Modulation

for
EE311

by
A. DAHIMENE

Institut de Génie Electrique et Electronique

Université M’hamed Bougara

Boumerdes
 Modulation
In this chapter, we are going to learn basic principles of
modulation. Modulation can be defined as the process of impressing
information from a modulating signal onto another signal called the
carrier. The resulting signal is called the modulated signal (in general,
it is a bandpass signal).

Typical modulation system

The reverse process is called demodulation. When the modulator
and the demodulator are located in the same apparatus, the system is
called a MODEM (MOdulator, DEModulator).
When we do not use modulation, the system is called a "baseband
communication" system. At that time, the baseband signal is
transmitted directly.
Before we go on in the development of modulation theory, we
have first to answer the following question: Why modulate?

1
 Historically, modulation has been introduced in order to use
reasonably sized antennas in radio communication. We know that the
physical size of an antenna is a fraction of the wavelength.
c
The wavelength is λ = , where c = 3×108 m/s is the speed of
f
light, f is the frequency of the signal. So, if we want to transmit a
baseband signal of 3 kHz by radio, the required wavelength is 100 km.
It is evident that it is very hard to build an antenna having many
kilometers of length. If we can transfer the information to a bandpass
signal with a carrier of 30 MHz, we obtain a wavelength of 10 meters.
A quarter wave antenna will be 2.5 meters. This is much more
reasonable. Modulation is also used to make the information fit the
communication channel. Sophisticated modulation schemes are
commonly used nowadays to transmit information. Techniques like
OFDM, Trellis Coding, CDMA, etc. are commonly used in everyday
communication systems.
Distortionless Communication
If we consider the whole communication system from the
baseband source signal to the baseband destination signal, all
communication systems can be considered as baseband. In this case, a
good communication system must be "distortionless". This means that
the destination signal must be a scaled (and maybe delayed) replica of
the source signal. If x(t) is the source signal and y(t) is the destination
one, we must have:
y (t ) = kx(t − τ ) . k is a constant, τ is a time delay.

2
In the frequency domain, we obtain:
Y ( f ) = ke − j 2π f τ X ( f ) . This means that the overall communication
system must behave like a filter (LTI system) with a transfer function:
Y( f )
H( f ) = = ke − j 2π f τ
X(f )
So, distortionless communication implies that the amplitude
response |H(f)| must be constant and that the phase response Arg[H(f)]
must be a linear function of the frequency. This means that all
frequencies must be delayed by the same amount. If the transfer
between the input and output signal is linear and time invariant but
without satisfying the above conditions, we say that the
communication system is subjected to "linear distortion". This
distortion can come from the amplitude response which is not constant
or from the phase response which is not linearly related to frequency
(phase or delay distortion).
This type of distortion can be cured or minimized by using a
filter called an "equalizer" at the output of the communication channel.
When the transfer function between the input and output is nonlinear,
we are in presence of "nonlinear distortion".
Harmonic distortion:
When we apply a pure sinewave at a frequency f0 to a linear
system, the output will be a sinewave at the same frequency. However,
if the system is nonlinear, the output will be a periodic waveform at
the same frequency, but it will not be sinusoidal anymore. So, we
observe harmonics at the output.

3
 Let the input be x(t ) = A cos ω0t , the output will be
∞
y (t ) = c0 + ∑ an cos ( nω0 + θ n ) , an = 2 cn and θ n = Arg [ cn ] .
n =1

The total harmonic distortion coefficient measures how far the

signal y(t) is from a sinewave. It is evaluated as:
∞

∑a 2
n
d = 100 × n=2
%
a12
It is the ratio of the rms value of all the harmonics of the signal
y(t) over the rms value of the fundamental.
Classification of modulation systems.
Depending on the modulating signal, we distinguish two different
types of modulation systems:
• Digital modulation systems: they are used to transmit digital
information through physical channels.
• Analog modulation systems: the modulating signal in this case is
a baseband analog signal.
We can also classify modulation according to the type of carrier used
(and therefore the modulated wave produced).
• Continuous wave (CW) modulation: The carrier is a sinewave
and the modulated signal is a narrow bandpass signal.
• Pulse modulation: The carrier is a periodic train of pulses. The
modulated signal will carry information about samples of the
signal.
We are going to analyze first analog CW modulation.

4
Analog CW modulation.
The modulating signal sɶ (t ) is assumed to be bounded. This
means that there exists a peak value sɶ (t ) max such

that: sɶ (t ) ≤ sɶ (t ) max for all t. We can thus define a normalized signal

sɶ (t )
s (t ) = and we have s (t ) ≤ 1.
sɶ (t ) max

The signal is also assumed to have an average value of zero. This

means that there is no delta impulse at the origin in its spectrum. It is
also assumed to be bandlimited to a maximum frequency W. In other
words, if S ( f ) = F [ s (t )] then S ( f ) = 0 for f > W .
In CW modulation, the modulated signal x(t) is a narrow
bandpass signal. This means that it can be expressed in either
quadrature form:
x(t ) = a (t )cos ω0t − b(t )sin ω0t
or in modulus/phase form:
x(t ) = r (t )cos (ω0t + ϕ (t ) )

ω0 = 2πf0 being the carrier frequency. Depending on the

modulation method, the information (signal s(t)) can affect a(t), b(t),
r(t) or ϕ(t).
I. Linear modulations
We are going to study in this part modulation methods where the
quadrature components a(t) and b(t) are linearly dependent on the
baseband signal s(t). A linear modulation method must satisfy the

5
superposition principle. If x1(t) is produced by s1(t) and x2(t) is
produced by s2(t), then a1x1(t)+a2x2(t) is produced by a1s1(t)+a2s2(t).
Before proceeding in the analysis of the different types of linear
modulation, we are going to study an "almost linear" one: The
Amplitude Modulation (AM).
Amplitude Modulation (AM)
In AM, the information s(t) is carried by the modulus r(t) of the
signal x(t). Since we have the constraint that r(t) must remain positive
all the time, we cannot simply make it proportional to s(t). We have to
add a constant in order to satisfy the above constraint.
r (t ) = A0 + ka sɶ (t )

A0 is a positive dc signal added to make r(t) ≥ 0, ka is a

proportionality constant and sɶ (t ) is the unnormalized signal. The
phase of the carrier ϕ(t) is constant and we use the value of zero. If we
introduce the normalized signal s(t), we can re-express r(t) as:
r (t ) = A0 + ka s (t ) sɶ (t ) max = A0 (1 + ms (t ) )

ka sɶ (t ) max
m=
A0
m is called the modulation index.
Since r(t) must be positive, we see that we must have 0 ≤ m ≤ 1.
If it happens that m exceeds 1, we say that we have overmodulation.
The AM signal is:
x(t ) = A0 (1 + ms (t ) ) cos ω0t

6
 Historically, amplitude modulation is the first modulation system
put into practice. It was used essentially because of the simplicity of
the receiver structure. It is easily verified that AM is not linear since it
does not satisfy the superposition principle. It is as linear as the
function f ( x) = ax + b . This function is not linear however it is
incrementally linear, i.e. an increment of the input is linearly related to
an increment of the output.

1.5

trough of modulation
1

0.5

A1 0 A2

-0.5

-1

-1.5
0 100 200 300 400 500 600 700 800 900 1000
peak of modulation

Sinewave modulated waveform

The signal s(t) can be an energy or a power type of signal. In the

first case, we can compute easily the spectrum of the modulated signal
as a function of the baseband modulating one.

