Dr Lee Peng Hin : EE6203 Computer Control Systems 77 Dr Lee Peng Hin : EE6203 Computer Control Systems 78

7 Controller Design • Step 2: Design an observer to esti-

mate the system states from avail-
Consider the discrete time system,
able measurements (if states are not
x(k + 1) = Ax(k) + Bu(k) (7.1) available). See Figure 6.
y(k) = Cx(k) (7.2)

• Step 1 : Design a state feedback con-

troller (assuming the system states
are avail for feedback). See Figure 5.

Figure 6: Simplified presentation of the system and the observer

Figure 5: Closed loop system with a linear state feedback law

Dr Lee Peng Hin : EE6203 Computer Control Systems 79 Dr Lee Peng Hin : EE6203 Computer Control Systems 80

• The two steps are independent (Sep- 8 Controllability

aration Principle). Combine the two
Consider the discrete time system,
steps with the system states replaced
by their estimates. The structure of x(k + 1) = Ax(k) + Bu(k) (8.1)

the feedback control loop is shown y(k) = Cx(k) (8.2)

in Figure 7. where x(k) is an n-vector. Assume ini-

tial state x(0) is given. From (4.3), the
state at time n is,

x(n)
n−1

= Anx(0) + A(n−i−1)Bu(i)(8.3)
i=0
= Anx(0) + A(n−1)Bu(0) +
· · · + Bu(n − 1) (8.4)
= Anx(0) + WcU (8.5)
Figure 7: A feedback control system using measurement
Dr Lee Peng Hin : EE6203 Computer Control Systems 81 Dr Lee Peng Hin : EE6203 Computer Control Systems 82

where
 
Wc = B AB A2B . . . An−1B (8.6)
 T
U = u(n − 1) . . . u(0) (8.7)

Hence,

WcU = x(n) − Anx(0) (8.8)

Recall that Wc is called the controlla-

bility matrix. If Wc has rank n, then
it is possible to find the control sig-
nals such that the initial state is trans-
ferred to the desired final state x(n) in
at most n sampling periods.
Dr Lee Peng Hin : EE6203 Computer Control Systems 83 Dr Lee Peng Hin : EE6203 Computer Control Systems 84

Definition : The pair (A, B) is control-

lable if it is possible to find a control
sequence
 
u(0), u(1), u(2), . . . , u(n − 1) which
allows the system to reach an arbitrary
final state x(n) in at most n sampling
periods from any initial state x(0).

Theorem : The pair (A, B) is control-

lable if and only if

rank Wc = n

where n is the order of the system.

Dr Lee Peng Hin : EE6203 Computer Control Systems 85 Dr Lee Peng Hin : EE6203 Computer Control Systems 86

Example 8.1. Consider

   
1 0 1
A= , B= (8.9)
0 1 1
Investigate the controllability of the sys-
tem.
 
Wc = B AB (8.10)
 
11
= (8.11)
11
Since |Wc| = 0, the system is uncon-
trollable.
Dr Lee Peng Hin : EE6203 Computer Control Systems 87 Dr Lee Peng Hin : EE6203 Computer Control Systems 88

Example 8.2. Consider the system at the second instant is

     
1 1 0 0
A= , B= , x(0) = x(2) = 
Bu(1)
 + ABu(0)
  (8.14)
0 1 1 0
0 1
(8.12) = u(1) + u(0) (8.15)
1 1
Find a control sequence, if it exists,  
u(0)
= (8.16)
which can drive the system to the de- u(1) + u(0)
 T
sired final state ξ = 1 1.2 . We For ξ = x(2), we obtain
have
  u(0) = 1, u(1) = 0.2
0 1
Wc = (8.13)
1 1
Since |Wc| = 0, the system is con-
trollable and there exists a control se-
quence which can drive the system to
the desired final state. The response of
Dr Lee Peng Hin : EE6203 Computer Control Systems 89 Dr Lee Peng Hin : EE6203 Computer Control Systems 90
Dr Lee Peng Hin : EE6203 Computer Control Systems 91 Dr Lee Peng Hin : EE6203 Computer Control Systems 92
Dr Lee Peng Hin : EE6203 Computer Control Systems 93 Dr Lee Peng Hin : EE6203 Computer Control Systems 94
Dr Lee Peng Hin : EE6203 Computer Control Systems 95 Dr Lee Peng Hin : EE6203 Computer Control Systems 96

Multivariable systems

When there are more than one inputs,

system is completely controllable if all
the states are controllable by at least
one of the inputs.
Example. Is the following two-input
system completely controllable ?

