Linear Control Systems Lecture # 25 Time-Varying Systems Change of Variables, Controllability & Observability
Linear Control Systems Lecture # 25 Time-Varying Systems Change of Variables, Controllability & Observability
Lecture # 25
Time-Varying Systems
Change of Variables,
Controllability & Observability
p. 1/24
Change of Variables
Consider the system
x = A(t)x
and the change of variables
x(t) = P(t)z(t)
where P(t) is nonsingular and continuously differentiable
for all t
x(t) = P(t) z(t) +
P(t)z(t)
A(t)x(t) = A(t)P(t)z(t)
x = A(t)x P(t) z(t) +
P(t)z(t) = A(t)P(t)z(t)
z(t) =
P
1
(t)A(t)P(t) P
1
(t)
P(t)
z(t)
def
= F(t)z(t)
p. 2/24
Let
A
(t, t
0
) and
F
(t, t
0
) be the transition matrices of
A(t) and F(t), respectively. What is the relationship
between
A
(t, t
0
) and
F
(t, t
0
)?
z(t) =
F
(t, t
0
)z(t
0
) =
F
(t, t
0
)P
1
(t
0
)x(t
0
)
On the other hand,
z(t) = P
1
(t)x(t) = P
1
(t)
A
(t, t
0
)x(t
0
)
P
1
(t)
A
(t, t
0
)x(t
0
) =
F
(t, t
0
)P
1
(t
0
)x(t
0
), x(t
0
)
P
1
(t)
A
(t, t
0
) =
F
(t, t
0
)P
1
(t
0
)
F
(t, t
0
) = P
1
(t)
A
(t, t
0
)P(t
0
)
p. 3/24
Periodic Systems
Consider the system x = A(t)x, where A(t) is a
continuous T-periodic function of t; i.e.,
A(t +T) = A(t), t
Let (t, ) be the transition matrix of A(t) and dene a
constant matrix R by
e
RT
= (T, 0)
There is always a matrix R (although it is not unique)
p. 4/24
Special Case: If (T, 0) is diagonalizable, then there is a
nonsingular matrix Q such that
Q
1
(T, 0)Q =
2
.
.
.
Take R
1
=
ln(
1
)
ln(
2
)
.
.
.
ln(
n
)
p. 5/24
Take R =
1
T
QR
1
Q
1
Then
exp(RT) = exp(QR
1
Q
1
) = Qexp(R
1
)Q
1
= Q
2
.
.
.
Q
1
= (T, 0)
p. 6/24
Lemma: Let P(t) = (t, 0)e
Rt
P(t) is nonsingular for all t
P(t +T) = P(t) for all t
Proof: P(t) is nonsingular because it is the product of two
nonsingular matrices
P(t +T) = (t +T, 0)e
R(t+T)
= (t +T, T)(T, 0)e
RT
e
Rt
= (t +T, T)e
RT
e
RT
e
Rt
= (t +T, T)e
Rt
Because A(t) is T-periodic, (t +T, T) = (t, 0)
P(t +T) = (t, 0)e
Rt
= P(t)
p. 7/24
P(t) =
t
[(t, 0)] e
Rt
+ (t, 0)
d
dt
e
Rt
= A(t)(t, 0)e
Rt
(t, 0)e
Rt
R
= A(t)P(t) P(t)R
Apply the change of variables x(t) = P(t)z(t)
z =
P
1
(t)A(t)P(t) P
1
(t)
P(t)
z(t)
= P
1
(t)
A(t)P(t)
P(t)
z(t)
= P
1
(t)P(t)R z(t) = R z(t)
Floquet decomposition: A T-periodic system x = A(t)x
can be transformed by a T-periodic change of variables
x(t) = P(t)z(t) into an equivalent time-invariant system.
p. 8/24
Example:
A(t) =
(1 + cos t) 0
0 (2 + cos t)
, T = 2
(t, 0) =
t
0
(1+cos ) d
0
0 e
t
0
(2+cos ) d
e
(t+sin t)
0
0 e
(2t+sint)
p. 9/24
Find R such that
e
RT
= (T, 0) =
e
(T+sin T)
0
0 e
(2T+sinT)
e
T
0
0 e
2T
R =
1 0
0 2
p. 10/24
P(t) = (t, 0)e
Rt
=
e
(t+sin t)
0
0 e
(2t+sint)
e
t
0
0 e
2t
e
sin t
0
0 e
sin t
p. 11/24
What is the effect of the change of variables on the
input-output response?
