FULL STATE FEEDBAK CONTROL:

Full-state variable feedback control is considered to achive the desired pole

locations of the closed-loop system.
The first step in the state variable design process requires us to assume that all
the states are available for feedback-that is, we have access to the complete
state, x(t), for all t. The system input u(t), is given by,

u K x
Determining the gain matrix K is the objective of the full-state feedback design
procedure. The beauty of the state variable design process is that the problem
naturally seperates into a full-state feedback component and an observer design
component. These two design procedures can occur independently, and in fact,
the seperation principle provides the proof that this approach is optimal. The
stability of the closed-loop system is guaranteed if the full-state feedback control
law stabilizes the system and the observer is stable.
Dorf and Bishop, Modern Control Systems

The full-state feedback block diagram is illustrated in Figure 1.

System Model

x A x B u
x

u
Kontrol Law

-K
Full-state feedback
Figure 1. Full-state feedback block diagram (with no reference input)

With the system defined by the state variable model

x A x B u
and the control feedback given by

u K x
we find the closed-loop system to be

x A x B u A x B K x A B K x
The characteristic equation associated with above equation is

det I A B K 0
If all the roots of the characteristic equation lie in the left-half plain, then the closed
loop is stable. In other words, for any initial condition x(t0), it follows that
Dorf and Bishop, Modern Control Systems

x ( t ) e A BK t x ( t 0 ) 0 as t
Given the pair (A,B), we can always determine K to place all the system
closed loop poles in the left half-plane if and only if the system is completely
controllable-that is, if and only if the controllability matrix PC is full rank (for a
SISO system, full rank implies that PC is invertible).
The addition of a reference input can be considered as

u ( t ) K x ( t ) N r ( t )
System Model
r(t)

x A x B u
x
Kontrol Law

-K

Where r(t) is the reference input. When r(t)=0 for all t>t0, the control design
problem is known as the regulator problem. That is, we desire to compute K
so that all initial conditions are driven to zero in a desirable fashion (as
determined by the design specifications).

When using this state variable feedback, the roots of the characteristic
equation are placed where the transient performance meets the desired
response.
Example:
Consider the third-order system with the differential equation
3

d y
d y
dy
5 2 3 2y u
3
dt
dt
dt
Dorf and Bishop, Modern Control Systems

d3y
d2y
dy
5 2 3 2y u
3
dt
dt
dt
We can select the state variables as x1=y, x2=dy/dt, x3=d2y/dt2. (Phase variables)

dy
x 1
x2
dt
d2y
x 2 2 x 3
dt
d3y
x 3 3 5 x 3 3 x 2 2 x1 u
dt
1
0
x 1 0
x 0

0
1
2

x 3 2 3 5

x1
x
2
x 3

0
0 u A x B u
1

y x1
and

y 1 0 0 x

If the state variable matrix is

K k1

k2

k3

and

u K x
then the closed-loop system is

x A x B K x A BK x
The state feedback matrix is

0
2 k1

0
3 k2

5 k 3

A BK

Dorf and Bishop, Modern Control Systems

and the characteristic equation is

det I A BK 3 5 k 3 2 3 k 2 2 k1 0
clc;clear
syms k1,k2,k3
mtrx=[0,1,0;0,0,1;(-2-k1),(-3-k2),(-5-k3)];
det(mtrx)
If we seek a rapid response with a low overshoot, we choose a desired
characteristic equation such that

2n
2

2
n

We choose =0.8 for minimal overshoot and n to meet the settling time
requirement.

ov e

/ 1 2

ov e

0.8 / 1 0.8 2

0.0152

If we want a settling time (with a 2% criterion) equal to 1 second, then

4
4
ts

1 n 6 rad / s
n 0.8 n
If we choose n=6 rad/s, the desired characteristic equation is

9.6 36 4.8 14.4 82.1 172.8

3

Comparing two characteristic equations yields

5 k 3 14.4
3 k 2 82.1
2 k1 172.8
Therefore, we require that k1=170.8, k2=79.1 and k3=9.4.

K 170.8 79.1 9.4

ACKERMANNS FORMULA:
For a single-input, single output system, Ackermanns formula is useful for
determining the state variable feedback matrix

K k1
where

k2 kn

u K x

Given the characteristic equation

q n 1n 1 n
The state feedback gain matrix is

K 0 0 1 P q A
1
C

where

q A A n 1A n 1 n 1A n I

where PC is the controllability matrix.

