ECE 486: Control Systems

I Lecture 22: controllability, stability, and pole-zero

cancellations; effect of coordinate transformations;
conversion of any controllable system to CCF.

Goal: explore the effect of pole-zero cancellations on internal

stability; understand the effect of coordinate transformations on
the properties of a given state-space model (transfer function;
open-loop poles; controllability).

Reading: FPE, Chapter 7

State-Space Realizations

u ẋ = Ax + Bu y
y = Cx

I a given transfer function G(s) can be realized using

infinitely many state-space models
I certain properties make some realizations preferable to
others
I one such property is controllability
Controllability Matrix
Consider a single-input system (u ∈ R):

ẋ = Ax + Bu, y = Cx x ∈ Rn

The Controllability Matrix is defined as

C(A, B) = B | AB | A2 B | . . . | An−1 B
 

We say that the above system is controllable if its

controllability matrix C(A, B) is invertible.
I As we will see later, if the system is controllable, then we
may assign arbitrary closed-loop poles by state feedback of
the form u = −Kx.
I Whether or not the system is controllable depends on its
state-space realization.
Example: Computing C(A, B)
Let’s get back to our old friend:
        
ẋ1 0 1 x1 0
x1
= + u, y= 1 1
ẋ2 −6 −5 x2 1 | {z } x2
| {z } |{z} C
A B

Here, x ∈ R2 =⇒ A ∈ R2×2 =⇒ C(A, B) ∈ R2×2

    
0 1 0 1
C(A, B) = [B | AB] AB = =
−6 −5 1 −5
 
0 1
=⇒ C(A, B) =
1 −5

Is this system controllable?

det C = −1 6= 0 =⇒ system is controllable

Controller Canonical Form
A single-input state-space model

ẋ = Ax + Bu, y = Cx

is said to be in Controller Canonical Form (CCF) is the

matrices A, B are of the form
   
0 1 0 ... 0 0 0
0 0 1 . . . 0 0  0
A =  ... ... ... . . . ... ...  , B =  ... 
   
   
   
0 0 0 . . . 0 1  0
∗ ∗ ∗ ... ∗ ∗ 1

A system in CCF is always controllable!!

(The proof of this for n > 2 uses the Jordan canonical form, we will
not worry about this.)
CCF with Arbitrary Zeros
s+1
In our example, we had G(s) = , with a
s2 + 5s + 6
minimum-phase zero at z = −1.
Let’s consider a general zero location s = z:
s−z
G(s) =
s2 + 5s + 6
This gives us a CCF realization
        
ẋ1 0 1 x1 0
x1
= + u, y = −z 1
ẋ2 −6 −5 x2 1 | {z } x2
| {z } |{z} C
A B

Since A, B are the same, C(A, B) is the same =⇒ the system is

still controllable.
A system in CCF is controllable for any locations of the zeros.
OCF with Arbitrary Zeros

Start with the CCF

        
ẋ1 0 1 x1 0
x1
= + u, y = −z 1
ẋ2 −6 −5 x2 1 | {z } x2
| {z } |{z} C
A B

Convert to OCF: (A 7→ AT , B 7→ C T , C 7→ B T )
        
ẋ1 0 −6 x1 −z
x1
= + u, y= 0 1
ẋ2 1 −5 x2 1 | {z } x2
| {z } | {z } C̄=B T
Ā=AT B̄=C T

We already know that this system realizes the same t.f. as the
original system.
But is it controllable?
OCF with Arbitrary Zeros
        
ẋ1 0 −6 x1 −z
x1
= + u, y= 0 1
ẋ2 1 −5 x2 1 | {z } x2
| {z } | {z } C̄=B T
Ā=AT B̄=C T

Let’s find the controllability matrix:

    
  0 −6 −z −6
C(Ā, B̄) = B̄ | ĀB̄ ĀB̄ = =
1 −5 1 −z − 5
 
−z −6
∴ C(Ā, B̄) =
1 −z − 5
det C = z(z + 5) + 6 = z 2 + 5z + 6 = 0 for z = −2 or z = −3

