Linear Control Systems

Lecture # 8
Observability
&
Discrete-Time Systems
p. 1/25
Denition: The system
x = Ax, y = Cx
or the pair (A, C), is said to be observable on [t
0
, t
f
] if any
initial state x(t
0
) = x
0
can be uniquely determined from
y(t) on [t
0
, t
f
]. It is said to be (re)constructible on [t
0
, t
f
] if
any nal state x(t
f
) = x
f
can be uniquely determined from
y(t) on [t
0
, t
f
]
Without loss of generality, take t
0
= 0
The observability Gramian of (A, C) is dened by
W
o
(0, t
f
) =
_
t
f
0
e
A
T
t
C
T
Ce
At
dt
p. 2/25
Lemma: W
o
(0, t
f
) is positive denite if and only if there is
no vector x
a
= 0 such that
Ce
At
x
a
0, t [0, t
f
]
Proof:
x
T
a
W
o
(0, t
f
)x
a
=
_
t
f
0
x
T
a
e
A
T
t
C
T
Ce
At
x
a
dt
x
T
a
W
o
(0, t
f
)x
a
= 0 Ce
At
x
a
0, t [0, t
f
]
p. 3/25
Theorem: The pair (A, C) is observable (constructible) on
[0, t
f
] if and only if the observability Gramian W
o
(0, t
f
) is
positive denite
Proof of sufciency: Suppose W
o
(0, t
f
) is positive denite
and let x
0
be the initial state
y(t) = Cx(t) = Ce
At
x
0
Multiply from the left by e
A
T
t
C
T
e
A
T
t
C
T
y(t) = e
A
T
t
C
T
Ce
At
x
0
Integrate from 0 to t
f
_
t
f
0
e
A
T
t
C
T
y(t) dt =
_
t
f
0
e
A
T
t
C
T
Ce
At
dt x
0
p. 4/25
_
t
f
0
e
A
T
t
C
T
y(t) dt = W
o
(0, t
f
)x
0
This equation has a unique solution
x
0
= W
1
o
(0, t
f
)
_
t
f
0
e
A
T
t
C
T
y(t) dt
Since
x(t
f
) = e
At
f
x
0
and e
At
f
is nonsingular, x(t
f
) is determined uniquely by
x(t
f
) = e
At
f
W
1
o
(0, t
f
)
_
t
f
0
e
A
T
t
C
T
y(t) dt
p. 5/25
Proof of Necessity: We want to show that positive
deniteness of W
o
(0, t
f
) is a necessary condition for
observability (constructibility) over [0, t
f
]
Suppose W
o
(0, t
f
) is not positive denite. Then, there is a
vector x
a
= 0 such that
Ce
At
x
a
0, t [0, t
f
]
For the case of observability, The output due to x(0) = x
0
is
Ce
At
x
0
and the output due to x(0) = x
0
+x
a
is
Ce
At
(x
0
+x
a
) = Ce
At
x
0
+Ce
At
x
a
= Ce
At
x
0
p. 6/25
The initial states x
0
and x
0
+x
a
produce the same output.
Hence, x
0
cannot be uniquely determined from the output
For the case of constructibility
y(t) = Ce
At
x(0) = Ce
At
e
At
f
x(t
f
)
x(t
f
) = x
f
and x(t
f
) = x
f
+e
At
f
x
a
correspond to the
same output because
Ce
At
e
At
f
_
x
f
+e
At
f
x
a
_
= Ce
At
e
At
f
x
f
+Ce
At
e
At
f
e
At
f
x
a
= Ce
At
e
At
f
x
f
+Ce
At
x
a
= Ce
At
e
At
f
x
f
p. 7/25
Lemma: The observability Gramian W
o
(0, t
f
) is positive
denite if and only if rank O = n, where
O =
_

_
C
CA
.
.
.
CA
n1
_

_
is the observability matrix (O is np n when A is n n
and C is p n)
p. 8/25
Proof:
W
o
(0, t
f
) is singular if and only if
there is x
a
= 0 such that Ce
At
x
a
0, t [0, t
f
]
C

k=0
t
k
k!
A
k
x
a
0, t [0, t
f
]
CA
k
x
a
= 0, k 0
CA
k
x
a
= 0, k = 0, 1, . . . , n 1
p. 9/25

_
C
CA
.
.
.
CA
n1
_

_
x
a
= 0
rank
_

_
C
CA
.
.
.
CA
n1
_

_
< n
Theorem: The pair (A, C) is observable (reconstructible)
if and only if rank O = n
p. 10/25
Duality
(A, B) controllable rank [B, AB, . . . , A
n1
B] = n
rank
_

_
B
T
B
T
A
T
.
.
.
B
T
(A
T
)
n1
_

_
= n

_
A
T
, B
T
_
observable
Similarly, observability of (A, C) is equivalent to
controllability of
_
A
T
, C
T
_
p. 11/25
Controllability and observability are invariant to state
transformations
{A, B, C, D} x = Pz {P
1
AP, P
1
B, CP, D}
rank
_
P
1
B, P
1
APP
1
B, . . . ,
_
P
1
AP
_
n1
P
1
B
_
= rank
_
P
1
B, P
1
AB, . . . , P
1
A
n1
B

