ELE 635 Communication Systems

Course Notes

M. Zeytinoglu

Department of Electrical and Computer Engineering

Ryerson University
Toronto, Ontario, Canada M5B 2K3

Email: [email protected]

Winter 2013
Contents

1 Introduction 1
1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Analog vs. Digital Information Systems . . . . . . . . . . . . . . . . . . . . . . . 2

2 Signals and Systems: A Brief Review 5

2.1 Signal Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Some Useful Signal Characteristics and Operations . . . . . . . . . . . . . . . . . 6
2.3 Special Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3.1 Unit Impulse Signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3.2 The Unit Rectangular Pulse Function: Π(t) . . . . . . . . . . . . . . . . . 8
2.3.3 The Unit Triangular Pulse Function: Δ(t) . . . . . . . . . . . . . . . . . . 9
2.3.4 The “Sinc” Function: sinc(t) . . . . . . . . . . . . . . . . . . . . . . . . . 9

3 Analysis and Transmission of Signals 11

3.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.1.1 Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.1.2 Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.1.3 Laplace Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3 Fourier Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3.1 Fourier Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3.2 Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3.3 Fourier Transform of a Periodic Function . . . . . . . . . . . . . . . . . . 22
3.4 Properties of the Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.5 Bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.6 Signal Transmission Through a Linear System . . . . . . . . . . . . . . . . . . . . 32
3.7 Frequency Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.8 Distortionless Transmission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.8.1 Approximation for Distortionless Transmission . . . . . . . . . . . . . . . 36
3.8.2 Response of an Ideal Lowpass Filter . . . . . . . . . . . . . . . . . . . . . 38
3.9 Some Practical Considerations for Bandpass Systems . . . . . . . . . . . . . . . . 42
3.10 Signal Distortion Over a Communication Channel . . . . . . . . . . . . . . . . . . 44
3.10.1 Linear Distortion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.10.2 Non-Linear Distortion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.10.3 Multipath Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

i
ii CONTENTS

3.10.4 Fading Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

4 Sampling 47
4.1 Ideal Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.1.1 Spectrum of a Sampled Waveform . . . . . . . . . . . . . . . . . . . . . . 48
4.2 The Sampling Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.3 Signal Recovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.4.1 Anti-aliasing filter specifications . . . . . . . . . . . . . . . . . . . . . . . 54
4.4.2 Choosing the sampling rate . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.4.3 Sampling of bandpass signals . . . . . . . . . . . . . . . . . . . . . . . . 58
4.5 Non-Ideal Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.6 Pulse Code Modulation (PCM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

5 Amplitude Modulation 63
5.1 Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.2 Double Sideband Amplitude Modulation . . . . . . . . . . . . . . . . . . . . . . . 66
5.2.1 Coherent Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.3 Generation of AM Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.3.1 Non-linear Modulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.3.2 Switching Modulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.4 Amplitude Modulation (AM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.4.1 Sideband and Carrier Power . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.4.2 AM Broadcasting Standards . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.4.3 Generation of AM Signals . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.4.4 Demodulation of AM Signals . . . . . . . . . . . . . . . . . . . . . . . . 81
5.5 Quadrature Amplitude Modulation (QAM) . . . . . . . . . . . . . . . . . . . . . . 83
5.6 Single Sideband Modulation (SSB) . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.6.1 Representation of Single Sideband Signals . . . . . . . . . . . . . . . . . 86
5.6.2 Generation of Single Sideband Signals . . . . . . . . . . . . . . . . . . . . 89
5.6.3 Demodulation of Single Sideband Signals . . . . . . . . . . . . . . . . . . 91
5.7 Vestigial Sideband Modulation (VSB) . . . . . . . . . . . . . . . . . . . . . . . . 91
5.7.1 Generation of VSB Signals . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.7.2 Demodulation of VSB Signals . . . . . . . . . . . . . . . . . . . . . . . . 93
5.7.3 Choosing Hv (f ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.7.4 Further Comments on VSB Modulation . . . . . . . . . . . . . . . . . . . 95
Chapter 1

Introduction

ELE 635 Communication Systems is an introductory course on communication systems. The

course will focus on the transmission of information and will investigate techniques applicable
to the analysis and synthesis of analog communication systems. In particular, we will study:

• mathematical tools (e.g. Fourier analysis);

• transmission media (e.g. communication channels);

• basic transmission and reception techniques (e.g. amplitude modulation (AM), frequency
modulation (FM) ... );

• performance analysis of standard communication systems;

• implementation examples (e.g. AM/FM radio broadcasting ...).

We will study these concepts from a systems point of view. In the laboratory we will look at the im-
plementation of basic communication systems elements including modulation, envelope detection,
phase locked loop (PLL), PLL as an FM signal detector, etc.

1.1 Overview
Figure 1.1 presents a generic communications system. Throughout the course we will investigate








   



 
   

Figure 1.1: Communication system.

1
2 CHAPTER 1. INTRODUCTION

the transmitter (TX), channel and receiver (RX) triplet as shown in the block diagram. Basic
functionality of these components are as follows.
Input Message: signal containing the raw information that we want to transmit, e.g. human voice
(speech), image, ...
Input Transducer: converts the raw input signal into an electrical signal (voltage, current, ...).
We will refer to the resulting signal as the baseband/input/message signal.
Transmitter (TX): converts the input signal (electronically, mechanically, physically, ...) to a
format that best matches the channel characteristics (fibre optic cable, electronic wave guide,
coaxial cable, ...); the transmitter uses modulation and frequency translation techniques.
Channel: Physical medium over which the signal will travel. Here, in the channel, a lot of “ugly”,
“nasty” things (from the message point of view) will happen, e.g. distortion, noise, interfer-
ence, fading, etc.
Receiver (RX) and the Output Transducer: These components reverse the modifications im-
plemented by the transmitter and the input transducer, respectively.
Our goal is to transmit the information contained within the input signal to the destination such that
the information contained in the received signal is either identical (ideally) or closely resembles the
information in the input signal. And of course, we want to achieve this objective most efficiently,
economically and effectively.
In this course, we will study analog communication systems, whereas the 4th year elective
courses will introduce digital communication systems. However, as digital communication sys-
tems also use analog signals which in turn rely on analog communication techniques, this course
will indeed function as an introduction to universal communication system fundamentals.

1.2 Analog vs. Digital Information Systems

Analog: We transmit the information carrying signal in raw format (even after the modifications
applied by the input transducer and the transmitter). Hence, any distortion affecting the sig-
nal in the channel will be part of the received signal and will irreversibly distort the signal.
By choosing the TX and the RX appropriately we can minimize distortion effects, but they
will always be part of the signal.

x(t) y(t)
5.0 V 5.03 V


t t
t0 t0

Figure 1.2: Analog signal transmission over a communication channel.

Example 1.1: If the level of the transmitted waveform at time t0 equals 5 V, and the
1.2. ANALOG VS. DIGITAL INFORMATION SYSTEMS 3

noise and/or distortion changes this value to 5.03 V, the distorted value is what we will
see/hear/receive at the RX. There is no going back to the true voltage value of 5 V as the
noise distorting the signal is not known.

Digital: In this case we transmit a “representation” of the signal x(t). Once again let us consider
the value of the signal at t0 :
x(t0 ) = 5V =⇒ {5}decimal ≡ {0101}binary =⇒ { −A, +A, −A, +A },
where we assume that each signal value is represented by a 4-digit binary number; a {1}binary
is represented by a rectangular pulse of +A amplitude and a {0}binary is represented by a
rectangular pulse of −A amplitude. The waveforms representing x(t0 ) and the received
waveform y(t) at the output of distorting and noisy channels are shown in Figure (1.3).



 



 

   
x(t0 )

  
 


y(t)   



  
 

 
  

y(t)  
 


  

Figure 1.3: Digital signal transmission over a noisy communication channel.

As we know the format of the transmitted waveform (positive pulses representing {1}binary
... but not the order of positive and negative pulses), we can still correctly identify the “most
4 CHAPTER 1. INTRODUCTION

likely” pulse train which must have been transmitted. Hence by inspecting y(t) we can
conclude that
y(t0) = {0101}binary = {5}decimal = x(t0 ).
The main advantage of the digital communication systems is the symbolic representation of
the signal value. Theoretically (and practically) it is possible to receive and decode a digital
signal such that y(t) = x(t) . This statement is a direct consequence of the underlying fact
that there is only a finite number of possible signals that we can transmit (e.g. there are two
possible signal levels in the case of binary coding ... M possible signal levels in the case of
M-ary coding).

Example 1.2:

  
    

 
  



  
 
  
 

 




 
  
 
  
 





  
 
  
  

  

Figure 1.4: Analog vs. digital signal transmission.

In our discussion we will refer to certain fundamental concepts that characterize the behaviour and
performance of communication systems. These concepts include:

• Bandwidth B;

• Signal power, noise power and signal-to-noise ratio, SNR;

• Shannon’s channel capacity theorem which relates B and SNR: C = B log2 (1+SNR) bits/s;

• Randomness or uncertainty;

• Redundancy;

• Modulation;

• Multiplexing.

In the following weeks we will introduce these concepts, discuss their significance and will use
them in the analysis and design of communication systems.
Chapter 2

Signals and Systems: A Brief Review

In this course we will mostly work with communication systems which fit into the block diagram
template shown in Figure 2.1:


  

     
x(t) h(t) y(t)
X(s) H(s) Y (s)
X(f ) H(f ) Y (f )

Figure 2.1: Basic communication system template.

We assume that the channel is modelled as a linear, time-invariant (LTI) system. This assumption
is usually made to simplify and to introduce a degree of mathematical tractability to the analysis
of the underlying communications problem. It is empirically shown that indeed a large class of
communication system can be very accurately represented by appropriate LTI models. The ELE
532 Signal and Systems course introduced various linear system analysis tools: differential equa-
tions, difference equations, Laplace transform, Fourier transform, etc. In the subsequent semester,
the ELE 639 Control Systems course used the Laplace transform as its principal analysis tool.
Whereas in the ELE 635 Communication Systems course we will mostly use the Fourier transform.
The Fourier transform will allow steady-state, sinusoidal analysis of the underlying signals and
systems. This steady-state analysis is mostly sufficient for the study of communication systems;
we will shortly justify this assertion.

