C B 0 @ 0 0 0 0A @ I @ 0 I A 0 0 0 0 C1 D11 D11
U= I ; T=
0 0 I 0 : 0 0 0 I

1 C C C A

and

Scherer

Hence SU = ker(UT ) \ S is described as

80 > I > <B A B B > > :@ C01

1 B1 C C A I C D11
0

1 2

9 > > = : ; C1 1 + (I ; D11 ) 2 = 0> : >

Therefore, we conclude that SU is complementary to S0 i I ; D11 is nonsingular. Moreover, if I ; D11 is non-singular, we arrive at the alternative description of SU as

0 B im B B @

A( ) (I ; D11 );1 C1 C1 + D11 (I ; D11 );1 C1

1 C C C A:

Then it is obvious that the following conditions are equivalent: The inequality (7) and N being negative de nite on SU , the inequality (9) and N + T T PT being negative on S , and, nally, the condition P 2 P and P being positive on ker(U ). The latter just follows from ! the observation that the kernel of U is nothing but the image of I . All this shows that Theorem 9 is a reformulation of Lemma A.1 what nishes the proof.

3.2 Robust Quadratic Performance

Suppose Ppi is a symmetric matrix that de nes a performance index. Then the robust quadratic performance (QP) speci cation on the channel i is formulated as follows: There exists an > 0 such that (10)

! Z 1 wi (t) !T Z1 w ( t ) i P dt ; wi (t)T wi (t) dt pi zi (t) zi (t)

0 0

holds for any trajectory of the uncertain system (3) with (0) = 0. The following technical hypothesis on the performance index is both indispensable to apply the full block S-procedure, and to arrive at nominal controller synthesis procedures that can be based on solving LMIs: (11)
pi Spi Ppi = Q T Spi Rpi

satis es Rpi 0:

Among others, this hypothesis holds true for the following very important special cases: Robust quadratic performance with ! ; I 0 Ppi = 0 1 I guarantees that the L2 -gain of the channel wi ! zi is robustly smaller than .

Robust Mixed and LPV Control with Full Block Scalings

0 ;I guarantees strict robust dissipativity of the channel Ppi = ; I 0 wi ! zi, generalizing positive realness. Based on Lyapunov arguments, one can easily obtain a su cient LMI condition for robust quadratic performance. Theorem 3.3. Suppose the interconnection (3) is well-posed, and suppose there exists an X satisfying (12)

0 B X >0 B B @

1T 0 0 X 0 0 C B C BX 0 0 0 C A B @ 0 0 Qpi Spi

T Rpi 0 0 Spi

1 10 I 0 C B A( ) B i ( ) C C<0 C B C B A A@ 0 I C

Ci(

Dii(

for all 2 . Then the uncertain system (3) is uniformly exponentially stable and satis es the robust quadratic performance speci cation for the channel wi ! zi . Proof. The left-upper block of (12) reads as

A(

!T
)

0 X X 0

A(

!
)

+ C i ( )T Rpi C i ( ) < 0:

At this point we exploit (11) to infer (7) such that we can conclude uniform robust exponential stability. The proof of robust performance is, again, ! straightforward. By continuity 0 0 and compactness of , we can add 0 I on the left-hand side of (12) for some small > 0 without violating the inequality. For any wi 2 L2 , let (:) and zi (:) be the corresponding state- and output trajectory of the uncertain ! ( t ) system (3) with (0) = 0. By right- and left-multiplication with w (t) i and its transpose, we infer

d (t)T X (t) + wi (t) zi (t) dt

!T

Ppi

wi (t) + w (t)T w (t) 0: i i zi (t)

Integrating over 0 T ] and letting T go to in nity implies the desired inequality (10) if we recall (0) = 0 and limT !1 (T ) = 0. On the basis of Lemma A.1, we can obtain with ease the equivalent multiplier test. Theorem 3.4. The interconnection (3) is well-posed and X satis es (12)

8 (13)

Scherer

if and only if there exists a P 2 P such that

0 B B B X >0 B B B B @

1T 0 0 C B X C B C B 0 C B C B C B0 C A B @0

10 I C B 0 0 0 0 A C B C B 0 Q S 0 0 C B C B C B 0 ST R 0 C CB 1 0 0 0 Qpi Spi A @ 0 T Rpi 0 0 0 0 Spi Ci X 0 0 0

0 0 0 0

B1 Bi C C I 0 C C < 0: C D11 D1i C C 0 I A Di1 Dii 1 C C C C C C C A

0 1

Proof. As in the proof of Theorem 3.2, apply Lemma A.1 to

00 B X B B 0 N =B B B 0 B @0

1 0 0 C B 0 0 0 0 B1 C B C B 0 0 0 0 I C S0 = im B C B C B 0 0 0 0 D 11 C B @ 0 0 0 0 Qpi Spi A T Rpi 0 0 0 0 Spi Di1 X 0 0 0

0 0 0 0

(14)

0 I B A B B 0 B S = im B B C1 B @ 0 Ci
U= I ; T=

B1 Bi C C I 0 C C C D11 D1i C C 0 I A Di1 Dii

0 0 I 0 0 0 : 0 0 0 I 0 0

0 1

and

If the disturbance wi that a ects the system is stochastic in nature, we can also consider the H2 criterion as a performance measure for the perturbed system. Let us assume that (0) = 0, that wi is white noise, and that we are interested in bounding the variance of the output zi by a given number . We need to assume that the uncertainty model and the controller are such that (15)

3.3 Robust H2 Performance

8 2 : Dii( ) = 0:

Then it is easy to arrive at the following analysis result.

