Chapter 2

Basics of Signals
2.1

What are Signals?

As mentioned in Chapter XX, a system designed to perform a particular task

often uses measurements obtained from the environment and/or inputs from a
user. These in turn may be converted into other forms. The physical variables
of interest are generally called signals. In an electrical system, the physical
variables of interest might be a voltage, current, amount of charge, etc. In
a mechanical system, the variables of interest might be the position, velocity,
mass, volume, etc. of various objects. Financial examples might include the
price of a stock, commodity, or option, an interest rate, or an exchange rate.
In performing its tasks, the system may need to manipulate or combine various
signals, extract information, or otherwise process the signals. These actions are
called signal processing or signal analysis.
A convenient abstraction is to model the value of a physical variable of
interest by a number. We are usually interested in the physical variable not
at just a single time, but rather at a set of times. In this case, the signal is a
function of time, say f (t). For example, f (t) might denote a voltage level, or
the velocity of an object, or the price of a stock at time t.
In some cases, we might be interested in measuring the quantity as a function of some variable other than time. For example, suppose we are interested
in measuring the water temperature in the ocean as a function of depth. In this
case, the signal is a function of a spatial variable, with f (x) denoting temperature at depth x.
A signal need not be a function of just a single variable. To continue the
example above, suppose we are interested in the temperature at particular points
in the ocean, not simply as a function of depth. In this case, we might let
f (x, y, z) denote the temperature at the point (x, y, z), so the signal is a function
1999-2002
c

by Sanjeev R. Kulkarni. All rights reserved.

Notes for ELE201 Introduction to Electrical Signals and Systems.
Thanks to Richard Radke for producing the figures.
Lecture

CHAPTER 2. BASICS OF SIGNALS

of three variables. Now, if we are also interested in how the temperature evolves
in time, the signal f (x, y, z, t) would be a function of four variables.
The word "Matlab"
3

intensity

0.05

0.1

0.15

0.2

0.25
0.3
seconds

0.35

0.4

0.45

0.5

Figure 2.1: Someone saying the word Matlab

Examples of signals that we will encounter frequently are audio signals,
images, and video. An audio signal is created by changes in air pressure, and
therefore can be represented by a function of time f (t) with f representing
the air pressure due to the sound at time t. An example of an audio signal
of someone saying Matlab is shown in Figure 2.1. A black and white image
can be represented as a function f (x, y) of two variables. Here (x, y) denotes a
particular point on the image, and the value f (x, y) denotes the brightness (or
gray level) of the image at that point.
An example of a black and white image is shown in Figure 2.2. A video can
be thought of as a sequence of images. Hence, a black and white video signal can
be represented by a function f (x, y, t) of three variables (two spatial variables
and time). In this case, for a fixed t, f (, , t) represents the still image/frame at
time t, while for a fixed (x, y), f (x, y, ) denotes how the brightness at the point
(x, y) changes as a function of time.
Three frames of a video of a commercial are shown in Figure 2.3. It turns
out that color images (or video) can be represented by a combination of three
intensity images (or video, respectively), as will be discussed later in Chapter
XX.

2.2

Analog and Digital Signals

Often the domain and the range of a signal f (x) are modeled as continuous.
That is, the time (or spatial) coordinate x is allowed to take on arbitrary values
(perhaps within some interval) and the value of the signal itself is allowed to
take on arbitrary values (again within some interval). Such signals are called

2.2. ANALOG AND DIGITAL SIGNALS

Figure 2.2: A gray-scale image.

0:13

0:14

Figure 2.3: Video frames from a commercial.

0:15

CHAPTER 2. BASICS OF SIGNALS

analog signals. A continuous model is convenient for some situations, but in

other situations it is more convenient to work with digital signals i.e., signals
that have a discrete (often finite) domain and range. Two other related words
that are often used to describe signals are continuous-time and discrete-time,
referring to signals where the independent variable denotes time and takes on
either a continuous or discrete set of values, respectively.
Sampled version of temperature graph
36

35

34

temperature

33

32

31

30

29

28

10
seconds

12

14

16

18

20

Figure 2.4: Sampling an analog signal.