2.1 Signal Measure

Given the fact that most signals of interest are rapidly varying functions of time “t”, how can we
measure the “size” of a given waveform? Remember that the signal g(t) is not necessarily a scalar
quantity. For a dc signal such as g(t) = 5 V, the concept of the size of the signal is well understood.

5
6 CHAPTER 2. SIGNALS AND SYSTEMS: A BRIEF REVIEW

Even for a sinusoidal signal such as g(t) = A cos ω0 t we can unambiguously refer to the size of
signal by referencing the amplitude of the sinusoidal oscillations. On the other hand, we have
to introduce a new measure to compare the “size” of an arbitrary waveform g(t). One possible
measure is the signal energy defined as:
 ∞
Eg = |g(t)|2dt. (2.1)
∞

Observe that for an arbitrary, possibly complex-valued signal g(t)

|g(t)|2 = g(t)g ∗(t). (2.2)

If g(t) is real-valued then g(t) = g ∗ (t). Thus,

|g(t)|2 = g(t)g ∗(t) = g 2 (t). (2.3)

There is a large class of important signals for which Eg is not finite, namely power signals. For
these signals we will use the signal power as a measure of the “size” of the signal:

1 T /2
Pg = lim |g(t)|2dt. (2.4)
T →∞ T −T /2

For example, periodic signals are typically power signals. Please note that Eg and Pg do not
indicate actual energy and power levels since they depend not only on the signal but also on the
load. Hence, if g(t) is a voltage waveform, then Eg and Pg represent the unit energy or power
delivered across a 1-Ω resistor.

2.2 Some Useful Signal Characteristics and Operations

Periodic Signals: The waveform g(t) is periodic with period T0 if g(t) = g(t + kT0 ), for all t and
k ∈ Z (the set of all integers: 0, ±1, ±2, . . .).

Energy Signals: The waveform g(t) is an energy signal if 0 ≤ Eg < ∞.

Power Signals: The waveform g(t) is a power signal if 0 ≤ Pg < ∞. Please note that a signal
cannot simultaneously be both an energy and a power signal.

Time shifting: g(t) −→ g(t − t0 ):

g(t) g(t − t0 )

b b + t0
t t
a a + t0

Figure 2.2: Time shifting operation.

2.3. SPECIAL FUNCTIONS 7

α<1 g(αt)
g(t)

t
a/α b/α
t
a b

β>1 g(βt)

t
a/β b/β

Figure 2.3: Time scaling operation.

Time Scaling: g(t) −→ g(αt) with α ∈ R. If α > 1, then g(αt) represents a time-compressed
(by a factor of α) version of g(t), whereas if α < 1, then g(αt) represents a time-stretched
(by a factor of α) version of g(t):
Time Reversal: g(t) −→ g(−t) represents a mirror image of g(t) with respect to the vertical axis
t = 0. Similarly, g(t0 − t) with t0 ∈ R first takes the mirror image of g(t) with respect to
the vertical axis and then shifts the resulting waveform by t0 units.

2.3 Special Functions

2.3.1 Unit Impulse Signal
The unit impulse signal δ(t) is one of the most useful signals that will allow us to formulate many
operations frequently encountered in the analysis of LTI systems. The unit impulse signal δ(t)
(also known as the Dirac-delta function, named after the quantum mechanist P.A.M. Dirac) has
zero amplitude everywhere except at t = 0 where its magnitude is infinitely large such that the
area underneath is unity. As such δ(t) belongs to the category of generalized functions.

Definition: The unit impulse signal δ(t) is defined as:

 ∞
δ(t) = 0, if t = 0; and δ(t)dt = 1. (2.5)
−∞

The Dirac-delta function is normally represented as a limiting case obtained by considering the
limit of a regular function, e.g. δ(t) = limτ →0 rect(t/τ ), where rect(t/τ ) is a rectangular pulse of
τ -duration centered at the origin.
Important Properties:

1. Symmetry: The δ-function has even symmetry, i.e., δ(t) = δ(−t).

8 CHAPTER 2. SIGNALS AND SYSTEMS: A BRIEF REVIEW

2. For an arbitrary function x(t) and constant K ∈ we have:


δ(t)x(t)dt = x(0);

δ(t − t0 )x(t)dt = x(t0 ), (the sifting property);

Kδ(t)dt = K.

3. Convolution with the δ-function:


δ(t − t0 ) ∗ x(t) = δ(λ − t0 )x(t − λ)dλ,

= x(t − λ)|λ=t0 ,
= x(t − t0 ).

δ(t − t0 ) x(t) x(t − t0 )

b b + t0
t t t
t0 a a + t0

Figure 2.4: Convolution with a δ-function.

2.3.2 The Unit Rectangular Pulse Function: Π(t)

We will use the notation Π(x)1 to refer to the unit rectangular pulse function of unit amplitude and
unit width, centered at the origin:
⎧ 1
⎨ 1, |t| < 2 ;
⎪
1
Π(t) = , |t| = 12 ; (2.6)
⎪
⎩
2
0, |t| > 12 .

Frequently, we will use the scaled version of the unit rectangular pulse as Π(t/τ ) where the scaling
parameter τ will represent the width of the pulse centered at origin.
1
We will also use the notation rect(t) interchangeably to represent the unit rectangular pulse function Π(t).
2.3. SPECIAL FUNCTIONS 9

Π(t/τ )

t
− τ2 τ
2

Figure 2.5: The rectangular pulse function Π(t/τ ).

2.3.3 The Unit Triangular Pulse Function: Δ(t)

The unit triangular pulse Δ(t) is a triangular shaped pulse of unit height and unit amplitude cen-
tered at the origin: 
1 − 2|t|, |t| ≤ 12 ;
Δ(t) = (2.7)
0, |t| > 12 .

Δ(t/τ )

t
− τ2 τ
2

Figure 2.6: The triangular pulse function Δ(t/τ ).

2.3.4 The “Sinc” Function: sinc(t)

Another function that we will frequently use is the sinc function defined as:
sin(t)
sinc(t) = . (2.8)
t

sinc(t)

−4π −3π −2π −π
π 2π 3π 4π

Figure 2.7: The function sinc(t).

Inspection of Equation(2.8) reveals the following key characteristics of the sinc function:
10 CHAPTER 2. SIGNALS AND SYSTEMS: A BRIEF REVIEW

• sinc(t) has even symmetry, i.e., sinc(t) = sinc(−t).

• sinc(t) = 0 when sin(t) = 0 except at t = 0 where it is indeterminate. This means that

sinc(t) = 0 for t = ±π, ±2π, ±3π, . . . .

• sinc(0) = 1. (As the denominator of sin(t)/t equals to 0 when t = 0, sinc(0) can be deter-
mined using the L’Hôpital’s rule.

• sinc(t) is the product of an oscillating signal (sin t) of period 2π and a monotonically de-
creasing function (1/t). Therefore, sinc(t) is an oscillating function with even symmetry
and decreasing amplitude. It has a unit amplitude at t = 0 and zero crossings at integer
multiples of π.
Chapter 3

Analysis and Transmission of Signals

3.1 Preliminaries

3.1.1 Fourier Series

Given a periodic waveform xp (t) with period T0 (and the corresponding fundamental frequency
f0 = 1/T0 such that ω0 = 2πf0 ) we can expand xp (t) in a Fourier series:


∞
xp (t) = Dn ejnω0 t (3.1)
n=−∞

where the coefficients of the Fourier series are evaluated as:


1
Dn = xp (t)e−jnω0 t dt. (3.2)
T0 T0

We will refer to the expansion of the periodic waveform xp (t) in terms of harmonically-related
complex exponentials as shown in Equations (3.1–3.2) as the complex Fourier series representation
of xp (t). In general, the {Dn } coefficients of the complex Fourier series expansion will be complex
valued. We will frequently use the nottaion:

xp (t) ⇐⇒ {Dn }n (3.3)

to represent the equivalence between the time domain (the function xp (t) itself) and the corre-
sponding Fourier domain representation (the {Dn } coefficients). In many cases (particularly, if the
waveform xp (t) has even or odd symmetry) we may find it convenient to expand xp (t) in terms of
an equivalent trigonometric Fourier series given as:


∞
xp (t) = a0 + (an cos nω0 t + bn sin nω0 t), (3.4)
n=1

11
12 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

with

1
a0 = xp (t)dt, (3.5)
T0 T0

2
an = xp (t) cos nω0 t dt, n = 1, 2, . . . (3.6)
T0 T0

2
bn = xp (t) sin nω0 t dt, n = 1, 2, . . . (3.7)
T0 T0

The integrals used in the evaluation of the complex or trigonometric Fourier series coefficients can
be taken over any interval of T0 duration as both the waveform xp (t) and the complex/trigonometric
basis functions of the Fourier series are periodic with period T0 . Also note that the complex Fourier
series expansion, Equations (3.1–3.2), and the trigonometric Fourier series expansion, Equations
(3.4–3.7) are equivalent and one can be easily converted into the other.

