Lecture Note

Controllability and Observability

1 State Diagram
Before exploring the topic on controllability and observability, let us discuss an important contin-
uation of the previous lectures, i.e., state diagram. A state diagram is nothing but a signal flow
graph which depicts the flow of signals from input to output while showing all the necessary states
and their derivatives. We can draw the state diagram either from the block diagram or from the
state model directly. Please note that a differentiator is denoted by s and an integrator is denoted
by 1/s in a state diagram. Consider a second order system in controllable canonical form as

x˙1 (t) = x2 (t)

x˙2 (t) = −α2 x1 (t) − α1 x2 (t) + u(t)
y(t) = β2 x1 (t) + β1 x2 (t)

The state diagram of the above is shown below.

2 Controllability and Observability

Controllability and observability are two important properties of state models which are to be
studied prior to designing a controller.

Controllability deals with the possibility of forcing the system to a particular state by appli-
cation of a control input. If a state is uncontrollable then no input will be able to control that
state. On the other hand whether or not the initial states can be observed from the output is
determined using observability property. Thus if a state is not observable then the controller will
not be able to determine its behavior from the system output and hence not be able to use that
state to stabilize the system.

1
2.1 Controllability
A control engineer asks a question if a dynamical model is controllable before designing a compen-
sator. Only when a dynamical system is controllable, a suitable controller is designed.
Consider a dynamical system

ẋ = Ax + Bu (1)
where A ∈ Rn×n and B ∈ Rn×p
Controllability: Definition
The state equation (1) ( or the pair (A, B) ) is controllable if for any initial state x(0) = x0
and any final state x1 , there exists an input u(t) that takes the system from x0 to x1 in the state
space in a finite time. Otherwise the state equation (1) is uncontrollable.
Theorem 1
The following three statements are equivalent.

1. state equation (1) ( or the pair (A, B) ) is controllable

2. The n × n controllability Gramian matrix

Z t Z t
Aτ T AT τ T
Wc (t) = e BB e dτ = eA(t−τ ) BB T eA (t−τ ) dτ (2)
0 0

is nonsingular for any t > 0.

3. The n × nm controllability matrix

UC = B AB A2 B ...... An−1 B
 
(3)
has rank n, i.e. full row rank.

proof
Part I: Assume Wc is nonsingular. Then show that the pair (A, B) is controllable. Part II:
Assume that the pair (A, B) is controllable. Then show that Wc is nonsingular.
Assume Wc is nonsingular. The response of the state equation (1)is given as
Z t
At
x(t) = e x(0) + eA(t−τ ) Bu(τ )dτ (4)
0
It is required to show that there exists a control input u that transfers the system from x0 to
x(t) = x1 in a finite time. Let the control input be given by
T (t−τ )
u(τ ) = −B T eA Wc −1 (t) eAt x(0) − x1
 
(5)

Inserting equation (5) in equation (4), one obtains

Z t 
T AT (t−τ )
At A(t−τ )
dτ Wc −1 (t) eAt x(0) − x1
 
x(t) = e x(0) − e BB e (6)
0

2
 This leads to following form

x(t) = x1 = eAt x(0) − Wc (t)Wc −1 (t) eAt x(0) − x1 = x1

 
(7)

only if Wc is nonsingular.
Part II: By contradiction: the pair (A, B) is controllable but Wc is singular
Since Wc is singular, then there exists a non-zero vector v such that v T Wc (t)v = 0. This implies
Z t Z t
T T A(t−τ ) T AT (t−τ ) T
v Wc (t)v = v e BB e vdτ = || B T eA (t−τ ) v ||2 dτ (8)
0 0

T
Thus B T eA (t−τ ) v = v T eA(t−τ ) B = 0
Let the system (1) has to move from x(0) = e−At v to x(t) = x1 = 0. Then
Z t
0=v+ eA(t−τ ) Bu(τ )dτ (9)
0

If multiplied by v T on both sides, then A square matrix M is symmetric if M = M T

Z t
T
0=v v+ v T eA(t−τ ) Bu(τ )dτ =|| v ||2 +0 (10)
0

This implies that v = 0 which is a contradiction to the assumption that v 6= 0.

Proof : Third Statement
If Wc is non-singular then the controllability matrix C has full row rank.
Consider that C does not have full row rank. Then there exits a non-zero vector v such that
v T C = 0. This implies that

v T Ak B = 0 k = 0, 1, 2, 3, ..., n − 1 (11)

Using Caley Hamilton Theorem, eAt = β0 I + β1 A + .... + βn−1 An−1 , one can easily show that
eAt B is linear combination of [B AB A2 B ...... An−1 B].Since v T Ak B = 0, so also v T eAt B = 0.
It is already shown that if v T eAt B = 0 for non-zero v, then Wc is singular. This result concludes
that if C has full row rank, then Wc has to be non-singular.
Let us assume converse: C has full row rank, but Wc (t) is singular.
Since Wc is singular so also v T eAt B = 0 holds true. At t = 0, v T B = 0
Differentiating v T eAt B and then setting t = 0, one gets v T AB = 0. Subsequent differential will
yield v T [B AB .... An−1 B] = v T C = 0. This contradicts the assumption that C has full row
rank.