x(t) = A(t)x(t) +B(t)u(t)
y(t) = C(t)x(t) +D(t)u(t)
Zero-state response:
y(t) =
t
t
0
C(t)(t, )B()u() d
x(t) = P(t)z(t)
x(t) = A(t)x(t) +B(t)u(t)
P(t) z(t) +
P(t)z(t) = A(t)P(t)z(t) +B(t)u(t)
p. 12/24
z =
P
1
(t)A(t)P(t) P
1
(t)
P(t)
z(t)
+P
1
(t)B(t)u(t)
= F(t)z(t) +P
1
(t)B(t)u(t)
y(t) = C(t)x(t) +D(t)u(t)
= C(t)P(t)z(t) +D(t)u(t)
{A, B, C, D} x = Pz {F, P
1
B, CP, D}
F = P
1
AP P
1
P
p. 13/24
C(t)P(t)
F
(t, )P
1
()B()
= C(t)P(t)P
1
(t)
A
(t, )P()P
1
()B()
= C(t)
A
(t, )B()
The input-output response is invariant to state
transformations
p. 14/24
What is the effect of the change of variables on internal
stability?
A(t) F(t) = P
1
(t)A(t)P(t) P
1
(t)
P(t)
F
(t, t
0
) = P
1
(t)
A
(t, t
0
)P(t
0
)
Lyapunov Transformation: A change of variables
x(t) = P(t)z(t) is a Lyapunov transformation if both P(t)
and P
1
(t) are bounded and continuously differentiable for
all t
Uniform stability and uniform asymptotic stability are
invariant to Lyapunov transformations
p. 15/24
Example:
A(t) =
(1 + cos t) 0
0 (2 + cos t)
e
sin t
0
0 e
sin t
1 0
0 2
z
p. 16/24
x(t) = P(t)z(t) is a Lyapunov transformation because
both P(t) and P
1
(t) are bounded and continuously
differentiable for all t
The system z = Rz is asymptotically stable
Hence, the system x = A(t)x is uniformly asymptotically
stable
p. 17/24
Controllability
Denition: The system x = A(t)x +B(t)u, or the pair
(A(t), B(t)), is said to be controllable on [t
0
, t
f
] if given
any initial state x
0
, there is a continuous control u(t) that
steers the state of the system from x(t
0
) = x
0
to
x(t
f
) = 0. It is said to be reachable on [t
0
, t
f
] if given any
nal state x
f
, there is a continuous control u(t) that steers
the state of the system from x(t
0
) = 0 to x(t
f
) = x
f
The Controllability Gramian of (A(t), B(t)) is dened by
W
c
(t
0
, t
f
) =
t
f
t
0
(t
0
, t)B(t)B
T
(t)
T
(t
0
, t) dt
In the time-invariant case we took t
0
= 0 & (t
0
, t) = e
At
p. 18/24
Lemma: W
c
(t
0
, t
f
) is positive denite if and only if there is
no vector x
a
= 0 such that
x
T
a
(t
0
, t)B(t) 0, t [t
0
, t
f
]
Theorem: The pair (A(t), B(t)) is controllable (reachable)
on [t
0
, t
f
] if and only if the controllability Gramian
W
c
(t
0
, t
f
) is positive denite
Proofs are the same as in the time-invariant case
p. 19/24
Example:
A =
0 0
0 1
, B =
1
e
2t
e
At
=
1 0
0 e
t
, (t
0
, t) =
1 0
0 e
(t
0
t)
(t
0
, t)B(t) =
1 0
0 e
(t
0
t)
1
e
2t
1
e
(t+t
0
)
Is there x
a
= 0 such that x
T
a
(t
0
, t)B(t) 0 for all
t [t
0
, t
f
]?
p. 20/24
x
a
=
, x
T
a
(t
0
, t)B(t) = +e
(t+t
0
)
Can we nd and (not both zero) such that
+e
(t+t
0
)
0 for all t [t
0
, t
f
] ?
The answer is NO. The system is controllable on any
interval [t
0
, t
f
]
p. 21/24
Alternative method:
W
c
(t
0
, t
f
) =
t
f
t
0
(t
0
, t)B(t)B
T
(t)
T
(t
0
, t) dt
=
t
f
t
0
1
e
(t+t
0
)
1 e
(t+t
0
)
dt
=
(t
f
t
0
) W
c12
W
c12
W
c22
W
c12
=
e
(t
f
+t
0
)
e
2t
0
W
c22
=
1
2
e
2(t
f
+t
0
)
e
4t
0
t
f
t
0
T
(t, t
0
)C
T
(t)C(t)(t, t
0
) dt
p. 23/24
Lemma: W
o
(t
0
, t
f
) is positive denite if and only if there is
no vector x
a
= 0 such that
C(t)(t, t
0
)x
a
0, t [t
0
, t
f
]
Theorem: The pair (A(t), C(t)) is observable
(constructible) on [t
0
, t
f
] if and only if the observability
Gramian W
o
(t
0
, t
f
) is positive denite
p. 24/24