Control Eng. Practice, Vol. 4, No. 7 pp.

1001-1007, 1996
Copyright © 1996 Elsevier Science Ltd
Pergamon Printed in Great Britain. All rights reserved
0967-0661/96 $15.00 + 0.00
PII:S0967-0661(96)00099-8

Hoo C O N T R O L OF A SATELLITE AXIS: LOOP-SHAPING,

CONTROLLER REDUCTION AND ~t-ANALYSIS

s. Le Ballois* and G. Duc**

*IUP G~nie Electrique, Rue d'Eragny, 95000 Neuville-sur-Oise, France
**SupJlec-Service Automatique, Plateau de Moulon, 91192 Gif-sur-Yvette, France (gilles.duc@supelec.fr)

(Received October 1995; in final form April 1996)

Abstract. This paper considers the control of one axis of the satellite SPOT4, which ex-
hibits lightly damped modes with large uncertainties in their natural frequencies. The pro-
posed solution involves Hoo synthesis with loop-shaping, modal reduction and I.t-analysis of
robustness. In both the eases considered (where the rate deviation is either measured, or
not), low-order controllers are obtained, which give the required performance while al-
lowing large parameter deviations.

Key Words. H~ Control, Satellite, Loop-Shaping, Reduction, ~t-Analysis

1. INTRODUCTION paper is to present the results and the knowledge

obtained in such a way.
The control of space structures is a challenging
problem: such processes generally present lightly
damped modes with large uncertainties, while per-
formanee requirements become higher. For that 2. PROCESS AND DESIGN OBJECTIVES
reason, the French Space Agency ("Centre National
d'l~tudes Spatiales") has been involved since 1990 in The problem considered here represents the control
a program to assess the potential benefits of new of one axis of the satellite SPOT4, for a given fixed
guidance, navigation and control (GNC) concepts. position of the solar array. The three dominant flex-
ible modes are retained in the model, with 5.10 -4
damping ratios and uncertainties up to +30% in the
As a part of this program, a collaboration has been natural frequencies:
developed with the Automatic Control Department
of Sup61ee ("l~cole Sup~tieure d'l~leetricit6"). The
focus on the attitude control of the Earth observation (1)
satellite SPOT4 was retained becauseit includes ~ ? = - M ( I + A)2rl - N ( I + A ) i I - q C ~
interesting genetic problems: many flexible modes
due to a large solar array, with badly known
flexibility and low damping ratios, a considerable with:
number of disturbances, and a high level of preci-
sion required. Cx : applied torque (Nm) ;
0 x : angular position (rd) ;
r/: state vector of flexible modes ;
In order to handle the difficulties step by step, a
SISO problem was first defined. The aim of this and:

1001
002 S. Le Ballois and G. Duc

pT=(_9.2673e_3 4.6895e-3 1.1338e-3) ; - angular rate deviation Ox less than + 5.10 -5 rd / s

