You are on page 1of 8

Automatica, Vol. 34, No. 7, pp.

889—896, 1998
( 1998 Elsevier Science Ltd. All rights reserved
PII: S0005–1098(98)00024–7 Printed in Great Britain
0005-1098/98 $19.00#0.00

Brief Paper

Robustness and Trade-offs in Repetitive Control*


RICHARD C. H. LEE- and MALCOLM C. SMITH-

Key Words—Feedback control; internal model principle; disturbance rejection; robust stability; transient
responses; loop shaping.

Abstract—We investigate some robustness compromises in- 2. Problem formulation


volved in achieving disturbance rejection for periodic distur- In this paper, we will restrict attention to (rational) scalar
bances. We consider asymptotic rejection or attenuation of plants and controllers in the standard feedback configuration
disturbance harmonics and transient response. Robustness is (Fig. 1). We choose to formulate the repetitive control problem
measured with respect to uncertainty in the gap metric. ( 1998 as the rejection or attenuation of periodic disturbances u at the
1
Elsevier Science Ltd. All rights reserved. plant output y . At the same time we wish to take account of the
1
robustness of the feedback loop. To this end we will use the gap
1. Introduction metric robustness margin as our measure of uncertainty toler-
A goal of repetitive control is to ensure tracking/disturbance ance which is given by:
rejection of periodic references/disturbances. If it is desired to

KA B K
achieve asymptotic tracking/rejection, the internal model prin- I ~1
b :
" (I!CP)~1(I, C) .
ciple (see Francis and Wonham, 1976) suggests that the control- P,C P H
=
ler needs to contain poles at the frequency of the periodic signal
and all its non-zero harmonics (possibly infinitely many). This For background on this robustness measure, the reader is refer-
was the original idea of Inoue et al. (1981) who proposed using red to Georgiou and Smith (1990), McFarlane and Glover
a positive feedback around a time delay to generate an internal (1990), Vinnicombe (1993, 1998) and Balas et al. (1995). Here we
model of the periodic disturbance. It was observed by a number briefly recall a few relevant facts. Let the plant P be written as
of authors that such a scheme lacks robustness (see Hara et al., a normalized coprime fraction: P"N/M where N, M3H and
=
1985, 1988; Cevic and Schumacher, 1997). Various modifications N*N#M*M"1. There exist º, »3H such that the Bezout
=
of the scheme were proposed to improve the situation which Identity M»!Nº"1 is satisfied. It is well-known that all
typically amount to a relaxation of the requirement for asymp- stabilizing controllers can be parameterized in the form
totic rejection of the higher harmonics (see e.g. Hara et al., 1985,
1988; Weiss, 1997). A detailed survey on the development of º#MQ
C"
repetitive control can be found in Hillerström and Walgama »#NQ
(1996).
In this paper we continue the work of Lee and Smith (1996) for some Youla parameter Q3H . Using these facts, we obtain:
=
and investigate when asymptotic disturbance rejection/attenu-
b "E(»#NQ, º#MQ)E~1
ation conflicts with a robustness requirement for the feedback P,C H
=
loop. To measure robustness we will use the gap metric uncer-
1
tainty measure. We will show that asymptotic rejection of a har- " . (1)
monic will not conflict with robustness if the plant magnitude is JEM*º#N*»#QE2L=#1
large enough at the relevant frequency (Theorem 3.1). We also
show that a reasonable level of asymptotic attenuation of peri- The optimal robustness margin of P is defined by
odic disturbances can always be achieved without serious ro-
b (P):" sup b .
bustness compromises (Theorem 3.2). We also investigate the 015 P,C
issue of transient response to periodic disturbances. We define C 45"-;
an input—output (power—energy) measure of the magnitude of From equation (1), this optimization problem can be converted
the transient response. In the case where asymptotic rejection of into the standard Nehari problem form:
a harmonic does not conflict with robustness, we show that to
approach optimal robustness necessitates an arbitrarily poor b (P)~2" inf EM*º#N*»#QE2L=#1.
015
disturbance transient response (Theorem 4.1). Finally, we con- Q|H=
sider a possible design approach for repetitive controllers based Convenient formulas for the computation of b (P) have been
around H loop shaping, to achieve an appropriate compro- 015
= derived in McFarlane and Glover (1990). In practice, it would be
mise between disturbance attenuation, transient response and necessary to consider a weighted robustness measure: b ,
robustness. This is illustrated in the context of a CD player WP,CW~Ç
for some weighting function fixed in advance, where
radial control loop application. ¼, ¼~13H , to take account of scaling or to reflect an a priori
=
view on the frequency weighting of uncertainties. However, for
the purpose of the theoretical development we will take ¼"1.
The reader is referred to Balas et al. (1995) and McFarlane and
Glover (1990) for the use of a weighted form of the robustness
optimization problem in H loop shaping design.
* Received 20 June 1997; recieved in final form 6 January =
It will be convenient to define a class of disturbance signals
1998. This paper was not presented at any IFAC meeting. This D which can occur at u by
paper was recommended for publication in revised form by n 1
Associate Editor André L. Tits under the direction of Editor
d(t)" + a e+uk t,
Tamer Basar. Corresponding author Professor Malcolm Smith k

