Eetop - CN - Digital Communication Systems Using MATLAB and Simulink

Contents
http://astro.temple.edu/~silage/
Digital Communication Systems Using MATLAB and Simulink
Bookstand Publishing 2009
MATLAB and Simulink models of baseband and bandpass modulation and demodulation systems, time,
frequency and code division multiplexing, synchronization and equalization, channel models, and baseband and
bandpass signal sampling in digital communication system design.
Embedded Design Using Programmable Gate Arrays

Bookstand Publishing, 2008
Embedded design with behavioral synthesis controller-datapath models in Verilog, C-to-Verilog translation, Xilinx
LogiCORE blocks and the PicoBlaze 8-bit soft-core processor utilizing the Xilinx Spartan-3E field programmable
gate array evaluation boards, the Xilinx ISE WebPACK and applications in digital signal processing,
communications and control.
http://astro.temple.edu/~silage/digitalcommSVU.htm
Digital Communication Systems Using SystemVue
Da Vinci Engineering Press, Thomson Delmar Publishing 2006
SystemVue models of baseband and bandpass modulation and demodulation systems, time, frequency and code
division multiplexing, synchronization and equalization, channel models, and baseband and bandpass signal
sampling in digital communication system design.
Syllabus
Chapter 1 - Introduction
Chapter 2 - Frequency Domain Analysis
Chapter 3 - Digital Baseband Modulation Techniques
Chapter 4 - Receiver Design
Chapter 5 - Digital Bandpass Modulation and Demodulation Techniques
Chapter 6 - Analog Modulation and Demodulation
Chapter 7 - Multiplexing Techniques
Chapter 8 - Analog-to-Digital and Digital-to-Analog Conversion
Download Complete MATLAB and Simulink models to accompany Digital Communication Systems using
MATLAB? and Simulink? can be downloaded in ZIP archive format here(~3 MB, digicommMS1.zip). The
ZIP archive files are password protected as described in Appendix A of the text.
Table of Contents
Chapter 1：Communication Simulation Techniques
Capabilities and Limitations of Simulation
Introduction to MATLAB? and Simulink?
Model Window
Temporal Display
Spectral Display
Correlation Display
Blocksets and Blocks
Data Types
Modulation
Analog Amplitude Modulation
Simulation of Coherent AM
Simulation of Noncoherent AM
Summary
References
Chapter 2：Baseband Modulation and Demodulation
Rectangular Pulse Amplitude Modulation
Simulation of Rectangular PAM
Rectangular PAM Power Spectral Density
Performance of Rectangular PAM in a Simple Receiver in AWGN
Performance of Filtered Rectangular PAM in a Simple Receiver in AWGN
Sinc Pulse Amplitude Modulation
Simulation of Sinc PAM
Sinc PAM Power Spectral Density
Performance of Sinc PAM in a Simple Receiver in AWGN
Raised Cosine Pulse Amplitude Modulation
Simulation of Raised Cosine PAM
Raised Cosine PAM Power Spectral Density
Performance of Raised Cosine PAM in a Simple Receiver in AWGN
Optimum Baseband Receiver: The Correlation Receiver
Correlation Receiver for Baseband Symmetrical Signals
Probability of Bit Error for Baseband Symmetrical Signals
Performance of Symmetrical PAM for the Optimum Receiver in AWGN
Correlation Receiver for Baseband Asymmetrical Signals
Probability of Bit Error for Baseband Asymmetrical Signals
Performance of Asymmetrical PAM for the Optimum Receiver in AWGN
Multilevel (M-ary) Pulse Amplitude Modulation
Simulation of M-ary Rectangular PAM
M-ary Rectangular PAM Power Spectral Density
Correlation Receiver for M-ary Baseband Signals
Probability of Bit Error for M-ary Baseband Signals
Performance of M-ary PAM for the Optimum Receiver in AWGN
Partial Response Signaling
Duobinary PAM Signaling
Simulation of Duobinary PAM
Simple Receiver for Precoded Duobinary Signals
Simple Receiver for Precoded Modified Duobinary Signals
Duobinary PAM Power Spectral Density
Performance of Duobinary PAM in a Simple Receiver in AWGN
Delta Modulation
Simulation of Delta Modulation
Eye Diagrams
Summary
References
Chapter 3：Bandpass Modulation and Demodulation
Optimum Bandpass Receiver: The Correlation Receiver
Correlation Receiver for Bandpass Symmetrical Signals
Probability of Bit Error for Bandpass Symmetrical Signals
Correlation Receiver for Bandpass Asymmetrical Signals
Probability of Bit Error for Bandpass Asymmetrical Signals
Binary Amplitude Shift Keying
Simulation of Binary ASK
Binary ASK Power Spectral Density
Performance of Binary ASK for the Optimal Receiver in AWGN
Binary Frequency Shift Keying
Simulation of Binary FSK
Binary FSK Power Spectral Density
Performance of Binary FSK for the Optimal Receiver in AWGN
Binary Phase Shift Keying
Simulation of Binary PSK
Binary PSK Power Spectral Density
Performance of Binary PSK for the Optimal Receiver in AWGN
Multilevel (M-ary) Amplitude Shift Keying
Simulation of M-ary ASK
M-ary ASK Power Spectral Density
Correlation Receiver for M-ary ASK Signals
Probability of Bit Error for M-ary ASK Signals
Performance of M-ary ASK for the Optimum Receiver in AWGN
Multilevel (M-ary) Frequency Shift Keying
Simulation of M-ary FSK
M-ary FSK Power Spectral Density
Correlation Receiver for M-ary FSK Signals
Probability of Bit Error for M-ary FSK Signals
Performance of M-ary FSK for the Optimum Receiver in AWGN
Multilevel (M-ary) Phase Shift Keying
Simulation of M-ary PSK
M-ary PSK Power Spectral Density
Probability of Bit Error for M-ary PSK Signals
Performance of M-ary PSK for the Optimum Receiver in AWGN
Quadrature Amplitude Modulation
Simulation of QAM
QAM Power Spectral Density
Probability of Bit Error for QAM Signals
Performance of QAM for the Optimum Receiver in AWGN
Differential Phase Shift Keying
Simulation of DPSK
DPSK Power Spectral Density
Probability of Bit Error for DPSK Signals
Performance of DPSK for the Optimum Receiver in AWGN
Noncoherent Demodulation of Binary Frequency Shift Keying
Simulation of Noncoherent Binary FSK Signals
Probability of Bit Error for Noncoherent Binary FSK Signals
Performance of Noncoherent Binary FSK Signals in AWGN
Noncoherent Demodulation of Binary Amplitude Shift Keying
Simulation of Noncoherent Binary ASK Signals
Probability of Bit Error for Noncoherent Binary ASK Signals
Performance of Noncoherent Binary ASK Signals in AWGN
Threshold for Demodulation of Noncoherent Binary ASK Signals
Constellation Plots
Summary
References
Chapter 4：Sampling and Quantization
Sampling Baseband Analog Signal
Companding
Line Codes
Power Spectral Density of Line Codes
Polar NRZ Line Code
Unipolar NRZ Line Code
Alternate Mark Inversion NRZ Line Code
Split-Phase NRZ Line Code
Return-to-Zero Line Codes
Simulation of Line Codes
Pulse Code Modulation
Differential Pulse Code Modulation
Simulation of DPCM
Sampling Bandpass Analog Signals
Summary
References
Appendix A：MATLAB? and Simulink? Model File Download Procedure
Appendix B：Complementary Error (Q) Function Table
Digital Communication Systems Using MATLAB and Simulink

Dennis Silage (Author)
http://www.amazon.com/Digital-Communication-Systems-MATLAB-Simulink/dp/1589096215
Digital Communication Systems Using SystemVue (Hardcover)

Dennis Silage (Author)
http://www.amazon.com/Digital-Communication-Systems-Using-SystemVue/dp/1584508507/ref=pd_sim_b_3
MATLAB/Simulink for Digital Communication (Paperback)

Won Y. Yang (Author), Yong S. Cho (Author), Won G. Jeon (Author), Jeong W. Lee (Author), Jong H. Paik
(Author), Jae K. Kim (Author), Mi-Hyun Lee (Author), Kyu I. Lee
(Author), Kyung W. Park (Author), Kyung S. Woo (Author)

Recently Viewed Items
MATLAB/Simulink for Digital Communication by Won Y. Yang
Digital Communication Systems Using SystemVue by Dennis Silage
Mastering Simulink by James Dabney
Modeling and Simulation In SIMULINK for En... by Mohammad Nuruz...
Contemporary Communication Systems Using MATLAB (Paperback)
John G. Proakis (Author) Masoud Salehi (Author) Gerhard Bauch (Author)
EE4512 Analog and Digital Communications
Laboratory
Dr. Dennis Silage
silage@temple.edu
• Course syllabus
• Course textbooks
• Course grades
• Course objectives
• Course
textbook
• Available as:
bound hardcopy
pdf file
Kindle download
YouTube video
blog session
twitter stream
text message
Morse code .wav file
• Course
textbook
– only
kidding!
• Course textbooks
• Communication
Systems provides
the theoretical
basis
• Digital
Communication
Systems provides
simulations for
insight and
experiential laboratories
in EE4513
• Course
syllabus
• All course
materials
are also
available on
Blackboard
• In-class problems and discussions form the basis for the

examinations
• SNAP quizzes (weekly, 15 minutes, 25%) and

examinations (three, hour exams, 35% and a two hour
final exam, 40%) are benchmarks of performance
• Blackboard is used for the

course and laboratory
documents and to
communicate assignments
• MATLAB and Simulink modeling of communication

systems in-class augments the text description
• MATLAB and Simulink modeling in EE4513 Analog and

Digital Communications Laboratory
• In-class discussions,
simulation studies and
problem solutions
Temple Technical Institute

Analog Communications
Laboratory 1962
• This course is the application of electrotechnology
• Professional career path in digital data communications
• Course sequence continues in EE4542

Telecommunication Engineering
• Read the assigned text

portions and review your
notes before class
• Weekly closed-book
but open notes SNAP
quizzes to assess Edwin and Marian Armstrong
performance with a portable radio 1923
• SNAP stands for Students’ Note-taking Achieves

Performance and the 15 minute quizzes are open-notes
but closed text book.
• Your organized notes are a compilation of reading the

text and thinking
about the concepts
and are an aid to
learning.
• The SNAP concept relates to the learning process

expressed in the Chinese proverb:
I hear and I forget,

I see and I remember,
I do and I understand.
• I hear and I forget, ← Listening to the lecture

I see and I remember, ← Reading the textbook
I do and I understand. ← Compiling the notes
The notes are to be

handwritten to insure
that the process works
for you. No xerox copies
or electronic versions
or textbooks are
permitted for use
during the SNAP quiz.
Leonardo Da Vinci and

his notebooks
• The 15 minute SNAP quizzes are

simple calculations and unit
conversions (calculators are
permitted) from formulas in your
notes and conceptual answers
to posed questions to assess
your understanding of the basic
concepts.
• The SNAP quiz is the assessment

of your preparation and organizes
your thoughts as part of the
engineering method (not that
shown here!).
EE4512 Analog and Digital Communications Chapter 1
Chapter 1
Introduction
Chapter 1
Introduction
• Components of a Communication
System
• Pages 1-5
• Modulation (partial definition):

3: to vary the amplitude, frequency, or phase of (a
carrier wave or a light wave) for the transmission
of information (as by radio)
• Carrier (partial definition):

2: an electromagnetic wave or alternating current whose
modulations are used as communications signals (as
in radio, telephonic, or telegraphic transmissions
• Analog modulation:
continuous
information such as
speech or video
encoded as an
amplitude
• Digital modulation:
discrete information
such as binary data
encoded as
a frequency shift
or
a phase shift
Source Simulink subsystems

Transmitter Receiver
User
Channel
with noise
MS Figure 1.16
Analog amplitude modulation (AM)

Simulink subsystems
Source
User
Channel
with noise
Coherent
demodulation
MS Figure 1.16

Source Simulink subsystems

User
Channel
Non-coherent demodulation with noise
MS Figure 1.20

Sinusoidal input and DSB AM

Simulink subsystem
Sinusoidal input and DSB AM

Carrier
DSB modulated
spectrum
DSB AM temporal signal and power spectrum

2 x carrier
DSB demodulated frequency
Original spectrum
spectrum
Demodulated DSB AM spectrum

Random data source Bit error rate

MS Figure 3.12
Digital binary phase shift keying (BPSK)

• Input binary data, rate rb = 1 kb/sec, bit time Tb = 1 msec
Tb = 1 msec
• Output binary data with no noise, note 2 bit delay in output

• Received signal with significant additive noise
Where’s the signal?

