EE550: Linear System Theory

End-Semester Examination Solution

Date: 23-11-2024 Marks: 60

Justifications are required for all answers to earn marks.

1. Answer the following questions: (5 × 2 = 10)
   
−2 1 0  
(a) Consider the system ẋ(t) = x(t) + u(t), y(t) = 1 0 x(t). Find the
0 −2 1
output response of its dual system for initial condition (1, 1) and step input.
Solution: Output response of the dual system is
Z ∞
T AT t T T
y(t) = B e x̄(0) + B eA (t−τ ) C T u(τ )dτ
0
−2t −2t
   −2t 
At e te AT t e 0
Since e = , we have e = . The response is:
0 e−2t te−2t e−2t
 −2t   Z t
e−2(t−τ )
  
  e 0 1 0 1
y(t) = 0 1 + dτ
te−2t e−2t 1 0 (t − τ )e−2(t−τ )
e −2(t−τ )
0
−2t
e−2t (1 − e−2t )/2
     1

    + e2
= 0 1 + 1 1 −2t t −2t = 0 1 1 3 −2t t −2t 2
te−2t + e−2t 4
− 4e − 2e 4
+ 4e + 2e
1 3 t
= + e−2t + e−2t
4 4 2
s+1
(b) Find a minimal realization of the transfer function G(s) = 2 .
s +s
Solution: Minimal realization is not possible as the numerator and denominator
are not coprime.
 
−1 1
(c) Consider the discrete-system x[k + 1] = Ax[k], where A = . Find x[100]
0 −1
using Caley-Hamilton theorem, if x[0] = (−2, 2).
Solution: x[100] = A100 x[0]. Note that
s100 = (s + 1)2 q(s) + α1 s + α0 , (1)
where (s + 1)2 is the characteristic polynomial of A. Evaluating eq. (1) and the
differentiation of eq. (1) at s = −1, we have
(−1)100 = −α1 + α0
100 × (−1)99 = α1 ⇒ α1 = −100 ⇒ α0 = −99.
Therefore, s100 = (s + 1)2 q(s) − 100s − 99. Using Cayley Hamilton’s theorem, we
know that (A + I)2 = 0. Therefore,
 
100 2 100 1 −100
A = (A + I) q(A) − 100A − 99I ⇒ A = −100A − 99I =
0 1
    
100 1 −100 −2 −202
Therefore, x[100] = A x[0] = =
0 1 2 2

1
 (d) Define the controllability matrix C of a system as shown below
   
−1 1 4 0 −1 1
C :=
1 1 0 0 1 1

Find a basis of the reachable space of the system.

Solution: Note that the controllability matrix is in the form of C = U ΣV T ,
hence
−1
from inspection we have a basis of the reachable space, i.e., img(C), is .
1
(e) For an LTI system, show that (A, B) is controllable if and only if (−A, B) is con-
trollable.
Solution:
 
C(−A,B) = B −AB · · · (−1)n−1 An−1 B
 
Im 0 ··· 0
   0 −Im
 ··· 0 
= B AB · · · An−1 B  .

.. .. ..
 ..

. . . 
0 0 · · · (−1)n−1 Im
| {z }
T

Therefore, C(−A,B) = C(A,B) T , where T is a square matrix of suitable dimension.

Since T is a diagonal matrix with only
1 and −1 in diagonal,
T is non-singular.
Therefore, rank (C)(−A,B) = rank C(A,B) . Hence, rank C(−A,B) = n if and only if


rank C(A,B) = n. This proves the statement.

2. Imagine you have designed a system governed by the following dynamical equations: (5)
        
x1 (t) −1 −1 0 x1 (t) 1 x 1 (t)
d   
x2 (t) =  0 −2 0 x2 (t) + 1 u(t), y(t) = 1 0 1 x2 (t)
dt
x3 (t) 0 0 −3 x3 (t) 1 x3 (t)

To test the system, you applied zero input, expecting a zero output as well. However,
to your surprise, the system’s output was the exponential function e−t + e−2t . This
unexpected behavior left you curious: What could have been the initial condition of the
system to produce this intriguing output?
Can you uncover the answer and find the initial state that made this possible?
Solution: Note that from the output we have

y(0) = 2, ẏ(0) = −e−t − 2e−2t |t=0 = −3, ÿ(0) = e−t + 4e−2t |t=0 = 5.

