EEP 3103

Modern Control Theory

M.Tech - 2019.S1

Module 2

Vinod V Nair
 State Observer(Module 2 contd.)
• State Observer. A state observer estimates the state variables based on
the measurements of the output and control variables.
• State observers can be designed if and only if the observability condition is
satisfied.
• The observer is a subsystem to reconstruct the state vector of the plant.
The mathematical model of the observer is basically the same as that of
the plant, except that we include an additional term that includes the
estimation error to compensate for inaccuracies in matrices A and B
• The estimation error or observation error is the difference between the
measured output and the estimated output. The initial error is the
difference between the initial state and the initial estimated state.

Where is the estimated state and is the estimated output

 State Observer (Module 2 contd.)
• The inputs to the observer are the output y and the control input u. Matrix
Ke, which is called the observer gain matrix, is a weighting matrix to the
correction term involving the difference between the measured output y
and the estimated output . This term continuously corrects the model
output and improves the performance of the observer
 Observable phase variable form of state model
.
X = AX + Bu Z=Po X will transform the state model to
y − CX + Du observable canonical form

Z = Po X
Po − 1 Z = Po − 1 Po X
X = Po − 1 Z
. .
−1
X = Po Z = A X + B u = A Po − 1 Z + B u
.
ie Z = Po A Po − 1 Z + Po B u = Ao Z + B o u
y = C Po − 1 Z + D u = C o Z + D u
Ao = Po A Po − 1
B o = Po B ;
C o = C Pc − 1
 0 0......0 − an 
1 0.......0 − a   bn 
b 
 n−1 
 n −1  Co =[0 0.....0 1]
Ao = .  B = . 
  o  
.  . 
0 0.......1 − a1   b1 
   

Determination of Po
Ch. Equation of original system is λI − A = λ n + a1λ n−1 + a2λ n−2 + .........an−1λ + an = 0

 a n − 1 a n − 2 ....... a 2 a 1 0 
a a ....... a 1 0 
 n − 2 n − 3 1 
. 
  Po = WQoT
W = . 
a1 1 ...........0 0 0  Qo = CT AT CT ( AT )2 CT .....( AT )n−1 CT 
 
1 0 ...........0 0 0 
 
 
Example
 Example

Define

The form is now observable canonical form

 Full Order State Observer
•

r U
+
x x y
+
-
B
+ ∫ C

•.

x̂x A
x̂

+
m
-
•

Bu ++ xˆ x̂ ŷ
B + ∫∫ ∫ C
+

OBSERVER A

K
Feedback of observed
state variable
 State Observer
Full-Order State Observer. The order of the state observer that
will be discussed here is the same as that of the plant. Assume
that the plant is defined by

and the observer model is defined by

Where is the estimated state and is the estimated output.The

inputs to the observer are the output y and the control input u. Matrix
K,, which is called the observer gain matrix, is a weighting matrix to
the correction term involving the difference between the measured
output y and the estimated output . This term continuously corrects
the model output and improves the performance of the observer
 State Observer
To obtain the state observer equation:

we see that the dynamic behaviour of the error vector is determined

by the eigenvalues of matrix . If matrix is a stable
matrix, the error vector will converge to zero for any initial error
vector e(0). That is x ( t ) will converge to x(t) regardless of the
values of x(0) and . If the eigenvalues of matrix are
chosen in such a way that the dynamic behaviour of the error vector
is asymptotically stable and is adequately fast, then any error vector
will tend to zero (the origin) with an adequate speed.
 State Observer
Dual Problem. The problem of designing a full-order observer becomes
that of determining the observer gain matrix , such that the error
dynamics defined by are asymptotically stable with
sufficient speed of response. (The asymptotic stability and the speed of
response of the error dynamics are determined by the eigenvalues of
matrix ) Hence, the design of the full-order observer becomes
that of determining an appropriate Ke, such that has desired
eigenvalues. Thus, the problem here becomes the same as the pole-
placement problem we discussed in an earlier section. In fact, the two
problems are mathematically the same and this property is called
duality.
 State Observer

Comparing the characteristic polynomials we get

relationship.
Also the matrices A, B and K in control become AT, CT and mT in estimation
due to duality..
State Observer Design - Comments
State Observer Design
State Observer Design
 State Observer Design

For the dual of this we have:

Since Ackermann’s formula for the primal system is

for the dual it becomes

 •Difference between measured and estimated outputs (error  m1 
signal) are fed back through constant error matrix “m” m 
•m is nx1 matrix  2
•Model is corrected continuously using the error signal m = . 
•LUENBEGER State Observer  
. 
. mn 
 
xˆ = Axˆ + Bu + m( y − yˆ )
x% = x − xˆ
  

x% = x − xˆ = Ax + Bu − [ Axˆ + Bu − m ( y − yˆ )]
= Ax + Bu − Axˆ − Bu − mC ( x − xˆ )
= A( x − xˆ ) − mC ( x − xˆ )
= ( A − mC )( x − xˆ )

ie x% = ( A − mC ) x%
Characteristic equation of the error vector is λ I − ( A − mC) = 0