Tel. #44 1223 332745; Fax #44 1223 332662; E-mail mcs@ +uk|)n
eng.cam.ac.uk. where ) "Mju
: , , ju N and a 3C. This allows the possibility
n 12 n k
- Department of Engineering, University of Cambridge, of taking an arbitrarily large number of harmonics from a peri-
Cambridge, CB2 1PZ, U.K. odic signal or to consider some almost periodic disturbances.

889
890 Brief Papers

Indeed, if the plant P is strictly proper then the robustness


margin will be zero if all harmonics are included (commented in
Hara et al., 1988). Even if P is invertible at infinity (unlikely in
practice) it is easy to check that the control system fails to be
w-stable in the sense of Georgiou and Smith (1989). Thus,
asymptotic rejection for all harmonics is not a practicable possi-
bility. However, no robustness compromise is required at
a given harmonic if
DP( ju )D
k 5b (P).
J1#DP( ju )D2 015
Fig. 1. Standard feebback configuration. k
We mention that the above theorem is closely related to results
of Cevik and Schumacher (1995) on the regulator problem with
robust stability. The next result investigates robustness compro-
mises if asymptotic attenuation of harmonics is required rather
We will consider disturbance rejection of signals in D from than asymptotic rejection.
n
the point of view of both asymptotic rejection or attenuation
(Section 3) as well as transient response (Section 4). ¹heorem 3.2. Suppose 0(a4b (P). For any e'0 there
015
exists a C such that b e5a!e and
e P,C
3. Asymptotic disturbance rejection and attenuation
The asymptotic rejection of disturbances u 3D at y DP(I!C P)~1( ju ) D4m (2)
1 n 1 e k k
requires that
for all ju 3) , if and only if
P(I!CP)~1( ju )"0 k n
k 1
m 50 when DP( ju )D5 ,
for all ju 3) . We define the quantity k k Ja~2!1
k n
m 5g(a, DP( ju )D) otherwise,
b (P, ) ):" sup Mb : P(I!CP)~1(s)"0 ∀s3) N k k
015 n P,C n
C 45"-; where
which is the optimal robustness margin subject to this distur- x(1!xJa~2!1)
bance rejection constraint. Using the fact that P(I!CP)~1" g(a, x):" .
1#x2
N(»#NQ) we see that
b (P, ) )~2" inf EM*º#N*»#QE2L=#1, Proof. The sufficiency part of the proof relies on the following
015 n standard boundary interpolation result (similar to the one used
Q| H=

Q(+uk)/Qk +uk |)n in the proof of Theorem 3.1). K
where the constraints Q( ju )"!(»/N)( ju )":Q are taken
k k k ¸emma 3.3. (Khargonekar and Tannenbaum, 1985). Let ¹
over all ju 3) such that P( ju )O0.
k n k 1
and ¹ be any stable real rational functions. Given a set of
2
numbers z , 2 , z 3C. Suppose E¹ #¹ QEH="f for some
¹heorem 3.1. Let P( ju )O0 for each ju 3) . Then
k k n 1 n 1 2
Q3H and Dz D4f for all k. Then for each d'0, there exists
= k