3.1.2 Fourier Transform

Let x(t) be a (possibly non-periodic) function satisfying the Dirichlet conditions1. We can deter-
mine the Fourier transform of x(t) to be represented by X(f ) using the Fourier integral:

X(f ) = F [ x(t) ], (3.8)

 ∞
= x(t)e−j2πf t dt, (3.9)
−∞

where ω = 2πf is the angular frequency measured in [rad/s]. We will also use the notation:
F
x(t) −→ X(f ),

to represent the origin/source of the Fourier transform X(f ). We can “recover” x(t) from its
Fourier transform X(f ) by using the inverse-Fourier transform operation:

x(t) = F −1[ x(t) ], (3.10)

 ∞
= X(f )ej2πf t df. (3.11)
−∞

We will also use the notation:

F −1
x(t) ←− X(f ),
to represent how we recover x(t) from its Fourier transform X(f ). Equivalently, we will use the
symbolic representation:
x(t) ⇐⇒ X(f ),
to show that x(t) and X(f ) form a Fourier transform pair.
1
The Dirichlet conditions required for the existence of the Fourier transform are: (1) The function x(t) must be
∞
absolutely integrable, i.e., −∞ |x(t)|dt < ∞, and (2) x(t) may have only a finite number of maxima and minima and
a finite number of discontinuities.
3.2. MOTIVATION 13

Remark: The definition of the Fourier transform (and the corresponding inverse Fourier trans-
form) can take many different forms. You may frequently find the normalization term 1/2π pre-
ceding the Fourier integral such that the Fourier and inverse-Fourier transforms are defined as:
 ∞
1
X(ω) = x(t)e−jωt dt,
2π −∞
 ∞
x(t) = X(ω)ejωtdω.
−∞

The above representation of the Fourier transform arises from the use of ω as the frequency variable
instead of f . Irrespective of which frequency variable is used, both representations are equivalent—
using substitution of variables with ω = 2πf and dω = d(2πf ) we can easily establish the equiv-
alence
√ of these representations. In other definitions of the Fourier transform you may even see
1/ 2π as the normalization factor symmetrically distributed to the definitions of both the forward
and inverse transforms.

3.1.3 Laplace Transform

We will use the notation X(s) = L[ x(t) ] to refer to the Laplace transform of the function x(t)
such that
x(t) ⇐⇒ X(s),
will form a Laplace transform pair.

3.2 Motivation
In this course our objective is the sinusoidal, steady-state analysis of signals and systems with
particular emphasis on communication systems. In particular, we assume that the systems we will
study are linear, time-invariant (LTI) systems. This is a simplifying assumption that nevertheless
allows us to model and study a wide range of practical communication systems with a great degree
of accuracy.

Let h(t) be the impulse response function of an Nth order LTI system with the transfer function
H(s) = L[ h(t) ]. Let x(t) be the input and y(t) be the corresponding output of the system. Then

y(t) = h(t) ∗ x(t).

Let x(t) = A sin ω0 t such that

Aω0
X(s) = ,
s2 + ω02
with
N(s)
H(s) = ,
(s + p1 ) · · · (s + pN )
where N(s) is the Laplace transform of the numerator of H(s). Assume the system poles are
simple, i.e., each has multiplicity 1, real-valued and pn < 0, for n = 1, 2, . . . , N to ensure the
14 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

stability of the system. Then,

Y (s) = H(s) X(s), (3.12)

N(s) Aω0
= , (3.13)
(s + p1 ) · · · (s + pN ) s + ω02
2

K0 K0∗ N
Kn
= + + (3.14)
s + jω0 s − jω0 n=1 s + pn

By evaluating the inverse Laplace transform of Y (s) given in Equation (3.14) we obtain the fol-
lowing expression for the output waveform:


N
−jω0 t
y(t) = K0 e + K0∗ ejω0 t + Kn e−pn t , (3.15)
n=1

N
= K sin(ω0 t + ϕ) + Kn e−pn t , (3.16)
n=1

where K = A|H(ω0)| and ϕ = arg[ H(ω0 ]. Since H(s) represents a stable system, we have:

lim Kn e−pn t = 0, for n = 1, 2, . . . , N.

t→∞

Therefore, the steady-state output of the system represented by yss (t) is described by the expres-
sion:

yss (t) = lim y(t),

t→∞
= K sin(ω0 t + ϕ),
= A|H(ω0)| sin ω0 t + arg[ H(ω0) ] .

Observations:

1. For a LTI system defined by the transfer function H(s), if x(t) is a sinusoid at frequency
ω0 , the steady-state system output yss (t) is also a sinusoid at the same frequency but with
amplitude and phase values modified by H(s) evaluated at s = jω0 . This approach forms
the basis for most tests used to determine if a system described by H(s) is linear.

2. If we restrict our analysis to sinusoidal steady-state analysis we do not need to know H(s)
over the entire s-plane since we only need to evaluate H(s) for s = jω, i.e., along the jω-
axis. In other words, all we need is the Fourier transform of the impulse response function
h(t) as
F [ h(t) ] = HF (f ) = H(s) s=j2πf ,

where we used the notation HF (f ) to explicitly refer to the Fourier transform of h(t).
3.2. MOTIVATION 15

Im[s]

HF (f ) = H(s)s=j2πf

Re[s]
H(s)

Figure 3.1: How to obtain the Fourier transform from the Laplace transform.

3. If we know HF (f ) then we can easily answer the the question: “What is the system output
when the input is A sin ω0 t?”. Based on the result stated above we can first evaluate HF (f )
at f0 and then express the steady-state system output as:

y(t) = A|HF (f0 )| sin 2πf0 t + arg[ HF (f0 ) ] .

This result is very useful if we work all the time with sinusoidal signals only. But simple
sinusoids are not very exciting signals. Do we really want to listen to a radio broadcasting
consisting of test tones, i.e., single frequency sinusoids, only? As most signals of interest
are highly complex, we may want to extend this important result to a much wider class of
complex signals.

4. The following table provides an overview of how we can approach to the problem of deter-
mining the steady-state output of a LTI system when the input is an arbitrary signal.

Input Signal Output Signal

x1 (t) y1 (t)
x2 (t) y2 (t)
αx1 (t) + βx2 (t) αy1 (t) + βy2 (t)
A sin(ω0 t + θ) A|HF (f0 )| sin ω0 t + θ + arg[ HF (f0 ) ]
ejω0 t |HF (f0 )|ej(ω0 t+arg[ HF (f0 ) ])
xp (t) = n Dn ejnω0 t yp (t) = n Dn |HF (nf0 )|ej(nω0 t+arg[ HF (nf0 ) ])
x(t) y(t) = F −1 [ HF (f )XF (f ) ] = HF (f )XF (f )ejωt df

Observe that the input-output relation shown in the third line is due to the linearity of the
system; xp (t) in the second last line is an arbitrary periodic function expanded in complex
Fourier series, and the corresponding system output yp (t) is obtained again using the linearity
of the system. And the last line generalizes the input-output relation to an arbitrary (non-
periodic) input signal and expresses the system output y(t) in terms of the system and input
Fourier transforms.
16 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

Thus, the Fourier analysis is a convenient and very powerful tool; it will allow us to deter-
mine the output of LTI systems excited by arbitrary input signals.

5. The Fourier representation (Fourier series for periodic and Fourier transform for non-periodic
signals) decomposes a waveform into a discrete or continuous sum of complex exponen-
tials. We also recall that for real-valued waveforms positive and negative frequencies are
by-products of the complex exponential representation. Negative frequencies have no phys-
ical meaning and a 2-sided spectrum of a real-valued signal can always be represented as a
1-sided spectrum.

6. The time-domain and frequency-domain representations of a signal are comparable to a rep-

resentation of an image in different file formats, e.g. TIFF, GIF, PNG, Raw, etc.; they all rep-
resent the same image and carry the same image information. Furthermore, one can transfer
one file format into another by using an appropriately designed file format converter, i.e.,
transformation.

7. The relative amplitude of frequency components at f  is proportional to |X(f )| (or if we

are using a Fourier series representation |Dn | where n f0 = f  ). The units of a Fourier
representation is determined by the waveform itself. Let the notation [ a ]u represent the
units associated with the signal a. If x(t) is a time waveform with [ x(t) ]u = gnats, then the
corresponding Fourier series coefficients have [ Dn ]u = gnats as x(t) = n Dn ejnω0 t and
[ X(f ) ]u = gnats/Hz since X(f ) = limT0 →∞ (T0 Dn ).

Questions: Why are we interested in steady-state , i.e. Fourier analysis? What happens to the
transients? Are the transients not important?

• Yes, we can ignore transients: For a stable LTI system, the transient response approaches to
zero as t → ∞, and the system output approaches the forced response. Contrary to analysis
and design of control systems, we can assume that the communication system has been in
operation for a long time such that all transient signals can be ignored . If you compare this
approach with that taken in a control system course, we understand why such a difference
exists. The very nature of a control system means that we initiate an action to perform a
task, hence we want to study the behaviour of the system as the action designed to generate
a certain response is applied to the system. In a communication system, however, this is
seldom the case.

In a typical communication system the duration of transients is very short relative to the
signal duration. Therefore, we can safely ignore the transients if they do not create problems
during the initial start-up phase.

• No, we cannot ignore transients: During the initial analysis, design and testing stages we
have to consider the effects of transients to ensure that they do not have any detrimental
effects on the system performance, e.g. dynamic behaviour of PLL circuits.
3.3. FOURIER ANALYSIS 17

3.3 Fourier Analysis

In this section we discuss important characteristics of the Fourier analysis and point out how the
Fourier transform can be considered a limiting case of the Fourier series representation. The dis-
cussion presented in these notes does not present a rigorous derivation of important results; rather,
it demonstrates key concepts by examples and observations. These notes should be used as a
supplement to the course reference text.

3.3.1 Fourier Series

Let xp (t) be a periodic function with period T0 seconds which also satisfies the Dirichlet conditions.
Therefore, we can expand xp (t) in a complex Fourier series with coefficients {Dn }. As an example
let us consider a periodic train of τ second wide rectangular pulses with amplitude A.

xp (t)

A









−T0 − T20 − τ2  τ
2
T0
T0
2

Figure 3.2: The periodic pulse train xp (t).

