ECE 486

STATE-SPACE REALIZATIONS

Fall 08

Reading: FPE, Section 7.4.1

Suppose we have a linear system with transfer function H(s) (which can be a matrix, in general). We have
seen that the transfer function is related to the matrices in the state space model via H(s) = C(sIA)1 B.
In this section, we examine the question: How many choices of A, B and C produce this transfer function
H(s)? We will see that there are multiple models (or realizations) that correspond to the same transfer
function.

Controllable Canonical Form

Example. Consider the system y (3) + 7

y + 14y + 8y = u
2u + 3u. Draw an all-integrator block diagram
for this system, and create the corresponding state-space model.
Solution.

To generalize the above model, consider the transfer function

H(s) =

an1 sn1 + an2 sn2 + + a1 s + a0

.
sn + bn1 sn1 + + b1 s + b0

Note that the above form also captures transfer functions that have numerator polynomials with degree
less than n 1 by setting the appropriate coefficients ai to zero. By using the same technique as in the
example above, an all-integrator block diagram for this transfer function is given by:

Based on this diagram, the state-variables evolve according to

x 1 = x2
x 2 = x3
..
.
x n1 = xn
x n = b0 x1 b1 x2 bn2 xn1 bn1 xn + u
y = a0 x1 + a1 x2 + + an2 xn1 + an1 xn .
In matrix-vector form, this corresponds to:

0
1
0
0

0
0
1
0

.
.
.
..
.
..
..
..
..
x =
.

0
0
0
1
b0 b1 b2 bn1


y = a0 a1 a2 an1 x .

.
x + .. u

0
1

This form is called the controllable canonical form (for reasons that we will see later). Note how the
coefficients of the transfer function show up in the matrix: each of the denominator coefficients shows up
negated and in reverse order in the bottom row of A. The matrix B consists of all zeros, except for the
last entry, which is 1. The C matrix has the numerator coefficients in reverse order. The most important
point is that the coefficients of the denominator polynomial are readily available in this form.

Modal Canonical Form

We will now develop another commonly used state-space realization.

Example. Consider again the system y (3) +7
y +14y +8y = u
2u+3u.

Find the partial fraction expansion

Y (s)
of the transfer function H(s) = U (s) , and then draw an all-integrator block diagram by linking together
block diagrams for each of the terms in the expansion. Use this block diagram to create a state-space
model.
Solution.

To generalize this realization, consider the transfer function

H(s) =

an1 sn1 + an2 sn2 + + a1 s + a0

.
sn + bn1 sn1 + + b1 s + b0

Suppose that the transfer function H(s) has n distinct poles p1 , p2 , . . . , pn , and consider the partial
fraction expansion of H(s):
k2
kn
k1
+
+ +
,
H(s) =
s + p1 s + p2
s + pn
where each of the ki s are some constants. By using the same technique as in the above example, the
all-integrator block diagram for this expansion is given by:

The state variables for this diagram evolve according to

x 1 = p1 x1 + k1 u
x 2 = p2 x2 + k2 u
..
.
x n = pn xn + kn u
y = x1 + x2 + + xn .
In matrix-vector form, this corresponds to:

p1
0
0

0
p
0
2

0
0 p3
x =

..
..
..

.
.
.
0
0
0


y = 1 1 1 1 x
4

..
.

0
0
0
..
.

pn

k1

k2

x + k3 u

..

.
kn

Note that in this realization, the matrix A is a diagonal matrix, with the poles on the diagonal. In the
state-space context, the poles of a linear system are also sometimes called the modes of the system, and
so this realization is called the modal form of the system. Note that the B matrix has the coefficients
of the partial fraction expansion as its entries, and the C matrix consists of all 1s (this is actually not
unique, as we can also come up with a modal realization where B has all ones, and C has the coefficients
of the expansion).
Note: The eigenvalues of a diagonal matrix (or, more generally, an upper or lower triangular matrix) are
simply the elements on the diagonal of the matrix. Thus, the modal form corroborates the fact that the
eigenvalues of A are the poles of the system.

Similarity Transformations

The above development shows that a given transfer function H(s) does not have a unique state-space
realization (there are at least two: the controllable canonical form and the modal canonical form). Exactly
how many realizations are there?
To answer this question, suppose we consider any particular realization
x = Ax + Bu
y = Cx
for the transfer function H(s) (so that H(s) = C(sI A)1 B). Now, let us choose an arbitrary invertible
n n matrix T, and define a new state vector
= Tx .
x
are simply linear combinations of the states in the vector x.
In other words, the states in the vector x
Since T is a constant matrix, we have
1
= Tx = TAx + TBu = TAT
+ |{z}
x
TB u
| {z } x

1
.
y = Cx = |CT
{z } x

Thus, after this transformation, we obtain the new state-space model

x + Bu

= A
x
x .
y = C
The transfer function corresponding to this model is given by

1 B
= CT1 (sI TAT1 )1 TB
H(s)
= C(sI
A)
= CT1 (sTT1 TAT1 )1 TB
= CT1 T(sI A)1 T1 TB
= C(sI A)1 B
= H(s) .
is the same as the transfer function
Thus the transfer function for the realization with state-vector x
= Tx is called a similarity
for the realization with state-vector x. For this reason, the transformation x
5

transformation. Since T can be any invertible matrix, and since there are an infinite number of invertible
n n matrices to choose from, we see that there are an infinite number of realizations for any given
transfer function H(s).
B,
C)
produce the same transfer function H(s), and since
Note: Since both of the sets (A, B, C) and (A,
we see that the following relationship holds:
the poles of H(s) are the eigenvalues of A and also A,

The eigenvalues of A are the same as the eigenvalues of TAT1 for any n n
invertible matrix T.

This is a well known result in matrix theory.

Example. Consider the state-space model


0
1
x =
2 3


y= 1 0 x .

 
0
x+
u
1



1 1
Perform a similarity transformation with matrix T =
, and verify that the new realization produces
0 1
the same transfer function as the above system.
Solution.