Then we must have

         
C y(0) 1 0 1 2 2
 CA  x(0) = ẏ(0) ⇒ −1 −1 −3 x(0) = −3 ⇒ x(0) = 1
CA2 ÿ(0) 1 3 9 5 0

3. Consider the following system

 
1 0 0 0
 0 2 1 0
ẋ(t) = 
 0
 x(t) + Bu(t).
0 2 0
0 0 0 −1
| {z }
A

2
 Find B ∈ R4×3 such that the elements of B are chosen from the set {0, 1}, and each column
of B has a one-norm equal to 1. Additionally, ensure that the pair (A, B) is controllable.
Using the identified B, assign the eigenvalues of the system to {−1, −1, −1, −1}. (8)
Solution: Since elements of B are binary and each column needs to have one-norm 1,
hence each column of B must have 1 in one place and zero elsewhere. We have to choose
the following B to make the system controllable.
 
1 0 0
0 0 0
B= 0 1 0


0 0 1

Thus, the system can be decoupled as

      
x1 1 0 0 0 x1 1 0 0  
u1
d  x
 2 =
  0 2 1 0 

 x2   0
+ 0 0  u2 
dt x3
   0 0 2 0   x3   0 1 0 
u3
x4 0 0 0 −1 x4 0 0 1
   
2 1 0
The decoupled systems are (A1 , B1 ) := (1, 1), (A2 , B2 ) := , , and (A3 , B3 ) =
0 2 1
(−1, 1). We design controller for each sub-system.

For (A1 , B1 ):
ẋ1 = x1 + u1 . Choose K1 = −2 and control law u1 = K1 x1 .
For
 (A2 ,B2 ):     
ẋ2 2 1 x2 0
= + u.
ẋ3 0 2 x3 1 2
      
0 1 1 −4 0 1 −2 −1
Let T = = . The desired matrix Ad = .
1 2 0 1 1 −2   1 0
Here Ab2 = T −1 A2 T = 4 −4 , B b = T −1 B2 = 1 .
1 0 0
b 2 T −1 = −9 −6 with control
   
Solving for Ad =A2 + B2 K2 ⇒ K2 = −6 3 ⇒ K2 = K
b b b b
x
law u2 = K2 2 .
x3
For (A3 , B3 ):
ẋ3 = −x1 + u3 . Choose K3 = 0 and control law u3 = K3 x3 .
Therefore,  
    x1
u1 −2 0 0 0  
x2 
u = u2  =  0 −9 −6 0  x3 
u3 0 0 0 0
x4

4. Consider the following system: (2 + 5 + 2 + 2= 11)

        
x1 (t) −2 0 0 0 x1 (t) 1 x1 (t)
d 
 2 = 4
x (t)  3 0 0 x2 (t) 0   x2 (t)
  +   u(t), y(t) = 2 1 0 0  
dt x3 (t)  1 0 3 0 x3 (t) 1 x3 (t)
x4 (t) 3 1 4 −2 x4 (t) 0 x4 (t)

Answer the following questions (justification required for answer)

3
 (a) Identify the controllable and uncontrollable eigenvalues of the system.
Solution:
   
0 0 0 0 1 −5 0 0 0 1
  4 5 0 0 0     4 0 0 0 0
A + 2I B =  1 0 5 0 1 
, A − 3I B = 
 1 0 0

0 1
3 1 4 0 0 3 1 4 −5 0
   
Here rank A + 2I B = 4 and rank A − 3I B = 3. Therefore, controllable
eigenvalues are {−2, −2, 3} and uncontrollable eigenvalue is 3.
(b) Design a state-estimator that settles to 98% of the actual states within 1 seconds
with no oscillations.
Solution: Note that  (A,C) is already in the standard block diagonal form where
3 0
the eigenvalues of are unobservable. Since 3 is unobservable, the system is
4 −2
not detectable. Hence, it is not possible to design an observer such that the states
and estimated states converge.
 T
(c) Is it possible to find a control input that steers the initial condition −2 4 4 7
 T
to 0 0 0 0 within any finite time. Provide justification for your answer.
Solution: If the initial condition is in the reachable space, then it can be steered to
 T
0 0 0 0 . Here
 
1 −2 4 −8
  0 4 4 28
CA,B = B AB A2 B A3 B =  1

4 10 34
0 7 0 56
 T  T
Since −2 4 4 7 ∈ img(CA,B ), the initial condition can be steered to 0 0 0 0 .
 T
(d) Is the state 0 0 −12 1 observable?
Solution: The observability matrix is
 
2 1 0 0
 0 3 0 0
OA,C = 12 9 0 0


12 27 0 0
 T
Since 0 0 −12 1 ∈ ker(OA,C ), the state is not observable.

