State Feedback Control

There are some interest to change the position of open loop poles
i.e improve performance, stability etc. This can be done by
introducing feedback gain K to the state equation 

D
x  Ax  Bu
y  Cx  Du
u

r +
x y
x

-
B 
 C 

+K
 Feedback Gain
• Introduce feedback gain K u   Kx  r

+ u
r x  Ax  Bu y

y  Cx  Du
-

• By substitution

x  Ax  B(r  Kx)  ( A  BK ) x  Br
 ACL x  Br
 Feedback Gain
• The output equation is given as

y  Cx  D(r  Kx)
y  (C  DK ) x  Dr
• Now the poles of the system can be determined from the
characteristic equation

I  A  BK  0
• Example: Stabilise the system and place the poles at -5 and -6
1 1  1 
x  x   u
1 2 0 
 pole placement example

• Locate the poles/eigenvalues

 1 1
det(sI  A)  I  A   (  1)(  2)  1  0
1   2
2  3  2  1  0
2  3  1  0
• system unstable. Introduce state feedback gain K

u   Kx   k1 k2  x
 Pole placement example

• ACL  A  BK
1 1  1
ACL       k1 k 2 
1 2 0
1  k1 1  k 2 
ACL  
 1 2 

  1  k1  1  k2
I  ACL   (  1  k1 )(  2)  (1  k 2 )  0
1 2
2  2    2  k1  2k1  (1  k 2 )  0
2  (k1  3)  (1  2k1  k 2 )  0
 pole placement example
• By choosing and we can put eig( ACL ) anywhere in the complex plane
• let poles located at -5 and -6. Desired close loop equation

( s  5)(s  6)  s  11s  30
2

k1  3  11 1  2k1  k 2  30
k1  14 1  28  k 2  30
• k2
Hence the feedback gain is  57

•
K  14 57
Can all poles be shifted using feedback gain? No
• What is the condition- system must be controllable
 Uncontrollable example
• Example
1 1 1
x  x   u
0 2  0 

u   Kx

ACL  A  BK
1 1 1
ACL       k1 k2 
0 2  0 
1  k1 1  k 2 
ACL  
 0 2 
 Uncontrollable example

  (1  k1 )  (1  k 2 )
I  ACL      (1  k1 ) (  2)  0
0  2
2  2  (1  k1 )(  2)  0
2  2    2  k1  2k1 )  0
2  3  k1  2  2k1  0
2  (3  k1 )  2(1  k1 )  0
• Factoring to give the poles ( s  1  k1 )( s  2)
• Feedback state cannot move the pole at s=2
 Preview on Pole Placement via
State Feedback
• Given a system (linear,
 time-invariant)
x(t )  Ax(t )  Bu (t )
• we can apply a linear feedback control law
u (t )   Kx(t )
to form a feedback loop

x(t )  ( A  BK ) x(t )
• the pole placement/eigenvalue assignment can be used to change
the open loop eigenvalues 1 ,  2 ,....,  n of matrix A to the desired
close loop eigenvalues of matrix (A-BK)
 Preview on Pole Placement via
State Feedback
• The characteristic equation of A
n 1
p( s )  sI  A  s  a1 sn
 ......  a n 1 s  a n
• will be transformed into characteristic equation of (A-BK)

pˆ ( s )  sI  A  BK  s n  aˆ1 s n 1  ......  aˆ n 1 s  aˆ n
• Condition of state feedback – Theorem

There exists a state feedback matrix K which assigns to the matrix A-

BK of the closed-loop system any arbitrary eigenvalues ˆ1 , ˆ2 ,...., ˆn
iff the state vector of the open loop system is controllable i.e
rank S = n where 
S  B AB A 2 B  A n 1 B 
 Base-Gura method
• Two more methods of calculating K are presented
• Method 1: The Base-Gura Formula

where
KW S  T

T 1
(aˆ  a )
1 a1  a n 1   aˆ1   a1 
0 1  a  aˆ  a 
W  n2 
aˆ   2  a   2
        
     
 0 0  1  aˆ n  a n 
p( s)  sI  A  s n  a1 s n 1  ......  a n 1 s  a n
pˆ ( s )  sI  A  BK  s n  aˆ1 s n 1  ......  aˆ n 1 s  aˆ n
 Base-Gura example

• Controllability matrix S B  AB A 2 B  A n 1 B 
• Example: Redo example above using Base-Gura

 1 1  1  p( s )  s 2  3s  1  0
x  x   u
1 2 0  pˆ ( s)  s 2  11s  30
1  3  1 0
W   W 
T

 0 1    3 1 

1 1 1 0
S  S 
T

 0 1 1 1 
 Base-Gura example

•  1 0 1 0  1 0
W S 
T T
    
  3 1  1 1    2 1 
1 1 0  1 0 
1
(W S )  
T T

1 2 1 2 1


 aˆ1   a1  11  (3) 14 

aˆ  a           
aˆ 2  a 2   30  (1)  29

KW S  T

T 1 1 0 14 
(aˆ  a )  
 2 1  29
 
 14  14 
   
28  29 57 in column vector
 Ackermann’s Formula
• Method 2: The Ackermann’s Formula

K T  e T S 1 pˆ ( A)
• where S is controllability matrix

and
S B  AB  A n 1 B 
e T   0 0  1

The last part came from the desired characteristic equation

characteristic equation
pˆ ( s )  s n  aˆ1 s n 1  ......  aˆ n 1 s  aˆ n
pˆ ( A)  A n  aˆ1 A n 1  ....  aˆ n 1 A  aˆ n I
 Example
• Example: Redo example using Ackermann’s Formula
 1 1  1
x(t )    x(t )   u (t )
1 2 0 
we like the poles to be at -5 and -6
Soln:
1 1 1  1
S  B AB 
1
  S  
 0 1  0 1 

1  1
e   0 1
T T
e S 1
  0 1     0 1
0 1 
 Solution to ackerman
pˆ ( s)  s 2  11s  30
pˆ ( A)  A 2  11 A  30 I
2
1 1  1 1 1 0 
   11   30 
1 2   1 2   0 1 
1 1  1 1 11 11  30 0 
       
1 2  1 2   11 22   0 30 
1  1 1  2 
 
1  2 1  4 
2 3 11 11  30 0  43 14 
       
 3 5   11 22   0 30   14 57 
 Solution to Ackerman
43 14 
K   0 1 
T
  14 57
14 57
• This method is numerically unstable when implemented on
computers.
• Matlab has command acker.m

• For the multivariable pole placement case: Please refer to

1) Kailath, T. (1980), Linear Systems, Prentice Hall (Chapter 13&14)
2) Rugh, W. J. (1996), Linear System Theory, Prentice Hall
(Chapter
6 and 7)