Example:
Consider the system

d2y
u(t)
2
dt

Y (s)
1
G (s) 2
U(s)
s

and determine the feedback gain to place the closed-loop poles at s=-1i.
Therefore, we require that
>>poly([-1+i -1-1i])

q ( ) 2 2
2

and 1= 2=2. With x1=y and x2=dy/dt, the matrix equation for the system G(s) is

dy
x1
x2
dt
d2y
x 2 2 u ( t )
dt

0 1
0
x
x u

0 0
1
Dorf and Bishop, Modern Control Systems

The controllability matrix is

0 1
PC B AB

1
0

Thus we obtain

0 1
K 0 1

1
0

where

q A

1 0 1 0 1
P

1 1 0 1 0
1
C

and

0 1
q A

0 0

0 1
1 0 2 2
2
2

0 0
0 1 0 2

Then we have
k1

k2

0 1 2 2
0 2
K 0 1
0 1
2 2

1 0 0 2
2 2
Note that computing the gain matrix K using Ackermanns formula requires the
1
use of PC . We see that complete controllability is essential because only then
1
we can guarantee that the controllabilty matrix PC has full rank and hence that PC
exists.
With Matlab
>>A=[0 1;0 0];
>>B=[0;1];
>>K=acker(A,B,[-1+i,-1-i])

Result:
K=
2

>>K=place(A,B,[-1+i,-1-i])
Dorf and Bishop, Modern Control Systems

OBSERVER DESIGN:
In the full-state feedback design procedure discussed in the previous section, it
was assumed that all the states were available for feedback at all times. This is a
good assumption for the control law design process. However, generally
speaking, only a subset of the states are readily measurable and available for
feedback. Having all the states available for feedback implies that these states
are measured with a sensor or sensor combinations. The cost and complexity of
the control system increase as the number of required sensor increases. So,
even in situations where extra sensors are available, it may not be cost-effective
to employ these extra sensors, if indeed, the control system design goals can be
accomplished without them. Fortunately, if the system is completely observable
with a given set of outputs, then it is possible to determine (or estimate) the
states that are not directly measured (or observed).

According to Luenberger, the full-state observer for the system

x A x B u
y Cx
is given by

x A x B u L y C x

where x denotes the estimate of the state x. The matrix L is the observer gain
matrix and is to be determined as part of the observer design procedure. The
observer is depicted in Figure 2. The observer has two inputs, u and y, and one
output, x .
u
Observer

x A x B u L~y

~y y C x

+
-

C
Figure 2. Observer.

Dorf and Bishop, Modern Control Systems

The goal of the observer is to provide an estimate

x so that x x as t .

Remember that we do not know x(t0) precisely; therefore we must

provide an initial estimate x ( t 0 ) to the observer. Define the observer
estimation error as

e( t ) x ( t ) x ( t )
The observer design should produce an observer with the property that e(t) 0
as t . One of the main results of systems theory is that if the system is
completely observable, we can always find L so that the tracking error is
asymptotically stable, as desired.

Taking the time-derivative of the estimation error in the error equation yields

e x x
and using the system model and the observer, we obtain

e A x B u A x Bu L y C x
Dorf and Bishop, Modern Control Systems

or

e ( t ) A L C e( t )

We can guarantee that e(t)0 as t for any initial tracking error e(t0) if the
characteristic equation

det I A L C 0
has all its roots in the left half-plane. Therefore, the observer design process
reduces to finding the matrix L such that the roots of the characteristic equation
lie in the left half-plane. This can always be accomplished if the system is
completely observable; that is, if the observability matrix, PO, has full rank.
Example:
Consider the second-order system

2 3
0
x
x u

1 4
1
y 1 0 x

In this example, we can only directly observe the state y=x1. The observer will
provide estimates of the second state, x2.

The observer design begins by checking the system observability to verify that an
observer can be constructed to guarantee the stability of the estimation error.
From the system model, we find that

2 3
A

1
4

and C 1 0

The corresponding observability matrix

C 1 0
PO

C
A
2
3

Since det PO=3, the system is completely observable. Suppose that the desired
charactersictic equation is given by

d 2 2n 2n

We can select =0.8 and n=10, resulting in an expected settling time of less
than 0.5 second. Computing the actual characteristic equation yields

det I A LC 2 L1 6 4 L1 2 3 L 2 1

where

L1
L
L2
Equating the coefficients yields the two equations

L1 6 16

4 L1 2 3 L 2 1 100
which, when solved, produces

L1 22
L
L 2 59

Dorf and Bishop, Modern Control Systems

The observer is thus given by

x 2 3 x 0 u 22 y x
1
1 4
1
59

e ( t ) A L C e( t )

2 3 22

e ( t )
1 0 e( t )

1 4 59

2 22 3 0
e ( t )
e( t )

1 59 4 0
20 3
e ( t )
e( t )

60 4

Dorf and Bishop, Modern Control Systems

Matlab Code
clc;clear
A=[-20 3;-60 4];B=[0;0];C=[0 0];D=[0];
sys=ss(A,B,C,D) %state-space model
x0=[1 -2]; %initial conditions
t=[0:0.01:1];
u=0*t; %zero input
[y,T,x]=lsim(sys,u,t,x0);
plot(T,x(:,1),T,x(:,2))
xlabel('Time (seconds)')

