Lecture – 20

Controllability and Observability of Linear

Time Invariant Systems

Dr. Radhakant Padhi

Asst. Professor
Dept. of Aerospace Engineering
Indian Institute of Science - Bangalore
Evaluation of Matrix Exponential eAt

Dr. Radhakant Padhi

Asst. Professor
Dept. of Aerospace Engineering
Indian Institute of Science - Bangalore
Method – 1: Power-series
2 2 3 3
A t A t
e = I + At +
At
+ +"
2! 3!

z This method is useful and accurate only if the

series truncates naturally. Otherwise, series
truncation introduces approximation error.
z Direct computation of eAt as power series is
computationally inefficient as well.

ADVANCED CONTROL SYSTEM DESIGN 3

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
 Method – 2:
Using Laplace Transform

e = L ( sI − A ) ⎤
⎡
At −1 −1
⎣ ⎦

z This method results in closed form expressions for

eAt, can be quite useful for small matrices.
z Numerical algorithms exist to evaluate
( sI − A) . However, its inverse still need to be found.
−1

z Can be quite cumbersome for large matrices.

ADVANCED CONTROL SYSTEM DESIGN 4

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Method – 3:
Using Similarity Transform
(Provided the matrix can be diagonalizable)
2 2 3 3
A t A t
e = I + At +
At
+ +" Similarity
2! 3!
Transformation:
−1
= PP + PDP t + −1 ( PDP −1
)( PDP −1
) t 2

+ " A = PDP
−1
2!
⎛ D 2 t 2 D 3t 3 ⎞ −1
= P ⎜ I + Dt + + + "⎟ P
⎝ 2! 3! ⎠
⎡eλ1t 0 0 ⎤
⎢ ⎥ −1
= P⎢ 0 % 0 ⎥P
⎢ 0 0 eλnt ⎥
⎣ ⎦
ADVANCED CONTROL SYSTEM DESIGN 5
Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Method – 4: Sylvester’s Formula
Case – 1: Distinct Eigenvalues

eAt satisfies the following determinant equation:

1 λ1 λ12 " λ1n −1 eλ1t
1 λ2 λ22 " λ2n −1 eλ2t
# # # # # =0
1 λn λn2 " λnn −1 eλnt
I A A2 " An −1 At
eN
Ultimate aim
i.e.

e At = α 0 ( t ) I + α1 ( t ) A + α 2 ( t ) A2 + " + α n −1 ( t ) An −1

ADVANCED CONTROL SYSTEM DESIGN 6

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
 Method – 4: Sylvester’s Formula
Case – 1: Distinct Eigenvalues

The coefficients α 0 ( t ) , α1 ( t ) ," , α n −1 ( t )

can be determined from the following set
of equations:
α 0 ( t ) + α1 ( t ) λ1 + α 2 ( t ) λ12 + " + α n −1 ( t ) λ1n −1 = eλ t 1

α 0 ( t ) + α1 ( t ) λ2 + α 2 ( t ) λ22 + " + α n −1 ( t ) λ2n −1 = eλ t 2

#
α 0 ( t ) + α1 ( t ) λn + α 2 ( t ) λ + " + α n −1 ( t ) λ
2
n
n −1
n =e λn t

ADVANCED CONTROL SYSTEM DESIGN 7

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
 Method – 4: Sylvester’s Formula
Case – 2: Repeated Eigenvalues

eAt satisfies the following determinant equation:

0 0 1 3λ1 "
( n − 1)( n − 2 ) λ n−3 t 2 λ1t Eigenvalues:
e
1
2 2 λ1 , λ1 , λ1 , λ4 ," , λn
0 1 2λ1 3λ12 " ( n − 1) λ1n−2 te λ1t  