Robust Mixed and LPV Control with Full Block Scalings

there exist X > 0, Z > 0 satisfying, for all

Theorem 3.5. Suppose the interconnection (3) is well-posed, and suppose

2 ,

(16)

0 I 1T 0 0 X 0 0 1 0 I 1 0 0 B B B X 0 0 0C A( ) B i ( ) C A( ) Bi( ) C C B B C B C<0 C B B C B @ 0 A I A @ 0 0 ; I 0 A@ 0 I C

Ci(

Dii(
I
)

0 0 0

Ci(

Dii(

(17)

Ci(

!T

;X
0

Z ;1

Ci(

!
)

< 0 Tr(Z ) < 1:

Then the uncertain system (3) is uniformly exponentially stable, and the variance of the output zi on the time-interval 0 1) is smaller than . Due to the zero row and column, the inequality in (16) could be simpli ed. We kept it in this form to display the similarity to the robust quadratic performance test. Proof. Uniform exponential stability follows as in the proof of Theorem 3.3. For verifying robust performance, one can easily rewrite (16) with Y = X ;1 to

A( )Y + Y A( Ci( )Y Ci(

)T + 1 Bi ( )Bi ( )T < 0:

Using Lemma A.2, (17) is seen to be equivalent to )T < Z Tr(Z ) < 1:

Let us now recall that the state covariance matrix of the uncertain system is given as the solution of the initial value problem _ = A( (t))K + KA( (t))T + Bi ( (t))Bi ( (t))T K(0) = 0: K

K(t) < Y for t 0 and hence we infer h i h i Tr Ci( (t))K(t)C i( (t))T Tr Ci ( (t))Y C i( (t))T < for all t 0

Standard comparison results for linear matrix di erential equations imply

what leads to the desired bound on the output variance. We end up with two inequalities in the parameter . Therefore, we have to apply Lemma A.1 to each of these inequalities individually. This leads to two independent multipliers to equivalently reformulate the robust H2 analysis condition to the multiplier version.

10

Scherer Theorem 3.6. The interconnection (3) is well-posed and

(16)-(17) if and only if there exist P1 P2 2 P such that

X , Z satisfy
0 1

(18)

0 I B A B B 0 B B B C 1 B @ 0 Ci

B1 D11 Di1
0

0 1T 0 0 X 0 0 0 B X 0 0 0 0 Bi C C B C B 0 C 0 0 Q1 S1 0 B B T R1 0 C B D1i C 0 0 S1 C B I A @ 0 0 0 0 ;I Dii 0 0 0 0 0

0 0 0 0 0 0

10 I C B A C B C B 0 C B C B C B C 1 C AB @ 0 Ci
0

B1 Bi C C I 0 C C<0 C D11 D1i C C 0 I A Di1 Dii

(19)

Instead of the stochastic interpretation, the criterion derived here admits as well a deterministic interpretation it guarantees a bound on the sum of the energy of the output responses to initial conditions taken as the columns of Bi ( (0)).

0 I 0 B 0 I B B @ C1 D11 Ci Di1 Remark.

1T 0 ;X 0 0 C B C B 0 QT2 S2 C A B @ 0 S2 R2
0 0

10 I C B C B 0 C AB @ C1 ; 1 0 Z Ci
0 0 0

1 I C C A < 0 Tr(Z ) < 1: D11 C Di1

3.4 Robust Generalized H2 Performance

kzi k1 < : 0<kw k2 <1 kwi k2
sup
i

The generalized H2 norm is the gain of the system mapping wi 2 L2 into zi 2 L1, i.e., the energy to peak gain 16]. The corresponding performance speci cation is to robustly guarantee the bound (20)

Note that the gain can only be nite if (15) holds what is assumed throughout this section. Theorem 3.7. Suppose the interconnection (3) is well-posed, and that there exists an X > 0 such that, for all 2 , the inequality (16) and (21)

Ci(

!T

;X 0
0

1I

Ci(

<0

Proof. As for quadratic performance, the proof of stability is obvious, and one can infer from (16) that there exists an > 0 such that

zi 2 L1 is smaller than .

hold true. Then (3) is robustly exponentially stable, and the gain of L2 3 wi !

d (t)T X (t) ( ; )w (t)T w (t) i i dt

Robust Mixed and LPV Control with Full Block Scalings

11

forR any wi 2 L2 . Integrating over 0 T ] leads to (T )T X (T ) ( ; T T ) 0 wi (t) wi (t) dt and hence (T )T X (T ) ( ; )kwi k2 2 for all T 0: The second inequality (21) implies C i ( )T C i ( ) < X and therefore, due to Dii( ) = 0, zi (T )T zi (T ) (T )T X (T ) for all T 0: Both relations taken together lead to (20). Similarly as for the H2 performance speci cation, we have to introduce two independent multipliers to equivalently reformulate these conditions to the corresponding multiplier versions. Theorem 3.8. The interconnection (3) is well-posed and X satis es (16) and (21) if and only if there exist multipliers P1 P2 2 P with (18) and 0 I 0 1T 0 ;X 0 0 0 1 0 I 0 1 B B B 0 Q2 S2 0 C 0 I C 0 I C B C B C B C < 0: (22) B C B C B T @ C1 D11 A @ 0 S2 R2 0 A @ C1 D11 C A 0 0 0 1I Ci Di1 Ci Di1 Let us nally include in our list the important time-domain speci cation to keep the peak-to-peak gain L1 3 wi ! zi 2 L1 smaller than . To be precise, we intend to robustly guarantee (23)

3.5 Robust Bound on Peak-to-Peak Gain

kzi k1 < : k 0<kw k1 <1 wi k1
sup
i

Again, simple Lyapunov arguments provide su cient conditions for this inequality to hold. Theorem 3.9. Suppose the interconnection (3) is well-posed. If there exist X , > 0, such that, for all 2 , (24) 0 I 1T 0 X X 0 0 1 0 I 1 0 0 B B B A( ) Bi( ) C X 0 0 0C A( ) B i ( ) C B C B C B C<0 B C B C B @ 0 A I A @ 0 0 ; I 0 A@ 0 I C Ci( ) Dii( ) 0 0 0 0 Ci( ) Dii( ) (25)

I 0 0 ) Dii ( ) Ci( ) Dii( ) then (3) is robustly exponentially stable, and the gain of L1 3 wi ! zi 2 L1 is robustly smaller than .

