Chapter 7

State Feedback Control

Design and State
Observer Design

EE4L005 DIGITAL CONTROL SYSTEMS

N. C. Sahoo
 Introduction
• The control design using state space approach is
often referred to as modern control design.

• The state variable methods appear to be much more

dependent on an accurate system model for design
process. An advantage of these methods is that the
system representation provides a complete (internal)
description of the system.
• In this chapter, state space design methods for
single-input single-output (SISO) systems will be
mainly discussed.
• A design method using pole placement or pole
assignment will be discussed.
• This method is similar to root-locus design in the
sense that the closed-loop poles may be placed at
desired locations.
• However, pole-placement design allows all closed-
loop poles to be placed at desired locations, whereas
the root-locus design allows only the two dominant
poles to be placed.
• However, there is a cost associated with placing all
closed-loop poles; it requires measurement and
feedback of all the state variables of the system.
• In many situations, all the state variables cannot be
measured because of cost consideration or because
of the lack of suitable transducers.
• In such cases, the non-measurable state variables
must be estimated from the measurable variables.
• Fortunately, the design can be separated into two
phases. In the first phase, the system is designed as
if all states of the system will be measured. The
second phase is concerned with the design of state
estimator, often called as observer.

Schematic diagram of state-feedback control system

 State Feedback
The state space model of the discrete-time system is
x(k  1)  Ax(k )  Bu(k )
y(k )  Cx(k )  Du(k )
where x is n 1 state vector, u is p  1 input vector, y
is q 1 output vector, and A, B, C, and D are real
constant matrices of appropriate dimensions.

• Let us assume a simple structure for the controller

for the plant:
u(k )  r(k )  Kx(k )
where r(k ) is the reference input and K is some real
constant matrix (feedback/gain matrix).
• Thus, the state model of the closed-loop system is
given by
x(k  1)  (A  BK)x(k )  Br(k )
y(k )  (C  DK)x(k )  Dr(k )

Structure of state feedback control system

 Pole Assignment by
State Feedback
Without loss of generality, let us assume D = 0. Thus,
for a SISO system, the system Eqs. are:
u(k )  r (k )  Kx(k ) ; K   K1 K2 Kn 
x(k  1)  (A  bK)x(k )  br (k ) y(k )  cx(k )

State feedback control for SISO system

 If the pair {A,b} is controllable, then the state
feedback control, as given above, can place/assign
the eigenvalues/poles of (A+bK) arbitrarily.

Proof: x(k  1)  Ax(k )  bu(k )

Let us first transform this into controllable canonical
form. It will be of general form as given below.
 0 1 0  0 0  0 
 0 0 1  0 0  0 
   
x(k  1)          x(k )    u (k )
   
 0 0 0  0 1  0
 
  n   n1   n2    2  1  1

 Ax(k )  bu(k )
• If such a transformation exists, the elements of last
row of matrix A correspond to the coefficients of the
characteristic polynomial, i.e.,
n1
zI  A  z  1 z
n
    n1 z   n
• In the following, it is shown that if the pair {A, b} is
controllable, it is possible to transform to canonical
form stated above.
• First, let us assume that such a transformation exists
and is given by
x(k )  Px(k )
where P can be expressed as:
  p11 p12  p1n   p1 
p p22  p2 n  p 2 
P  21  
       
   
 pn1 pn 2  pnn  p n 

p i   pi1 pi 2  pin  ; i  1,2,, n

Thus,
x1 (k )  p11 x1 (k )  p12 x2 (k )    p1n xn (k )  p1x(k )
• Shifting operation on this equation gives
x1 (k  1)  p1x(k  1)  p1Ax (k )  p1bu (k )
x1 (k  1)  x2 (k ) is a function of x(k ) only.
Thus,
p1b  0 and x2 (k )  p1Ax ( k )
• Shifting operation on this equation gives
x2 (k  1)  p1Ax (k  1)  p1A x(k )  p1Abu (k )
2

Therefore, p1Ab  0 and x3 ( k )  p 1 A x ( k )

