7 Structures and state-space realizations

The transfer function representation generalizes the frequency response characteriza-

tion of LTI systems’s input-output behavior and offers more insights into system char-
acteristics than the latter. As shown in Chapter 6, we can study the properties such as
stability and casuality of an LTI system with the distribution of its transfer function’s
poles easily. As to be seen in this chapter, such a representation also allows us to study
the relationship between the total system and its subsystems using linear algebra and
the computational structures for system implementation.
The models that we have considered so far are mathematical representations of
input-output behavior of a system under study. Another important topic in this chap-
ter is to introduce a new type of model that is specified in terms of a collection of inter-
mediate signals of the system. These signals are called the state variables and describe
the internal behavior of the system. The model defined in terms of the states variables
is called state-variable representation. This type of representation is a very powerful
tool in system modeling and signal processing. The objective here is to define the state
model and to study the basic properties of such a model. A more profound develop-
ment of this topic can be found in advanced textbooks.
The development of this chapter is organized as follows. In Section 7.1, we study
the block-diagram representation of LTI systems in terms of transfer functions. The
issue on how to implement a causal LTI system of a rational transfer function is dis-
cussed in Section 7.2, where some classical structures¹ for LTI systems are developed.
Section 7.3 is devoted to developing the concept of state-space representation, in
which how to set up state-space equations and some basic properties of state mod-
els are discussed. Section 7.4 deals with discretizing a continuous-time state model,
which allows us to study such a system via its discrete-time counterpart using digital
processing techniques. To end this chapter, a summary is given in Section 7.5.

7.1 Block-diagram representation

As mentioned before, a system, no matter it is LTI or not, usually consists of several

subsystems interconnected in some manner. The basic interconnections, as shown in
Figure 7.1, include cascade, parallel and feedback.
Consider the situation, where both subsystem Sk for k = 1, 2 are LTI with hk the
unit impulse response of Sk . Then we have the following.

1 In the sequel, a system structure means a way in which the output of a system is computed with an
input signal.
 7.1 Block-diagram representation | 227

x y1 y
- S1 - S2 -

(a)

x e y
- S1 - +m- S1 -
y1
−
6
x ? y
- +m-

y2 6
y2
- S2 S2 
(b) (c)

Fig. 7.1: Three basic interconnections for block-diagram representation of systems: (a) cascade;
(b) parallel; (c) feedback.

Cascade interconnection: Looking at Figure 7.1(a), we have y1 = h1 ∗ x, y = h2 ∗ y1 ,

equivalently in transform-domain, Y1 = H1 X, Y = H2 Y1 , that is

Y = H1 H2 X ≜ HX ↔ y = h ∗ x, (7.1)

where h is the inverse transform of H = H1 H2 . According to Theorems 2.1 and 2.2, the
overall system is LTI and its system transfer function is H = H1 H2 . Such a claim is
consistent with the associativity of convolution discussed in Chapter 2.

Parallel interconnection: It follows from Figure 7.1(b) that y = h1 ∗ x + h2 ∗ x

y = (h1 + h2 ) ∗ x ↔ Y = (H1 + H2 )X ≜ HX. (7.2)

With the same argument, we conclude that the overall system is LTI with a system
transfer function H = H1 + H2 , which is consistent with the distributivity of convolu-
tion discussed in Chapter 2.

Feedback interconnection: With e and y2 indicated in the block-diagram shown by

Figure 7.1(c), we have

y = h1 ∗ e, e = x − y2 , y2 = h2 ∗ y.

Equivalently, Y = H1 E, E = X − Y2 , Y2 = H2 Y and hence the transfer function of

the overall system H = Y/X is
H1
H= . (7.3)
1 + H1 H2
When the plant H1 is unstable, one can design a controller H2 such that the over-
all system (i.e. the closed-loop system) is stable. For example, h1 [n] = 2n u[n]. Let
228 | 7 Structures and state-space realizations

h2 [n] = α n u[n]. It is easy to show that

H1 (z) 1/2(1 − α z−1 )

H(z) = = .
1 + H1 (z)H2 (z) 1 − (1 + α /2)z−1 + α z−2

According to the stability triangle (6.31), the unstable plant h1 [n] can be stablized
if and only if the parameter α of the 1st order controller h2 [n] satisfies

|α | < 1, |1 + α /2| < 1 + α ⇒ 0 < α < 1.

More complicated block-diagrams can be formed with these basic blocks and an-
alyzed in a similar way. See the chapter tutorial questions.

