ME 547: Linear Systems

State-Space Canonical Forms

Xu Chen

University of Washington

UW Linear Systems (ME) State-space canonical forms 1 / 31

Goal
The realization problem:
I existence and uniqueness: the same system can have infinite amount
of state-space representations: e.g.
( (
ẋ = Ax + Bu ẋ = Ax + 12 Bu
y = Cx y = 2Cx

B
    
A B A 2
packed representation: Σ1 = , Σ2 =
C D 2C D
I canonical realizations exist
I relationship between different realizations?
I example problem for this set of notes:

b2 s 2 + b1 s + b0
G (s) = . (1)
s 3 + a2 s 2 + a1 s + a0

UW Linear Systems (ME) State-space canonical forms 2 / 31

Outline

1. CT controllable canonical form

2. CT observable canonical form

3. CT diagonal and Jordan canonical forms

4. Modified canonical form

5. DT state-space canonical forms

6. Similar realizations

UW Linear Systems (ME) State-space canonical forms 3 / 31

Controllable canonical form (ccf)
Choose x1 such that

/ 1 x1
/ b s2 + b s + b /y
u 2 1 0 (2)
s3 + a2 s2 + a1 s + a0

U(s) ...
X1 (s) = ⇒ x 1 + a2 ẍ1 + a1 ẋ1 + a0 x1 = u
s3 2
+ a2 s + a1 s + a0
Let x2 = ẋ1 , x3 = ẋ2 . Then ẋ3 = −a2 x3 − a1 x2 − a0 x1 + u.
Y (s) = b2 s 2 + b1 s + b0 X1 (s) ⇒ y = b2 ẍ1 +b1 ẋ1 +b0 x1


|{z} |{z}
x3 x2

Putting in matrix form yields

      
x (t) 0 1 0 x1 (t) 0
d  1
x2 (t)  =  0 0 1  x2 (t)  +  0  u(t) (3)
dt
x3 (t) −a0 −a1 −a2 x3 (t) 1
 
  x1 (t)
y (t) = b0 b1 b2  x2 (t) 
x3 (t)
UW Linear Systems (ME) State-space canonical forms 4 / 31
Block diagram realization of controllable canonical forms
      
x1 (t) 0 1 0 x1 (t) 0
d 
x2 (t)  =  0 0 1   x2 (t)  +  0  u(t) (4)
dt
x3 (t) −a0 −a1 −a2 x3 (t) 1
 
  x1 (t)
y (t) = b0 b1 b2  x2 (t) 
x3 (t)
b2

+
b1
U (s) + Y (s)
+ 1 X3 1 X2 1 X1
b0
s s s
−
+
a2

+
a1

a0
UW Linear Systems (ME) State-space canonical forms 5 / 31
General ccf

For a single-input single-output transfer function

bn−1 s n−1 + · · · + b1 s + b0
G (s) = + d,
s n + an−1 s n−1 + · · · + a1 s + a0

we can verify that

 
0 1 ··· 0 0 0
 0 0 ··· 0 0 0 
.. .. .. .. ..
   
Ac Bc
 
Σc = = 
 . . ··· . . .
(5)
Cc Dc

 0
 0 ··· 0 1 0 

 −a0 −a1 ··· −an−2 −an−1 1 
b0 b1 ··· bn−2 bn−1 d

realizes G (s). This realization is called the controllable canonical form.

UW Linear Systems (ME) State-space canonical forms 6 / 31

1. CT controllable canonical form

2. CT observable canonical form

3. CT diagonal and Jordan canonical forms

4. Modified canonical form

5. DT state-space canonical forms

6. Similar realizations

UW Linear Systems (ME) State-space canonical forms 7 / 31

Observable canonical form (ocf)
b2 s 2 + b1 s + b0
Y (s) = U(s)
s 3 + a2 s 2 + a1 s + a0
a2 a1 a0 b2 b1 b0
⇒ Y (s) = − Y (s) − 2 Y (s) − 3 Y (s) + U(s) + 2 U(s) + 3 U(s).
s s s s s s