7
 Starting from x(t), we obtain
x(t ) = A0 (1 + ms (t ) ) cos ω0t = A0 cos ω0t + A0 ms (t )cos ω0t
giving
A0 A A A
X(f ) = δ ( f − f 0 ) + 0 δ ( f + f0 ) + 0 mS ( f − f 0 ) + 0 mS ( f + f 0 )
2 2 2 2
This relation is shown graphically below.
S(f)

−W W f

X(f)

A0
A0
δ(f + f0) δ(f − f0)
2 2
A0
A0
m S(f + f0) m S(f − f0)
2 2

−f0 f0 − W f0 f0 + W f

The above sketch shows the transformation from the baseband

signal s(t) to the bandpass signal x(t). We can also remark that if we
do not want to have a superposition of shifted spectra (aliasing), we
must have f 0 > W . It is also apparent that the bandwidth B of the
modulated signal is twice the bandwidth of the baseband signal.
B = 2W

8
 If we consider only the positive half of the spectrum, we remark
that the Hermitian symmetry of S(f) is translated to f0. So, the positive
half is composed of two halves: the upper sideband above the carrier
frequency and the lower sideband below the carrier frequency.
Power Computation
In order to analyze power signals, we may assume that s(t) is
periodic. We can start the analysis with the simplest real periodic
signal: the sinewave. So, let us assume that s (t ) = cos ωm t where

ωm < ω0.
x(t ) = A0 (1 + m cos ωmt ) cos ω0t = A0 cos ω0t + A0 m cos ωmt cos ω0t
Using trigonometric identities, we obtain;
A0 A
x(t ) = A0 cos ω0t + m cos (ω0 − ωm ) t  + 0 m cos (ω0 + ωm ) t 
2 2
The signal in this case is composed of 3 sinewaves: the carrier with
A0
amplitude A0 and the two sidebands with amplitude m each. The
2
spectrum consists of only Dirac impulse functions. A more general
case is the one of a bandlimited periodic signal. We can express s(t) as:
N
s (t ) = ∑ ak cos ( kωmt + θ k )
k =1

The signal has a zero dc value and the number of harmonics N is

W 
given by N =   . The notation   stands for the floor (i.e. the
 fm 
integer just below) of the number written inside. The modulated signal
is now given by:

9
 N
x(t ) = A0 cos ω0t + A0 m∑ ak cos ( kωmt + θ k ) cos ω0t
k =1

A0 N
x(t ) = A0 cos ω0t + m∑ ak cos (ω0 − kωm ) t + θ k 
2 k =1
A0 N
+ m∑ ak cos (ω0 + kωm ) t + θ k 
2 k =1
The above formula is general enough to allow us to compute the
power of the modulated signal. If we assume that the different
sinewaves are independent, the total power will be given by the sum
of the power of the different components.
A02 A02 2 N 2
Px = + 2 × m ∑ ak
2 8 k =1

A02
In the above relation, we can recognize the carrier power Pc = and
2
A02 2 N 2
the sideband power Psb = m ∑ ak . So, the total power of the signal
8 k =1

is: Px = Pc + 2 Psb . The sideband power can also be expressed as a

ak2N
function of the power of the normalized baseband signal Ps = ∑ ,
k =1 2

A02 2
i.e. Psb = m Ps . So, in terms of the carrier power and the sideband
4
power, we obtain:
A02 2 A02 A02 2
Px = Pc + m Ps = + m Ps
2 2 2

10
Given that the signal s(t) is normalized with a maximum value of 1, its
1 1
Tm ∫Tm ∫
power is less than 1 ( Ps = ≤ s (t ) max dt = 1 ). The
2 2
s (t ) dt
Tm Tm

modulation index m is also less or equal to 1. This means that the

power transmitted by the two sidebands is smaller than the power used
to transmit a carrier that conveys no information. More than 50% of
the total power is used to transmit the carrier. We can evaluate the
efficiency of the system using the following efficiency coefficient:
2 Psb m 2 Ps
η= =
Pc + 2 Psb 1 + m 2 Ps
η cannot exceed the value of 1/2.
1 m2
Example: if s (t ) = cos ωm t , Ps = , then η = . If we use m = 1,
2 2 + m2
1
the efficiency becomes η = . In general, AM systems use a
3
modulation index smaller than 1. The typical value is 30%. AM is
used to transmit audio signals that have an average power Ps << 1. In
this case, the efficiency drops to a very small value.
Broadcast AM radio use carrier frequencies in the medium waves
(530 kHz to 1.71 MHz) or in short waves (3 to 30 MHz). We can
encounter stations transmitting in long waves (148.5 kHz to
283.5 kHz). The bandwidth B allocated to every channel is 10 kHz.
This means that each sideband occupies a bandwidth W of 5 kHz.
AM production:
The following block diagram reproduces the defining equation of AM.

11
 sɶ (t )

dc A cos ω0t

Another structure can be derived by manipulating the defining

equation.
x(t ) = A0 (1 + ms (t ) ) cos ω0t = A0 cos ω0t + A0 ms (t )cos ω0t

sɶ (t )

A cos ω0t

These block diagrams are essentially theoretical. In practice, radically

different methods are used. A common technique is to use a saturating
class C amplifier. The theory behind this type of amplifier will be
covered in a future course. Another method for AM production is to
use a memoriless nonlinear amplifier.
Consider a system with the following input output transfer:
z = a0 + a1w + a2 w2

Let w = s1 + s2 , then z = a0 + a1s1 + a1s2 + a2 s12 + a2 s22 + 2a2 s1s2

If s1 (t ) = A cos ω0t and s2 (t ) = sɶ (t ) , then

12
z (t ) = a0 + a1 A cos ω0t + a1sɶ (t ) + a2 A2 cos 2 ω0t + a2 sɶ 2 (t ) + 2a2 Asɶ (t )cos ω0t
1 1
Now, cos 2 θ = + cos 2θ , so z(t) is the sum of four different
2 2
components:
a2 A2
Dc component: a0 +
2
Baseband component: a1sɶ (t ) + a2 sɶ 2 (t )

Component around ω0: a1 A cos ω0t + 2a2 Asɶ (t )cos ω0t

a2 A2
Component at 2ω0: cos 2ω0t
2
If we use a bandpass filter tuned at f0, we can select the component
around ω0. This component is:
x(t ) = a1 A cos ω0t + 2a2 Asɶ (t )cos ω0t
 2a 
= a1 A 1 + 2 sɶ (t )  cos ω0t
 a1 
= A0 (1 + ms (t ) ) cos ω0t

2a2 s (t ) max
In the above expression, A0 = a1 A and m = . In order to
a1
specify the filter, we have to compute the spectrum of the signal z(t).
 a2 A 2 
Z ( f ) =  a0 + δ ( f )
 2 
+ a1Sɶ ( f ) + a2  Sɶ ( f ) ∗ Sɶ ( f ) 
aA
+ 1 [δ ( f − f 0 ) + δ ( f + f 0 ) ] + a2 A  Sɶ ( f − f 0 ) + Sɶ ( f + f 0 ) 
2
a2 A 2
+ [δ ( f − 2 f0 ) + δ ( f + 2 f0 )]
4

13
 Z(f)

a2 A2 a1 A 2
a0 + δ ( f − f0 ) a2 A
δ ( f − 2 f0 )
2 2 4
a2  Sɶ ( f ) ∗ Sɶ ( f ) 
a2 ASɶ ( f − f 0 )
a1Sɶ ( f )

W 2f0
2W f0 f
f0 − W f0 + W

Positive frequency spectrum of Z(f)

The convolution term is the Fourier transform of the square a2 sɶ 2 (t ) . If
we look at the above spectrum, we remark that the AM signal is
present around f0. It has a bandwidth B = 2W. We notice also that we
must have f0 > 3W if we want to avoid overlap of spectra. If the
nonlinear system contains a higher degree, then this condition will
change.
There exist many other methods for producing AM signals. They are
better analyzed in an electronic circuit course.
AM demodulation:
There exist several techniques for AM modulation. They fall into two
different classes: Homodyne (Synchronous, Coherent) demodulation
and Non Coherent demodulation.
Coherent demodulation:
In this technique, we multiply the received AM signal by a carrier
generated at the receiver. The local carrier must have the same
frequency and same phase as the one of the AM signal. Let the

14
received signal be: x(t ) = A0 [1 + ms (t ) ] cos ω0t and the locally generated

carrier: y (t ) = B cos (ω0 + ∆ω ) t + ϕ0  . ∆ω is a frequency error and ϕ0 is a

phase error. The result of the product is:

z (t ) = A0 B [1 + ms (t ) ] cos ω0t cos (ω0 + ∆ω ) t + ϕ0 
A0 B AB
= [1 + ms(t )] cos ( (∆ω )t + ϕ0 ) + 0 [1 + ms(t )] cos ( 2ω0t + (∆ω )t + ϕ0 )
2 2
It is composed of two components. If we use a lowpass filter with a
bandwidth W, we will recover the first term. If we want to demodulate
the signal, the frequency error ∆ω must be zero and the phase error ϕ0
must be as small as possible (far from the value of π/2). The recovered
signal will be a signal proportional to s(t) plus a dc component. A
capacitor in series is enough to eliminate the dc component. The
locally generated carrier on the other hand must be exactly
synchronized to the received carrier. In most implementations, we can
use the received carrier if the baseband signal is itself bandpass. This
is the case of most audio1 and speech2 signals.

Narrow Lowpass
bandpass filter filter
at f0 cutoff W

The above figure shows a typical synchronous demodulator.

Non coherent demodulator:

1
Audio signals are usually bandpass between 50 Hz and 15 kHz.
2
Speech signals are bandlimited between 300 and 3400 Hz.

15
The envelop detector is commonly use in AM receivers and is in fact
the first demodulator in the history of radio communication.