qT =(-1.1905e-2 8.4060e-3 1.6255e-3); - control torque Cc less than +0.2Nm

-settling time: from T+I 1, 0 x and Ox have to be
( 7.4267e-1-1.9034e-1-7.3047e-1"~
M = | - 2 . 0 4 7 2 e - 2 9.3464e + 0 1.5336e+ 0| ; less than 10% of the above maximal deviations.
k.- 1.4908e- 2 2.9100e- 1 2.8964e + lJ
Two case studies have been considered:
( 9.6885e-4 -8.1434e-5 -1.3613e-4~
N=|-2.6707e-5 3.9987e-3 2.8581e-4| ; - case 1: only the angle 0 x is measured
\-1.9448e- 5 12450e- 4 5.3979e- 3J
- case 2: both Ox and Ox are measured.
fl= 4.5216e-4 ;
I the 3x3 Identity matrix;
3. DESIGN OF THE CONTROL LAW
A a 3x3 diagonal matrix with --03 _<Ati _<03.
3.1 Loop-Shaping H,~ Design
k 0.165 s time delay is added to take into account
he discretisation of the control law, together with The method of McFarlane and Glover (1990, 1992)
ensor and actuator delays. Fig. 1 shows the Bode has been chosen as the synthesis procedure. The
liagram of different models, for deviations up to reasons for this choice are as follows :
:30% in the natural frequencies.
- Hoo methods have proved to be powerful for very
lightly damped plants, see e.g. (Safonov, et al.,
7he control variable is a torque C c generated by a
1991; Zhou, et al., 1992).
eaction wheel. The main disturbance Ca is a torque
tpplied at the plant input (i.e. Cx = Cc + Ca), which - The method of McFarlane and Glover requires one
}ccurs when the video recorder starts (see Fig. 2). first to shape the plant model P0, using compen-
sators W1, Wz (see Fig. 3.a). This task is easily
me design objectives are as follows: the same con- done by classical compensators (lead, lag, PI .... ),
roller has to ensure, for the largest possible devia- and following the usual rules of automatic control.
ions of the natural frequencies:
- The following Hooproblem is then considered:
closed-loop stability
angle deviation 8x less than +12.10 -5 rd inf (q(I-PaK)-l(Pa I) oo (2)
I)
K s t a b i l i s i n g ~,

tO o Log Magnitude
i I I I I III I I I I I I III I where the criterion includes the four main transfers
I i I I III I I I I [ I III I
of the closed-loop plant, from the input and output
I I I I I Clll I
disturbances to the outputs of the controller and
LO"4 . . . . . . . . . . _ _ 2 _ the plant respectively (see Fig. 3.b).
- The optimal value Yminof the Hoo criterion (2) can
10 "1 10 0 101
Frequency (radi~/see) be computed in advance: no y-iteration is requir-
Ph~e (~g~s)
ed. Furthermore, the value obtained for Yminis a
I I I I I I I I1[I I
300
- -£ - 7 -,-, - -~ -~-7~5 7~- - - S - robustness and performance indicator (one has
I I I [ [ I I il I
~00 -- --I-- -- i- --I--I -- -- A LI-- -- - J. -- usually to try to obtain 7minbetween 2 and 3).
; I i i i , ,, I III I
[ I ~ I I I Ill I I Ill i
100 --I P --I --F--I-I + I-I - - -- P -- + -I- P I--I ----~-
i I I t I I Ill I I t I I IIII
0 I. I. I .I I I . ill . I I i I I [Ill I

10 "1 i00 10 I
F r e q u e n c y (mdimkq/se¢)

?ig. 1. Bode plots, for -0.3 <Aii <0.3 a) loop-shaping

e1 e2

0.5--
i Calm)
0.1 time (s)
,,< I 1'2>
I )
T T+0.1 T+5 b) H~ design

Fig. 2. Torque disturbance Fig. 3. Loop-shaping Hoo design

• iii
 Ho~Control of a Satellite Axis 1003

- The actual controller is WlKW2, where the com- 3.3 g-analysis

putation of K involves two classical LQG-type
Riccati equations; they are known to be better g-analysis (Doyle, 1982, Fan, et al., 1991) is then
conditioned than, for example, the ones involved used to check the robustness: the closed-loop system
by the g-iteration approach (Doyle, et aL, 1989), is put in the standard form of Fig. 4, where all un-
especially for lightly damped plants. Numerical certainties are incorporated in A(s). For each
difficulties are therefore avoided. s o ~ C , a general form isA(s0)cA__, where A__ is the
set defined as:
3.2 Controller Reduction
A={diag{A~'...AC,8~Icl...SClc~,tSrlI~...srIr~}},
The controller K obtained in this way is of the same
order as P~, which can therefore be high. It is thus 8; 87 (6)
desirable to apply a reduction procedure.
which includes C full complex blocks, c complex
Instead of using the popular balanced truncation repeated scalars and r real repeated scalars. Roughly
method (Anderson and Liu, 1989), a modal method speaking, the A,c. 's correspond to unmodeled dy-
is chosen (Siret, et al., 1979), which allows one to namics, the 6 r ' s to parameter variations, whereas
select the most important poles of the controller, the 6 c,
i s will be used in this paper to specify some
which will be exactly retained in the reduced-order stability margin (see Section 4).
one.
Define the structured singular value as:
Assuming that the system to be reduced has distinct
eigenvalues 21... Am, 2m+1... An, where 21... 2 m are
A t ( M ) = ( ~ f {~(A); I - MA singular}) -1 (7)
the eigenvalues to be retained, it can be written:

