Chapter 1.

Background
Principles in Digital Communication

Chapter 1-1. Fourier Transform

1
A. Definition

2
 Formal Definition of Spectrum
Given a time-domain signal s(t), its Fourier transform is defined as follows.

A signal in time The frequency domain

spectrum of s(t)
 
s (t ) S( f ) =  s (t ) e
− j 2 ft
dt
−

S(f) is referred to as the Fourier spectrum (or spectrum) of s(t). Sometimes we

will write
S(f) = F (s(t)) or s (t ) . S ( f )
As explained earlier, frequency is a generalized concept of “color”. Spectrum is
a generalized concept of the color description of a light signal.

3
 Useful Fourier Transform Pairs
A At sinc( f t )

-t/2 t/2 t
 -2t -1/t 1/t 2/t
f

At sinc( tt ) duality A

t f
-2t -1/t 1/t 2/t -t/2 t/2

A A d(f)
f
0 t
duality
A
d(t) t A

0 f
cos (2f0t) 0.5

t -f0 0 +f0 f

sin (2f0t) 0.5j -0.5j

t -f0 0 +f0 f

4
Example: A Rectangular Pulse
Consider s(t) as a rectangular pulse shown here. The corresponding Fourier
transform S(f) is given by:
 s(t)
S ( f ) =  s (t )e − j 2 ft
dt A
−
t /2 t /2
e − j 2 ft

− j 2 ft
= Ae dt = A
t
− /2
− j 2 f −t / 2
-t/2 0 t/2 t
e − j f t − e + j f t t
= A
− j 2 f
sin( f t )
= At  S(f)
 ft
Define sinc(x)  sin(x)/x. We then have At
f
S ( f ) = At sinc( f t ) -2/t -1/t 1/t 2/t

Question: (a) If t increases or reduces, what happens to the spectrum?

(b) Is the spectrum symmetric? 5
 Some Useful Relationships
The basic relationship
e jx = cos( x) + j sin( x) (1)
e− jx = cos( x) − j sin( x) (2)
Adding (1) and (2), we have
1 jx
cos( x) = (e + e− jx ) (3)
2
Subtracting (2) from (1), we have
1 jx
sin( x) = (e − e − jx ) (4)
2j

6
More Precise Drawing

s(t)
1

-1/2 0 1/2 t

S(f)

7
 An Important Observation
A KEY observation from the spectrum analysis is that the narrower a pulse,
the wider its spectrum. This creates the fundamental problem in commutation
systems, i.e., if we want to increase transmission speed using narrow pulses,
we have to use more spectrum.
s(t) s(t)
A A

t t
t
t

S(f) S(f)
At At
f f
-1/t 1/t -1/t 1/t

8
Example: The Spectrum of a Constant

A time-domain signal: s(t)=A


The corresponding Fourier transform: S ( f ) = A  e − j 2 ft dt
−
 

e  1dt = 
− j 2 0 t
When f=0: dt =
− −
 

  (cos(−2 ft ) + j sin(−2 ft ))dt = 0

− j 2 ft
When f≠0: e dt =
− −

Therefore S(f)=Ad(f) S(f)=Ad(f)

s(t)=A A

0 f
t

Notes: You will learn the definition of d(f) in signal analysis. d(f) means a
very large signal at f=0 (i.e., zero frequency) and nothing at other frequencies.9
 About d(x)
d(x) is a very unusual function. It can be 1/e →∞
defined as follows.
 at x =0 d(x)
d ( x) = 
0 at x  0


 d ( x)dx = 1
−

In words, d(x) is a very, very “tall” and

“narrow” pulse. The area under the d(x) is 1. area =1

This is a quite strange definition. However, this

definition is very useful in describing the
x
behavior at a particular “point”. The simplified
e→0
notation of d(x) is an arrow.
d(x)
0 x 10
 Example: The Spectrum of d(t)
For s(t)=d(t), we have



− j 2 ft
S( f ) = d (t ) e dt
−
Note that d(t) is zero except around t=0, and ej2ft=1 when t=0. We thus have
0+
S ( f ) =  d (t ) 1dt
0−
According to the definition of d(t), we have
0+
S ( f ) =  d (t )dt = 1
0−

s(t)=Ad(t)
A S(f)=A

t 0 f

11
 Duality Property
We can put the previous two results together.

S(f)=Ad(f)
s(t)=A A

0 f
t

s(t)=Ad(t)
A S(f)=A

t 0 f

We can see that if we swap the time and frequency domains, the resultant two
functions still form a pair of Fourier transforms. This is a special case of the duality
property. 12
B. Magnitude, Power and Energy Spectra

13
 Different Spectrums
S(f)
spectrum of s(t) f
-2/t -1/t 1/t 2/t
|S(f)|

f
magnitude spectrum -2/t -1/t 1/t 2/t
|S(f)|2

f
energy spectrum
-2/t -1/t 1/t 2/t
|S(f)|2/T

f
power spectrum
-2/t -1/t 1/t 2/t

Note: In this course, “spectrum” may mean any one above.

