state space analysis - 20.

1
20. STATE SPACE ANALYSIS
20.1 INTRODUCTION
- state equations can be converted to transfer functions. The derivation follows.
Topics:
Objectives:
state space analysis - 20.2
- state equation coefficient matrices can be transformed to another equivalent for,
if the state vector is rearranged.
x

Ax Bu + =
y Cx Du + =
State equations as functions of time
sX X
0
AX BU + =
Y CX DU + =
In the s-domain
X sI A ( ) BU X
0
+ =
X sI A ( )
1
BU sI A ( )
1
X
0
+ =
Y C sI A ( )
1
BU sI A ( )
1
X
0
+ ( ) DU + =
Y C sI A ( )
1
B D + ( )U C sI A ( )
1
X
0
+ =
Assuming the system starts at rest,
Y C sI A ( )
1
B D + ( )U =
Y
U
---- C sI A ( )
1
B D + ( ) =
(the transfer function)
state space analysis - 20.3

x

Ax Bu + =
y Cx Du + =
State equations as functions of time
z

T
1
Ax Bu + ( ) =
This can be used to calculate new coefficient matrices
x Tz =
Map one state vector to another one
z

T
1
ATz Bu + ( ) =
z

T
1
ATz T
1
Bu + =
z

T
1
ATz T
1
Bu + = A T
1
AT = B T
1
B =
z

Az Bu + =
y Cx Du + =
The output equation becomes
y CTz Du + =
y CTz Du + = C CT = D D =
y Cz Du + =
The equivalent set of state equations is,
A T
1
AT = B T
1
B = z

Az Bu + =
C CT = D D = y Cz Du + =
state space analysis - 20.4
The transfer function form can be put into a matrix form. In this case the denom-
inator is the characteristic equation.
The free (homogeneous) response of a system can be used to find the state transi-
tion matrix.
G C sI A ( )
1
B D + =
G C sI A ( )
1
B D + =
The two following transfer functions are equivalents
G CTT
1
sI A ( )
1
TT
1
B TT
1
D + =
This can be shown by applying the transformation matrix
G CT ( ) T
1
sI A ( )
1
T ( ) T
1
B ( ) D + =
G CT ( ) sI A ( )
1
( ) T
1
B ( ) D + =
G C sI A ( )
1
( )B D + G = =
G C sI A ( )
1
B D +
sI A ( ) B
C D
sI A
------------------------------------
poles
zeros
-------------- = = =
The transfer function can be said to be equivalant to the determinants of the matrix form.
sI A ( ) B
C D
sI A
------------------------------------
sI A ( )D B ( )C
sI A
--------------------------------------------- - sI A ( )D BC sI A ( ) + = =
sI A characteristic equation homogeneous = =
state space analysis - 20.5
The forced response (particular) response of the system can be found using con-
volution,
As an example the homogeneous/free response of the system is shown below.
X
h
sI A X
0
=
x
h
t ( ) L
1
sI A X
0
[ ] =
The homogeneous equation can be written in the s-domain, and then converted to time.
x
h
t ( ) L
1 I
s
--
A
s
---
A
2
s
2
------
A
3
s
3
------ + + + + X
0
=
x
h
t ( ) e
At
x
0
= e
At
transition matrix =
e
At
I At
1
2!
---- -
\ .
| |
A
2
t
2 1
3!
---- -
\ .
| |
A
3
t
3
+ + + + =
aside: This expansion is a McLaurin (Taylor) series.
The homogeneous equation can be written in the s-domain, and then converted to time.
x

Ax Bu + =
y Cx Du + =
x t ( ) e
At
x 0 ( ) e
A t ( )
Bu ( ) d
0
t

+ =
y t ( ) Ce
At
x 0 ( ) Ce
A t ( )
Bu ( ) d
0
t

Du t ( ) + + =
initial
response
impulse
response
state space analysis - 20.6
The forced/particular solution is shown below,
If a matrix is diagonizable, the diagonal matrix can be found with the following
technique. This can be used for more advanced analysis techniques to create diagonal (and
F Mx

=
x

v =
v
F
M
----- =
d
dt
-----
x
v
0 1
0 0
x
v
0
1
M
---- -
F + =
A
0 1
0 0
=
sI A ( )
1
s
1 0
0 1
0 1
0 0