2

• Continuing this process, we obtain

n2
p1 A b0 and xn (k )  p1A n1x(k )
Shifting operation on this equation gives
xn (k  1)  p1A n x(k )  p1A n1bu (k )
 Comparing this with the last equation of the
transformed set of equations, we obtain
n 1
p1 A b 1
• Thus,
 p1 
 pA 
x(k )  Px(k )   1  x( k )
  
 n 1 
p1A 
where p1 must satisfy the conditions
n2 n 1
p1b  p1Ab    p1A b  0 ; p1 A b 1
• This transformation transforms the original system
equations
x(k  1)  Ax(k )  bu(k ) to
x(k  1)  PAP 1x(k )  Pbu (k )
On comparison, 0  p1b 
0  p Ab 
Pb      1 
   
   n 1 
1 p1A b 

 p1 b Ab A b   0 0
n1
1
 p1S  0 0  1
 p1  0 0  1S 1
• S is the controllability matrix, which is nonsingular
because of the assumption of controllability of the
system.
• Thus, the state equations of the system can be
transformed to this canonical form by this
transformation.
The transformation matrix P is given by
 p1 
 pA 
P 1 
;
p 1  0 0  1S 1

  
 n 1 
 1
p A 
• Thus, the feedback control law becomes
u (k )  r (k )  Kx(k )  KP 1x(k )  r (k )  Kx(k )  r (k )
where K  KP 1  K1 K2  Kn 
• With this control law, the transformed system
becomes
x(k  1)  [A  bK]x(k )  br (k ) 
 0 1 0 0 0  0
 0 0 1 0 0  0
  
  x (k )    r (k )
   
 0 0 0 0 1  0
  n  K1  n 1  K 2  n  2  K 3  2  K n 1 1  K n  1 
• Since the coefficients K i ’s are arbitrarily chosen real
numbers, the coefficients of the characteristic
equation polynomial of ( A  b K ) can be given any
desired value, i.e., the desired closed-loop poles can
be located at arbitrary locations.
• It is subject to complex conjugate pairing: coefficients
of a characteristic equation polynomial will be real
only if the complex poles occur in conjugate pairs.
• The system poles are invariant under equivalence
transformation.
• Assume that the desired characteristic polynomial of
(A+bK) and hence ( A  b K ) is:
n 1
z  1 z
n
    n1 z   n
 It can be seen that this requirement is met if K is
chosen as
K   n   n  n1   n1  1  1 
• The last step in the design algorithm is to transform
the feedback control law to the original coordinates,
i.e., to obtain K from K . This is done by
K  KP

• This proves that if the system is controllable, its

eigenvalues can be arbitrarily located by state
feedback. Thus, an unstable system can be
stabilized by complete state feedback if all the states
are controllable.
Ackermann’s Formula:
• In general, the techniques for developing state
models for plants do not result in the controllable
canonical form. There are other approaches for
determination of state-feedback gain matrix K.
• One such well-known formula is known as
Ackermann’s Formula.
K  KP
 p1 
 pA 
P 1  ; p1  0 0  1S 1
  
 n 1 
p1A 
From these Eqs.,
 0  0 1S 1 
 
0  0 1S A 
1
K   n   n  n1   n1  1  1 
  
 1 n 1 
0  0 1S A 

 0  0 1S 1[(1  1 ) A n1  ( 2   2 ) A n2    ( n   n )I]

• From Cayley-Hamilton Theorem,

A n  1A n1     n1A   n I  0

  A n  1A n1     n1A   n I

• Combining these equations,
K  0  0 1S 1Φ( A)
n 1
where Φ ( A )  A n
  1 A     n1A   n I

• These two equations describe the Ackermann’s

formula for determination of the state feedback gain
matrix K.
Deadbeat Control:
• A case of special interest occurs when a state
feedback control law is chosen which places all the
closed-loop poles at the origin, i.e.,
det[ zI  ( A  bK )]  z  0
n

• According to Cayley-Hamilton theorem,

( A  bK )  0
n

This implies that the force-free response of the closed-

loop system x(k  1)  ( A  bK)x(k )  br (k ) is:

x(k )  ( A  bK ) x  0
k 0
for kn
 0
• In other words, any initial state x is driven to zero in
(at most) n steps.
• The feedback control law that assigns all the
closed-loop poles to the origin is therefore a
deadbeat control law.
Example:
A deadbeat control law for the following system is to
be determined.
x(k  1)  Ax(k )  bu(k )

1 0.0787  0.0043 
where A  ; b 
 0 0.6065   0.0787 
• The system characteristic polynomial is
( z  1)( z  0.6065 )  z 2  1.6065 z  0.6065  z 2  1 z   2
The controllability matrix is:
0.0043 0.0105 
S  b Ab    
 0.0787 0.0477 
• The vector p1 which is the last row of S 1 is:
p1  126 .9355  6.9355 

 p1  126.9355  6.9355 
Thus, P     
 1  
p A 126 .9355 5.7834 
• The transformation x  Px converts the system to
the following controllable canonical form
 0 1  0 
x(k  1)    x( k )   u
 0.6065 1.6065  1

• It is immediately seen that in terms of the

transformed state, the deadbeat control law is
given by

u  Kx  r where K  0.6065  1.6065

• In terms of the original state, we have

u  Kx  r
where K  KP   126.9385  13.4974 