Professional Documents
Culture Documents
Observer Design
1. Introduction
2. Controllability
3. Observability
4. Useful Transformation
6. State Observers
7. Servo Systems
VI.1. Introduction
Controllability and observability play an important role in the optimal control of
multivariable systems.
Pole placement will be used to design the controller and the observer.
1
VI.2. Controllability
A control system is said to be completely state controllable if it is possible to transfer the
system from any arbitrary initial state to any desired state in a finite time period.
u n 1T
u n 2T
n
xnT G x0 H GH G H
n 1
u 0
2
G = n n matrix
H = n r matrix
T = Sampling Period
1 0
2
P 1GP
0 n
Note: if the eigenvalues of G are distinct then the eigenvectors of G are distinct. The
converse is not true.
The condition for complete state controllability is that, if the eigenvectors of G are
distinct, then the system is completely state controllable if and only if no row of
P 1 H has all zero elements.
If the G matrix does not possess distinct eigenvectors, then diagnalization is impossible.
We may transform G into a Jordan canonical form.
1 1 0
1 1
1
2 1
J
2
3 0
4
5
0 6
S 1GS J
If we define a new state vector x̂ by xkT Sx kT , substitute into 6.1,
3
x k 1T S 1GSx kT S 1 Hu kT
6.2
Jx kT S 1 Hu kT
The condition for the complete state controllability of the system of above equation is as
follows:
The system is completely state controllable if and only if 1) no two Jordan blocks in J of
6.2 are associated with the same eigenvalues, 2) The elements of any row of S 1 H that
corresponds to the last row of each Jordan block are not all zero and 3) The elements of
each row of S 1 H that correspond to distinct eigenvalues are not all zero.
x1 k 1 a b x1 k 1
x k 1 c d x k 1u k
2 2
x k
y k 1 0 1
x 2 k
Determine the conditions on a,b,c and d for complete state controllability.
1 a b
rank H GH rank 2 ab cd
1 c d
Where
T = Sampling Period
Define output controllability matrix C H CGH CG n 1 H
The condition for complete output controllability is that the matrix
4
C H
CGH CG n 1 H be of rank m, or that
Rank C H CGH CG n 1 H m
The system defined by equation 6.3 and 6.4 is said to be completed output controllable,
or simply output controllable, if it is possible to construct an unconstrained control signal
u kT defined over a finite number of sampling periods 0 kT nT such that, starting
from any initial output y 0 , the output y kT can be transferred to the desired point
y f in the output space in at most n sampling periods.
u n 1T
u n 2 T
xnT G n x0 H GH G n 1 H
u 0
Thus
u n 1T
u n 2T
Cx nT CG n x0 C H CGH CG n 1 H
u 0
So:
u n 1T
u n 2T
n
y nT CG x0 C H CGH CG H
n 1
u 0
Where
5
G = n n matrix
H = n r matrix
C = m n matrix
D = m r matrix
T = Sampling Period
Define output controllability matrix D C H CGH CG n 1 H
The condition for complete output controllability is that the matrix
D C H CGH CG n 1 H be of rank m, or that
Rank D C H CGH CG n 1 H m
u n 1T
u n 2 T
Cx nT CG n x0 C H CGH CG n 1 H
u 0
x Ax Bu
y Cx Du
6
A= n n matrix
B= n r matrix
C= m n matrix
D= m r matrix
rank B AB A n 1 B n
Output controllability
rank D CB CAB CA n 1 B m
7
VI.3. Observability
Consider the unforced discrete time control system defined by
Where
T = Sampling Period
The system is said to be completely observable if every initial state x0 can be
determined from the observation y kT over a finite number of sampling periods.
Or
rank C * G * C * G *n1 C * n
Alternative form of the condition for complete observability
xk 1T GxkT 6.9
y kT CxkT 6.10
Define :
xkT Px kT
8
1n 0
P
n
1 n 2
GP
n
0 n
1n 0 1n x1 0
n
n
y nT CP P GP x 0 CP
1 n2 x 0 CP 2 x 2 0
n n
0 n n x n 0
The system is completely observable if and only if none of the columns of the m n matrix CP
consists of all zero elements.