⎡ ⎤
0.5 0.5 0
⎢ ⎥
x(k + 1) = ⎣ 0 1 0 ⎦ x(k) +
0.83 −2.16 −0.33
⎡ ⎤
3 1
⎢ ⎥
⎣ 2 0 ⎦ u(k)
−1 1
Dr Lee Peng Hin : EE6203 Computer Control Systems 97 Dr Lee Peng Hin : EE6203 Computer Control Systems 98

Soln : 9 Observability
⎡ ⎤ ⎡ ⎤
2.25 0.5 2.25 0.25
⎢ ⎥ 2 ⎢ ⎥ • Observability is a property of a sys-
AB = ⎣ 2 0 ⎦;A B = ⎣ 2 0 ⎦
−1.5 0.5 −1.75 0.25 tem which is strongly related to the
The controllability matrix, ability to determine the initial state
 
Wc = B AB A B 2 x(0) on the basis of the input and
⎡ ⎤
3 1 2.5 0.5 2.25 0.25 output data.
⎢ ⎥
= ⎣ 2 0 2 0 2 0 ⎦
• Consider the system,
−1 1 −1.5 0.5 −1.75 0.25
rank of Wc is at most 3. x(k + 1) = Ax(k), x(0) = xo(9.1)
R1 − R3 = 2R2. Hence, the three y(k) = Cx(k) (9.2)
rows at not linearly independent and
• Assume that y(0), . . . , y(k − 1) are
rank Wc is at most 2, so system is not
given for a certain k. To simplify
completely controllable.
derivations, assume that u(k) = 0.
Dr Lee Peng Hin : EE6203 Computer Control Systems 99 Dr Lee Peng Hin : EE6203 Computer Control Systems 100

• • Condition (9.4) holds true if and only

y(0) = Cx(0) if k = n, i.e.
⎡ ⎤
y(1) = Cx(1) = CAx(0) C
⎢ ⎥
.. ⎢ CA ⎥
rank Wo = rank ⎢
⎢ ..
⎥=n
⎥
⎣ ⎦
y(k − 1) = CAk−1x(0) CAn−1
or ⎡ (9.5)
⎤ ⎡ ⎤
C y(0)
⎢ ⎥ ⎢ ⎥ • Wo is known as the observability ma-
⎢ CA ⎥ ⎢ y(1) ⎥
⎢ ⎥ x(0) = ⎢ ⎥
⎢ .. ⎥ ⎢ .. ⎥ trix.
⎣ ⎦ ⎣ ⎦
CA k−1 y(k − 1)
(9.3)
x(0) can be obtained from (9.3) if
and ⎡only if ⎤
C
⎢ ⎥
⎢ CA ⎥
⎢ ⎥ has rank n. (9.4)
⎢ .. ⎥
⎣ ⎦
CA k−1
Dr Lee Peng Hin : EE6203 Computer Control Systems 101 Dr Lee Peng Hin : EE6203 Computer Control Systems 102

• Definition : The pair (A, C) is ob-

servable if there exists a finite N such
that the knowledge of the inputs
u(0), u(1), . . . , u(k − 1) and the out-
puts y(0), y(1), . . . , y(k − 1) is suffi-
cient to determine the initial states
x(0) of the system.

• Theorem : The pair (A, C) is observ-

able if and only if

rank Wo = n

where n is the order of the system.

Dr Lee Peng Hin : EE6203 Computer Control Systems 103 Dr Lee Peng Hin : EE6203 Computer Control Systems 104
Dr Lee Peng Hin : EE6203 Computer Control Systems 105 Dr Lee Peng Hin : EE6203 Computer Control Systems 106
Dr Lee Peng Hin : EE6203 Computer Control Systems 107 Dr Lee Peng Hin : EE6203 Computer Control Systems 108

Example 9.1. Consider

 
1.1 −0.3  
A= , C = 1 −0.5
1 0
(9.6)
Investigate the observability of the sys-
tem.
 
