GENERAL I ARTICLE

Discrete-Time Systems
1. Why do We Celebrate Birthdays Once a Year?

A Ramakalyan, P Kavitha and S Harini Vijayalakshmi

In this article we introduce digital control and con-

trast it with analog control. Traditionally, control
systems have been designed using analog techniques.
But today, because of the explosive growth of digi-
tal technology, controllers are typically implemented
as programmable digital hardware or as programs
on digital computers. The concepts and techniques
for analysis and design of digital controllers are the Ramakalyan is currently
central concerns of this article. In this part we in- with the Department of
troduce discrete-time signals and provide examples. Instrumentation and
Control Engineering at
We also describe discrete-time systems in terms of
REC Tiruchi. Kavitha
difference equations and look at their solutions via and Harini were final year
transform methods. students in the same
Department at the time
Introduction
of writing this article.
In the article on Systems and Control Engineering[l], we
have seen a Proportional plus Derivative (P D) Controller.
The actual physical implementation of that need not concern
us, but some remarks are in order. Assuming that error and
its derivative can be measured, it is necessary to take a linear
combination of these in order to determine the actuating
signal u(t). Such combinations are readily carried out by
Kavitha is now with
circuits built out of devices called Operational Amplifiers. Hindustan Computers
Refer [2] for more details on operational amplifiers. Ltd. at Chennai.

A more modern alternative, especially for larger systems,

is to convert position and velocity to digital form and to
use a computer to calculate the necessary controls. For
use by a digital computer, continuous-time signals known
as analog signals need to be sampled and then converted
to discrete-time signals by an operation called analog-to-
digital (AID) conversion. After processing, the discrete-time Harini is now at Texas
A & M University doing
signals should be converted back to continuous-time by a
MS with DSP as major.
digital-to-analog (D I A) converter. Such a configuration is
called a Sampled-data System. Figure 1 shows a standard

-Fe-b-r-U-Qr-Y--2-0-00------------~-------------------------------3--9
-R-ES-O-N-A--N-C-E--I
 GENERAL I ARTICLE

+
.Q1ND H H Digital
Processor
D/A
H Plant
I
I
Figure 1.
sampled data feedback control system.

Although the expression digital control system refers to a

control system with some part in digital form (and that gen-
erally means a sampled -data digital control system), occa-
sionally w.e may have a plant completely described by digital
mathematics, and when this plant is controlled by a digital
compensator, the system is entirely digital.

Discrete-Time Signals

A signal j (t) is said to be a continuous-time signal or ana-

log signal when it is defined at every instant of time i.e.,
\;j t E ~.

On the other hand, if the signal is defined only at discrete

instants of time and not elsewhere i.e., t takes on only the
discrete values t = kT for some range of integer values of
k, the signal j (kT) is said to be a discrete-time signal. Dis-
crete time signals arise in many areas of science, engineering,
business and economics.

One of the most common ways in which discrete-time sig-

nals can be obtained is in sampling continuous time signals.
As illustrated in Figure 2, suppose that j(t), (t E ~)
is applied to an electronic switch that is closed for a mo-
ment every T seconds; the output of the switch can then
be viewed as a discrete-time signal j(kT). The resulting
discrete-time signal is called the sampled version of the orig-
inal continuous-time signal j(t). The instants at which data
appear in a discrete-time signal are called the sampling in-

f(t) i---x------;
! /
f(kT)
T~ _-L-~~_ _ _ __
~
Figure 2.

-40-------------------------------~--------------------------------
RESONANCE I February 2000
 GENERAL I ARTICLE

stants; the time interval T between two subsequent sam-

pling instants is called the sampling interval. The amount
of time during which the switch is closed is (assumed to be)
much smaller than T. We emphasize here that the sampling
process described is called uniform sampling since T is con-
stant. Non-uniform or multi-rate sampling is also used in
some applications but is beyond the scope of this article.

The sampling theorem suggests the sampling period required

to reconstruct the original continuous-time signal from its
sampled version.

Shannon's Sampling Theorem

Let f (t) be a continuous-time signal whose frequency spec-

trum is band limited to w (in radians per second). Then,
f(t) can be reconstructed from its sampled sequence f(kT)
if the sampling period T is smaller than ~. In other words,
the sampling rate must be at least twice the bandwidth or
the highest frequency component in the system.

Smaller sampling rates are used in communication systems

and larger sampling rates are used in control systems. Typ-
ically, audio signals are sampled at 44.1 kHz.