Note: if the ith column of CP consists of all zero elements, then the state variable xi 0 will not
appear in the output equation and therefore can not be determined from observation of y kT .
Thus x(0), which is related to x 0 by P, can not be determined.
If the matrix G involves multiple engenvalues, then G may be transformed into Jordan canonical
form:
S 1GS J
If we define a new state vector x̂ by xkT Sx kT , substitute into 6.7,
x k 1T S 1GSx kT
Jx kT
n
y kT CSx kT y nT CS S 1GS x 0
The condition for the complete observalibility of the system of above equation is as
follows:
The system is completely observable if and only if 1) no two Jordan blocks in J are
associated with the same eigenvalues, 2) none of the columns of CS that corresponds to
the first row of each Jordan block consists of all zero elements 3) The elements of each
column of CS that correspond to distinct eigenvalues are not all zero.
x1 k 1 a b x1 k 1
x k 1 c d x k 1u k
2 2
9
x k
y k 1 0 1
x 2 k
Determine the conditions on a, b, c and d for complete observability.
1 a
rank C * G * C * rank 2b0
0 b
Note: a necessary and sufficient condition for complete observability is that no pole-zero
cancellation occur in the pulse transfer function. If cancellation occurs, the canceled mode can
not be observed in the output.
Principle of Duality.
Where
T = Sampling Period
Where
x kT =n-vector (state vector at kth sampling instant)
u kT =m-vector (control signal at kth sampling instant)
y kT =r-vector (output vector at kth sampling instant)
G *= conjugate transpose of G
10
H *= conjugate transpose of H
C*= conjugate transpose of C
T = Sampling Period
Note: The analogy between controllability and observability is referred to as the principle of
duality, due to kalman. The principle of duality states that the system S1 defined by equations
6.11-12 is completely state controllable (observable) if and only if system S2 defined by
equation 6.13-14 is completely observable (state Controllable).
rank H GH G n 1 H n
2) A necessary and sufficient condition for complete observability is that
rank C * G * C * G *n1 C * n
For system S2:
rank C * G * C * G *n1 C * n
2) A necessary and sufficient condition for complete observability is that
rank H GH G n 1 H n
Complete Observability Linear Time-Invariant Continuous-Time Control system.
x Ax
y Cx
A = n n matrix
C = m n matrix
11
rank C * A * C * A *n1 C * n
Note: It can be shown that a system that is completely state controllable and completely
observable in the absence of sampling remains completely state controllable and completely
observable after introduction of sampling if and only if, for every eigenvalue of the characteristic
equation for the continuous-time control system, the relationship
Re i Re j
Implies
2 n
Imi j
T
Where T is the sampling period and n 1,2, It is noted that, unless the system contains
complex poles, pole-zero cancellation will not occur in passing from the continuous-time to the
discrete time case.
x1 0 2 x1 0
x 2 0 x 1u
2 2
x
y 1 0 1
x2
Determine the controllability and observability of the continuous time system and the
corresponding discrete time control system.
0 2
rank B AB rank 2 controllable
1 0
1 0
rank C * A * C * rank 2 observable
0 2
12
cos 2T sin 2T
G T e AT
cos 2T
,
sin 2T
cos 2T 1 sin 2T 1 0 2 0 0.5 cos 2T 0.5
H T (e AT I ) A 1 B
sin 2T cos 2T 1 4 2 0 1 0.5 sin 2T
x1 k 1 cos 2T sin 2T x1 k 0.5 cos 2T 0.5
x k 1 sin 2T cos 2T x k u k
2 2 0.5 sin 2T
x kT
y kT 1 0 1
x 2 kT
2 n
Imi j 4
T
Check:
0.5 cos 2T 0.5 0.5 cos 2 2T 0.5 cos 2T 0.5 sin 2 2T
rank H GH rank
0.5 sin 2T 0.5 sin 2T
n
Above matrix rank is 2 only if 0.5 sin 2T 0 T
2
13
VI.4. Useful Transformation
Transforming state-space equations into canonical forms
xk 1 Gxk Hu k 6.15
y k Cxk Du k 6.16
We will transform the equations 6.15-6.16 into the following three canonical forms:
1) Controllable canonical form 2) observable canonical form, 3) diagonal or Jordan
canonical form.