C
Wo = (9.7)
CA
 
1 −0.5
= (9.8)
0.6 −0.3
Since |Wo| = 0, the system is unob-
servable.
Dr Lee Peng Hin : EE6203 Computer Control Systems 109 Dr Lee Peng Hin : EE6203 Computer Control Systems 110

Example 9.2. Consider

 
1 0  
A= , C= 01 (9.9)
1 1
The output sequence of the system is
{y(0), y(1)} = {1, 1.2}. Find the ini-
tial state if u(0) = u(1) = 0 .
 
0 1
Wo = (9.10)
1 1
Since |Wo| = 0, the system is observ-
able. Hence, the initial conditions can
be determined. From (9.3),
    
0 1 x1(0) 1
=
1 1 x2(0) 1.2
which gives

x1(0) = 0.2, x2(0) = 1

Dr Lee Peng Hin : EE6203 Computer Control Systems 111 Dr Lee Peng Hin : EE6203 Computer Control Systems 112
Dr Lee Peng Hin : EE6203 Computer Control Systems 113 Dr Lee Peng Hin : EE6203 Computer Control Systems 114
Dr Lee Peng Hin : EE6203 Computer Control Systems 115 Dr Lee Peng Hin : EE6203 Computer Control Systems 116
Dr Lee Peng Hin : EE6203 Computer Control Systems 117 Dr Lee Peng Hin : EE6203 Computer Control Systems 118

10 Loss of controllability/observability Example 10.1. The SS model of a har-

due to sampling monic oscillator with transfer funuc-
2
tion H(s) = 2ω 2 is
s +ω
• When sampling a continuous time    
• 0 ω 0
system, the resulting discrete-time x(t) = x(t) + u(t)
−ω 0 ω
system matrices depend on the sam-  
y(t) = 1 0 x(t)
pling period T .
Investigate the controllability and ob-
• For a discretized system to be con-
servability of the discrete time system
trollable, it is necessary that the ini-
whose states are sampled with sampling
tial continuous-time system be con-
time T .
trollable.
The discrete time model is
 
• Controllability (and observability) may cos ωT sin ωT
x(k + 1) = x(k)
be lost for certain values of the sam- − sin ωT cos ωT
 
pling period. 1 − cos ωT
+ u(k)(10.1)
sin ωT
 
y(k) = 1 0 x(k) (10.2)
Dr Lee Peng Hin : EE6203 Computer Control Systems 119 Dr Lee Peng Hin : EE6203 Computer Control Systems 120

and

|Wc| = − sin ωT (1 − cos ωT )

|Wo| = sin ωT

Hence, the controllability and observ-

ability of the discrete time system is
lost when ωT = qπ, where q is an in-
teger, although the respective contin-
uous time system is both controllable
and observable.
Dr Lee Peng Hin : EE6203 Computer Control Systems 121 Dr Lee Peng Hin : EE6203 Computer Control Systems 122
Dr Lee Peng Hin : EE6203 Computer Control Systems 123 Dr Lee Peng Hin : EE6203 Computer Control Systems 124

11 Pole placement Problem formulation :

Design a state feedback controller,

The linear state feedback controller shown
i.e. find a vector K in (11.1) such
in Figure 5 is expressed as
that the poles of the closed loop
u(k)
system (11.3) take on desirable pre-
= −Kx(k) ⎡ ⎤(11.1) assigned values.
x1(k)
 ⎢
 ⎢ x (k) ⎥
⎥ The characteristic polynomial of the closed
= − k1 k2 . . . kn ⎢ ⎥
2
⎢ .. ⎥(11.2)
⎣ ⎦ loop system is
xn(k)
α(z) = det [zI − A + BK] (11.4)
The above controller is called a regula-
= z n + β̂1z n−1 + . . .
tor since the reference input r(k) = 0.
Sub (11.1) into (7.1), the closed loop +β̂n−1 z + β̂n (11.5)

system is given by the homogeneous Suppose that the closed loop poles are
equation, desired to be at,

x(k + 1) = [A − BK]x(k) (11.3) pi = p1, p2, . . . , pn (11.6)