• Output binary data with bit errors due to additive noise

End of Chapter 1
Introduction
EE4513 Analog and Digital Communications Laboratory Chapter 1
Chapter 1
Communication Simulation
Techniques
• Introduction to MATLAB and
Simulink
• Blocksets
• Simulation displays
• Simulink
Library
Browser:
Commonly
Used Blocks
• Simulink
Library
Browser:
Communications
Blockset
• Simulink
Library
Browser:
Signal
Processing
Blockset
• Simulink
Library
Browser:
Simulink
Extras
• Simulink Library Browser and the model window
MS Figure 1.2
• Source parameter block for the Random Integer

Generator block and the Simulink model window
MS Figure 1.3
• Configuration parameter window for the Simulink

simulation
MS Figure 1.4
• Sinusoid and half wave rectified sinusoid Simulink

temporal model
MS Figure 1.6
• Sinusoid and half wave rectified sinusoid temporal display

in the Simulink Figures window MS Figure 1.7
• Half wave rectified sinusoid Simulink spectral display

model
MS Figure 1.8
• Scope and axis properties parameter windows of the

Spectrum Scope block MS Figure 1.9
• Scaled power spectral display of a half wave rectified

sinusoid, fo = 1 kHz
MS Figure 1.10
• Gaussian noise source and lowpass filtered power

spectral density display MS Figure 1.11
• Gaussian noise source and LPF PSD MS Figure 1.12
Flat PSD
LPF PSD, fcutoff = 11.25 kHz

• Gaussian noise source and lowpass filtered

autocorrelation display MS Figure 1.13
• Gaussian noise source and lowpass filtered power

autocorrelation display MS Figure 1.14
Uncorrelated
Correlated
End of Chapter 1
Communication Simulation
Techniques
Chapter 2
Frequency Domain Analysis
Chapter 2
• Why Study Frequency Domain
Analysis?
• Pages 6-13
Why frequency domain

analysis?
• Allows simple algebra

rather than time-domain
differential equations
to be used
• Transfer functions can 500, 1500 and 2500 Hz

be applied to transmitter,
communication channel
and receiver
• Channel bandwidth,
noise and power are MS Figure 4.2 and Figure 4.3
easier to evaluate
Why frequency domain

analysis?
• A complex signal
consisting of the sum of
three sinusoids is
difficult to discern in the
temporal domain
but easy to identify in the 500, 1500 and
spectral domain 2500 Hz
MS Figure 4.2 and Figure 4.3

• Example 2.1 Input sum of three sinusoids, temporal

display
Butterworth LPF
9 pole, fcutoff = 1 kHz
500
Hz
1500
Hz
2500 EX21.mdl
Hz
MS Fig 4.8 modified

• Example 2.1 Input sum of three sinusoids
• Output after Butterworth LPF

• Example 2.1 Input sum of three sinusoids, temporal

display
Since this is a MATLAB and Simulink time-based

simulation, an analog low pass filter is used here
Butterworth LPF
9 pole, fcutoff = 1 kHz
EX21.mdl
MS Fig 4.8 modified

• Example 2.1 Input sum of three sinusoids, spectral

display
EX21spectral.mdl
500
Hz
1500
Hz
2500
Hz Butterworth LPF
9 pole, fcutoff = 1 kHz MS Fig 4.11 modified
• Input power spectral density of the sum of three sinusoids
34 dB
• Output power spectral density after Butterworth LPF
Attenuation (decibel dB)

–36 dB
• Butterworth Filters
Stephen Butterworth was a

British physicist who invented
the Butterworth filter, a class
of circuits that are used to
filter electrical signals. He
worked for several years 1885-1958
at the National Physical
Laboratory (UK), where he did
theoretical and experimental
work for the determination of
standards of electrical
inductance and analyzed the
electromagnetic field around submarine cables.
• Example 2.2
10 MHz sinusoid
with additive
white Gaussian
noise (AWGN)
Communications channel
EX22.mdl
• Example 2.2 Simulink design window
Configure Simulation Simulate
T = 0.01 sec
EX22.mdl
• Example 2.2 Simulink design window
Configure Simulation Tsimulation = 10-8 = 10 nsec

fsimulation = 1/Tsimulation = 108 = 100 MHz T = 0.01 sec
• Example 2.2 10 MHz sinusoid with AWGN
• Power spectral density of 10 MHz sinusoid with AWGN
10 MHz Here’s the signal!

Chapter 2
• The Fourier Series
• Pages 13-38
• Fourier Series
Jean Baptiste Joseph Fourier was

a French mathematician and physicist
who is best known for initiating the
investigation of Fourier Series and its
application to problems of heat flow.
The Fourier transform is also named
in his honor.
1768-1830
∞
s(t) = X0 + ∑ Xncos(2π n fo t + φn )
n=1
• Fourier series coefficients:

trignometric an bn
polar Xn
complex cn
X0 = a0 Xn = an2 + bn2
| c n | = Xn / 2 Xn = | 2 c n |
Xo = co
MATLAB and Simulink simulation
can provide the magnitude of
the complex Fourier series
coefficients for any periodic
waveform
• Example 2.3 modified
Temporal display
of a complex pulse
as the addition of
two periodic pulses
EX23.mdl
• Example 2.3 modified Simulink design window
Configure Simulation Simulate
T = 4 sec
EX23.mdl
Example 2.3 modified Simulink design window
Configure Simulation fsimulation = 1024 Hz

Tsimulation = 1/fsimulation = 0.976562 msec T = 4 sec
Temporal display scopes

Simulink discrete
pulse generators
EX23.mdl

fS = 1024 Hz
Tsimulation = 1/fsimulation =
0.976562 msec
First Pulse
EX23.mdl

fS = 1024 Hz
0.976562 msec
Second
Pulse
EX23.mdl
Period To = 4 sec
Fundamental frequency
fo= 0.25 Hz
Sample based simulation:
Period = 4096 samples
must be 2N for FFT
Pulse width = 1024
samples
Sample time = 0.978562
msec
4/4096 = 9.78562 x 10-4
The default display or autoscaling for the temporal scope

does not have a uniform amplitude.
EX23.mdl
Right-Click on
the vertical
axis and use
Axes properties
to change the
amplitude
scaling
• Example 2.3 modified First periodic pulse
T = 1 sec
T = 4 sec
• Second periodic pulse
T = 2 sec
• Example 2.3 modified Sum of periodic pulses
T = 1 sec T = 4 sec
T = 2 sec
EX23.mdl

Spectral display scope
Spectral display
of a complex pulse
as the addition of
two periodic pulses
fS = 1024 Hz
T = 4 sec = 4096
samples
fS = fsimulation for
conveninence
Input signal is
non-framed based
so buffer input is
required
fS = 1024 Hz
T = 4 sec = 4096
samples
Amplitude scaling
magnitude-squared
Scaled | FFT |2 using Simulink Plot Tools. In Command

Window:
>>plottools on Default spectral display
plottools
Add data point markers (o)

and change the frequency
axis to 5 Hz
Chapter 2
• Power in the Frequency Domain
• Pages 38-52
• Example 2.3 modified Unscaled | FFT |2
fs / 2 = 512 Hz
500 Hz
• Scaled | FFT |2
Spectral Resolution MS Eq 1.1

∆f = fs / N = 1024/4096 = 0.25 Hz
5 Hz
• Example 2.3 modified Scaled | FFT |2
The Fourier series components are discrete. However, the

Spectrum Scope does not read out the values. An
interactive cursor
is available in the
Figures window.
Copy and Paste

the Spectrum
Scope plot as
Figure 1
The interactive cursor can lock-on to the markers and read

out the data values sequentially. The interactive cursor is
evoked from the toolbar for a Figure plot.
x = 0 (f = 0 Hz)
y = 2362 = | FFT(f = 0) |2
The magnitude squared data values are divided by the

number of points in the FFT N = 4096 to get the value. The
magnitude of the DC (0 Hz) value is then:
(2362/4096)0.5 = (0.56767)0.5 = 0.7594 = c0 = X0
The DC value of the complex pulse is calculated as 0.75
The first five spectral components can be read, computed

and then compared to a direct calculation of cn.
f Hz Data n Component Spectral Resolution

0 2362 0 0.7594 MS Eq 1.1
0.25 1038 1 0.5034 ∆f = fs / N
0.5 103.8 2 0.1592 ∆f = 1024/4096
0.75 115.3 3 0.1678 ∆f = 0.25 Hz
1 0 4 0
• Example 2.7
Rectangular pulse
train
• Example 2.7
Amplitude = 3
Period To = 10 msec
fo= 100 Hz
Pulse width = 2 msec
Duty cycle = 2/10 = 0.2
Period = 4096 samples
Pulse width = 819 samples
819/4096 ≈ 0.2
Sample time = 2.441 µsec
10-2/4096 = 2.441 x 10-6
• Example 2.7
Period To = 10 msec
Since the Spectrum
Scope requires 2N =
4096 samples, the
simulation time =
10 msec divided by
the sample time =
2.441 µsec must be
≥ 4096
10-2/2.441 x 10-6 =
4096.68
• Example 2.7 One cycle of periodic pulse
3 Tpulse = 2 msec
DC value = 0.6 To = 10 msec
• Magnitude squared of the Fast Fourier Transform | FFT |2
| FFT (f = 0) |2 = 1447
(1447/4096)0.5 = (0.353271)0.5 = 0.5944

• Example 2.7 | FFT |2
First spectral null at 1/Tpulse = 1/2 msec = 500 Hz

Successive nulls at n/Tpulse = n x 500 Hz
Hard to discern when the nulls are when plotted on a

linear scale
500 Hz 1 kHz 1.5 kHz

First spectral null at 1/Tpulse = 1/2 msec = 500 Hz

Successive nulls at n/Tpulse = n x 500 Hz
Easier to see when the nulls are plotted on a decibel (dB)

scale where –40 dB ≈ 0
500 Hz 1 kHz 1.5 kHz
-40 dB
• de·ci·bel (dĕs'ə-bəl, -bĕl')

n. (Abbr. dB)
A unit used to express relative difference in power or
intensity, usually between two acoustic or electric signals,
equal to ten times the common logarithm of the ratio of the
two levels.
The bel (B) as a unit of measurement

was originally proposed in 1929 by
W. H. Martin of Bell Labs. The bel
was too large for everyday use, so
the decibel (dB), equal to 0.1 B,
became more commonly used. Alexander Graham Bell
1847-1922
Spectrum Scope axis

properties
• Example 2.7 | FFT |2 (Data/4096)0.5
f Hz Data Component n Calculated cn

0 1447 0.5944 0 0.6
100 1290 0.5612 1 0.561
200 844 0.4540 2 0.454 cf S&M
300 376 0.3030 3 0.303 pp 33-35
400 81 0.1406 4 0.140
500 ≈0 0 5 0
cn X0 = a0 Xn = an2 + bn2
Xo = co | c n | = Xn / 2 Xn = | 2 c n |
• Example 2.8
Rectangular pulse trains

with pulse period of 0.5 sec
and pulse widths of
0.0625 sec
0.125 sec
0.250 sec
Period = 216 = 65536
samples
Pulse widths = 8192,
16384, 32768
Sample time = 7.629 µsec EX28.mdl
0.5/65536 = 7.629 x 10-6
• Example 2.8
Spectrum Scope has
inherent amplitude and
frequency axes scaling
(but not cursor readout)
EX28.mdl
First null 16 Hz = 1/τ τ = 0.0625 sec
S&M p. 36-37

• Average Normalized (RL = 1Ω) Power

X0 = a0 Xn = an2 + bn2
Fourier series components
cn from simulation | c n | = Xn / 2 Xn = | 2 c n |
Periodic signal as a Xo = co
∞
frequency domain s(t) = X0 + ∑ Xncos(2π nfo t + φn )
representation n=1
Average (RL = 1Ω)

power in the signal PS as
a time domain or t o +T ∞ 2
1 X
frequency domain PS =
T ∫ s 2
(t) dt = X 2
0 + ∑
n=1 2
n
representation to
Parseval’s Theorem
• Average (RL ≠ 1Ω) Power
The average non-normalized (RL ≠ 1Ω) power in the

signal PS as a time domain or frequency domain
representation
∞
s(t) = X0 + ∑ Xncos(2π nfo t + φn )
n=1
to +T
1 1  2 ∞ Xn2 
PS = ∫  X0 + ∑ 
2
s (t) dt =
T RL to
RL  n=1 2 
Parseval’s Theorem
• Parseval’s Theorem
Marc-Antoine Parseval des Chênes was

a French mathematician, most famous
for what is now known as Parseval’s
Theorem, which presaged the
equivalence of the Fourier Transform.
A monarchist opposed to the French
Revolution, Parseval fled the country 1755-1836
after being imprisoned in 1792 by Napoleon for
publishing tracts critical of the government.
t o +T ∞ 2
1 X
PS = ∫ s2 (t) dt = X02 + ∑ n
T to n=1 2
• Examples 2.9 and 2.10
Normalized power
spectrum of a
rectangular pulse
and a Butterworth
low pass filtered
(LPF) rectangular
pulse
EX29.mdl
Amplitude = 3
Period To = 10 msec
fo= 100 Hz
Pulse width = 2 msec
Duty cycle = 2/10 = 0.2
Period = 4096 samples EX29.mdl
Pulse width = 819 samples
819/4096 ≈ 0.2
Sample time = 10-2/4096 = 2.441 x 10-6 = 2.441 µsec
Sampling rate = 1/ 2.441 x 10-6 ≈ 409.6 kHz
Since this is a MATLAB and Simulink sample based

simulation (period = 4096 samples, pulse width = 819
samples) a digital low pass filter design is used here (a
time-based simulation requires an analog filter design)
EX29.mdl
• Example 2.9 and 2.10 Digital Low Pass Filter Design
fs = 409600
fcuttoff = 300 Hz
• Power Spectral Density

The normalized (RL = 1 Ω) power spectral density
(PSD) for periodic signals is discrete because of the
fundamental frequency fo = 1/To = 100 Hz here.
However, for aperiodic signals the PSD is conceptually

continuous. Periodic signals contain no information
and only aperiodic signals are, in fact, communicated.
• Power Spectral Density
Simulated by | FFT |2 in
MATLAB and Simulink to
obtain cn from which we
can derive Xn
| FFT |2 ≈ PSD
X0 = a0 Xn = an2 + bn2
Xo = co | c n | = Xn / 2 Xn = | 2 c n |
• Power Spectral Density Simulated by | FFT |2

The dB scale for PSD is more prevalent for analysis
Linear
Spectral nulls
dB
• Power Spectral Density Low Pass Filtered
Unfiltered
dB
LPF
fcutoff = 300 Hz
dB
• Bandwidth
The bandwidth of a signal is the width of the frequency

band in Hertz that contains a sufficient number of the
signal’s frequency components to reproduce the signal
with an acceptable amount of distortion.
Bandwidth is a nebulous term and

communication engineers must
always define what if meant by
bandwidth in the context of use.
• Total Power in the Signal
Parseval’s Theorem allows us to determine the total

normalized power in the signal without the infinite sum of
Fourier series components by integrating in the temporal
domain:
t o +T ∞ 2
1 X
PS = ∫ s2 (t) dt = X02 + ∑ n
T to n=1 2
From Example 2.10 the total

normalized (RL = 1Ω) power in the
signal is 1.8 V2 (not W) and the percentage of the total
power in the signal in a bandwidth of 300 Hz is
approximately 88% (S&M p. 45).
Chapter 2
• The Fourier Transform
• Pages 52-69
• Example 2.12
Spectrum of a
simulated single
pulse from a very
low duty cycle
rectangular pulse
train EX212.mdl
• Example 2.12 Simulated single pulse
duty cycle = 10-3/1 = 0.001 = 0.1%

pulse width = 1 msec
pulse period = 1 sec
•| FFT |2

The magnitude of the Fourier Transform of a single pulse
is continuous and not discrete since there is no Fourier
series representation. In the MATLAB and Simulink
simulation here the data points are very dense and
virtually display a continuous plot.
• Example 2.12 Simulated single pulse | FFT |2