Now let us define xk Tx k
Then eq. 15-16 become
x k 1 T 1GTx k T 1 Hu k Gx k Hu k
y k CTx k Du k Cx k Du k
Where, G T 1GT and H T 1 H , C CT , D D , or
14
x1 k 1 0 1 0 0 x1 k 0
x k 1 0 0 1 0 x2 k 0
2
u k
xn 1 k 1 0 0 0 1 xn 1 k 0
xn k 1 an an 1 an 2
a1 xn k 1
x1 k
x k
y k bn anb0 bn 1 an 1b0 b1 a1b0 2 Du k
xn k
Where bk are those coefficients appearing in the numerator of the following pulse
transfer function.
1
C ZI G H D C ZI G H D n0
1 b z n b1 z n 1 bn 1 z bn
z a1 z n 1 an 1 z an
B) Observable canonical form
Q WN * , N C * G * C *
1
G *n1 C *
0 0 0 an
1 0 0 an 1
Q 1GQ G 0 1 0 an 2
0 0 1 a1
bn anb0
bn 1 an 1b0
Q H H
1
b1 a1b0
And
CQ C 0 0 0 1
By defining xk Qx k
Then eq. 15-16 become
x k 1 Gx k Hu k
y k Cx k Du k
15
Where, G Q 1GQ and H Q 1H , C CQ , D D , or
x1 k 1 0 0 0 an x1 k bn anb0
x k 1 1 0 0 an 1 x2 k bn 1 an 1b0
2
0 1 0 an 2 u k
xn 1 k 1 xn 1 k b2 a2b0
xn k 1 0 0 1
a1 xn k b1 a1b0
x1 k
x k
y k 0 0 0 1 2 Du k
xn k
P1 0 0
0 P2 0 0
P 1GP
0 0
0 0 Pn
By defining xk Px k
Then eq. 15-16 become
x k 1 Gx k Hu k
y k Cx k Du k
Where, G P 1GP and H P 1H , C CP , D D , or
16
x1 k 1 P1 0 0 0 x1 k 1
x k 1 0 P2 0 0 x 2 k 2
2
0 0 0 0 u k
x n 1 k 1 x n 1 k n 1
xn k 1 0 0 0
Pn x n k n
x1 k
x k
y k 1 2 n 2 Du k
xn k
Where the i and i are constants such that i i will appear in the numerator of the term
1
when the pulse transfer function is expanded into partial fractions as follows:
z pi
1
1
C ZI G H D C ZI G H D 1 1 2 2 n n D
z p1 z p2
z pn
In many cases we choose 1 2 n 1 .
Remarks: The sufficient and necessary conditions for the system to be completely state
controllable is that i 0 and sufficient and necessary conditions for the system to be
completely observable is i 0 .
The case for multiple eigenvalues pi of matrix G then Jordan canonical form will be
formed. ( skipped)
Invariance property of the rank condition for the controllability matrix and
observability matrix.
1) For controllability matrix M H GH G n 1H
~ ~
Let P be a transformation matrix and P 1GP G , P 1H H
~
P 1G 2 P P 1GPP 1GP G 2
~
P 1G 3 P P 1GPP 1GPP 1GP G 3
Then:
~
P 1G n 1P P 1GPP 1GPP 1GP G n 1
17
P 1M P 1 H GH G n 1H
P 1H P 1GH P 1G n 1H
P 1
H P GPP H P G n 1PP 1H
1 1 1
~ ~~ ~ ~
H GH G n 1H
~
M
~
Since matrix P is no singular, rank M= rank M
Similarly, for the observability matrix
N C * G * C * G * C *
n1
1 ~ ~
Let P be a transformation matrix and P GP G , CP C
P * N P * C * P * G * C * P * G * C *
n 1
Then
~
~ ~
~ n 1 ~
C * G *C * G * C *
~
N
~
Since P is non singular, rank N= rank N
18
VI.5. Design Via Pole Placement
Remarks: if the system is completely state controllable, then poles of the closed-loop
system may be placed at any desired locations by means of state feedback through an
appropriate state feedback gain matrix.