0 I B @ 0

Ci(

10 1T 0 ; X 0 0 I C B C B A @ 0 ( ; )I 10 A @ 0

1 C A<0

12

Scherer

Proof. As before we observe that (24) implies

d T T T dt (t) X (t) + (t) X (t) ; wi (t) wi (t) 0: R t e; (t; ) kw ( )k2 d and hence We infer (t)T X (t) i 0
(t)T X (t)

kwi k2 0: 1 for all t

i

The inequality (25) leads (due to ; > 0) to 1 kz (t)k2 (t)T X (t) + ( ; )kw (t)k2

for some small > 0. If we combine we infer kzi k2 ( ; ) kw i k 2 1 1 and hence (23). The reformulation into the corresponding multiplier version is, again, straightforward. Theorem 3.10. The interconnection (3) is well-posed and (24)-(25) hold if and only if there exist multipliers P1 P2 2 P that satisfy (26) 0 1T 0 X X 0 0 0 0 1 0 I 0 0 1 B C B B A B1 Bi C X 0 0 0 0 0C B C B C B C B C B C B C 0 0 Q S 0 0 0 I 0 B C B C B C 1 1 B C B C B C<0 T B C B C B 0 0 S R 0 0 C D D 1 1 11 1i C 1 B C B C B @ A @ 0 0 0 0 ; I 0 A@ 0 0 I C A 0 0 0 0 0 0 Ci Di1 Dii (27)

(t)T X (t) + ( ; )kwi k2 1

0 B B B B B @

1T 0 ; X C B 0 C B C B C B 0 C A B @ 0
0

Q2 T S2
0 0

0 S2 0 R2 0 0 ( ; )I 0 1I 0 0

0 0 0

10 I C B 0 C B C B C B C1 C AB @ 0 Ci

D11 D1i C C C < 0: 0 I A Di1 Dii

0 0 C C

Note that X and enter these inequalities non-linearly. In an analysis problem, it is advisable to search for the best upper bound on the peak-topeak gain by xing > 0, and minimizing the bound over the LMIs (26)-(27) (what is a convex optimization problem) to get the optimal value ( ). Then one can perform an additional line-search over to further minimize ( ) and to get to the smallest achievable bound.

4 Dualization

On the basis of Lemma A.2, one can dualize all robust performance tests we have provided in the previous section. Let us concentrate on quadratic

Robust Mixed and LPV Control with Full Block Scalings

13

performance only with an index matrix

Qpi Spi T Rpi Spi

that is non-singular

and whose inverse is denoted as ~ pi S ~pi ! Q T R ~pi ~ pi = S

Qpi Spi T Rpi Spi

!;1

Note that this property can be enforced by a slight perturbation that neither changes the problem formulation nor its solution. For an arbitrary matrix M , we observe that (28) im M I
(k+l) :

!?

I = im ;M T
:

Hence the set of dual scalings has to be de ned as

8 < (k+l) ~ := P ~ P : 2R

~=P ~T 8 2 P

!T
T

~ P ;

!
T

<0

9 =

~2P ~ is partitioned in the same fashion as P . and each P Corollary 4.1. There exists an X and a multiplier P 2 P with (12) if ~2P ~ and only if there is a dual Lyapunov matrix Y > 0 and a dual multiplier P with 0 1T 0 0 Y 0 0 0 0 1 0 ;AT C T C T 1 1 i C C B B B C 0 0 0 0 Y 0 I 0 0 C B B C B C C B B B C T T T ~ ~ C B B C 0 0 0 0 Q S ;B1 ;D11 ;Dj1 C B C C B B C>0 B C T C B B ~ ~ B C 0 0 S R 0 0 CB 0 I 0 C B C B C T ;DT C ~ pi S ~pi C @ A B AB @ ;BiT ;D1 @0 0 0 0 Q j i A T R ~pi ~ pi 0 0 I 0 0 0 0 S
The Lyapunov matrices and multipliers are related as ~ ;1 : X = Y ;1 and P = P Proof. Suppose X and P satisfy (12). Since we conclude that R > 0. Hence we infer

= 0 is in the value set , 0 01 0 0C C 0 0C C 0: C I 0C C 0 0A 0 I

00 B I B B 0 B B B 0 B @0

0 0 1T 0 0 B X 0 0C C B C B 0 0 0C B B C B 0 I 0C C B 0 0A @ 0 0 0 0 I

X 0 0 0

10 0 C B 0 0 0 0 I B C B C 0 Q S 0 0 B C B C C B 0 ST R 0 0 C B @ A 0 0 0 Qpi Spi 0 T Rpi 0 0 0 Spi {z } 0

0 0 0 0
N

14

Scherer
T

maximal dimension. Moreover, since (13) implies DI P DI < 0 11 11 P is non-singular. Hence the same is true of N . Due to (28) (after a row permutation), the proof is nished by applying Lemma A.2. The set of multipliers P is obviously convex. However, since this set is described by in nitely many inequalities, and since it is not obvious, in general, how to reduce to nitely many conditions, it is hard to even decide whether a matrix P is indeed contained in P or not. This precludes the veri cation of the robust performance tests in Section 3 by standard algorithms. This is the motivation to con ne, at the expense of conservatism, the search to a smaller set of multipliers that admits a simple description, preferably in terms of nitely many LMIs. For that purpose we assume that the value set is the convex hull of nitely many prespeci ed matrices: (29) = Cof
1