G H
DP( ju )D a Q 3H such that
b (P, ) )" min b (P), k . d =
015 n 015 J1#DP( ju )D2
+uk|)n k E¹ #¹ Q EH=4f#d
1 2 d
and
Proof. By writing the inner—outer factorization of M*º#N*»" (¹ #¹ Q )( ju )"z
¹B~1 where ¹, B3H and B is the Blaschke product whose 1 2 d k k
=
zeros are the unstable poles of M*º#N*», we have for all ju 3) .
k n
DM*º#N*»#QD( ju)"D¹#BQD( ju) Sufficiency: We first choose a compensator C "(º#Q M)/
0 0
for all u3RXM$RN. So, for each ju 3) , (»#Q N) such that b~1 "a~1 (or equivalently, EM*º#
k n 0 P,CÒ
N*»#Q EL="Ja~2!1). Using the same inner-outer
D¹#BQD( ju )"DM*º#N*»#Q D( ju ) 0
k k k factorization as in the proof of Theorem 3.1 (namely,
M*º#N*»"¹B~1 for some ¹3H and all-pass B3H ),
1 = =
" . we have E¹#BQ EH="Ja~2!1. Define
DP( ju )D 0
k

A B
M*
The theorem is then proved by using the standard result on z "!
: B ( ju ) for DP( ju ) DJa~2!1*1, (3)
k N k k
boundary interpolation problem (Khargonekar and Tasnnen-
baum, 1985), namely

A B
DPDJa~2!1
inf E¹ #¹ QEH= z "!
: B ( ju )
1 2 k (1#DPD2)NM k
Q| H∀=
Q(+uk)/Qk +uk |)n
for 0(DP( ju )DJa~2!1(1. (4)
G
" max inf DD¹ #¹ QEH , D¹ #¹ QD( ju )
+uk |)n Q|H=
1 2 = 1 2 k H k
Then, we can check that Dz D4Ja~2!1 for each k. So, for any
k
K e'0, by Lemma 3.3, there exists a Q 3H such that
where ¹ , ¹ 3H . e =
1 2 =

SA B
1 2
Remark 3.1. The analogous constraints on Q3H in the matrix
= E¹#BQ EH=4 !1 (5)
case are Q( ju )"!(N*N)~1N*»( ju ) and we also get e a!e
k k

G H
p (P( ju )) and the interpolation constraints
b (P, ) )" min b (P), .*/ k ,
015 n 015 J1#p2 (P( ju )) (¹#BQ ) ( ju )"z ∀ju 3)
+uk|)n .*/ k e k k k n
(6)
if the matrix P( ju ) has full column rank for each ju 3) . are satisfied. We will now choose
k k n
º#Q M
The above theorem illustrates the impossibility of achieving C" e .
asymptotic disturbance rejection of general periodic inputs. e »#Q N
e
Brief Papers 891

Then inequality (5) is equivalent to b e5a!e. We will now


P,C
check that the constraints (6) are sufficient for inequality (2) to
be satisfied. Firstly, we can verify using the properties of the
coprime factors that equation (6) implies
(»#Q N)( ju )"(M*#z NB~1)( ju ).
e k k k
Thus, if DP( ju )DJa~2!151 then equation (3) gives
k
P(I!C P)~1( ju )"N(»#Q N) ( ju )"04m .
e k e k k
If 0(DP( ju )DJa~2!1(1 then equation (4) gives
k

A B
1!DPDJa~2!1
(»#Q N) ( ju )" ( ju )
e k (1#DPD2)M k

which implies that


DP(I!C P)~1( ju )D"DN(»#Q N) ( ju )D"g(a, DP( ju ) D)4m . Fig. 2. The function g(a, x) for a"0.3.
e k e k k k
Finally, if P( ju )"0, no interpolation constraint on the
k
controller parameter is required in order to satisfy
P(I!C P)~1 ( ju )"0.
e k
and
Necessity: Clearly, for any ju 3) such that DP( ju )D51/ G(s)!G( ju )
k n k y' (s)" + k a.
(Ja~2!1), there is nothing to prove. So, without loss of gener- 53 s!ju k
+uk|)n k
ality, assume that DP( ju ) D(1/(Ja~2!1) for all ju 3) . Then
k k n Now suppose that G(s)"D#C(sI!A)~1B where A is stable.
for sufficiently small e'0,
Let ¸ be the observability gramian matrix with respect to this
0
1 state-space realization, i.e.
DP( ju )D(
k J(a!e)~2!1 =
for all ju 3) . Select such an e'0 and a C such that
k n e
¸ "
0 P0 etA C*CetA dt50
*