In a discrete-time signal, the amplitude can assume any

value in a continuous range. A digital signal is a discrete-
time signal with quantized amplitude i.e., the amplitude can
assume values only from a finite set. Thus, a digital signal
is discretized in time and quantized in amplitude, whereas
a discrete-time signal is discretized in time but need not be
quantized in amplitude. Discrete-time is frequently used in
theoretical study, while digital is used in connection with
hardware or software realizations. In practical usage, the
terms discrete-time and digital are often interchanged. In
the present article, we refer to discrete-time. Refer to Fig-
ure 3 for different signals.

A very common example of a discrete-time signal is our age.

We do grow older every moment but we celebrate our birth-
days only once in a year (and increment our age by one),
i.e., the sampling period is one year! The plots of age(t) and
age(kT) are shown in Figure 4. In this example, we intend

-R-ES-O-N--A-N-C-E--I-F-eb-r-U-a-rY--2-0-0-0-----------~-------------------------------4-1
 GENERAL I ARTICLE

CT Sampled
Data Digital

Figure 3. to emphasize the notion of· biological growth that generally

prevails in individuals. An answer to the question "how old
are you?" is always a positive integer!
Other examples of discrete-time signals are daily tempera-
ture (T = 1 hour), banking transactions (T = ~ year), census
(T = 5 years) etc. Whatever may be the sampling interval,
it is apparent that we are much more comfortable dealing
with discrete-time signals. One reason might be, perhaps,
that there is always an inherent delay in processing the sig-
nals.
Notice that f(kT) simply becomes a sequence of numbers.
The standard convention (which is followed here) is not to
show the dependence on T in the notation for the sampled
signal. Equivalently, we may assume that the sampling pe-
riod is normalised to unity.

Discrete-Time Systems
A discrete-time system is defined as a system whose inputs
and outputs are discrete-time signals. In Part 2 of [1], we
have discussed several properties of systems and the convo-
lution integral description of continuous-time systems.

Figure 4.

-4-2------------------------------~~----------R-E-SO--N-A-N-CE---I-F-eb-r-u-ar-Y-2-0-0-0
 GENERAL I ARTICLE

All those concepts transfer directly from the analog world to

discrete-time systems with a little modification as follows.

As depicted in Figure 2, the AID conversion itself may be

viewed as a system so that f (k) is the convolution of f (t)
and a train of impulses separated by T units of time.

00

f(k) = I: f(t) 8(t - k). (1)

k = -00
Further if we assume the properties of linearity, time invari-
ance and causality we can see that
k
y(k) = I: h(k - i) u(i)
i =0
where u( ) is the input sequence, h( ) is the impulse response,
and y( ) is the output sequence.

Example 1: Consider a savings account in a bank. Assume

that the interest is compounded monthly at r%.

If we deposit u(O) rupees at the beginning of a month, then

y(O) = u(O). Hence, y(l) = y(O) + l~O u(O) is the balance
at the beginning of next month. It is now easy to see that
k
y(k) = .2: (1 + (l~O))k - i u(i)
'/, = 0
is the total amount in the account if Rs.u(i) is deposited at
the beginning of every month.

We can also write the same convolution summation using

induction and arrive at the following equation. We leave
the derivation to the interested reader.

r
y(k + 1) - (1 + 100) y(k) = u(k + 1) ;k =0,1,2, ......
(2)
This is called a first order linear difference equation with
constant coefficients. Linear difference equations are to dis-
crete systems as linear differential equations are to continu-
ous systems. Differential equation has initial conditions as

-R-ES-O-N-A-N-C-E--I-F-e-br-U-a-rY--2-00-0-----------~~-----------------------------43--
 GENERAL I ARTICLE

derivative terms evaluated at say t = 0 while a difference

equation has initial conditions as shifted terms, such as y(O),
y ( -1), and y ( - 2) for a third order difference equation, for
example. Owing to several reasons such as number of coeffi-
cients, number of operations and the presence of initial con-
ditions, the difference equation is preferable to convolution
describing a Linear Time Invariant Lumped (LTIL) system.
Notice that a similar argument is applicable to differential
equations.

Example 2 : The difference equation

y(k + 1) (1 + b - d) y(k) + u(k)

describes the population mod~l of any country where b is

the birth rate, d is the death rate and u(k) is the number of
net immigrants entering the country in year k.