14
 Power Spectrum
|S(f)|2

energy spectrum f
-2/t -1/t 1/t 2/t
|P(f)|2
power spectrum f
-2/t -1/t 1/t 2/t
Definition: power spectrum = energy spectrum /T

In the above, T is the time span of a signal. For example, T=12t for a signal
contains 12 pulses as shown below. Note that this definition is somewhat
ambiguous as the zero part can also be counted in the signal. However, no matter
how to choose T, the shape of the power spectrum is the same as the energy
spectrum. Sometimes we will call it “normalized power spectrum” since T is just
a normalization constant. t

15
12t
 Example
power spectrum
s(t)
A


f
-t/2 0 t/2 t -2/t -1/t 1/t 2/t
t

power spectrum = energy spectrum/ time

For the above signal,
S ( f ) = At sinc( f t )
Therefore
power spectrum = energy spetrum/time = S ( f ) / t = A2t sinc2 ( f t )
2

We can compare this with the average power spectrum of a stream of pulses
(each with duration t) derived later. They are the same.
16
 Example
Example 2.8: The waveform s(t) shown in the left figure is passed through a
filter with the transfer function shown in the right figure.
s(t) 1 H(f)
2 volts
f

-2 msec
0
2 msec t
-250 Hz 250 Hz
low-pass filter
?
Find the magnitude spectra of this filter’s input and output.

Solution: Refer to the previous result. Substitute A = 2 volts, t = 4 msec into

S ( f ) = At sinc( f t ) = 0.008sinc(0.004 f )
The spectra of the filter’s input and output are shown below.

0.008 volts/Hz 1 H(f) 0.008 volts/Hz

f f f
-500 Hz -250 Hz 250 Hz 500 Hz -250 Hz 250 Hz -250 Hz 250 Hz
input low-pass filter output 17
 C. Additional Reading:
Properties of Fourier Transform

18
 Properties of Fourier Transform
Fourier transform has many interesting properties. These properties are very
important for signal analysis and for communication systems. EE3118 provides
detailed discussions on these properties, so we will only give some of the proofs.

The proofs of these properties are very mathematical. Please do not worry if you
have difficulty in understanding the proofs. I expect that you should gain better
understanding in EE3118.

The questions in Tutorial 1 are also designed to help you to understand the
properties of Fourier transform. Please go through these tutorial questions.

19
 Useful Operations
s(t)
A
Original signal: -a 0 b t

s(0.5t)
A
Time scaling: -2a 0 2b t

s(t-T)
A
Time delay:
T-a T T+b t
s(t)sin(2f0t)
A
Modulation: -a b t

A s(-t)
Mirror:
-b 0 a t

s(-t-T)
A
Delay and mirror: 20
-(T+b) -(T-a) t
 Some Important Properties
1. Time Shift: s(t-t)  S(f) exp(-j2ft)
2. Frequency Shift: s(t) exp(-j2f0t)  S(f+f0)
3. Linearity: s1(t) + s2(t)  S1(f)) + S2(f).
4. Modulation 1: s(t)cos(2f0t)  0.5[S(f-f0)+S(f+f0)].
Modulation 2: s(t)sin(2f0t)  -0.5j[S(f-f0)-S(f+f0)].
1  f 
5. Time scaling: s (at )  S 
|a|  a 
6. Symmetry: For real s(t), S(-f)=S*(f) and so |S(f)|=|S(-f)|.
 
The total energy E =
 s (t ) dt =  S ( f ) df
2 2
7. Parseval Theorem:
− −
8. Duality: S(t)  s(-f)

21
In the above, we assume that s(t)  S(f).
 Proof for Time Shift Property

F ( s(t − t )) = 
−
s (t − t )e − j 2 ft dt

Let x =t -t 
=  s ( x ) e − j 2 f ( x +t ) dx
−

=  s ( x ) e − j 2 fx  e − j 2 f t dx
−

=  s ( x ) e − j 2 fx dx  e − j 2 f t
−

= F ( s(t ))  e − j 2 f t

Therefore s(t-t)  S(f) exp(-j2ft)

22
 Proof for Frequency Shift Property

F ( s(t )e − j 2 f t ) =
0

−
s (t )e − j 2 f0t e − j 2 ft dt


=  s ( t ) e − j 2 ( f + f0 )t dt
−

=  s ( t ) e − j 2 ft dt Relacing f by f + f 0
−

= S ( f + f0 )

Therefore s(t)exp(-j2f0t)  S(f+f0).