b e5a!e. We know (see Vinnicombe, 1998) that


P,C and A*¸ #¸ A#C*C"0. Then
0 0
DP(I!C P)~1( ju )D
e k G(s)!G( ju )
k "!C(sI!A)~1( ju I!A)~1B
k
K K
1#x( ju )DP( ju )DJ(a!e)~2!1 s!ju
" k k )P( ju ) k
1#DP( ju )D2 k for each k and the time domain transient is
k
for some Dx( ju )D41. Thus,
k y (t)"!C etA + ( ju I!A)~1Ba
m 5DP(I!C P) ~1( ju )D 53 k k
k e k +uk|)n
1!DP( ju )DJ(a!e)~2!1 for t50. Define the vector
5 k )DP( ju )D
k

AB
1#DP( ju )D2 a
k 1
a:" F
"g(a!e, DP( ju )D).
k a
n
Since this holds for all sufficiently small e'0, then
m 5g(a, DP( ju )D). K and the n]n Hermitian non-negative definite matrix
k k
M:"(B*(!ju I!A*)~1¸ ( ju I!A)~1B) 4 4 .
The above theorem tells us a minimum modulus for the plant k 0 l 1 k,l n
in the case of asymptotic rejection, together with the asymptotic We now consider the square of transient energy:
attenuation which is consistent with a certain robustness mar-
gin, for plant input disturbances. For example if we require n n
Ey (t)E2" + + a*B*(!ju I!A*)~1
a"0.3 (a reasonable level in practice) then asymptotic rejection 53 2 k k
of a harmonic is possible providing DP( ju )D'0.314, and other- k/1l/1
k ]¸ ( ju I!A)~1Ba "a*Ma.
wise, attenuation to less than 7.68% of the disturbance magni- 0 l l
tude is always possible as shown by Fig. 2. Note that g(a, x) has
Note that
a maximum of

A B
1 T 1@2
g "
a
.!9 2(1#J1!a2)
at x "
a
.!9 1#J1!a2
,
EaE"pow(d(t)):" lim
T?=
¹
0
P
Dd(t)D2 dt ,

where E)E denotes the Euclidean vector norm. If we define


so even with a"1, a 50% reduction is always possible. We EGE D to be the maximal transient energy Ey (t)E for signals
530 53 2
mention that a 50% reduction in the steady state was shown to d(t)3D n with pow(d(t))41, then
be possible together with robustness to time delays in Hiller- n
ström and Sternby (1996). EGE D "Jj1 (M),
530 n
4. ¹ransient performance where jM ()) denotes the maximum eigenvalue. Our next result
Consider a signal d(t)3D which is input to a SISO stable shows generally that arbitrarily poor transient performance
n results from demanding both optimal robustness and asymp-
system G(s) (e.g. in our case we will take G"P(I!CP)~1). The
output yL (s) can be decomposed into a steady state y' (s) and totic disturbance rejection.
44
a finite energy transient component y' (s). That is: yL (s)"y' (s)#
53 44
y' (s) where ¹heorem 4.1. Suppose
53 DP( ju )D
G( ju ) k
y' (s)" + k a b (P)(
015
44 s!ju k J1#DP( ju )D2
+uk |)n k k
892 Brief Papers

for all ju 3) . Consider any sequence of stabilizing com- where EHE2 ":= : DH(p#ju)D2 du. We now claim that, for
k n p,2 ~=
pensators C with b Pb (P) as mPR and m sufficiently large,
m P,Cm 015
P(1!C P)~1( ju )"0 for each ju 3) and each m. Then
m k k n
K K
P(I!C P)~1!P(I!C P)~1
EP(I!C P)~1E D "R 015 m 41. (10)
m 530 n s!ju
k p,2
as mPR. To see this, note first that DP(I!C P)~1D and DP(I!C P)~1D
015 m
are bounded above by b (P)~1#1, for all sufficiently large m.
Proof. A direct consequence of the assumed inequalities is that 015
Thus both
b (P, ) )"b (P) as seen in Theorem 3.1. Next note that
015 n 015