7.2 Structures of LTIs with a rational transfer function

The block-diagram representation of an LTI system is based on the input/output re-

lation of the system and does not provide the details about how the system is con-
structed.
Look at the transfer functions specified by (6.33) and (6.35). Under the assumption
that M = N, they can be unified as

∑Nk=0 βk ρ −k
H(ρ ) = , (7.4)
1 + ∑Nk=1 αk ρ −k

with ρ −1 = s−1 , z−1 , representing the transfer function of the integrator and the (unit)
delay element, respectively, both can be implemented physically using electronic cir-
cuits.
Consider a well-designed causal LTI system having a transfer function H(ρ )
of form (7.4). It follows from H(ρ ) = Y(ρ )/X(ρ ) that [1 + ∑Nk=1 αk ρ −k ]Y(ρ ) =
∑Nk=0 βk ρ −k X(ρ ), i.e.
N N
Y(ρ ) = ∑ βk ρ −k X(ρ ) − ∑ αk ρ −k Y(ρ )m,
k=0 k=1

which corresponds to the system structure shown in Figure 7.2(a). Such a structure is
usually referred to as direct form-I (DFI) structure.
X(ρ )
Alternatively, define an intermediate variable: w ↔ W(ρ ) ≜ N −k
. Then
1+∑k=1 αk ρ
Y(ρ ) = ∑Nk=0 βk ρ −k W(ρ ) and hence
N
{
{
{
{ W(ρ ) = X(ρ ) − ∑ αk ρ −k W(ρ )
{ k=1
{
{ N
{
{ −k
{ Y(ρ ) = ∑ βk ρ W(ρ ).
{ k=0
 7.2 Structures of LTIs with a rational transfer function | 229

x β y x w β y
-0 - +l- +l - - - +l - -0 - +l -
? 6 6 ? 6 ? 6
−1 −1 −1
ρ ρ ρ
β1 −α1 −α1 w1 β
?- - +l +l ? +l
  ∙? -1 - +l
? 6 6 ? 6 ? 6
−1
ρ ρ −1 ρ −1

β2 −α2 −α2 w2 β
?- - +l +l ? 
+l  ∙? -2 - +l
? 6 6 ? 6 ? 6
−1
ρ ρ −1 ρ −1

β −α3 −α3 w3 β
?-3  ?  ∙? -3
(a) (b)
Fig. 7.2: Two classical structures for 3rd order LTI systems. (a) DFI structure; (b) DFII structure.

These two equations correspond to the structure specified by Figure 7.2(b), which is
obtained by implementing the first equation/system using the DFI structure (with
β0 = 1, βk = 0, ∀ k ≠ 0). Note that this DFI can provide the intermediate sig-
nals: wk ↔ ρ −k W(ρ ), k = 0, 1, ⋅ ⋅ ⋅ , N, we can obtain the output in a linear com-
bination of these signals easily. The structure by Figure 7.2(b) is traditionally referred
to as direct form-II (DFII) structure.
It should be noted that both structures by Figure 7.2(a) and Figure 7.2(b), though
very different, represent one and the same system that has a transfer function H(ρ ).
Comparing the two structures, we can see that the structure by the DFII structure re-
quires N less integrators/delay elements than the DFII. In that sense, the former is
more efficient as far as implementation cost is concerned.

Example 7.1: Specify the DFII structure for each of the following causal LTI systems:
d2 y(t) 2
– dt2
+ 2y(t) = 3 ddtx(t)
2 − dx(t)
dt
+ 5x(t);
– y[n] − 1.3y[n − 1] + 0.4y[n − 2] = −7x[n] + 10x[n − 1].

Solution: For the continuous-time system, we have N = 2 and

3s2 − s + 5 3 − s−1 + 5s−2
H(s) = = .
2
s +2 1 + 0s−1 + 2s−2
With α1 = 0, α2 = 2, β0 = 3, β1 = −1 and β2 = 5, the DFII structure can be specified.
See Figure 7.3(a).
The 2nd system is a discrete-time system of order N = 2. The corresponding trans-
fer function is
−7 + 10z−1
H(z) = .
1 − 1.3z−1 + 0.4z−2
With α1 = −1.3, α2 = 0.4, β0 = −7, β1 = 10 and β2 = 0, the corresponding DFII
structure can be specified as shown in Figure 7.3(b).
230 | 7 Structures and state-space realizations

x(t) 3 x[n] −7
- +l - - - +l - - +l - - - +l -
6 ? 6y(t) 6 ? 6y[n]
s−1 z−1
−1 1.3 10
? - - +l +l  ∙? -
? 6 6 ?
s−1 z−1
−2 5 −0.4
 ? -  ?
(a) (b)
Fig. 7.3: DFII structures for Example 7.1: (a) for H(s); (b) H(z).

For a given transfer function H(ρ ) of an LTI system, there are other structures than the
DFI and DFII that can be used to implement the same system. This is shown by the
following example.

Example 7.2: Figure 7.4 is a 2nd order causal discrete-time system implemented with a
lattice structure. Determine the transfer function of this system and the constraints on
the structure parameters Km , m = 1, 2, ψl , l = 0, 1, 2 that ensure the system stability.

x[n] f [n] f [n]

- +m 2 - +m 1 -
I K2
− @ I K1
− @
@ @
@ K @ K

 @ I
 @ I
+m @2 z−1
 +m @1 z−1

w2 [n] w1 [n]
ψ2 ? ψ1? ?
ψ0
? ? y[n]
- +m - +m -

Fig. 7.4: A lattice structure for 2nd order discrete-time system.