Where
Figure 6.1
Then the system becomes a closed loop control system and its state equation becomes
19
2) The sufficient conditions. We will approve that if the system is completely state
controllable, then we will find matrix K that make the eigenvalues of G HK as
desired.
ZI G z n a1 z n 1 an 1 z an 0
Figure 6.2
We define a transformation matrix T as follows:
T MW , where M H GH G n 1H , which is of rank n,
And
an 1 an 2 a1 1
a an 3 1 0
n2
W
a1 1 0 0
1 0 0 0
20
0 1 0 0
0 0 1 0
T 1GT G
0 0 0 1
an an 1 an 2 a1
And
0
0
T 1H H
0
1
Next we define K KT n n 1 1
0 0 0 0 0
0 0 0 0 0
Then HK n n 1 1
0 0 0 0 0
1 n n 1 n 2 1
The characteristic equation ZI G HK
Becomes as follows:
ZI G HK ZI G HK
1 0 0 0 0 1 0 0 0 0 0 0
0
1 0 0 0 0 1 0 0 0 0 0
z
0 0 0 0 0 0 0 1 0 0 0 0
0 0 0 1 an a n 1 a n2 a1 n n 1 n 2 1
z 1 0 0
0 z 1 0
0 0 0 1
a n n a n 1 n 1 a n2 n2 z a1 1
z n a1 1 z n 1 a n 1 n 1 z a n n 0
21
z u1 z u2 z un z n 1 z n 1 n 1 z n 0
1 a1 1
2 a2 2
n an n
Hence, from the equation we have
1
K KT
n n 1 1 T 1
n an n 1 an 1 1 a1 T 1
Ackermann’s Formula
By using the state feedback u k Kxk , we wish to place the closed loop poles at u1 , u2 ,un .
That is, we desire the characteristic equation to be
ZI G HK z u1 z u2 z un z n 1z n 1 n 1z n 0
Let us define G G HK
Since Cayley-hamilton theorem states that G satisfies its own characteristic equation, we have
G G n 1G n 1 n 1G n I 0
We now use this equation to derive ackerman’s formula.
II
G G HK
G 2 G HK G HK G 2 GHK HK G
G 3 G HK G HK G HK G HK G 2 GHK HKG G 3 G 2 HK GHKG HKG 2
G n G HK G HK G HK G n G n 1HK GHKG n 2 HKG n 1
Multiply the above equations in order by n , n 1 ,1 ,1 and adding the results
Left side is G =0.
Right side can be written as
22
G an 1HK an 2GHK an 2 HKG HKG n 1 G n 1HK
an 1K an 2 KG KG n 1
n2
G H GH G H n 1
a
n 2 K an 3 K G K G 0
K
Since the system is completely state controllable, the controllability matrix
H GH G n 1 H is of rank n and its inverse exists.
We can have
an 1K an 2 KG KG n 1
n2
a n 2 K a n 3 K G K G
H GH G n 1H
1
G
K
We have
K 0 0 1 H GH G n 1H 1
G 6-19
Once the desired characteristic equation is selected, there are several different ways to determine
the corresponding state feedback matrix K for a completely controllable system.
Method 1 :
K KT 1
n n 1 1 T 1
n an n 1 an 1 1 a1 T 1
Where
ZI G z n a1 z n 1 an 1 z an 0
23
T MW , where M H GH G n 1H , which is of rank n,
And
an 1 an 2 a1 1
a an 3 1 0
n2
W
a1 1 0 0
1 0 0 0
Method 2 : The desired state feedback gain matrix K can be given by Ackermann’s formula.
We have
K 0 0 1 H GH G n 1H 1
G
Method 3:
If the desired eigenvalues u1 , u2 ,un are distinct, then the desired state feedback gain matrix K
can be given as follows:
K 1 1 11 2 n
1
i G ui I 1 H , i 1,2, n
i G i H , i 1,2, n
Method 4: if the order of the system is low, substitute K into the characteristic equation.
