Electrical Engineering Department

University of Hail
Hail-2019

Advanced Digital Control Syst

EE554
Chapter 4
Controllability and Observability of digital linear systems

Dr Mourad Kchaou
Electrical Engineering Department
Hail University

Friday 20th March, 2020

Presented by Dr Mourad v. (b.2003201353)

Outline

1 Reachability and controllability

Reachability
Controllability
Controllability

2 Observability

3 Canonical Decomposition
Reachability and controllability

Reachability and controllability

Reachability and controllability

I In this part, we will discuss two fundamental concepts of
control theory: those of controllability and
observability. These concepts are able to describe the
interaction in a system between the external environment
(inputs and outputs) and the internal variables (states)
I Controllability is the property that indicates if the
behaviour of a system can be controlled by acting on its
inputs.
I Observability is the property that indicates if the internal
behaviour of a system can be detected at its outputs.

3 / 1
Reachability and controllability

Reachability and controllability

I The reachability problem is to find the set of all the final

states x(k1 ) reachable starting from a given initial state
x(k0 ).
I A state x(k1 ) of a dynamic system is reachable from the
state x(k0 ) in the time interval [k0 , k1 ] if there exists an
input function u(k) that transfer x0 to x1 in a finite time
k1 < ∞

4 / 1
Reachability and controllability

Reachability

x(k) u1 (k) u2 (k)

k
k1 k2
Figure: A reachable state x1

5 / 1
Reachability and controllability

Reachability

I The controllability problem is to find the set of all the

initial states x(k0 ) controllable to a given final state x(k1 )
I A state x(k0 ) is called controllable if there exists an input
that transfers the state from x(k0 ) to the zero state in
some finite time.

6 / 1
Reachability and controllability

Reachability

Consider an LTI discrete-time system described by state space

model.
(
x(k + 1) = Ax(k) + Bu(k)
(1)
y(k) = Cx(k) + Du(k)

The solution is given by

k−1
X
x(k) = Ak x(0) + Ak−i−1 Bu(i) (2)
i=0

7 / 1
Reachability and controllability

Reachability

The states x(k) obtained starting from the initial condition

x(0) = 0 and considering only the forced evolution of the
system is given:
k−1
X
x(k) = Ak−i−1 Bu(i) (3)
i=0
 
u(k − 1)
u(k − 2)
 
= B AB A2 B . . . Ak−1 B 

..  (4)
 . 
u(0)
= RU (5)

A solution U exists if and only rank(R) = n

8 / 1
Reachability and controllability

Reachability

Definition
The Reachability matrix in k steps is defined as

R(k) = B AB A2 B . . . Ak−1 B
 
(6)

Theorem
The system (A, B) is completely reachable if only if
rank(R(n)) = n

9 / 1
Reachability and controllability

Controllability
Definition
A state x0 = x(0) is controllable to zero in k steps if it exists
an input sequence u(0), u(1), · · · , u(k − 1) which brings the
initial state x0 to a final state equal to the origin x(k) = 0 in
the time interval [0, k]

k−1
X
0 = Ak x(0) + Ak−i−1 Bu(i) (7)
i=0

k−1
X
− Ak x(0) = Ak−i−1 Bu(i) = RU (8)
i=0
Equation (8) admits a solution if and only if
Ak x(0) ∈ Im(R), ∀x0 ∈ Rn
10 / 1
Reachability and controllability

Controllability

I For discrete linear systems the reachability and

controllability properties are NOT equivalent:
I The reachability implies the controllability
I The controllability does not imply the reachability
I If the reachability condition is satisfied rank(R) = n and
rank(A) = n, a system is controllable . In this case we
have rank(A−k R) = rank(R) = n
I If, however, rank(A) < n, then controllability does not
imply reachability

11 / 1
Reachability and controllability

Controllability
Example
Find the vector u(k), such
 that the system reach, from the
1
origin, the state x(2) =
1
   
0 1 0
x(k + 1) = x(k) + u(k) (9)
−1 −3 1

 
  0 1
R = B AB = (10)
1 −3

the system is reachable

  since rank(R)
 = 2. the input vector is
u(1) 4
given by U = = R−1 x(2) =
u(0) 1
12 / 1
Reachability and controllability

Gilbert’s test for Controllability

The Gilbert’s method is an alternate test of the controllability of a
given system.
As already discussed, It is possible to transform the A matrix into a
diagonal matrix using a transformation matrix T .
Let x = T z. System (20) can be transformed to

z(k + 1) = Λz(k) + T −1 Bu(k) (11)

where Λ = T −1 AT is a diagonal matrix. The ith equation can be

written as

zi (k + 1) = λi zi (k) + α1i u1 (k) + α2i u2 (k) + . . . + αmi um (k) (12)

If all the elements of ith row of T −1 B, i.e.,

α1i = α2i = . . . = αmi = 0 (13)

then the system is uncontrollable.

13 / 1
Reachability and controllability

Gilbert’s test for Controllability

I Hence, the system is the controllable if and only if the

input matrix of the diagonal model has no zero
rows.
I Assume the system is transformed to Jordon’s canonical
form.
Thus, the system is controllable if the elements of the row
of T −1 B corresponding to the last row of any Jordan block
are not all zero.