Dr Lee Peng Hin : EE6203 Computer Control Systems 125 Dr Lee Peng Hin : EE6203 Computer Control Systems 126

Then the desired closed loop charac- Q : Under what circumstances will there
teristic polynomial can be formed as exists a state feedback matrix gain K

αc(z) such that the poles of the closed loop

system
= (z − p1)(z − p2) . . . (z − pn(11.7)
)
= z n + β1z n−1 + · · · + βn−1z x(k + 1) = [A − BK]x(k) (11.9)

+βn (11.8) can be placed arbitrary.

Equate Ans : If the pair (A, B) is controllable.

(11.5) ≡ (11.8)

and solve for K. This design method is

known as pole placement method.
Dr Lee Peng Hin : EE6203 Computer Control Systems 127 Dr Lee Peng Hin : EE6203 Computer Control Systems 128

Example 11.1. Consider the discrete-time stant (τ = −T

ln r ) of 1sec for T = 0.1sec.
SS model of the servo motor obtained
x1(k) = position (angle) of motor shaft
in Example 2.1.
  which can be measured
1 0.0952
x(k + 1) = x(k) + x2(k) = shaft velocity, can also
0 0.905
 
0.00484 be measured
u(k) (11.10)
0.0952
  The controllability matrix
y(k) = 1 0 x(k) (11.11)  
Wc = B AB (11.12)
Design a state feedback controller so  
0.00484 0.0139
that the closed loop poles are at = (11.13)
0.0952 0.0862
0.888 ± j0.173 = 0.905 ±0.193 rad Since |Wc| = 0, the system is control-
 
lable. For K = k1 k2 ,
These poles will give a damping ratio
(δ = √ −2ln r ) of 0.46 and a time con-  − A + BK] =
[zI 
ln r+θ 2
z − 1 + 0.00484k1 −0.0952 + 0.00484k2
0.0952k1 z − 0.905 + 0.0952k2
Dr Lee Peng Hin : EE6203 Computer Control Systems 129 Dr Lee Peng Hin : EE6203 Computer Control Systems 130

Thus, α(z) in (11.5), The controller is given by

 
  x (k)
1
α(z) = z 2 +(0.00484k1 +0.0952k2−1.905)z u(k) = − 4.52 1.12
x2(k)
+ 0.00468k1 − 0.0952k2 = −4.52x1(k) − 1.12x2(k)
+ 0.905 (11.14)

and the desired poles polynomial αc(z)

in (11.8),

αc(z) = z 2 − 1.776z + 0.819 (11.15)

Equating (11.14) with (11.15),

0.00484k1 + 0.0952k2 − 1.905 = −1.776

0.00468k1 − 0.0952k2 + 0.905 = 0.819

Solving these equations,

k1 = 4.52, k2 = 1.12
Dr Lee Peng Hin : EE6203 Computer Control Systems 131 Dr Lee Peng Hin : EE6203 Computer Control Systems 132
Dr Lee Peng Hin : EE6203 Computer Control Systems 133 Dr Lee Peng Hin : EE6203 Computer Control Systems 134
Dr Lee Peng Hin : EE6203 Computer Control Systems 135 Dr Lee Peng Hin : EE6203 Computer Control Systems 136
Dr Lee Peng Hin : EE6203 Computer Control Systems 137 Dr Lee Peng Hin : EE6203 Computer Control Systems 138

Example 11.2. A system is modeled as

   
• −1 0 1
x(t) = x(t) + u(t)
1 0 0
 
y(t) = 1 0 x(t)

Sampling the above system using a ZOH

with sampling period T secs gives the
discrete time model as

x(k + 1) = Φ(T )x(k) + Θ(T )u(k)

 
y(k) = 1 0 x(k)

where
 
e−T 0
Φ(T ) =
1 − e−T 1
 
1 − e−T
Θ(T ) =
T − 1 + e−T
Now, design a state feedback controller
Dr Lee Peng Hin : EE6203 Computer Control Systems 139 Dr Lee Peng Hin : EE6203 Computer Control Systems 140

of the form   Equating (11.17) and (11.18),


 x (k)
1
u(k) = − k1 k2 (11.16) − e−T + (1 − e−T )k1 − 1
x2(k)
so that the closed loop poles are at + (T − 1 + e−T )k2 = p1 + p2 (11.19)