3 times
λ1t
1 λ1 λ 2
λ 3
" λ n −1
e
=0
1 1 1
λ4t
1 λ4 λ4
2
λ 4
3
" λ 4
n −1
e
# # # # " # #
1 λn λn2 λn3 " λnn −1 eλnt
I A A2 A3 " An −1 e At
i.e.
e At = α 0 ( t ) I + α1 ( t ) A + α 2 ( t ) A2 + " + α n −1 ( t ) An −1
ADVANCED CONTROL SYSTEM DESIGN 8
Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
 Method – 4: Sylvester’s Formula
Case – 2: Repeated Eigenvalues
The coefficients α 0 ( t ) , α1 ( t ) ," , α n −1 ( t )
can be determined from:

α 2 ( t ) + 3α 3 ( t ) λ1 + " +
( n − 1)( n − 2 ) t 2
α n −1 ( t ) λ1n −3 = eλ t 1

2 2
α1 ( t ) + 2α 2 ( t ) λ1 + 3α 3 ( t ) λ12 + " + ( n − 1) α n −1 ( t ) λ1n − 2 = teλ1t
α 0 ( t ) + α1 ( t ) λ1 + α 2 ( t ) λ12 + " + α n −1 ( t ) λ1n −1 = eλ t 1

α 0 ( t ) + α1 ( t ) λ4 + α 2 ( t ) λ42 + " + α n −1 ( t ) λ4n −1 = eλ t 4

#
α 0 ( t ) + α1 ( t ) λn + α 2 ( t ) λn2 + " + α n −1 ( t ) λnn −1 = eλ t 1

ADVANCED CONTROL SYSTEM DESIGN 9

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Method – 4: Sylvester’s Formula
Example
⎡0 1 ⎤
A=⎢ ⎥ , λ1,2 = 0, −2
⎣0 −2 ⎦
To compute e At using Sylvester's formula, we have
1 λ1 eλ1t 1 0 1
1 λ2 eλ2t = 1 −2 e −2t = 0
I A e At I A e At
Expanding the determinant
−2e At + A + 2 I − Ae −2t = 0
⎡ −2 t ⎤
e = ( A + 2 I − Ae ) =
At 1 −2 t ⎢ 1
1
2
(1 − e )⎥
2 ⎢ −2 t ⎥
⎣0 e ⎦
ADVANCED CONTROL SYSTEM DESIGN 10
Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Controllability of Linear Time Invariant Systems

Dr. Radhakant Padhi

Asst. Professor
Dept. of Aerospace Engineering
Indian Institute of Science - Bangalore
Controllability
• A system is said to be controllable at time t0 if
it is possible by means of an unconstrained
control vector to transfer the system from any
initial state X 0 to any other state in a finite
interval of time

• Controllability depends upon the system matrix

A and the control influence matrix B
ADVANCED CONTROL SYSTEM DESIGN 12
Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Graphical Meaning

Xf

Must happen in finite time.

X0

ADVANCED CONTROL SYSTEM DESIGN 13

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Condition for Controllability:
(single input case)
System: X = AX + Bu
t
Solution: X (t ) = e At X (0) + ∫ e A( t −τ ) Bu (τ ) dτ
0

Assuming X (t1 ) = 0, t1

0 = eAt1 X (0) + ∫ eA(t1−τ ) Bu(τ ) dτ

0
t1

X (0) = − ∫ e− Aτ Bu (τ ) dτ
0
ADVANCED CONTROL SYSTEM DESIGN 14
Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
 Condition for Controllability:
(single input case)
n −1
e − Aτ = ∑ α k (τ ) Ak (Sylvester's formula )
k =0
t1 n −1 t1

X (0) = − ∫ e − Aτ Bu (τ ) dτ = −∑ Ak B ∫ α k (τ ) u (τ ) dτ
0 k =0 0
t1
n −1
= − ∑ Ak B β k where β k  ∫ α k (τ ) u (τ ) dτ
k =0 0