We can expand xp (t) in a complex Fourier series n Dn ejnω0 t with ω0 = 2π/T0 such that:

1 T0 /2
Dn = xp (t)e−jnω0 t dt,
T0 −T0 /2
 τ /2
A
= e−jnω0 t dt,
T0 −τ /2

τ /2
A e−jnω0t
= ,
T0 −jnω0
−τ /2

A  e−jnω0τ /2 − ejnω0 τ /2 
= ,
T0 −jnω0
2A  ej(.) − e−j(.) 
= ,
nω0 T0 2j
A  πτ 
= sin n ,
nπ T0
τ  πτ 
= A sinc n .
T0 T0
18 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

Thus given the Fourier series representation

 
τ  πτ 
xp (t) ←→ A sinc n (3.17)
T0 T0
n

a graphical depiction of the spectrum of xp (t) is shown in Figure (3.3).

A Tτ0

f
 2π 

× T0

Figure 3.3: Spectrum of xp (t).

Remarks and Observations:

• Magnitude and Phase Spectra: We can separate the information contained in {Dn } into a
corresponding magnitude spectrum { Dn } and a phase spectrum {arg[Dn ]}. In this example
all Dn ’s are real valued; some are positive and some are negative. Observe that if α+ is a
positive real number, then arg[α+ ] = arg[|α+ |ej0] = 0. Conversely, if α− is a negative real
number, then arg[α− ] = arg[|α− |e±jπ ] = ±π.

 


π f

−π


 


Figure 3.4: Magnitude and phase spectra of xp (t).

• Where did the negative frequencies come from? Consider a real-valued periodic signal
signal xp (t) with the Fourier series expansion

xp (t) = Dn ejnω0 t = · · · + D−n e−jnω0 t + Dn ejnω0 t + · · · (3.18)
n
3.3. FOURIER ANALYSIS 19

with D−n = Dn∗ . Let Dn = Dn ejφn where ϕn = arg[Dn ]. Observing that D−n = Dn e−jφn ,
we can reformulate the two terms on the right-hand side of Equation( 3.18 ) as:
D−n e−jnω0t + Dn ejnω0 t = Dn∗ e−jnω0t + Dn ejnω0 t (3.19)
= Dn e−j(nω0 t+ϕn ) + Dn ej(nω0 t+ϕn ) , (3.20)
 jnω0 t 
e + e−jnω0 t
= 2 Dn , (3.21)
2
= 2 Dn cos(nω0 t + ϕn ). (3.22)
Thus, any component with a negative frequency −nω0 can be combined with its counter-
part term at the positive frequency nω0 to yield the sinusoidal component cos(nω0 t + φn ).
Hence, the negative frequency components are essential; they simply result from the com-
plex Fourier series expansion of xp (t). If we had expanded xp (t) into a trigonometric Fourier
series then the combined terms as shown in Equation (3.22) would have already been built
in into the calculation of the {an } and {bn } coefficients of the trigonometric Fourier series
expansion.
• One-sided √ rms-spectrum: Let x(t) be the sinusoidal signal A cos ω0 t with the rms value
xrms = A/ 2. The expansion of x(t) into a complex Fourier series can be easily accom-
plished using the Euler’s formula such that
A jω0 t A −jω0 t
x(t) = e + e .
2 2

Spectrum analyzers (such as the ones we will be using in the laboratory) typically display
one-sided rms-spectra. For an arbitrary periodic function xp (t) with complex Fourier series
coefficients {Dn } we have the Fourier series expansion:

∞
xp (t) = Dn ejnω0 t ,
n=−∞
∞ 
 
= |D0 | + D−n e−jnω0 t + Dn ejnω0 t ,
n=1

∞
= |D0 | + 2|Dn | cos(nω0 t + ϕn ),
n=1

where as before ϕn = arg[Dn ]. Therefore, the rms values corresponding to each term in the
above Fourier series expansion are calculated as:
 
|D0 | rms = |D0 |,
  √
2|Dn | rms = 2|Dn |/ 2,
√
= 2|Dn |.
The two-sided magnitude spectrum of xp (t) together with the one-sided rms-spectrum is
shown in Figure (3.5).
20 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

|D0 | |D0 |
√
2|D1 |
|D−1 | |D1 | √
2|D2 |
|D−2 | |D2 |










f f

[ × f0 ]
[ × f0 ]




Figure 3.5: (a) Two-sided magnitude spectrum and (b) one-sided rms-spectrum of xp (t).

3.3.2 Fourier Transform

A non-periodic signal can be obtained from a periodic signal by letting the period of the signal
approach to infinity. Consequently, we can show that the Fourier transform of such a non-periodic
function can be derived by observing the behaviour of the Fourier series coefficients of a cor-
responding periodic function as the waveform period is increased. To illustrate this point let us
consider the following example. Let xp (t) be a periodic train of rectangular pulses of τ second
duration, amplitude A and period T0 . In Section 3.3.1 we presented xp (t) and shown that the
corresponding Fourier series coefficients are given by the expression:
Aτ  πτ 
Dn = sinc n , n = 0, ±1, ±2, . . . (3.23)
T0 T0
Now consider the corresponding non-periodic function x(t) = AΠ(t/τ ). Its Fourier transform
X(f ) can be easily calculated as:
 ∞
t
X(f ) = A Π( )e−j2πf t dt,
−∞ τ
 τ /2
=A e−j2πf t dt,
−τ /2

τ /2
e−j2πf t
=A ,
−j2πf
−τ /2

A  ejπf τ − e−jπf τ 
= ,
πf 2j
sin(πf τ )
=A ,
πf
= Aτ sinc(πf τ ).
This result establishes the Fourier transform pair:
t
A Π( ) ⇐⇒ Aτ sinc(πf τ ). (3.24)
τ
3.3. FOURIER ANALYSIS 21

A Π(t/τ ) Aτ sinc(πf τ )
Aτ
A

t f
− τ2 τ
2 − τ1 1
τ

Figure 3.6: The rectangular pulse function AΠ(tτ ) and its Fourier transform.

As we compare the Fourier series coefficient given in Equation (3.23) and the Fourier transform
in Equation (3.24) we observe that

∞
n
lim T0 Dn δ(f − ) = X(f ) (3.25)
T0 →∞
n=−∞
T0

when we recognize that the Dn coefficients located at discrete frequency locations n/T0 will con-
verge to the continuous frequency variable f as T0 → ∞. The relation established in Equation
(3.25) holds in general and indicates how the Fourier series and transform results are related.

Properties of the δ-function allow us to determine some useful Fourier transform pairs:

F [δ(t)] = δ(t)e−j2πf t dt = e−j2πf t |t=0 = 1. (3.26)

−1
F [δ(f )] = δ(f )ej2πf t df = ej2πf t |f =0 = 1. (3.27)

We can now amend our Fourier transform tables by adding the transform pairs:

δ(t) ⇐⇒ 1 (3.28)
1 ⇐⇒ δ(f ) (3.29)

We can continue with the computation of other useful Fourier transform pairs by using the above
results and Fourier transform properties:

−1
F [δ(f − f0 )] = δ(f − f0 )ej2πf t df = ej2πf0 t , (3.30)

F [δ(f + f0 )] = δ(f + f0 )ej2πf t df = e−j2πf0 t ,
−1
(3.31)
1  1
F −1 δ(f − f0 ) + δ(f + f0 ) = ej2πf0 + e−j2πf0 (3.32)
2 2
= cos 2πf0 t, (3.33)

which results in the following entry in the Fourier transforms table:

1 
cos 2πf0 t ⇐⇒ δ(f + f0 ) + δ(f − f0 ) . (3.34)
2
22 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

Similarly, by first expressing sin 2πf0 t as (ej2πf0 t − e−j2πf0 t )/2j we can show that:
j 
sin 2πf0 t ⇐⇒ δ(f + f0 ) − δ(f − f0 ) . (3.35)
2

1
2

 


cos 2πf0 t     

sin 2πf0 t
π
2

  


   
− π2

Figure 3.7: Fourier series expansion of a periodic train of δ-functions.

As expected the Fourier transforms of the trigonometric functions cos and sin have the same mag-
nitude but different phase spectra.
Observe that τ in the definition of the single pulse function x(t) defines the pulsewidth. And the
first zero crossing of the corresponding Fourier transform is at f = 1/τ . Thus, if we consider the
main lobe of the sinc function, i.e., [0, 1/τ ] as the range of frequencies in x(t), we observe that
this bandwidth is inversely proportional to τ and thus it increases with decreasing τ . This result is
expected since decreasing τ values describe shorter pulse duration (in time-domain), and of course
shorter duration pulses imply larger signal bandwidth (in frequency-domain). This reciprocal re-
lation between signal duration and signal bandwidth will be one of the important relations in the
analysis and design of communication systems and protocols.
AΠ(t/β)
Aα
AΠ(t/α) A
Aβ

t f
α β β α
− − 1 1
2 2 2 2
α β

Figure 3.8: Changes in the Fourier transform of Π(tτ ) as a function of the pulse width.

3.3.3 Fourier Transform of a Periodic Function

Let xp (t) be a periodic function with period T0 . xp (t) has the Fourier series expansion {Dn }n
where 
1
Dn = xp (t)e−j2πnt/T0 dt.
T0 T0
3.3. FOURIER ANALYSIS 23

Let us also consider x(t) as a non-periodic signal which is equivalent to one period of xp (t) as
shown in Figure (3.9). We observe that

xp (t)

t
-T0 0 T0

x(t)

t
0

Figure 3.9: Extraction of x(t) from xp (t).


xp (t), |t| ≤ T0 /2;
x(t) = (3.36)
0, otherwise;

and xp (t) = n x(t − nT0 ). Since x(t) is a non-periodic function we can compute its Fourier
transform as  ∞
X(f ) = x(t)e−j2πf t dt.
−∞
But we also have

1 T0 /2
Dn = xp (t)e−j2πnt/T0 dt,
T0 −T0 /2

1 T0 /2
= x(t)e−j2πnt/T0 dt,
T0 −T0 /2
 ∞
1
= x(t)e−j2πnt/T0 dt,
T0 −∞

1 n
= X( ).
T0 T0
Using the linearity of the Fourier transform we can also write
   
F xp (t) = F Dn ej2πnt/T0 , (3.37)
n

= Dn δ(f − nf0 ), (3.38)
n

1  n
= X( ) δ(f − nf0 ). (3.39)
T0 n T0
24 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

Example 3.1: Let xp (t) be a periodic train of δ-functions with unit amplitude defined as: xp (t) =
n δ(t − nT0 ), where T0 is the period, f0 = 1/T0 is the fundamental frequency and ω0 =
2πf0 . As xp (t) is a periodic function, it can be expanded in an exponential Fourier series
with coefficients:

1
Dn = xp (t)e−jn2πf0 t dt,
T0 T0

1
= δ(t)e−jn2πf0 t dt,
T0 T0
1
= .
T0
Thus, the Fourier series expansion of xp (t) is also a train of δ-functions with constant am-
plitude 1/T0 uniformly spaced at integer multiples of the fundamental frequency f0 .