The OCF realization of the transfer function

s−z
G(s) = 2 is not controllable when z = −2 or −3,
s + 5s + 6
even though the CCF is always controllable.
Beware of Pole-Zero Cancellations!
The OCF realization of the transfer function
s−z
G(s) =
s2 + 5s + 6
is not controllable when z = −2 or −3, even though the CCF is
always controllable.
Let’s examine G(s) when z = −2:
s − z  s+
 2 1
G(s) = 2
= =
s + 5s + 6 z=−2 (
s + 2)(s + 3) s+3
 

— pole-zero cancellation!
For z = −2, G(s) is a first-order transfer function, which can
always be realized by this 1st-order controllable model:
1
ẋ1 = −3x1 + u, y = x1 −→ G(s) =
s+3
Beware of Pole-Zero Cancellations!!
We can look at this from another angle: consider the t.f.
1
G(s) =
s+3
We can realize it using a one-dimensional controllable
state-space model

ẋ1 = −3x1 + u, y = x1

or a noncontrollable two-dimensional state-space model

        
ẋ1 0 −6 x1 2
x1
= + u, y= 0 1
ẋ2 1 −5 x2 1 x2

— certainly not the best way to realize a simple t.f.!

Thus, even the state dimension of a realization of a given t.f.
is not unique!!
Beware of Pole-Zero Cancellations!!
Here is a really bad realization of the t.f.
1
G(s) = .
s+3
Use a two-dimensional model:

ẋ1 = −3x1 + u
ẋ2 = 100x2
y = x1

I x2 is not affected by the input u (i.e., it is an

uncontrollable mode), and not visible from the output y
I does not change the transfer function
I ... and yet, horrible to implement: x2 (t) ∝ e100t
The transfer function can mask undesirable internal state
behavior!!
Pole-Zero Cancellations and Stability
I In case of a pole-zero cancellation, the t.f. contains much
less information than the state-space model because some
dynamics are “hidden.”
I These dynamics can be either good (stable) or bad
(unstable), but we cannot tell from the t.f.
I Our original definition of stability (no RHP poles) is flawed
because there can be RHP eigenvalues of the system matrix
A that are canceled by zeros,yet they still have dynamics
associated with them.
Definition of Internal Stability (State-Space Version): a
state-space model with matrices (A, B, C, D) is internally
stable if all eigenvalues of the A matrix are in LHP.
This is equivalent to having no RHP open-loop poles and no
pole-zero cancellations in RHP.
Coordinate Transformations

Now that we have seen that a given transfer function can have
many different state-space realizations, we would like a
systematic procedure of generating such realizations, preferably
with favorable properties (like controllability).
One such procedure is by means of coordinate transformations.
Coordinate Transformations
x2

x̄ 1
x̄ 2

x1
0

x 7−→ x̄ = T x, T ∈ Rn×n nonsingular

x = T −1 x̄ (go back and forth between the coordinate systems)
Coordinate Transformations
For example,
     
x1 x̄1 x1 + x2
7−→ =
x2 x̄2 x1 − x2

This can be represented as

 
1 1
x̄ = T x, where T =
1 −1

The transformation is invertible: det T = −2, and

  1 1

−1 1 −1 −1 2 2
T = = 1 1
det T −1 1 2 −2

Or we can see this directly:

x̄1 + x̄2 = 2x1 ; x̄1 − x̄2 = 2x2

Coordinate Transformations and State-Space Models
Consider a state-space model

ẋ = Ax + Bu
y = Cx

and a change of coordinates x̄ = T x (T invertible).

What does the system look like in the new coordinates?

x̄˙ = T˙x = T ẋ (linearity of derivative)

= T (Ax + Bu)
= T (AT −1 x̄ + Bu) (x = T −1 x̄)
−1
=T
| AT
{z } x̄ + |{z}
TB u
Ā B̄
y = Cx
−1
= CT
| {z } x̄
C̄
Coordinate Transformations and State-Space Models

T
ẋ = Ax + Bu −−−−→ x̄˙ = Āx̄ + B̄u
y = Cx y = C̄ x̄

where

Ā = T AT −1 , B̄ = T B, C̄ = CT −1

What happens to
I the transfer function?
I the controllability matrix?
Coordinate Transformations and State-Space Models
T
ẋ = Ax + Bu −−−−→ x̄˙ = Āx̄ + B̄u
y = Cx y = C̄ x̄
where Ā = T AT −1 , B̄ = T B, C̄ = CT −1

Claim: The transfer function doesn’t change.