Sometimes a signal that starts out as an analog signal needs to be digitized
(i.e., converted to a digital signal). The process of digitizing the domain is called
sampling. For example, if f (t) denotes temperature as a function of time, and
we are interested only in the temperature at 1 second intervals, we can sample
f at the times of interest as shown in Figure 2.4.
Another example of sampling is shown in Figure 2.5. An original image
f (x, y) is shown together with sampled versions of the image. In the sampled
versions of the image, the blocks of constant intensity are called pixels, and the
gray level is constant within the pixel. The gray level value is associated with
the intensity at the center of the pixel. But rather than simply showing a small
dot in the center of the pixel, the whole pixel is colored with the same gray level
for a more natural appearance of the image. The effect of more coarse sampling
can be seen in the various images. Actually, the so-called original image in
Figure 2.5a is also sampled, but the sampling is fine enough that we dont notice
any graininess.
The process of digitizing the range is called quantization. In quantizing a
signal, the value f (x) is only allowed to take on some discrete set of values (as
opposed to the variable x taking on discrete values as in sampling).
Figure 2.6 shows the original temperature signal f (t) (shown previously in
Figure 2.4) as well various quantized versions of f . Figure 2.7 shows the image
from Figure 2.2 and various quantized versions. In the quantized versions of the
images, the gray levels can take on only some discrete set of values. Actually,

2.2. ANALOG AND DIGITAL SIGNALS

No subsampling

4 x 4 blocks

8 x 8 blocks

16 x 16 blocks

Figure 2.5: Sampling an image.

CHAPTER 2. BASICS OF SIGNALS

Unquantized signal

32 levels

0.8

0.8

0.6

0.6

0.4

0.4

0.2

0.2

0.5

1.5

0.5

16 levels
1

0.8

0.8

0.6

0.6

0.4

0.4

0.2

0.2

0.5

1.5

1.5

8 levels

1.5

0.5

Figure 2.6: Quantized versions of an analog signal.

the so-called original image is also quantized, but because of the resolution
of the printer and limitations of the human visual system, a technique known
as halftoning (discussed in Chapter XX) can be used so that we dont notice
any artifacts due to quantization. It is typical in images to let the gray level
take on 256 integer values with 255 being the brightest gray level and 0 the
darkest. In Figures 2.7d-f there are only 8, 4, and 2 gray levels respectively, and
quantization artifacts become quite noticeable.
Sampling and quantization to digitize a signal seem to throw away much
information about a signal, and one might wonder why this is ever done. The
main reason is that digital signals are easy to store and process with digital
computers. Digital signals also have certain nice properties in terms of robustness to noise, as well discuss in Section XX. However, there are also situations
in which analog signals are more appropriate. As a result there is often a need
for analog-to-digital conversion and digital-to-analog conversion (also written
A/D and D/A conversion). In digitizing signals, one would also like to know
how much information is lost by sampling and quantization, and how best to
do these operations. The theory for sampling is clean and elegant, while the
theory for quantization is more difficult. It turns out that choices for sampling
rates and number of quantization levels also depend to a large extent on system
and user requirements. For example, in black-and-white images, 256 gray levels
is adequate for human viewing much more than 256 would be overkill, while
much less would lead to objectionable artifacts. We defer a more detailed consideration of sampling and quantization until Chapter XX after we have covered

2.2. ANALOG AND DIGITAL SIGNALS

256 levels

32 levels

16 levels

8 levels

4 levels

2 levels

Figure 2.7: Quantized versions of a gray-scale image.

CHAPTER 2. BASICS OF SIGNALS

some additional background material.

Before discussing some basic operations on signals, we describe a fairly common notational convention which we will also follow. Continuous-time signals
will be denoted using parentheses, such as x(t), while discrete-time signals will
use brackets such as x[n]. This convention also applies even if the independent variable represents something other than time. That is, y(u) denotes a
signal where the domain is continuous, while y[k] indicates a discrete domain,
whether or not the independent variables u and k refer to time. Often the letters
i, j, k, l, m, n are used to denote a discrete independent variable.

2.3

Some Basic Signal Operations

In addition to the obvious operations of adding or multiplying two signals, and

differentiating or integrating a signal, certain other simple operations are quite
common in signal processing. We give a brief description of some of these here.
The original signal is denoted by x(t).
Original signal, f(x)
6
4
2
0
8

2
0
2
Amplitudeshifted signal, f(x)+1.5

0
2
Timeshifted signal, f(x+3)

6
4
2
0

6
4
2
0
2

Figure 2.8: Amplitude- and time-shifted versions of a signal.