2 = Ax+ Be=(Ao 1
0~ (BI~ and let ~ = sup At(M(jo)). The closed-loop system
A2JX+(82) (3) well
y = Cx + De = (C 1 C2)x+De remains stable for all A(s) such that A(s0)cA__ for
each So s C , and such that Ila(s)L < 1 / ~
with Al:diag{~l.../~m} , A2:diag{~ra+l...~n}.
Criteria are given by Siret, et aL(1979) to choose
conveniently the 2i's to be retained. 4. DESIGN 1: CONTROL FROM 0 x

The reduced-order system is then: The first step is to choose a precompensator W~ to

be applied to the nominal model P0 of the plant. To
{~ = Al z + l~ e this end, a lead filter was designed to obtain a
-20 dB/decade roll-off around the crossover fre-
= H x + Ee (4)
quency, whose value has been chosen centred be-
tween the first and second flexible mode:
where matrices E and H are selected to obtain the
same static gain for systems (3) and (4), and to CO0dB = 1.5 rd/s
optimise the integral of the square error between
their step responses: ~ ( s ) =3747
1+sx/6/1.5 (8)
1+ s /1546
H=CXLr(LXLr) 1
The resulting value for Yminis 2.19. The Hoo con-
E = D _ C A _ I B + H A I _ I B 1 with : (5)
troller K is then computed for the shaped plant
L=(1 m O) ; A X + X A T + A - I B B r A - T = o . P, = P0 ~ (for this step, the time delay is replaced
by a lst-order Pad6 approximant). The controller
order is 10, which is equal to the order of Pa. The
final controller obtained for the plant is W1K.

At this stage, the fact has to be emphasised that due

to the presence of lightly damped poles and zeros,
the controller has to be reduced with special care, to
obtain satisfactory results. Consider first the Bode
Fig. 4. g-analysis plot of controller Wl K (see Fig. 5). It clearly shows
1004 S. Le Ballois and G. Duc
Log Magnitude 5o M a g n i t u d e ( d B )

40 - - , - - , - - - , - - - , - - - , - - - , -
I I I I q
30 - -- ] - -~ -- - -I - - - I-- -- -- I-- - -- i- 7 - - S --
20 __ __ A _ _ A I ~ j __ _ _ I __ __ L __
,:-~,,-,, __~,_'~
_ _

, , , ,.~.~..~_,~

10.2 10" | 100 101 102 103 I P t I I

P h u e (degrees) Frequency (radians/see)
200 r i [ iiii11 i i T IIlllt 1 111HII T ~ TIIITI i 11 TTI[[] I 40 -~---~---~ ....
I ~ t ~ at~rredu~don ~lmi
Illllll I IIIIIII I [ IIIIIlii I
~ I Itlllll I I I IIII I I IIIIII I I [11111 I I I I I I H ~
-5o _ _ 2 _ '_ J _ _ 4 i °
I I I~llJll I I 11 I[[11 I 11111111 I I IIIIIt
0 ~ I $11111
1 II I II Ii iil~f ll~dtlclloll i i i I -8(]0 -700 -4500 -500 -400 -300 -200 -100 0 IOO
ITIITI- 1 ~ IYl Phase (degrees)

II I ~ @ ~ i I I. . 11111111
. l ~ l J l lI l t l
II I
Fig. 6. Design 1: Nichols plot (reduced controller)-
. • . . . . . . ., . . . . . . . , . .