20 3
e ( t )
e( t )

60 4

The response of the

estimation error to an
initial error of

1
e

is shown in the figure.

e1

0
-1
-2

e2

-3
-4
-5

0.2

0.4
0.6
Time (seconds)

0.8

Ackermanns formula can also be employed to place the roots of the observer
characteristic equation at the desired locations. Consider the observer gain
matrix

L1
L
L 2

Ln
and the desired observer characteristic equation

p n 1n 1 n 1 n
The s are selected to meet given performance specifications for the observer.
The observer gain matrix is then computed via

L p A PO1 0 1

where PO is the observability matrix

Dorf and Bishop, Modern Control Systems

p A A n 1A n 1 n 1A n I
Example:
Consider the second-order system given in previous example. The desired
characteristic equation is given

p 2 2 n 2n
where =0.8 and n=10, hence, 1=16 and 2=100. Computing p(A) yields

2 3
p A

1
4

2 3
1 0 133 66
16
100

1
4
0
1

22
177

We have the observability matrix

C 1 0
PO

C
A
2
3

which implies that

0
PO

2 3 1 3
1

Using Ackermanns formula yields the observer gain matrix

133 66
L p A P 0 1

22
177

1
O

1
0
2 3 1 3

0 22
1 59

COMPENSATOR DESIGN: INTEGRATED FULL-STATE FEEDBACK AND OBSERVER

The state variable compensator is constructed by appropriately connecting the fullstate feedback conrol law to the observer. The compensator is shown in Figure 3.
System model

x A x B u

Observer
Conrol Law

-K

Compensator

x A x B u L~y

~y y C x

+
-

Figure 3. State variable compensator employing full-state feedback in series with a full-state

Our strategy was to design the state feedback control law as u(t)=-Kx(t), where
we assumed that we had access to the complete state x(t). Then we designed
(t)
an observer to provide an estimate of the state
. It seemsxreasonable
that
we can employ the state estimate in the feedback control law in place of x(t). In
other words, we can consider the feedback law

u ( t ) K x ( t )
The feedback gain matrix K was designed to guarantee stability of the closedloop system; that is, the roots of the charactersitic equation

det I A BK 0
are in the left-half plane. Under the assumption that the complete state x(t) is
available for feedback, the feedback control law (with properly designed gain
matrix K) leads to the desired results that x(t)0 as t for any initial
condition x(t0). We need to verify that, when using the feedback control law for
observed states, we retain the stability of the closed-loop system.

Dorf and Bishop, Modern Control Systems

Consider the observer again

u ( t ) K x ( t )

x A x B u L y C x
Substituting the feedback law and rearranging terms in the observer yields the
compensator system

x A BK LC x L y
u K x
Notice that the system in the equation has the form of a state variable model with
input y and output u as illustrated in Figure 3.
Computing the estimation error using the compensator yields

e x x Ax Bu A x Bu Ly L C x
or

e A LC e
Dorf and Bishop, Modern Control Systems

This is the same result as we obtained for the estimation error. The estimation
error does not depend on the input as seen in the error equation, where input
terms cancel.
Recall that the underlying system model is given by

x Ax Bu
y Cx
Substituting the feedback law

u ( t ) K x ( t )
into the system model

x Ax Bu A x B K x
and with
we obtain

x x e
x A B K x B K e

We can write the equations in matrix form

BK
x A B K
e

0
A

L
C

x
e

u ( t ) K x ( t )
Recall that our goal is to verify that, with
, we retain
stability of the closed-loop system and the observer. The characteristic
equation associated with the matrix equation
det I A BK det I A LC
So if the roots of det[I-(A-BK)]=0 lie in the left half-plane, and if the roots of
det [I-(A-LC)]=0 lie in the left half-plane, then the overall system is stable.
Therefore, employing the strategy of using the state estimates for the
feedback is in fact a good strategy.

The fact that the full-state feedback law and the observer can be designed
independently is an illustration of the seperation principle.
Dorf and Bishop, Modern Control Systems

The design procedure is summarized as follows

1. Determine K such that det[I-(A-BK)]=0 has roots in the left-half plane and
place the poles appropriately to meet the control system design
specifications. The ability to place the poles arbitrarily in the complex plane
is guaranteed if the system is completely controllable.

2. Determine L such that det[I-(A-LC)]=0 has roots in the left-half plane and
place the poles to achieve acceptable observer performance. The ability to
place the poles arbitrarily in the complex plane is guaranteed if the system is
completely observable.

3. Connect the observer to full-state feedback law using

u ( t ) K x ( t )

Dorf and Bishop, Modern Control Systems