C R

The envelop detector is composed of a diode, a resistor and a

capacitor. It is essentially a peak detector. This means that the time
constant of the circuit (RC) must be much larger than the period of the
carrier (1/f0). On the other hand, the circuit should not distort the
information signal s(t). This implies that the time constant must be
smaller than the period of the highest frequency present in the
signal (W).
1 1
> RC >>
W f0

You will have the occasion to experiment this circuit in the lab. If the
above condition is satisfied, the signal obtained at the output will be
proportional to the envelop of the AM signal r(t). Here again a dc
blocking capacitor is needed to eliminate the dc value present in the
demodulated signal.
Double sideband suppressed carrier modulation (DSB-SC):
In AM, we spend more than half of the total power transmitting a
carrier that conveys no information. The following method transmits
just the sidebands without transmitting the carrier. The DSB-SC signal
is then:

16
 x(t ) = A0 s (t )cos ω0t

We see immediately than this method is a linear modulation scheme.

Furthermore, from the defining equation, the DSB-SC modulated
signal is a bandpass signal written in quadrature form. a(t ) = A0 s (t ) and
b(t ) = 0 . In the frequency domain, the spectrum of the DSB-SC signal

is obtained by a straightforward application of the modulation theorem.

A0
X(f )= [ S ( f − f 0 ) + S ( f + f 0 )]
2
We see that all the power is in the transmitted sidebands. There is no
transmitted carrier. The transmitted power is:
1 2
Px = 2 Psb = A0 Ps
2
Most transmitters are limited by the peak power they can transmit.
The peak power is given by the square of the maximum of the
envelop Ppeak = r (t ) max . For DSB-SC, the maximum of the envelop is A0
2

while for AM this maximum is A0(1 + m). So, for AM,

Ppeak = A02 (1 + m ) and for DSB-SC, Ppeak = A02 . The ratio of sideband
2

power over the peak power for the two modulations is given by:
 Ps
 DSB-SC
Psb  4
=
Ppeak m 2 Ps
 AM
 4 (1 + m )
2

So, for a given peak power, a DSB-SC transmitter produces more than
four times the sideband power of an AM transmitter.

17
Except for a missing impulse, the spectrum of DSB-SC and the one of
AM look alike, however, in the time domain, the is a fundamental
difference. The DSB-SC envelop and phase are given by:
0 s (t ) > 0
r (t ) = A0 s (t ) ϕ (t ) = 
π s (t ) < 0

Every time the signal s(t) changes sign, the modulated signal
undergoes a phase reversal.
1

0.8

0.6

0.4

0.2

-0.2

-0.4

Phase reversals -0.6

-0.8

-1
0 0.5 1 1.5 2 2.5

We see that we cannot demodulate a DSB-SC signal with a simple

envelop detector. We need a more sophisticated demodulator.
The most commonly used demodulator for DSB is the homodyne one.
However, since there is no carrier transmitted along with the signal,
the local carrier generation is more complex.
Let x(t ) = A0 s (t )cos ω0t . The signal s(t) is assumed to be a power
signal with zero average. If we square the signal x(t), we obtain:
A02 s 2 (t )
x (t ) = A s (t )cos ω0t =
2 2 2
0
2
(1 + cos 2ω0t ) . The signal s2(t) is
2
completely positive. This means that it has an average value that is

18
different from zero. We can express it as: s 2 (t ) =< s 2 (t ) > + s1 (t ) . The
signal s1(t) has a zero average. In the frequency domain, we obtain:
A02 < s 2 (t ) > A02
Z( f ) = δ ( f ) + S1 ( f )
2 2
A02 < s 2 (t ) >
+ [δ ( f − 2 f0 ) + δ ( f + 2 f0 )]
4
A02
+ [ S1 ( f − 2 f 0 ) + S1 ( f + 2 f 0 ) ]
4
We observe a spectrum around 2f0 that is practically the one of an AM
A02 < s 2 (t ) >
signal with a carrier area of . So, we can use a narrow
4
bandpass filter tuned at 2f0 to extract a carrier. The filter will be
followed by a frequency divider by 2 (a simple D flip-flop).

Narrow
( )² Bandpass Filter ÷2 LPF
at 2f0

Squaring loop demodulator

In the above system, there remains a small problem. When we divide a
frequency by two, we have an ambiguity of π in the phase. This is due

to the fact that

( 2ω0t + 2kπ ) = ω t + kπ . This means that we can have
0
2
a signal reversal at the demodulator. If the destination of the
demodulator is the human ear, this reversal will not be noticed by the
auditor. However, if the system is used to transmit data, and if we
assign "1" to a positive value and "0" to a negative value, then the data

19
will be negated. One way to prevent this is to send a prefix word
known to the receiver. If it is received correctly, we keep the output of
the squaring loop. Otherwise, we invert the output carrier from the
squaring loop.
One way to avoid problems in carrier recovery is to send a subcarrier
at a frequency related to the one we want to recover.
Single Sideband Modulation (SSB):
When we are transmitting real signals in DSB-SC, the two sidebands
are related and if we know one, we can deduce the other. So, this
modulation method transmits only one of the two sidebands, either the
upper sideband (USB-SSB) or the lower sideband (LSB-SSB).
Basically, an SSB modulator can be implemented using a DSB-SC
one followed by a sideband filter.

s(t)
Sideband
filter

A0 cos ω0t

SSB modulator
It is quite simple to represent the different operations in the frequency
domain. The following sketch shows a USB-SSB signal in the
frequency domain.

20
 S(f)

−W W f

X(f)

−f 0 −W −f 0 f0 f0+W f

USB-SSB Spectrum
We can observe from the above sketch that the bandwidth of the SSB
signal is the same as the one of the baseband signal. So, for the same
information, the SSB modulated signal uses half the bandwidth of the
DSB modulated signal. This is why SSB is used in crowded spectrum
environment such as amateur radio. It has been used also in Frequency
Division Multiplexing (FDM) systems to transmit different voiceband
signals3. If we observe the following figure, we can observe that the
different shifted spectra do not overlap. They can be transmitted using
a single wire. To avoid any problem in carrier recovery, a subcarrier is
usually transmitted in a separate channel.

3
A voiceband signal is a signal that conveys human speech. Its spectrum is essentially different from zero in a
band between 300 and 3400 Hz.

21
 Band Limiting Filters
SSB Filter
DSBSC
4kHz
CH1 8.6 → 15.4kHz 12.3 → 15.4kHz
m1(t)

300Hz 3400kHz
f1 = 12kHz

4kHz

12.6 → 19.4kHz 16.3 → 19.4kHz

CH2
m2(t)
300Hz 3400kHz

f1 = 16kHz

Σ
Increase in 4kHz steps

FDM OUT
12 – 60kHz
4kHz

CH12 52.6 → 59.4kHz

m12(t)
56.3 → 59.4kHz
300Hz 3400kHz

f12 = 56kHz

12 Channel FDM system

To see how SSB is demodulated, we have to express the SSB signal in
the time domain. In order to do so, it is simpler to obtain first the
corresponding analytic signal X + ( f ) and then use the fact that

x(t ) = Re [ x+ (t )] . When we observe the USB-SSB spectrum, we notice

that the analytic signal X + ( f ) is just the analytic signal S+ ( f ) shifted

and scaled in the frequency domain. So, X + ( f ) = A0 S + ( f − f 0 ) . In the

time domain, it gives x+ (t ) = A0 s+ (t ) exp [ jω0t ] . Replacing s+ (t ) by its

expression s (t ) + jsˆ(t ) , we obtain:

x(t ) = A0 Re ( s (t ) + jsˆ(t ) )( cos ω0t + j sin ω0t ) 
= A0 [ s (t )cos ω0t − sˆ(t )sin ω0t ]

22
The above expression is the one of a USB-SSB modulated signal. It is
a simple matter to show that the expression of the LSB-SSB
modulated signal is:
x(t ) = A0 [ s (t )cos ω0t + sˆ(t )sin ω0t ]
The above two expressions suggest that SSB modulators can be
implemented using the following block diagram:

A0 cos ω0t
s(t)
π ∓
2
Hilbert
transform

The minus sign is for USB-SSB while the plus is for LSB-SSB. This
method of SSB production is called the Phasing Method.
SSB Demodulation:
We are going to consider only the coherent demodulation method. A
general SSB signal is: x(t ) = A0 [ s (t )cos ω0t ∓ sˆ(t )sin ω0t ] . We

multiply this signal by a carrier y (t ) = B cos (ω0 + ∆ω ) t + ϕ0  . Let us

consider first the case of zero frequency offset (∆ω = 0).