The signal a + x(t) where a is some number is just adding a constant signal
to x(t) and simply shifts the range (or amplitude) of the signal by the amount
a. A somewhat different operation is obtained when one shifts the domain of
the signal. Namely, the signal x(t t0 ) is a time-shift of the original signal x(t)
by the amount t0 . Its like a delayed version of the original signal. Figure 2.8
shows amplitude and time-shifted versions of a signal.

2.3. SOME BASIC SIGNAL OPERATIONS

A periodic signal
0.7

0.6

0.5

0.4

0.3

0.2

0.1

20

40

60

80

100
seconds

120

140

160

180

200

Figure 2.9: A periodic signal.

For some signals, appropriate time shifts can leave the signal unchanged.
Formally, a signal is said to be periodic with period P if x(t P ) = x(t) for all
t. That is, the signal simply repeats itself every P seconds. Figure 2.9 shows an
example of a periodic signal.
Amplitude scaling a signal to get ax(t) is simply multiplying x(t) with a
constant signal a. However, a rather different operation is obtained when one
scales the time domain. Namely, the signal x(at) is like the original signal,
but with the time axis compressed or stretched (depending on whether a > 1 or
a < 1). Of course, if a = 1 the signal is unchanged. Figure 2.10 shows the effects
of amplitude and time scaling. For negative values of a, the signal is flipped (or
reflected) about the range axis, in addition to any compression or stretching.
In particular, if a = 1, the signal is reflected about the range axis, but there
is no stretching or compression. For some functions, the reflection about the
range axis leaves the function unchanged, that is, the signal is symmetric about
the range axis. Formally, the property required for this is x(t) = x(t) for all
t. Such functions are called even. A related notion is that of an odd function,
for which x(t) = x(t). These functions are said to be symmetric about the
origin, meaning that they remain unchanged if they are first reflected about
the range axis and then reflected about the domain axis. Figure 2.11 shows
examples of an even function and an odd function.
The signal x(y(t)) is called the composition of the two functions x() and
y(). For each t, it denotes the operation of taking the value y(t) and evaluating
x() at the time y(t). Of course, we can get a very different result if we reverse
the order and consider y(x(t)).
One other operation that is extremely useful is known as convolution. We
will defer a description of this operation until Section XX.

10

CHAPTER 2. BASICS OF SIGNALS

Original signal, f(x)

8
6
4
2
0
10

0
Amplitudescaled signal, 2f(x)

10

10

0
Timescaled signal, f(x/2)

10

10

0
Timescaled signal, f(2x)

10

10

10

8
6
4
2
0

8
6
4
2
0

8
6
4
2
0

Figure 2.10: Amplitude- and time-scaled versions of a signal.

2.4. NOISE

11
An even function

0.6
0.4
0.2
0
-0.2
-0.4
-0.6
-50

-40

-30

-20

-10

10

20

30

40

50

10

20

30

40

50

An odd function
0.6
0.4
0.2
0
-0.2
-0.4
-0.6
-50

-40

-30

-20

-10

Figure 2.11: Examples of even and odd functions.

2.4

Noise

In many applications desired signals are subject to various types of degradations.

These degradations can arise from a variety of sources such as limitations of the
sensing device, random and/or unmodeled fluctuations of underlying physical
processes, or environmental conditions during sensing, transmission, reception,
or storage of the data. The term noise is typically used to describe a wide range
of degradations.
It is often useful to try and model certain properties of the noise. One
widely used model is to assume that the original (desired) signal is corrupted
by additive noise, that is, by adding another unwanted signal. Of course, if
we knew the noise signal that was added, we could simply subtract it off to
get back the original signal, and the noise would no longer be an issue. Unfortunately, we usually do not have such detailed knowledge of the noise signal.
More realistically, we might know (or assume) that the noise satisfies certain
properties without knowing the exact values of the noise signal itself. It is very
common to model the noise as random, and assume that we know something
about the distribution of the noise. For example, we might assume that the
noise is randomly (uniformly) distributed over some interval, or that it has a
Gaussian (normal) distribution with a known mean and variance. Even this
minimal type of knowledge can be extremely useful as we will see later.
Robustness to effects of noise can be a major design consideration for certain
systems. This can be one reason why for many applications a digital system

12

CHAPTER 2. BASICS OF SIGNALS

Original signal
1

0.8

0.6

0.4

0.2

0.5

1.5

1.5

Noisy signal
1

0.8

0.6

0.4

0.2

0.5

Figure 2.12: Adding noise to an analog signal.