• If m is chosen so that (A-mC) has stable and fast roots x% (t )

will decay to ‘0’ irrespective of x% (0)
Ie. x̂ will converge to x regardless of xˆ (0 )
% (t ) converges to ‘0’ and is independent of
• Error x
forcing function ‘u’

• Selection of ‘m’ is same as selection of K in regulator

• Useful only if the system is completely observable

 Determination of Gain Matrix m
Method I
1. Determine the ch. Polynomial of original (1)and desired (2) system

λI − A =λn +a1λn−1 +a2λn−2 +.........an−1λ +an (1)

n n−1 n−2
(λ −µ1)(λ −µ2).......(λ −µn) = λ +αλ
1 +α2λ +....+αn−1λ +αn (2)
2. Determine the transformation matrix Po which transforms the original state model to
observable phase variable form
α n − an 
α − a 
 n−1 n 
3. Observer gain matrix m  
−1 .
m = Po  
 . 
α 2 − a2 
 
α1 − a1 
 Method II
1. Determine ch. polynomial with (a)state observer and (b) desired c.l poles

λ I − ( A − mC) = λ I − A + mC = λ n + b1λ n−1 + b2λ n−2 + .... + bn−1λ + bn (1)

(λ − µ1 )(λ − µ2 ).......(λ − µn ) = λ n + α1λ n−1 + α2λ n−2 + .... + αn−1λ + αn (2)
b1 , b2 ,...bn Are functions of m1, m2 ,...mn

2. Equate coefficients and solve for m1 , m2 ,...mn

 Method III
1. Determine ch. polynomial with desired c.l poles

(λ − µ1 )(λ − µ2 ).......(λ − µn ) = λ n + α1λ n−1 + α2λ n−2 + .... + αn−1λ + αn

2. Determine the matrix φ ( A) using the coefficients of desired ch.
polynomial
φ ( A ) = A n + α 1 A n −1 + α 2 A n − 2 + .... + α n −1 A + α n I

3. Calculate the observer gain matrix using Ackermann’ formula

−1
C  0 
 CA  0 
   
.  . 
m = φ ( A)    
.  . 
 CA n − 2  0 
   
 CA 
n −1
1 
 State Observer – Design Problem

Design a full-order state observer, assuming that the system

configuration is identical to the standard form. Assume that the
desired eigenvalues of the observer matrix are µ1 = -10, µ2 = -10
The design of the state observer reduces to the determination of an
appropriate observer gain matrix Ke.
 State Observer – Design Problem
Method 1: We shall determine the observer gain matrix by using the
transformation approach .The given system is already in the OCF. Therefore
 State Observer – Design Problem
Method 2

By comparing with standard form we have

 State Observer – Design Problem
• Method 3 : Use Ackermann’s formula
 Minimum Order-State Observer
Minimum-Order Observer,
• The observers discussed thus far are designed to reconstruct all the state
variables. In practice, some of the state variables may be accurately
measured. Such accurately measurable state variables need not be
estimated. Suppose that the state vector x is an n-vector and the output
vector y is an m-vector that can be measured. Since m output variables
are linear combinations of the state variables, m state variables need not
be estimated. We need to estimate only n – m state variables. Then the
reduced-order observer becomes an (n - m)th-order observer.
• Such an (n - m)th-order observer is the minimum-order observer.
 Minimum Order-State Observer
Consider the system

where the state vector x can be partitioned into two parts xa, (a scalar)
and xb.Here the state variable xa , is equal to the output y and thus can be
directly measured, and xb is the un-measurable portion of the state vector.
Then the partitioned state and output equations become

And we have

A tabular listing of necessary

substitutions for writing the
Observer equation for the
Minimum-Order State Observer is
given alongside.
 Transfer Function
Transfer Function of the Observer-Based Controller.

By taking the Laplace transform of, assuming a zero initial condition,

and solving for , we obtain
 Summary
Summary of State-Space Design Method
1. The state-space design method based on the pole-placement-combined
with observer approach is very powerful. It is a time-domain method. The
desired closed loop poles can be arbitrarily placed, provided the plant is
completely state controllable.
2. If not all state variables can be measured, an observer must be
incorporated to estimate the un-measurable state variables.
3. In designing a system using the pole-placement approach, several different
sets of desired closed-loop poles need be considered, the response
characteristics compared, and the best one chosen.
4. The bandwidth of the observer controller is generally large, because we
choose observer poles far to the left in the s plane. A large bandwidth
passes high frequency noises and causes the noise problem.
 Summary
5. Adding an observer to the system generally reduces the stability margin. In
some cases, an observer controller may have zero(s) in the right-half s
plane, which means that the controller may be stable but of non-minimum
phase. In other cases, the controller may have pole(s) in the right-half s
plane—that is, the controller is unstable. Then the designed system may
become conditionally stable.
6. When the system is designed by the pole-placement-with-observer
approach, it is advisable to check the stability margins (phase margin and
gain margin), using a frequency-response method. If the system designed
has poor stability margins, it is possible that the designed system may
become unstable if the mathematical model involves uncertainties.
7. Pole placement approach provides higher order compensators which are
costlier to implement so that if classical approaches are possible it may be
economical to go for these.