23
 Proof for Linearity Property


− j 2 ft
F ( s1 (t ) + s2 (t )) = ( s1 (t ) + s 2 (t )) e dt
−
 

 
− j 2 ft
= s1 (t )e dt + s2 (t )e − j 2 ft dt
− −

= F ( s1 (t )) + F ( s2 (t ))

Therefore s1(t) + s2(t)  S1(f)) + S2(f).

24
 Modulation Property
How to draw s(t)cos(2f0t)?

s(t) A

-a 0 b t

cos(2f0t)

-a b t

s(t)cos(2f0t) A

-a b t
-A

25
 Proof for Modulation Property 1
(
Since cos(2 f 0t )) = e+ j 2 f0t + e− j 2 f0t / 2 , we have )

F ( s(t ) cos(2 f 0t )) =  s(t ) cos(2 f t )dt
−
0

( )
= 0.5  s ( x ) e + j 2 f0t + e − j 2 f0t e − j 2 ft dt
−
 

(
= 0.5  s ( x ) e − j 2 ( f − f 0 ) t
)dt + 0.5  ( s ( x ) e − j 2 ( f + f 0 ) t
)dt
− −
 
= 0.5  s ( x ) e − j 2 ft
dt Relacing f by f − f 0 + 0.5  s ( x ) e − j 2 ft dt Relacing f by f + f 0
− −

= 0.5S ( f − f 0 ) + 0.5S ( f + f 0 )

Therefore s(t)cos(2f0t)  0.5[S(f-f0)+S(f+f0)].

26
 Proof for Modulation Property 2
( )
Since sin(2 f 0t )) = e+ j 2 f0t + e− j 2 f0t / 2 j , we have

F ( s(t ) sin(2 f 0t )) =  s(t ) sin(2 f t )dt
−
0

( )
= −0.5 j  s ( x ) e + j 2 f0t − e − j 2 f0t e − j 2 ft dt
−

 

(
= −0.5 j   s ( x ) e − j 2 ( f − f 0 ) t
) ( )
dt −  s ( x ) e − j 2 ( f + f0 )t dt 
 − − 
 

= −0.5 j   s ( x ) e − j 2 ft
dt Relacing f by f − f0 −  s ( x ) e − j 2 ft dt Relacing f by f + f 0 
 − − 
= −0.5 j ( S ( f − f 0 ) − S ( f + f 0 ) )

Therefore s(t)sin(2f0t)  -0.5j[S(f-f0)+S(f+f0)].

27
 Spectrums of cos and sin Functions
Let s(t)=1 in the previous examples. As we have seen,
S(f)=d(f)
s(t)=1 A

0 f
t
We can write cos(2f0t)=1 cos(2f0t).Then using the modulation property, we get
the spectrums of cos and sin functions as follows
0.5 0.5 S(f)
cos(2f0t)  0.5(d(f-f0)+d(f+f0))
-f0 0 +f0 f
0.5j -0.5j S(f)
sin (2f0t)  -0.5j(d(f-f0)-d(f+f0))
-f0 0 +f0 f
The magnitude spectra of both cos and sin are the same. The above
transform is the foundation for modulation to be discussed later.
28
 Proof for Scaling Property

If a>0, we have F ( s(at )) = 
−
s (at )e − j 2 ft dt

Let x = at 

 s ( x) e
− j 2 f ( x / a )
= d ( x / a)
−

=  s ( x ) e − j 2 ( f / a ) x d ( x / a )
−

= (1/ a) S ( f / a)


Note that this is – for a <0.
If a<0, we have F ( s(at )) = 
−
s (at )e − j 2 ft dt

Let x = at −

 s ( x) e
− j 2 f ( x / a )
= d ( x / a)



 s ( x) e
− j 2 f ( x / a )
= d ( x / (−a ))
−

= (1/ | a |) S ( f / a)
29
 Proof for Symmetry Property
If s(t) is real, then

S (− f ) =  s ( t ) e − j 2 ( − f )t dt
−

=  s ( t ) e + j 2 ft dt
−

=  s ( t ) e − j 2 ft dt
−

= S( f )

Based on the above, we have |S(f)|=|S(-f)|. Note that this is true only when s(t)
is real. (Why?)

30
 Summary
You should learn Fourier analysis in more depth in an other module “Signal
Analysis”. In this course, our discussion on Fourier analysis is very brief. We
focus on its application in communication systems.

The properties of Fourier Transform are vital to all aspects in modern

engineering and science. We will see the applications of these properties
throughout your studying in university and your future engineering career.
Please try to understand them and try to memories them

A final remark: This is a course on “communication” rather than “signal

analysis”. Hence we are mostly interested in the principles about spectrum
but not the details. For example, we are interested in the rough “shape” of the
spectrum so that we can design a system accordingly. The exact mathematical
expression of the spectrum is usually less important.

-5000Hz -1500Hz 1500Hz 5000Hz f

31