NPTEL

Control Engineering
Sept.-Dec 2020
Solution: Assignment 12
Common Data for Q1-Q3
Through the first three questions, we will explore the controllability properties of the following state-space
canonical forms.

i) Diagonal form/ Modal canonical form

   
λ1 0 0 b1
ẋ =  0 λ2 0  x + b2  u
0 0 λ3 b3

ii) Jordan canonical form    

λ1 1 0 c1
ẋ =  0 λ1 1  x + c2  u
0 0 λ1 c3
iii) Controllable canonical form    
0 1 0 0
ẋ =  0 0 1  x + 0 u
a3 a2 a1 1

1. Which of the following options has the conditions for the Modal canonical form to be controllable?
(a) λi 6= 0 for i ∈ {1, 2, 3}
(b) Always controllable
(c) λi 6= 0 and bi 6= 0 for i ∈ {1, 2, 3}
(d) λi are distinct and bi 6= 0 for i ∈ {1, 2, 3} [Correct]
Solution : The determinant of the controllability matrix is −b1 b2 b3 (λ1 − λ2 )(λ2 − λ3 )(λ3 − λ1 ) =
0. Therefore, the modal canonical form is controllable if and only if the eigenvalues are distinct
(λ1 6= λ2 , λ2 6= λ3 and λ3 6= λ1 ) and bi 6= 0 for i ∈ {1, 2, 3}. Therefore, Option (d) is correct.
2. Which of the following options has the conditions for the Jordan canonical form to be controllable?
(a) c3 6= 0 [Correct]
(b) ci 6= 0 for i ∈ {1, 2, 3}
(c) λ1 6= 0 and ci 6= 0 for i ∈ {1, 2, 3}
(d) Always controllable
Solution : The controllability matrix of the Jordan Canonical form is
c1 c1 λ1 + c2 c1 λ21 + 2c2 λ1 + c3
 

C = c2 c2 λ1 + c3 c2 λ21 + 2c3 λ1  .

c3 c3 λ 1 c3 λ21
The determinant of the controllability matrix is −c33 . Therefore c3 6= 0 is a necessary and sufficient
condition for the Jordan canonical form to be controllable. Therefore Option (a) is correct.

1
3. Which of the following are conditions for the controllable canonical form to be controllable?
(a) a3 6= 0
(b) a1 6= 0
(c) ai 6= 0 for i ∈ {1, 2, 3}
(d) Always controllable [Correct]
Solution : The controllability matrix is
 
0 0 1
C = 0 1 a1  .
1 a1 a2 + a21

This matrix is always invertible, irrespective of the values taken by a1 , a2 and a3 , as the det(C) =
−1 6= 0. Therefore, the system is always controllable. This is the very reason for naming this
canonical form as controllable canonical form. Option (d) is correct.

4. Consider the LTI system

 
0 a1 0
ẋ = 0 0 a2  x + Bu, u ∈ R, a1 , a2 6= 0.
0 0 0

The minimum number of non-zero elements in the matrix B such that the above system is control-
lable is . [Correct answer : 1; Range 0.99 to 1.01]
Solution : Let [b1 b2 b3 ]> be the B matrix. The controllability matrix is
 
b1 b2 a1 b3 a1 a2
C = b2 b3 a2 0 .
b3 0 0

The determinant is −b33 a1 a22 . With respect to the matrix B, it suffices to have just b3 6= 0 for the
system to be controllable. Therefore, the minimum number of non-zero elements in the matrix B
which renders the given system controllable is 1.
5. Consider the following system
 
0 0 a3  
ẋ = 1 0 a2  x, y = 0 0 1 x.
0 1 a1

The system is observable for

(a) a1 = a3 6= 0, a2 6= 0
(b) a2 = 0
(c) a1 = 0, a2 = a3 6= 0
(d) Observable for all a1 , a2 , a3 [Correct]
Solution : This system is the observable canonical form for continuous-time LTI systems. This
is the dual of the controllable canonical form. In the previous assignment, we showed that the
controllable canical form is always controllable. By duality, we can conclude that the observable
canonical form is always observable. Therefore, Option (d) is correct.

2
 Just for verification, we can compute the observability matrix.
 
0 0 1
O = 0 1 a1  .
1 a1 a2 + a21
This matrix is always invertible as the det(O) = −1 6= 0. Therefore, the system is always observable
irrespective of the values taken by a1 , a2 and a3 . This is the very reason because of which this
canonical form is called the observable canonical form. Option (d) is correct.
6. Consider the following system
 
7 0 0  
ẋ = 0 1 0 x, y = c1 c2 c3 x
0 0 4
The system is observable for
(a) c1 6= 0, c3 6= 0, c2 6= 0 [Correct]
(b) c2 = 0
(c) c1 = 0, c2 = c3 6= 0
(d) unobservable for all c1 , c2 , c3
Solution : The system is in the diagonal canonical form. If the diagnonal entries are distinct, the
diagonal form is controllable when c1 6= 0, c3 6= 0, c2 6= 0.
By duality, we can conclude that, the given system is observable when c1 6= 0, c3 6= 0, c2 6= 0.
Therefore, Option (a) is correct.
Just to verify, we can compute the observability matrix.
 
c1 c2 c3
O =  7c1 c2 4c3  .
49c1 c2 16c3
The determinant of O is 54c1 c2 c3 . For the determinant to be non-zero c1 , c2 and c3 should be
non-zero. Therefore, the given system is observable when c1 6= 0, c3 6= 0, c2 6= 0. Option (a) is
correct.
Common data for Q7 - Q15
Consider the system ẋ = Ax + Bu where
   