Solution: In order to determine the transfer function of a given system structure, it is

helpful to define a set of intermediate variables. As a general rule, we take the output
of each delay element as such a variable. See w1 [n] and w2 [n] in Figure 7.4.
As observed, the other intermediate signals can be represented in terms of these
variables and the input x[n]. For example, the signals f1 [n], f2 [n] in Figure 7.4 are given
by
f2 [n] = x[n] − K2 w2 [n], f1 [n] = f2 [n] − K1 w1 [n].
 7.2 Structures of LTIs with a rational transfer function | 231

Then the next step is to derive the expression for wk [n + 1], k = 1, 2 in terms of
wk [n], k = 1, 2 and x[n]. Clearly,

w1 [n + 1] = f1 [n]
= −K1 w1 [n] − K2 w2 [n] + x[n],
w2 [n + 1] = w1 [n] + K1 f1 [n]
= (1 − K21 )w1 [n] − K1 K2 w2 [n] + K1 x[n].

Note that y[n] = ψ0 w1 [n + 1] + ψ1 w2 [n + 1] + ψ2 w2 [n] + K2 f2 [n] and hence

y[n] = 𝜈1 w1 [n] + 𝜈2 w2 [n] + 𝜈0 x[n],

where
{
{ 𝜈0 = ψ0 + ψ1 K1 + K2
{
{ 𝜈1 = ψ1 (1 − K21 ) − ψ0 K1
{
{
{ 𝜈2 = ψ2 − K2 (K2 + ψ0 + ψ1 K1 ).
In summary, this structure of the system can be described completely with

{
{ w1 [n + 1] = −K1 w1 [n] − K2 w2 [n] + x[n]
{
{ w2 [n + 1] = (1 − K21 )w1 [n] − K1 K2 w2 [n] + K1 x[n] (7.5)
{
{
{ y[n] = 𝜈1 w1 [n] + 𝜈2 w2 [n] + 𝜈0 x[n].

Applying z-transform to the above equations yields

{
{ zW1 (z) = −K1 W1 (z) − K2 W2 (z) + X(z)
{
{ zW2 (z) = (1 − K21 )W1 (z) − K1 K2 W2 (z) + K1 X(z) (7.6)
{
{
{ Y(z) = 𝜈1 W1 (z) + 𝜈2 W2 (z) + 𝜈0 X(z).

The two unknowns W1 (z), W2 (z) can be solved from the first two linear equations:
z
W1 (z) = X(z)
z2 + K1 (1 + K2 )z + K2
K1 z + 1
W2 (z) = X(z),
z2 + K1 (1 + K2 )z + K2
Y(z)
and hence with H(z) = X(z)
, we finally have the transfer function

𝜈0 z2 + [𝜈0 K1 (1 + K2 ) + 𝜈1 + 𝜈2 K1 ]z + (𝜈2 + 𝜈0 K2 )
H(z) =
z2 + K1 (1 + K2 )z + K2
β0 + β1 z−1 + β2 z−2
≜ . (7.7)
1 + α1 z−1 + α2 z−2
232 | 7 Structures and state-space realizations

As a causal LTI system, it is stable if and only if its poles are all inside the unit cir-
cle. Note α1 = K1 (1 + K2 ), α2 = K2 . The conditions specified by (6.31) are equivalent
to
|Km | < 1, m = 1, 2. (7.8)
which is the necessary and sufficient conditions for this system to be stable.

For a 2nd order system with H(z) given by (7.7), we can implement it using either
the DFII structure by Figure 7.2(b) or the lattice structure by Figure 7.4 if the implemen-
tation is carried out with infinite precision. In practice, the parameters in a structure,
say K1 , K2 , ψ0 , ψ1 and ψ2 in the lattice structure, are in the decimal format. For exam-
ple,
1
K1 = = 0.333 ⋅ ⋅ ⋅ 3 ⋅ ⋅ ⋅
3
= 0 × 2−1 + 1 × 2−2 + ⋅ ⋅ ⋅ + 0 × 2−2k+1 + 1 × 2−2k + ⋅ ⋅ ⋅
= (0101010101 ⋅ ⋅ ⋅ 01 ⋅ ⋅ ⋅ )2 .

Such a number is of infinite number of digits in both decimal and binary formats.
When a digital device (say a micro-processor) of 8 bits is used for implementation, the
actually implemented parameter is
4
∗ −2k+1
K1 = (01010101)2 = ∑ [0 × 2 + 1 × 2−2k ]
k=1

= 2−2 + 2−4 + 2−6 + 2−8 = 0.332 031 25,

which is different from the desired one. The difference is referred to as finite word
length (FWL) error:

Δ K1 = K1 − K∗1 = 0.001 302 083 333 33.

Such an order of small errors may cause a dramatic degradation in performance

to the system if the structure used is not properly chosen.
Lattice structures are important in many applications, including digital filter syn-
thesis, owing to their excellent finite wordlength (FWL) properties. The following ex-
ample demonstrates the superior performance of the lattice structure in reducing the
FWL effects over that of the DFII structure.