Original signal, quantized to 8 levels

1
0.8
0.6
0.4
0.2
0

0.5

1.5

Noisy quantized signal

1
0.8
0.6
0.4
0.2
0

0.5
1
Original signal, recovered by requantization

1.5

0.5

1.5

1
0.8
0.6
0.4
0.2
0

Figure 2.13: Adding noise to a quantized signal.

2.5. SOME COMMON SIGNALS

13

might be preferred over an analog one. Of course, the power of digital computing
is also a key reason for the prevalence of digital systems, and robustness to noise
is one factor that makes digital computing so reliable. Figures 2.12 and 2.13
illustrate the effect of adding noise on an analog signal and a quantized (although
still continuous-time) signal. Without further knowledge of signal and noise
characteristics, the noise cannot be removed from an analog signal since any
possible value could be a valid value for the signal. On the other hand, if we
know the original signal is quantized (so it takes on only a discrete set of values),
then depending on the noise level, it may be possible to remove much of the
noise by simply re-quantizing the noisy signal. This process simply maps the
observed signal values to one of the possible original levels (for example, by
selecting the closest level).

2.5

Some Common Signals

Here we briefly define some signals that we will commonly encounter. Perhaps
the most basic and frequently used signal is a sinusoid defined by
x(t) = A sin(t)
and shown in Figure 2.14.
The sinusoid x(t) = A sin( t)

-A
2 /

2 /

Figure 2.14: The sinusoid x(t) = A sin(t).

Here A is the amplitude, and is the radian frequency. The units of
are radians/sec so that when multiplied by time t (in sec) we get radians. An
equivalent form for the sinusoid that is often used is
x(t) = A sin(2f t).
The frequency f is in units of Hertz (abbreviated Hz) which is sec1 , or often
called cycles per second. Of course, f and are related by = 2f . Also,

14

CHAPTER 2. BASICS OF SIGNALS

its clear that since sin( + /2) = cos , we could have equivalently written the
sinusoid as
x(t) = A cos(2f t /2).
Up to this point, we have only considered real-valued signals. Although
physical quantities can generally be represented in terms of real-valued signals,
it turns out to be extremely useful to consider signals taking on complex values.
The most basic complex-valued signal we will use is the complex exponential
ejt . (Note that here
we have used the symbol j instead of i to denote the
imaginary number 1. This is common in electrical engineering since the
symbol i has traditionally been used to represent an electrical current.) The
well-known Euler identity can be used to write the complex exponential in terms
of standard sinusoids. Namely,
ejt = cos(t) + j sin(t).
As with sinusoids, the complex exponential can also be written in terms of
frequency in Hertz rather than radian frequency.
Some other signals that we will use on occasion and therefore give special
symbols to are the step function, ramp, square wave, triangle wave, and the sinc
function (pronounced like sink). These signals are defined by

0 if t < 0
step(t) =
1 if t 0

0 if t < 0
ramp(t) =
t if t 0

1 if 1/2 t < 1/2

rect(t) =
0 otherwise

1 |t| if 1 t 1
tri(t) =
0 otherwise
and
sinc(t) =

sin(t)
t

and are shown in Figure 2.15.

2.6

Delta Functions

The notion of a delta function is extremely useful in the analysis of signals

and systems, although it may feel unnatural on first exposure. Although the
concept of the delta function can be made completely rigorous, rather than get
side-tracked with too much mathematical detail and sophistication, our aim
here is to provide some intuition and ability to work with the delta function.
On the other hand, it is important to have enough rigor so that this important
tool is used properly.