20O]o-2 ]o-~ ~o°

10
1
|0
2
10
3

Frequency(radians/sec) extraction of matrix Ar can be obtained using the

method of Morton (1985). Matrix H(s) (see Fig. 7)
Fig. 5. Design 1: Bode plots of controller Wl K is then obtained by adding the Pad6 approximant
representing the time-delay. In Fig. 7, G(s) is the
that it includes three pairs of lightly damped poles, model of the plant perturbed by real uncertainty Ar .
each of them close to a pair of lightly damped zeros,
whose natural frequencies are very close to those of When all the uncertainties are real,/t is known to be
the nominal model. This phenomenon is a common
not necessarily a continuous function in co ; such a
feature of central Hoo controllers, even if exact can-
case is therefore to be avoided. A complex uncer-
cellations do not occur (which is the case with the
applied method).
tainty, 8c, is thus introduced, with s c ~ R as scaling
factor (see Fig. 7): uncertainty so8c can be viewed as
In terms of robustness, this tendency can be very a gain/phase perturbation on the control applied to
damaging. In fact, it can be verified that +10% the plant. Furthermore, the transfer function from vc
deviations of each of the flexible mode frequencies to z~ is Sc(1-WIKG)-I: if e.g. p ( M ( j c o ) ) < l , the
yield closed-loop instability. The reason is that such
H~ norm (I-WIKG) -1 ~ is guaranteed to be less
deviations modify the alternation of poles and zeros
of the controller and the plant, significantly altering than 1/s c , for all perturbed plants; or equivalently
the open-loop frequency response. the modulus margin (i.e. the distance between the
Nyquist plot and critical point -1, see Fig. 8) is
This lack of robustness can be easily rectified by greater than s~ for all perturbed plants.
applying the reduction procedure presented in Sec-
tion 3 to the H~o controller K. In fact, each combi-
natioxi of lightly damped poles and zeros can be
removed without significantly altering the frequency
response of the controller (see Fig. 5). The corre-
sponding Nichols chart is given in Fig. 6. The re-
suiting controller order is then reduced to 4.

To perform Ix-analysis, scaling factors on each un-

certainty are fn'st introduced in (1):

S=diag{sl,s2,s3} Si ~R
A=SA, "%=diag{S,,S2,83}8, eR (9) Fig. 7. Ix-analysis: real and complex uncertainties

such that if e.g. fii ~ [-1 +1], the frequency of mode

i can take any value between co,O-s/) and co,(l+si),
where coi is the nominal value.