S&M p. 56-60 Aτ = 1(10-3) = 10-3
S(f) = Aτ sinc(π f τ) (co/65536)0.5 = 10-3
co = 0.065536
co = 0.065536

S&M p. 56-60
S(f) = Aτ sinc(π f τ)
Zero-crossing at integral multiples of 1/τ = 1/10-3 = 1000 Hz
0.005
1000 2000 3000 4000 5000 6000 7000

Hz
duty cycle = 10-2/1 = 0.01 = 1%

pulse width = 10 msec
pulse period = 1 sec
•| FFT |2

S&M p. 56-60
S(f) = Aτ sinc(π f τ)
Zero-crossing at integral multiples of 1/τ = 1/10-2 = 100 Hz
0.05
100 200 300 400 500 600 700 Hz

• Properties
of the
Fourier
Transform
• Properties
of the
Fourier
Transform
Modulation
principle
Chapter 2
• Normalized Energy Spectral Density
• Pages 60-65
• Normalized Energy
If s(t) is a non-periodic, finite energy signal (a single

pulse) then the average normalized power PS is 0:
t o +T
1 finite value
PS = lim ∫
2
s (t) dt = lim = 0 V2
T →∞ T T →∞ T
to
∞
ES = ∫
−∞
s2 (t) dt V 2 - sec
However, the normalized energy ES for the same s(t) is

non-zero by definition (S&M p. 60-61).
• Parseval’s Energy Theorem
Parseval’s energy theorem follows directly then from the

discussion of power in a periodic signal:
∞ ∞
ES = ∫ s2 (t) dt = ∫
2 2
| S (f) | df V - sec
−∞ −∞
• Energy Spectral Density
Analogous to the power spectral density is the energy

spectral density (ESD) ψ(f). For a linear, time-invariant (LTI)
system with a transfer function H(f), the output ESD which is
the energy flow through the system is:
ψOUT(f) = ψIN(f) | H(f) |2

• Energy Spectral Density
The energy spectral density (ESD) ψ(f) is the magnitude

squared of the Fourier transform S(f) of a pulse signal s(t):
ψ(f) = | S(f) |2
The ESD can be approximated by the magnitude squared

of the Fast Fourier Transform (FFT) in a MATLAB and
Simulink simulation as described in Chapter 3.
ψ(f) ≈ | FFT |2
End of Chapter 2
Chapter 3
Digital Baseband Modulation
Techniques
Chapter 3
Techniques
• Goals in Communication System
Design
• Pages 75-76
• Communication System Design

MS Figure 2.1
Source Transmitter
Channel Receiver
Noise
An information source (voice, music, video, text, signals

or images) is modulated in the transmitter .

MS Figure 2.1
Source Transmitter
Channel Receiver
Noise
The output of the transmitter is inputted to the physical

transmission medium (wires, fiber optics, wireless or
acoustic) as a channel with additive noise.

MS Figure 2.1
Source Transmitter
Channel Receiver
Noise
The output of the channel is inputted to the receiver

which attempts to demodulate the signal and recover the
data.
Chapter 2
Baseband Modulation and
Demodulation
• Rectangular Pulse Amplitude
Modulation (PAM)
• Pages 18-20
• Rectangular Pulse Amplitude Modulation (PAM)
Random Integer Generator simulates a binary information

source, bit rate rb = 1 kb/sec
MS Figure 2.1
Random Integer
Generator from
the Communications
Blockset, Comm
Sources
Random Integer
Generator block from
the Communications
Blockset, Comm
Sources
binary
random seed
rb = 1 kb/sec
Tb = 1 msec
Binary data [0,1] Binary Polar PAM transmitter ± 5 V
MS Figure 2.1
Transmit data = ([0,1] – 0.5)10 V = ± 5 V

Additive White Gaussian Noise (AWGN) Channel

provides a random noise source, mean = 0 V, variance
σ2 = 0.5 V
MS Figure 2.1
AWGN Channel
block from the
Communications
Blockset, Channels
AWGN Channel
block from the
Communications
Blockset,
Channels
random seed
variance σ2
Scope displays Transmit and Receive temporal data
MS Figure 2.1
• Output of Binary Polar PAM Transmitter
• Input of Binary PAM Receiver with AWGN

MS Figure 2.1
Simple PAM receiver consists of a Sample and Hold

block with synchronous timing at Tb /2 provided by a
pulse generator. This induces a delay of Tb /2.
/2
The Sign and Lookup Table blocks convert the

received data to the correct format (see MS p. 19).
The Sign block

converts the
variable amplitude
bipolar received
data from the
Sample-and-
Hold block to
trinary data
[–1, 0, 1]
The Lookup Table

block converts the
trinary data [–1, 0, 1]
to a replica of the
bipolar transmitted
signal [± 5] for
comparison by
mapping 0 to +5
• Input of Binary PAM Receiver with AWGN
• Output of Sample and Hold block

Tb / 2 delay
Tb
• Output of Sample and Hold block

Tb / 2 delay
• Output of Sign and Lookup Table blocks

Tb / 2 delay
• Transmitted PAM signal
• Received PAM signal with no errors

Tb / 2 delay
Chapter 3
Techniques
• Baseband Modulation Using
Rectangular Pulses and Binary
Pulse Amplitude Modulation
• Pages 76-88
• PAM generation of data sequence 00, 01, 10 and 11
00
01
Fig35.mdl 11
MS for S&M Figure 3-5
10
Pulse Generator block

outputs a –1 V pulse
with a width of 20 msec
and a nominal pulse
period of 200 msec
Transport Delay block

delays the input by
20 msec using a sample
data buffer automatically
set
00
01
Time-based
sampling,
sampling rate
fS = 50 kHz,
sampling period 11
TS = 1/ fS = 20 µsec Fig35.mdl
MS for S&M Figure 3-5 10
• Data 00 PAM Output
τ 2τ τ = 20 msec
τ 2τ τ = 20 msec
τ 2τ τ = 20 msec
τ 2τ τ = 20 msec
• Spectrum of PAM data sequence 00, 01, 10 and 11
00
01
ESD ≈ | FFT |2
11
Fig35spectrum.mdl
MS for S&M Figure 3-5 10

Spectrum Scope requires a discrete or sampled signal

input. Pulse Generator block changed to sampled-based
signal. Integer Delay block used to delay the pulse.
Fig35spectrum.mdl

Pulse Generator block

changed to sampled-based
signal.
Sample time TS = 20 µsec
Pulse period = 500 000
samples (10 sec)
Pulse width = 1000
samples (20 msec)
Integer Delay block for

sampled-based signal.
Sample time TS = 20 µsec
Delay = 1000
samples (0.2 sec)
Spectrum Scope simulation for N = 218 = 262 144

samples.
samples
Simulation time T = N / fS = 262 144 / 50 000 = 5.24288
sec
Frequency resolution ∆f = fS / N = 50 000/262 144 =
0.1907 Hz
Fig35spectrum.mdl
• Data 00 PAM simulated ESD Output | FFT |2
τ = 20 msec
1/2τ 1/τ S&M Figure 3-6a
25 Hz 50 Hz
• Data 00 PAM simulated ESD Output | FFT |2 dB

τ = 20 msec
1/τ S&M Figure 3-6b
50 Hz

τ = 20 msec
1/τ S&M Figure 3-6c
50 Hz

τ = 20 msec
1/2τ 1/τ S&M Figure 3-6d
25 Hz 50 Hz

• Sum of the equally likely simulated ESD ≈ | FFT |2 for data

00, 01, 10 and 11 unscaled by probability p = 0.25
S&M Figure 3.7

Ψ(f) = 2A2 τ2 sinc2(π f τ)
3/τ
150 Hz
2/τ
100 Hz
1/τ
50 Hz
• Power of a Series of Pulses
The power of a series of pulses is the energy in a series

of pulses divided by the time to transmit the series of
pulses (S&M p. 83).
Power = Energy / Transmit Time
This concept can be extended to find the average

normalized power spectral density (PSD) G(f) of a series
of rectangular pulses:
G(f) = n A2 τ2 sinc2(π f τ) / n τ = A2 τ sinc2(π f τ)

• The total power in a series of rectangular pulses is a

function of the bandwidth of the data transmission.
S&M Figure 3-10

Ψ(f) = 2A2 τ2 sinc2(π f τ)
3/τ
96.5%
2/τ
95%
1/τ
90%
• The rectangular pulse width τ is the entire bit time Tb

here and is optimum in the bandwidth sense.
Table 3-1 Bandwidth of a Binary Rectangular PAM

Signal as a Percentage of the Total Power (S&M p. 86,
MS p. 22) with τ = Tb = 1 / rb
Bandwidth (Hz) Percentage of Total Power

1/Tb 90%
1.5/Tb 93%
2/Tb 95%
3/Tb 96.5%
4/Tb 97.5%
5/Tb 98%
Chapter 2
Demodulation
• Rectangular PAM
Power Spectral Density
• Pages 20-22
• For an ensemble (collection) of pseudo-random

rectangular PAM data signals, the power spectral
density (PSD) is used since the bit period Tb is finite
and not infinite as for a single pulse with its ESD.
• The pulse width τ can be less than the bit period Tb but
this is not bandwidth efficient.
• The MATLAB and Simulink
simulation of a binary
rectangular PAM transmitter
(MS Figure 2.1) is modified
for a variable pulse width
τ ≤ Tb to verify these
concepts.
• Rectangular PAM with variable pulse width τ ≤ Tb = 1

msec
Pulse train, 0 to 1 V, pulse period 1 msec, pulse width

0.5 msec (50% duty cycle)
Multiplier
Transmitter of MS Figure 2.1 modified

MSFig21mod.mdl
• Rectangular PAM with

variable pulse width
τ ≤ Tb = 1 msec
Pulse train = 0 to 1 V
Pulse period = 1 msec
MSFig21mod.mdl
• Rectangular PAM with

variable pulse width
τ ≤ Tb = 1 msec
Pulse width = 50%

(0.5 msec)
Phase delay = 0.5 msec
MSFig21mod.mdl
• Top: Binary data source. Middle: pulse train. Bottom:

rectangular PAM transmit data output with 50% duty cycle.
Data
Tb
Duty Cycle
Tb /2
Product
with Gain
• PSD of rectangular PAM
Spectrum scope block calculates the PSD of the output of

the rectangular PAM transmitter with variable duty cycle
τ / Tb
Fig21modspec.mdl
• Scaled PSD of rectangular PAM τ / Tb = 0.5, Tb = 1 msec
The impulse spectral terms

δ[(2n–1) fo] where fo = 1 / Tb =
1 kHz are due to the periodic
signal imposed by the product
modulator with 50% duty cycle
first-null bandwidth = 2 kHz = 1/τ = 2/T

• Scaled PSD of rectangular PAM τ / Tb = 1, Tb = 1 msec
No impulse spectral terms in

the PSD for the rectangular
PAM signal because the
signal is only random with
100% duty cycle (τ = Tb)
first-null bandwidth = 1 kHz = 1/τ = 1/Tb

Chapter 2
Demodulation
• Performance of Rectangular
PAM in a Simple Receiver in
AWGN
• Pages 66-69
• The performance of rectangular PAM (τ = Tb, 100%

duty cycle) in a simple receiver (single point sampling at
Tb / 2) in AWGN is assessed by the bit error rate (BER)
The received binary data is compared bit-by-bit to the
transmitted binary data by the Error Rate Calculation
block
MS Figure 2.7
• Rectangular PAM with AWGN and BER analysis
The Error Rate

Calculation block delays
the transmitted binary
data to correlate it with
the received binary data
The BER output is

displayed numerically
• Rectangular PAM with AWGN and BER analysis
The computation delay

allows the BER analysis
to be delayed before
starting if warranted (no
delay is used here)
• Spectrally white Gaussian noise is added in the channel

with variance σ2 in volts by the AWGN Channel block
The initial seed is

the MATLAB
variable randseed
so that each
simulation is
unique
• The signal-to-noise ratio (SNR) is the metric for the BER

analysis. The signal power in the rectangular PAM signal
is A2 / RL and the noise power is σ2 / RL. The SNR then
is:
SNR = (A2 / RL )/(σ2 / RL) = A2 / σ2

SNR = 10 log10 [A2 / σ2] dB
MS Figure 2.7
• The signal power in the rectangular PAM signal is A2 / RL

W (watts) and the normalized power (RL = 1Ω) is
A2 V2 (volts squared).
• Although it is correct to use the units V2 for normalized

power and then divide by RL to convert to W, the
convention is to use the term normalized power W even
though the units are V2. Here A = ± 5 V and the
normalized power = 25 W.
MS Figure 2.7
• Rectangular PAM performance in a simple receiver with

AWGN (MS p. 24)
Table 2.2 Observed BER as a Function of SNR in an
Unfiltered Rectangular PAM Digital Communication
System, Normalized Signal Power = 25 W.
SNR dB AWGN σ2 V2 BER

∞ 0 0
13.98 1 0
12.21 1.5 0
10.97 2 2 × 10-4
6.98 5 1.24 × 10-2
3.98 10 5.64 × 10-2
3.19 12 7.43 × 10-2
• The BER performance of a rectangular PAM with AWGN

can be improved with the addition of a low-pass filter
(LPF) in the receiver. The LPF passes only the bandwidth
required for the modulated signal.
MS Figure 2.9
• The low-pass filter in the rectangular PAM receiver is

an Analog Filter Design block
• The available simulated analog low-pass filter types are

Bessel, Butterworth, Chebyshev types I and II, and
Elliptic. The Chebyshev LPF with fcutoff = 1200 Hz, in-band
ripple = 0.1 dB displays the maximum roll-off attenuation
with a trade-off of the in-band ripple.
• Rectangular PAM performance in a LPF receiver with

AWGN (SVU p. 73)

LPF (9-pole Chebyshev, 0.1 dB ripple, fcutoff = 1.2 kHz)
Binary Rectangular PAM Digital Communication
System, Normalized Signal Power = 25 W.