We leave it to the reader to model the vacillating mathe-

matician problem [3] as a discrete-time system.

General Forms of Difference Equations

An nth order difference equation may be written, typically,

either as

y(k + n) + an-l y(k + n - 1) + + aO y(k) =

bm u(k + m) + bm-l u(k + m - 1) + .. + bo u(k) (3)
which is called the advanced form, or as

y(k) + an-l y(k 1) + + ao y(k - n) =

bm u(k + m n) + + bo u(k - n) (4)

which is called the delayed form where nand m are fixed
integers and k ranges over -n, - n + 1, 0, 1, 2,
The coefficients ai and bi are real constants, not necessarily
non-zero. We assume that the coefficient of y (k + n) in (3)
or y(k) in (4) has been normalized to one. If m > n, it can
be verified that the system is not causal. Thus, for causal
systems, we require n ~ m.

Both the forms are widely used in applications. The ad-

vanced form is used in control and filter design, the delayed

-44------------------------------~-------------------------------
RESONANCE I February 2000
 GENERAL I ARTICLE

form is used in estimation, identification and DSP. To find

the response y(k), k = 0, 1, 2 ... excited by an input u(k),
we need n initial conditions y(j), j = - 1, - 2, ... - n.
(It is often assumed that u(k) = 0 for k < 0). Once the
initial conditions are given, the response can be obtained by
direct substitution. The solution, however, is not in closed
form and is difficult to infer from the general properties of
the system.

The Z-Transform

N ow let us look at a transform called the Z-transform to

study the system. Recall, the Laplace transform of a signal
f(t) given by
00
F(s) = 10 j(t) e- st

for t > O. Now that the continuous-time signal is dis-

cretized, the integral simply reduces to a summation over
the index k, i.e,
00

F(s) 2:: f(kT) e- skT

k = 0

Let us define a new complex variable

sT
z = e (5)

Accordingly the transformation becomes

00

F(z) = L f(kT) z-k (6)

k = 0

and F(z) is called the Z-transform of f(kT). Several prop-

erties analogous to the Laplace transform such as linearity,
frequency shifting, time shifting, initial value theorem, final
value theorem etc. are also applicable for the Z-transform.
We may apply the Z-transform to the difference equation
and obtain a closed form solution. We have given some im-
portant properties of the Z-transform and the Z-transform
of some signals in Box 1. Analogous to continuous-time sys-
tem discussed in [1], we get the zero input response (due to
initial conditions alone) and the zero state response (due to
externally applied input alone). We also have the notions of

-R-ES-O-N-A-N--C-E-I--Fe-b-rU-a-r-Y-2-0-0-0-----------~-------------------------------4-5
 GINlRAL I ARTICLE

Box 1. Z-Transforms.

An extensive list of Z-transforms may be found in [4]. However, a few signals and
their Z-transforms are given below.
Unit impulse Z {8 (k)} = 1
U nit step sequence Z {q(k)} = z: I
eak Z e ak = _ z _a
z - e
cos (w k) Z cos(w k) = z(z - cos(w))
z2 - 2 z cos(w) + I
f(k - i) z -i F (z) + z -i f ( -1) + ... + f ( - i)
f(k + i) ziF(z) - zif(O) - ... - z f(i - 1)
initial value f (0) lim z _ 00 F ( z )
final value f (00 ) lim z _ I (z - 1) F (z )

transfer function (the ratio of the transform of output to the

transform of input with all initial conditions equal to zero),
and complete characterization. In other words, Z-transforms
are to discrete-time systems as Laplace transforms are to
continuous-time systems. However, in the present article, we
shall employ a new technique to solve the difference equa-
tion. This technique is called the State-Space description of
LTIL systems.

Description of Systems in State Space

For the sake of clarity, let us assume that the order of the
system is 2 and that all biS except bO are zero in' (3). The
system is now

y(k -+: 2) + at y(k + 1) + ao y(k) = bO u(k). (7)

Let us now define 2 variables Xl (k) and X2 (k) as

Xl (k) = y(k), x2 (k) = Xl (k + 1) (8)

so that in terms of these variables the system (7) may be

rewritten as

X2 (k + 1) = -an Xl (k) - al X2 (k) + bo u(k). (9)

-46----------------------------~~-----------R-E-S-O-N-A-N-C-E-I--Fe-b-ru-a-rv--20-0-0
 GENERAL I ARTICLE

Together with the definitions of the 2 variables in (8) we

may write

X (k + 1) = A X (k) + B u(k), (10)

and
y(k) = C X (k) (11)
where
o
-ao o]
(12)
The following observations may be readily made.