P K K
p`j= P(I!C P)~1 2

K K
G(s)!G( ju ) 2 015 ds
k "(k, k)th entry of M4EGE2 D , s!ju
s!ju 530 n p`jK2 k
k H
2
and
where G"P(I!C P)~1. It will be sufficient to prove that any
m

P K K
diagonal entry of M tends to infinity under the conditions of the p~jK2 P(I!C P)~1 2
theorem. 015 ds
s!ju
Suppose to the contrary that there exists a subsequence ~j= k
C "(º#MQ )/(»#NQ ) (which we will still refer to
m m m can be made arbitrarily small (and similarly with C replaced
015
as C for convenience) such that b Pb (P),
m P,Cm 015 by C ) for sufficiently large K . Further
m 2
P(I!C P)~1( ju )"0 ∀ju and there exists a constant K '0
m k k 1

P K K
such that p`jK2 P(I!C P)~1!P(I!C P)~1 2
015 m ds

K K
P(I!C P)~1 s!ju
m 4K (7) p~jK2 k
s!ju 1 tends to zero as mPR, for any fixed K . These establish
k H
2 2
equation (10). Thus, equations (9) and (10) together imply that
for all m. First, there exists a unique Q 3H such that (Zhou
015 =

K K
et al., 1996) P(I!C P)~1
m
E(», º)#Q (N, M)EH="b (P)~1. s!ju
k
015 015
H
2

K K
Next, the set of H functions: MS "(», º)#Q (N, M): m51N P(I!C P)~1
m
= m m 5
is a normal family on the right half plane. Thus, there exists s!ju
S such that S PS uniformly on compact subsets of the k p,2
-*. m -*.

K K
right half plane. First note that S 3H . This follows since P(I!C P)~1
-*. = 015
ES (s)E must be bounded above by b (P)~1 on any compact 5
-*. 015 s!ju
k p,2
subset of the right half plane. Now define

K K
P(I!C P)~1!P(I!C P)~1

A B
!º ! 015 m
Q "S 3H . s!ju
-*. -*. » = k p,2
5K #1
We claim that 1
for sufficiently large m, which contradicts equation (7). K
S "(», º)#Q (N, M)
-*. -*.

A B
!º 5. Experimental results of a compact disc player
"(», º)#S (N, M). (8) We will now illustrate the application of Theorems 3.1 and 4.1
-*. »
to the radial control loop of a compact disc (CD) player. The
This follows since
experiments used a Philips CD930 player which was specially

A B
!º modified to allow identification and external control for the
S "(», º)#S (N, M)
m m » radial (and focus) loop. System identification and controller
implementations were done by a digital signal processor
for all m, and S PS uniformly on compact subsets of
m -*. (TMS320C40 processor) at a sampling frequency of 20 kHz.
the right half plane. From equation (8) we obtain In order to read data from the optical compact disc, the laser
E(», º)#Q (N, M)EH "b (P)~1 which means that
-*. = 015 spot must follow the movement of the track accurately in both
Q "Q since Q is unique. A further property of Q is that
-*. 015 015 015 the radial and focus directions. This has to be done within
(Zhou et al., 1996) a radial tolerance of $0.1 lm (and a focus accuracy of
E((», º)#Q (N, M))( ju)E"b (P)~1 $1 lm). An audio CD has constant pit density over the whole
015 015 of the disc, which necessitates a variable rotation speed of the
for all u. For each k, the condition disc, decreasing from 8.3 Hz at the innermost track to 3.6Hz at
the outermost track. Experiments were carried out with an
P(I!C P)~1( ju )"0 eccentric test disc operating at an intermediate track where the
015 k
rotation speed was around 6 Hz (denote u "2n ) 6). The radial
is equivalent to 0
plant is basically a double integrator (with parasitic resonance
Q ( ju )"!»/N( ju ),
015 k k dynamics at high frequency). For the track and disc considered,
which implies
the open-loop plant was found to have a gain crossover fre-
b (P)~1"E((», º)#Q (N, M))( ju )E quency of around 400 Hz. Therefore, the open-loop nominal
015 015 k radial plant P(s) was chosen to be
"J1#D(M*º#N*»#Q ) ( ju )D2
015 k (2n ) 400)2
P(s)" .
J1#DP( ju )D2 s2
" k
DP( ju )D
k This transfer function is the one seen from the externally imple-
( b (P)~1. mented controller which means that the inputs and outputs are
015 in volts and are normalized to the 0—10 V range. Taking account
Therefore, we must have P(I!C P)~1( ju )O0. In this case, of this normalization, the unweighted b was chosen as the
015 k P,C
we can find p'0 such that robustness measure.
The existing Philips radial controller is a PI plus lead com-