ZI G HK =0 and then matches the coefficients of powers in z of this characteristic equation
with equal powers in z of the desired characteristic equations.
xk 1 Gxk Hu k
1 1 0
Where G , H ,
1 1 2
Note that
24
z 1 1
ZI G z 2 2z 2
1 z 1
Hence a1 2, a2 2
Determine a suitable feedback gain matrix such that the system will have the closed loop
pole at z 0.5 j 0.5, z 0.5 j 0.5,
Method 1:
0 2
M H GH , full rank controllable
2 2
a 1 2 1
W 1 , full rank
1 0 1 0
0 2 2 1 2 0
T MW
2 2 1 0 2 2
1
0 2
0 1 G
2 2
0.5 0.5
0 1 G
0.5 0
0.5 0 G
0.5
2
1 1 1 1 1
G G G 0.5I
2
0.5I
1 1 1 1 1 0.5
0.5 1
K 0.5 0G 0.5 0 0.25 0.5
1 0.5
Method 3:
25
K 1 11 2
1
i G ui I 1 H , i 1,2
1 G u1I 1 H
1
1 1 0
0.5 j 0.5I
1 1 2
1
0.5 j 0.5 1 0
1 0.5 j 0.5 2
1 2
1 0.5 j 1 j
2 G u2 I 1 H
1
1 1 0
0.5 j 0.5I
1 1 2
1
0.5 j 0.5 1 0
1 0.5 j 0.5 2
1 2
1 0.5 j 1 j
1
2 2 1 j 2
1 0.5 j
1 0.5 j 1 1 0.5 j 1 0.5 j
1 2 1
1 j 1 j 3.2 j 1 j 2
1 0.5 j 1 0.5 j 1 0.5 j 1 0.5 j
K 1 11 2
1
1 j 2
1 1 0.5 j 1 0.5 j
1 1 0.25 0.5
3.2 j 1 j 2
1 0.5 j 1 0.5 j
Method 4:
For lower order system, it will be simpler to substitute the K into the characteristic
equation.
26
ZI G HK
z 0 1 1 0
k1 k2
0 z 1 1 2
z 1 1
1 2k1 z 1 2k2
z 2 2k2 2z 2 2k2 2k1 0
The desired characteristic equation for the desired system is
ZI G z 0.5 j 0.5z 0.5 j 0.5 z 2 z 0.5 0
k1 0.25 , k2 0.5
Deadbeat Response.
If the eigenvalues of G HK lie inside the unit circle, then the system is asymptotically
stable.
xk 0, for k q, q n
Nilpotent matrix:
0 1 0 0
0 0 1 0
N nn , we have N n 0
0 0 0 1
0 0 0 0
27
Recall:
ZI G HK ZI G HK
1 0 0 0 0 1 0 0 0 0 0 0
0
1 0 0 0 0 1 0 0 0 0 0
z
0 0 0 0 0 0 0 1 0 0 0 0
0 0 0 1 a n a n 1 an2 a1 n n 1 n 2 1
z 1 0 0
0 z 1 0
0 0 0 1
a n n a n 1 n 1 an2 n2 z a1 1
z n a1 1 z n 1 a n 1 n 1 z a n n 0
when u1 , u2 ,un =0 we can easily get
G HK
0 1 0 0 0 0 0 0
0 0 1 0 0 0 0 0
0 0 0 1 0 0 0 0
a n a n 1 a n2 a1 n n 1 n 2 1
0 1 0 0
0 0 1 0
0 0 0 1
0 0 0 0
Which is a nilpotent matrix.
Thus we have G HK 0
n
28
Remarks:
If the desired eigenvalues are all zeros then any initial state x0 can be brought to the origin in
at most n sampling periods and the response is deadbeat, provided the control signal u k is
unbounded.
In deadbeat response, the sampling period is the only design parameter. The designer must
choose the sampling period carefully so that an extremely large control magnitude is not required
in normal operation of the system.
Trade off must be made between the magnitude of the control signal and the response speed.
xk 1 Gxk Hu k
1 1 0
Where G , H ,
1 1 2
Note that
z 1 1
ZI G z2 2z 2
1 z 1
Hence a1 2, a2 2
Determine a suitable feedback gain matrix such that the system will have the closed loop
pole at z 0 z 0 , which is dead beat response.
