1

Robust Control( H

/sliding mode
controller) with an application to
multi-input power system stabilisation
Magdi S. Mahmoud, and Omer M. Elagemi
!
AbstractRobust control theory is a method that ex-
plicitly deals with uncertainty in its approach to con-
troller design.In this work, H/sliding mode controller
under a worst-case scenario is described and de-
signed.This is done in order to eliminate the prob-
lem associated with unmeasurable state variables in
the traditional sliding-mode controller, and to improve
the time-domain performance under a worst-case sce-
nario. The control input, derived by Lyapunovs second
method, keeps the system stable. Proof of closed-
loop stability is done by an additional form of a sliding
surface and an estimation error.
Index Terms: H,Robust control,Sliding mode con-
trol,Stability
1 INTRODUCTION
Robust control design is the search for
controllers which can cope with a de-
signer specied range of process dy-
namics or disturbance signals. It refers
to the control of unknown plants with
unknown dynamics subject to unknown
disturbances. Clearly, the key issue
with robust control systems is uncer-
tainty and how a controller can be
designed which will maintain perfor-
mance/or stability in the face of ex-
ternal disturbances. A controller de-
signed under a set of parameters will
be deemed robust if it is able to main-
tain acceptable performance and grace-
ful degradation under different assump-
tions provided those assumptions are
within set limits.Examples that have
found widespread use are the opera-
tional ampliers and bipolar transistors.
There are a variety of techniques that
have been developed for robust control
to handle uncertainties that often occurs
at high frequency. In this paper, we con-
sider sliding-mode control [l] and H
optimisation-based control [2-5] which
have been developed as effective con-
trollers under uncertainties. The tradi-
tional sliding-mode controller (SMC) is
associated with the problem of unmea-
surable state variables and unsatisfac-
tory time-domain performance under a
worst-case situation. In order to improve
and solve this problems, a standard SMC
[3-5] is extended to the H/sliding-
mode controller and is the subject of this
work. The controller so designed by this
hybridization is applied to power system
stabilization and the linear time-domain
simulation is done in a power system.
2 ROBUST CONTROL TECH-
NIQUES
2.0.1 sliding mode control
In sliding-mode control, the system state
is made to follow a specic trajectory
dened by the sliding surface in the
state space of the system. It can switch
from one continuous structure to another
based on the current position in the state
space.The selection of the maximum and
the minimum values guarantees sliding
mode operation for any operating point
2.0.2 Hoptimisation-based control
The Hdesign methodology originated
with the seminal work of Zames[2] and
since then, considerable progress has
been made in the development of the
theory. The key idea of the H, control
problem is to nd an internally stabil-
ising controller for the controlled plant
such that the -norm of the closed-loop
transfer function is below a given level
(a positive scalar). They are used in
control theory to synthesize controllers
achieving stabilization with guaranteed
performance. To use H methods, a
control designer expresses the control
problem as a mathematical optimization
problem and then nds the controller
that solves this. These methodologies
have been applied for the stabilisation of
power systems [6-8, 9-11].
It is important to keep in mind that
the resulting controller is only optimal
with respect to the prescribed cost func-
tion and does not necessarily represent
the best controller in terms of the usual
performance measures used to evaluate
controllers such as settling time, energy
expended and so on.
3 PROPOSED H/SMC AND
STABILITY PROOF
An H/sliding mode controller
(H/SMC) is proposed to solve the
problem associated with unmeasurable
state variables in the traditional sliding
mode controller (SMC), and to improve
the time-domain performance under a
worst-case scenario.
The state equations under worst-case
conditions can be expressed as
x(t) = Ax(t)+B1Wworst(t)+B2u(t)
z(t) = C1x(t)+D11Wworst(t)+D12u(t)
y(t) = C2x(t)+D21Wworst(t)+D22u(t)
where x R
n
,W worst R
m
1,u R
m
2,
z R
p
1,y R
p
2,A is the n x n system
matrix, B1 is the n x m1 exogenous input
matrix, B2 is the n x m2 control matrix,
C1 is the p1 x n regulated output matrix,
C2 is the p2 x n output or measurement
matrix, D11 is the p1 x ml regulated di-
rect feed forward matrix, D12 is the p1x
m2 regulated direct feed-forward matrix,
D2l is the p2 x ml output direct feed
forward matrix, and D22 is the p2 x m2
output direct feed-forward matri.
Assumptions I : The following rank
conditions are made [3-5].
Al. The pair (A, B2) is stahilizable and
(C2, A) is detectable. A2. Rank D12 =
m2, rank D21=p2.
A3. Rank
_
A jwI B2
C1 D12
_
= n +m2
A4. Rank
_
A jwI B1
C2 D21
_
= n +p2
for all frequencies.
AS. Dl1=0 and D22=0.
The standard H estimator equation
under a worst-case
situation can be expressed as[3 5].