\ .
|
| |
1
s 1
0 s
1
s 1
0 s
s
2
0
-------------
s
s
2
----
1
s
2
----
0
s
2
----
s
s
2
----
1
s
---
1
s
2
----
0
1
s
---
= = = = =
Given a second order differential equation, the state matrix can be found,
This can be used to find the inverse matrix,
e
At
L
1
sI A ( )
1
[ ]
1 t
0 1
= =
The function of time can be found assuming an initial position of 10 and velocity of 5.
x
h
t ( )
x
v
e
At
10
5
1 t
0 1
10
5
10 5t +
5
= = = =
F Mx

=
Given a second order differential equation, the state matrix can be found,
state space analysis - 20.7
separable) system matrices.
example,
T
1
AT =
Given a matrix A, a purely diagonal matrix may be found,
e
At
Te
t
T
1
=
This may also be written as,
e
t
e

1
t

n
t
=
e
At
I At
1
2!
-----A
2
t
2
+ + + =
This can also be applied to the
e
At
I TT
1
t
1
2!
---- - TT
1
( )
2
t
2
+ + + =

n
= A TT
1
=
I A 0 =
Note: the Eigenvalues are the values
found in the characteristic/homoge-
neous equation solution.
I A ( )v 0 =
The Eigenvalues are then used to find the Eigenvectors,
Av v =
The Eigenvectors are then combined into a single matrix, this is the transition matrix.
Eigenvalues can be found using the following relationship. If any of the Eigen-
values are repeated the Jordan normal form is required.
state space analysis - 20.8
If there are repeated Eigenvalues in the system the Jordan Form can be used.
A
1 2
3 4
=
Given the state matrix
I A
1 0
0 1
1 2
3 4

1 + 2
3 4
= =
The eigenvalues can be calculated,
1 + ( ) 4 ( ) 6 ( )
2
3 2 + 2 ( ) 1 ( ) = = =
Therefore the Eigenvalues are,
1 2
3 4
u
11
u
21

1
u
11
u
21
2
u
11
u
21
= =
An eigenvector can be calculated using the Eigenvalue 2.
u
11
2u
21
+ 2u
11
= 2 u
21
3u
11
=
v
1
u
11
u
21
2
3
= =
1 2
3 4
u
12
u
22

2
u
12
u
22
1
u
12
u
22
= =
Another eigenvector can be calculated using the Eigenvalue 1.
u
12
2u
22
+ u
12
= u
22
u
12
=
v
2
u
12
u
22
1
1
= =
T
v
1
v
2
2 1
3 1
= =
The Eigenvectors can be combined to a single matrix.
To check,
A
u
11
u
21
A
u
12
u
22
, 2
2
3
1
1
1
, =
1 2
3 4
2
3
1 2
3 4
1
1
,
4
6
1
1
, =
4
6
1
1
,
4
6
1
1
, =
verified

1
2 =
2
1 =
Av
i

i
v
i
=
state space analysis - 20.9
J

1
1 0 0 0
0
2
1 0 0
0 0
3
0 0

0 0 0
n 1
1
0 0 0 0
n
= Jordan block
A
2 1 1
1 2 1
1 1 2
=
Consider the example below with repeated eigenvalues.
I A
2 1 1
1 2 1
1 1 2
3 + ( )
2
0 ( ) = =
Find a matrix, that has repeated eigenvalues, and is not singular
state space analysis - 20.10
1
T
11
T
21
T
31
2 1 1
1 2 1
1 1 2
T
11
T
21
T
31
2 T
11
T
21
T
31
+ +
T
11
2T
21
T
31
+
T
11
T
21
2T
31
+
= =
Solve for the Jordan block one column at a time,
1
T
12
T
22
T
32
2 1 1
1 2 1
1 1 2
T
12
T
22
T
32
2 T
12
T
22
T
32
+ +
T
12
2T
22
T
32
+
T
12
T
22
2T
32
+
= =
0
T
13
T
23
T
33
2 1 1
1 2 1
1 1 2
T
13
T
23
T
33
2 T
13
T
23
T
33
+ +
T
13
2T
23
T
33
+
T
13
T
23
2T
33
+
= =
T
11
2T
11
T
21
T
31
+ + =
T
21
T
11
2T
21
T
31
+ + =
T
31
T
11
T
21
2T
31
+ + =
T
12
2T
12
T
22
T
32
+ + =
T
22
T
12
2T
22
T
32
+ + =
T
32
T
12
T
22
2T
32
+ + =
4T
13
2T
13
T
23
T
33
+ + =
4T
23
T
13
2T
23
T
33
+ + =
4T
33
T
13
T
23
2T
33
+ + =
T
21
=
T
21
=
T
31
=
FIX
state space analysis - 20.11
The Eigenvectors can be used to calculate the system response.
zeros of state space functions can be found using the state matrices.
Given the following Eigen values and vectors the free response is,
v
1
2
3
= v
2
1
1
=
1
2 =
2
1 =
x
h
t ( ) v
i
e