0 2 2 1 2 0
T MW
2 2 1 0 2 2
1 0, 2 0
1
2 0 0.5 0
K 2 a 2 1 a1 T 1
0 2 0 (2) 2 2 0 1
2 2 0.5 0.5
x1 k 1 1 1 1 x1 k 1 0 x1 k
x k 1 T 1 1T x k T 20 1T x k
2 2 2
0.5 0 1 1 2 0 x1 k 0.5 0 0 0 2 0 x1 k
0.5 0.5 1 1 2 2 x 2 k 0.5 0.5 0 2 2 2 x 2 k
0 1 x1 k
0 0 x 2 k
x1 0 a
For any initial state given by
x 2 0 b
29
x1 1 0 1 a b x1 2 0 1 b 0
x 1 0 0 b 0 and x 2 0 0 0 0
2 2
Thus the state X k for k=2,3,4… becomes zero and the response is indeed deadbeat.
Figure 6.3
ZI G HK 0
If the system is completely state controllable, then the feedback gain matrix K can be determined
to yield the desired closed-loop poles.
Remarks: the state feedback can change the characteristic equation for the system, it also
changes the steady state gain of the entire system. K 0 can be adjusted such that the unit-step
response of the system at steady state is unity.
30
VI.6. State Observers
Note:
1) In practice, not all state variables are available for direct measurement.
2) In many practical cases, only a few state variables of a given system are measurable.
3) Hence, it is necessary to estimate the state variables that are not directly measurable.
Such estimation is commonly called observation.
A state observer, also called a state estimator, is a subsystem in the control system that
performs an estimation of the state variables based on the measurements of the output and
control variables.
Full order state observation: estimate all n state variables regardless of whether some
state variables are available for direct measurement.
Minimum order state observation: observation of only the unmeasurable state variables.
Reduced order state observation: observation of all unmeasurable state variables plus
some of the measurable state variables.
Figure 6.4
31
xk state vector at kth sampling instant
u k control vector at kth sampling instant
y k output vector (m-vector)
G n n non singular matrix
H n r matrix
C m n matrix
Similarly
xk 2 G 3 xk 1 G 3 Hu k G 2 Hu k 1 G 1 Hu k 2
y k 2 Cx k 2 CG 3 xk 1 CG 3 Hu k CG 2 Hu k 1 CG 1Hu k 2
y k n 1 Cx k n 1 CG n xk 1 CG n Hu k CG n 1Hu k 1 CG 1Hu k n 1
Combining them
y k CG 1 CG 1H 0 0 u k
y k 1 2 2
CG xk 1 CG H CG H 1
0 u k 1
CG 1H
n n
y k n 1 CG CG n 1H CG H u k n 1
1
CG H
Or
32
CG 1 y k CG 1H 0 0 u k
2 2
CG xk 1 y k 1 CG H 0 u k 1
1
CG H
CG 1H
n n
y k n 1 CG H CG n 1H CG H u k n 1
1
CG
Note the right side of the equation is entirely known. Hence xk 1 can be determined if
and only if
CG 1 CG n 1
2 n2
CG is full rank. Or CG is full rank since G is not singular
n
CG C
Or C * G * C * G * C * G *
2 n 1
C * is of full rank.
CG 1 y k CG 1 CG 1H 0 u k
1 1
0
2 2 2
CG y k 1 CG CG H CG H 1
0 u k 1
xk 1
CG 1H
n n n
CG y k n 1 CG CG H CG H CG 1H u k n 1
n 1
Remarks:
1) xk 1 can be determined provided the system is completely observable.
2) in the presence of disturbance and measurement noise, xk 1 can not be estimated
accurately.
3) If C is not 1 by n matrix but is a m by n matrix, then the inverse of the matrix
CG 1
2
CG is not defined. To cope with this, we will use dynamic model.
n
CG
x k 1 G~
~ x k Hu k 6-25
y k Cx k
~ ~ 6-26
33
Where matrices G, H, and C are the same as those original system.
Let us assume that the dynamic model is subjected to the same control signal u k as the
original system.
x k and xk
If initial conditions are same for the original one as the dynamic one, then ~
will be same. Otherwise, x k and xk will be different.