1
T0

t f
    [× T0 ]
     [× f0 ]

Figure 3.10: Fourier series expansion of a periodic train of δ-functions.

Example 3.2: Let xp (t) be a periodic train of rectangular pulses of τ second duration, amplitude
A, period T0 and Fourier series coefficients:
Aτ  πτ 
Dn = sinc n , n = 0, ±1, ±2, . . .
T0 T0
Let x(t) = AΠ(t/τ ) be the non-periodic waveform extracted from xp (t) with the Fourier
transform
X(f ) = Aτ sinc(πf τ ).
Using these previously established results and the relation presented in Equation (3.39) we
can express the Fourier transform of xp (t) as:
  1  n
F xp (t) = X( ) δ(f − nf0 ),
T0 n T0
Aτ  πnτ
= sinc( ) δ(f − nf0 ).
T0 n T0

Example 3.3: Let xp (t) = n δ(t − nT0 ) be a periodic train of unit impulse functions spaced T0
seconds apart. To determine its Fourier transform Xp (f ), we first isolate one period of xp (t)
3.4. PROPERTIES OF THE FOURIER TRANSFORM 25

such that x(t) = δ(t) which has the Fourier transform X(f ) = F [δ(t)] = 1. Therefore,
  1  n
F xp (t) = X( ) δ(f − nf0 ),
T0 n T0
1 
= δ(f − nf0 ).
T0 n

Hence, the Fourier transform of a periodic train of unit impulse functions spaced T0 seconds
apart is also a train of impulse functions scaled in amplitude by 1/T0 and spaced f0 = 1/T0
Hz apart. This is an important result that we will refer to when discussing sampling of analog
waveforms. Example 2.1 illustrates this result and depicts the Fourier transform of a periodic
train of impulse functions in Figure (3.10).

3.4 Properties of the Fourier Transform

There are many useful properties of the Fourier transform. You have seen these properties (and
their variants in the context of different frequency transformations such as the Laplace transform).
We will not discuss the derivation of these properties in any detail, it will be your responsibility
to read and understand them. Please refer to Section 3.3 of the course reference text for a detailed
discussion, and to Table 3.2 Properties of Fourier Transform Operations presented on page 123
of course reference text. We will only discuss the frequency-shifting/modulation property as
modulation is the basic operation that underlies almost all communication systems, as in amplitude
modulation (AM) and frequency modulation (FM).

Frequency-Shifting property
Let x(t) be a waveform with the Fourier transform X(f ). Then

x(t) ej2πf0 t ⇐⇒ X(f − f0 ). (3.40)

Proof:

  ∞
F x(t) e j2πf0 t
= x(t) ej2πf0 t e−j2πf t dt, (3.41)
−∞
 ∞
= x(t) e−j2π(f −f0 )t dt, (3.42)
−∞

= X(f − f0 ). (3.43)

Thus, the time-domain multiplication of x(t) with the complex exponential ej2πf0 t will result in
frequency-shifting of X(f ) such that it will be centered on f0 . Similarly, if were to multiply x(t)
with e−j2πf0 t then X(f ) will be shifted in frequency and centered on −f0 . We can now use this
property together with the linearity of the Fourier transform to determine the effect of multiplying
26 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

a signal with a sinusoid. Using Euler’s identity we first expand cos 2πf0 t = (ej2πf0 t + e−j2πf0 t )/2
such that:
 ∞
 
F x(t) cos 2πf0 t = x(t) cos 2πf0 t dt, (3.44)
−∞
 ∞
1
= x(t) ej2πf0 t + e−j2πf0 t e−j2πf t dt, (3.45)
2 −∞
 ∞
1
= x(t) e−j2π(f −f0 )t + x(t) e−j2π(f +f0 )t dt, (3.46)
2 −∞

1 
= X(f − f0 ) + X(f + f0 ) . (3.47)
2
Similarly, the effect of multiplying x(t) with sin 2πf0 t = (ej2πf0 t − e−j2πf0 t )/2j becomes:
  1  
F x(t) sin 2πf0 t = X(f − f0 ) − X(f + f0 ) , (3.48)
2j
1 
= X(f − f0 )e−jπ/2 + X(f + f0 )ejπ/2 . (3.49)
2
Of course, we could have derived these results using the Fourier transforms of sinusoids shown in
Equations (3.34–3.35) together with the frequency convolution property, F [x1 (t)x2 (t)] = X1 (f ) ∗
Xf (f ). Figure (3.11) demonstrates the effects of multiplying a signal with a sinusoid. If the spec-
trum of the signal x(t) covers the frequency band [−B, B], then F [x(t) cos ω0 t] and F [x(t) sin ω0 t]
are both centered at ±f0 and occupy the frequency bands [−B −f0 , −f0 +B] and [f0 −B, f0 +B].
Frequency-shifting the signal up and down the frequency band is known as modulation.

x(t)
X(f )
A

t f
0 0 B
x(t) cos ω0 t F[ x(t) cos 2πf0 t ]
A/2
t f

2B
x(t) sin ω0 t F[ x(t) sin 2πf0 t ]
π/2
t f
−f0 f0
−π/2

  
 

Figure 3.11: Demonstration of the frequency-shifting property.

3.5. BANDWIDTH 27

Example 3.3: We now want to extend the preceding discussion to demonstrate how we can
combine two signals for transmission over a communication channel. Let x1 (t) and x2 (t) be
two waveforms with spectra X1 (f ) and X2 (f ), respectively, and let y(t) = x1 (t) cos 2πf1 t+
x2 (t) cos 2πf2 t. We want to determine any constraints on the modulating frequencies f1 and
f2 such that the spectra of X1 (f ) and X2 (f ) can be recovered from Y (f ) = F [y(t)], i.e., we
should be able to identify the spectra of the constituent signals within Y (f ).

X1 (f ) cos 2πf1 t

x1 (t)
f
−B1 B1
y(t)
X2 (f )
x2 (t)

f
−B2 B2 cos 2πf2 t

Figure 3.12: Combining signals per Example 3.3.

The first step of our analysis will be to sketch the spectrum of the composite signal Y (f ),
identify all critical frequencies and attempt to formulate constraints on the modulating fre-
quencies f1 and f2 . In particular, a key consideration for identifying X1 (f ) and X2 (f ) in
Y (f ), is that the frequency-shifted versions of X1 (f ) and X2 (f ) should not overlap:

Case 1: f1 > f2
f2 + B2 ≤ f1 − B1 =⇒ f1 − f2 ≥ B1 + B2 ;

Case 2: f1 < f2
f1 + B1 ≤ f2 − B2 =⇒ f2 − f1 ≥ B1 + B2 .

Combining these observation into one inequality yields the result:

f1 − f2 ≥ B1 + B2 .

This example illustrates frequency division multiplexing, a technique that allows multiple
signals to share the same channel: each signal is allocated part of the frequency spectrum.

3.5 Bandwidth
In our discussion of the frequency-shifting/modulation property we have observed that to transmit a
signal x1 (t) which covers the frequency band [0, B] (we consider only the positive frequency values
to determine the frequency content of a signal or system) over a communications channel, the
channel should accommodate the [0, B] frequency band, i.e., the channel should have a minimum
bandwidth of B Hz. On the other hand, to transmit the modulated waveform x(t) cos ω0 t we
28 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

F[ x1 (t) cos 2πf1 t ]

A1 /2

f
−f1 f 1 − B1 f 1 f 1 + B1

F[ x2 (t) cos 2πf2 t ]

A2 /2

f
−f2 f 2 − B2 f2 f 2 + B2
Y (f )

f
−f2 −f1 0 f1 f2
f 1 + B1 f 2 − B2

Figure 3.13: Spectra of the modulated and combined waveforms in Example 3.3 (Case 2).

require a communications channel with twice the bandwidth as the modulated waveform occupies
the frequency band [f0 − B, f0 + B].

The concept of bandwidth is an important element in the study of communication systems.

We will now proceed to define and discuss the concept of bandwidth and the classification of
signals based on their frequency contents. In particular, bandwidth is a measure of the extent of
significant spectral components of the signal for positive frequencies. The definition of bandwidth
for systems is a measure of the range of frequencies thac the system can handle. What we mean by
“handled” will be explained as we discuss the concept of distortionless transmission. However, as
a brief introduction to the concept of the bandwidth of a system, you may think system bandwidth
as being the frequency-domain equivalence of dynamic range.

Baseband/Lowpass Signals: A baseband/lowpass waveform has significant spectral magnitude

components for frequencies in the vicinity of the frequency origin (f = 0) with the spectral
magnitude components being negligible elsewhere. Observe that if a signal x(t) is strictly
band-limited, then |X(f )| = 0 for f > Bx as demonstrated in Figure (3.14a). However,
Figure (3.14b) also demonstrates a baseband/lowpass signal; in this case |X2 (f )| rapidly
approaches zero yet never becomingzero beyond a certain frequency. Below, we will discuss
in some detail how to measure the bandwidth of such not strictly band-limited signals.