i
t
i 1 =
n

2
3
e
2t
1
1
e
1t
+
2e
2t
e
t
+
3e
2t
e
t
+
= = =
state space analysis - 20.12
The following relationship locates the zero frequencies in state equations.
s
0
I A B
C D
0 =
Consider the example beginning with state equations,
s
0
1 0
0 1
0 1
3 4

\ .
|
| |
5
5

1 0
0
0 =
A
0 1
3 4
= B
5
5
= C
1 0
= D 0 =
x

v 5F =
v

3x 4v 5F + =
d
dt
-----
x
v
0 1
3 4
x
v
5
5
F + =
x
1 0
x
v
0F + =
This can be put into the matrix form,
s
0
1 5
3 s
0
4 + 5
1 0 0
0 =
s
0
0 ( ) 1 ( ) 4.5 ( ) 5 s
0
4 + ( ) ( ) + 0 =
24.5 5s
0
=
s
0
4.9 =
There is one zero at -4.9 rad/sec. The order of the polynomial determines the
number of zeros. If the polynomial has no s terms, then there are no zeros.
state space analysis - 20.13
20.2 OBSERVABILITY
- a system is observable iff the system state x(t) can be found by observing the
input u and output y over a period of time from x(t) to x(t+h).
- If an input is to be observable it must be detectable in the output. For example
consider the following state equations.
d
dt
-----
x
v
a b
c d
x
v
e
f
u + =
y
g h
x
v
j
k
u + =
if the parameters c, d or f were not in the final expression for y, the second state equa-
tion would be said to be not observable. This will happen when an output parameter
is zero in the C matrix and an equation is decoupled in the A matrix, i.e. the same
variable column is zero.
For example,
d
dt
-----
x
v
1 0
0 a
x
v
3
b
u + =
y
5 0
x
v
6
u + =
G C sI A ( )
1
B D +
5 0
s 0
0 s
1 0
0 a

\ .
|
| |
1
3
b
6
+ = =
G
5 0
s 1 0
0 s a + \ .
|
| |
1
3
b
6
+
5 0
s a + 0
0 s 1
s 1 ( ) s a + ( )
---------------------------------
\ .
|
|
|
|
| |
3
b
6
+ = =
G
5 0
1
s 1
----------- 0
0
1
s a +
-----------
3
b
6
+
5 0
3
s 1
-----------
b
s a +
-----------
6
+
15
s 1
----------- 6 + = = =
Note: the variables a and b both dissapeared because this system is not
observable.
state space analysis - 20.14
This often happens when a system has elements that are decoupled, or when a
pole and zero cancel each other.
Observability can be verified formally for an LTI system with the following rela-
tionship.
Another theorem for testing observability is given below. If any of the states sat-
isfies the equation it is unobservable.
Ce
At
x
*
0 = t 0
The system is unobservable if the equation below is true for any state,
A
1 2
3 4
=
One example,
x
*
1
0

0
0
1

0
0

1
, , =
C
5 6
=
Ce
At
x
*
5 6
e
t
e
2t
e
3t
e
4t
1
0
0
1
,
)
`

5e
t
6e
3t
+ ( ) 5e
2t
6e
4t
+ ( )
1
0
0
1
,
)
`

= =
Ce
At
x
*
5e
t
6e
3t
+ ( ) 5e
2t
6e
4t
+ ( )
1
0
5e
t
6e
3t
+ ( ) 5e
2t
6e
4t
+ ( )
0
1
,
)
`

