ELEC 3035, Lecture 7: Observer design

Ivan Markovsky

Observers

Observer design by pole placement

Duality between observer and controller design

Pole placement by output feedback

ELEC 3035 (Part I, Lecture 7) Observer design 1 / 12

 General observer design problem
Given dynamical system Bext with two types of external variables:

observed variables w w
Bext
z
to-be-estimated variables z

find system (called observer) accepting w and producing z

w Observer z

We will consider the case: B = Bi/s/o (A, B, C, D), w = (u, y ), z = x.

Lecture 6: nonrecursive, feedforward observer for the initial state x(0)

Now our goal is recursive feedback observer for the current state x(t)
ELEC 3035 (Part I, Lecture 7) Observer design 2 / 12
 Output feedback control
separation and certainty principles
We can extend a given state-feedback controller
u = Kx
to output feedback controller
u = K xb
by using the observer state estimate xb in place of x.

u Plant y

xb
K Observer

Output feedback controller

ELEC 3035 (Part I, Lecture 7) Observer design 3 / 12
 Internal model and feedback principles
The observer design is based on the following principles:

1. Internal model: the model run by u, gives an estimate xb for x

2. Feedback: correct the estimate xb, so that the error

x(t) xb(t) =: e(t) 0 as t

Let the feedback be a linear function of the output error

feedback correction = L(y yb)

Then the observer for the model Bi/s/o (A, B, C, D) is

xb = Axb + Bu L(y yb)

yb = C xb + Du

ELEC 3035 (Part I, Lecture 7) Observer design 4 / 12

 Error dynamics

Our goal is to choose L, so that the state error e(t) 0 as t .

The dynamics of e is

e = (x xb)
= Ax + Bu Axb Bu + L(y yb)
= A(x xb) + LC(x xb)
= (A + LC) e
| {z }
Ao

i.e., e Bss (Ao ) an autonomous LTI system.

Therefore, e(t) 0 as t is equivalent to stability of Bss (Ao ).

ELEC 3035 (Part I, Lecture 7) Observer design 5 / 12

 Comparison with state-feedback stabilization

In the state feedback stabilization problem we have

x = Ax + Bu and u = Kx

which gives an autonomous LTI closed loop system

x = (A + BK ) x
| {z }
Ac

and the aim is to choose K , so that Bss (Ac ) is stable.

ELEC 3035 (Part I, Lecture 7) Observer design 6 / 12

 Observer design by pole placement

The condition e(t) 0 as t is a minimum requirement.

In fact we want e(t) 0 fast

(possibly in a finite (small) number of steps deadbeat observer)

The error dynamics is governed by the poles of the matrix

Ao := A + LC

so for desired error dynamics we can

select desired pole locations of Ao and choose L to achieve them.

ELEC 3035 (Part I, Lecture 7) Observer design 7 / 12

Duality of the observer PP and controller PP problems
Observer PP problem: Choose L, so that


det zI (A + LC) = pdes (z)

Controller PP problem: Choose K , so that

det zI (A + BK ) = pdes (z)

Observer PP is not a new problem:



det zI (A + LC) = det zI (A + LC)
 
= det zI (A + C L )
 
e +B
= det zI (A eKe)

= observer PP is controller PP for the dual system.

ELEC 3035 (Part I, Lecture 7) Observer design 8 / 12
The results for state feedback PP can be restated for observer PP:

Theorem: The eigenvalues of A + LC can be assigned choosing L

to any locations in C if and only if A, C is observable.

Observer canonical form Controller canonical form

Lemma:
Let A, c and A , c be two observable pairs and
assume that A and A have the same char. polynomials.
Then there is a unique similarity transformation given by the matrix

1
T := O(A , c ) O(A, c)

such that
T 1 AT = A and cT = c .

ELEC 3035 (Part I, Lecture 7) Observer design 9 / 12

 Closed-loop system with output feedback controller
Consider the closed loop system

u Plant y

xb
K Observer

Output feedback controller

where
Plant: x = Ax + Bu, y = Cx + Du

Observer: xb = Axb + Bu L(y C xb Du)

State feedback controller: u = K xb

ELEC 3035 (Part I, Lecture 7) Observer design 10 / 12

Feedback controller:
xb = (A + LC)xb + (B + LD)u Ly , u = K xb
= (A + LC + BK + LDK )xb Ly

Note: the feedback controller is a dynamical system

    
x A BK x
Closed-loop system: =
xb LC A + LC + BK xb

Note: closed-loop system order = plant order + controller order

    
x A + BK BK x
Error equation: =
e 0 A + LC e

ELEC 3035 (Part I, Lecture 7) Observer design 11 / 12

 Example: output feedback deadbeat control
5 5
x 10 x 10
1 2

0 1

1 0

u
y

2 1

3 2

4 3
5 10 15 20 5 10 15 20
t t

15th order single-input open-loop system, 30 order closed-loop system

(The same system as the one used in the example of Lecture 1)

ELEC 3035 (Part I, Lecture 7) Observer design 12 / 12