By linearising equations (1) using a first-order ap-

proximation:
Ill/
{ 'Ox - p r rI+ PC~,
=

~I=_M(I + 2SAr)rI_ N ( I +SAr)il_qCx ,(10) Fig. 8. Guaranteed modulus margin

 H,. Control of a Satellite Axis 1005

Evaluation o f g ( M ( j c o ) ) has been performed using given in Fig. 11: prior to the loop-shaping realised
s I = 0.3, s 2 = s 3 = 0.2, s c = 0.1. Fig. 9 gives the by precompensator Wl, a rate feedback is introduced
upper bound o f It obtained frequency by frequency by the gain K i. Such a collocated feedback is a well-
(Balas, et al., 1993). According to the maximum of known technique to control flexible structures
(Safonov, et aL, 1991; Chiang, et aL, 1993). Choos-
this plot, which is 0.92, closed-loop stability is
ing K i = 2250 gives a crossover between modes 2
guaranteed with a modulus margin at least equal to
and 3 for the inner loop.
0.1/0.92 = 0.109, for frequency deviations up to:
- _+0.3/ 0.92 = +_32.6% for the mode 1 This structure allows one to remove the lead filter
previously introduced in Wl: the precompensator
- _+02,/ 0.92 = +_21.7% for modes 2 and 3.
includes only two gains Kp = 2000, K, = 1, corre-
sponding to the position and rate feedback respec-
Note that without reduction of the controller, the
tively (with K r << K i to modify the inner loop
maximum of It goes to 3.8, which allows only
slightly). The resulting value for %minis only 1.74. A
+7.9%, +5.3% and +_5.3% frequency deviations.
9th-order Hoo controller K is then computed for the
nominal model P0 with inner loop gain K i and
Finally, Fig. 10 gives the responses of 0 x , bx and
precompensator WI =[ K p K , ] .
C c to the disturbance defined by Fig. 2, for devia-
tions o f - 3 0 % , 0% and +30% on the frequency of
each flexible mode. From these figures, it can be As for Design 1, this controller is reduced by re-
seen that the design objectives are almost satisfied, moving all sets of closely related badly damped
although the angular rate overtakes the allowed poles and zeros. A non-dominant real pole also
deviation somewhat, after T + 11. having been removed, the reduced controller K has
order 2. The quality of the reduction procedure is
underlined in Fig. 12, which gives the Bode plots of
5. DESIGN 2: CONTROL FROM 0 x A N D b x the SISO transfer function from b x to C c , i.e.
T(s)=WlK(s)[1/s 1 ] r - K i , before and after re-
Sensitivity to parameter variations can be reduced
duction. Clearly this frequency response is not al-
by using an inner loop. The control structure is
tered by the reduction procedure.

r l+rl+ i 11111 lit

The corresponding Nichols chart is given in Fig. 13.
t IIIII
I lllll
I lllIl
I llltl
III
111
It can be noted that the region of high closed-loop
0.8
I IIIII I IIII+ III gain is avoided better than with Design 1.
. . . . ~ Y I-I F] -- -- -- 7 I'7 ~ . . . . . T 17
I IIIII I 11111 III
I t IIII llill III

I,,,, ,+
I I IIII IIIII Ill
0.6 . . . . 7 7 - 1 7 f7 - - - 1717 . . . . . 7"17
I Itlll I lllll III
~x
I IIIII 1111 III

0.4 I ~ T I B N - - ~ - ]1~1--1 R . . . . . TI-I

>
I I llIII i I IIIII Ill
i I II411 I I I Ill Ill
I I IIIII I I II II Ill
0.2
. . . . . . TI lIlllll] . . . . . q] T Ii - I ~ . . . . . . . ]1-1111 '*
I IIII ~ I IIII III
1 IIII I I IIII III
0 , - . . . . . . . . . . . . . . i , . . . . . .

10"1 10 o 101 iO 2
F ~ l u a ' l ~ J (r~li~,~/f~)

Fig. 9. Design 1: It plot for 30%, 20%, 20% Fig. 11. Control structure for Design 2

14 x 10 "5 angular position ~(rd) 5 x 10 "5 a n g u l a r rate Ox(rd/s) control torque Ce ( N m )

0.04
0.02
0
-0.02
-0.04
-0.06
-0.08
-0.1
-0.12
- 3 , , , , , •
-0.14
-2 -4
30 0 5 10 15 20 25 30 o ~ io 1'5 io 15 30
time(see) time(see) time(sec)