x(t) = A x(t)+B
1

W
worst
(t)+Z

K
e
(y(t) y(t))
where

W
worst
(t) =
2
.B1
T
X

x(t)
y(t) = [C2 +
2
.D21B1
T
X

] x(t)
in equation.5 is a positive scalar value
and is
iterated until the desired specication is
obtained.
The controller gain K
c
is
K
c
=

D12(B2
T
x

+D12
T
C1)
2
where

D12 = (

D12
T
D12)
1
The estimator gain K
e
is
K
e
= (yC2
T
+B1D21
T
)

D21
where

D21 = (D12D11
T
)
1
the term Z is
Z

= (I
2
.Y

)
1
The controller Riccati equation termX

is given by
X

= Ric
_
AB
2

D
12
D
12
T
C
1

2
B
1
B
1
T
-B
2

D
12
B
2
T
-
T

C
1
(AB
1

D
12
T

D
12
T
C
1
_
where

C1 = (I D12

D12D12
T
)C1
The estimator Riccati equation term Y

is given
by
Y

= Ric
_
(A B
1

D
21
D
21
T
C
2
)
T

2
.C
1
C
1
T
C
2
T

D
21
C
2

B
1

B
1
T
(A B
1

D
12
T

D
12
T
C
1
_
Where

B1 = B1(I D21
T

D21D21)
The estimated control input vector under worst-
case
conditions is
u
c
(t) = K
c
x(t)
where x(t) R
n
is the estimated state variables,
and K
c
is
the input gain of the worst-case control. The
internally
stabilising control gain using the packed matrix
notation
is
K
H

c
(s) =
_
A
1
Z

K
e
K
e
0
_
where
A
1
:= AB
2
K
e
Z

K
e
C
2
+

2
.(B1B1
T
Z

K
e
D21B1
T
)X

The closed-loop system can be expressed as

x(t)
_
=
_
A B2K
e
Z

K
e
C2 A
2
_ _
x(t)
x(t)
_
+
_
B1
Z

K
e
D21
_
W
worst
(t)
_
z(t)
y(t)
_
=
_
C1 D12K
w
C2 0
_ _
x(t)
x(t)
_
+
_
0
D21
_
W
worst
(t)
where
A
2
:= AB
2
K
w
+
2
.B
1
B
T
1
X

K
e
(C
2
+
2
.D
21
B
T
1
X

)
Remark: A stabilising compensator can be
obtained if and only if there exist positive semi
denite solutions to the two Riccati equations
(X

)<
2
where (A) = spectral radius
ofA =largest eigenvalue ofA =
max
(A).
From eqns. 1 and 5, the state equation is
expressed as
x(t) = Ax(t) +B
1
W
worst
(t) +B
2
u(t)
= [A+B
1
(
2
.B
1
T
X

)]x(t) +B
2
u(t)
The sliding-surface vector and the differential
sliding surface vector can be expressed as
x(t) = G
T
x(t)

x(t) = G
T
x(t)
wherex R
n
is the state variable, G
T
is the
sliding surface
gain expressed in the Appendix (Section 8.2).
Substituting eqn. 22 into eqn. 24 gives
x(t) = G
T
x(t)
= G
T
[(A + B
1
(
2
.B
1
T
X

))]x(t) +
B
2
u
equal
WSMC
(t)] = 0
To obtain the average control input in the sliding
mode, an equal control input u
equal
WSMC
(t) under
a worst case scenario from eqn. 25 is obtained
by
u
equal
WSMC
(t) = (G
T
B
2
)
1
[G
T
(A +
B
1
(
2
.B
1
T
X

))]x(t)
= K
equal
WSMC
x(t)
where
K
equal
WSMC
:= (G
T
B
2
)
1
[G
T
(A+B
1
(
2
.B
1
T
X