Example 1:
Given a dynamical system    
−0.5 0 0.5
ẋ = x+ u (12)
0 −1 1
find the input force u that will transfer x(0) = [10 − 1]T to x(2) = [0 0]T in two seconds.
Solution:

3
 Z 2
 −0.5τ  
 e−0.5τ 0
 
e 0 0.5 
Wc (2) = 0.5 1 dτ (13)
0 0 e−τ 1 0 e−τ
 
0.2162 0.3167
This results Wc (2) =
0.3167 0.4908
T
If one selects the control input u(t) = −B T eA (tf −t) Wc −1 (tf ) eAtf x(0) − xf , where tf = 2 is
 

the final time and xf = x(2) is the final state, then one will get u(t) = −58.82e0.5t + 27.96et .

Example 2
Consider the system ẋ = Ax + Bu, y = Cx. where
   
−2 1 1
A= , B= and C = [0 1]
1 −2 1
Y (s)
Show if the system is controllable. Find the transfer function U (s)
. Can you see any connection
between controllability and the transfer function?
Solution: The controllability matrix is given by


1 −1
U = [B AB] =
1 −1

Its determinant kU k = 0 ⇒ U has a rank 1 < 2. Thus the system is not controllable. The transfer
function
 −1  
Y (s) −1 s + 2 −1 1 1
G(s) = = C(sI − A) B = [01] =
U (s) −1 s + 2 1 s+1

Although state model is of order 2, the transfer function has order 1. The eigenvalues of A
are λ1 = −1 and λ2 = −3. This implies that transfer function is associated with pole-zero
cancellation where the pole is at −3. Since one of the dynamic mode is cancelled, the system
became uncontrollable.

2.2 Observability
Given the dynamical system
ẋ = Ax + Bu
(14)
y = Cx
the system is observable if for any unknown initial state x(0), there exists a finite t1 > 0 such that
the knowledge of the input u and the output y over [0, t1 ] suffices to determine uniquely the initial
state x(0). Otherwise the system is not observable.
Theorem 2:
Following statements are equivalent.

1. The system (14) or the pair (A, C) is observable

4
 2. The n × n observability Gramian matrix
Z t
T
Wo (t) = eA τ C T CeAτ dτ (15)
0

is nonsingular for any t > 0.

3. the np × n matrix 

C
 CA 
 
 . 
O=
 . 
 (16)
 
 . 
CAn−1
has rank n, i.e., full column rank.

2.3 Theorem of Duality

The Pair (A, B) is controllable if and only if the pair (AT , B T ) is observable.

Try to prove the theorem on observability and the Theorem of Duality.

2.4 Discrete Time State Variable Model

A discrete time state varible model is described by

x(k + 1) = Ax(k) + Bu(k) (17)

y(k) = Cx(k) + Du(k)

Note that D does not play any role in determining controllability. The above model or the pair
(A, B) is said to be controllable if for any initial state x(0) = x0 and any final state x1 , there
exists an input sequence of finite length that takes the system from x0 to x1 . Otherwise the state
equation (17) is uncontrollable. In literature, there are three different controllability definitions.

1. Transfer any state to any state (reachability)

2. Transfer any state to the zero state (controllability to origin)

3. Transfer zero state to any state (controllability from origin)

For continuous time case all the three definitions are equivalent and we term it as controllability
(in some book reachability). But for discrete time case definitions (1) and (3) are equivalent but
(2) is not not. In some books, controllability refers to the second definition while reachability refers
to the first or third.

5
Example:

Take the following state equation

   
2 1 −1
x(k + 1) = + u(k)
0 0 0

The reachability matrix is  

−1 −2
U = [B AB] =
0 0
The rank of U is 1, thus, the model is not reachable. However, for any x1 (0) = α and x2 (0) = β,
the input u = 2α + β transfers x(0) = [α β]T to x(1) = [0 0]T . Thus, the model is controllable to
origin or simply controllable.

Theorem: The discrete state variable model (17) or the pair (A, B) is controllable to origin
if and only if range(An ) ⊂ range(U ).

3 Stabilizablity and Detectability

If for a state model, the uncontrollable states are stable then the model is called stabilizable. It is
an important property for designing a controller since allows us to design a controller even if the
model is not fully controllable (but stabilizable).

Example:

Consider the state equation    

−3 0 0
ẋ(t) = x(t) + u(t) (18)
1 −2 1
Controllability matrix 

0 0
U = [B AB] =
1 −2
is singular. So, the system is uncontrollable. The state equation can also be written as

x˙1 (t) = −3x1 (t)

x˙2 (t) = x1 (t) − 2x2 (t) + u(t)

Careful observation of the above model reveals that the state x1 (t) is uncontrollable since the
derivative of x1 (t) depends only on x1 (t), neither on u(t), nor on x2 (t). Now, the first state equa-
tion describes the dynamis and it is clear that x1 (t) is a stable state. Thus, the model is stabilizable
although not controllable.

Similarly when the unobservable states of a state model are stable, it is known as detectable.
It is a desirable property for designing an observer.