∞ 0 0
10.00 2.5 0
0.96 20 0
–3.01 50 1 × 10-4
–6.02 100 6.3 × 10-3
–9.03 200 3.63 × 10-2
• Comparison of BER performance in binary rectangular

PAM in an unfiltered and LPF receiver:
Unfiltered Receiver
6.98 5 1.24 × 10-2
3.98 10 5.64 × 10-2
LPF Receiver
–6.02 100 6.3 × 10-3
–9.03 200 3.63 × 10-2
• Rectangular PAM ± 5 V, 1 kb/sec with AWGN, σ2 = 10 V2
Simple receive data sampling occurs at the mid-point Tb /2
Tb + 5V
-5 V
sampling points at (2n+1)Tb /2

Chapter 3
Techniques
• Pulse Shaping to Improve
Spectral Efficiency: Sinc Pulses
• Pages 89-101
• A practical sinc-shaped pulse has a finite duration T =

± 4 Tb. Zero-crossings occur at ± n / rb = ± n Tb
MS Figures 2.15 and 2.16
binary 1 rb = 1 kb/sec
binary 0
2/rb 3/rb
1/rb 4/rb
• Bandwidth of a sinc-shaped PAM pulse is rb / 2 Hz which

is 50% of the first-null bandwidth of a rectangular PAM
pulse (rb Hz).
+5 V
binary 0
2/rb
0
1/rb 8/rb
3/rb
binary 1
4/rb
–5 V rb = 1 kb/sec
• Since the practical sinc-shaped pulses have zero-

crossings at multiples of 1 / rb = Tb there is no interference
between adjacent pulses if sampled at Tb.
startup sampling points at nTb

1/rb binary 1
binary 0
• Sinc PAM Power Spectral Density (S&M p. 90,

MS p. 29-30)
The power spectral density (PSD) for a sinc PAM signal is

derived from the energy spectral density (ESD) for a
single sinc pulse. The ESD ψ(f) for a sinc pulse with a
peak amplitude A and data rate rb is:
ψ(f) = (A / rb)2 = A2 Tb2 – rb/2 ≤ f ≤ rb/2
As for rectangular pulses earlier, the normalized PSD G(f)

of a series of sinc pulses is:
G(f) = n A2 Tb2 / n Tb = A2 Tb – rb/2 ≤ f ≤ rb/2
where 1 / rb = Tb
Chapter 2
Demodulation
• Sinc Pulse Amplitude
Modulation
• Pages 27-33
• MATLAB and Simulink generation of a practical (finite

duration) sinc PAM signal uses the impulse response of
a filter (MS p. 27-29).
MS Figure 2.14
sinc impulse filter
data source
impulse train
• The Random Integer Generator block data has an

amplitude of 0, 1 V and a period of 1 msec (1/rb). The
output is offset and scaled to provide a ± 1 V data
source.
• The Pulse Generator

block has an amplitude of
1 V, a period of 1 msec or
a frequency of 1 kHz
(rb) and a pulse width of 2%
of the period (20 µsec)
which is the Simulink
simulation time.
• Simulink Comm Filters in the Communications

Blockset provides the Raised Cosine Transmit Filter
which can generate
a sinc when the
rolloff factor α = 0.
• Raised Cosine Transmit

Filter block generates a
sinc with a rolloff factor
α = 0. The group delay
of 4 symbols specifies
the ± 4Tb duration.

Filter block upsampling
factor 50 specifies the
duration of a symbol
Tb = 50 x 20 µsec =
1000 µsec = 1 msec,
where 20 µsec is the
Simulink simulation
time. The filter gain of 5
sets the peak amplitude
to ± 5 V.
• MATLAB and Simulink sinc PAM transmitter can be used

to verify the PSD.
MS Figure 2.19
• PSD of sinc PAM, rb = 1 kb/sec
first null bandwidth rb /2 = 500 Hz

sinc2 shaped PSD
non-ideal sinc(t)
• PSD of rectangular PAM, rb = 1 kb/sec
first null bandwidth rb = 1000 Hz

• PSD of sinc PAM, rb = 1 kb/sec
relatively flat PSD
sinc2 shaped PSD

• The performance of practical sinc PAM in a simple

receiver (single point sampling at Tb ) in AWGN is
assessed by the bit error rate (BER)
• The MATLAB and Simulink simulation of binary sinc
PAM is used to assess performance.
MS Figure 2.20
• The sinc PAM transmitter is included as a Simulink

subsystem.
MS Figure 2.20
• The MATLAB and Simulink simulation of binary sinc

PAM includes a Chebyshev low-pass filter (LPF) with
fcutoff = 600 Hz to improve BER performance. The
bandwidth is rb /2 = 500 Hz, rb = 1 kb/sec here.
MS Figure 2.20
• The signal power of the practical sinc PAM is

problematical because of the complex shape and finite
duration. A Simulink simulation can be used to compute
the root-mean-square (RMS) of the sinc PAM transmit
output.
MS Figure 2.21
• The Simulink RMS

block from the
Signal Processing
Blockset, Statistics
computes the RMS.
• The measured normalized power of the practical sinc

PAM signal is [RMS]2 = 4.9342 = 24.3 W. The
normalized power of the rectangular PAM signal is 25
W exactly. The sinc PAM power is required for the
computation of SNR in the BER analysis.
MS Figure 2.21
• Sinc PAM performance in a simple receiver with AWGN

(MS p. 31-33)

LPF (9-pole Chebyshev, 0.1 dB ripple, fcutoff = 600 Hz)
Binary Sinc PAM Digital Communication System,
Normalized Signal Power ≈ 24.3 W.

∞ 0 0
0.85 20 0
−3.13 50 4 × 10-3
−6.14 100 3.5 × 10-3
−9.15 200 2.14 × 10-2
−13.13 500 9.31 × 10-2
• Comparison of BER performance in binary sinc and

rectangular PAM with an LPF receiver:
Sinc PAM
−6.14 100 3.5 × 10-2
−9.15 200 2.14 × 10-2
Rectangular PAM
–6.02 100 6.3 × 10-3
–9.03 200 3.62 × 10-2
Chapter 3
Techniques
• Pulse Shaping to Improve
Spectral Efficiency: Raised Cosine
Pulses
• Pages 101-111
• The raised cosine PAM pulse has a finite duration by

definition. Zero-crossings occur at ± n / rb = ± n Tb
+5 V
binary 0
2/rb
0
1/rb 8/rb
3/rb binary 1
4/rb
• Bandwidth of a raised cosine pulse is rb / 2 + β Hz where

β is the damping factor and 0 ≤ β ≤ rb / 2.
+5 V
binary 0
2/rb
0
1/rb 8/rb
3/rb binary 1
4/rb
• The rolloff factor α of a raised cosine pulse is related to

the damping factor β: α = 2 β / rb 0≤α≤1
β = α rb / 2 0 ≤ β ≤ rb / 2
+5 V
binary 0
2/rb
0
1/rb 8/rb
3/rb binary 1
4/rb
• Since the raised cosine pulses have zero-crossings at

multiples of 1 / rb = Tb there is no interference between
adjacent pulses if sampled at Tb.
startup sampling points at nTb

1/rb binary 1
binary 0
• Raised Cosine PAM Power Spectral Density (S&M p. 102,

MS p. 36-38)
The normalized power spectral density (PSD) G(f) for a

raised cosine PAM signal is derived from the energy
spectral density (ESD) for a single raised cosine pulse
with a peak amplitude A and data rate rb = 1 / Tb is:
G(f) = A2 / rb | f | ≤ rb / 2 – β
G(f) = 4 A2 Tb [1 + cos { π / 2 β ( | f | – rb/2 + β ) } ] 2
rb / 2 – β ≤ f ≤ rb/2 + β
G(f) = 0 | f | ≤ rb / 2 + β
Chapter 2
Demodulation
• Raised Cosine Pulse Amplitude
Modulation
• Pages 81-87
• MATLAB and Simulink generation of raised cosine PAM

signal uses the impulse response of a filter (MS p. 33-36).
MS Figure 2.22
raised cosine impulse filter
data source
impulse train
• The Random Integer Generator block data has an

amplitude of 0, 1 V and a period of 1 msec (1/rb). The
output is offset and scaled to provide a ± 1 V data
source.
• The Pulse Generator

block has an amplitude of
1 V, a period of 1 msec or
a frequency of 1 kHz
(rb) and a pulse width of 2%
of the period (20 µsec)
which is the Simulink
simulation time.
• Simulink Comm Filters in the Communications

Blockset provides the Raised Cosine Transmit Filter
which generates a
raised cosine filter
when the rolloff factor
0 < α ≤ 1 (α ≠ 0)

Filter block is generated
with a rolloff factor α = 0.5
The group delay of 4
symbols specifies the
± 4Tb duration.
• MATLAB and Simulink raised cosine PAM transmitter

can be used to verify the PSD.
MS Figure 2.22
• PSD of sinc PAM, rb = 1 kb/sec, α = 0
first null bandwidth rb /2 = 500 Hz
• PSD of raised cosine PAM, rb = 1 kb/sec, α = 0.5, β = 250
first null bandwidth rb /2 + β = 750 Hz

α = 2 β / rb
• PSD of raised cosine PAM, rb = 1 kb/sec, α = 0.5, β = 250
first null bandwidth rb /2 + β = 750 Hz
first null bandwidth rb = 1000 Hz

• The performance of raised cosine PAM in a simple

receiver (single point sampling at Tb ) in AWGN is
assessed by the bit error rate (BER)
• The Simulink simulation of binary raised cosine PAM
is used to assess performance.
MS Figure 2.28
• The MATLAB and Simulink simulation of raised cosine

PAM includes a Chebyshev low-pass filter (LPF) fcutoff =
900 Hz, in-band ripple = 0.1 dB to improve BER
performance. The first null bandwidth is 750 Hz.
MS Figure 2.28
• The computed normalized power of the raised cosine

PAM signal is [RMS]2 = 4.672 = 21.8 W. The normalized
power of the sinc and rectangular PAM signals are 24.3
and 25 W. The raised cosine PAM power is required
for the computation of SNR in the BER analysis.
MS Figure 2.21rcos
• Raised cosine PAM performance in a simple receiver with

AWGN (MS p. 38-40)

LPF (9-pole Chebyshev, 0.01 dB ripple, fcutoff = 900 Hz)
Binary Raised Cosine PAM Digital Communication
System, Normalized Signal Power ≈ 21.8 W.

∞ 0 0
3.38 10 0
0.37 20 4 × 10-4
−3.61 50 5.8 × 10-3
−6.62 100 2.48 × 10-2
−9.62 200 6.75 × 10-2
• Comparison of BER performance in binary sinc and

raised cosine PAM with an LPF receiver:
Sinc PAM
−3.13 50 4 × 10-4
−9.15 200 2.14 × 10-2
Raised Cosine PAM
SNR dB AWGN σ V BER
–3.61 50 5.8 × 10-2
–9.62 200 6.75 × 10-2
End of Chapter 3
Techniques
Chapter 4
Receiver Design
Chapter 4
Receiver Design
• Probability of Bit Error
• Pages 124-149
• Probability of Bit Error
The low pass filtered and sampled PAM signal results in

an expression for the probability of bit error Pb (S&M p.
124-127). A is the amplitude at the sampling point and γ
is the attenuation of the channel (0 ≤ γ ≤ 1)
P{ ith bit in error } = P(bi = 0) P{ no[ (i-1)Tb + Tb/2 ] < –γA }

+ P(bi = 1) P{ no[ (i-1)Tb + Tb/2 ] ≥ γA }
• Review of Probability and Stochastic Processes

(S&M p. 127-132)
Probability distribution function FX(a) = P{ X = a }
Probability density function fX(x) = d FX(x) / dx
Mean (or expected value) µX = ∫ x fX(x) dx

∞
Variance σX2 = ∫ (x – µX) fX(x) dx
–∞
E { (X – µX)2 }
• Review of Probability and Stochastic Processes

(S&M p. 127-132)
Joint probability
distribution function FX,Y(a, b) = P{ X = a and Y = b }
Joint probability
density function fX,Y(x, y)) = ∂2 FX,Y(x, y) / ∂x ∂y
Conditional probabilities P { X > a and event Z } =

P { event Z } P{ X > a | event Z }
Chapter 4
Receiver Design
• Examining Thermal Noise
• Pages 132-136
• Johnson–Nyquist noise or thermal noise is the electronic

noise generated by the thermal agitation of the charge
carriers (usually the electrons) inside an electrical
conductor at equilibrium.
equilibrium
This thermal noise was first measured by John B. Johnson

at Bell Labs in 1928. He described his findings to Harry
Nyquist, also at Bell Labs, who was able to explain the
results.
Harry Nyquist 1984-1995

1889-1976
• Thermal (or Gaussian) noise is approximately white,

meaning that the power spectral density is equal
throughout the frequency spectrum. Additionally, the
amplitude of the signal has very nearly a Gaussian
probability density function with mean µn = 0.
S&M Figure 4-3