14 / 1
Observability

Observability

The solution for the output is

k−1
X
y(k, x0 , u(k)) = CAk x(0) + CAk−i−1 Bu(i) + Du(k) (14)
i=0

Definition
The pair of states x1 = x2 is called indistinguishable from the
output y, if for any input sequence u(k)

y(k, x1 , u(k)) = y(k, x2 , u(k)), ∀k ≥ 0 (15)

A linear system is called (completely) observable if no pair of

states are indistinguishable from the output

15 / 1
Observability

Observability

Consider the problem of reconstructing the initial condition x0

from n output measurements, applying a known input sequence

y(0) = Cx0 + Du(0)

y(1) = CAx0 + CBu(0) + Du(1)
..
. (16)
k−2
X
k−1
y(k − 1) = CA x(0) + CAk−i−2 Bu(i) + Du(k − 1)
i=0

16 / 1
Observability

Observability

Let
   
C y(0) − Du(0)
  CA  y(1) − CBu(0) − Du(1) 
O= , Y =  ..
   
.. 
  . . 
k−1 Pk−2 k−i−2
CA y(k − 1) − i=0 CA Bu(i) − Du(k − 1)
(17)

The initial state x0 is determined by solving the linear system

Y = Ox0 (18)

17 / 1
Observability

Observability

Definition
The observability matrix is defined as
 
C
 CA 
O(k) =  .  (19)
 
 .. 
CAk−1

If we assume perfect knowledge of the output (i.e., no noise on

output measurements), we can always solve the system
Y = Ox0 . In particular:
I There is only one solution if rank(O(n)) = n
I There exist infinite solutions if rank(O) < n.
18 / 1
Observability

Observability

Theorem
The system (A, C) is observable if only if rank(O(n)) = n

19 / 1
Canonical Decomposition

Canonical Decomposition

The Canonical Decompositions of state equations will

establish the relationship between Controllability,
Observability , and a transfer matrix and its minimal
realisations

20 / 1
Canonical Decomposition

Canonical Decomposition
Consider an LTI discrete-time system described by state space
model.
(
x(k + 1) = Ax(k) + Bu(k)
(20)
y(k) = Cx(k) + Du(k)

Let x̂ = P x, where P is a nonsingular matrix. Then the state

equation
(
x̂˙ = Âx̂(t) + B̂u(t)
(21)
y(t) = Ĉ x̂(t) + D̂u(t)

with  = P AP −1 , B̂ = P B, Ĉ = CP −1 and D̂ = D is
equivalent to (20). All properties of (20) including stability,
controllability, and observability, are preserved in the
transformed system (21).
21 / 1
Canonical Decomposition

Canonical Decomposition

Theorem
Consider the n-dimensional state equation in (20) and assume
that the system is not controllable i.e.

Rank(R(n)) = Rank( B AB A2 B . . . An−1 B ) = n1 < n

 

(22)

We form the n × n matrix

P −1 = q1 q1 qn1
 
. . . qn (23)

where the first n1 columns are any n1 linearly independent

columns of C, and the remaining columns can arbitrarily be
chosen as long as P is nonsingular.

22 / 1
Canonical Decomposition

Canonical Decomposition
Theorem
Then the equivalence transformation x̂ = P x transforms (20) to
      
x̂c (k + 1) Âc Â12 x̂c (k) B̂c
= + u(k)
x̂ĉ (k + 1) 0 Âĉ x̂ĉ (k) 0
  (24)
  x̂c (k)
y = Ĉc Ĉĉ + D̂u(k)
x̂ĉ (k)

where Âc is n1 × n1 and Âĉ is (n − n1 ) × (n − n1 ) . The

n1 -dimensional subequation of (24)
(
x̂c (k + 1) = Âc x̂c (k) + B̂c u(k)
(25)
yc (k) = Ĉc x̂c (k) + D̂c u(k)

is controllable and has the same transfer function as the original state
equation (20).
23 / 1
Canonical Decomposition

Canonical Decomposition
Theorem
Consider the n-dimensional state equation in (20) and assume
that the system is not controllable i.e.
 
C
 CA 
Rank(O(n)) = Rank( . ) = n2 < n (26)
 
 . .
CAn−1

We form the n × n matrix

 T
P = p1 p2 pn−2 . . . pn (27)

where the first n2 columns are any n2 linearly independent rows

of O, and the remaining columns can arbitrarily be chosen as
long as P is nonsingular. 24 / 1
Canonical Decomposition

Canonical Decomposition
Theorem
Then the equivalence transformation x̂ = P x transforms (20) to
      
x̂o (k + 1) Âo 0 x̂o (k) B̂o
= + u(k)
x̂ō (k + 1) Â21 Âō x̂ō (k) B̂ō
  (28)
  x̂o (k)
y(k) = Ĉo 0 + D̂u(k)
x̂ō (k)

where Âo is n2 × n2 and Âō is (n − n2 ) × (n − n2 ) . The

n2 -dimensional subequation of (28)
(
x̂o (k + 1) = Âo x̂o (k) + B̂o u(k)
(29)
yo (k) = Ĉo x̂o (k) + D̂o u(k)

is observable and has the same transfer function as the original state
equation (20).
25 / 1
Canonical Decomposition

z Domain Controllability and Observability Test

Let be the transfer function of a single-input single-output

system

H(z) = c(zI − A)−1 b (30)

Theorem
If there are no zero-pole cancellations in the transfer function
H(z) of a single-input single-output system, then the system is
both controllable and observable. If the zero-pole cancellation
occurs in , then the system is either uncontrollable or
unobservable or both uncontrollable and unobservable.

26 / 1
Canonical Decomposition

27 / 1
Canonical Decomposition

28 / 1