~
Remarks: although the state xk may not be measurable the output y k is measurable.
The dynamic model does not use the measured output y k .
The dynamic model of equation 6-25 is modified into the following form:
x k Hu k Ke y k C~
x k 1 G~
~ x k , where matrix Ke serves as a weighting
matrix.
Figure 6.5
34
y k Cxk 6-28
u k Kx k
Where
We assume that the system is completely state controllable and completely observable,
but xk is not available for direct measurement.
Following figure shows a state observer incorporated into the system of previous figure.
Figure 6.6
u k K~
x k
From above figure we have
x k 1 G~
~ x k Hu k Ke y k C~
x k 6-29
Which can be rewritten into
x k 1 G K eC ~
~ x k Hu k K e y k 6-30
35
x k 1 is one sampling
Equation 6-30 is called a prediction observer since the estimate ~
period ahead of the measurement y k .
Error Dynamics of the full order state observer
By use principle of duality, the condition for complete observability for the system
defined by
xk 1 Gxk Hu k
y k Cxk
1 1 0
Where G , H , C 0 2,
1 1 2
Design a full order state observer, assuming that the system configuration is identical to
the in above figure. The desired eigenvalues of the observer matrix are
z 0.5 j 0.5, z 0.5 j 0.5, and desired ch equation is
36
z 2 z 0.5 0
37
Define Q WN *
1
, where N C * G * C * G * C * G *
2 n1
C*
an 1 an 2 a1 1
a an 3 1 0
n2
W
a1 1 0 0
1 0 0 0
0 0 0 an
1 0 0 an 1
Q 1GQ 0 1 0 an 2 6-40
0 0 1 a1
CQ 0 0 0 1 6-41
x k Q k
~
Now define ~
x k Hu k Ke y k C~
x k 1 G~
~ x k
will be changed into
G KeC ~ x k Hu k KeCxk
k 1 Q 1 G KeC Q k Q 1 Huk Q 1 KeCQ k
~ ~
6-42
Subtract 6-42 from 6-38, we have
k 1 k 1 Q 1GQ Q 1KeCQ k k
~ ~
6-43
Define ek k k
~
38
n
Let us write Q Ke n 1 , then
1
1
0 0 n
0
n 0
0 0 n 1
Q 1KeCQ n 1 0 0 0 1 0 0 0 n2
1 0 0 0 1
And
0 0 0 an n
n 1
0 0 an 1 n 1
Q 1 G KeC Q Q 1GQ n 1 0 0 0 1 0 1 0 an 2 n 2
1 0 0 1 a1 1
The ch equation becomes
zI Q 1 G KeC Q 0
z 0 0 an n
1 z 0 an 1 n 1
0 1 0 an 2 n 2 0
0 0 1 z a1 1
or
zI Q 1 G KeC Q z n a1 1 z n 1 an 1 n 1 z an n 0 6-45
Suppose the desired characteristic equation for error dynamics is
z n 1 z n 1 n 1 z n 0 6-46
1 a1 1
2 a2 2
n an n
39
Hence, from the equation we have
1 1 a1
2 2 a2
n n an
n n n
1
Since Q Ke n 1
, we have Ke Q n 1
WN
* 1 n 1
1 1 1
40
1
C 0
CG 0
K e G
n 1
CG 1
Summary: The full order prediction observer is given by equation
x k 1 G~
~ x k Hu k Ke y k C~
x k
G KeC ~ x k Hu k KeCxk
The observed state feedback is given by
u k K~x k
If we have the feedback equation substituted into the observer equation, we obtain
x k 1 G K eC HK ~
~ x k Hu k K e y k
Similar to the state feedback, four methods will be used to determine the observer feedback gain
Ke .
Method 1 :
n an n an
a
1 n 1 an 1
Ke Q n 1 n 1
WN *
1 a1 1 a1
1
Where Q WN * , N C * G * C * G * C * G *
2 n1
C*
an 1 an 2 a1 1
a an 3 1 0
n2
W
a1 1 0 0
1 0 0 0
ZI G z n a1 z n 1 an 1 z an 0
Note: if the system is already in an observable canonical form, then the matrix K e can be
determined easily, because matrix WN * becomes an identity matrix, thus WN * I
1 1
41
Method 2 : The desired observer feedback gain matrix K e can be given by Ackermann’s
formula.