This implies that S as de ned in (14) is, in fact, a negative ! subspace of ! N of

5 How to Verify the Robust Performance Tests

:::

N g:

Let us then introduce the multiplier set

P1 := fP 2 P : Q < 0g which satis es P1 P :

Note that, in general, this is a strict inclusion such that the restriction of the search to P1 introduces conservatism. However, a straightforward convexity argument (based on Q < 0) reveals that

P 2 P1 if and only if Q < 0

!T

> 0 for all = 1 : : : N:

Hence, the set P1 can indeed be fully described by nitely many linear matrix inequalities, and the search for P 2 P1 renders our robust performance test veri able by solving standard LMI problems. Remark. If D11 = 0 such that the parameters enter (3) a nely, we observe that all the robust performance inequalities already imply that any multiplier in P satis es Q < 0 and is, hence, contained in P1 . Hence, in this case, P and P1 conincide and no conservatism has been introduced. It is possible to reduce the conservatism for block-diagonal uncertainties. For that purpose let us assume that

8 0 I > < B 11 . .. => =@ : 0

0 1

m Im

9 > = C A : i 2 ;1 1]> :

Robust Mixed and LPV Control with Full Block Scalings

15

Then we obtain (29) where the N = 2m generators are de ned by letting each i vary in f;1 1g. Let us now partition the left-upper block Q of the multiplier P as what de nes m diagonal blocks Q1 : : : Qm . If we introduce

P2 := fP 2 P : Q < 0
we infer

= 1 : : : mg satisfying P1 P2 P

P 2 P2 i Q < 0

= 1 ::: m

!T
I

!
I

>0

= 1 ::: N

on the basis of a simple multi-convexity argument. Again, the set P2 is described in terms of nitely many LMIs. As expected, one can demonstrate by simple examples that the set P2 is in general larger than P1 , and that P2 can lead to less conservative robust performance tests. All these techniques apply in the same fashion to the corresponding dual inequalities as provided in Section A.2 for the corresponding sets of dual ~, P ~1 and P ~2 . multipliers P We have listed several important analysis tests for robust performance. In a typical multi-objective robust controller design problem one tries to nd a controller that meets a selection of all these speci cations on various channels of the controlled system. In order to render the underlying analysis problems algorithmically tractable, we constrain ourselves to the multiplier set P1 or, for block-diagonal uncertainties, to P2 respectively. Even for the nominal performance speci cations, the corresponding multiobjective control problems are still hard and mostly open. It is wellknown that the main di culty arises due to the fact that each of the performance speci cation requires, in general, a di erent Lyapunov matrix to render it satis ed. The presently known design techniques do not allow to easily overcome this obstacle. This has been the motivation to consider, instead, the so-called mixed control problems in which the goal is to render the speci cations satis ed with a common Lyapunov function X for all speci cations under investigation. To be speci c, we consider the robust mixed QP/H2 problem: Try to robustly achieve quadratic performance with index Ppi on channel wi ! zi and an H2 bound on channel wj ! zj . Note that the generalization to any other combination of (possibly repeated) performance speci cations on other channels is obvious. The robust mixed QP/H2 control problem hence aims at nding a controller, multipliers P P1 P2 2 P and a Lyapunov matrix X as well as an auxiliary variable Z > 0 that render the LMIs (13) and (18)-(19) with i replaced by j satis ed.

6 Mixed Robust Controller Design

16

Scherer

Quite recently it has been observed 12, 6, 20, 18] how to step in a formal manner from the analysis to the corresponding synthesis inequalities. This procedure is based on transforming the Lyapunov matrix X and the controller parameters as (30)

6.1 Synthesis Inequalities for Output Feedback Control

X Ac Bc Cc Dc ! X Y K L M N =: v

into the new symmetric blocks X , Y and the transformed controller parameters K , L, M , N that are collected in the variable v. Let us now introduce the functions ! Y I X (v ) = I X and

A(v) Bj (v) C i (v) Dij (v) =

0 1 0 1 ! ! AY A Bj 0 B K L I 0 0 B C B C = @ 0 XA XBj A + @ I 0 A M N 0 C Fj CiY Ci Dij 0 Ei

that are a ne in v. In order to transform the synthesis inequalities, one needs to nd a formal congruence transformation involving Z such that the blocks in the analysis inequalities transform as

! Z T XZ ! X (v )

! ! XA XBj ! Z T XA]Z Z T XBj ] : Ci Dij Ci Z Dij

Then it su ces to perform the substitution

Z T XZ

! ! Z T XA]Z Z T XBj ] ! A(v) Bj (v) CiZ Dij C i (v) Dij (v)

to arrive at the synthesis inequalities in the new variables v and all other variables that appear in the analysis test (such as multipliers and auxiliary parameters). Once having solved the synthesis inequalities, the inversion of (30) leads back to the Lyapunov matrix and to the desired controller parameters. This inverse is easily calculated by nding non-singular matrices U , V with I ;XY = UV T and solving the equations

K L M N

Y V I 0

0 + U XB = XAY 0 0 0 I

I 0 X= X U

Ac Bc Cc Dc

VT 0 CY I

Robust Mixed and LPV Control with Full Block Scalings

17

for X and Ac , Bc, Cc , Dc . If performing these two steps for the robust mixed QP/H2 problem, one arrives at the synthesis inequalities