−p1 and − p2 and

The closed loop characteristics polyno- (−e−T +(1−e−T )k1)(−1+(T −1+e−T )k2)
mial is −(1−e−T )k2(−(1−e−T )+(T −1+e−T )k1)
α(z) = z 2 + (−e−T + (1 − e−T )k1 − 1+ = p1p2 (11.20)
(T − 1 + e−T )k2)z
Solving (11.19) and (11.20) gives k1 and
+ (−e−T + (1 − e−T )k1) k2. Obviously, if the system order is
(−1 + (T − 1 + e−T )k2) more than 2, solving these two equa-
− (1 − e−T )k2(−(1 − e−T ) tions is rather tedious.
+ (T − 1 + e−T )k1) (11.17) But if the state space model is in the
The desired closed loop polynomial is controllable canonical form (Ac, Bc), then

αc(z) = z 2 + (p1 + p2)z + p1p2 (11.18) the problem of finding K is much sim-
Dr Lee Peng Hin : EE6203 Computer Control Systems 141 Dr Lee Peng Hin : EE6203 Computer Control Systems 142

pler. In this case, and

⎡ − BK =
A ⎤
x(k
⎡ + 1) ⎤ 0 1 0 0 ... 0
⎢ ⎥
⎢ 0 0 1 0 ... 0 ⎥
⎢ ⎥
0 1 0 0 ... 0 ⎢ ⎥
⎢ ⎥ ⎢
⎢
0 0 0 1 ... 0 ⎥
⎥
⎢ 0 0 1 0 ... 0 ⎥ ⎢ ..
.
..
.
..
.
..
. ... ..
. ⎥
⎢ ⎥ ⎢
⎢
⎥
⎥
⎢ 0 ⎥ ⎣ 0 0 0 0 ... 1 ⎦
⎢ 0 0 1 ... 0 ⎥ −(a0 + k1 ) −(a1 + k2 ) −(a2 + k3 ) −(a3 + k4 ) . . . −(an−1 + kn)
=⎢ . .. .. .. . ⎥ x(k)
⎢ . ... . ⎥ so that
⎢ ⎥
⎢ 0 0 0 0 ... 1 ⎥
⎣ ⎦
α(z) = |zI − A + BK|
−a0 −a1 −a2 −a3 . . . −an−1
⎡ ⎤
0 = z n + (an−1 + kn)z n−1 + · · · +
⎢ ⎥
⎢0⎥ (a1 + k2)z + (a0 + k1)
⎢ ⎥
+⎢ .⎥
⎢ . ⎥ u(k)
⎢ ⎥ The desired closed loop characteristic
⎣0⎦
1 polynomial is
The characteristic equation is αc(z) = (z − p1)(z − p2) . . .
|zI − A| = z n + an−1z n−1 (z − pn) (11.22)
+ · · · + a1z + a0 = 0 (11.21) = z n + β1z n−1 + · · · + βn−1z
+βn (11.23)
Dr Lee Peng Hin : EE6203 Computer Control Systems 143 Dr Lee Peng Hin : EE6203 Computer Control Systems 144

and the gains can be calculated by equat- 12 Ackermann’s formula

ing (11.21a) and (11.23) to yield
In general, a plant need not be rep-
ki+1 = βn−i − ai; i = 0, 1, . . . n − 1
resented by a model in CCF. A more
(11.24)
practical procedure for calculating K is
by use of Ackermann’s formula (proof
omitted).
Let the desired characteristic polyno-
mial be

αc(z) = z n + β1z n−1 + . . .