The result of the product contains terms at low frequency and terms
around 2f0. The low frequency component is:
A0 B
[ s(t )cos ϕ0 ∓ sˆ(t )sin ϕ0 ]
2

23
So, the output of the coherent demodulator will contain a linear
combination of s (t ) and sˆ(t ) . If the end destination is the human ear,
this signal will sound exactly as s(t) alone. This is due to the fact that
the human ear is insensitive to phase shifts in the signal. In other cases,
the phase error cannot be tolerated.
The analysis of the frequency error is easier to study in the frequency
domain. Using the modulation theorem, the result of the SSB signal
multiplied by a carrier is:
1 1
F  x(t ) cos (ω0 + ∆ω ) t  = X ( f − f 0 − ∆f ) + X ( f + f 0 + ∆f )
2 2
Starting from a USB-SSB signal with the spectrum shown below:

X(f)

−f 0 −W −f 0 f0 f0+W f

We obtain the following spectrum after elimination of the components

around 2f0:

S(f)

−W − ∆f −∆f ∆f W + ∆f f

24
We remark that all the frequencies of the message are translated by a
constant shift. This constant shift does not make the speech
unintelligible. However, when ∆f is positive, it makes everybody
sound like "Donald Duck", hence the name; Donald Duck distortion.
On the other hand, music will be completely distorted since the
harmonic relations between notes will disappear.
Advantages and disadvantages of SSB:
We see that SSB is a linear modulation system that saves on
bandwidth. The transmission bandwidth is equal to the signal one.
However, in order to achieve this result we need very complex
hardware.
In the filtering method, we have to transmit completely one sideband
and eliminate completely the other. This means that the transition
region of the filter is zero. The only way to achieve reasonable filters
is to use this method for signals that have no energy around zero
frequency.

X(f) Filter
Amplitude
transfer

f 0 −W f0 f0+W f

If we want to apply the phasing method, we will encounter the same

problem. It is impossible to build a filter that phase shifts all
frequencies from zero to W. This means that the phase response of the

25
filter has also a zero transition band. We can approximate the Hilbert
transformer if the signal s(t) has the same character as above. It must
not have any energy around zero. So, SSB is useful if we want to
transmit speech. Audio signals can be transmitted at the expense of a
quite complicated hardware. Data cannot be transmitted in the shape
of a sequence of pulses. This signal possesses power at zero frequency.
If we want to transmit signals that have spectra that are different from
zero around dc, one solution is to use Vestigial Sideband.
Vestigial Sideband (VSB):
In VSB modulation, we use filter that transmit most of one sideband
and a very small amount of the other (a vestige).
In order to determine the filter characteristics, we must analyze a
complete modulation and demodulation system. The demodulation
method is always coherent. We multiply the received signal by a
carrier B cos ω0t and we lowpass filter the result to eliminate terms
around 2ω0 .

A B C D E
VSB Filter Lowpass
Filter

A0 cos ω0t B cos ω0t

In the above block diagram, we must determine the signals at different

points.
At A, we have the baseband signal xA(t) = s(t) with spectrum
XA(f) = S(f).

26
At B, we obtain the DSB-SC signal xB(t) = A0s(t)cosω0t with spectrum
A0 A
XB( f ) = S ( f − f0 ) + 0 s( f + f0 ) .
2 2
At C, we have the VSB signal obtained by filtering the DSB-SC signal.
We are going to characterize it in the frequency domain only:
A0
XC( f ) = [ H ( f ) S ( f − f0 ) + H ( f ) S ( f + f0 )]
2
At D, we use the modulation theorem of Fourier transforms and we
obtain:
A0 B
XD( f ) = [ H ( f − f0 )S ( f − 2 f0 ) + H ( f − f0 )S ( f )
4
+ H ( f + f 0 ) S ( f ) + H ( f + f 0 ) S ( f + 2 f 0 )]
The lowpass filter eliminates all the terms around ±2f0. So, the signal
A0 B
at E will be: X E ( f ) = [ H ( f − f0 )S ( f ) + H ( f + f0 )S ( f )]
4
If we want to have a distortionless transmission, this signal must be
proportional to s(t). This means that:
H ( f + f 0 ) + H ( f − f 0 ) = constante
After some manipulations, we obtain that the transfer function of the
filter must satisfy:
H ( f 0 + x) + H ∗ ( f 0 − x) = 2 Re [ H ( f 0 )] for f around f0.

If H ( f ) = R ( f ) + jX ( f ) , then
R( f 0 + x) + R ( f 0 − x ) = 2 R( f 0 )
X ( f 0 + x) = X ( f 0 − x)

27
 Re[H( f )]

f
f0

Im[H( f )]

f0 f

The above graph shows the different symmetries that the real and
imaginary part of the transfer function must satisfy. The real part must
show an odd symmetry with respect to the point (f0, Re[H(f0)]) while
the imaginary part must have the vertical line passing by f0 as a
symmetry axis.
The VSB signal has been characterized in the frequency domain. We
have seen that it can be demodulated using coherent demodulation.
We can also have the expression of the VSB signal in the time domain.

28
Being a general bandpass signal, it can be expressed in quadrature
form and it is completely described by its complex envelop. The VSB
signal appears at point C in our block diagram. Its complex envelop is
given by the filtering of the complex envelop of the DSB-SC signal at
B by the lowpass equivalent filter.
The complex envelop of the signal at B is mxB (t ) = A0 s (t ) with a
spectrum M xB ( f ) = A0 S ( f ) . The lowpass equivalent filter H lp ( f ) is

the filter H ( f ) translated down. Using the symmetries derived above,

we can express the equivalent lowpass filter transfer function as:
H lp ( f ) = R ( f 0 ) + A( f ) + jX lp ( f ) = R ( f 0 ) [1 + jQ( f )]

In the above expression, A(f) is an odd function (R(f) translated down

to f = 0 and shifted down by R(f0)) while Xlp(f) is an even function.
This implies that Q(f) satisfies Q( f ) = Q∗ (− f ) . The complex envelop
of the VSB signal is then M x ( f ) = A0 R ( f 0 ) S ( f ) [1 + jQ( f )] or in the

time domain: mx (t ) = A0 R ( f 0 ) s (t ) + jA0 R ( f 0 )F -1

[Q( f ) S ( f )] . The
function Q(f)S(f) satisfies the condition of Hermitian symmetry. This
implies that its inverse Fourier transform is real. Let
q(t ) = F −1
[Q( f ) S ( f )] then the complex envelop of the VSB signal is:
mx (t ) = A0 R ( f 0 ) [ s (t ) + jq (t ) ] and the VSB signal can be written as:

x(t ) = A0 R ( f 0 ) [ s (t )cos ω0t − q (t )sin ω0t ] .

The signal q(t) is the output of the filter with transfer function Q(f).
A( f ) X lp ( f )
Q( f ) = +
jR( f 0 ) R( f 0 )

29
Two extreme cases are interesting:
If we want to keep the upper sideband and eliminate completely the
2 R( f 0 ) f > f0
lower one, we must have H ( f ) = 
 0 0 < f < f0

This implies that Q( f ) = − j sgn( f ) and q(t ) = sˆ(t ) . The VSB signal in
this case is a USB-SSB signal.
The other extreme case is when we want to keep both sidebands. At
that time, Q(f) = 0 and the signal is just a DSB-SC one.
In our analysis, we have assumed that we favor the upper sideband.
We can obtain the same results for the lower sideband. The modulated
signal bandwidth is intermediate: W < B < 2W.
Envelop demodulation of linear modulation + carrier:
If we add a large amplitude carrier to the inphase component of a
bandpass signal (DSB, SSB, VSB) we obtain:
x(t ) = B cos ω0t + A0 s (t )cos ω0t ± A0 q (t )sin ω0t
The envelop of the signal is:

A02 q 2 (t )
r (t ) = ( B + A0 s(t ) ) + A q (t ) = B + A0 s (t ) 1 +
2 2 2

( B + A0 s (t ) )
0 2

If B >> A0 , the expression inside the absolute value will always be

A02 q 2 (t )
positive and r (t ) ≈ ( B + A0 s (t ) ) + ≈ B + A0 s (t ) .This
2 ( B + A0 s (t ) )
technique is used in the transmission of analog television where the
TV signal is transmitted in VSB+Carrier.