In a block diagram, the above looks like

b2

b1

U (s) + 1 + 1 + 1 Y (s)
b0
s s s
− − −

a2

a1

a0

UW Linear Systems (ME) State-space canonical forms 8 / 31

Observable canonical form
b2

b1

U (s) + + Y (s)
+ 1 X3 1 X2 1 X1
b0
s s s
− − −

a2

a1

a0

Here, the states are connected by

Y (s) = X1 (s) y (t) = x1 (t)
sX1 (s) = −a2 X1 (s) + X2 (s) + b2 U(s) ẋ1 (t) = −a2 x1 (t) + x2 (t) + b2 u(t)
sX2 (s) = −a1 X1 (s) + X3 (s) + b1 U(s) ⇒ ẋ2 (t) = −a1 x1 (t) + x3 (t) + b1 u(t)
sX3 (s) = −a0 X1 (s) + b0 U(s) ẋ3 (t) = −a0 x1 (t) + b0 u(t)

UW Linear Systems (ME) State-space canonical forms 9 / 31

Observable canonical form


 = −a2 x1 (t) + x2 (t) + b2 u(t)
ẋ1 (t)

ẋ (t)
2 = −a1 x1 (t) + x3 (t) + b1 u(t)


 ẋ = −a0 x1 (t) + b0 u(t)
3 (t)
y (t) = x1 (t)

   
−a2 1 0 b2
⇒ẋ(t) =  −a1 0 1  x(t) +  b1  u(t)
−a0 0 0 b0
| {z } | {z }
Ao Bo
 
y (t) = 1 0 0 x(t)
| {z }
Co

The above is called the observable canonical form realization of G (s).

Exercise
Verify that Co (sI − Ao )−1 Bo = G (s).
UW Linear Systems (ME) State-space canonical forms 10 / 31
General ocf

In the general case, the observable canonical form of the transfer function

bn−1 s n−1 + · · · + b1 s + b0
G (s) = +d
s n + an−1 s n−1 + · · · + a1 s + a0
is  
−an−1 1 · · · 0 0 bn−1
 −an−2 0 · · · 0 0 bn−2 0 
.. .. .. .. ..
   
Ao Bo
 
Σo = = 
 . . ··· . . . .

(6)
Co Do  −a1 0 · · · 0 1 b1 
 
 −a0 0 · · · 0 0 b0 
1 0 ··· 0 0 d

UW Linear Systems (ME) State-space canonical forms 11 / 31

1. CT controllable canonical form

2. CT observable canonical form

3. CT diagonal and Jordan canonical forms

4. Modified canonical form

5. DT state-space canonical forms

6. Similar realizations

UW Linear Systems (ME) State-space canonical forms 12 / 31

Diagonal form
When
B(s) b2 s 2 + b1 s + b0
G (s) = = 3
A(s) s + a2 s 2 + a1 s + a0
and the poles p1 6= p2 6= p3 , partial fractional expansion yields
k1 k2 k3 B(s)
G (s) = + + , ki = lim (s − pi ) ,
s − p1 s − p2 s − p3 p→pi A(s)
namely + sX1 1 X1
k1
s
+
p1
U (s) + Y (s)
+ sX2 1 X2
k2
s
+ +
p2

+ sX3 1 X3
k3
s
+
p3
UW Linear Systems (ME) State-space canonical forms 13 / 31
Diagonal form
+ sX1 1 X1
k1
s
+
p1
U (s) + Y (s)
+ sX2 1 X2
k2
s
+ +
p2