Example 7.3: Consider an 8th order narrow-band band-pass filter generated with
MATLAB command [V β , V α ] = ellip(4, 0.25, 40, [0.4 0.5]). The infinite precision mag-
nitude response is shown in Figure 7.5 with solid line.
For the DFII structure, the corresponding parameters are given in V α and V β :
T
Vα = [ 1 α1 α2 α3 α4 α5 α6 α7 α8 ]
T
V β = [ β0 β1 β2 β3 β4 β5 β6 β7 β8 ] ,
 7.2 Structures of LTIs with a rational transfer function | 233

10

−10

−20
20log|H(ejΩ)|

−30

−40

−50

−60

−70

−80

−90
0 1π 2π 3π 4π π
5 5 5 5
Ω

Fig. 7.5: Magnitude responses for Example 7.3: solid line for the desired response; dot-dashed line
for the DFII structure with Bc = 9 bits; dashed line for the lattice structure with Bc = 9.

and the parameters of the lattice structure can then be computed using the relation-
ship between the parameters of the two structures.
For every parameter of each set above, one bit is used for the sign, the number of
bits for the integer and fractional parts is denoted with BI and BF . So, the total number
of bits for implementing a parameter is Bc = 1 + BI + BF . With the same Bc = 8 bits, the
DFII structure requires BI = 1, BF = 6, while the lattice structure needs BI = 0, BF = 7.
The magnitude response of the filter implemented with truncated parameters for
the DFII structure as well as for the lattice structure is calculated and plotted in Fig-
ure 7.5 with dotted line and dot-dashed line, respectively. Clearly, the lattice structure
is better than the DFII structure in terms of reducing the FWL effect on the frequency
response.

In-depth discussions on the topic of LTI system structures can be found from the
textbooks [5, 6]. The systems we have studied so far are assumed to be time invariant.
There are many occasions where the parameters of systems are designed to be time-
varying in order to adapt changes of signal characteristics and environments. These
are more advanced topics [7, 8].
234 | 7 Structures and state-space realizations

7.3 State-space variable representation

The transfer function representation of LTI systems focuses on the input-output rela-
tion and hence it is by nature a black box representation. In this section, we introduce
an alternative approach, called state-space realization or internal description. In this
approach, it involves not only the input-output relationship but also an additional set
of internal signals, called state variables.
It should be noted that the discussions in this section are for discrete-time signals
and systems. The related topics for the continuous-time counterparts will be given in
Section 7.4.

7.3.1 State model and state-space realizations

Reconsider the system in Example 7.2. As indicated in (7.5), such a system can be de-
scribed with two intermediate variables w1 [n] and w2 [n] in such a way that for any
given n0 , the system output y[n] for n > n0 is completely determined by w1 [n0 ], w2 [n0 ]
and the input signal x[n] for n > n0 . The signals w1 [n] and w2 [n] are usually referred
to as states of the system.
A generalized (7.5) is

w[n + 1] = f (w[n], x[n], n)

{ (7.9)
y[n] = g(w[n], x[n], n),

where w[n] is the the state vector

T
w[n] ≜ [ w1 [n] w2 [n] ⋅ ⋅ ⋅ wN [n] ] , (7.10)

with T denoting the transpose operator. Usually, the 1st equation is called state updat-
ing equation and the 2nd one, output equation.
The state vector w[n0 ] actually represents the total effect of the past input signal
(i.e. x[n], ∀ n ⩽ n0 ) on the future output of the system (i.e. y[n], ∀ n > n0 ).
The functions f (w, x, n) and g(w, x, n) may be nonlinear in w and x, in which case
the system is nonlinear. It can be shown that the system is linear if and only if

w[n + 1] = A[n]w[n] + B[n]x[n]

{ (7.11)
y[n] = C[n]w[n] + d[n]x[n],

where the quadruple (A, B, C, d) is called a state-space realization of the system, with

A[n] ∈ RN×N , B[n] ∈ RN×1 , C[n] ∈ R1×N , d[n] ∈ R (7.12)

having nothing to do with the input signal. The integer N, indicating the number of
state variables, is the dimension of the system.
 7.3 State-space variable representation | 235

If (A, B, C, d) are all constant, the state-space equations

w[n + 1] = Aw[n] + Bx[n]

{ (7.13)
y[n] = Cw[n] + dx[n]

represents an LTI system.

T
Look at (7.5). Denote w[n] ≜ [ w1 [n] w2 [n] ] . It can be rewritten into

{ a11 a12 1
{
{ w[n + 1] = [ ] w[n] + [ ] x[n]
a21 a22 K1
{
{
{
{ y[n] = [ 𝜈1 𝜈2 ] w[n] + 𝜈0 x[n],

with a11 = −K1 , a12 = −K2 , a21 = 1 − K21 , a22 = −K1 K2 . This is a state-space
realization of the 2nd order LTI system (7.7) that can be implemented with the structure
depicted in Figure 7.6.
a
11

1 ? 𝜈
-- +m - z−1 w1 [n]-
∙ -1
a12
x[n]
-∙ 6  -
? y[n]
- +m-
 𝜈0
? a21 6
-- +m - z−1 -∙ -
K1 w2 [n] 𝜈2
6 a
22

Fig. 7.6: The equivalent state-space realization of the lattice structure shown in Figure 7.4.