Y I I X

>0

0 B B B B B B B @ 0 B B B B B B B @

1T 0 0 C B I C B C B 0 C B C B C B0 C A B @0

1T 0 0 C B I C B C B 0 C B C B C B0 C A B @0

0 0 0 0 0 0

I 0 0 Q S ST R
0 0 0 0

0 Qpi T Rpi 0 Spi 0 0 0 0 0 0

0 0 0 0

10 I 0 C B A B1 C B C B 0 I C B C B C CB B C 1 D11 Spi A @ 0 0
0 0 0 0

0 0 0 Z ;1 C j Dj 1 where we suppress the variable v for reasons of space. Unfortunately, testing the feasibility of these inequalities does, in general, not amount to solving a convex optimization or LMI problem. Hence one has to resort to controller scalings iteration as they are known from -theory 13]. One might suggest the following procedure: Consider the scaled uncertainty set r and try to maximize r such that the synthesis inequalities are satis ed for the value set r . Start with a nominal design for r = 0. Then the iteration proceeds as follows: In the rst step, x K , L, M , N and nd X , Y and multipliers P , P1 , P2 which correspond to r such that the synthesis inequalities hold and r is maximal. If parametrizing the multipliers by nitely many LMIs, testing the feasibility for xed r corresponds to an analysis problem and hence reduces to a standard LMI feasibility test the maximization of r can be performed by bisection. In the second step, one xes P , P1 , P2 , and maximizes r by varying the whole variable v. For xed r, this corresponds to a nominal performance design problem (such that it reduces to an LMI feasibility test), and the maximization of r is done by bisection. One can immediately advise many variations of this principal procedure. In particular, one might choose other combinations of xed and varying parameters in the optimization steps to increase r 18].

0 I 0 1T 0 ;X 0 0 B B 0 I C B C B 0 QT2 S2 B C @ C 1 D11 A B @ 0 S2 R2

0 0 0 0 0 Q1 S1 0 T 0 S1 R1 0 0 0 0 ; I 0 0 0 0 0 0 0 0 0 0

I 0

10 I 0 C B A B1 C B C B 0 I C B C B C B C 1 D11 C AB @ 0 0

Bi C C 0 C C<0 C D1i C C I A C i Di1 Dii Bj 0 D1j I Djj

0 1

0 1

C j Dj 1

10 I 0 1 C B 0 I C C B C < 0 Tr(Z ) < 1 C B A @ C 1 D11 C A

C j Dj1

C C C C <0 C C C A

18

Scherer

6.2 Synthesis Inequalities for State-Feedback Control

The procedure described in the previous section applies literally to statefeedback design. One just needs to employ the functions ! ! A ( v ) B ( v ) B AY + BM j j X (v) = Y C i(v) Dij (v) = CiY + EiM Dij

and the equations to reconstruct the Lyapunov matrix and the (static) controller read as X = Y ;1 Dc = MY ;1: As such, the synthesis inequalities are not a ne in all variables. However, as demonstrated for the analysis inequalities, one can straightforwardly dualize the synthesis inequalities that we have derived above. Due to the mere fact that B j (v) and D ij (v) do actually not depend on v, the dual inequalities de ne convex constraints even if letting both the multipliers, the auxiliary parameter Z , and the whole parameter v vary in fact, they can be easily rearranged to become a ne in all unknowns. Hence, what is known from single-objective control problems or for block-diagonal multipliers completely generalizes to robust mixed control problems with full block scalings. It is well-known how to eliminate the controller parameters in single-objective nominal design problems. This is also possible in single-objective robust control problems. Let us consider, as an example, the robust quadratic performance problem with index Pp on the channel w2 ! z2 . We will provide a variation of the standard procedure based on the projection lemma that leads, even for a general quadratic performance index, to particularly simple formulas. We just apply Lemma A.3 and observe we can work with basis matrices K1 and K2 of the kernels of T E T and ker C F1 F2 ker B T E1 2 respectively. Then we arrive at the following equivalent synthesis test: Find ~2P ~ that satisfy X , Y and multipliers P 2 P , P (31) (32)

6.3 Elimination of Controller Parameters

Y I I X

>0

0 B B B T K2 B B B B @

1T 0 0 C B I C B C B 0 C B C B C B 0 C A B @0

0 0 0 0 0 0

I 0 0 0 Q ST
0 0 0 0

0 S 0 R 0 0 Qp T Rp 0 Sp

10 I 0 0 1 C B XA XB1 XB2 C C C B C C B 0 I 0 C C B C K2 < 0 C B C B C D D 1 11 12 C C C B Sp A @ 0 0 I A

0 0 0 0

C2 D21 D22

Robust Mixed and LPV Control with Full Block Scalings

19

(33)

0 B B B TB K1 B B B @

1T0 0 B C I B C B C B 0 C B C B C B0 C AB @0

0 0 0 0 0 0

I 0 0 0
0 ~ Q ~T S 0 0 0 ~ S ~ R 0 0

0 0 0 ~p Q T ~p S

0 0 0 0 ~p S ~p R

10 ;Y AT ;Y C1T ;Y C2T C B C I 0 0 B C B T T T C ; B ; D ; D B 1 11 21 C B C B I 0 C B 0 T ;D C T T ; D A@ ;B2 12 22

0 0

1 C C C C K1 > 0 C C C A

and the duality coupling condition (34)

! !;1 ~ S ~ Q Q S : ~T R ~ = ST R S

Non-convexity enters via the generally non-convex constraint (34). Although this result has not appeared in the literature, it is a straightforward (and notationally simple) extension of those in 15, 10] to robust quadratic performance. The main purpose for being so detailed is to point out the relation to LPV control in the next section. In LPV control, it is assumed that the parameters (t) that enter the system are not unknown but that they can be measured on-line. This allows, among other applications, to approach (robust) gain-scheduling synthesis problems for non-linear control systems. So far, the LPV problem has been solved for blockdiagonal parameter matrices with the corresponding block-diagonal scalings 1, 9, 14, 21]. In 19] we have sketched how to extend these techniques to full block scalings, and in this chapter we give for the rst time the full problem solution. Adjusted to the structure of (1), we assume that the measured parameter curve enters the controller also in a linear fractional fashion. Hence an LPV controller is de ned by scheduling the LTI system (35)

7 LPV Design with Full Scalings

x _ c = Ac xc + Bc

y wc

u zc
c(

= Ccxc + Dc

y wc

with the actual parameter curve as

wc =

(t))zc :

The controller is hence parameterized through the matrices Ac, Bc , Cc, Dc , and through a possibly non-linear scheduling function c( ). Note that previous approaches were based on c( ) = such that the controller is scheduled with an identical copy of the parameters. However, full block scalings require the extension to more general scheduling function that will - a posteriori - turn out to be quadratic functions.