Proof:
Ḡ(s) = C̄(Is − Ā)−1 B̄

−1
= (CT −1 ) Is − T AT −1 (T B)
−1 −1 −1 −1


= CT T IT s − T AT TB
−1 −1
T (Is − A) T −1
 
= CT TB
−1 −1 −1
=CT
| {z T} (Is − A) T| {z T} B
I I
−1
= C (Is − A) B ≡ G(s)
Coordinate Transformations and State-Space Models

T
ẋ = Ax + Bu −−−−→ x̄˙ = Āx̄ + B̄u
y = Cx y = C̄ x̄
where Ā = T AT −1 , B̄ = T B, C̄ = CT −1

The transfer function doesn’t change.

In fact:
I open-loop poles don’t change
I characteristic polynomial doesn’t change:

det(Is − Ā) = det(Is − T AT −1 )

= det T (Is − A)T −1
 

= det T · det(Is − A) · det T −1

= det(Is − A)
Coordinate Transformations and State-Space Models

T
ẋ = Ax + Bu −−−−→ x̄˙ = Āx̄ + B̄u
y = Cx y = C̄ x̄
−1 −1
where Ā = T AT , B̄ = T B, C̄ = CT

Claim: Controllability doesn’t change.

Proof: For any k = 0, 1, . . .,

Āk B̄ = (T AT −1 )k T B = T Ak T −1 T B = T Ak B (by induction)

n−1
Therefore, C(Ā, B̄) = [T B | T AB | . . . | T A B]
n−1
= T [B | AB | . . . | A B]
= T C(A, B)

Since det T 6= 0, det C(Ā, B̄) 6= 0 if and only if det C(A, B) 6= 0.

Thus, the new system is controllable if and only if the old one is.
Coordinate Transformations and State-Space Models
T
ẋ = Ax + Bu −−−−→ x̄˙ = Āx̄ + B̄u
y = Cx y = C̄ x̄
where Ā = T AT −1 , B̄ = T B, C̄ = CT −1

Note: The controllability matrix does change:

C(Ā, B̄) = |{z}

T C(A, B)
| {z } | {z }
coord.
new trans. old

m
T = C(Ā, B̄) [C(A, B)]−1

This is a recipe for going from one controllable realization of a

given t.f. to another.
CCF is the most convenient controllable realization of a given
t.f., so we want to convert a given controllable system to CCF
(useful for control design).
Example: Converting a Controllable System to CCF

Note!! The way I do this is different from the textbook.

   
−15 8 1
Consider A = , B= (C is immaterial).
−15 7 1
Convert to CCF if possible.

Step 1: check for controllability.

      
−15 8 1 −7 1 −7
AB = = =⇒ C=
−15 7 1 −8 1 −8
det C = −1 – controllable
Example: Converting a Controllable System to CCF
Step 2: Determine desired C(Ā, B̄).
We need to figure out Ā and B̄.
For CCF, we must have
   
0 1 0
Ā = , B̄ = ,
−a2 −a1 1

so we need to find the coefficients a1 , a2 .

Recall: the characteristic polynomial does not change:

det(Is − A) = det(Is − Ā)

   
s + 15 −8 s −1
det = det
15 s−7 a2 s + a1
(s + 15)(s − 7) + 120 = s(s + a1 ) + a2
s2 + 8s + 15 = s2 + a1 s + a2
Example: Converting a Controllable System to CCF
Step 2: Determine desired C(Ā, B̄).
We need to figure out Ā and B̄.
For CCF, we must have
   
0 1 0
Ā = , B̄ = .
−a2 −a1 1

We have just computed

   
0 1 0
Ā = , B̄ =
−15 −8 1

Therefore, the new controllability matrix should be

 
0 1
C(Ā, B̄) = [B̄ | ĀB̄] =
1 −8
Example: Converting a Controllable System to CCF
Step 3: Compute T .
Recall: T = C(Ā, B̄) · [C(A, B)]−1
 
1 −7
C(A, B) =
1 −8
 −1
−1 1 −7
[C(A, B)] =
1 −8
   
1 −8 7 8 −7
= =
−1 −1 1 1 −1
 
0 1
C(Ā, B̄) =
1 −8
  
0 1 8 −7
T =
1 −8 1 −1
 
1 −1
=
0 1
In the next lecture, we will see why CCF is so useful.