As can be seen, this state-space realization was derived from the lattice structure
shown in Figure 7.4 and hence is called the equivalent state-space realization of the
latter. Though different, both represent the one and the same 2nd order system of
transfer function H(z) by (7.7).

7.3.2 Construction of an equivalent state-space realization

As shown in Example 7.3, the procedure to find a state-space realization for an (causal)
LTI system represented by a structure contains the following steps:
– to define the state variables {wk [n]} as the output signals of all delay elements z−1 ;
– to derive the expression for each wm [n + 1]—the input of the mth delay element,
as a linear combination of the state variables {wk [n]} and input x[n]. This yields
the 1st equation of (7.13);
– to express the output y[n] in terms of the state vector w[n] and the input signal
x[n]. That leads to the 2nd equation of (7.13).
236 | 7 Structures and state-space realizations

Consider the class of 3rd order systems described with the following transfer function

β0 + β1 z−1 + β2 z−2 + β3 z−3

H(z) =
1 + α1 z−1 + α2 z−2 + α3 z−3
𝛾1 z−1 + 𝛾2 z−2 + 𝛾3 z−3
= β0 + ,
1 + α1 z−1 + α2 z−2 + α3 z−3
where 𝛾k ≜ βk − β0 αk , k = 1, 2, 3.
Now, let us find a state-space realization for such a system. First of all, we note
that the direct-form-II structure of this system is given in Figure 7.2(b) with ρ = z−1 ,
where wk = wk [n], k = 1, 2, 3 are the state variables. Clearly,

w1 [n + 1] = −α1 w1 [n] − α2 w2 [n] − α3 w3 [n] + x[n]

w2 [n + 1] = w1 [n]
w3 [n + 1] = w2 [n].

We note that the output y[n] = β0 w1 [n + 1] + β1 w1 [n] + β2 w2 [n] + β3 w3 [n]. With

w1 [n + 1] replaced by the expression obtained above, y[n] can be written as

y[n] = 𝛾1 w1 [n] + 𝛾2 w2 [n] + 𝛾3 w3 [n] + β0 x[n]

T
Denote w[n] ≜ [ w1 [n] w2 [n] w3 [n] ] . The four equations obtained above can be
arranged in the form (7.13), where the corresponding realization is given by

{ −α1 −α2 −α3 1

{
{ [ ] [ ]
{
{A = [ 1 0 0 ], B = [0]
{ .
{
{ [ 0 1 0 ] [0]
{
{
{ C = [ 𝛾1 𝛾2 𝛾3 ] , d = β0

It is easy to understand that for the N-th-order system

∑Nk=1 𝛾k z−k
H(z) = d +
1 + ∑Nk=1 αk z−k

we have a state-space model of form (7.13) with the following realization

{ −α1 −α2 −α3 ⋅ ⋅ ⋅ −αN−1 −αN 1

{
{ [ ] [ ]
{
{ [ 1 0 0 ⋅ ⋅ ⋅ 0 0 ] [ 0]
{
{ [ ] [ ]
{
{ [ 0 1 0 0 ⋅⋅⋅ 0 ] , B = [ .. ]
] [
{
{A = [ ]
[ . . . . . . ] [.] .
{
{ [ . . . . . . ] [ ] (7.14)
{
{ [ . . . . . . ] [0]
{
{
{
{ [ 0 0 0 ⋅⋅⋅ 1 0 ] [0]
{
{
{
{ C = [ 𝛾1 𝛾2 𝛾3 ⋅ ⋅ ⋅ 𝛾N−1 𝛾N ] , d = β0

This state-space realization is called the controllable realization of the system

H(z).
 7.3 State-space variable representation | 237

7.3.3 Similarity transformations

Now, let us consider the relationship between a state-space realization (A, B, C, d) with
the transfer function H(z).
Denote X(z), Y(z) and W(z) as the z-transform of x[n], y[n] and w[n], respectively.
It follows from (7.13) that

zW(z) = AW(z) + BX(z)

{
Y(z) = CW(z) + dX(z).