µn = 0 σn = 1
• Since thermal noise has a Gaussian probability density

function the probability that a noise voltage n(t) at time to
will be less than or equal to a threshold –γA is (S&M Eq.
4.27):
− γA
1  (x - µn )2 
P{ n(t o ) ≤ −γA } = FX (− γA) = ∫
−∞ 2π σ n
exp 
 2σ n 
2  dx
and the probability that a noise voltage n(t) at time to

will be greater than a threshold γA is (S&M Eq. 4.28):
∞
1  (x - µn )2 
P{ n(t o ) > γA } = 1 − FX (γA) = ∫
γA 2π σ n
exp 
 2σ n 
2  dx
• The probabilistic properties of

thermal noise do not change with
time (stationarity). Thermal noise
is an insidious property of
communication systems that limits
the speed of reliable data transmission
and the detection of weak signals.
• A MATLAB and Simulink simulation verifies the spectral

characteristics of thermal noise and the performance of
low-pass filtered fcutoff = 11.25 kHz thermal noise.
MS Figure 1.11
• Thermal noise PSD = No, | f | → ∞ MS Figure 1.12
No (single-sided spectrum)
No No
• Thermal noise LPF PSD = No, | f | < 11.25 kHz
No 11.25 kHz
• Thermal noise PSD = No, | f | → ∞ MS Figure 1.12
uncorrelated
• Thermal noise autocorrelation MS Figure 1.14
uncorrelated
• Thermal noise LPF PSD = No, | f | < 11.25 kHz
uncorrelated
• LPF thermal noise autocorrelation MS Figure 1.14
correlated
• Simulink Histogram block from the Statistics, Signal

Processing Blockset is analogous to the pdf. The Vector
Scope block from the Sinks, Signal Processing Blockset,
displays the histogram. Simulink blocks to calculate the
mean µ and variance σ2 are in Statistics.
NoiseHistogram.mdl
• Thermal noise histogram display
µ = 10
σ2 = 1 Gaussian pdf
• LPF fcutoff = 1 kHz thermal noise histogram display
µ = 10.42 pdf remains

σo2 = 0.64 Gaussian
after LPF
• LPF thermal noise has an average normalized power:
σo2 = No fcutoff (S&M p. 135)
The probability that the ith-bit is received in error is:
P(bi = 0) P{ n(t o ) < − γA } + P(bi = 1) P{ n(t o ) ≥ γA } =

− γA
1  (x - µo )2 
P(bi = 0) ∫
−∞ 2π σ o
exp 
 2σ o 
2  dx +
∞
1  (x - µo )2 
P(bi = 1) ∫
γA 2π σ o
exp 
 2σ o 
2  dx
• If and only if the binary thresholds are symmetrical (–γA

and γA), the P(bi = 0) + P(bi = 1) = 1 and the Gaussian
normal pdfs are symmetrical the probability that the ith-bit
is received in error becomes (S&M p. 136):
P(bi = 0) P{ n(t o ) < − γA } + P(bi = 1) P{ n(t o ) ≥ γA } =

∞
1  (x - µo )2 
∫
γA 2π σ o
exp 
 2σ o 
2  dx
-γA γA
Chapter 4
Receiver Design
• Gaussian Probability
Density Function, Probability of
Bit Error
• Pages 137-149
• Gaussian (normal)
probability
density
function (pdf)
• Gaussian (normal)
probability
distribution
function
• Gaussian (Normal) Probability Distribution
Abraham de Moivre was a French

mathematician famous for de Moivre's
formula, which links complex numbers
and trigonometry, and for his work on
the normal distribution and probability
theory in 1734. He wrote a book on
probability theory entitled The Doctrine
of Chances which was said to be highly
prized by gamblers. 1667-1754
Gauss rigorously justified and extended the work in 1809.

• Gaussian (Normal) Probability Distribution
Johann Carl Friedrich Gauss was a

German mathematician and scientist
who contributed significantly to many
fields, including number theory,
geometry, electrostatics, astronomy
and optics.
1777-1855
• Gaussian pdfs
µ = 0, σ = 1
S&M Figure 4-6a
µ = 1.6, σ = 1
S&M Figure 4-6b
• Gaussian pdfs
µ = 0, σ = 2
S&M Figure 4-6c
µ = 1, σ = 2
S&M Figure 4-6d
• The probability of bit error is the area under the Gaussian

pdf from the threshold a to ∞ which could be tabulated.
However, the probability of bit error is determined by three
independent variables (a, µ and σ) and this would be an
unwieldy table.
Q-function
S&M Figure 4-7

• Rather, construct a single table with µ =0 and σ = 1 for the

probability of bit error as the area under the Gaussian
pdf as a function of the threshold a only which is known as
the complementary error or Q-function.
Q-function
S&M Figure 4-7

• The
Q-function
for µ =0
and σ = 1
as a function
of the
threshold a
is listed in
Table 4-1
(S&M
p. 141) and
Appendix B
(MS p. 185-
186)
• Reading the
Q-function
table
correctly
•What is
Q(1.82)?
• What is
Q(2.63)?
• What is
Q(3.18)?
• Reading the
Q-function
table
correctly
• Q(1.82) =
0.0344
• Q(2.63) =
0.0043
• Q(3.18) =
0.0007
• What if µ ≠ 0? It can
be shown that the
Q-function table
remains valid if
the threshold
a
variable in the table
is changed from a
to a – µ. Note that
the areas under the
Gaussian pdfs are the
same.
S&M Figure 4-8a a–µ

Figure 4-8b
• What if µ ≠ 0 and
σ ≠ 1? It can also
be shown that the
Q-function table
remains valid if
a
the threshold
variable in the table
is first changed from
a to a – µ and …
S&M Figure 4-9a a–µ

Figure 4-9b
• … then the threshold

variable in the
Q-function table
is changed from
a – µ to (a – µ) / σ.
Note that argument a–µ
on the x-axis is
compressed by σ
and the y-axis is
expanded by σ.
S&M Figure 4-9b

(a –µ) / σ
Figure 4-9c
• For simple baseband PAM the probability of bit error Pb is

expressed by the Q-function:
 noise margin of 
 
sampled value
Pb = Q  

average normalized noise power at
 
 the input to the single point sampler 
sampling at  
Tb/2 binary rectangular PAM data with AWGN
Chapter 4
Receiver Design
• Optimal Receiver: The Matched
Filter or Correlation Receiver
• Pages 149-161
• The simple baseband PAM receiver structure is:
But is this the best that there is? What about sampling an
odd number of times (like 3) during each bit time Tb?
Tb
• Although sampling can be increased to a very large, odd

number of samples during Tb, there is an optimal way:
Since LPF in PAM improved performance, assume that

the optimal processing is a linear filter H(f) (S&M p. 150)
• After development (S&M p. 150-153) the optimal

processing is a linear filter Ho(f)
The optimal linear filter Ho(f) has an impulse response

ho(t) and is known as a matched filter since the processing
is matched to input signal s(t):
Ho (f) = k S∗ (f) exp( − j 2π f iTb )
ho (t) = F-1 { Ho (f) } = k F-1 { S∗ (f) exp( − j 2π f iTb ) }
ho (t) = k s(iTb − t) impulse response
matched to s(t)
• The impulse response of the optimum filter ho(t) is a

scaled (by k), time delayed (by iTb) and time reversed
(function of iTb– t):
ho (t) = k s(iTb − t)
• When optimum processing is used the argument inside

the Q-function is maximized (S&M p. 153-154) and the
probability of bit error Pb is:
 | r(iTb ) ∗ h(iTb ) | 
Pb = Q  maximum 
 σp
 
 | r(iTb ) * h(iTb ) | 2 
Pb = Q  maximum 
 σ 2

 p 
 2 Eb 
Pb = Q   S&M Eq. 4.58
 No
 
where Eb is the energy per bit of the received signal.
• The optimum filter Ho(f) is equivalent to the correlation

receiver (S&M p. 155-156).
Optimum Filter ho (t) = k s(iTb − t)
Correlation Receiver
• Since the optimum filter Ho(f) and the correlation receiver

are equivalent, with s1(t) = s(t) for a matched filter and
r(t) = γ s(t) where γ is the communication channel
attenuation, the energy-per-bit Eb is (S&M p. 156,
Eq 4.62):
iTb iTb
Eb = ∫
(i-1)Tb
γ s(t) γ s(t) dt = γ 2 ∫
(i-1)Tb
s2 (t) dt
• The expected or mean value ai(iTb) is the output of the

correlation receiver when r(t) = γ si(t) and n(t) = 0 where γ
is the communication channel attenuation.
iTb
ai (iTb ) = ∫
(i-1)Tb
γ si (t) s1(t) dt S&M Eq. 4.67
• Pb = Q( √(2 Eb / No) ) and the ratio Eb / No can be

expressed in dB: 10 log10 (Eb / No ). The resulting plot of
Pb verses Eb / No in dB is a characteristic of binary
symmetric
PAM with
AWGN. S&M Figure 4-14
• Symmetric binary PAM

implies that the two
transmitted signals for
binary 1 and binary 0
s(t) and the resulting
Threshold = 0
outputs a(iTb) from the
correlation receiver
are equal in
magnitude but
opposite in sign:
sbi=1(t) = – sbi=0(t)
S&M Figure 4-15

• The probability of bit

error for equally-likely
binary symmetric
PAM is the sum of
the error regions
shown.
P(bi=0) = P(bi=1) = 0.5
S&M Figure 4-16

• The binary PAM signals are symmetrical and the

threshold is 0 (equidistant from the means or expected
values ± a(iTb) ). The error regions are equal in area.
S&M Figure 4-16

• The probability of bit

error does not
minimize if the
threshold is
misadjusted (τ ≠ 0).
With a misadjusted
threshold the apriori
probabilities are now
important since the
area of the error
regions are no longer
equal.
S&M Figure 4-17

• The correlation receiver is also known as the integrate-

and-dump which describes the process.
matched filter or
MS Figure 2.29
Chapter 2
Demodulation
• Optimum Binary Baseband
Receiver: The Correlation
Receiver
• Pages 40-41
• The correlation receiver can be simulated in Simulink:
MS Figure 2.29
• The Simulink
Integrate and Dump
block is in the
Comm Filters,
Communications
Blockset
• The parameters of the Simulink Integrate and Dump

block are the integration period and offset in simulation
samples.
fsimulation = 50 kHz
0.02 msec
Tb = 1 msec
Integration period =
Tb/Tsimulation = 1/ 0.02 =
50 samples
Offset = 0 samples
• The complete binary symmetrical rectangular PAM digital

communication system with BER analysis and optimum
receiver.
MS Figure 2.30
BER
• Observed BER as a function of SNR for binary

rectangular PAM in a LPF simple receiver (LPF) and the
optimum correlation receiver (CR) with normalized signal
power = 25 W (MS Table 2.3 p. 27 and Table 2.6 p. 43).
SNR dB AWGN σ2 V2 BER (LPF) BER (CR)
∞ 0 0 0
3.98 10 0 0
0.96 20 0 0
−3.52 50 1 × 10-4 0
−6.02 100 6.3 × 10-3 1 × 10-4
−9.03 200 3.63 × 10-2 6.4 × 10-3
−13.01 500 1.185 × 10-1 6.02 × 10-2
−16.02 1000 1.345 × 10-1
• The energy per bit Eb = 2.5 × 10-2 V2-sec (S&M Eq. 4.62,
p. 156) for rectangular ± 5 V PAM with the channel
attenuation γ = 1:
iTb iTb
Eb = ∫
(i-1)Tb
γ s(t) γ s(t) dt = γ 2 ∫
(i-1)Tb
s2 (t) dt
The observed bit error rate (BER) can be compared to

the theoretical probability of bit error Pb (S&M Eq. 4.58, p.
154) to validate the basic simulation.
 2 Eb 
Pb = Q  
 No
 
• The ratio of the energy per bit Eb to the noise power

spectral density No (Eb / No) rather than the SNR is the
conventional metric used for BER performance.
iTb iTb
Eb = ∫
(i-1)Tb
γ s(t) γ s(t) dt = γ 2 ∫
(i-1)Tb
s2 (t) dt
S&M Eq. 4.62
 2 Eb 
Pb = Q   S&M Eq. 4.58
 No
 
• The BER and Pb comparison (MS Table 2.7, p. 44):
Table 2.7 Observed BER and Theoretical Pb as a

Function of Eb / No in a Binary Symmetrical Rectangular
PAM Digital Communication System with Optimum
Receiver
Eb / No dB BER Pb
∞ 0 0
10 0 4.05 × 10-6
8 0 2.06 × 10-4
6 2.5 × 10-3 2.43 × 10-3
4 1.35 × 10-2 1.25 × 10-2
2 3.95 × 10-2 3.75 × 10-2
0 8.09 × 10-2 7.93 × 10-2
Chapter 4
Receiver Design
• Correlation Receiver for Asymmetric
PAM, Optimum Thresholds,
Synchronization
• Pages 162-173
• Asymmetric PAM
signals do not have
equal means or
expected values of
the output of the
correlation receiver:
| a2(iTb) | ≠ | a1(iTb) |
S&M Figure 4-19

• The optimum S&M Figure 4-19

threshold τopt is
again equidistant
between the means
or expected values of
the output of the
correlation receiver:
a2 (iTb ) + a1(iTb )
τ opt =
2
S&M Eq. 4.71
Here | a2(iTb) | + τopt = a1(iTb) – τopt and the threshold is

equidistant from the means or expected values.
• If the apriori
probabilities are not
equal then the
optimum threshold
τopt is not equidistant
from the means or
expected value of
output of the
correlation receiver.
An asymmetric binary
PAM signal is shown:
a2(iTb) ≠ a1(iTb)
S&M Figure 4-23

• The optimum
threshold τopt where
the apriori
probabilities are
(1 – M) and M
(which sums to 1) is:
2 σ o2 ln ( M / (1− M ) ) + a2 (iTb )2 − a1(iTb )2

τ opt =
2 [a2 (iTb ) − a1(iTb )] S&M Eq. 4.85
if M = 0.5 then: τ opt =
2 S&M Eq. 4.71
• Asymmetric PAM with S&M Figure 4-19

an optimum threshold
τopt has a probability
of bit error Pb:
 a1(iTb ) − a2 (iTb ) 
Pb = Q  
 2 σo 
 a (iT ) − a (iT ) 2 
 [ 1 b 2 b ]  2No
S&M Eq. 4.78
Pb = Q σ =
 4 σ o2  o
2 and Eq. 4.79
 