X1 (f ) X2 (f )

f f
Bx

  
    

  
   

Figure 3.14: Spectral characteristics of baseband/lowpass signals.

3.5. BANDWIDTH 29

Bandpass Signals: A bandpass waveform has spectral magnitude components that are non-zero
for frequencies in some band concentrated at ±fc with fc  0. The spectral magnitude
components are negligible elsewhere. We recall that if x(t) is a baseband signal with band-

Xbp (f )

f
−fc fc

Figure 3.15: Spectrum of a bandpass signal.

width B-Hz, then x(t) cos ωc t and x(t) sin ωc t are bandpass signals both with bandwidth 2B.
Further, the modulated waveforms x(t) cos ωc t and x(t) sin ωc t occupy the frequency band
|f ± fc | ≤ 2B. Therefore, a generic bandpass signal can be represented as

xbp (t) = xc (t) cos ωc t + xs (t) sin ωc t, (3.50)

where xc (t) and xs (t) are baseband signals each with bandwidth B-Hz. Observe that the
spectrum of xbp (t) is not necessarily symmetric with respect to ±fc . We can also express
xbp (t) as:
xbp (t) = E(t) cos ωc t + ϕ(t) , (3.51)
with

E(t) = x2c (t) + x2s (t), (3.52)
 x (t) 
s
ϕ(t) = arctan . (3.53)
xc (t)
Since xc (t) and xs (t) are baseband signals, so are E(t) and ϕ(t). In particular E(t) is the
slowly-varying (baseband/lowpass) envelope signal and ϕ(t) is the slowly-varying phase
signal. Consequently, the bandpass signal xbp (t) will appear as a high-frequency signal with
slowly varying amplitude and slowly varying frequency.

E(t) Xbp (f )
xbp (t)

t f
−fc fc

Figure 3.16: Time- and frequency-domain display of xbp described in Equations (3.50–3.53).

Strictly Band-limited Signals: If a signal is strictly band-limited, then the definition of band-
width is obvious. A strictly band-limited signal will have non-zero spectral magnitude com-
ponents only over a finite frequency band. Therefore, for a strictly band-limited baseband
30 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

signal xlp (t) with |Xlp (f )| = 0 for |f | ≥ Bx , the signal bandwidth equals Bx -Hz. Similarly
for strictly band-limited bandpass signal xbp (t) with |Xbp (f )| = 0 for |f − fc | ≥ Bx , the
signal bandwidth equals 2Bx -Hz.



Xlp (f ) Xbp (f )

f f
0 Bx fc − Bx fc fc + Bx

Bandwidth = Bx Bandwidth = 2Bx

Figure 3.17: Bandwidth measurement for strictly band-limited signals.

Bandwidth Measurement for Not Strictly Band-limited Signals: If the signal is not strictly
band-limited, then we can use different criteria to measure the signal bandwidth. We will
now introduce two of the most commonly used criteria.

• Main-Lobe Bandwidth Measurement: In this case, the bandwidth is defined as the

bandwidth of the main-lobe of the magnitude spectrum of the signal with the under-
standing that the main-lobe of the magnitude spectrum will include the significant fre-
quency components. For example, if xlp (t) = AΠ(t/τ ) then Xlp (f ) = Aτ sinc(πτ f )
such that the main-lobe of magnitude spectrum |Xlp (f )| covers the baseband frequen-
cies [0, 1/τ ] and therefore the main-lobe bandwidth of xlp (t) is (1/τ )-Hz. In the case
of a bandpass signal such as xbp (t) = xlp (t) cos ωc t = AΠ(t/τ ) cos ωc t, we measure
the frequency band occupied by the main-lobe of the modulated waveform (centered
about fc ). In this particular example the main-lobe of the magnitude spectrum |Xbp (f )|
covers the frequency band [fc −(1/τ ), fc +(1/τ )]. Therefore, the main-lobe bandwidth
of xbp is 2/τ .




Xlp (f ) Xbp (f )
Aτ A τ2
1
1
τ fc − 1 fc + τ
τ
f f

fc
1 2
Bandwidth = τ Bandwidth = τ

Figure 3.18: Bandwidth measurement for not strictly band-limited signals.

3.5. BANDWIDTH 31

• 3-dB/Half Power Bandwidth Measurement: An alternative measure of signal band-

width is the so called 3-dB bandwidth. Typical (but not all) lowpass signals have mag-
nitude spectra which are maximum at the origin and decrease monotonically with in-
creasing frequency. Similarly, typical bandpass signal have magnitude spectra which
are maximum at the centre frequency fc , which decrease monotonically as the fre-
quency variable moves away from fc . For such signals we measure the 3-dB bandwidth
as the frequency where the power spectrum drops 3-dB relative to its maximum (if the
signal is x(t) then |X(f )|2 is its power spectrum). Let xlp (t) be lowpass/baseband sig-
nal with power spectrum |Xlp (f )|2 such that maxf |Xlp (f )|2 = |Xlp (0)|2 . Observe that
the 3-dB drop in the power spectrum relative to its maximum is equivalent to halving
the maximum power measured in decibels2 expressed using the unit notation [dB].
 |X (0)|2   
= 10 log10 |Xlp (0)|2 − 10 log10 2,
lp
10 log10
2
 
= 10 log10 |Xlp (0)|2 − 3 dB.
Figure (3.19) depicts the measurement of the 3-dB bandwidth for baseband signals. We

|X(f )|
|X(0)|
√
|X(0)|/ 2
f

|X(0)|2 |X(f )|2 measured in [dB]



|X(0)|2 /2
f

f0
3-dB Bandwidth = f0

Figure 3.19: 3-dB Bandwidth measurement for baseband signals.

measure the 3-dB bandwidth of bandpass signals using a process similar to the one we
used to measure the main-lobe bandwidth, namely the measurements are taken with
respect to the centre frequency of the bandpass signal. Figure (3.20) demonstrates this
process.
• There are other definitions of bandwidth measurement criteria, e.g. the rms-bandwidth.
We will define and explain such criteria as required.

Remarks:

• The bandwidth and time-duration of a signal are inversely related and cannot be indepen-
dently altered, i.e., [Bx ][time-duration] ≥ K where Bx is the signal bandwidth and K is
constant. This inequality is another manifestation of the Heisenberg uncertainty principle.
2
For a given quantity x let xlin be its measurement using a linear scale and let xdB be its measurement using the
dB-scale, then xdB = 10 log10 [ xlin ].
32 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

|X(f )|2 measured in [dB]

|X(0)|2


|X(0)|2 /2
f

fL fc fU

3-dB Bandwidth = fU − fL

Figure 3.20: 3-dB Bandwidth measurement for bandpass signals.

• If x1 has a bandwidth of B1 -Hz and x2 has a bandwidth of B2 -Hz, then x1 x2 has a a band-
width of (B1 + B2 )-Hz. We can easily validate this result by using the frequency convolution
property of the Fourier transform, namely F [x1 (t)x2 (t)] = X1 (f ) ∗ Xf (f ).

3.6 Signal Transmission Through a Linear System

Let h(t) be the impulse response of a LTI system. By definition, h(t) is the response of the system
at rest, i.e., h(t) is the system output observed when all system initial conditions are zero with a
unit impulse function δ(t) applied to the input of the system at time t = 0. The impulse response
function h(t) fully describes the system and allow us to determine the output of the system for an
arbitrary input function using the convolution operation: if x(t) is the input, then the system output
equals y(t) = h(t) ∗ x(t). Where does the convolution operation come from?

1. If the system input equals Kδ(t − t0 ), the sytem output is Kh(t − t0 ). This result is simply
due to the LTI characteristic of the system (a scaled and time-shifted unit impulse function
will generate a scaled and time-shifted impulse response function as the system output).

2. We can approximate any arbitrary wave form x(t) as

 
x(t) ≈ x(λ)Δλ δ(t − λ).
λ

Δτ
x(λ)
x(t)

t
λ

Figure 3.21: Approximation of an arbitrary waveform x(t).

3.7. FREQUENCY RESPONSE 33

3. If the input to the system is [x(λ)Δλ] δ(t − λ), then the system output equals [x(λ)Δλ] h(t −
λ) where we used the definition of the impulse response and the LTI characteristics of the
system. Using the linearity of the system once again, we observe that if λ [x(λ)Δλ] δ(t−λ)
is the system input then λ [x(λ)Δλ] h(t − λ) will be the corresponding output.
4. The accuracy of the approximation of x(t) will improve as Δλ → 0. In this case we will
have:
 
y(t) = lim [x(λ)Δλ] h(t − λ) ,
Δλ→0
λ
 ∞
= x(λ)h(t − λ)dλ,
−∞

= h(t) ∗ x(t).

Important system properties can be deduced from the impulse response function h(t).

Causality: If h(t) = 0 for t < 0, then the system will be causal. The causality condition is
necessary for system realizability.
BIBO Stability: A system is bounded-input, bounded-output (BIBO) stable if maxt |y(t)| < ∞
when maxt |x(t)| < ∞. We can show that the system will be BIBO stable, if |h(t)|dt < ∞.

3.7 Frequency Response

Earlier we provided a physical interpretation of what X(f ) represents: for a given waveform x(t)
(voltage, current,...), X(f ) represents a density function describing the contribution of the contin-
uous frequency components to the composition of x(t). We will now use this interpretation and
extend it to the description of a LTI system. In particular, we will use the term frequency response
to describe a LTI system. In Section 3.2 we have shown that a sinusoidal input (complex or real)
to a LTI system generates a amplitude-scaled and phase-shifted sinusoid at the same frequency as
the input. Let x(t) = ejω0 t be the input to a LTI system decribed by the impulse response function
h(t), then the system output is given by the exporession:
y(t) = |H(f0 )|ej(2πf0 t+arg[ H(f0 ) ]) , (3.54)
where  
H(f ) = F h(t) , (3.55)
is the frequency response function of the system. H(f ) exerts its characteristics on the system
output by modifying the amplitude and phase of the frequency components in x(t).