30
Exponential modulation

Instantaneous frequency
Up to now, we have defined the frequency as the speed of rotation of a
phasor (constant frequency phasor) φ (t ) = A0 exp  j (ω0t + θ 0 )  . We

are going to generalize this definition to general complex functions of

the real variable t. Consider such function: z (t ) = r (t )exp ( jθ (t ) ) . It is
obvious that it is a generalization of the constant frequency phasor. It
can be represented graphically as a vector with modulus (amplitude)
r(t) and argument θ(t).

r (t )

θ (t )

The argument θ(t) of z(t) is called the instantaneous phase. Since this
phase varies, the generalized phasor is going to rotate. However, it is
not going to rotate at a constant speed. We can thus define an
instantaneous speed of rotation for this function. It is the instantaneous
frequency:
dθ
ω (t ) = rd/s
dt

31
We can measure this frequency in Hertz:
ω (t ) 1 dθ
f (t ) = =
2π 2π dt
In this section, we are interested in constant amplitude phasors.
Furthermore, we assume that the instantaneous frequency has an
average value f0 with a deviation around it d(t):
f (t ) = f 0 + d (t )
The average value of d(t) is zero. This means that the instantaneous
phase can be expressed as:
θ (t ) = 2π f0t + φ (t ) = ω0t + φ (t )
φ (t ) is called the instantaneous phase deviation and we have:
1 dφ
d (t ) =
2π dt
To this generalized phasor, we can associate the following signal:
x(t ) = Re [ w(t ) ] = r (t )cos (ω0t + φ (t ) )
This signal has the general shape of a bandpass signal. In order for x(t)
to be bandpass, its quadrature components must be bandlimited to a
frequency W < f0. The quadrature components are:
a(t ) = r (t ) cos φ (t ) and b(t ) = r (t )sin φ (t ) .
This condition is not always satisfied. However, in practical
exponential modulation, the carrier f0 is usually very high (hundreds of
MHz). So, we can consider that the obtained modulated signals are
bandpass signals.

32
Frequency and Phase Modulation (FM & PM)
For both phase and frequency modulation, the modulated signal must
have constant amplitude. The information is carried in the phase
deviation. These modulations are called "exponential Modulation"
because the signal has always the following shape:
x(t ) = Re  A0 exp ( jφ (t ) ) exp ( jω0t ) 

Phase modulation is a modulation process that makes the phase

deviation φ(t) proportional to the baseband signal s(t).
φ (t ) = kφ sɶ (t ) = (∆φ ) s (t )
The constant ∆φ = kφ sɶ (t ) max is called the maximum phase deviation.

In order to avoid phase ambiguity, this constant cannot exceed π.

0 ≤ ∆φ ≤ π

The expression of a real phase modulated signal is:

x(t ) = A0 cos (ω0t + (∆φ ) s (t ) )
Frequency modulation, on the other hand, is a modulation system
where the frequency deviation is made proportional to the information
signal.
d (t ) = k f sɶ (t ) = (∆f ) s (t )

The constant ∆f = k f sɶ (t ) max is called the maximum frequency

deviation. The instantaneous frequency is:

f (t ) = f 0 + (∆f ) s (t )

33
In order to have always a positive frequency, we must have ∆f ≤ f 0 .
t
The instantaneous phase deviation is: φ (t ) = 2π (∆f ) ∫ s (λ )d λ . The

lower bound of the integral is not indicated to take into account any
initial phase. Sometimes, the lower bound is assumed to be −∞. So,
the expression of a frequency modulated signal is:

( t
x(t ) = A0 cos ω0t + 2π (∆f ) ∫ s (λ )d λ )
If we look at the relations that exist between the phase and the
frequency, we remark that the two modulations are related. In fact, we
can built a frequency modulator using a phase modulator, a frequency
demodulator using also a phase demodulator and vice versa.

x(t )
s (t ) Phase
Integrator Modulator
Frequency
Modulator

x(t )
s (t ) Frequency
Differentiator Modulator
Phase
Modulator

34
 x(t ) Phase s (t )
Demodulator Differentiator
Frequency
Demodulator

x(t ) Frequency s (t )
Demodulator Integrator
Phase
Demodulator

The above four figures show how we can build one type of modulator
or demodulator using the other.
Exponential modulation is a highly nonlinear modulation. This means
that it is very hard to relate the spectrum of the modulated waveform
with the one of the baseband as we did with the linear modulations. So,
an analysis of the modulated signal in the frequency domain is quite
difficult in the general case.
There are two special cases where this analysis is not very
complicated: the narrowband phase and frequency modulation where
the phase deviation is very small and the sinusoidal modulation where
the baseband signal is sinusoidal.
Narrowband phase and frequency modulation
The modulated signal in this case has the general shape of:

35
 x(t ) = A0 cos (ω0t + φ (t ) ) along with φ (t ) max << 1

Developing the cosine, x(t) becomes:

x(t ) = A0 cos φ (t )cos ω0t − A0 sin φ (t )sin ω0t

Using the fact that φ (t ) max << 1 , we have: cos φ (t ) ≈ 1 and

sin φ (t ) ≈ φ (t ) , giving: x(t ) = A0 cos ω0t − A0φ (t )sin ω0t .

A0φ(t)

A0

The above phasor diagram illustrates that the signal x(t) is the
projection on the real axis of the sum of two phasors rotating at the
same speed ω0 and making an angle of 90° between them. We see that
this phase shift produces the phase modulation. Furthermore, the
Fourier transform of the expression of x(t) can be evaluated. So, if
Φ ( f ) = F [φ (t )] , then

A0 A
X(f ) = [δ ( f − f0 ) + δ ( f + f0 )] − 0 [Φ( f − f0 ) − Φ( f + f0 )] .
2 2j
If the signal is PM, then φ (t ) = (∆φ ) s (t ) giving Φ ( f ) = (∆φ ) S ( f ) . So,
if the signal is bandlimited to W, then the PM signal will be limited to
a bandwidth B = 2W.
t
If the signal is FM, then φ (t ) = 2π (∆f ) ∫ s(λ )d λ giving

(∆f )
Φ( f ) = S ( f ) . Here also the bandwidth of the FM signal is 2W.
jf

36
Sinusoidal modulation
The other case that has a simple analytic expression is when the
modulating signal is sinusoidal. When a signal is sinusoidal, its
derivative is also sinusoidal. So, we can use the same analysis for both
frequency and phase modulation. The modulated signal in both cases
is x(t ) = A0 cos (ω0t + φ (t ) ) . φ (t ) = (∆φ ) s(t ) for PM and
t
φ (t ) = 2π (∆f ) ∫ s(λ )d λ for FM.

For FM modulation, we assume that s (t ) = cos ωm t . This gives:

2π (∆f ) ∆f
φ (t ) = sin ωmt . β = is called the modulation index. So,
ωm fm
x(t ) = A0 cos (ω0t + β sin ωm t ) .

For PM modulation, the modulating signal is s (t ) = sin ωmt . The

instantaneous phase deviation becomes: φ (t ) = (∆φ )sin ωmt . In this
case, the modulation index is β = (∆φ ) and we obtain the same
expression. So, for both cases, the expression of the modulated signal
is:
x(t ) = A0 cos (ω0t + β sin ωmt ) = A0 Re exp ( j β sin ωmt ) exp ( jω0t ) 

In the above expression, the function exp ( j β sin ωmt ) is periodic with

2π
a period Tm = . It can be developed in Fourier series. The
ωm
development is:
+∞
exp ( j β sin ωmt ) = ∑J n ( β )exp ( jnωmt )
n =−∞

37
The Fourier coefficients J n ( β ) are the Bessel functions of the first

kind of order n and argument β. These functions are tabulated and can
be easily computed. They appear as solutions of differential equations.
For positive order, we can use the following Mc Lauren series:

β  (−1) k  β 
n ∞ k

J n (β ) =  
2
∑  
k =0 k !( n + k )!  2 

and when n is negative, we use the following relation:

J − n ( β ) = (−1) n J n ( β )
The following figure shows the behavior of the first 6 Bessel functions.

They look like damped sinewaves. From the Mc Lauren series we can
deduce their properties for β around 0. For very small values of β, we
have the following approximations:
J 0 (β ) ≈ 1

1β 
n

J n ( β ) ≈   for n > 0
n!  2 
This means that the only functions that we should consider around
β
zero are J0 and J1. So, for β < 0.1, J 0 ( β ) ≈ 1 and J1 ( β ) ≈ .
2

38
The following table gives the value of the first Bessel functions.

The exponentially modulated signal is:

+∞
x(t ) = A0 ∑ J n ( β ) cos (ω0 + nωm ) t 
n =−∞

We see that the signal contains a large number of components around

the carrier frequency. For example, let us assume β = 1. Then only J0,
J1 and J2 are significantly different from zero. We can then write:
2
x(t ) ≈ A0 ∑ J n (1) cos [ (ω0 + nωm )t ]
n =−2

The modulated signal is the sum of five sinewaves. The spectrum is

displayed below. The values of the different amplitudes are read from
the above table.