 Ed  iTb
∫ {γ ( s (t) − s (t)) }
2
Pb = Q  where Ed = dt
 2 N  1 2
 o  (i-1)Tb
• The optimum correlation receiver for asymmetric binary

PAM uses the difference signal s1(t) – s2(t) as the
reference:
S&M Figure 4-20

• The optimum correlation receiver can be reconfigured as

an alternate but universal structure which can be used for
both asymmetric or symmetric binary PAM signals:
S&M Figure 4-21

• If the apriori
probabilities are not
equal then the
optimum threshold
τopt is not equidistant
from the means or
expected value of
output of the
correlation receiver.
An asymmetric binary
PAM signal is shown:
a2(iTb) ≠ a1(iTb)
S&M Figure 4-23

• The optimum
threshold τopt where
the apriori
probabilities are
(1 – M) and M
(which sums to 1) is:
2 σ o2 ln ( M / (1− M ) ) + a2 (iTb )2 − a1(iTb )2

τ opt =
2 [a2 (iTb ) − a1(iTb )] S&M Eq. 4.85
if M = 0.5 then: τ opt =
2 S&M Eq. 4.71
• The probability
of bit error Pb then
is:
S&M Eq. 4.86
 a1(iTb ) − τ opt   τ opt − a2 (iTb ) 

Pb = M Q   + (1− M ) Q  
 σo   σo 
 2   
( a1(iTb ) − τ opt )  ( τ opt − a2 (iTb ) )
2

Pb = M Q  + (1− M ) Q  
2 No  2 No 
   
   
• If the apriori S&M Figure 4-19

probabilities are
equal (M = 0.5):
τ opt =
2
and Pb becomes:
   
( a1(iTb ) − τopt ) ( opt 2 b ) 
2 2
τ − a (iT )
Pb = 0.5 Q   + 0.5 Q 
  
 2 No   2 No 
   
 a2 (iTb ) ) 
( a1(iTb ) -
2
 Ed 
Pb = Q  = Q  S&M p. 168
 2 No   2N
   o 

Function of Ed / No in an Asymmetrical Binary
Rectangular PAM Digital Communication System with
Optimum Receiver.
Ed / No dB BER Pb
∞ 0 0
12 2.5 × 10-3 2.53 × 10-3
10 1.28 × 10-2 1.25 × 10-2
8 3.59 × 10-2 3.75 × 10-2
6 8.05 × 10-2 7.93 × 10-2
4 1.334 × 10-1 1.318 × 10-1
2 1.856 × 10-1 1.872 × 10-1
0 2.362 × 10-1 2.394 × 10-1
• Since for binary rectangular PAM Ed = 4 Eb (S&M p. 168)

the BER performance for asymmetric PAM is comparable
to symmetric PAM if a Ed / No is reduced by 6 dB
(10 log (4) = 6) :
Ed / No dB BER Pb
Asymmetric PAM 10 1.28 × 10-2 1.25 × 10-2

(MS Table 2.8, 8 3.59 × 10-2 3.75 × 10-2
p. 49) 6 8.05 × 10-2 7.93 × 10-2
Eb / No dB BER Pb
Symmetric PAM 4 1.35 × 10-2 1.25 × 10-2

(MS Table 2.7, 2 3.95 × 10-2 3.75 × 10-2
p. 44) 0 8.09 × 10-2 7.93 × 10-2
Chapter 2
Demodulation
• The Correlation Receiver for
Baseband Asymmetrical
Signals
• Pages 44-47
• The optimum threshold τopt requires the additive Gaussian

noise variance σ2 as processed by the correlation
receiver or σo2:
2 σ o2 ln ( M / (1− M ) ) + a2 (iTb )2 − a1(iTb )2
τ opt =
2 [a2 (iTb ) − a1(iTb )] σo
MS Figure 2.31
• The optimum threshold τopt also requires the apriori

probabilities P1 = M and P0 = M – 1:
2 σ o2 ln ( M / (1 − M ) ) + a2 (iTb )2 − a1(iTb )2
τ opt =
2 [a2 (iTb ) − a1(iTb )]
Here P1 = 0.4998 = M and P0 = 1 – M = 0.5004
Mean
MS Figure 2.32
• The Mean and

Variance blocks
are in the Statistics,
Signal Processing
Blockset
• The parameter of the

Mean block can be set
for running mean or an
average over the
simulation period. The
result can be shown by
a Display block from the
Simulink Blockset.
MS Figure 2.32
• The parameter of the

Variance block can be set
for running variance or an
average over the
simulation period. The
result can be shown by
a Display block from the
Simulink Blockset.
• The Simulink simulation for binary asymmetrical PAM

with BER analysis has a threshold adjustment based on
the estimate of τopt from measurements of σo and M.
2 σ o2 ln ( M / (1− M ) ) + a2 (iTb )2 − a1(iTb )2
τ opt =
2 [a2 (iTb ) − a1(iTb )]
Reference
Threshold
MS Figure 2.33
• The Simulink simulation for the alternative and universal

structure for the correlation receiver:
MS Figure 2.34
S&M Figure 4-21

• Synchronization (not considered here) provides timing

recovery or the exact beginning and end of a bit time Tb:
S&M Figure 4-21
start of Tb?
• Synchronization of both symbol time and carrier frequency

and phase is one of the advanced topics in EE4542
Telecommunications Engineering. The effect is called jitter.
Chapter 4
Receiver Design
• Multi-level PAM (M-ary PAM)
• Pages 200-206
• Multi-level (M-ary) PAM is another means to minimize the

bandwidth required for a data transmission rate rb b/sec.
Rather than transmitting a binary signal in a bit time Tb,
send a multi-level (usually a power-of-2) signal during the
same period called the symbol time TS.
A multi-state comparator determines the received symbol

which is then decoded to the received bits.
S&M Figure 4-48
M=4
• The three
optimum
thresholds if
the apriori
probabilities
are equally
likely
(Pi = 0.25) S&M Figure 4.49
are: a1(iTS ) + a2 (iTS )
τ opt1 =
2
a2 (iTS ) + a3 (iTS )
τ opt2 = S&M Eq. 4.136
2
a3 (iTS ) + a4 (iTS )
τ opt3 =
2
• The
probability
of symbol
error Ps,
where M = 2n M=4
is the
number of
levels,
can be 2 (M − 1)  Ed,symbol 
shown to be: Ps = Q  S&M Eq. 4.139
M  2 No 
where: 
i Ts
∫ {γ s (t) - s (t) }

2
Ed,symbol = j k dt S&M Eq. 4.140
(i-1) Ts
• There are
2 (M – 1)
error
regions
due to only M=4
adjacent
regions
being
misinterpreted with M = 2n equally probable symbols. The
probability of occurrence Pj for a misinterpreted symbol is
also equally likely and is:
1
Pj =
2 (M − 1) SVU Eq. 2.41
• The amplitudes of the M-ary (M = 4) rectangular PAM

signal is ± 3A and ± A. The energy difference for a
symbol Ed,symbol = 4γ2 A2 TS (S&M Eq. 4.140):
i Ts
∫ {γ s (t) - s (t) }

2
Ed,symbol = j k dt
(i-1) Ts
The average received energy per symbol Eavg,symbol =

5γ2 A2 TS (S&M Eq. 4.141)
11
S&M Figure 4-50 modified
10
01
M=4
00
• Then Ed,symbol = 0.4 Eavg,symbol and substitute (M = 4):
3  Ed,symbol 
Ps = Q  S&M Eq. 4.139b
2  2 No 
3  0.4 Eavg,symbol 
Ps = Q   S&M Eq. 4.142a
2  No 

11
10
01
M=4
00
• The average energy per bit Eb = 0.5 Eavg,symbol since M =4

and there are two bits per symbol:
3  0.4 Eavg,symbol 
S&M Eq. 4.142a
Ps = Q  
2  No 

3  0.8 Eb  S&M Eq. 4.142b

Ps = Q  
2  No 
11
10
01
M=4
00
• The symbols transmitted can have 1 bit in error 4 / 6 of

the time and 2 bits in error 2 / 6 of the time:
Transmitted Di-Bit Received Di-Bit Bits In Error

00 01 1
01 00 1
01 10 2
10 01 2
10 11 1
11 10 1
4 2
Pb,4-level = P(1 of 2 bits in error) + P(2 of 2 bits in error)
6 6
4 1 2 2
Pb,4-level = Ps + Ps = Ps S&M Eq. 4.143
62 6 3
• The probability of bit error Pb,4-symbol = 2 Ps / 3 but can that

be improved? Change the assignment of symbols to di-
bits as a Gray code and there is only 1 bit in error for
each of the six error regions and Pb,4-symbol = Ps / 2
Transmitted Di-Bit Received Di-Bit Bits In Error
01 00 1
00 01 1
00 10 1
10 00 1
10 11 1
11 10 1
6
Pb,4-level = P(1 of 2 bits in error)
6
6 1 1
Pb,4-level = Ps = Ps
62 2
• The simple change to a Gray coded symbol improves the

probability of bit error Pb:
23  0.8 Eb   0.8 Eb  Straight binary

Pb = Q  = Q  
3 2  No   No  coded
13  0.8 Eb  3  0.8 Eb 
Pb = Q  = Q   Gray coded
2 2  No  4  No 
11
S&M Figure 4-50
10
00
01
• Frank Gray was a researcher at Bell Labs who made

numerous innovations in television and is remembered for
the Gray code. The Gray code
appeared in his 1953 patent and
is a binary system often used in
electronics. Gray also conducted
pioneering research on the
development of television. He
proposed an early form of the
flying spot scanning system for
TV cameras in 1927, and helped
develop a two-way mechanically
scanned TV system in 1930. Frank Gray
1894-1964
Chapter 2
Demodulation
• Multilevel (M-ary) Pulse
Amplitude Modulation
• Pages 49-55
• The Simulink simulation of a 4-level straight binary (not

Gray coded) rectangular PAM system with BER analysis:
MS Figure 2.35
BER
bit to symbol converter symbol to bit converter

• Although Simulink
provides a Bit to
Integer block and
an Integer to Bit
block from the
Utility Blocks,
Communication
Blockset, these
blocks use a vector
of bits rather than
from and to a serial
bitstream.
• A bit to symbol converter Simulink subsystem can be

configured from Downsampler blocks and a Delay block
from the Signal Operations, Signal Processing Blockset
and the Gain and Sum blocks from the Simulink, Math
Operations to produce the 4-level PAM signal (0, 1, 2 or
3) from the input di-bit (00, 01, 10, or 11) with the most
significant bit (MSB) first.
MS Figure 2.36
• The Simulink bit to symbol converter can be easily

reconfigured to produce an 8-level PAM signal (0, 1, 2,
…7) from the input tri-bit (000, 001, … 111) with the
most significant bit (MSB) first.
MS Figure 2.37
• A symbol to bit converter Simulink subsystem can be

configured from Constant blocks, Pulse Generator blocks
and a Multiport Switch block from the Signal Routing,
Simulink Blockset to produce the di-bit (MSB first) from the
4-level PAM signal.
MS Figure 2.40
• A symbol to bit converter Simulink subsystem can be

easily reconfigured to produce the tri-bit (MSB first) from
an 8-level PAM signal.
MS Figure 2.41
• Binary data to 4-level PAM converter
MS Figure 2.36
• Transmit binary data, rb = 1 kb/sec
01
MSB LSB
• Transmit 4-level PAM signal, rS = 500 symbols/sec
1
• M-ary PAM transmits n bits per symbol (M = 2n) but has the
same rectangular pulse shape as binary PAM. The
normalized power spectral density for M-ary PAM PSDM has
the same sinc shape as that for binary PAM PSDB but uses
the symbol time TS rather than the bit time Tb:
2
PSDM (f) = A avg TS sinc ( π TS f )
PSDB (f) = A 2 Tb sinc ( π Tb f )
• Observed BER and theoretical Pb as a function of Eb / No

in a 4-level rectangular straight-binary coded PAM digital
communication system with optimum receiver (MS Table
2.11 p. 58).
Eb / No dB BER Pb
∞ 0 0
10 2.7 × 10-3 2.4 × 10-3
8 1.28 × 10-2 1.25 × 10-2
6 3.42 × 10-2 3.75 × 10-2
4 7.52 × 10-2 7.78 × 10-2
2 1.291 × 10-1 1.320 × 10-1
0 1.842 × 10-1 1.867 × 10-1
• The M-ary PAM PSD uses the average amplitude Aavg,

where Pj is the apriori probability of occurrence of the
amplitude Aj.
M
2
A avg = ∑ A 2j Pj
j=1
• The M-ary (M = 4) PAM PSD can be determined by a

Simulink model
Fig239.mdl
MS Figure 2.39
• The Simulink simulation of a 4-level straight binary (not

Gray coded) rectangular PAM system measures the BER
performance.
MS Figure 2.35
BER
• The rectangular M-ary pulse width is the entire symbol
time TS and is optimum in the bandwidth sense. The
bandwidth is similar to that for binary PAM but with Tb =
TS (cf. S&M Table 3-1 p. 86).
Table 2-9 Bandwidth of a Binary Rectangular M-ary PAM

Signal as a Percentage of the Total Power (MS p. 53)