K K
P(I!C P)~1
015 5K #2 (9) pensator designed by classical methods so as to stabilize the
s!ju 1 marginally stable open loop plant and to achieve asymptotic
k p,2
Brief Papers 893

disturbance rejection at dc. When normalized as in external where ) "M0, ju , !ju N and a 3C. It can easily be checked
3 0 0 k
implementation: using Theorem 3.1 that
(s#314.5)(s#862.1) b (P)"b (P, ) )"0.3827.
C "!1.747 . 015 015 3
P)*-*14 s(s#6622) We now apply the H loop shaping method for robust control-
=
This controller gives a gap metric robustness margin of ler design (Balas et al., 1995; McFarlane and Glover, 1990) in
rejecting periodic sinusoids of frequencies in ) . Consider a set
b
P,CP)*-*14
"0.2054. The closed loop P(I!C
P)*-*14
P)~1 has 3
a bandwidth 473.0 Hz. Steady-state attenuation between 3.6 and of weighting functions
8.3 Hz, using this Philips controller, is less than !2.88 dB (as (s#b) (s#b)2#u2
shown in Fig. 8). Note that the radial error RAD ERR contains ¼ (s)" ] 0
b s s2#u2
a periodic signal of fundamental frequency 6 Hz as well as 0
harmonics at frequencies 18 and 36 Hz (see Figs 3 and 5). This is parameterized by b3[0.5, 15000]. The final weighted loop-
caused by the eccentric rotation and the imperfect spiral track of shaping controller C is constructed as a product of the weight
the CD. For the purpose of illustration we will consider the b
¼ and the optimal robustness controller of the weighted plant
asymptotic rejection of the fundamental 6 Hz frequency only. b
¼ P. Different values of b will lead to different controllers with
We therefore consider disturbances of the form b
different robustness margins b and different power-energy
P,Cb
d(t)" + a e+ukt, induced norms EGE D (see Figs 6 and 7). Here, we take
k 530 n
G"P(I!C P)~1 for the measure of the power-energy
+uk|)3 0.3

Fig. 3. The physical radial error (RAD ERR) using the dSPACE implemented Philips controller C .
P)*-*14

Fig. 4. The physical radial error (RAD ERR) using the dSPACE implemented loop shaping controller C .
0.3
894 Brief Papers

Fig. 5. The spectra of the physical radial errors (RAD ERR) using the dSPACE implemented controllers. C : dashed ‘‘--’’,
P)*-*14
C : solid ‘‘—’’.
0.3

Fig. 6. The relation between b and the parameter b.


P,C

induced norms corresponding to a disturbance entering at the ness and power-energy induced norm are given by b "0.3
P,CÒ?Ê
plant input. As seen in Fig. 7 and proved in Theorem 4.1, the and EP(I!C P)~1E D "0.09805, which are almost the
0.3 530 n
power-energy induced norm blows up when the gap metric same as for the seventh order controller. This closed loop has
robustness margin b b is close to the optimal value (this corres- a desirable bandwidth of 514.8 Hz (Fig. 8). It can also be seen
P,C
ponds to small values of b). The opposite happens for large from Fig. 8 that a steady state attenuation to less than !48 dB
values of b. So, the region which is of practical interest is the is achieved for the frequency range 3.6—8.3 Hz. We can observe
middle range where both the robustness margin and the transi- that the 6 Hz component in the radial error RAD ERR using
ent performance are satisfactory. For instance, with the desired this loop-shaping controller is rejected (see Figs 4 and 5). On the
robustness margin 0.3 and the parameter b"374, the final other hand, the loop-shaping controller performs less well in
fourth order loop shaping controller (after model reduction from attenuating high frequency disturbances.
a seventh order controller) is
6. Conclusion
(s#284.5$j69.99)(s#470.7$j433.4) In this paper, we have formulated the repetitive control prob-
C (s)"!3.152 .
0.3 s(s2#u2)(s#6605) lem in an input—output (H -control) context. Trade-offs in
0 =
disturbance rejection/attenuation, both asymptotic and transi-
In fact, we can see that this controller is of minimum order in ent, have been considered against a robustness requirement.
rejecting frequencies in ) for a double integrator. The robust- A design approach has been suggested and applied to the
3
Brief Papers 895