1. An nth order difference equation is split into n first

order difference equations.

2. The vector X comprising the variables Xi, i = 1,2,3,

... n is called the state vector. If X has n state vari-
ables, then it has the dimension n x 1. Notice that
the initial conditions remain intact. They are simply
renamed as X (O)! Each of the first order difference
equations describes the evolution of an internal state
of the system. Accordingly, the name of the variable
is state variable and hence the vector X is state vec-
tor. Equation (10) which involves the dynamics of the
state vector is called the state equation. It describes
the relationship between the input and the state.

3. For an nth order system, the square matrix A, called

the system matrix, has the dimension n x n. (What
are the eigen values of A?)

4. If there is a single input, then the dimension of B is

n x 1. I t is easy to see that the dimension of B
becomes n x r if the system is driven by r inputs,
.
packed Into a vector U- = [Ul, U2, .... u r ]T

5. If there is a single output, then C is simply a row

vector; otherwise its dimension is m x n, where m
is the number of outputs of the system. The algebraic
equation (11) which gives the output in terms of the
state vector is called the output equation.

--------~-------­
RESONANCE I February 2000 47
 GENERAL I ARTICLE

6. Without loss of generality, we may also include u in

the output equation with an appropriate weight i.e.,
y = c X + D u. (When does D come into the
picture?)

State-variable equations are also called state-space equations

because the state vector forms a linear space. By now, the
reader should appreciate the obvious advantage of such a
description of a system in terms of state variables. Unlike
transfer functions, which are limited to single input single
output or SISO systems, state-space equations are readily
extendable to multi-input multi-output or MIMO Systems.

Originally, the idea of state space was given by Poincare.

But it was Rudolf Kalman who had exploited this descrip-
tion in the 1960s. This approach introduced the concepts of
controllability and observability, which led to a better under-
standing of the structure of systems, and a number of new
results, such as state feedback and state estimator. Refer to
Part 1 of [1] for a brief history of modern control.

Applying Z-Transform to State Space Equations: By

applying the Z-transform to the state-space equations (10)
and (11), we get
y (z) = C (ZI - A)-l [Z X (0) + B (j (z)] + D 0- (z)
For a 8180 system, we may easily verify that

Y(z) = H(z) = C (ZI - A)-l B + D

U(z)
(with the initial conditions assumed to be zero.) This is a
fundamental equation relating the state-space equation and
discrete transfer function. (What is the characteristic poly-
nomial of the system?)

An important point is that the state space description of

a system is not unique. The one presented in this article,
called the controllable canonical form" is simple and is in-
tuitively appealing. Physically, not all the state variables
defined as before may have a meaning. However, mathe-
matically, we may perform similarity transformations to get
a variety of models for the same system.

--------~--------
48 I February 2000 RESONANCE
 GENERAL I ARTICLE

Suggested Reading

[1] A Ramakalyan and J R Vengateswaran, Systems and Control Engineer-

ing,Resonance, Vol. 4, Nos. 1,3 & 5, 1999.
[2] Kapil Krishnan Manu and R Ramaswamy, Chaos, Resonance, Vol. 3,
Nos. 4,6,10,1998.
[3] KB Athreya, The Vacillating Mathematician Problem, Resonance, Vol.
2, No.1, 1997.
Address for Correspondence
[4] C T Chen, System and Signal Analysis, Holt, Rinehart.nd Winston,
A Ramakalyan
N.Y.,1994. (Also available from Oxford Univ. Press priced in rupees.)
Department of Instrumentation
[5] J G Proakis and D G Manolakis, Digital Signal Processing: Principles,
and Control Engineering,
Algorithms, and Applications, 3/e, Prentice Hall oflndia, 1997.
Regional Engineering College
[6] A V Oppenheim and R W Schafer, Discrete-time Signal Processing,
Tiruchirapalli 620 015, India .
Prentice Hall, Englewood Cliffs, NJ, 1989. (Also available in Prentice
Email:rkalyn@rect.ernet.in
Hall of India).

Solution to Crossword Puzzle

(Published in the previous issue)

(Cross Word Puzzle Contributed by Vinita Krishnamurthy)

-RE-S-O-N----A-N~C-E--I-F-eb-r-U-a-rY--2-0-0-0-----------~~-------------------------------4-9