20

Scherer

The goal is to construct an LPV controller that renders the quadratic performance speci cation with index Pp for the channel w2 ! z2 for all possible parameter curves satis ed. As standard, the solution of this problem is obtained with a simple trick. In fact, the controlled system can, alternatively, be obtained by scheduling the LTI system 0 x _ 1 0 A B1 0 B2 B 0 1 0 x 1 B B B z1 C C1 D11 0 D12 E1 0 C w1 C C B C B C B C B C B C B z 0 w 0 0 0 0 I C B C B C B c c (36) = C B C B C B C B C B C B z C D 0 D E 0 w 2 2 21 22 2 2 C B C B B A @ y A @ C F1 0 F2 0 0 A @ u C wc 0 0 I 0 0 0 zc with the parameters as ! ! ! w ( t ) 0 z 1 1 (37) wc = 0 zc c ( (t)) and then controlling this parameter dependent system with the LTI controller (35). Hence, with the chosen controller structure, the LPV problem we started out with is equivalently reformulated to a robust performance design problem as discussed previously: Find an LTI controller (35) that renders the system (36)(37) uniformly exponentially stable such that the robust quadratic performance speci cation for w2 ! z2 with non-singular index Pp is satis ed. Due to the fact that the parameters are measured on-line, the uncertainties enter the system in a very speci c form what is re ected in the particular structure of the describing matrices in (36) and of the parameters in (37). This particular structure implies that the synthesis inequalities related to this robust performance problem are standard LMI problems. Hence they can be solved (without conservatism) using existing algorithms. For guaranteeing robust stability and performance of the closed-loop system, we employ extended multipliers adjusted to the extended uncertainty structure that are given as (38) 0 1

! B Q Q12 S S12 C Qe Se = B Q21 Q22 S21 S22 C Pe = S B T Re @ A with Qe < 0 Re > 0 R R12 C e
R21 R22

and that satisfy (39)

0 B 0 B B @I

0 C B 0 B c( ) C P C B e 0 A @ I I 0

1 0

0 C c( ) C A > 0 for all 0 C

2 :

Robust Mixed and LPV Control with Full Block Scalings

21

~e = Pe;1 are partitioned similarly as The corresponding dual multipliers P (40) 0 Q ~ S ~12 1 ~ Q ~ 12 S ! B ~ 21 Q ~ 22 S ~e ~21 S ~22 C ~e S Q B C with Q ~e = Q ~e < 0 R ~ e > 0: = P B T R ~e ~e ~ R ~ 12 C @ A S R ~ 21 R ~ 12 R As indicated by this notation, it will turn out that the LPV synthesis ~e inequalities will be only in uenced by the multipliers blocks in Pe and P without indices, and they will be actually identical to those of the robust control problem apart from the coupling condition (34). Therefore, testing these synthesis conditions indeed amounts to solving a standard LMI problem. Theorem 7.1. There exists a controller (35) and a scheduling function such that the system (36)-(37) controlled with (35) satis es the analysis conditions for robust quadratic performance with multipliers (38)-(39) if and ~2P ~1 that satisfy the linear only if there exist X , Y and scalings P 2 P1 , P matrix inequalities (31)-(33). Proof. The proof of `only if' is straightforward: Eliminate the controller parameters in the analysis inequalities. Due to the speci c structure of the describing matrices in (36), the resulting synthesis inequalities simplify to (31)(33) such that only the multiplier parts ! ! ~ ~ Q S Q S ~ = ~T ~ P = S T R and P S R appear. The constructive proof of `if' is more involved. Let us assume that we have found a solution to (31)-(33). First step: Extension of Scalings. Let us de ne the matrices

Z= I 0

~= and Z

~ ) is the orthogonal complement of with the row partition of P . Note that im(Z im(Z ), and recall ~T P Z ~ > 0 as well as Z T PZ ~ <0 Z ~T P ~Z ~ > 0: (41) Z T PZ < 0 Z ~ , we try to nd an extension Pe with (38) such that For the given P and P ~ ~ as in (40). After a the dual multiplier Pe = Pe;1 is related to the given P suitable permutation, this amounts to nding an extension ! ! !;1 ~ P T P T P (42) = T T T T NT T T T T NT with where the speci c parametrization of the new blocks in terms of a non-singular matrix T and some symmetric N will turn out convenient. Such an extension is

22

Scherer

very simple to obtain. However, we also need to obey the positivity/negativity constraints in (38) that amount to (43) and (44) ~ 0 Z ~ 0 Z

Z 0 0 Z

!T

P T T T T T NT P T T T T T NT

Z 0 0 Z

<0

!T

! ~ ! Z 0 > 0: ~ 0 Z

~ ;1 is non-singular. We can assume w.l.o.g. (perturb, if necessary) that P ; P Then we set ~ ;1 );1 N = (P ; P and observe that (42) holds for any non-singular T . The main goal is to adjust T to render (43)-(44) satis ed. We will in fact ~ of T = (T1 T2 ). Due to (41), construct the subblocks T1 = TZ and T2 = T Z the conditions (43)-(44) read in terms of these blocks as (45) h T ~ (Z ~T P Z ~ );1 Z ~ T T2 > 0: T1 N ; Z (Z T PZ );1 Z T T1 < 0 and T2T N ; Z