1/TS 90%
1.5/TS 93%
2/TS 95%
3/TS 96.5%
4/TS 97.5%
5/TS 98%
• PSDM M = 4, rb = 1 kb/sec, first null bandwidth = 500 Hz
500 Hz
• PSDB rb = 1 kb/sec, first null bandwidth = 1 kHz
1 kHz
Chapter 2
Demodulation
• Performance of M-ary Pulse
Amplitude Modulation
• Gray Encoded Data
• Pages 57-60
• The Simulink simulation of a 4-level straight binary Gray

encoded rectangular PAM system with BER analysis:
MS Figure 2.42
BER
bit to Gray encoded Gray encoded symbol
symbol converter to bit converter
• The M = 4 level symbol (0, 1, 2 and 3) is Gray encoded by

a Lookup Table Block from the Simulink Blockset.
MS Figure 2.43 M = 4 symbol
bit to Gray encoded

symbol converter
• The M = 4 level symbol

(0, 1, 2 and 3) is Gray
encoded by a Lookup
Table Block from the
Simulink Blockset.
• The M = 4 level symbol (0, 1, 2, and 3) is Gray encoded by

a Lookup Table Block by mapping [0, 1, 2, 3] to [0, 1, 3, 2]
Gray coding 00 → 00
01 → 01
10 → 11
11 → 10
• The M = 4 level Gray encoded symbol is decoded as di-bits

by a Lookup Table Block from the Simulink Blockset.
M = 4 symbol MS Figure 2.44
Gray encoded symbol

to bit converter
• The M = 4 level symbol (0, 1, 2, and 3) is Gray decoded by

01 → 01
10 → 11
11 → 10
• The PSD of a Gray coded rectangular PAM signal is the

same as a straight-binary rectangular PAM signal. The
coding does not
change the PSD.
The PSD is only
affected by the pulse
shape and data rate.
• PSDM M = 4, rb = 1 kb/sec, first null bandwidth = 500 Hz
500 Hz
• The Simulink simulation of a 4-level straight binary Gray

encoded rectangular PAM system measures the BER
performance.
MS Figure 2.42
BER
• Observed BER and theoretical Pb as a function of Eb / No

in a 4-level rectangular Gray coded PAM digital
communication system with optimum receiver (MS Table
2.12 p. 60).
Eb / No dB BER Pb
∞ 0 0
10 1.7 × 10-3 1.8 × 10-3
8 8.6 × 10-3 9.3 × 10-3
6 2.92 × 10-2 2.81 × 10-2
4 5.93 × 10-2 5.84 × 10-2
2 9.56 × 10-2 9.90 × 10-2
0 1.441 × 10-1 1.404 × 10-1
• Observed BER as a function of Eb / No for binary

rectangular PAM for straight-binary coded (SBC) and
Gray coded (GC) data (MS Table 2.11 p. 58 and MS
Table 2.12 p. 60).
Eb / No dB BER (SBC) BER (GC)
∞ 0 0
10 2.7 × 10-3 1.7 × 10-3
8 1.28 × 10-2 8.6 × 10-3
6 3.42 × 10-2 2.92 × 10-2
4 7.52 × 10-2 5.93 × 10-2
2 1.291 × 10-1 9.56 × 10-2
0 1.842 × 10-1 1.441 × 10-1
Although a slight improvement, Gray coding improves

BER performance without an increase in Eb.
End of Chapter 4
Receiver Design
Chapter 5
Digital Bandpass Modulation
and Demodulation
Techniques
Chapter 5
and Demodulation
Techniques
• Binary Amplitude Shift Keying
• Pages 212-219
• The analytical signal for binary amplitude shift keying

(BASK) is:
sBASK(t) = sbaseband(t) sin (2π fC t) (S&M Eq. 5.1)
The signal sbaseband(t) can be any two shapes over a bit time
Tb but it is usually a rectangular signal of amplitude 0 for a
binary 0 and amplitude A for binary 1. Then BASK is also
known as on-off keying (OOK). MS Figure 3.5
Tb
0 1 1 0 1
• The binary amplitude shift keying (BASK) signal can be

simulated in Simulink.
sBASK(t) = sbaseband(t) sin 2π fC t (S&M Eq. 5.1)
Sinusoidal carrier fC = 20 kHz, Ac = 5 V
Multiplier
BASK signal
baseband binary PAM signal 0,1 V, rb = 1 kb/sec

• A BASK signal is a baseband binary PAM signal multiplied

by a carrier (S&M Figure 5-3).
Unmodulated sinusoidal carrier
Baseband binary PAM signal
BASK signal
• The unipolar binary PAM signal can be decomposed into a

polar PAM signal and DC level (S&M Figure 5-4).
Unipolar binary PAM signal 0→1V
Polar binary PAM signal ± 0.5 V
DC level 0.5 V
• The spectrum of the BASK signal is (S&M Eq. 5.2):
SBASK(f) = F( sASK(t) ) = F( sbaseband(t) sin (2π fC t) )

SBASK(f) = 1/2 j (Sbaseband(f – fC) + Sbaseband(f + fC) )
The analytical signal for the baseband binary PAM signal is:
sbaseband(t) = sPAM(t) + A/2 (S&M Eq. 5.3)

Sbaseband(f) = SPAM(f) + A/2 δ(f) (S&M Eq. 5.4)
Therefore by substitution (S&M Eq. 5.5):
SBASK(f) = 1/ 2j ( SPAM(f – fC) + A/2 δ(f – fC)

– SPAM(f + fC) – A/2 δ(f + fC) )
• The bi-sided power spectral density PSD of the BASK

signal is (S&M Eq. 5.7):
GBASK(f) = 1/4 GPAM(f – fC) + 1/4 GPAM(f + fC)

+ A2/16 δ(f – fC) + A2/16 δ(f + fC)
For a rectangular polar PAM signal (± A):
GPAM(f) = (A/2)2 / rb sinc2 (π f / rb) (S&M Eq. 5.8)
MS Figure 3.7
• The single-sided power spectral density PSD of the BASK

signal is:
GPAM(f) = (A/2)2 / rb sinc2 (π f / rb)

GBASK(f) = 1/2 GPAM(f + fC) + A2/8 δ(f + fC)
Carrier 20 kHz MS Figure 3.7
1 kHz rb = 1 kHz
sinc2
• The bandwidth of a BASK signal as a percentage of total

power is double that for the same bit rate rb = 1/Tb binary
rectangular PAM (MS Table 2.1 p. 22)
(MS Table 3.1 p. 91).

2/Tb 90%
3/Tb 93%
4/Tb 95%
6/Tb 96.5%
8/Tb 97.5%
10/Tb 98%
Chapter 5
and Demodulation
Techniques
• Binary Phase Shift Keying
• Pages 219-225
• The analytical signal for binary phase shift keying

(BPSK) is:
sBPSK(t) = sbaseband sin (2π fC t + θ) (S&M Eq. 5.11)

sbaseband(t) = + A bi = 1 sbaseband(t) = – A bi = 0
0° MS Figure 3.13
+180° Tb 0° 0° +180°
0 0 1 1 0
• The BPSK signal initial phase θ = 0°, +A is a phase shift =

0° and –A is a phase shift = +180°
sBPSK(t) = sbaseband sin (2π fC t) (S&M Eq. 5.11)

sbaseband(t) = + A bi = 1 sbaseband(t) = – A bi = 0
0° MS Figure 3.13
+180° Tb 0° 0° +180°
0 0 1 1 0
• The binary phase shift keying (BPSK) signal can be

PM modulator
BPSK
signal
Fig312.mdl

• The Phase Modulator

block is in the
Modulation,
Communication
Blockset but as an
analog passband
modulator not a digital
baseband modulator.
• The Phase Modulator block has the parameters of a

carrier frequency fC in Hz, initial phase in radians and the
phase deviation constant in radians per volt (rad / V).
fC = 20 kHz
initial phase φo = π
phase deviation kp = π / V
• The Random Integer Generator outputs 0,1 V and with a

initial phase = π and a phase deviation constant = π/V,
the phase output φ of the BPSK signal is:
bi = 0 φ = π + 0(π/V) = π
bi = 1 φ = π + 1(π/V) = 2π = 0
• The spectrum of the BPSK signal is (S&M Eq. 5.13):
SBPSK(f) = F( sPSK(t) ) = F(sbaseband(t) sin 2π fC t)

SBPSK(f) = 1/2 j (Sbaseband(f – fc) + Sbaseband(f + fC) )
The analytical signal for the baseband binary PAM signal is:
sbaseband(t) = sPAM(t) (S&M Eq. 5.12)

Sbaseband(f) = SPAM(f)
Note that there is no DC level in sPAM(t) and therefore by

substitution:
SBPSK(f) = 1/ 2j ( SPAM(f – fC) – SPAM(f + fC) )

• The bi-sided power spectral density PSD of the BPSK

signal is (S&M Eq. 5.13)
GBPSK(f) = 1/4 GPAM(f – fC) + 1/4 GPAM(f + fC)
For a rectangular polar PAM signal (± A):
GPAM(f) = A2 / rb sinc2 (π f / rb) (S&M Eq. 5.8 modified)
MS Figure 3.14
• The single-sided power spectral density PSD of the BPSK

signal is:
GBPSK(f) = 1/2 GPAM(f + fC)
GPAM(f) = A2 / rb sinc2 (π f / rb)
No carrier MS Figure 3.14
rb = 1 kHz
sinc2
• The bandwidth of a BPSK signal as a percentage of total

power is double that for the same bit rate rb = 1/Tb binary
rectangular PAM (MS Table 2.1 p. 22) and the same as
BASK (MS Table 3.1 p. 91)
(MS Table 3.5 p. 100)

2/Tb 90%
3/Tb 93%
4/Tb 95%
6/Tb 96.5%
8/Tb 97.5%
10/Tb 98%
Chapter 5
and Demodulation
Techniques
• Binary Frequency Shift Keying
• Pages 219-225
• The analytical signal for binary frequency shift keying

(BFSK) is:
sBFSK(t) = A sin (2π (fC + ∆f) t + θ) if bi = 1

sBFSK(t) = A sin (2π (fC – ∆f) t + θ) if bi = 0
MS Figure 3.9
fc – ∆f
fc – ∆f fc + ∆f fc + ∆f Tb fc – ∆f
0 1 1 0 0
• The BFSK signal initial phase θ = 0°
sBFSK(t) = A sin (2π (fC + ∆f) t) if bi = 1

sBFSK(t) = A sin (2π (fC – ∆f) t) if bi = 0
MS Figure 3.9
fc – ∆f
fc – ∆f fc + ∆f fc + ∆f Tb fc – ∆f
0 1 1 0 0
• The binary frequency shift keying (BFSK) signal can be

BFSK
signal
FM Modulator
Fig38.mdl

• The Frequency Modulator

block is in the
Modulation,
Communication
Blockset but as an
analog passband
modulator not a digital
baseband modulator.
• The Frequency Modulator block has the parameters of a

carrier frequency fC in Hz, initial phase in radians and the
frequency deviation constant in Hertz per volt (Hz/V).
fC = 20 kHz
initial phase = 0
frequency deviation =
2000
• The Random Integer Generator outputs 0,1 V but is offset

to ±1 and with a initial phase = 0 and a frequency
deviation constant = 2000 Hz/V, the frequency shift ∆f of
the BFSK signal is:
bi = 0 di = –1 ∆f = 0 – 1(2000 Hz/V) = –2000 Hz

bi = 1 di = +1 ∆f = 0 + 1(2000 Hz/V) = +2000 Hz
• The BFSK signal can be decomposed as (S&M Eq. 5.14):
sBFSK(t) = sbaseband1(t) sin (2π (fC + ∆f) t + θ) +

sbaseband2(t) sin (2π (fC – ∆f) t + θ)
1 0 0 1 1
fc + ∆f fc – ∆f fc – ∆f fc + ∆f fc + ∆f
• The BFSK signal is the sum of two BASK signals:
sBFSK(t) = sbaseband1(t) sin (2π (fC + ∆f) t + θ) +

sbaseband2(t) sin (2π (fC – ∆f) t + θ)
From the linearity property, the resulting single-sided PSD

of the BFSK signal GBFSK(f) is the sum of two GBASK(f)
PSDs with f = fC ± ∆f:
GBFSK(f) = (A/2)2 / 2 rb sinc2 (π f / rb) + A2/8 δ(f)

fc – ∆f fc + ∆f
• The single-sided power spectral density PSD of the BFSK

signal is:
GBFSK(f) = (A/2)2 / 2rb sinc2 (π (fC + ∆f) / rb) + A2/8 δ(fC + ∆f)
+ (A/2)2 / 2rb sinc2 (π (fC – ∆f)/ rb) + A2/8 δ(fC – ∆f)
carriers MS Figure 3.10
rb = 1 kHz
∆f = 2 kHz
sinc2
• Minimum frequency shift keying (MFSK) for BFSK occurs

when ∆f = 1/2Tb = rb/2 Hz.
GBFSK(f) = (A/2)2 / 2rb sinc2 (π (fC + ∆f) / rb) + A2/8 δ(fC+ ∆f)
+ (A/2)2 / 2rb sinc2 (π (fC – ∆f)/ rb) + A2/8 δ(fC – ∆f)
∆f = 500 Hz rb = 1 kHz
sinc2
• This BFSK carrier frequency separation 2∆f = 1/Tb = rb Hz is

the minimum possible because each carrier spectral impulse
is at the null of the PSD of the other decomposed BASK
signal and thus is called minimum frequency shift keying
(MFSK).
2∆f = 1000 Hz rb = 1 kHz

sinc2
• The bandwidth of a BFSK signal as a percentage of total

power is greater than that of either BASK or BPSK by 2∆f
Hz for the same bit rate rb = 1/Tb (MS Table 3.3 p. 95).
For MFSK 2∆f = rb = 1/Tb Hz.