))]
3
The estimated equal sliding-mode input
u
equal
WSMC
(t) under
a worst case is dened as
u
equal
WSMC
(t) = K
equal
WSMC
x(t)
In eqn. 29, the K
equal
WSMC
is called equal gain
under worst SMC.
To determine a control law that keeps the system
on (x(t)) = 0, we introduce Lyapunovs second
method
V (x(t)) =
2
(x(t))/2
The time derivative of V(x(t)) can be expressed
as

V (x(t)) = (x(t)) (x(t)) = G

V (x(t)) = G
T
x(t)G
T
([A+B1(
2
.B1
T
X

))]
x(t) +B2u
equal
WSMC
(t)] 0
where the u
WSMC
(t) is the input vector of a
worst-case SMC. Selection of the maximum and
the minimum values in the following manner
will guarantee sliding-mode operation for any
operating point. Eqn. 32 can be reduced as the
control input with switching function
u
+
WSMC
(G
T
B2)
1
[G
T
(A+B1(
2
.B1
T
X

))]x(t)
for
G
T
x(t)>0
u

WSMC
(G
T
B2)
1
[G
T
(A+B1(
2
.B1
T
X

))]x(t)
for
G
T
x(t)<0
Equations 33 and 34 can be formed as the control
input with
sign function
u
sign
WSMC
(t) = (G
T
B2)
1
[G
T
(A+B1(
2
.B1
T
X

))]Xx(t)sign((x(t))) Equation 35 can be sim-

))]
is the input gain of the worest-case SMC. Finally,
the
estimated control input vector of the proposed
H/SMC
is expressed as
u
sign
H/SMC
(t) = K
WSMC
x(t)sign(( x(t)))
_
x(t)

x(t)
_
=
_
A B
2
K
WSMC
sign(( x(t)))
Z

K
c
C
2
A
2
_
_
x(t)
x(t)
_
+
_
B1
Z

K
c
D
2
1
_
W
worst
(t)
_
z(t)
y(t)
_
=
_
C1 D12K
WSMC
sign(( x(t)))
C2 0
_ _
x(t)
x(t)
_
+
_
0
D21
_
W
worst
(t)
where
A
2
:= A+B2K
WSMC
sign(( x(t)))
+
1
.B1B1
T
X

K
c
(C2+
2
.B1
T
D21X

)
Theorem 1 : Consider the state equs. 1-3 and
the H estimator state eqn. 4 for the regulation
problem under a worst case. The estimated
sliding-mode control law with sign function
that keeps the system stable is guaranteed
asymptotically stable for eqn. 1
u
sign
H/SMC
(t) = K
equal
WSMC
x(t)sign(( x(t)))
K
equal
WSMC
:= (G
T
B2)
1
[G
T
(A+B1(
2
.B1
T
X

))]
subject to sign (( x(t))) = +1 for ( x) 0
sign (( x(t))) = 0 for ( x) = 0
sign (( x(t))) = 1 for ( x) 0
Proof: Given in the Appendix (Section 8.1). The
algorithm for the realisation of the proposed
H /SMC can be summarised as follows: (i)
Set the controller state equation under the worst
case. (ii) Set the H estimator state equation
under the worst case. (iii) Check if assumptions
1 (rank conditions) are satised. If they are not,
reformulate the problem by adding weights or
adding (ctitious) inputs or outputs. (iv) Select a
large positive value of . (v) Solve the two Riccati
equations. Determine if the solutions are posi-
tive semi-denite; also, verify that the spectral
radius condition is met. (vi) If all the proceeding
conditions are satised, lower the value of.
Otherwise, increase it. Repeat steps (iii) and (v)
until either an optimal or satisfactory solution
is obtained. (vii) Choose the equation of sliding
hyperplane expressed in the Appendix (Section
8.2). (viii) Compute the estimated control input
with the sign function. (ix) Apply the estimated
control input to the plant. The block diagram
4
Fig. 1. Block diagram of proposed
H/SMC
TABLE 1
Initial condition data
M T
do
X
d
w
o
P V Q V
9.26 7.76 0.973 0.19 377 0.75 0.025 1.05
K
A
T
A
R B K
c
T
p
T
1
T
2
50.0 0.05 -0.034 0.262 14.0 1.0 0.6850 0.1
of the proposed H/SMC is shown in Fig. 1
which contains two main blocks, the controlled
plant and the H/SMC. The lower block is
categorised in two parts. The rst part is the
signal that is estimated. The second part is the
control signal.
3.0.3 Numerical example
The linearised fourth-order power system model
for a single machine to the innite bus system
is shown in Fig. 2. Descriptions of the detailed
notation in a power system are found in [12]. A,
B1 and B2 are formed as
A =
_
D/M K
1
/M K
2
/M 0
w
o
0 0 0
0 K
4
/T
do
1/T
do
K
3
1/T
do
0 K
A
K
5
/T
A
K
A
K
6
/T
A
1/T
A
_
B1 =
_
1/M 0 0 0