+ βn−1z + βn (12.1)

The controller K can be computed as

 
K = 0 . . . 0 1 Wc−1αc(A) (12.2)

where Wc is the controllability matrix

Dr Lee Peng Hin : EE6203 Computer Control Systems 145 Dr Lee Peng Hin : EE6203 Computer Control Systems 146

and Example 12.1. Consider the servo motor

design example in Example 11.1 where
αc(A) = An + β1An−1  
1 0.0952
+ · · · + βn−1A + βnIn (12.3) x(k + 1) = x(k)
0 0.905
 
(12.2) is derived by Ackermann in 1972. 0.00484
+ u(k)(12.4)
0.0952
We have,

αc(z) = z 2 − 1.776z + 0.819 (12.5)

Thus,
 2  
1 0.0952 1 0.0952
αc(A) = − 1.776
0 0.905 0 0.905
+0.819I
 2  (12.6)
0.043 0.01228
= (12.7)
0 0.03075
Also,
 
 −1 −95.13 15.34
Wc−1 = B AB =
105.1 −5.342
Dr Lee Peng Hin : EE6203 Computer Control Systems 147 Dr Lee Peng Hin : EE6203 Computer Control Systems 148

So, from (12.2), 13 Deadbeat control

   
−1
K = 0 1 Wc αc(A) = 4.52 1.12 The solution x(k) of the closed loop
(12.8) system
The answer is the same as that ob-
x(k + 1) = [A − BK]x(k) (13.1)
tained in Example 11.1.
is
In MATLAB,
x(k) = [A − BK]kx(0) (13.2)
K = acker(A, B, p); for ≤ 10 state variables
• If eigenvalues are such that
K = place(A, B, p); more general
|λi[A − BK]| < 1 ∀i = 1, . . . , n, then
x(k) → 0 as k → ∞, i.e. stable.

• If

λi[A − BK] = 0, ∀ i = 1, . . . , n
(13.3)
then

– stable.
Dr Lee Peng Hin : EE6203 Computer Control Systems 149 Dr Lee Peng Hin : EE6203 Computer Control Systems 150

– x(k) becomes 0 for k ≤ n, i.e. acteristic polynomial is

closed loop system will arrive and αc(z) = z n
remain at rest in at most n sam- = det [zI − A + BK](13.4)
pling periods, after an impulse dis-
– This is always possible provided
turbance in the process state.
that (A, B) is controllable.
– this type of control is called dead-
• Then, by Ackermann’s formula,
beat control.  
K = 0 . . . 0 1 Wc−1An (13.5)
• Deadbeat control is a phenomenon
of discrete time systems. No such To show that the response to any ini-
equivalent phenomenon for continu- tial state x(0) is deadbeat when (13.3)
ous time systems. holds :

• To achieve deadbeat control, must Consider a controllable system

determine K such that λi[A − BK] = x(k + 1) = Ax(k) + Bu(k) (13.6)

0 ∀i = 1, . . . , n, i.e. the desired char- then, there exists a similarity trans-
formation matrix that transforms the
Dr Lee Peng Hin : EE6203 Computer Control Systems 151 Dr Lee Peng Hin : EE6203 Computer Control Systems 152

above into the controllable canonical fies,

 
form (CCF), KP = k1 k2 . . . kn
 
w(k) = P−1x(k) (13.7) = −a0 −a1 . . . −an−1

where

w(k + 1) = Acw(k) + Bcu(k)

and (Ac, Bc) is in CCF. Using state

feedback,

u(k) = −Kx(k)
= −KPw(k)

Then,

w(k + 1) = (Ac − BcKP)w(k)

From (11.24) and since Ac is in CCF

and αc(z) = z n, the matrix KP satis-
Dr Lee Peng Hin : EE6203 Computer Control Systems 153 Dr Lee Peng Hin : EE6203 Computer Control Systems 154