Example 3.4:

Hence, H(f ) summarizes how the system will affect different frequency components as they pass
through the system. Using Fourier analysis, we can write
Y (f ) = H(f )X(f ).
34 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

 



f1

f2
H(f )

f3 

|H(f )|






f1 f2 f3

Figure 3.22: Determining the frequency/magnitude response of a LTI system.

In our analysis the most important elements are H(f ), the system bandwidth Bh , the bandwidth
of the input signal Bx , and how Bx is related to Bh . As a further example, consider a case as
demonstrated in the above figure where a unit pulse function of τ -second duration is applied to a
lowpass system with adjustable bandwidth.

3.8 Distortionless Transmission

Distortionless transmission is an ideal case that is physically impossible to implement (as we will
show). However, studying the properties of such an ideal case allows us to establish a benchmark
against which we can measure the performance of non-ideal but realizable systems.

Definition: A distortionless system with input x(t) and y(t) satisfies the input-output relation:

y(t) = K x(t − t0 ), (3.56)

where K ∈ is the gain factor, and t0 ∈ is the system delay.

The frequency domain characteristics of a distortionless system can be obtained by taking the
Fourier transform of Equation (3.56) and then determining the corresponding systems function
3.8. DISTORTIONLESS TRANSMISSION 35

y1 (t)
Bh1

x(t)


y2 (t)

H(f )
Bh2


τ


y3 (t)
Adjustable Bandwidth Bh
Bh1 > Bh2 > Bh3 Bh3

Figure 3.23: Effect of system bandwidth on the system response to a unit pulse.

H(f ). As we have
Y (f ) = H(f )X(f ),
with
Y (f ) = F [ y(t) ], (3.57)
= F [ Kx(t − t0 ) ], (3.58)
= K F [ x(t − t0 ) ], (3.59)
= KX(f )e−j2πf t0 , (3.60)
it then follows:
Y (f )
H(f ) = = K e−j2πf t0 . (3.61)
X(f )

   

|H(f )| arg[ H(f ) ]

 
slope = −2πt0

Figure 3.24: Magnitude and phase response of an ideal distortionless transmission system.

Since having a physically realizable with H(f ) = K for −∞ < f < ∞ is not possible, in practice
we require that |H(f )| = K only over a specific frequency band, let’s say [fL , fU ]. In this case,
we may consider the H(f ) of the corresponding “ideal band-limited” distortionless system will be
the same as depicted in Figure (3.24) but seen through a narrow window over the frequency band
[fL , fU ]. Figure (3.25) demonstrates this behavior.
36 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

   |H(f )|


−fU −fL fL fU
 
arg[ H(f ) ]

slope = −2πt0

Figure 3.25: Frequency and phase response functions of an ideal distortionless transmission system band-
limited to the frequency band [fL , fU ].

3.8.1 Approximation for Distortionless Transmission

In many cases achieving exactly a distortionless transmission system even over a limited frequency
range may not be a realizable option. As an example consider the following RC-lowpass system:

vi (t)
vo (t)

Figure 3.26: The RC-lowpass system.

with
1
Hrc (f ) =
1 + j2πf RC
such that
1
|Hrc (f )| =  , and arg[Hrc (f )] = − tan−1 (2πf RC). (3.62)
1 + (2πf RC)2

We can also express the magnitude and phase response functions in terms of the 3-dB bandwidth
parameter f0 . Let f0 be the the 3-dB bandwidth of Hrc such that:

|Hrc (0)|2 1 1 1
|Hrc (f0 )|2 = =⇒ = =⇒ f0 = . (3.63)
2 2 1 + (2πf0 RC)2 2πRC
Equation (3.62) can then be re-written in terms of f0 :
1
|Hrc (f )| =  , and arg[Hrc (f )] = − tan−1 (f /f0 ). (3.64)
1 + (f /f0 )2
3.8. DISTORTIONLESS TRANSMISSION 37

|Hrc (f )|
√

1/ 2

f0
arg[ Hrc (f ) ]


−π/4
−π/2

Figure 3.27: Magnitude and phase spectra of the RC-lowpass system.

Clearly, Hrc (f ) does not represent a distortionless system. Therefore, let us consider the case
where we want to have a system which satisfies the requirements for distortionless transmission
within specified constraints. In particular:
• Consider only the magnitude response, |Hrc (f )|;
• Allow maximum deviation from from distortionless transmission to be no more than 10%
for f ∈ [0, 10] kHz.
As |Hrc (f )| is a monotonically decreasing function of f with |Hrc (f )|max = |Hrc (0)| = 1, we can
solve for the system parameter RC (or f0 ) such that:
1 1
 = 0.9 with f1 = 10 kHz and f0 = . (3.65)
1 + (f1 /f0 )2 2πRC
Solving for for the system parameter RC we obtain RC = 7.7 × 10−6. We can now approximate
the output of Hrc for f ∈ [0, 10] kHz by taking the phase response into consideration:
We determine the phase response at f1 = 10 kHz, by evaluating arg[Hrc (f1 )] = − tan−1 (f1 /f0 ) =
−1
−0.451, where f0 = 2π × 7.7 × 10−6 . We can now approximate arg[Hrc ] over the frequency
band [0, 10] kHz by a straight-line (linear phase response) with slope:
arg[Hrc (f1 )] − arg[Hrc (0)] −0.451
slope = = = −0.451 × 10−4 . (3.66)
f1 − 0 104
From Equation (3.61) we recognize that the system delay t0 and the slope of the linear phase
response (or its approximation as we are discussing right now) are related through the relation:
slope = −2πt0 . (3.67)
Solving for t0 we obtain t0 = 7.2 μs. To conclude, the RC-lowpass system described by the
frequency response function Hr c(f ) approximately provides distortionless transmission for f ∈
[0, 10] kHz such that the the system output is given by the expression:
y(t) ≈ Kx(t − t0 ),
with K ∈ [0.9, 1] and t0 = 7.2 μs.
38 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

|Hrc (f )|






 f0

arg[ Hrc (f ) ]



−π/2

Figure 3.28: Approximating the RC-lowpass system for a distortionless transmission system.

3.8.2 Response of an Ideal Lowpass Filter

Let us consider an ideal lowpass filter (LPF) with the magnitude and phase response function: The

    

|HLP (f )| arg[ HLP (f ) ]

 
slope = −2πt0

Figure 3.29: Magnitude and phase response of an ideal lowpass filter.

corresponding frequency response is then given by the expression:

 f 
HLP (f ) = Π e−j2πf t0 . (3.68)
2B
Using the table of Fourier transforms (Lathi and Ding, Table 3.1, p. 107) and the table of properties
of Fourier transform operations (Lathi and Ding, Table 3.2, p. 123) and we determine the impulse
response function of the ideal lowpass filter from Equation (3.68):

hLP (t) = 2B sinc 2πB(t − t0 ) . (3.69)

Observations:

• hLP (t) represents a non-casual system, therefore an ideal lowpass filter is not realizable.

• As B → ∞, hLP (t) → δ(t − t0 ), such that y(t) ≈ x(t − t0 ). If the input is a single pulse,
then the output will be identical to the input delayed by t0 . (While we are discussing the
3.8. DISTORTIONLESS TRANSMISSION 39

hLP (t)



t0
1 1
t0 − 2B t0 + 2B

Figure 3.30: Impulse response function of an ideal lowpass filter.

case where the input is a single pulse, this observation will be correct with any arbitrary
input.) However, for a signal with B < ∞, the input pulse will be spread-out as a result of
transmission through a lowpass filter with finite B.

x(t) x(t)
HLP (f ) y(t)

Figure 3.31: Response of a lowpass filter to a rectangular pulse.

• We can truncate hLP (t) such that hLP (t) = 0, for t < 0. This operation will convert the
non-causal ideal lowpass filter into a casual and therefore realizable approximation.

• In particular, for system realizability we require

◦ In time domain: h(t) = 0, for t < 0—causality;

◦ In frequency domain: | log H(ω)|/(1 + ω 2 )dω < ∞—Paley-Wiener condition.

Observe that any H(f ) = 0 over a finite frequency band violates the Paley-Wiener condition.

In realizable filters it is not possible to have the passband and stopband ranges adjacent to each
other (this would imply a Π(f ) type of a frequency response which would result in a corresponding
impulse response function with infinite support, and hence will result in a non-causal system).
Also, rather than having |H(f )| = K for f ∈ passband and |H(f )| = 0 for f ∈ stopband, we
specify maximum and minimum attenuation levels specified over these frequency bands. In the
context of frequency-selective filters, we will refer to:

• passband(s), transition band(s), stopband(s);

• maximum deviation/tolerance over passband(s);
• minimum attenuation/tolerance over stopband(s).
40 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

|HLP (f )|

Gp




−Gs

 Bp Bs

   
 


Figure 3.32: Specifications of a realizable lowpass filter.

Frequency-selective filters are mostly specified in terms of their magnitude response functions.
Once the filter is designed to satisfy the magnitude response constraints, we study the correspond-
ing phase response to ensure that the phase distortion (caused by non-linear phase response) re-
mains within acceptable limits. As an example consider a typical, realizable lowpass filter with
the magnitude response constraints specified as shown. The magnitude response function should
remain outside of any cross-hatched regions shown in Figure (3.32). Figure (3.32) also shows that
the following parameters describe a realizable lowpass filter:
Passband: The frequency range [0, Bp ] Hz. (We provide information only for f > 0 as magnitude
response function of the filter will have even symmetry about f = 0).

Passband Tolerance, Gp : This parameter is typically measured in dB and represents the maxi-
mum allowable deviation from ideal distortionless transmission conditions, i.e. |H(f ) = 1
or 0 dB. Ideally, we want Gp to be as close to 1 (or 0-dB) as possible. This parameter is also
know as the ripple factor.

Transition Band: The frequency range [Bp , Bs ] Hz.