T
(1)
B2 =
_
0 0 0 K
A
/T
A

T
(2)
Data for initial conditions are listed in Table 1.
A, B2, C1, and C2 are given by
A =
_

_
0 0.0615 0.1048 0
376.991 0 0 0
0 0.0674 0.1957 0.1289
0 50.0410 845.38 20.0
_

_
Fig. 2. Block diagram of linearised power
system model for small-signal stability
study
B2 =
_
0 0 0 1000

T
C1 = diag(80000)
C2 =
_
1 0 0 0

h controller gain is obtained by

h estimator gain is obtained by

T
Sliding surface gain is obtained by
G = 1e + 004[1.2187 0.00120.0169O.OOO1]
T
4 SIMULATION RESULTS
We applied the new controller designed to a
power system for small signal to check its stabil-
ity. The power system is a single phase machine
as oppose to three phase system. The model of
the power system is shown in Figure 2
4.1 Under normal load condition
4.1.1 Without control
The model is simulated rst without any control
and the result is shown in g 3. It can be seen
from the gure that the system is unstable with-
out any control as it grows unbounded and this
is not desired.
5
Fig. 3. Angular velocity waveform without
any control
Fig. 4. Angular velocity waveforms under
normal load operation
Conventional PSS
4.1.2 with conventional control
A PID controller is applied to the model and
simulation is carried out. The result is shown in
Fig 4. It is seen that though this perform better
than without control ,it fall short of expectation
with regard to the time domain performance.
4.1.3 With Hcontrol
Further on, H innity control is applied to the
model and the simulation performed. It is seen
that the result gets better but still unsatisfactory
as shown in Fig 5
Fig. 5. Angular velocity waveforms under
normal load operation
H C-PSS
4.1.4 with H/Sliding mode control
Finally we apply the developed controller which
is a combination of h innity/sliding mode con-
trol and the result is shown in g 6. It can be
seen that the system perform very well as the
time domain performance greatly improved.
Fig. 6. Angular velocity waveforms under
normal load operation
Proposed H/SMC PSS
4.2 Simulation test under normal
load operation
4.2.1 Simulation test under load variation
The dynamic simulation to show the robust
property of the proposed controller under load
variation (normal load and light load) is shown
in Fig. 5. It is shown that the proposed
H/SMC-PSS is less sensitive to load variation
(P = 0.7 and then P = 0.4).
Fig. 7. Angular velocity waveform without
any control
Fig. 8. Angular velocity waveforms
under normal load operation Proposed
H/SMC PSS
Fig. 9. Angular velocity waveforms for
proposed H/SMC under load variation
e Normal f 50 % vanation
4.3 Simulation under parameter vari-
ation of inertia moment coefcient M
The dynamic simulation result in Fig. 6 shows
the angular velocity waveform under parameter
variation of the inertia moment coefcient M (50
6
% overestimation). Also it is shown that the pro-
posed H/SMC-PSS is less sensitive to parameter
variation of the inertia moment coefcient M
(M=9.26 and to M= 13.89).
5 CONCLUSIONS
We have studied a new hybrid controller in
this work. Specically, an h innity/sliding
mode controller has been developed and
designed.Applied to PSS, it solved the problem
associated with unmeasurable state variable in
traditional sliding mode controller. Fig. 4 shows
the angular velocity response under normal load
operation (P=0.7). It is obvious that the system
response with the proposed H/SMC-PSS is
better than the case without any control with
the conventional PSS, and comparable to the
standard H-C-PSS. The proposed controller is
obtained with the value of the output vector by
measuring angular velocity only, and eliminates
the need to measure all the state variables in a
traditional SMC.
An H/sliding mode controller solved the
problem associated with unmeasurable state
variables in the traditional sliding mode con-
troller and to improved the time-domain perfor-
mance under a worst-case scenario. The stabil-
ity proof was achieved by Lyapunovs function
candidate using an additional form of sliding-
surface vector and estimation error. The linear
time-domain simulation in studying a small-
signal power system has shown satisfactory per-
formance under load variation (light load), and
under parameter variation of inertia moment
coefcient M (50% over estimation).
REFERENCES
UTKlN, V I.: Variable structure systems with
sliding modes, IEEE Trans. Autom. Control,
1977, AC-22, (2), pp. 212-222
ZAMES, G.: Feedback and optimal sensitivity:
model reference transformations, multiplica-
tive semi-norm, and approximate inverses,
IEEE Trons. Autom. Control, 1981, AC-26, pp.
301-320
DOYLE, J. C., GLOVER, K., KHARGONEKAR, P.
P., and FRANCIS, B. A.: State-space solutions
to standard H
2
and H control problems,
IEEE Trans.Autom.Control, 1989,34, (8), pp.
831-847
SHAHIAN, B., and HASSUL, M.: Control sys-
tem design using MATLAB (Prentice-Hall
Press, London, 1993)
ZHOU, K., DOYLE, J. C., and GLOVER, K.: Ro-
bust and optimal control (Prentice-Hall Press,
New Jeaey, 1996)
CHAN, W. C., and HSU, Y. Y.: An optimal
variable structure stabilizer for power system
stabilization, IEEE Trans. Power App. Syst.,
1983, PAS-102, pp. 1738-1746
KOTHARI, M. L., NANDA, J., and BHAT-
TACHARYA, K.: Design of variable structure
power system stabilisers with desired eigen-
values in the sliding mode, IEE Proe C. Gene?
Transm. Dislrib., 1993, 140, (4), pp. 263-268
LEE, S. S., and PARK, J. K.: Design of reduced-
order observer-based variable structure power
system stabiliser for unmeasurable state vari-
ables,IEEE Proc. Gener. Transm. Distrib., 1998,
145, (4), pp. 525-530
OHTSUKA, K.: et al. An H optimal control
theory-based generator control system, IEEE
Trans. Energy Conv., 1992, 7, (1), pp. 108-11
CHEN, S., and MALIK, 0. P.: H optimization-
based power system stabilizer design, IEE
Proc. Gener Transm. Distrib., 1995,142, (2), pp.
179184
AHMED, S. S., CHEN, L., and PETROIANU,
A.: Design of sub-optimal Hexcitation con-
trollers, IEEE Trans. Power Syst., 1996, 11,
(1),pp. 312-318
YU, Y. N.: Electric power system dynamics
(Academic Press, New York, 1983)
6 APPENDIX
6.0.1 Proof of theorem 1
Let us dene the estimation error equation and
the differential estimation error equation
e = x x
e = x