1 0 1 1
A = −2 1 0 , B = 0 .
0 2 1 1
 
Using Ackerman’s formula, find a state-feedback controller K = k1 k2 k3 such that the control
law u = −Kx places the closed-loop poles at −5, −3, −1.
7. k1 = . [ans = 16, range = 15.9 to 16.1]
8. k2 = . [ans = -14, range = -14.1 to -13.9]
9. k3 = . [ans = -4, range = -4.1 to -3.9]
Using a similarity transformation, convert the system to its controllable canonical form denoted by
   
0 1 0 0
x̂˙ =  0 0 1  x̂ + 0 u
a1 a2 a3 1
Enter the following entries of the matrix.

3
10. a1 = . [ans = -3, range = -3.1 to -2.9]
11. a2 = . [ans = -3, range = -3.1 to -2.9]
12. a3 = . [ans = 3, range = 2.9 to 3.1]
 
Using the controllable canonical form obtained above, find a state-feedback controller K = k1 k2 k3
such that the control law u = −Kx places the closed-loop poles of the original system, i.e., eigen-
values of A − BK, at −6, −4, −2.
13. k1 = . [ans = 30, range = 29.9 to 30.1]
14. k2 = . [ans = -20.5, range = -20.6 to -20.4]
15. k3 = . [ans = -15, range = -15.1 to -14.9]
Solutions: The desired characteristic polynomial is

φ(s) = (s + 5)(s + 3)(s + 1) = s3 + 9s2 + 23s + 15.

Thus,
 
44 24 44
φ(A) = A3 + 9A2 + 23A + 15 = −88 44 −24 .
−48 88 44

Further, the controllability matrix is given by

 
1 2 3
A2 B = 0
 
C= B AB −2 −6 .
1 1 −3

By Ackerman’s formula, we have

1 C −1 φ(A) = 16 −14 −4
   
K= 0 0

Alternatively, you may use the MATLAB command acker.

To convert the system to the controllable canonical form, we use the procedure described in the
lectures. The characteristic polynomial of A is given by

det(sI − A) = s3 − 3s2 + 3s + 3.

Comparing with s3 + b1 s2 + b2 s + b3 , we have b1 = −3, b2 = 3, b3 = 3. With

 
b2 b1 1
W = b1 1 0 ,
1 0 0

we have
 
0 −1 1
P = CW =  0 −2 0 .
−3 −2 1

Using the transformation P x = x̂, we have

   
0 1 0 0
 = P −1 AP =  0 0 1 , B̂ = P −1 B = 0 .
−3 −3 3 1

4
 Note that the eigenvalues of A − BK are same as the eigenvalues of P −1 (A − BK)P = Â − B̂ K̂,
where we have defined K̂ = k̂1 k̂2 k̂3 = KP . Thus, it is enough to design a K such that
eigenvalues of  − B̂ K̂ are placed at −6, −4, −2.
     
0 1 0 0   0 1 0
 − B̂ K̂ =  0 0 1 − 0 k̂1 k̂2 k̂3 =  0 0 1 .
−3 −3 3 1 −3 − k1 −3 − k2 3 − k3

It is easy to verify that the characteristic polynomial of  − B̂ K̂ is given by s3 − (3 − k3 )s2 −

(−3 − k2 )s − (−3 − k1 ). Since we want the closed-loop poles to be placed at −6, −4, −2, the desired
characteristic polynomial is given by (s + 6)(s + 4)(s + 2) = s3 + 12s2 + 44s + 48. Comparing the
two characteristic polynomials, we have
 
K̂ = 45 41 15 .
Thus,
K = K̂P −1 = 30
 
−20.5 15 .

Common data for Q16- Q18

Consider the system ẋ = Ax + Bu, y = Cx with
   
1 −1 2 1  
A = 1 0 −1 , B = 2 , C = 1 0 0
1 1 3 1
Design a controller K such that the control input u = −K x̂ places the poles of A − BK at
{−2, −1 ± j}. x̂ is the estimated value of x obtained using the below observer design specifications:
(I) x̂ is the state of a full-order observer whose poles are located at {−5, −4, −3}.
16. Total number of state variables in the output-feedback (comprising the plant and the observer)
system defined by (I) are . [ans = 6, range = 5.99 to 6.01]
 
17. Let K = k1 k2 k3 be the controller gain. k3 = . [ans = 5.53, range = 5 to 6]
 
l1
18. Let L = l2  be the feedback-gain of the observer. l1 = . [ans = 16, range = 15.5 to 16.5]
l3
Solution: For a full-order observer described by Scheme (I), the dynamics of the system are given
by
    
ẋ A − BK BK x
=
ė 0 A − LC e
as derived in the lectures. Thus, the total number of state variables is 2n = 6.
Next, we design the controller gain and the observer gains for each of the schemes which place the
eigenvalues at the desired locations. Note that the design of the controller K is independent of the
design of the observer. Hence, the value of K is same for both the schemes. This value can be
computed in MATLAB using
K = acker(A,B,[-2 -1+i -1-i]).
The observer gains can be computed in a similar manner.
L = acker(A’,C’,[-5 -4 -3])’
 
  16
We get, K = 1.8112 0.3287 5.5315 and L =  0.9091 .
46.4545