2.6. DELTA FUNCTIONS

15

step(t)

ramp(t)

-1

-1

-2
-2

-1

-2
-2

-1

rect(t)
2

-1

-1

-1

tri(t)

-2
-2

-2
-2

-1

sinc(t)
1

0.5

-0.5
-5

-4

-3

-2

-1

Figure 2.15: (a) Step function. (b) Ramp function. (c) Rectangle function. (d)
Triangle function. (e) Sinc function.

16

CHAPTER 2. BASICS OF SIGNALS

The delta function in continuous-time is also called the Dirac delta function,
unit impulse function, or sometimes just the impulse function. It is defined
implicitly through its behavior under integration as follows:
Definition: (t) is the Dirac delta function if it satisfies
Z
f (t)(t) dt = f (0)

(2.1)

for every function f (t) that is continuous at t = 0.

From this definition we can infer the following two properties of the delta
function. First, by considering the function f (t) = 1 in Equation (2.1), we get
Z
(t) dt = 1
(2.2)

This result implies that the area under the delta function is equal to 1.
The second property gives the value of (t) for t 6= 0. Suppose (t) took on
positive values in even a very small interval away from t = 0. Then we could
choose a function f (t) that also took positive values inside a portion of this same
interval, but with f (t) = 0 elsewhere (including t = 0) and with f (t) continuous
at t = 0. However, in this case the left hand side of Equation (2.1) must be
positive, but the right hand side is 0. Therefore, (t) cannot take on positive
values in any interval. A similar argument leads us to the conclusion that (t)
cannot take on negative values in any interval. Thus, (t) = 0 for t 6= 0.
These two results (namely, that the area under (t) is 1 and that (t) = 0 for
all t 6= 0) are inconsistent with our usual notions of functions and integration.
If (0) was any finite value, then the area under (t) would be zero. Strictly
speaking, (0) is undefined although it is convenient to think of (0) = . Thus,
although we call (t) the delta function, it is technically not a function in the
usual sense. It is what is known as a distribution. However, it turns out that
for many manipulations we can treat (t) like a function.
It is also convenient to have a graphical representation as shown in Figure
2.16. The arrow indicates that the value at t = 0 is infinite (or undefined), with
the height of the arrow indicating the area under (t). To depict A(t) where
A is some constant, we would draw the height of the arrow to be A.
It is sometimes also helpful to think of (t) as a limit of a sequence of
approximating functions. Consider the function a rect(at). This has area 1, but
if a > 1 it is more concentrated around t = 0. As we let a we get a
sequence of approximations as shown in Figure 2.17, which intuitively get closer
and closer to (t). In fact, it is not hard to verify that for f (t) continuous at
t = 0 we have
Z
f (t) a rect(at) dt f (0) as a

so that in the limit a the defining property of (t) is indeed satisfied.

It turns out that many other choices for the approximating functions will also
work if the area is 1 and scaling is done to get concentration at t = 0.

2.6. DELTA FUNCTIONS

17

The delta function

3

2.5

1.5

0.5

0
1

0.8

0.6

0.4

0.2

0.2

0.4

0.6

0.8

Figure 2.16: Representation of a delta function.

a = 1/2

a=1

3.5

3.5

2.5

2.5

1.5

1.5

0.5

0.5

0
-2

-1

0
-2

-1

a=2
4

3.5

3.5

2.5

2.5

1.5

1.5

0.5

0.5

0
-2

-1

a=4

0
-2

-1

Figure 2.17: The delta function is a limit of rectangle functions with area 1.

18

CHAPTER 2. BASICS OF SIGNALS

The primary motivation for introducing the delta function is that it is a

useful tool in analyzing systems and signals as well see in Chapter XX. There
is also a physical motivation involving modeling of physical phenomena that
is actually closely related to the analytical motivation. In some situations we
would like to model physical phenomena that occur in a time interval short
compared to the resolution of a measuring device. For example, we may be
interested in the energy entering a camera due to a flash of light as the camera
shutter opens and closes very quickly. For a fixed shutter speed, many models
for the light source will be good enough to represent a flash or impulse of
light. However, if we fix a particular function to model the flash of light, it may
not represent a true flash for faster shutter speeds. Modeling the flash of light
as a delta function is an idealization that works for any (non-infinite) shutter
speed.
A number of properties of (t) can be obtained directly from the definition by
utilizing the usual rules of integration. Intuition can also sometimes be gained
by considering approximations of (t), although a formal justification generally
requires verifying the defining property.
For example, since (t) = 0 for all t 6= 0, it seems obvious that (t) is even.
However, since (t) is technically not a function, we should really verify directly
that (t) behaves just as (t) in the defining property of Equation (2.1). This
can be done by a simple change of variable as follows. For any function f (t)
continuous at t = 0, we have
Z
Z
f (t)(t) dt =
f (u)(u) du = f (u) |u=0 = f (0)