x //= Ax+B1W
worst
+B2 u
equal
WSMC
(t)
A x B2
u
equal
WSMC
(t) B1

W
worst
Z

K
e
(y y)
= (AZ

K
e
C2)x (AZ

K
e
C2) x
+B1W
worst
Z

K
e
D21W
worst
B1

W
worst
+Z

K
e
D21

W
worst
let x = e + x e = (A Z

K
e
C2)x (A
Z

K
e
C2)(e +x)
+B1
2
B1
t
X

x Z

K
e
D21
2
B1
t
X

x B
1

2
B1
t
X

x +Z

K
e
D
2
1

2
B1
t
X

x
= [A
2
B1B1
T
X

K
e
(C2
2
D21B1
T
X

)]e
Lyapunouves fuction candidate with the
addition of the sliding surface and the
7
estimation error is chosen
V = (1/2)
T
+ (1/2)e
T
e

V =
T
+e
T
e
=
T
(G
T

x) e
T
[A
2
B1B1
T
X

X(Z

K
e
(C2
2
D21B1
T
X

))]e
=
T
G
T
[A x + B2(K
equal
WSMC
zsign(( x(t))) +
B1

W
worst
+Z

K
e
C2x +Z

K
e
D21
W
worst
Z

K
e
C2 x Z

K
e
D21

W
worst
+Z

K
e
D22K
equal
WSMC
zsign(( x(t))] e
T
[A
2
B1B1
T
X

K
e
(C2
2
D21B1
T
X

)]e
let
W
worst
(t) =
2
.B1
T
X

x(t)
and

W
worst
(t) =
2
.B1
T
X

x(t)