+ sX3 1 X3
k3
s
+
p3

The state-space realization of the above is

   
p1 0 0 1  
A =  0 p2 0  , B =  1  , C = k1 k2 k3 , D = 0.
0 0 p3 1

UW Linear Systems (ME) State-space canonical forms 14 / 31

Jordan form

If poles repeat, say,

b2 s 2 + b1 s + b0 b2 s 2 + b1 s + b0
G (s) = = , p1 6= pm ∈ R,
s 3 + a2 s 2 + a1 s + a0 (s − p1 )(s − pm )2

then partial fraction expansion gives


k1
 = lims→p1 G (s)(s − p1 )
k1 k2 k3
G (s) = + + w/ k2 = lims→pm G (s)(s − pm )2
s − p1 (s − pm )2 s − pm  d
G (s)(s − pm )2

k3 = lims→pm ds


UW Linear Systems (ME) State-space canonical forms 15 / 31

Jordan form

k1 k2 k3
G (s) = + +
s − p1 (s − pm )2 s − pm
has the block diagram realization:
+ 1 X1
k1
s
+
p1
U (s) + Y (s)
k3
+
X3
+ 1 + 1 X2
k2
s s
+ +
pm pm

UW Linear Systems (ME) State-space canonical forms 16 / 31

Jordan form
+ 1 X1
k1
s
+
p1
U (s) + Y (s)
k3
+
X3
+ 1 + 1 X2
k2
s s
+ +
pm pm

The state-space realization of the above, called the Jordan canonical form,
is
   
p1 0 0 1  
A =  0 pm 1  , B =  0  , C = k1 k2 k3 , D = 0.
0 0 pm 1

UW Linear Systems (ME) State-space canonical forms 17 / 31

1. CT controllable canonical form

2. CT observable canonical form

3. CT diagonal and Jordan canonical forms

4. Modified canonical form

5. DT state-space canonical forms

6. Similar realizations

UW Linear Systems (ME) State-space canonical forms 18 / 31

Modified canonical form
If the system has complex poles, say,
b2 s 2 + b1 s + b0 k1 αs + β
G (s) = = +
3 2
s + a2 s + a1 s + a0 s − p1 (s − σ)2 + ω 2
then we have
+ 1 X1
k1
+ s

p1
U (s) Y (s)
+
k3
+
X3
+ 1 + 1 X2
ω k2
+ s + s
+
σ σ
−
ω

where k2 = (β + ασ)/ω and k3 = α.

UW Linear Systems (ME) State-space canonical forms 19 / 31
Modified canonical form
+ 1 X1
k1
+ s

p1
U (s) Y (s)
+
k3
+
X3
+ 1 + 1 X2
ω k2
+ s + s
+
σ σ
−
ω

⇒ modified Jordan form:

   
p1 0 0 1  
A=  0 σ ω , B = 0  , C = k1 k2 k3 , D = 0.
 
0 −ω σ 1

UW Linear Systems (ME) State-space canonical forms 20 / 31

1. CT controllable canonical form

2. CT observable canonical form

3. CT diagonal and Jordan canonical forms

4. Modified canonical form

5. DT state-space canonical forms

6. Similar realizations

UW Linear Systems (ME) State-space canonical forms 21 / 31

DT state-space canonical forms
I The procedures for finding state space realizations in discrete time is
similar to the continuous time cases. The only difference is that we use

Z {x(k + n)} = z n X (z),

instead of
dn
 
L x(t) = s n X (s),
dt n
assuming zero state initial conditions.
I Fundamental relationships:

x (k) / z −1 / x (k − 1)

X (z) / z −1 / z −1 X (z)

x (k + n) / z −1 / x (k + n − 1)

UW Linear Systems (ME) State-space canonical forms 22 / 31

DT ccf

b2 z 2 + b1 z + b0
G (z) =
z 3 + a2 z 2 + a1 z + a0
I Same transfer-function structure ⇒ same A, B, C , D matrices of the
canonical forms as those in continuous-time cases
I Controllable canonical form:
      
x1 (k + 1) 0 1 0 x1 (k) 0
 x2 (k + 1)  =  0 0 1   x2 (k)  +  0  u (k)
x3 (k + 1) −a0 −a1 −a2 x3 (k) 1
 
  x1 (k)
y (k) = b0 b1 b2  x2 (k) 
x3 (k)