With the first equation above, we have W(z) = (zI − A)−1 BX(z), where I is the
identity matrix of dimension N. Therefore,

Y(z) = C(Iz − A)−1 BX(z) + dX(z),

and hence H(z) = Y(z)/X(z) is given by

H(z) = d + C(zI − A)−1 B. (7.15)

Once a state-space realization of an LTI system is given, its transfer function can
be obtained directly with (7.15).
Both the transfer function and the state model can represent an LTI system com-
pletely. It is interesting to note that for a given system the transfer function is unique,
while the state-space realizations are not. In fact, denote w[n] ̃ = Tw[n], where T ∈
N×N
R is any nonsingular matrix. The state-space equations (7.13) become

̃ + 1] = Ã w[n]
w[n ̃ ̃
+ Bx[n]
{
y[n] = C̃ w[n]
̃ + dx[n].

where (A,̃ B,̃ C,̃ d) given by

à = T −1 AT, B̃ = T −1 B, C̃ = CT, (7.16)

̃
satisfying H(z) = d + C(zI − A)̃ −1 B.̃ This means that the quadruple (A,̃ B,̃ C,̃ d) is also a
state-space realization of the same system. Such a matrix T is referred to as a similarity
transformation.

Example 7.4: Find the controllable realization for the following digital filter

(1 + z−1 )2
H(z) = ,
1 − z−1 + 14 z−2

and hence the realization that is obtained from the controllable realization with the
similarity transformation
1 2
T=[ ].
0 1
238 | 7 Structures and state-space realizations

Solution: First of all,

1 + 2z−1 + z−2
H(z) =
1 − z−1 + 14 z−2
(1 − z−1 + 14 z−2 ) + [(2 + 1)z−1 + (1 − 14 )z−2 ]
=
1 − z−1 + 14 z−2
3z−1 + 34 z−2
=1+ .
1 − z−1 + 14 z−2

Therefore, the controllable realization is

1 − 14 1
Ac = [ ] , Bc = [ ]
1 0 0
Cc = [ 3 3
4
], d = 1.

1 −2
Noting T −1 = [ ], one has
0 1

−1 − 94 −1 − 17
A = T −1 Ac T = [ ]T = [ 4 ]
1 0 1 2

1
B = T −1 Bc = [ ], C = TCc = [ 3 27.
4
]
0
So, one can implement this second-order filter using either the controllable real-
ization (Ac , Bc , Cc , 1) or the realization (A, B, C, 1) obtained above.

Before turning into the next section, it is interesting to note that the 1st equation
of (7.13) suggests

w[n] = Aw[n − 1] + Bx[n − 1] = A[Aw[n − 2] + Bx[n − 2]] + Bx[n − 1]

= A2 w[n − 2] + ABx[n − 2] + Bx[n − 1] = ⋅ ⋅ ⋅
n
= An−n0 w[n0 ] + ∑ An−k Bx[k − 1], (7.17)
k=n0 +1

and hence y[n] = Cx[n] + dx[n] is equal to

n
y[n] = CAn−n0 w[n0 ] + ∑ CAn−k Bx[k − 1] + dx[n]
k=n0 +1

≜ yzi [n] + yzs [n] (7.18)

for any given integer n0 . This result implies that if we consider the output y[n] just for
n ⩾ n0 , then yzi [n] is the zero-input response of the system while yzs [n] is the zero-state
response.
 7.4 Discretizing a continuous-time state model | 239

Furthermore, the unit impulse response h[n] of the (causal) system with a stat-
space realization (A, B, C, d) is

{ 0, ∀n<0
{
h[n] = { d, n=0 (7.19)
{ n−1
{ CA B, ∀ n > 0.

7.4 Discretizing a continuous-time state model

Applying the same procedure above, we can obtain the state model for the N-th-order
causal LTI system of transfer function

∑Nk=1 𝛾k s−k
H(s) = d + .
1 + ∑Nk=1 αk s−k

In fact, this model is of exactly the same form as (7.13):

{ dw(t)
{ = Aw(t) + Bx(t)
{ dt (7.20)
{
{ y(t) = Cw(t) + dx(t),
where (A, B, C, d) is a state-space realization and satisfies

H(s) = d + C(sI − A)−1 B. (7.21)

As in the discrete-time case, the realizations of H(s) are not unique and one of
such realizations is the controllable realization given by (7.14), which can be obtained
directly from the transfer function H(s). In addition, if (A, B, C, d) is a state-space real-
ization of H(s), so is (A,̃ B,̃ C,̃ d) that is obtained with (7.16) for any nonsingular matrix
T of dimension N.

Example 7.5: Let us consider the RLC series circuit shown in Figure 1.32. Derive a state
model for this system.

Solution : As known before, the system can be represented with the following LCCDE

d2 y(t) R dy(t) 1 1
+ + y(t) = x(t),
dt2 L dt LC LC
where y(t) is the voltage across the capacitor and x(t) is the applied voltage excitation.
Clearly, the transfer function of this circuit is

1/LC (1/LC)s−2
H(s) = = ,
s2 + (R/L)s + (1/LC) 1 + (R/L)s−1 + (1/LC)s−2
240 | 7 Structures and state-space realizations

and its controllable realization is

−R/L −1/LC 1
Ac = [ ] , Bc = [ ]
1 0 0
Cc = [ 0 1/LC ] , d = 0.

The physical meaning of the states of this state-space model is not clear.
Now, if we choose the current i(t) and y(t) as the states:

w1 (t) = i(t), w2 (t) = y(t).