5. Consider the system in Figure (1) with states (x1 , x2 , x3 ). Find the input u(t) such that
the system reaches equilibrium from (0, 0, 1) within 2 seconds. (7)
Solution: State-space representation of the system from the block diagram is:
      
x1 −2 0 0 x1 0
d   
x2 = 0 −1 0 x2  + 0 u.
dt
x3 0 1 −1 x3 1

4
 Figure 1: Block diagram of a system

The controllability Gramian is:

Z 2
T
Wc (2) = eAt BB T eA t dt
0
Z 2 e−2t 0
    −2t 
0 0  e 0 0
e−t 0  0 0 0 1  0 e−t te−t  dt

=  0
0 0 te−t e−t 1 0 0 e−t
   
Z 2 0 0 0 0 0 0
= 0 0 0  dt = 0 0 0 
0 −2t 1−e−4
0 0 e 0 0 2
 T  2
T
We want Wc η = 0 0 1 . Therefore, η = 0 0 1−e−4 . Therefore, desired input:
T (2−t)
u(t) = B T eA η
 −2(2−t)  
e 0 0 0
2e−(2−t)
e−(2−t) (2 − t)e−(2−t)   0  =
 
= 0 0 1  0 for t ⩾ 0.
−(2−t) 2 1 − e−4
0 0 e 1−e−4

6. Design an attitude control system for a rigid, single-axis satellite given in Figure 2.
The system’s input is the control torque, T , and the output is the angular position, θ.
The dynamics of the satellite are governed by the equation: T = 10 θ̈. where T is the
torque applied about this axis, θ̈ is the angular acceleration of the satellite. (9)
The control system design must meet the following requirements:
ˆ Closed-loop system specifications:
- Damping Ratio (ζ): Achieve critically damped behavior.

5
 - Settling Time: 1 second, defined as the time required for the system to reach and
remain within 2% of the final value.
ˆ Observer Design: - The measurement of all states are available.
ˆ Constraints: - Ensure that the system dynamics are stable and meet the given time
and damping requirements without introducing oscillations.

Figure 2: Satellite control schematics

The goal is to design a control law and observer that allow precise control over the
satellite’s orientation about the single axis, meeting the specified performance criteria.
Solution: Controller design: Given ζ = 1 (critical damping) and ζω4n = 1 ⇒ ωn = 4.
The desired pole location are −4, −4. Hence the desired system matrix in controller
canonical form is:  
−8 −16
Ad =
1 0
The system equation is T = 10θ̈. Let x1 = θ, x2 = θ̇.
      
ẋ1 0 1 x1 0
= + 1 T
ẋ2 0 0 x2 10
1 1
    
0 1 0 0
Let P = 1 10 = 1 10 . Using P z = x as a transformation:
10
0 0 1 10
0
   
0 0 1
ż = z+ T
1 0 0

Then using the identity:

     
0 0 1 b −8 −16  
b = −8 −16 .
+ K = Ad = ⇒K
1 0 0 1 0

b −1 = −160 −80 . Therefore, the control law is u = −160 −80 x.

   
Thus, K = KP

Observer design: Since the measurement of all states are available, there is no requirement
of an observer design. The states can directly be used in the controller.

6
7. Let CA,B and OA,C be the controllability and observability matrix of a system such that
       

 1 0   
 2 1 
   
0 0 0 5
    
img(CA,B ) = span   ,   and ker(OA,C ) = span   ,  
 

 1 1   
 3 1 

0 1 1 0
   

Find a basis of the set of states that are both controllable and unobservable. (5)
Solution: Let v ∈ img(CA,B ) ∩ ker(OA,C . Then
                
1 0 2 1 1 0 −2 −1 α1 0 α1 2
0 0 0 5 0 0 0 −5  α2  0 α2  1
v= 1 α1 +1 α2 = 3 β1 +1 β2 ⇒ 1 1 −3 −1  β1  = 0 ⇒  β1  = 1
               

0 1 1 0 0 1 −1 0 β2 0 β2 0
 T
Therefore, v = 2 0 3 1 . Hence,
 
2
0
img(CA,B ) ∩ ker(OA,C = span 
3


8. Consider the following system and find it’s Kalman decomposition canonical form. (5)
   
−5 0 0 0 1
 0 1 0 0 1 u(t), y(t) = 1 0 1 0 x(t) + u(t)
    
ẋ(t) = 
 0 x(t) +
0 0 0 0
0 0 0 −2 0

Answer the following from the canonical form:

(a) Is the system stabilizable?

(b) Is the system detectable?
(c) Is the Lyapunov stable?
(d) Is the system BIBO stable?
(e) What are poles of the system?

Solution: The state space representation is already in the Kalman decomposition

canonical form.
(a) The system is not stabilizable, since 0 is an uncontrollable eigenvalue.

(b) The system is not detectable, since 1 is an unobservable eigenvalue.

(c) Since one of the eigenvalues 1 of the system matrix is in the RHP of complex plane,
the system is not Lyapunov stable.

(d) & (e) The eigenvalue of the system that is both controllable and observable is −5.
Therefore, −5 is the pole of the system. Hence, the system is BIBO stable as the pole is
in LHP of the complex plane.