= − ⎡⎣ B AB " A B ⎤⎦ [ β 0 n −1
β1 " β n −1 ]
T

[ β0 β1 " β n −1 ]
T
This system should have a non-trivial solution for

ADVANCED CONTROL SYSTEM DESIGN 15

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
 Controllability
Result: If the rank of CB  ⎡⎣ B AB " An −1 B ⎤⎦ is n,
then the system is controllable.
Example:
⎡ x1 ⎤ ⎡− 1 0 ⎤ ⎡ x1 ⎤ ⎡2⎤
⎢ x ⎥ = ⎢ 0 − 2⎥ ⎢ x ⎥ + ⎢1 ⎥ u
⎣ 2⎦ ⎣ ⎦ ⎣ 2⎦ ⎣ ⎦
⎡ ⎡ 2 ⎤ ⎡ −1 0 ⎤ ⎡ 2 ⎤ ⎤ ⎡ 2 −2 ⎤
CB = ⎢ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎥=⎢ ⎥
⎣⎣ ⎦ ⎣
1 0 −2 ⎦ ⎣ ⎦⎦ ⎣
1 1 −2 ⎦
rank ( CB ) = 2 ∴ The system is controllable.
ADVANCED CONTROL SYSTEM DESIGN 16
Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Output Controllability
Result: X = AX + BU
Y = CX + DU

X ∈ Rn , U ∈ Rm , Y ∈ R p

If the rank of CB  ⎡⎣CB CAB " CAn −1 B D ⎤⎦ is p,

then the system is output controllable.
Note: The presence of DU term in the output equation
always helps to establish output controllability.

ADVANCED CONTROL SYSTEM DESIGN 17

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Observability of Linear Time Invariant Systems

Dr. Radhakant Padhi

Asst. Professor
Dept. of Aerospace Engineering
Indian Institute of Science - Bangalore
Observability
• A system is said to be observable at time
t0 if, with the system in state X(t0) ,it is
possible to determine this state from the
observation of the output over a finite
interval of time

• Observability depends upon the system

matrix A and the output matrix C
ADVANCED CONTROL SYSTEM DESIGN 19
Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
 Observability
If the rank of OB  ⎡⎢C T AT C T (A ) C T ⎤⎥ is n,
n −1
Result: " T
⎣ ⎦
then the system is observable.
Example:
⎡ x 1 ⎤ ⎡ −1 0 ⎤ ⎡ x1 ⎤ ⎡ 2 ⎤ ⎡ x1 ⎤
⎢ x ⎥ = ⎢ 0 −2 ⎥ ⎢ x ⎥ + ⎢1 ⎥ u y = [1 0] ⎢ ⎥
⎣ 2⎦ ⎣ ⎦ ⎣ 2⎦ ⎣ ⎦ ⎣ x2 ⎦
⎡ ⎡1 ⎤ ⎡ −1 0 ⎤ ⎡1 ⎤ ⎤ ⎡1 −1⎤
OB = ⎢ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎥=⎢ ⎥
⎣⎣ ⎦ ⎣
0 0 −2 ⎦ ⎣ ⎦⎦ ⎣
0 0 0 ⎦
rank ( OB ) = 1 ≠ 2 ∴ The system is NOT observable.
ADVANCED CONTROL SYSTEM DESIGN 20
Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Controllability and Observability
in Transfer Function Domain
z The system is both controllable and
observable if there is no Pole-Zero
cancellation.

z Note: The cancelled pole-zero pair

suppresses part of the information about
the system

ADVANCED CONTROL SYSTEM DESIGN 21

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
 Principle of Duality
System S1: X = AX + BU CB = ⎡⎣ B AB A2 B " An −1 B ⎤⎦

Y1 = CX OB = ⎡C T
⎣
T
AC T T2
A C T
" A T n−1
CT ⎤
⎦

System S2: Z = AT Z + C T V CB = ⎡C T AT C T AT C T " AT C T ⎤

2 n−1

⎣ ⎦
Y2 = BT Z OB = ⎡⎣ B AB A2 B " An −1 B ⎤⎦

The principle of duality states that the system S1 is controllable if

and only if system S2 is observable; and vice-versa!
Hence, the problem of observer design for a system is actually a
problem of control design for its dual system.