Fig. 10. Design 1: responses to torque disturbance

1006 S. Le Ballois and G. Duc
L o g IVlilPlitude I t r i i T I~wll f i i I lllll i i i i i,~l
5 , I I I I IIIII I I I I Illil i I I I IIII
E --I - T 4 ~ B l ~ l ~ _~1--I~ ~ l ~ i ~ ~ ~ ~ I_~1~'t-- ~ ~ ~ l ~ l ~ l ~ "~" ~l ~1~11~ I L I I I IIIII I 1 I I I IIII I I I IIIII
[ - I - I 7 ~ i 7 1I~T I I - - I - I -f 17 I1GI-- G 7 T i-Plrll - q ~ T I T I rfl - 7 G T I T I FIt I I I I I IIII I I I I I IIII I I I I I III
:,:',: :,:: a:,: :
4i I I I IIIII~I I I lllill~/'l~l'Oiilll~0111 I IIIIIII I I l lllllJ I I I I 1fill I I I IIIII I I I I IIII
I I I I Illll I t IIIII I I I I / Iit
~o ~ ~ _~,~,___-,~_~~-'F- -3 ~ :"~,'~,- ~ ~ ~ , ~ , ~ - ~ ~ ~,=,~ I I I I IIIII I I IIIII I I I I I
---I--i3 Bill-- Ei_ - - I-- 3 --7 ~ i ~ : l ~ l - - -S ~ ~-i'-7/-I~1 ~. _~ 3 El~..t
--I -'1__. 0 I ~ i - - ~1 _--I 7 _~,_-- _7 ~ T ~ ! R ' I - - 7 7 T ITI-- 7 7 r l T i
06 --- ~--- ~-~ ,,-~,---~- H,---~- + ~+, ~
_--I _ 1 J i l l t i l _ _1 - I - i i l i i l _ J .A 2. i i l ~ l ~ ~. J ~- Ili HI ~ . . . . I I IIIII I IIII I I I I
I IIIII I III I I I i iii
[ Itl I I I I III
lO _-.'" . . . . . . . . -1 ........ O ........ I ......... 2' ...... 3
10 1O I0 l0 10 10
I I I I IIIII I I I I I Iiii I I I i
Phase ( d q l r e m ) Frequency (mli~s/sec)
I I I I IIIII I i I I I Illl I I I I III
0 ~ , ]~,,. r ~ ~,,. ~ ~ ~ H . , ~ ~.a..~ 1
0.2 . _ _ _1 _ .L J I. £ 1 J Ll ~ _ J _ £ J ~ ± Ig Lj _ • ±JjLLI

l
I I I[11111 I I III I
I I I I IIIII I I I I IIIII I I ~ I III
I I IIIIIII I I II
I I I I IIIII I I I I LLIII I I I I II]
I I IIIIIII I I
I I I I IIIII I I I I IIIII I J I I III
-50 -I-i7 nFnl- - 0 I I I 111111 1 1 I IIIILl 1 l i Ill I
I I iiiiiii 10 "1 10 o 10 1 102
I I Illllli Frequency (r~li~ns/se~)

" -2 -I 0 1 2 3
10 10 1O l0 l0 10 Fig. 14, Design 2: p plot
Frequency (rlidienit~ec)

Fig. 12. Design 2: Bode plots of transfer CclO~ Cc to the torque disturbance defmed by Fig. 2, for
the same deviations (-30%, 0%, +30%) as in Fig.
10. It can be seen that this second design signifi-
6o ~ (dB)
I I I I I I I I
can@ reduces the position and rate deviations,
SO - - T - - T - - T - - T - - T - - T
I I I I I I
. . . . . V-
I
without increasing the control level.
4o --T--T--T--T--T--T-- -?-
30 _ _ Z _ _ L _ _ J _ _ _ £ _ _ L _ _ L _ L _
I I I I I I I

6. C O N C L U S I O N
io
__
I
+ _ _
,
1/3d B
+_.-_.+,
K
~,/ oJlt? 7, . . . . .
I
i
~_,_t~71i__'~_._J
I /
, I ~,:'B'~r~.
/~\
I /
i I ,,7
I /

I I ~ ~ /I I I ~1 = I • Results obtained during part of a collaboration be-

-10 - - T - - Y - - T - - T - - 1 - - - ] - - V -
I I t I I I tween Supdlec and CNES have been presented: the
~ - - T -'- r - -
.3o T2. . .° . . g- problem was to control one axis of the SPOT4 sat-
ellite despite lightly damped modes with large un-
-4O
400 -700 -600 -500 -400 -3~ -200 -I~ 0 I~
Phase (degrees)
certainties in their natural frequencies.