where the first equality is obtained by the change of variable u = t, and the
second equality follows from the definition of (u). The conclusion is that (t)
satisfies the required property of (t), and so (t) = (t).
By the change of variable u = at and considering the cases a > 0 and a < 0
separately, it is easy to show that
(at) =

1
(t).
|a|

By the change of variable u = t t0 , it follows that

Z
x(t)(t t0 ) dt = x(t0 ).

Therefore, the time-shifted delta function (tt0 ) behaves like we would expect.
This property is sometimes called the sifting property of the delta function. The
natural graphical depiction of (t t0 ) is shown in Figure 2.18.
We now turn to the discrete-time delta function, also called the Kronecker
delta function. The Kronecker delta function is denoted by [n] and defined as

1 if n = 0
[n] =
0 otherwise

2.6. DELTA FUNCTIONS

19

Timeshifted delta function, (t0.6)

3

2.5

1.5

0.5

0
1

0.8

0.6

0.4

0.2

0.2

0.4

0.6

0.8

Figure 2.18: Time-shifted delta function.

Discretetime delta function

3

2.5

1.5

0.5

0.5

1
1

0.8

0.6

0.4

0.2

0.2

0.4

0.6

0.8

Figure 2.19: Discrete-time delta function.

20

CHAPTER 2. BASICS OF SIGNALS

Figure 2.19 shows the graph of [n]. Hence, in discrete-time, the delta function is in fact a function in the proper sense. There are none of the mathematical
subtleties/difficlties associated with the continuous-time delta function. In fact,
[n] is rather simple to work with.
Many properties of (t) have analogous counterparts in discrete-time, and
the discrete-time properties are generally easier to verify. For example, the
result

X
f [n][n] = f [0]
n=

follows trivially from the definition. Recall that in continuous time, the analogous property was actually the definition. Also trivial is the fact that [n] is
an even function of n = . . . , 1, 0, 1, . . .. It is easy to see that the time-shifted
delta function [n n0 ] satisfies the discrete-time sifting property

f [n][n n0 ] = f [n0 ].

n=

It turns out that for some properties the discrete-time counterpart is not analogous. For example, in discrete-time if a is an integer we have [an] = [n].

2.7

2-D Signals

One useful notion that arises in two (and higher) dimensions is separability. A
function f (x, y) is called separable if it can be written as f (x, y) = f1 (x)f2 (y).
Many of the commonly encountered 2-D functions are simply separable extensions of the corresponding 1-D functions. For example, the 2-D version of the
complex exponential is
ej(1 x+2 y) = ej1 x ej2 y
where 1 and 2 are the radian frequencies in the x and y directions, respectively.
That is, the 2-D complex exponential is simply the product of a 1-D complex
exponential in each direction.
Likewise, the 2-D Dirac delta function (x, y) is given by
(x, y) = (x)(y)
Formally, (x, y) would actually be defined by the property
Z Z
f (x, y)(x, y) dx dy = f (0, 0)

for any function f (x, y) continuous at (0, 0), but showing equivalence with the
separable expression is straightforward.
Similarly, in the discrete case, the Kronecker delta [m, n] is defined by

1 if m = n = 0
[m, n] =
0 otherwise

2.7. 2-D SIGNALS

21

Clearly the Kronecker delta is also separable

[m, n] = [m][n].
Many other 2-D functions like rect(x, y) and sinc(x, y) have natural separable
extensions from the 1-D versions as well.
Separability can offer some nice analytical and computational advantages.
For example, terms involving the two independent variables can sometimes be
separated, reducing a two-dimensional analysis to two separate one-dimensional
analyses. This can result in computational savings by allowing processing to be
done along the two dimensions separately.