UW Linear Systems (ME) State-space canonical forms 23 / 31

DT ocf

b2 z 2 + b1 z + b0
G (z) =
z 3 + a2 z 2 + a1 z + a0
I Observable canonical form:
      
x1 (k + 1) −a2 1 0 x1 (k) b2
 x2 (k + 1)  =  −a1 0 1   x2 (k)  +  b1  u (k)
x3 (k + 1) −a0 0 0 x3 (k) b0
 
  x1 (k)
y (k) = 1 0 0  x2 (k) 
x3 (k)

UW Linear Systems (ME) State-space canonical forms 24 / 31

DT diagonal form

b2 z 2 + b1 z + b0
G (z) =
z 3 + a2 z 2 + a1 z + a0
I Diagonal form (distinct poles):

k1 k2 k3
G (z) = + +
z − p1 z − p2 z − p3

      
x1 (k + 1) p1 0 0 x1 (k) 1
 x2 (k + 1)  =  0 p2 0   x2 (k)  +  1  u (k)
x3 (k + 1) 0 0 p3 x3 (k) 1
 
  x1 (k)
y (k) = k1 k2 k3  x2 (k) 
x3 (k)

UW Linear Systems (ME) State-space canonical forms 25 / 31

DT Jordan form 1

b2 z 2 + b1 z + b0
G (z) =
z 3 + a2 z 2 + a1 z + a0
I Jordan form (2 repeated poles):

k1 k2 k3
G (z) = + 2
+
z − p1 (z − pm ) z − pm

      
x1 (k + 1) p1 0 0 x1 (k) 1
 x2 (k + 1)  =  0 pm 1   x2 (k)  +  0  u (k)
x3 (k + 1) 0 0 pm x3 (k) 1
 
  x1 (k)
y (k) = k1 k2 k3  x2 (k) 
x3 (k)

UW Linear Systems (ME) State-space canonical forms 26 / 31

DT Jordan form 2
b2 z 2 + b1 z + b0
G (z) =
z 3 + a2 z 2 + a1 z + a0
I Jordan form (2 complex poles):

k1 αz + β
G (s) = +
z − p1 (z − σ)2 + ω 2

      
x1 (k + 1) p1 0 0 x1 (k) 1
 x2 (k + 1)  =  0 σ ω   x2 (k)  +  0  u (k)
x3 (k + 1) 0 −ω σ x3 (k) 1
 
  x1 (k)
y (k) = k1 k2 k3  x2 (k) 
x3 (k)

where k2 = (β + ασ)/ω, k3 = α.
UW Linear Systems (ME) State-space canonical forms 27 / 31
Exercise

obtain the controllable canonical form:

I G (s) = 1+2zz −1 −z −3
−1 +z −2

UW Linear Systems (ME) State-space canonical forms 28 / 31

1. CT controllable canonical form

2. CT observable canonical form

3. CT diagonal and Jordan canonical forms

4. Modified canonical form

5. DT state-space canonical forms

6. Similar realizations

UW Linear Systems (ME) State-space canonical forms 29 / 31

Relation between different realizations
Similar realizations
Given one realization Σ of a transfer function G (s) and a nonsingular
T ∈ Rn×n , we can define new states:

Tx ∗ = x.

Then
d
ẋ(t) = Ax(t) + Bu(t) ⇒ (Tx ∗ (t)) = ATx ∗ (t) + Bu(t),
dt
ẋ ∗ (t) = T −1 ATx ∗ (t) + T −1 Bu(t)

∗
⇒Σ :
y (t) = CTx ∗ (t) + Du(t)
namely
T −1 AT T −1 B
 
Σ∗ =
CT D
also realizes G (s) and is said to be similar to Σ.
UW Linear Systems (ME) State-space canonical forms 30 / 31
Relation between different realizations
Similar realizations

Exercise (Another observable canonical form.)

Verify that the following realize the same system

−a2 1 0 b2 0 −a0 b0
   
0
 −a1 0 1 b1  0 −a1 b1 
∗  1

Σ=  −a0 0 0 b0  , Σ =  0
 
1 −a2 b2 
1 0 0 d 0 0 1 d

UW Linear Systems (ME) State-space canonical forms 31 / 31