So, which is a nilpotent matrix. (If there

is a positive k such that Ak = 0, then
⎡ c − BcKP)
(A ⎤
0 1 0 0 ... 0 A is called nilpotent). So,
⎢ ⎥
⎢ 0 0 1 0 . . . 0 ⎥
⎢ ⎥
⎢ 0 ⎥ (Ac − BcKP)n = 0
⎢ 0 0 1 ... 0 ⎥
=⎢ . .. .. .. . . . .. ⎥
⎢ . ⎥
⎢ ⎥ In terms of the original state,
⎢ 0 0 0 0 ... 1 ⎥
⎣ ⎦
−a0 −a1 −a2 −a3 . . . −an−1 x(n) = [A − BK]nx(0)
⎡ ⎤
0 = [PAcP−1 − PBcK]nx(0)
⎢ ⎥
⎢0⎥  = [P(Ac − BcKP)P−1 ]nx(0)
⎢ ⎥
⎢ . ⎥
− ⎢ . ⎥ −a0 −a1 . . . −an−1
⎢ ⎥ = 0
⎣0⎦
1
⎡ ⎤
0 1 0 0 ... 0
⎢ ⎥
⎢0 0 1 0 ... 0⎥
⎢ ⎥
⎢0 0 0 1 ... 0⎥
⎢ ⎥
=⎢. . . . . ⎥
⎢ . . . . . . .. ⎥
⎢ ⎥
⎢0 0 0 0 ... 1⎥
⎣ ⎦
0 0 0 0 ... 0
Dr Lee Peng Hin : EE6203 Computer Control Systems 155 Dr Lee Peng Hin : EE6203 Computer Control Systems 156

Example 13.1. Consider a double-integrator crete time SS model as,

 
system, 1 T
x(k + 1) = x(k) +
1 d2 y(t) 
0 1

Y (s) = 2 U (s) or = u(t) 2
s dt T /2
u(k) (13.8)
For T
     
x1(t) y(t) y(k) = 1 0 x(k) (13.9)
x(t) = = •
x2(t) y(t)
Determine a vector K for deadbeat con-
the continuous time SS model is
    trol. Determine u(0) and u(1) for var-
• 0 1 0
x(t) = x(t) + u(t)
0 0 1 ious values of T with initial conditions
   T
y(t) = 1 0 x(t) x(0) = 1 1 . We have,
 
 −1 1 3
− T 2 2T
Sampling the system with a ZOH with Wc−1 = B AB = 1 1
sampling period T , we obtained the dis- 2 − 2T T
So, from (13.5),
   
−1
K = 0 1 Wc A = 2 1 3
T 2 2T
(13.10)
Dr Lee Peng Hin : EE6203 Computer Control Systems 157 Dr Lee Peng Hin : EE6203 Computer Control Systems 158

and k = 0,
   
1 T 1 3
[A − BK] = 2 4 u(0) = −Kx(0) = − +
− T − 12
1 T2 2T
(13.12)
Note that [A − BK]q = 0 for q > 1.
k = 1,
Therefore, x(k) = 0 for k > 1 which  
1 1
u(1) = −Kx(1) = + (13.13)
means that the system will come to a T2 2T
standstill at the second sampling in- Using (13.12) and (13.13), we can sum-

stant. marises u(0) and u(1) for various val-

From (4.3), ues of T as :

Table 1. Values of u(0) and u(1) for various T
x(k) = [A − BK]kx(0) (13.11)
T 2 1 0.5
for k = 1, u(0) -1 -2.5 -7
 
1 T u(1) 0.5 1.5 5
x(1) = 2+ 4
− 12 − T1 Figure 8 shows the trajectories of the

and from output and input u(kT ) for T = 0.5, 1

 T
and 2 secs when x(0) = 1 1 .
u(k) = −Kx(k)
Observations from Figure 8.
Dr Lee Peng Hin : EE6203 Computer Control Systems 159 Dr Lee Peng Hin : EE6203 Computer Control Systems 160

• Smaller the value of T , the bigger

the value of u(kT ) and vice versa.

– i.e. the faster we desired the re-

sponse to settle to zero, the greater
the control effort.

• It is thus important to choose the

sampling period carefully when us-
ing deadbeat control.

– e.g. select T to satisfy the Shan-

non sampling criterion while keep-
ing the magnitude of u(kT ) within
acceptable limits.

Figure 8: Response for deadbeat control in Example 13.1

Dr Lee Peng Hin : EE6203 Computer Control Systems 161 Dr Lee Peng Hin : EE6203 Computer Control Systems 162
Dr Lee Peng Hin : EE6203 Computer Control Systems 163 Dr Lee Peng Hin : EE6203 Computer Control Systems 164