Then
dw1 (t)
L = x(t) − Rw1 (t) − w2 (t)
dt
dw2 (t)
C = w1 (t).
dt
The corresponding state-space realization is obviously given by

− R − 1L 1
à = [ 1L ], B̃ = [ L ], C̃ = [ 0 1 ] , d = 0.
C
0 0

The state variables defined in such a way have a clear physical meaning. What is
the similarity transformation matrix T between the two realizations?

Applying the Laplace transform to the 1st equation (i.e. the forced equation) of
(7.20) leads to sW(s) = AW(s) + BX(s) and hence

W(s) = (sI − A)−1 BX(s).

Therefore, w(t) can be obtained with the inverse Laplace transform.

Alternatively, define a matrix exponential function in t:

M 2 t2 M k tk
eMt ≜ I + Mt + + ⋅⋅⋅ + + ⋅⋅⋅ , (7.22)
2! k!

where M ∈ RN×N is a constant matrix. It can be shown that

deMt
= MeMt , eM(t+τ ) = eMt eMτ ,
dt

which implies that eMt can be treated as if it were scalar.

It turns out that the 1st equation of (7.20) can be re-arranged into

dw(t) d −At
e−At [ − Aw(t)] = e−At Bx(t) ⇔ [e w(t)] = e−At Bx(t).
dt dt
 7.4 Discretizing a continuous-time state model | 241

t
The latter simply tells e−At w(t) − e−At0 w(t0 ) = ∫t e−Aτ Bx(τ )dτ , that is,
0

w(t) = e A(t−t0 )
w(t0 ) + ∫ eA(t−τ ) Bx(τ )dτ (7.23)
t0

for any t0 given. In addition, the output of the system is then given by

t
A(t−t0 )
y(t) = Ce w(t0 ) + ∫ CeA(t−τ ) Bx(τ )dτ + dx(t). (7.24)
t0

Like the discrete-time case, the first term of the above equation is the zero-input
response, while the 2nd one is the zero-state response of the system.
Note the system is causal. When x(t) = δ (t), w(0− ) = 0 and hence it follows from
(7.24) that
t

y(t) = ∫ CeA(t−τ ) Bδ (τ )dτ + dδ (t) = h(t).

0−

Therefore,
h(t) = CeAt Bu(t) + dδ (t). (7.25)
The continuous-time state-space model (7.20) can be studied via a discrete-time
counterpart. In fact, let t = (n + 1)T, t0 = nT and assume that x(t) is piece-wise con-
stant such that x(t) = x(nT), nT ⩽ t < (n + 1)T, ∀ n. It then follows from (7.23) and
(7.24) that

w[n + 1] = As w[n] + Bs x[n]

{ (7.26)
y[n] = Cs w[n] + ds x[n],

where w[n] ≜ w(nT), x[n] ≜ x(t), y[n] ≜ y(t) and

(n+1)T T

Bs ≜ ∫ eA(T+nT−τ ) Bdτ = ∫ eAξ Bdξ

nT 0
(7.27)
AT
As ≜ e , Cs ≜ C, ds ≜ d.

This discrete-time system is a consequence of the time discretization of the

continuous-time system H(s). Since the former can be evaluated recursively, the dis-
cretization yields actually a numerical method for solving the state equations of an
LTI continuous-time system.
It should be noted that w[n], y[n] computed with (7.26) are equal to the w(t), y(t)
of (7.20) for t = nT, respectively, if the input signal x(t) is piece-wise constant. If this
condition is not satisfied, w[n], y[n] are just approximate values for w(nT), y(nT). Theo-
retically, the accuracy of such an approximation will be improved as T is made smaller.
242 | 7 Structures and state-space realizations

7.5 Summary

In this chapter, we have examined three different descriptions of LTI systems: the
block-diagram, the structures, and the stat-space realizations. These descriptions are
totally equivalent in the sense that they represent one and the same transfer function.
However, different descriptions offer different insights into system features.
The block-diagram representation provides the interconnection of the (LTI) sub-
systems. Such an interconnection is formed of the basic connections such as cascade,
parallel and feedback. The transform-domain approach allows us to study the rela-
tionship between the overall system and the sub-systems using simple algebra oper-
ations. Application areas of such representation include system controls, where regu-
lations and feedbacks play an important role, and signal processing. All these will be
dealt with in the later chapters.
For a given transfer function of a system (or a sub-system of the entire system),
there exists an infinite number of structures that can be used for implementation on a
digital device (say a digital signal processor) or simulation on a computer. It is worthy
to note that different structures, though equivalent in infinite precision case, have dif-
ferent numerical properties. Such a feature implies that for a given application, differ-
ent structures can yield quite different performance. For the topic on optimum system
structures, we refer to [5], [6].
The state-space description is another very powerful tool for system modeling and
signal analysis. One of the remarkable features of the state model is that for a given
linear system there exists an infinite number of state-space realizations and that any
two such realizations are related with a similarity transformation. By nature, the state-
space realizations belong to the scope of system structures. As this particular subset
of system structures can be characterized analytically, the optimal structure problem
is more tractable over this subset. One successful application of the system structure
theory is the so-called FWL state-space realization design of digital filters. See [5], [6].
The fundamentals of signals and systems that are intended to offer in this text-
book have been provided by now. For in-depth studies and applications, please refer
to more specified courses.