If we denote by n+(S ), n;(S ) the number of positive, negative eigenvalues of the symmetric matrix S , we hence have to calculate n; (N ; Z (Z T PZ );1 Z T ) ~ (Z ~T P Z ~ );1 Z ~ T ). Simple Schur complement arguments reveal that and n+ (N ; Z

n;
equals

Z T PZ Z T Z N

~ ;1 Z ) + n; (N ): n;(N ; Z (Z T PZ );1 Z T ) + n;(Z T PZ ) and n;(Z T P ~T P Z ~ > 0, we infer Z T P ~ ;1 Z < 0, and hence By Lemma A.2 and Z T T ; 1 ~ n;(Z PZ ) = n;(Z P Z ). This leads to ~ (Z ~T P Z ~ );1 Z ~ T ) = n+ (N ): n; (N ; Z (Z T PZ );1Z T ) = n;(N ) and n+(N ; Z This implies that there exist T1 , T2 with n; (N ), n+(N ) columns that satisfy (45). Since n+(N ) + n;(N ) equals the number of rows of T , these two blocks indeed de ne a square T that can be assumed (after perturbation, if necessary) to be non-singular. This nishes the construction of the extended multiplier (38) where we observe that the dimensions of Q22 /R22 equal the number of columns of T1 /T2 which are, in turn, identical to n; (N )/n+ (N ).

Robust Mixed and LPV Control with Full Block Scalings

23

Second Step: Construction of the scheduling function. Let us recall that !T ! !T ! I I ~ I P I > 0 and ; T P ; T < 0 on . By Lemma A.3, we can hence infer that, for each 2 , there indeed

exists a c( ) that satis es (39). Due to the structural simplicity, we can even provide an explicit formula which shows that c( ) can be selected to depend smoothly on . Indeed, by a straightforward Schur-complement argument, (39) is equivalent to

0 U 11 B U B 21 B @ W11 +
W21

W22 + c( )

U12 U22 W12

(W11 + )T
T W12 V11 V21

1 (W22 + c( ))T C C C A>0 V12

T W21

V22

T Q;1 Se > 0, V = ;Q;1 > 0, W = Q;1 Se . Since there exists for U = Re ; Se e e e a solution, the inequality can be rearranged to

W22 + c( ) V22 ; ! T U W U11 21 12 ; W21 V21 W11 +

U22

!;1
V11

T U12 W21 W12 V12

>0

and it is clear that one solution is obtained by rendering the (2 1)-block to vanish:
c(

) = ;W22 + W21 V21

W11 +

U11

!;1

V11

U12 W12 :

Third Step: LTI controller construction. After having reconstructed the scalings, the design of the LTI part of the controller amounts to solving a nominal quadratic performance problem what can be done along standard lines, as e.g. shown in 20]. Remark. The proof reveals that the scheduling function c( ) has a ~ ;1 (if many rows/colums as there are negative/positive eigenvalues of P ; P ~ ;1 assuming w.l.o.g. that the latter is non-singular.) If it happens that P ; P is posititve or negative de nite, there is no need to schedule the controller at all then we obtain a controller that solves the robust quadratic performance problem. ~2P ~1 to The search for X and Y and for the multipliers P 2 P1 and P satisfy (31)-(33) amounts to testing the feasibility of a standard LMI. Moreover, the controller construction in the proof of Theorem 7.1 is constructive. Hence we conclude that we have found a full solution to the quadratic performance LPV control problem (including L2 -gain and dissipativity speci cations) for

Note that c( ) has dimension n;(N ) n+ (N ).

24

Scherer

full block scalings Pe that satisfy Qe < 0. The more interesting general case without this still restrictive negativity hypotheses is dealt with in a forthcoming paper. In this chapter we have discussed how to handle multi-objective robust performance analysis and synthesis problems for systems that are a ected by time-varying parametric uncertainties. We introduced a general technique, the so-called full block S-procedure, that allows to formally introduce full block multipliers in order to reduce the conservatism that could result from the restriction to standard block-diagonal multipliers. Finally, we provided a solution to the single-objective LPV control problem if employing a subclass of full block scalings the fully general case is left to a forthcoming paper.

8 Conclusions

Suppose S is a subspace of Rn , T 2 Rl n is a full row rank matrix, and U Rk l is a compact set of matrices of full row rank. De ne the family of subspaces

A Auxiliary Results on Quadratic Forms A.1 A Full Block S-Procedure

implicit negativity condition

SU := S \ ker(UT ) = fx 2 S : UTx = 0g = fx 2 S : Tx 2 ker(U )g indexed by U 2 U . Suppose N 2 Rn n is a xed symmetric matrix. The goal is to render the 8U 2 U : N < 0 on SU

explicit. We want to relate this property, under certain technical hypotheses, to the existence of a symmetric multiplier P that satis es As a technical hypothesis, we require that all subspaces SU are complementary to a xed subspace S0 S that has the two properties dim(S0 ) k and N 0 on S0 : In the intended applications, S is an unperturbed system, T picks the interconnection variables that are constrained by the uncertainties, the elements of U 2 U de ne kernel representations of the uncertainties, and SU is the uncertain system. The complementarity condition amounts to a wellposedness property of the uncertain system description, and the non-negativity of N is a condition on the performance index of interest. Lemma A.1. The two conditions (46)

N + T T PT < 0 on S and 8U 2 U : P > 0 on ker(U ):

8U 2 U : SU \ S0 = f0g and N < 0 on SU

Robust Mixed and LPV Control with Full Block Scalings

25

hold i there exists a matrix P that satis es

(47)