We have
1
C 0
CG 0
K e G
n 1
CG 1
G G 1G n 1 n 1G n I
n
Method 3:
If the desired eigenvalues u1 , u2 ,un of matrix G K eC are distinct, then the observer
feedback gain matrix K e may be given by the equation as follows:
1
1 1
1
Ke 2
n 1
1
Where 2 satisfy the equation
n
i C G ui I 1 , i 1,2, n
Method 4: If the order of the system is low, substitute K e into the characteristic equation.
ZI G K eC =0 and then matches the coefficients of powers in z of this characteristic
equation with equal powers in z of the desired characteristic equations.
42
Example 6.7 Consider the system defined by
xk 1 Gxk Hu k
y k Cxk
1 1 0
Where G , H , C 0 2,
1 1 2
Design a full order state observer, assuming that the system configuration is identical to
the in above figure. The desired eigenvalues of the observer matrix are
z 0.5 j 0.5, z 0.5 j 0.5, and desired ch equation is
z 2 z 0.5 0
ZI G z 2 2 z 2 0
Method 1:
Method 2:
Method 3:
Method 4:
Consider the completely state controllable and completely observable system defined by
the equations
43
xk 1 Gxk Hu k
y k Cxk
x k , we have
For the state feedback control based on the observed state ~
u k Kx k
~
ZI G HK HK
0
0 ZI G K eC
Pole placement design and the observer design are independent of each other.
Remarks: The poles of the observer are usually chosen so that the observer response is
much faster than the system response. A rule of thumb is to choose an observer response
at least four to five times faster than the system response ( or deadbeat response).
Current observer:
In the prediction observer the observed state ~ x k is obtained from measurements of the
output vector up to y k 1 and of the control vector up to u k 1 . A different
formulation of the state observer is to use y k for the estimation of ~ x k . This can be
done in two steps. First step we determine z k 1 , an approximation of xk 1 based
on ~x k and u k . In the second step, we use y k 1 to improve xk 1 .
44
Minimum order observer:
State feedback control scheme where the feedback state consists of the measured portion
of the state and the observed portion of the state obtained by use of the minimum order
observer.
45
46
VI.7. Servo Systems
Note: It is generally required that the system has one or more integrators within the
closed loop, to eliminate the steady state error to step inputs.
47
vk 1 vk r k 1 y k 1
vk r k 1 C Gxk Hu k 6-51
CGxk vk CHu k r k 1
We then have
u k 1 K 2 xk 1 K1vk 1
K 2 xk 1 K1 CGxk vk CHu k r k 1
K 2 Gxk Hu k K1 CGxk vk CHu k r k 1 6-52
K 2 K 2G K1CG xk I m K 2 H K1CH u k K1r k 1
x G H x 0
u K K G K CG
I m K 2 H K1CH u K1r
6-55
2 2 1
xe k 1 G H xe k
u k 1 K K G K CG I m K 2 H K1CH ue k
e 2 2 1
6-56
G H xe k 0
wk
0 0 ue k I m
x k
Where wk K 2 K 2G K1CG I m K 2 H K1CH e 6-57
ue k
Define
48
xe k
k , n+m vector
ue k
G H
G , n m n m matrix
0 0
0
H , n m m matrix
Im
K K 2 K 2G K1CG I m K 2 H K1CH
G I n H , m n m matrix
K 2 K1 0 I m
CG CH
Then we have k 1 G k Hwk 6-58
And wk K k
Thus, 6-58 will be completely state controllable, K can be designed, and K 2 K1 can be
obtained using following equation:
G I n H G I n H
K K 2 K1 0 I m K 2 K1 K 0 I m
CG CH CG CH
Example B-6-17 : Figure shows a servo system where the integral controller has a time delay of
one sampling period. Determine the feedforward gain K1 and the feedback gain K2 such that the
response to the unit step sequence input is deadbeat.
Solution:
49
50