ADVANCED CONTROL SYSTEM DESIGN 22

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Stabilizability and Detectability
z Stabilizable system: Uncontrollable
system in which uncontrollable part is
stable

z Detectable system: Unobservable

system in which the unobservable
subsystem is stable

ADVANCED CONTROL SYSTEM DESIGN 23

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Example
Ref: B. Friedland, Control System Design, McGraw Hill, 1986

System Dynamics
⎡ x1 ⎤ ⎡ 2 3 2 1 ⎤ ⎡ x1 ⎤ ⎡ 1 ⎤
⎢ x ⎥ ⎢ −2 −3 0 0 ⎥ ⎢ x ⎥ ⎢ −2 ⎥
⎢ 2⎥ = ⎢ ⎥ ⎢ 2⎥ + ⎢ ⎥u
⎢ x3 ⎥ ⎢ −2 −2 −4 0 ⎥ ⎢ x3 ⎥ ⎢ 2 ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢⎣N x4 ⎥⎦ ⎣ −2 −2 −2 −5⎦ ⎢⎣ x4 ⎥⎦ ⎣ −1⎦
 
N N
X A X B

Output Equation
y = [ 7 6 4 2] X
  

C
ADVANCED CONTROL SYSTEM DESIGN 24
Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Example
Ref: B. Friedland, Control System Design, McGraw Hill, 1986

Transfer Function:
y (s) ( s + 2 )( s + 3)( s + 4 ) 1
= C ( sI − A ) B =
−1
=
u (s) (
s + 1)( s + 2 )( s + 3)( s + 4 ) ( s + 1)
 

pole-zero cancellation

Implication: What appears to be a fourth-order system, is

actually a first-order system! Hence, there is
either loss of controllability or observability
(or both).

Question: Is this system stabilizable?

ADVANCED CONTROL SYSTEM DESIGN 25
Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Example
Ref: B. Friedland, Control System Design, McGraw Hill, 1986

Define X = TX . Then
X = TX = T ( AX + Bu )

( )
X = TAT −1 X + TB u
( )
Let
⎡4 3 2 1⎤ ⎡ −1 0 0 0 ⎤ ⎡1 ⎤
⎢3 3 2 1⎥⎥ ⎢ 0 −2 0 0 ⎥ ⎢0 ⎥
T =⎢ ⇒ TAT −1 = ⎢ ⎥ , TB = ⎢ ⎥
⎢2 2 2 1⎥ ⎢ 0 0 −3 0 ⎥ ⎢1 ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣1 1 1 1⎦ ⎣ 0 0 0 −4 ⎦ ⎣0 ⎦

ADVANCED CONTROL SYSTEM DESIGN 26

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Example
Ref: B. Friedland, Control System Design, McGraw Hill, 1986

⎡ x1 ⎤ ⎡ − x1 + u ⎤
⎢ ⎥ ⎢ ⎥
x
⎢ 2⎥ = ⎢ −2 x2 ⎥
, y = CX = CT −1 X = x1 + x2
⎢ x3 ⎥ ⎢ −3 x3 + u ⎥
⎢ ⎥ ⎢ ⎥
⎢⎣ x4 ⎥⎦ ⎢⎣ −4 x4 ⎥⎦
Implications:
x1 : Affected by the input; visible in the output
x2 : Unaffected by the input; visible in the output
x3 : Affected by the input; Invisible in the output
x4 : Unaffected by the input; Invisible in the output
ADVANCED CONTROL SYSTEM DESIGN 27
Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Block Diagram:

ADVANCED CONTROL SYSTEM DESIGN 28

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
Where do uncontrollable or
unobservable systems arise?

z Redundant state variables

z Physically uncontrollable system

z Too much symmetry

ADVANCED CONTROL SYSTEM DESIGN 29

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
References

z K. Ogata: Modern Control Engineering,

3rd Ed., Prentice Hall, 1999.

z B. Friedland: Control System Design,

McGraw Hill, 1986.

ADVANCED CONTROL SYSTEM DESIGN 30

Dr. Radhakant Padhi, AE Dept., IISc-Bangalore
ADVANCED CONTROL SYSTEM DESIGN 31
Dr. Radhakant Padhi, AE Dept., IISc-Bangalore