39
 f0 − fm

f0 − 2 fm f0 f0 + fm f0 + 2 fm

Single sided spectrum of the signal

The above spectrum is approximately bandlimited. For the
exponentially modulated signal, we can use as transmission bandwidth
the band of frequency that contains most of the power of the signal.
The total power of the transmitted waveform can be computed as
follows. From the series development of x(t), we obtain:
2
 +∞ 
P =< x (t ) >=< A  ∑ J n ( β )cos (ω0 + nωm ) t   >
2 2
0
 n=−∞ 
If the different sinewave are independent, the total power is the sum of
the individual powers. We obtain:
A02 +∞ 2 A02
P= ∑ J n (β ) = 2
2 n=−∞
We have used the following property of the Bessel functions:
+∞

∑J
n =−∞
2
n (β ) = 1

40
If we keep n components on each side of the carrier, we obtain:

A02 n 2
P ( n, β ) = ∑ J k (β )
2 k =− n
The ratio of this power to the total power is:
P ( n, β ) n
= ∑ J k2 ( β )
P k =− n

This ratio is very close to 1 ( ≈ 0.95 ) for n =  β + 1 . So, the band of

frequency components containing the sidebands from f 0 − nf m to

f 0 + nf m can be used as an approximation for the bandwidth of the
exponentially modulated waveform for sinusoidal modulating
waveform. So the transmission bandwidth is approximately:
B = 2nf m = 2( β + 1) f m
∆f
For FM signal, β = , we obtain:
fm
B = 2(∆f ) + 2 f m
The above rule is called the "Carson's Rule". It has been found
empirically that this rule can be applied even if the modulating signal
is not sinusoidal. In this case, we can define the "Deviation Ratio";
∆f
δ= where W is the bandwidth of the baseband modulating signal.
W
Carson's rule generalizes to B = 2(δ + 1) f m = 2(∆f ) + 2W . For

example, broadcast FM uses a value of ∆f = 75 kHz for transmitting a

baseband signal having a bandwidth W = 15 kHz. This gives a
transmission bandwidth B = 2×75 + 2×15 = 180 kHz. The normalized

41
bandwidth is 200 kHz. Carson's rule underestimates it, but the error is
small.
A general remark about FM and PM is that the bandwidth of the
resulting signal is in general much larger than 2W. This is due to the
fact that these modulations are highly non-linear. Another remark
about FM is that it is very resistant to perturbations induced by noise
and interference. So, we can say that FM protects the information of
the signal at the expense of a bandwidth increase.
β
When β is very small, we can use J 0 ( β ) ≈ 1 and J1 ( β ) ≈ to express
2
the modulated signal:
β β
x(t ) = A0 cos ω0t + A0 cos (ω0 + ωm ) t  − A0 cos (ω0 − ωm )t ) 
2 2
giving x(t ) = A0 [ cos ω0t − β sin ωmt sin ω0t ] = A0 [ cos ω0t − β s (t )sin ω0t ]
which is the narrow band approximation.
Filtering the FM signal
Being a non linear modulation, the usual method of filtering the
complex envelop of the FM signal by the equivalent lowpass filter
does not work for general filter shapes. In some specific cases, this
technique can be used. In order to use it, this FM signal must be
bandpass. In this case, the complex envelop is easily extracted.
x(t ) = A0 cos (ω0t + φ (t ) ) giving a complex envelop

mx (t ) = A0 exp ( jφ (t ) ) . One case where this technique can be used is

the case of a filter with an amplitude response of the type:

42
H ( f ) = M ( f ) exp ( jθ ( f ) ) where M ( f ) = M 0 + k ( f − f 0 ) for f > 0

and θ ( f ) = −2πτ g ( f − f 0 ) + θ 0 . The equivalent lowpass filter is:

H lp ( f ) = [ M 0 + kf ] exp ( − j 2πτ g f ) exp( jθ 0 ) . In the frequency domain,

the complex envelop of the output is:

M y ( f ) = M x ( f ) H lp ( f ) = [ M 0 + kf ] exp ( − j 2πτ g f ) exp( jθ 0 ) M x ( f )

So:
M y ( f ) = M 0 exp ( jθ ) M x ( f ) exp ( − j 2πτ g f )
+ kf exp ( jθ ) M x ( f )exp ( − j 2πτ g f )

 dx(t ) 
Given that F  = 2π jfX ( f ) , the multiplication by f in the
 dt 
frequency domain becomes a differentiation in the time domain. So,
 k dmx (t − τ g ) 
my (t ) =  M 0 mx (t − τ g ) +
π  exp ( jθ )
 2 j dt 
From the expression of the input complex envelop, we obtain:
dmx (t − τ g ) dφ (t − τ g )
= jA0 exp ( jφ (t ) ) and finally:
dt dt
 k dφ (t − τ g ) 
my (t ) = A0  M 0 +
π  exp ( jφ (t − τ g ) ) exp ( jθ )
 2 dt 
The output signal is then:
 k dφ (t − τ g ) 
y (t ) = A0  M 0 +
π  cos (ω0t + φ (t − τ g ) + θ )
 2 dt 
dφ (t )
If the signal is FM, = 2π (∆f ) s (t ) . In this particular case:
dt

43
 (
y (t ) = A0  M 0 + k (∆f ) s (t − τ g )  cos ω0t + 2π (∆f ) ∫
t −τ g
s (λ ) d λ + θ )
We remark that the output signal has two different modulations: FM
and AM. The information signal is contained in the envelop of the
output signal. So, an envelop detector will demodulate the signal.
If the filter has a different transfer function, we can use the concept of
"quasi-static" approximation. If the carrier frequency is much higher
than the baseband modulating frequency, then we can safely assume
that the frequency is constant over a quite long time. The FM signal
will behave almost like a constant frequency sinewave. We know that
if the input of a filter with transfer function H(f) is a pure sinewave
A0 cos ( 2π f 0t ) , the output is also a sinewave at the same frequency:

y (t ) = A0 H ( f 0 ) cos  2π f 0t + Arg [ H ( f 0 ) ] . In the quasi- static

approximation, we replace f 0 by the instantaneous frequency f (t ) .

So, given f (t ) = f 0 + (∆f ) s (t ) , we obtain:

y (t ) = A0 H ( f 0 + ( ∆f ) s (t ) cos  2π f 0t + 2π ( ∆f ) ∫ s (λ )d λ + Arg [ H ( f 0 + ( ∆f ) s (t ) )]

t

 
Example:
Using H ( f ) = M ( f ) exp ( jθ ( f ) ) with M ( f ) = M 0 + k ( f − f 0 ) for

f > 0 along with θ ( f ) = −2πτ g ( f − f 0 ) + θ 0 , we obtain:

H ( f (t )) = M 0 + k (∆f ) s (t ) and Arg [ H ( f (t )) ] = −2πτ g (∆f ) s (t ) + θ 0 .

We see that we obtain the same amplitude variation as in the previous

case (except for a time delay).

44
FM through nonlinear system
Consider the following memoriless nonlinear system.

x(t ) T [.] y (t )

If the input is x(t ) = A0 cosθ (t ) , the output will be

y (t ) = T [ A0 cos θ (t )] . The output signal T [ A0 cosθ (t )] is not periodic

as a function of the variable t, it is periodic, with a period 2π if we

consider it function of θ. We can thus develop the output y in Fourier
series.
a0 +∞
y (t ) = + ∑ an cos nθ (t )
2 n=1
The Fourier series is a cosine series because the input is A0 cosθ . It is

even. So, T [ A0 cosθ ] is also even. The coefficients are given by:

1 π
an =
π ∫ π T [cosθ ] cos nθ dθ
−

Replacing the argument θ by its value leads to:

a0 +∞
y (t ) = + ∑ an cos [ nω0t + nφ (t ) ]
2 n=1
The above signal is a superposition of an infinite number of
exponentially modulated waveforms. For the FM case, we can write:
a0
y (t ) =
2 ( t
) ( t
+ a1 cos ω0t + 2π (∆f ) ∫ s (λ )d λ + a2 cos 2ω0t + 2π (2∆f ) ∫ s (λ )d λ + ⋯ )
We can remark that the output is a sum of FM signals at carrier

45
frequencies nf0 each one with a maximum deviation n∆f. If the
different spectra do not intersect, we can select one of them using a
bandpass filter tuned at nf0 and obtain at the output:

( t
z (t ) = an cos nω0t + 2π (n∆f ) ∫ s(λ ) d λ )
FM Generation
Direct Method
The FM signal can be generated directly using a Voltage Controlled
Oscillator (VCO) like the one used in the lab generators (GW-Instek
GFG8255A). The output signal is a sinewave with an instantaneous
frequency given by f = f v + km vin . The frequency fv is called the free
running frequency and the constant km is called the VCO gain (It is
measured in Hz/Volts). In general, these VCOs can use a variable
reactance in a parallel RLC circuit used to tune an oscillator. We can
use "varactors" for example. The output frequency of this type of
oscillator is the resonant frequency of the RLC circuit.
1
f = where C = C0 − cvin . The input voltage is proportional to
LC
1
−
1 1  c  2
s(t). vin = Vin max s (t ) , so: f (t ) = = 1 − C vin (t )  .
LC LC0  0 
1
c 1
<< 1 , we can use the approximation [1 − ε ] 2 ≈ 1 + ε .
−
If vin (t )
C0 max
2

 c 
The instantaneous frequency is given by: f ≈ f v 1 + vin  . The free
 2C0 

46
 1 cf
running frequency is f v = and the VCO gain is km = v . The
LC0 2C0

maximum frequency deviation is ∆f = Vin max km . There exist a large

number of VCO circuits. The most common ones (the ones that are
found in integrated circuits and in signal generators) generate
triangular waves using integrators or capacitors charged by controlled
current sources. A good reference is "K. K. Clarke & D. T. Hess,
Communication Circuits: Analysis and Design, 2nd ed. Krieger Pub
Co, 1994" which is the textbook for the communication circuit course.