Fig. 13. Design 2: Nichols plot (reduced controller) The proposed solution involves three different tools.
First, a controller is determined by using the loop-
A IX-analysis has been performed, by using s I = 0.3, shaping Hoo design of McFarlane and Glover, which
$2 ----53 ----0.2, s c = 0.3. Fig. 14 gives the upper bound exhibits mathematical properties and is easy to use.
obtained for/4 whose maximal value is 0.76. From Special care is then taken to reduce the controller,
this result closed-loop stability is guaranteed with at using a modal reduction method. Finally IX-analysis
least 0:3 / 0.76 = 0.395 modulus margin, for fre- allows one to guarantee simultaneously, some sta-
quency deviations up to: bility margin and parameter deviations.

- x~0.3/ 0.76 = ~_39.5% for mode 1 Following this procedure, low-order controllers
- !~0.2/0.76 = ~24.1% for modes 2 and 3. have been obtained, which give the required per-
formance in the face of up to +30% deviation in the
frequencies of the modes.
Finally, Fig. 15 gives the responses of 0 x , 0x and

12x 10-S angularposifion ~(rd) 5 x 10 -5 angular rate #x(i'd/s) controltorque Cc (Nm)

0.04 . . . . .

°71
-..0~| r - - . t - -
| |

-0.~

-0.08

-0,12~
-I).1

"0.14 1
"'"

. . . . .
0 5 I0 15 20 25 30 0 5 I0 15 20 25 30 0 5 I0 15 20 25 30
time (see) time (see) time (see)

Fig. 15. Design 2: responses to torque disturbance

 HooControl of a Satellite Axis 1007

7. REFERENCES McFarlane, D. and K. Glover (1990). Robust Con-

troller Design Using Normalised Coprime Factor
Anderson, B.D.O. and Y. Liu (1989). Controller Plant Descriptions. Lecture Notes in Control and
reduction, concepts and approaches. IEEE Trans. Information Sciences, Springer Verlag.
Autom. Contr., 34, 802-812. McFarlane, D. and K. Glover (1992). A loop-sha-
Balas, G.J., J.C. Doyle, K. Glover, A. Packard and ping design procedure using H~o synthesis. IEEE
R. Smith (1993)./t-Analysis and Synthesis Tool- Trans. Autom. Contr., 37, 759-769.
box. The Math Works Inc. Morton, B.G. (1985). New applications of ~t to real
Chiang, R.Y., M.G. Safonov, K. Haiges, K. Madden parameter variation problems. 24th IEEE Conf.
and J. Tekawy (1993). A fixed controller for a on Decision and Control, Ft Lauderdale, FL,
supermaneuverable fighter performing the Herbst 242-250.
maneuver. Automatica, 29, 111-127. Safonov, M.G., R.Y. Chiang and H. Flashner (1991).
Doyle, J.C. (1982). Analysis of control systems with Hoo robust control synthesis for a large space
structured uncertainty, lEE Proc. Part D, 129, structure. J. Guidance, Contr. & Dynamics, 14,
242-250. 513-520.
Doyle, J.C., K. Glover, P.P. Khargonekar and B.A. Siret, J.M., G. MichaYlesco and P. Bertrand (1979).
Francis (1989). State-space solutions to standard On the use of aggregation techniques. In Hand-
H 2 and H~o control problems. IEEE Trans. book of Large Scale Systems Engineering Appli-
Autom. Contr., 34, 831-846. cations, 20-37, North-Holland.
Fan, M.K.H., A. Tits and J.C. Doyle (1991). Ro- Zhou, T., M. Fujita and F. Matsumura (1992). Ro-
bustness in the presence of mixed parametric un- bust control of a two-axis, magnetic suspension,
certainty and unmodeled dynamics. IEEE Trans. flexible-beam system based on Ho~ optimization
Autom. Contr., 36, 25-38. theory. Int. J. of Robust and Nonlinear Control,
2, 165-182.