8U 2 U : N + T T PT < 0 on S and P > 0 on ker(U ):

The dualization of robust performance tests is most easily achieved with the following well-known auxiliary result that is the abstract version of a dualization argument in 10] which is based on manipulating block matrices. Lemma A.2. Suppose that N is symmetric and non-singular, and that S is a negative subspace of N of maximal dimension. (N is negative de nite on S , and the number of negative eigenvalues of N coincides with the dimension of S .) Then N ;1 is positive de nite on S ?:

A.2 Dualization

A.3 Solvabiltiy Test for a Quadratic Inequality

The following explicit solvability characterization for a quadratic matrix inequality serves to conveniently eliminate controller parameters in synthesis tests. Consider the quadratic inequality

(48)

I AT XB + C

!T

Q S ST R

I AT XB + C

<0

in the unstructured unknown X . We assume that

R 0 and that

Q S ST R Q S ST R

is non-singular.

By Lemma A.2, we can dualize to equivalently reformulate the inequality as

;B T X T A ; C T
I

!T

!;1

;B T X T A ; C T
I

> 0:

The solvability test makes use of matrices A? and B? whose columns form a basis of the kernels of A and B respectively. Lemma A.3. The quadratic inequality (48) has a solution X if and only if

B? CB?
and

!T

Q S ST R Q S ST R

B? CB?

<0

;C T A?
A?

!T

!;1

! ;C T A? > 0: A?

26

Scherer

The proof of `only if' is obvious. The proof of the converse can be based on the projection lemma and is constructive if a solution is known to exist, one can explicitly calculate it.

References

1] P. Apkarian and P. Gahinet, A convex characterization of gain-scheduled H1 controllers, IEEE Trans. Automat. Control, 40 (1995), pp. 853-864. 2] P. Apkarian, P. Gahinet, and G. Becker, Self-scheduled H1 control of linear parameter-varying systems, Proc. Amer. Contr. Conf., (1994), pp. 856-860. 3] G. Becker, A. Packard, D. Philbrick, and G. Balas, Control of parametricallydependent linear systems: A single quadratic Lyapunov approach, Proc. Amer. Contr. Conf., San Francisco, CA, (1993), pp. 2795-2799. 4] G. Becker, A. Packard, Robust performance of linear parametrically varying systems using parametrically dependent linear feedback, Systems Control Lett., 23 (1994), pp. 205-215. 5] S.P. Boyd, L. El Ghaoui, E. Feron, and V. Balakrishan, Linear Matrix Inequalities in Systems and Control Theory. SIAM Studies in Applied Mathematics 15, SIAM, Philadelphia, 1994. 6] M. Chilali, P. Gahinet, H1 Design with Pole Placement Constraints: an LMI Approach, IEEE Trans. Automat. Control, 41 (1996), pp. 358-367. 7] P. Gahinet, A. Nemirovskii, A.J. Laub, and M. Chilali, The LMI control toolbox, Proc. 33rd IEEE Conf. Decision Contr., (1994), pp. 2038-2041. 8] L. El Ghaoui and J.P. Folcher, Multiobjective robust control of LTI systems subject to unstructured perturbations, Systems Control Lett., 28 (1996), pp. 23-30. 9] A. Helmersson, Methods for Robust Gain-Scheduling, Ph.D. Thesis, Linkoping University, Sweden, 1995. 10] T. Iwasaki and S. Hara, Well-posedness of feedback systems: insights into exact robustness analysis and approximate computations, IEEE Trans. Aut. Control, 43 (1998), pp. 619-630. 11] I.E. Kose, F. Jabbari, W.E. Schmittendorf, A direct characterization of L2-gain controllers for LPV systems, Proc. IEEE Conf. Decision Contr., (1996), pp. 39903995. 12] I. Masubuchi, A. Ohara, N. Suda, LMI-based controller synthesis: a uni ed formulation and solution, Proc. American Contr. Conf., (1995), pp. 3473-3477. 13] I. Masubuchi, A. Ohara, N. Suda, Robust multi-objective controller design via convex optimization, Proc. IEEE Conf. Decision Contr., (1996), pp. 263-264. 14] A. Packard, Gain-scheduling via linear fractional transformations, Systems Control Lett., 22 (1994), pp. 79-92. 15] A. Rantzer and A. Megretski, System analysis via integral quadratic constraints, Proc. IEEE Conf. Decision Contr., (1994), pp. 3062-3067. 16] M.A. Rotea, Generalized H2 control, Automatica, 29 (1993), pp. 373-385. 17] C.W. Scherer, Mixed H2 =H1 Control, in: A. Isidori, Ed., Trends in Control, A European Perspective, Springer-Verlag, Berlin 1995, pp. 173-216. 18] C.W. Scherer, From LMI Analysis to Multichannel Mixed LMI Synthesis: A General Procedure, Selected Topics in Identi cation, Modelling and Control, 8 (1995), pp. 1-8. 19] C.W. Scherer, Robust generalized H2 control for uncertain and LPV systems with general scalings, Proc. IEEE Conf. Decision Contr., (1996), pp. 3970-3975. 20] C.W. Scherer, P. Gahinet, M. Chilali Multi-Objective Output-Feedback Control via LMI Optimization, IEEE Trans. Automat. Control 42 (1997), pp. 896-911.

Robust Mixed and LPV Control with Full Block Scalings

27

21] G. Scorletti, L. El Ghaoui, Improved linear matrix inequality conditions for gain scheduling, Proc. IEEE Conf. Decision Contr., (1995), pp. 3626-3631. 22] H. Tokunaga, T. Iwasaki, S. Hara, Multi-objective robust control with transient speci cations, Proc. IEEE Conf. Decision Contr., (1996), pp. 3482-3483.