(The part between the two ♦ is for reading only)

Frequency mixing
In this part, we introduce an important technique used in receiver and
transmitter design: Frequency mixing. The mixer is a device capable
of changing a carrier frequency for any type of modulation. It is based
on the frequency translation theorem of Fourier theory. We start first
with real signal mixing.
Consider a general bandpass signal xrf (t ) = r (t ) cos ωrf t + φ (t )  . If we

multiply this signal by a sinewave xlo (t ) = B cos ωlot , the result is:

z (t ) = xrf (t ) × xlo (t ) = Br (t ) cos ωrf t + φ (t )  cos ωlot

Using trigonometric identities, we see that this signal is the sum of

two bandpass signals:

r (t )cos (ωrf + ωlo ) t + φ (t )  + r (t )cos (ωrf − ωlo ) t + φ (t )  .

B B
z (t ) =
2 2

47
Using a bandpass filter tuned at either the sum or the difference
frequency, we obtain a bandpass signal having the same complex
envelop (i.e. the same information) but a different carrier frequency.
The new frequency is usually called "intermediate frequency" fif.
If fif = f rf + flo , we say that we are doing "up mixing". On the other

hand, if fif = f rf − flo or fif = flo − f rf , we say that we are performing

down mixing.
♦The "mixer" is an important electronic subsystem in any
communication receiver or transmitter. It is the basic building block of
the "superheterodyne" receiver. This concept of receiver was
introduced in order to solve the very complex problem of amplifying
and selecting one radio station among a large number of stations
transmitting at different frequencies.
The first solution that comes to mind is to use a "tunable" bandpass
filter. However, the construction of a very selective tunable bandpass
filter is very complex. Furthermore, due to component aging, such
system is prone to random changes and mistuning after a certain time.
It is much easier to build a fixed frequency very selective filter. So,
instead of translating the center frequency of a tunable filter before the
different signals, it is much easier to translate the frequency of the
signals before the center frequency of a fixed bandpass filter. This is
the concept of the super heterodyne receiver. The superheterodyne
receiver is composed of a tunable local oscillator ganged with a wide
band tunable RF amplifier, a mixer and a fixed frequency IF amplifier.
It is built using the block diagram shown below.

48
Let us assume we are using down mixing. The rf amplifier pre-selects
a band of frequencies containing a small number of stations around
the station at frequency fc. The bandwidth BRF is large compared to the
bandwidth B required by the modulation used (FM, AM, any linear
one) but smaller than 2fIF, the intermediate frequency. Using down
mixing, we must have:
f IF = f c − f1 giving f c = f1 + f IF or f IF = f1 − f c giving f c = f1 − f IF .
From the above two relations, we see that if the rf filter does not
exist, then we can receive two different stations if we simply use the
local oscillator for tuning. These two stations are separated by 2fIF.
These two frequencies are called "image frequencies". The job of the
tunable rf amplifier is to eliminate one of them so that it will not
interfere with the station that we want to receive. Intermediate
frequency for broadcast receivers has been standardized to the values
of 455 kHz for AM and 10.7 MHz for FM.

49
Another technique is used to avoid this problem of image frequencies.
It is based on frequency translation using complex phasors. Consider
the high frequency signal: xrf (t ) = r (t ) cos ωrf t + φ (t )  . The associated

analytic signal is: xrf + (t ) = xrf (t ) + jxˆrf (t ) = r (t )exp ( jφ (t ) ) exp ( jωrf t ) .

If we multiply this signal by the phasor: exp ( − jωlot ) , we obtain the

( )
signal: z (t ) = r (t )exp ( jφ (t ) ) exp j (ωrf − ωlo ) t . The real part is:

xif (t ) = Re [ z (t )] = r (t ) cos (ωrf − ωlo ) t + φ (t )  . This is the correct

translated signal. So, the process of performing the above operation is:
xif (t ) = Re ( xrf (t ) + jxˆ (t ) ) ( cos ωlo t − j sin ωlot )  giving:

xif (t ) = xrf (t )cos ωlot + xˆrf (t )sin ωlot

This leads to the following block diagram:

cos ωlot

π
2

π
2
Imageless Mixer
The above circuit can be used without any image rejection filter before.
♦

50
Indirect FM generation
This technique of FM generation is the one that is commonly used in
FM transmitters. This is due to the fact that the carrier frequency and
the maximum frequency deviation can be set with high precision. It is
based on a Narrow Band Frequency modulator cascaded with
nonlinear amplifiers that are used as frequency multipliers. Mixers are
also used to translate the carrier because frequency multiplication
leads usual to impractically high carrier frequencies. A general block
diagram of such system is:

∫ (.)du
s (t ) t
NBPM × N1 ×N 2

cos 2π f1t cos 2π f 2t

The frequency multipliers are implemented using a nonlinear

amplifier (Class C) followed by a narrow bandpass filter tuned at the
proper harmonic.
If we consider the above block diagram, the carrier frequency is given
by f 0 = N 2 ( N1 × f1 − f 2 ) or f 0 = N 2 ( f 2 − N1 × f1 ) . If the frequency

deviation at the output of the NBFM is ∆f1 , the final deviation is

∆f = N1× N 2 × ∆f1 . In general the frequency multiplication is
achieved by a cascade of frequency doublers and triplers. It is
impossible to achieve an efficient amplifier if the multiplication factor
is larger than three.

51
FM Demodulation
FM demodulation by differentiation (FM to AM conversion):
If we compute the derivative of an FM signal, we obtain:
d  dφ 
 A0 cos (ω0t + φ (t ) )  = − A0  ω0 +  sin (ω0t + φ (t ) )
dt  dt 
dφ
Since the signal is an FM one, = 2π (∆f ) s (t ) , we get:
dt
d
dt (
 A0 cos (ω0t + φ (t ) )  = − A0 (ω0 + 2π (∆f ) s (t ) ) sin ω0t + 2π (∆f ) ∫ s (λ )d λ
t
)
We can see that the output of a differentiation circuit will produce a
modulation of the amplitude. This modulation can be detected using
any AM demodulator.
The following circuit is a differentiator built using an OP-Amp
simulated using Multisim.
XSC1
G

A B

R1

1kΩ

C1 U1
1 2

V1 1nF
5V
FM
1kHz 100 Hz OPAMP_3T_VIRTUAL
0

52
The above picture displays the output signal. We clearly see the two
modulations: AM and FM. The signal produced by the FM source is a
sinewave modulated FM signal with β = 5, carrier frequency = 1 kHz
and modulating frequency = 100 Hz.
♦Differentiator using an op-amp
i1
R

2
C
3 1

i1
Vin 0 Op-Amp

Vout

The Operational Amplifier is an amplifier that possesses a very high

gain, differential input and very high input impedance. If the voltage

53
at the inverting input is v− , the input at the non-inverting input is v+

and the gain of the op-amp is G, its output is given by G ( v+ − v− ) . G

is assumed very large. This means that a very small difference will
produce a measurable output. We can safely assume that this
difference is zero. At that time, the inverting input is practically at
ground in the above schematic. This implies that the capacitor is in
dVin
parallel with Vin. So, i1 = C . The input impedance of the op-amp
dt
is very large. The same current will flow though the resistance R. So,
Vout = − Ri1 . After the elimination of i1, we obtain:
dVin
Vout = − RC
dt
♦
There exist a large number of other FM demodulators. The interested
student should consult the previous reference (Clarke & Hess).
The Phase Locked Loop
The phase locked loop (PLL) is a feedback system composed
principally of a voltage controlled oscillator (VCO), a phase detector
(PD) and a lowpass filter (LPF). The phase detector is usually
modeled as a multiplier.