= rank
_
P
1
[B, AB, . . . , A
n1
B]
_
= rank [B, AB, . . . , A
n1
B]
Matlab:
O = obsv(A, C); rank(O)
p. 12/25
Mathematical Preliminaries: Consider the equation
y = Mx
where M is an mn matrix. The equation has a solution
if and only if y can be expressed as a linear combination of
the columns of M
rank M = rank [M, y]
The range space of M, denoted by R(M), is the set of all
linear combinations of the columns of M. It is a subspace
of R
m
of dimension equal to rank M. The equation
y = Mx has a solution if and only if y R(M). The
equation y = Mx has a solution for every y R
m
if and
only if
rank(M) = m R(M) = R
m
p. 13/25
Given that y = Mx has a solution. Is the solution unique?
Suppose there are two solutions x
a
and x
b
such that
y = Mx
a
, y = Mx
b
M(x
a
x
b
) = 0
The null space of M, denoted by N(M), is the set of all
vectors x R
n
such that Mx = 0. It is a subspace of R
n
of dimension (n rank M)
The solution of y = Mx is unique if and only if there is no
vector x = 0 such that Mx = 0
Equivalently, rank M = n
p. 14/25
Discrete-Time Systems
The denitions of controllability (reachability) and
observability (constructibility) are the same as in the
continuous-time case
Let us study steering the state of the system
x(k + 1) = Ax(k) +Bu(k)
Find u(k) to steer the state of the system from x
0
to x
f
in
nite time
x(k) = A
k
x(0) +
k1

j=0
A
kj1
Bu(j)
p. 15/25
x(k) A
k
x
0
= [B, AB, . . . , A
k1
B]
_

_
u(k 1)
u(k 2)
.
.
.
u(0)
_

_
Can we choose u(0), . . . , u(k 1) such that x(k) = x
f
for
some k?
The answer is Yes if for some k,
x
f
A
k
x
0
R
_
[B, AB, . . . , A
k1
B]
_
p. 16/25
Due to Cayley-Hamilton theorem
R
_
[B, AB, . . . , A
k1
B]
_
= R
_
[B, AB, . . . , A
n1
B]
_
for all k n
Thus, we take k = n and rewrite the equation as
x
f
A
n1
x
0
= C
_

_
u(n 1)
u(n 2)
.
.
.
u(0)
_

_
def
= C U
p. 17/25
Reachability: x
0
= 0 and x
f
is any vector in R
n
x
f
= C U
There is a solution for every x
f
if and only if
R(C) = R
n
rank C = n
Controllability: x
f
= 0 and x
0
is any vector in R
n
A
n
x
0
= C U
There is a solution if and only if A
n
x
0
R(C)
rank C = n is a sufcient condition for the existence of a
solution. Is it necessary?
p. 18/25
If A is nonsingular, then the vector A
n
x
0
can be any vector
in R
n
. In this case, the condition rank C = n is necessary
If A is singular, the equation
A
n
x
0
= C U
may be solvable even if rank C < n
p. 19/25
Example:
A =
_
1 1
0 0
_
, B =
_
1
0
_
C = [B, AB] =
_
1 1
0 0
_
rank C = 1
p. 20/25

_
+
0
_
=
_
1 1
0 0
_ _
u(1)
u(0)
_
( +) = u(0) +u(1)
There are innitely many solutions. Two examples are
u(0) = , u(1) = 0
u(0) = , u(1) =
p. 21/25
Summary:
rank C = n is a necessary and sufcient condition for
reachability of x(k + 1) = Ax(k) +Bu(k)
rank C = n is a sufcient condition for controllability of
x(k + 1) = Ax(k) +Bu(k)
If A is nonsingular, then rank C = n is a necessary and
sufcient condition for controllability of
x(k + 1) = Ax(k) +Bu(k)
Remark: Strictly speaking, we should say that the pair
(A, B) is reachable if and only rank C = n. However, by
abuse of notation, we will continue to call the pair (A, B)
controllable if and only if rank C = n. This will keep the
discussion of controllability the same for both
continuous-time and discrete-time systems
p. 22/25
Observability: Consider the system
x(k + 1) = Ax(k), y(k) = Cx(k)
and study the question of uniquely determining x(0) from
the output y(k)
_

_
y(0)
y(1)
.
.
.
y(k 1)
_

_
=
_

_
Cx(0)
CAx(0)
.
.
.
CA
k1
x(0)
_

_
=
_

_
C
CA
.
.
.
CA
k1
_

_
x(0)
p. 23/25
The equation will have a unique solution x(0) if and only if
rank
_

_
C
CA
.
.
.
CA
k1
_

_
= n
By Cayley-Hamilton theorem
rank
_

_
C
CA
.
.
.
CA
k1
_

_
= rank
_

_
C
CA
.
.
.
CA
n1
_

_
, k n
p. 24/25
Take k = n and write the equation as
Y = O x(0)
where
Y =
_

_
y(0)
y(1)
.
.
.
y(n 1)
_

_
The equation has a unique solution if and only if
rank O = n
p. 25/25