=
Both values are non-zero, so both states are observable.
Ce
At
x
*
5e
t
6e
3t
+ ( ) 5e
2t
6e
4t
+ ( ) , { } =
A
1 0
0 2
=
One example,
C
3 0
=
Ce
At
x
*
3 0
e
t
0
0 e
2t
1
0
0
1
,
)
`

3e
t
0
1
0
0
1
,
)
`

3e
t
0 , { } = = =
The second state is not observable.
state space analysis - 20.15
Yet another test for observability is,
If a system in unobservable, it is possible to make it observable by changing the
model.
A pole-zero cancellation is often the cause of the loss of observability.
If all unstable modes are observable, the system is detectable.
20.3 CONTROLLABILITY
- a system is controllable iff there is an input u(t) that will cause the system to go
from any initial state to any final state in a finite time.
- stabilizable if it is controllable or if the uncontrollable nodes are stable.
- If an input is to be observable it must be detectable in the output. For example
consider the following state equations.
C
CA

CA
n 1
x
*
0 =
rank M
o
( ) rank
C
CA

CA
n 1
n = =
state space analysis - 20.16
This can be verified with the
d
dt
-----
x
v
a b
c d
x
v
e
f
u + =
y
g h
x
v
j
k
u + =
if the parameters c, d or f were not in the final expression for y, the second state equa-
tion would be said to be not uncontrollable. This will happen when an output
parameter is zero in the C matrix and an equation is decoupled in the A matrix, i.e.
the same variable column is zero.
For example,
d
dt
-----
x
v
1 0
a b
x
v
2
0
u + =
y
3 4
x
v
5
u + =
G C sI A ( )
1
B D +
3 4
s 0
0 s
1 0
a b

\ .
|
| |
1
2
0
5
+ = =
Note: the variables a and b both dissapeared because this system is not con-
trollable.
G
3 4
s 1 0
0 s b +
1
2
0
5
+
3 4
s B + 0
0 s 1
s 1 ( ) s b + ( )
---------------------------------
2
0
5
+ = =
G
3 4
1
s 1
----------- 0
0
1
s b +
-----------
2
0
5
+
3
s 1
-----------
4
s b +
-----------
2
0
5
+
6
s 1
----------- 5 + = = =
state space analysis - 20.17
Another test for controllability is,
Yet another test for controllability is,
x
*
( )
T
e
At
B 0 = t 0
The basic relationship is,
1
0
0
1
,
)
`

T
e
1 2
3 4
t
5
6
1 0 0 1
,
)
`

e
1t
e
2t
e
3t
e
4t
5
6
=
An example is,
A
1 2
3 4
= B
5
6
=
1 0 0 1
,
)
`

5e
1t
6e
2t
+
5e
3t
6e
4t
+
5e
1t
6e
2t
+ 5e
3t
6e
4t
+ , { } =
The results are both non-zero, so both states are controllable.
1
0
0
1
,
)
`

T
e
1 0
0 2
t
0
3
1 0 0 1
,
)
`

e
1t
0
0 e
2t
0
3
=
An example is,
A
1 0
0 2
= B
0
3
=
1 0 0 1
,
)
`
0
3e
2t
0 3e
2t
, { } =
The first state has a zero, so it in not controllable.
x
*
( )
T
B AB A
n 1
B
state space analysis - 20.18
For a system to be controllable, all of the states must be controllable.
If a system in uncontrollable, it is possible to make it observable by changing the
model.
A pole-zero cancellation is often the cause of the loss of observability.
if all unstable modes are controllable, the systems is said to be stabilizable.
The principle of Duality requires that a system be completely observable to be
controllable.
20.4 OBSERVERS
Observers estimate system state variables when not all of the variables are
directly observable.
Observers use a limited set of system states that are available to identify other
system states that are not observable.
The separation principle ensures that the observer cannot effect the stability of
the system it is observing.
20.5 SUMMARY

rank M
c
( ) rank
B AB A
n 1
B
n = =
state space analysis - 20.19
20.6 PRACTICE PROBLEMS
20.7 PRACTICE PROBLEM SOLUTIONS
20.8 ASSIGNMENT PROBLEMS
1.
20.9 BIBLIOGRAPHY
How, J, "16.31 Feedback Control Course Notes", MIT Opencourseware website.