xc (t )
LPF Ka
y (t )
v(t )

VCO

54
The input signal is xc (t ) = Ac cosθ c (t ) and the output of the VCO is
v(t ) = Av cosθ v (t ) . The output of the multiplier is
Ac Av
z (t ) = cos (θ c (t ) − θ v (t ) ) + cos (θ c (t ) + θ v (t ) )  . The lowpass filter
2 
eliminates the sum term and filters the difference term. So, we can
consider that the output of the lowpass filter (after amplification) is
A A 
y (t ) = K a h(t ) ∗  c v cos (θ c (t ) − θ v (t ) )   .
 2 
π
Let us introduce a variable ε such that: θ v (t ) = θ c (t ) − ε (t ) + . The
2
output of the lowpass filter becomes:
A A 
y (t ) = K a h(t ) ∗  c v [sin ε (t ) ]
 2 
We have a one to one relationship between y(t) and ε(t) if
π π
− ≤ ε (t ) ≤ . The relation becomes linear if ε << 1. To simplify the
2 2
analysis of the PLL and eliminate the effect of the amplitudes, let us
make Ac = 2 and Av = 1. The input signal has a carrier frequency f0 .
This makes θ c (t ) = ω0t + φ (t ) . The VCO free running frequency fv is
shifted from f0 by an offset f ∆ : f v = f 0 − f ∆ . The instantaneous phase
π π
of the VCO is: θ v (t ) = 2π f v t + φv (t ) + . The constant is added in
2 2
order to introduce the variable ε(t) in the following expressions. The
instantaneous phase deviation φv (t ) is produced by the output signal

55
 •
y(t): φ v (t ) = 2π K v y (t ) 4. The constant Kv is the VCO gain expressed in

Hz/Volts. Using the definition of ε(t), we can write:

π
ε (t ) = θ c (t ) − θ v (t ) +
2
Replacing each phase, we obtain:
ε (t ) = 2π f ∆ t + φ (t ) − φv (t )
After differentiation:
• •
ε (t ) = 2π f ∆ + φ (t ) − 2π K v y (t )
However, y(t) is the output of the lowpass filter amplified by Ka. So:
+∞
y (t ) = K a ∫ h(τ )sin ( ε (t − τ ) ) dτ
−∞

The PLL is thus governed by the following integro-differential

equation:
• +∞ •
ε (t ) + 2π K ∫ h(τ )sin ( ε (t − τ ) ) dτ = φ (t )
−∞

K = KaKv is called the "Loop Gain". The above equation is a nonlinear

equation that is quite complex to solve. In our course, we are going to
consider two different cases: The first order loop (with no filtering) to
analyze in a simple manner the "locking mechanism" and the linear
approximation when sin ε ≈ ε .
Frequency Acquisition
Consider a filter with a transfer function H ( f ) = 1 over the frequency
band of interest. The impulse response will be a Dirac impulse:
h(t ) = δ (t ) and we will have:

4
A dot above a function means that the function is differentiated.

56
 +∞
∫ h(τ )sin ( ε (t − τ ) ) dτ = sin ( ε (t ) )
−∞

We obtain the following first order differential equation:

• •
ε + 2π K sin ε = φ (t ) + 2π f ∆
Let us assume that we apply the signal xc (t ) = 2cos (ω0t + φ0 ) at the

time t = 0. In this case, φ (t ) = φ0 and its derivative is zero. We also

assume that the free running frequency of the VCO is different from
the carrier we want to lock on ( f ∆ ≠ 0 ). The differential equation
•
becomes ε + 2π K sin ε = 2π f ∆ for t ≥ 0. This equation can be written
as:
•
ε f
= − sin ε + ∆
2π K K
This equation relates the derivative of ε with ε. The set of points
 ε• , ε  is called the phase plane.
  In our case, it is better to use
 
 ε• 
 , ε  . This describes a single curve with t as a parameter. As t
 2π K 
 
varies, the point is going to move along the curve shown below. The
•
points on the curve where ε = 0 are called "equilibrium points". We
distinguish two different types of equilibrium points: A point is called
stable if after a small perturbation around the point, the trajectory in
the phase plane will go back to the equilibrium point. The point is

57
called unstable if after a small perturbation, the trajectory will go
away from the equilibrium point.
The following mechanical analogy will show the difference between
the points.

Stable Equilibrium
Unstable Equilibrium

If a trajectory starts at an equilibrium point, it will remain there.

•
ε
2π K

f∆
K
ε

Stable Equilibrium Point Unstable Equilibrium Point

Phase Plan Plot

 ε• 
The above curve shows the locus of the points  , ε  as t goes
 2π K 
 
from zero to infinity. The trajectory starts at an initial point ε0 and it
will move. It is easy to see that for any starting point, the trajectory
will move toward a stable equilibrium point. However, we can have

58
equilibrium points if and only if the curve intersects the ε axis. This is
f∆ •
possible if ≤ 1 . So, if this condition is satisfied, then lim ε (t ) = 0
K t →∞

and the steady state value of ε will be:

f∆
ε st = sin −1
K
The steady state value of y(t) will be:
f∆
yst = K a sin ε st =
Kv
π
and since θ v (t ) = θ c (t ) − ε (t ) + , the steady state output of the VCO
2
will be:
 π
vst (t ) = cos  ω0t + φ0 − ε st + 
 2
If the loop gain is very large, then εst will be very small. The VCO
output will have exactly the same frequency as the input signal with a
π
phase shift that is practically .
2
f∆
If > 1 , it is impossible to have a lock. The trajectory will not
K
intersect the ε axis and there is no equilibrium point and no steady
state solution. The VCO frequency will keep changing.
In order to have a lock, the VCO free running frequency must be in
the following range: [ f 0 − K , f 0 + K ] . This range is called the "lock
range". The value found is valid for a first order loop only. It will be
different for another filter.

59
FM Demodulation
In order to analyze the PLL when the input is frequency modulated,
we assume that the VCO free running frequency is the same as the
input carrier frequency. This means that f ∆ = 0 . We also assume that ε

remains small all the time. We can replace sinε by ε in our analysis.
With these hypotheses, we have:
xc (t ) = 2cos (ω0t + φ (t ) )

 π
v(t ) = cos  ω0t + φv (t ) + 
 2
+∞ +∞
and y (t ) = K a ∫ h(τ )sin ( ε (t − τ ) ) dτ ≃ K a ∫ h(τ )ε (t − τ )dτ
−∞ −∞

ε (t ) = φ (t ) − φv (t )
Since y is linearly related to ε and φv is also linearly related to y, it is
more interesting to use the phase deviations as primary variables
instead of using xc and v. the different relationships are better
described by the following block diagram.

φ (t ) + ε (t )
LPF Ka
y (t )
−
φv (t )

VCO

The above block diagram is completely linear. The VCO transfer

function is given by the following relation:

60
 t
φv (t ) = 2π K v ∫ y (u )du

This gives the following relation in the frequency domain:

Kv
Φv ( f ) = Y( f )
jf
The closed loop transfer function (in the frequency domain) is:
1 jfH ( f )
Y( f ) = Φ( f )
Kv H ( f ) + j f
K
where H ( f ) = F [ h(t ) ] .
t
If the input is an FM wave, φ (t ) = 2π∆f ∫ s (u )du , then

∆f
Φ( f ) = S( f )
jf
The transfer function becomes:
∆f H( f )
Y( f ) = S( f )
Kv H ( f ) + j f
K
We see that the output signal is the baseband information signal
filtered. If the loop gain is very high, the output signal will be
proportional to s(t).
∆f
y (t ) ≈ s (t )
Kv
So, if we can make sure that the error signal is small at all times, the
PLL can be used with advantage as a frequency demodulator.
Another important application of the PLL is the implementation of
frequency synthesizers.

61
Frequency Synthesis
When a PLL is locked, the frequencies of the signals arriving at the
two inputs of the phase detector (multiplier) are equal.

f1
÷ N1 LPF Ka
y (t )

f2
÷N2 VCO

The blocks labeled ÷Nk are frequency dividers (usually implemented

by presettable logic counters). At the inputs of the phase detector
(assuming that the PLL is locked), we can write:
f1 f
= 2
N1 N 2
N2
So, the VCO will produce: f 2 = f1 . This means that we can
N1
produce a signal with a frequency that can be set using digital
hardware and that can be very stable if the reference oscillator
producing f1 is very stable. This technique of frequency generation is
commonly used in modern receivers. The PLL is usually built in a
microcontroller.

62