Fig. 7. The relation between the power—energy induced norm EGE D and the robustness margin, parameterized by b.
530 n

Fig. 8. The Bode diagram for the closed loops. P(1!C P)~1: dashed ‘‘- -’’; P(1!C P)~1: solid ‘‘—’’.
P)*-*14 0.3

control of a CD player. The fourth order loop-shaping control- Francis, B. A. and W. M. Wonham (1976). The internal model
ler was implemented using DSP in real time with satisfactory principle of control theory. Automatica, 12, 457—465.
robustness and transient decay. Georgiou, T. T. and M. C. Smith (1989). w-stability of feedback
systems. Syst. Control ¸ett., 13, 271—277.
Georgiou, T. T. and M. C. Smith (1990). Optimal robustness
References in the gap metric. IEEE ¹rans. Autom. Control, AC-35,
Balas, G. J., J. C. Doyle, K. Glover, A. Packard and R. Smith 673—686.
(1995). k-Analysis and Synthesis ¹oolbox ºser’s Guide: For Hara, S., T. Omata and M. Nakano (1985). Synthesis of repeti-
ºse with MA¹¸AB. The MathWorks, Inc. tive control systems and its application. Proc. 24th IEEE
Cevik, M. K. K. and J. M. Schumacher (1995). The regu- Conf. on Decision and Control, pp. 1387—1392.
lator problem with robust stability. Automatica, 31, Hara, S., Y. Yamamoto, T. Omata and M. Nakano (1988).
1393—1406. Repetitive control system: a new type servo system for peri-
Cevik, M. K. K. and J. M. Schumacher (1997). The robust odic exogenous signals. IEEE ¹rans. Autom. Control, AC-33,
regulation problem with robust stability. Preprint. 659—668.
896 Brief Papers

Hillerstro( m, G. and J. Sternby (1996). Robustness properties of McFarlane, D. C. and K. Glover (1990). Robust Controller
repetitive controllers. Int. J. Control, 6, 939—961. Design ºsing Normalized Coprime Factor Plant Descriptions,
Hillerstro( m, G. and K. Walgama (1996). Repetitive control Lecture Notes in Control and Information Sciences, Vol. 138,
theory and applications—a survey. Proc. 13th IFAC ¼orld Springer, Berlin.
Congress, Vol. D, 1—6. Vinnicombe, G. (1993). Frequency-domain uncertainty and
Inoue, T., M. Nakano and S. Iwai (1981). High accuracy control the graph topology. IEEE Trans. Autom. Control, AC-38,
of servomechanism for repeated contouring. 10th Annual 1371—1383.
Symp. on Incremental Motion Control Systems and Devices, Vinnicombe, G. (1998). ºncertainty and feedback—H loop
=
pp. 285—292. shaping and the l-gap metric, Imperial College Press, in press.
Khargonekar, P. P. and A. Tannenbaum (1985). Non-euclidian Weiss, G. (1997). Repetitive control systems: old and new ideas.
metrics and the robust stabilization of systems with parameter In Control in the 21st Century. C. I. Byrnes, B. N. Datta, D. S.
uncertainty. IEEE Trans. Autom. Control, AC-30, 1005—1013. Gilliam and C. F. Martin, Eds, pp. 389—404. Birkha( user
Lee, R. C. H. and M. C. Smith (1996). Some remarks on Verlag, Boston.
repetitive control. Proc. of 35th IEEE Conf. on Decision and Zhou, K., J. Doyle and K. Glover (1996). Robust and Optimal
Control, pp. 2023—2024. Control, Prentice-Hall, Englewood Cliffs, NJ.

You might also like