7.6 Problems

Problem 7.1: A system is shown in Figure 7.1, where H1 and H2 are the system transfer
functions of two LTI sub-systems, respectively. Show that the system x → y is LTI and
determine the transfer function.

Problem 7.2: Find the expression for the transfer function of LTI system given in the
black-diagram shown in Figure 7.8.
 7.6 Problems | 243

- +m - H1
y1
6 ?
x ?
- +m- y

? 6y2
- +m - 6
H2

Fig. 7.7: Block-diagram for the system in Problem 7.1.

x y
- - H1 - +m- H2 -
6

- +m H3

Fig. 7.8: Block-diagram for the system in Problem 7.2.

Problem 7.3: The block-diagram of a system is drawn in Figure 7.9, where H1 (s) =
s
s2 +4s+4
and H2 (s) = κ are the transfer functions of the two casual LTI subsystems,
respectively.
(a) Determine the transfer function of the overall system.
(b) Find the range for the paramter κ that ensures the stablity of the entire system.

x y
- +m- H1 - H2 -
6

Fig. 7.9: Block-diagram for the system in Problem 7.3.

Problem 7.4: The block-diagram of a system is drawn in Figure 7.10, where

s+1 10
H1 (s) = , H2 (s) = , H3 (s) = κ
s s2 + s
are the transfer functions of the three casual LTI subsystems, respectively.
244 | 7 Structures and state-space realizations

H3 

x ? y
- +m- H1 - +m - H2 -∙ -
− 6

Fig. 7.10: Block-diagram for the system in Problem 7.4.

(a) Determine the transfer function of the overall system.

(b) Find the range for the paramter κ that ensures the stablity of the entire system.

Problem 7.5: Consider the feedback system depicted in Figure 7.1(c), where the trans-
fer functions of S1 and S2 are given by H1 (s) = κ1 and H2 (s) = −κ2 e−sτ with the constant
τ > 0, respectively.
(a) Determine the transfer function of the overall system H(s) = Y(s)/X(s).
(b) What are the contraints on the parameters κ1 and κ2 in order to ensure H(s) stable?
Sketch the corresponding diagram of pole distribution.
(c) Assume that H(s) is stable. Determine the unit impulse response of H(s).

Problem 7.6: Draw the direct-form I and direct-form ćò structures for each of the causal
LTI systems given:
d
(a) dt
y(t) + 2y(t) = 2x(t);
d2 d d
(b) dt2
y(t) + 5 dt y(t) + 4y(t) = dt
x(t) + 2x(t);
(c) y[n] + 0.5y[n − 1] = 4x[n] − x[n − 1];
(d) 2y[n] + y[n − 1] − 2y[n − 3] = x[n] + 6x[n − 1] + x[n − 2].

Problem 7.7: Figure 7.11 depicts two structures, each of which represents a causal LTI
system.
(a) Determine the equivalent state-space equations of each structure.
(b) Find the expression for the transfer function of each system and hence the system
equation relating x and y.
(c) Check the stability of each system.

Problem 7.8: Given a causal LTI system described with the following state-space equa-
tions:

d −1 0 1
w(t) = [ ] w(t) + [ ] x(t)
dt 1 −3 0
y(t) = [ −1 2 ] w(t) + 2x(t).
 7.6 Problems | 245

x(t)
2 ? −1 ? y(t)
? ?
+m - s−1 - +m - s−1 ∙- +m -
I
@
6 −2
@ 
4 6

(a)

2 ? −1 ?
x[n] ? ? y[n]
∙ - +m - z−1 - +m - z−1 ∙- +m -
I
@
6 6
@ 
1/6 6
−1/4
∙
(b)
Fig. 7.11: Structures of the systems for Problem 7.7.

Determine the output of this system when it is excited with input x(t) = u(t) and the
T
initial condition x(0) = [ 1 2 ] .

Problem 7.9: Given a causal LTI system described with the following state-space equa-
tions:

0 1 0
w[n + 1] = [ ] w[n] + [ ] x[n]
−6 5 1
y[n] = [ 1 1 ] w[n] + x[n].

Determine the output of this system when it is excited with input x[n] = u[n] and
T
the initial condition x[n] = [ 1 2 ] .

Problem 7.10: Figure 7.12 is a resistor-capacitor-inductor circuit. Assume R = 1, L = 1,

C = 1. Find a state-space equation for this analog circuit and then discretize it using
T = 1.
246 | 7 Structures and state-space realizations

R L

i(t)
+ +

x(t) ∼ C y(t)
− −

Fig. 7.12: An RLC analog circuit for Problem 7.10.