Stopband: The frequency range [Bs , ∞) Hz.

Stopband Attenuation, Gs : This parameter represents the minimum attenuation over the stop-
band; it is typpically specified in dB. Iddeally, we want Gs to be as close to 0 (or if specified
in dB, as large as possible). Please note that Gs measures attenuation, therefore it is a positive
quantity. For example, a Gs = 40 dB implies that the lowpass filter will exert a minimum of
40 dB attenuation ( measured relative to the passband gain) over the stopband.
In general, the filter order increases when [Gp is small] or [Gs is large] or the transition band,
[Bp , Bs ], is narrow or the stopband edge frequency Bp is small. The most commonly utilized
frequency-selective filter structures are lowpass, highpass, bandpass, bandstop and allpass filters.
For example, a typical bandpass filter may have the following magnitude response specifications
as shown in Figure (3.33). There are many analog filter design methods/families. Some of the
3.8. DISTORTIONLESS TRANSMISSION 41

|HBP (f )|

Gp1
Gp2




−Gs

 Bp1 Bs1 Bs2 Bp1

      

  

   

 

 
Figure 3.33: Specifications of a realizable bandstop filter.

most commonly used ones methods include: Butterworth, Chebchev-I, Chebchev-II, Elliptic and
Bessel-Thompson among others. Each of these filter design methodss result in polynomials that
define the pole-zero locations of the filter in the s-plane. When you study a text on analog filter
design techniques, you will notice that all design techniques refer to lowpass filters only. So, how
can we design the other types of frequency-selective filters?
Discussion:
• A comparison of the frequency responses for lowpass and highpass filters (to simplify the
discussion assume that both systems have phase responses that are identical to zero over the
entire frequency band) yield the relationship HHP (f ) = 1 − HLP (f ).
HLP (f ) HHP (f )



 

Figure 3.34: Lowpass to highpass transformation.

• Bandpass and bandstop filters are combinations of lowpass and highpass filters.
• While most common filter structures are frequency-selective, there are other “filters” as well:
allpass filters (used as phase conditioners or to create minimum-phase systems required for
system inevitability), phase-shift system/Hilbert transformer (used to create analytic signals),
differentiator, etc..
• Digital filter design techniques open up many other possibilities. The design of frequency-
selective infinite impulse response (IIR) digital filters rely mostly on existing analog filter de-
sign methodologies (e.g., Butterworth, Chebchev-I/II, ...). On the other hand, finite impulse
42 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

response (FIR) digital filters allow to achieve perfect linear phase (in addition to meeting the
magnitude response specifications) and are particularly suitable for efficient implementation
on high-speech digital signal processors.

3.9 Some Practical Considerations for Bandpass Systems

In a previous section we have analyzed an RC-circuit as a simple first-order lowpass filter. We
now extend the discussion to an RLC-circuit and define the simplest bandpass system.

vi (t)
vo (t)

Figure 3.35: The RLC-Parallel resonant system.

Parallel Resonant/Tank/Tuned Circuit: The transfer function of the RLC-circuit equals:

sL
H(s) = . (3.70)
s2 (RLC) + sL + R
We can also express the frequency response function of the RLC circuit in an equivalent form as:
1
H(f ) = , (3.71)
f f0
1+jQ −
f0 f
where

C
Q=R , (Quality Factor); (3.72)
L
1 1
f0 = √ , (Resonant Frequency). (3.73)
2π LC

From the magnitude response curve shown in Figure (3.36) we observe that
f0
B = fU − fL = (3.74)
Q
with
1
fU = f0 1 + ; (3.75)
2Q
1
fL = f0 1 − . (3.76)
2Q
3.9. SOME PRACTICAL CONSIDERATIONS FOR BANDPASS SYSTEMS 43

|H(f )|2

 




  f0  
1 1
fL = f0 1 − 2Q fU = f0 1 + 2Q

Figure 3.36: The magnitude response of the parallel resonant system.

• Observe that the quality factor parameter Q defines the bandwidth of the RLC-circuit. As
Q increases, the system becomes more selective.

• Practical circuits usually have Q ∈ [10, 100]. In other words, the 3-dB bandwidth of a
bandpass system falls within 1%–10% of its centre frequency f0 .

• The quantity B/f0 is known as the the fractional bandwidth of a bandpass system. Designing
a distortionless bandpass system can be very challenging if B/f0  1 or B/f0  1. For
ease of implementation, a bandpass system should have 0.01  B/f0  0.1. Thus, large

Frequency Band Carrier Frequency f0 Bandwidth B

Longwave radio 100 kHz 2 kHz

Shortwave radio 5 MHz 100 kHz
VHF 100 MHz 2 MHz
μWave 5 GHz 100 MHz
Optical 5 × 1014 Hz 1013 Hz

Table 3.1: Selected carrier frequencies and nominal bandwidth (based on B/f0 ≈ 0.02). [A.B. Carlson,
Communication Systems, Third Edition, McGraw-Hill, Inc., page 195]

bandwidths to be used for signal transmission (more data payload with faster data rates)
require higher frequencies.

• Earlier we defined lowpass/baseband/bandpass signals and systems. One additional termi-

nology we will frequently use is that of a narrowband system.

Definition 3.1. A bandpass system is a narrowband system if its bandwidth is small com-
pared to its centre frequency.

• In communication problems, the information source outputs usually a baseband signal (e.g.
a TTL waveform from a digital circuit, voice signal from a microphone, a video signal from a
44 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

  

g(t) x(t)
 
 
 
  
 

fc


Figure 3.37: Preparing the baseband signal for transmission.

camera ...) In a communication system, it is the transmitter that modulates a high-frequency

carrier signal with the baseband information signal, so that the fractional bandwidth of re-
sulting modulated waveform will be compatible with the above discussed ranges.

3.10 Signal Distortion Over a Communication Channel

3.10.1 Linear Distortion
In its simplest form we define linear distortion as any deviation from the conditions of distortion-
less transmission we defined earlier.

Example 3.5: Let x(t) be arbitrary, periodic signal with a Fourier series expansion {Dn }n . If
all frequency components in x(t) as defined by the Fourier series coefficients are scaled and
delayed by the same amount, then the system output corresponding to x(t) will also be scaled
and delayed. However, if the individual components are scaled and/or delayed differently
(e.g. in the case of a non-uniform magnitude response or a non-linear phase response),
then the output of the system will be distorted relative to the system input. This type of

x(t)

H(f ) y(t)

Figure 3.38: Effets of linear distortion on TDM signals.

distortion can create significant problems for time-division multiplexed (TDM) signals as
signals occupying adjacent TDM channels will mutually interfere.
3.10. SIGNAL DISTORTION OVER A COMMUNICATION CHANNEL 45

3.10.2 Non-Linear Distortion

Let H(f ) be the frequency response of a system with x(t) and y(t) be the system input and output
signals, respectively. Let y(t) = g[x(t)] with g[.] being a non-linear function. If we expand g[.] in
a McLaurin series:  
y(t) = g x(t) = a0 + a1 x(t) + a2 x2 (t) + · · · .
Observe that if x(t) is band-limited to B Hz, then xn (t) is band-limited to nB Hz. Consequently,
the spectrum of the system output is spread out well beyond the spectrum of the input signal.
If spectrum-spreading distortion occurs as a result of non-linear distortion, it may likely create
problems with frequency-division multiplexed systems.

Example 3.6: Let x(t) = A cos ω1 t + B cos ω2 t with

y(t) = g[x(t)] = x(t) + 2x2 (t).

Substituting the expression for x(t) into the system input-output equation (after some simple
arithmetic operations and use of trigonometric identities) results in the following expression
for the system output:
   2
y(t) = A cos ω1 t + B cos ω2 t + 2 A cos ω1 t + B cos ω2 t ,
= A cos ω1 t + B cos ω2 t + 2A2 cos2 ω1 t + 2B 2 cos ω2 t + 4AB cos ω1 t cos ω2 t,
= A cos ω1 t + B cos ω2 t + A2 + A2 cos 2ω1t + B 2 + B 2 cos 2ω2 t
+ 2AB cos(ω1 + ω2 )t + 2AB cos(ω1 − ω2 )t.

If f1 = 4 kHz and f2 = 3 kHz, the input and output spectra will be as shown: Observe that

Bx
X(f )



  

Y (f )




      
By

Figure 3.39: Spectra of x(t) and y(t) = g[x(t)]. (Magnitudes of spectral components are not drawn to
scale.)

x(t) requires a transmission bandwidth of Bx = 1 kHz whereas y(t) requires a transmission

bandwidth of By = 8 kHz due to the spectrum-spreading effects of the non-linear system.
Most importantly, we observe that the system output contains frequency components that are
not part of the system input. This is an example of so-called intermodulation distortion
46 CHAPTER 3. ANALYSIS AND TRANSMISSION OF SIGNALS

(IM distortion). This approach (feeding an input signal consisting of two closely spaced
sinusoids and searching the system output for frequency components that are not part of the
system input) is the foundation of many test systems designed to measure the IM distortion.
As such, the IM distortion is an inherent measure of the system non-linearity.

3.10.3 Multipath Effects

Multipath transmission takes place when the transmitted signal arrives to the receiver by two or
more paths with different delays. Multipath transmission may happen in cases when we transmit
signal over a cable that has impedance mismatch with the transmitter and/or receiver structures.
More frequently, we encounter multipath transmission in mobile communication systems where
the receiving antenna receives not only a direct signal but also its reflections from nearby objects.
The direct and reflected signals may interfere with other constructively or destructively.

 



 


α τ + Δτ

x(t) y(t)

Figure 3.40: Modelling multipath distortion.

3.10.4 Fading Channels

Fading channels are non-stationary channels where the channel transmission characteristics vary
randomly due to changing propagation conditions. In particular the channel may exert random
attenuation of the signal, a phenomenon known as fading. Many communication systems employ
automatic gain control subsystems to counter the effects of such fading channels.