V =
T
G
T
[A x +B2(K
equal
WSMC
xign(( x(t))) +B1
2
.B1
T
X

x(t) +Z

K
e
C2x +Z

K
e
D21

2
.B1
T
X

x Z

K
e
C2 x
Z

K
e
D
2
1
2
.B
T
1
X

x(t)e e
T
[A
2
B
1
B1
T
X

K
e
(C2
2
D21B1
T
X

)]e
=
T
G
T
[AB2K
equal
WSMC
sign(( x(t))
+B1
2
.B1
T
X

] x +
T
G
T
[Z

K
e
(C2x +Z

K
e
D21
2
.B1
T
X

)]e
e
T
[A
2
B1B1
T
X

K
e
(C2

2
D21B1
T
X

)]e
let K
equal
WSMC
:= (G
T
B2)
1
[G
T
(A+B1
T

2
]

V =
T
G
T
[AB2(G
T
B2)
1
[G
T
(A+B1
2
B1
T
X

)
sign(( x(t)) +B1
2
.B1
T
X

] x+

T
G
T
[Z

K
e
C2x +Z

K
e
D21
2
.B1
T
X

x Z

K
e
(C2 D21
2
.B1
T
X

)]e e
T
A[
2
B1B1
T
X

K
e
(C2
2
D21B1
T
X

)]e
=
T
G
T
A
[1 sign(( x(t))] x + [
T
G
T
B1
2
.B1
T
X

X(1 sign(( x(t))] x +

K
e
(C2
2
D21B1
T
X

)]e
e
T
[A
2
B1B1
T
X

K
e
(C2
2
D21B1
T
X

)]e]
the estimation error ise 0ast 0.

V =
T
G
T
A[1 sign(( x(t))] x
+[
T
G
T
B1
2
.B1
T
X

X
(1 sign(( x(t))] x 0
subject to,if >0,

V = 0 if = 0,

V = 0 if <0,

V
2KG
T
A x 2KG
T
B1
2
B1
T
X

x<0
k is positive constant. The condition is satised
on negative denite and is asymptotical stable.
This completes the proof of this theorem.
6.1 S
lidinghyperplane design in eqn. 23 From the equ.
22 x(t) = (A+B1(
2
.B1
T
X

))x(t) +B2u(t)
= A
d
x(t) +B2u(t)
whereA
d
= A + B1(
2
.B1
T
X

).The
transformed stateq(t)by the nonsingular
matrix M is given by
q(t) = Mx(t), x(t) = M
1
q(t), MB = [0 : B2]
T
whereM is thenxn nonsingular matrix,
andB2ismXm matrix. By differentiating eqn.
54, we get
q(t) = M x(t)
q(t) = MA
d
M
1
q(t) +MB2u(t)
By partitioningqsuch thatq = [q
1
q
2
]
T
whereq
1
is a
(n m)x1 column vector and q
2
is anx1 column
vector, eqn. 56 reduces to
_
_
q
1
(t)
.....
q
2
(t)
_
_
=
_
_
A
1
1 A
1
2
..... ....
A
2
1 A
2
2
_
_
_
_
q
1
(t)
.....
q
2
(t)
_
_
+
_
_
0
.....
B2
_
_
u(t)
where A
1
1 is (n m)x(n m), A
1
2is(n
m)xm, A
2
1 is mx(n m), and A
2
2ismxm. From
the rst part of eqn. 57,
q
1
(t) = A
1
1q
1
(t) +A
1
2q
2
(t)
From equation 23 and 54, (q(t)) = [G
1
1G
1
2]q(t)
= G
1
1q
1
(t) +G
1
2q
2
(t) = 0
= (x(t)) = G
T
M
1
q(t) = 0
Comparing eqn.59 with eqn. 61
[G
1
1G
1
2] = G
T
M
1
From eqn.60, q
2
(t) is obtained by
q
2
(t) = G
1
1
2G
1
1q
1
(t)
6.1.1 Use the desired eigenvalue assign-
ment
q
2
(t) = [A
1
1 A
1
2G
1
1]q
1
= A
c
q
1
If the pair(A
1
1, A
1
2) is controllable by a suitable
choice of the vectorG
1
1, the eigenvalues of ma-
trix A
c
may be placed arbitrarily in the complex
plane.
8
6.1.2 Determine sliding hyperplane
From eqn. 62, let G
1
2 = I, and
G
T
= [G
1
1I]M
Thus, the sliding-surface gain has been obtained.
9