2∆f + 2/Tb 90%

2∆f + 3/Tb 93%
2∆f + 4/Tb 95%
2∆f + 6/Tb 96.5%
2∆f + 8/Tb 97.5%
2∆f + 10/Tb 98%
Chapter 5
and Demodulation
Techniques
• Coherent Demodulation of
Bandpass Signals
• Pages 225-236
• The development of the optimum receiver for bandpass

signals utilizes the same concepts as that for the optimum
baseband receiver:
Optimum Filter ho (t) = k s(iTb − t)
Correlation Receiver
• The optimum filter Ho(f) and the correlation receiver

are equivalent here also, with s1(t) = s(t) for symmetrical
signals and r(t) = γ s(t) + n(t) where γ is the communication
channel attenuation and n(t) is AWGN. The energy per bit
Eb and the probability of bit error Pb is (S&M p. 226):
iTb iTb
Eb = ∫
(i-1)Tb
γ s(t) γ s(t) dt = γ 2 ∫
(i-1)Tb
s2 (t) dt
 2 Eb 
Pb = Q  
 No
 
• The matched filter or correlation receiver is a coherent

demodulation process for bandpass signals because not
only is bit time (Tb) as for baseband signals required but
carrier synchronization is also needed. Carrier
synchronization requires an estimate of the transmitted
frequency (fC) and the arrival phase at the receiver (θ):
s1(t) = sin(2π fC t + θ)
• BPSK signals are symmetrical with:

s1T(t) = – s2T(t) = A sin(2π fc t) S&M Eqs. 5.15-5.19
iTb
∫ [ ± A γ sin (2π f t)]

2
Eb, BPSK = C dt
(i-1)Tb
iTb
2
γ A 2
γ 2 A 2Tb
Eb, BPSK = ∫ [1− cos (4π fC t)] dt =
2 (i-1)Tb
2
 2 Eb   γ 2 A 2Tb 
Pb, BPSK = Q
  = Q  
No  No 
   
• For this analysis of Eb, PSK for BPSK signals it is assumed

that the transmitter produces an integer number of cycles
within one bit period Tb:
S&M Eq. 5.17
2 2 iTb
γ A
Eb, BPSK =
2 ∫ [1− cos (4π f t)] dt
(i-1)Tb
C
0
2 2 2 2 iTb
γ A Tb γ A
Eb, BPSK =
2
−
2 ∫
(i-1)Tb
cos (4π fC t) dt
• However, even for a non-integer number of cycles within

one bit period Tb if 1 / fC = TC << Tb:
S&M Eq. 5.17
iT b
∫
(i-1)Tb
cos (4π fC t) dt << Tb insignificant
2 2 2 2 iTb
γ A Tb γ A
Eb, BPSK =
2
−
2 ∫
(i-1)Tb
cos (4π fC t) dt

and s1(t) = sin (2π fc t)
iTb
ai (iTb ) = ∫
(i-1)Tb
γ si (t) s1(t) dt S&M Eq. 4.67
τ opt = = 0 S&M Eq. 4.71
2

τ opt = 0
S&M Figure 4-16

• For this analysis of Ed, ASK for BASK signals it is assumed

within one bit period:
S&M Eq. 5.27
2 2 iTb
γ A
Ed, BASK =
2 ∫ [1− cos (4π f t)] dt
(i-1)Tb
C
0
2 2 2 2 iTb
γ A Tb γ A
Ed, BASK =
2
−
2 ∫
(i-1)Tb
cos (4π fC t) dt
• BASK OOK signals are not symmetrical with:

s1T(t) = A sin(2π fct) s2T(t) = 0 S&M Eqs. 5.22, 5.23
and s1(t) = sin(2π fct) s2(t) = 0
iTb
ai (iTb ) = ∫ γ si (t) [ s1(t) − s2 (t)] dt

(i-1)Tb
S&M Eq. 5.24-5.26
iTb
γ A Tb
a1(iTb ) = ∫ γ A s (t) dt = 2
1 a2 (iTb ) = 0
(i-1)Tb
2
a2 (iTb ) + a1(iTb ) γ A Tb
τ opt = =
2 4
• BASK signals in general may not be symmetrical with:

s1T(t) = A1 sin(2π fc t) s2T(t) = A2 sin(2π fc t)
and s1(t) – s2(t) = sin (2π fc t) where the amplitude is arbitrary.
iTb
ai (iTb ) = ∫ γ si (t) [ s1(t) − s2 (t)] dt S&M Eq. 4.67

(i-1)Tb
iTb
γ A i Tb
ai (iTb ) = ∫ γ A i s (t) dt = 2
1 S&M Eq. 4.71
(i-1)Tb
2
a2 (iTb ) + a1(iTb ) γ ( A1 + A 2 )Tb
τ opt = =
2 4
• BFSK signals are not symmetrical with:

sT(t) = A sin(2π (fc ± ∆f) t) S&M Eq. 5.30
iTb
∫ [ A γ (sin (2π fC + ∆f ) t − sin (2π fC − ∆f ) t ))] dt

2
Ed, BFSK =
(i-1)Tb
Ed, BFSK = γ 2 A 2 Tb if fC + ∆f = n1 / Tb and fC − ∆f = n2 / Tb

S&M Eq. 5.31
 Ed, FSK   2 2
γ A Tb 
Pb, BFSK = Q  = Q  S&M Eq. 5.32
 2 No   2 No 
   
• For this analysis of Ed, FSK for BFSK signals it is assumed

within one bit period:
S&M Eq. 5.31
2 2 iTb
0
γ A
Ed, BFSK = γ A Tb − ∫ cos (4π (fC + ∆f ) t) dt
2 2
2 (i-1)Tb 0
iT
γ2 A 2 b
− ∫
2 (i-1)Tb
cos (4π (fC − ∆f ) t) dt
iTb 0
− γ2 A 2 ∫
(i-1)Tb
sin (2π (fC + ∆f ) t) sin (2π (fC − ∆f ) t) dt
• BFSK signals are not symmetrical with:

sT(t) = A sin(2π (fc ± ∆f) t) S&M Eq. 5.30
and s1(t) – s2(t) = sin(2π (fc + ∆f) t) – sin(2π (fc – ∆f) t)
iTb
ai (iTb ) = ∫ γ si (t) [ s1(t) − s2 (t)] dt S&M Eq. 5.31

(i-1)Tb
iTb
γ A i Tb
ai (iTb ) = ∫ γ A i s (t) dt = 2
i S&M Eq. 5.33
(i-1)Tb
2
a2 (iTb ) + a1(iTb ) γ ( A1 − A 2 )Tb
τ opt = = = 0
2 4
if A1 = A 2
• A comparison of coherent BPSK, BFSK and BASK

illustrates the functional differences, but BFSK and BASK
uses Ed and not Eb:
 2 Eb   γ 2 A 2Tb  2 2
Pb, BPSK = Q  = Q   E γ A Tb
 No  No  b, BPSK = 2
   
 Ed, FSK   γ 2 A 2Tb 
Pb, BFSK = Q  = Q  Ed, BFSK = γ 2 A 2Tb
 2 No   2 No 
   
 Ed, ASK   γ 2 A 2Tb  γ 2 A 2Tb
Pb, BASK = Q  = Q  Ed, BASK =
 2 No   4 No  2
   
• The normalized Eb, FSK = Eb, PSK = γ2 A2 Tb / 2 (S&M Eq.

5.24) and Eb, ASK = γ2 A2 Tb / 4 (S&M Eq. 5.36) so that:
 2 Eb, PSK   γ 2 A 2Tb  γ 2 A 2Tb

Pb, BPSK = Q  = Q  Eb, BPSK =
 No   No  2
   
 Eb, FSK   γ 2 A 2Tb  γ 2 A 2Tb

Pb, BFSK = Q  = Q  Eb, BFSK =
 No   2 No  2
   
 Eb, ASK   γ 2 A 2Tb 
Pb, BASK = Q  = Q  γ 2 A 2Tb
    Eb, BASK =
 No   4 No  4
Thus there are no practical advantages for either coherent
BFSK or BASK and BPSK is preferred (S&M p. 236).
• For the same Pb BPSK uses the least amount of energy,

BFSK requires twice as much and BASK four times as much
energy:
 2 Eb, PSK   γ 2 A 2T 
Pb, BPSK = Q   = Q b

 N   N 
 o   o 
Argument of
 Eb, FSK   2 2
γ A Tb  Q should be
Pb, BFSK = Q  = Q 
 No   2 No  as large as
   
possible to
 Eb, ASK   γ 2 A 2Tb  minimize Pb
Pb, BASK = Q  = Q 
 No   4 No 
   
Chapter 3
Bandpass Modulation and
Demodulation
• Optimum Bandpass Receiver:
The Correlation Receiver
• Pages 81-85
• The matched filter or

for bandpass
symmetrical
signals can be
simulated in
Simulink:
MS Figure 3.1
• The matched filter or

for bandpass
asymmetrical
signals can also
be simulated
in Simulink:
MS Figure 3.2
• The alternate but

universal structure
which can be used
for both asymmetric
or symmetric binary
bandpass signals can
be simulated in
Simulink: MS Figure 3.3
Chapter 3
Demodulation
• Binary Amplitude Shift Keying
• Pages 86-92
• Binary ASK (OOK) coherent digital communication system

with BER analysis:
Threshold
MS Figure 3.4

Function of Ed / No in a Binary ASK Digital
Communication System with Optimum Receiver
Ed / No dB BER Pb
∞ 0 0
12 2.9 × 10-3 2.53 × 10-3
10 1.12 × 10-2 1.25 × 10-2
8 3.46 × 10-2 3.75 × 10-2
6 7.65 × 10-2 7.93 × 10-2
4 1.335 × 10-1 1.318 × 10-1
2 1.863 × 10-1 1.872 × 10-1
0 2.387 × 10-1 2.394 × 10-1
Chapter 3
Demodulation
• Binary Phase Shift Keying
• Pages 98-103
• Binary PSK coherent digital communication system with

BER analysis:
MS Figure 3.12
• The BER and Pb comparison (SVU Table 3.5, p. 167):

Function of Eb / No in a Binary PSK Digital
Eb / No dB BER Pb
∞ 0 0
10 0 4.05 × 10-6
8 1 × 10-4 2.06 × 10-4
6 2.5 × 10-4 2.41 × 10-3
4 1.31 × 10-2 1.25 × 10-2
2 3.35 × 10-2 3.75 × 10-2
0 8.19 × 10-2 7.93 × 10-2
Chapter 3
Demodulation
• Binary Frequency Shift Keying
• Pages 92-98
• Binary FSK coherent digital communication system with

BER analysis:
fC–∆f
fC+∆f
MS Figure 3.9

Function of Ed / No in a Binary FSK (MFSK) Digital
Ed/No dB BER Pb
∞ 0 0
12 2.5 × 10-3 2.5 × 10-3
10 1.29 × 10-2 1.25 × 10-2
8 3.50 × 10-2 3.75 × 10-2
6 8.04 × 10-2 7.93 × 10-2
4 1.352 × 10-1 1.314 × 10-1
2 1.833 × 10-1 1.872 × 10-1
0 2.456 × 10-1 2.393 × 10-1
• The BER and Pb performance comparison for BASK,

BPSK and BFSK (MFSK):
Ed / No dB BER Pb
10 1.12 × 10-2 1.25 × 10-2 BASK
8 3.46 × 10-2 3.75 × 10-2
Eb / No dB BER Pb
10 0 4.05 × 10-6 BPSK
8 1 × 10-4 2.06 × 10-4
Ed / No dB BER Pb
10 1.29 × 10-2 1.25 × 10-2 BFSK
8 3.50 × 10-2 3.75 × 10-2
• BER and Pb comparison using Eb with Eb, ASK = γ2 A2 Tb2 / 4

and thus reduced by 10 log (0.5) ≈ –3 dB or:
Eb / No dB BER Pb
7 1.12 × 10-2 1.25 × 10-2 BASK
5 3.46 × 10-2 3.75 × 10-2
Eb / No dB BER Pb
10 0 4.05 × 10-6 BPSK
8 1 × 10-4 2.06 × 10-4
Eb / No dB BER Pb
10 1.29 × 10-2 1.25 × 10-2 BFSK
8 3.50 × 10-2 3.75 × 10-2
BASK performs better than BFSK but BPSK is the best.
Chapter 5
and Demodulation
Techniques
• Differential (Noncoherent) Phase
Shift Keying
• Pages 267-271
• Differential (noncoherent) phase shift keying (DPSK) is

demodulated by using the received signal to derive the
reference signal. The DPSK protocol is:
Binary 1: Transmit the carrier signal with the same phase

as used for the previous bit.
Binary 0: Transmit the carrier signal with its phase shifted
by 180° relative to the previous bit.
• The one-bit delayed reference signal ri-1(t) is derived from

the received signal ri(t) and if the carrier frequency fC is an
integral multiple of the bit rate rb:
ri−1(t) = γ A sin (2π fC (t − Tb ) + θ) S&M Eqs.
ri−1(t) = γ A sin (2π fC t + θ) 5.88 and 5.89
The output of the integrator for a binary 0 and binary 1
then is z(iTb) = ± γ2 A2 Tb / 2 (S&M Eqs. 5.91 and 5.93)
• DPSK signals have an equivalent bit interval TDPSK = 2 Tb.

The probability of bit error for DPSK signal is different than
that for coherent demodulation of symmetric or asymmetric
signals and is:
1  E DPSK  1  Eb, DPSK 

Pb, DPSK = exp  −  = exp  − 
2  2 No  2  N o 
γ 2 A 2 Tb S&M Eq. 5.102
Eb, DPSK =
2
Chapter 3
Demodulation
• Differential Phase Shift Keying
• Pages 130-135
• Binary DPSK noncoherent digital communication system

with BER analysis:
one-bit
continuous
BPF delay
MS Figure 3.33
• Binary DPSK noncoherent digital communication system

differential binary encoder Simulink Subsystem:
XOR
one bit
sample delay MS Figure 3.34
• Simulink Logic and

Bit Operations
provides the
Logical Operator
block:
• Simulink Logical Operator blocks can be selected to

provide multiple input AND, OR, NAND, NOR, XOR,
NXOR, and NOT functions:
• The Logic and Bit Operations

can be configured as scalar
Boolean binary (0, 1) or
M-ary (0, 1…M−1) vector
logic functions. Here scalar
Boolean binary data is used.
• The XOR logic generates the DPSK source coding:
Table 3.16 Input Binary Data bi, Differentially Encoded

Binary Data di, and Transmitted Phase φi (Radians) for a
DPSK Signal.
bi di-1 di φi one-bit startup
1 0
1 1 1 0 XOR logic
0 1 0 π 0 0 1
0 0 1 0 0 1 0
1 1 1 0 1 0 0
0 1 0 π 1 1 1
0 0 1 0
1 1 1 0
1 1 1 0
• The Signal Processing

Blockset provides the
Filtering, Analog Filter
Design block:
• The Signal Processing Blockset provides the analog

bandpass filter (BPF) specified as a 9-pole Butterworth
filter with cutoff frequencies of 19 kHz and 21 kHz centered
around the carrier frequency fC = 20 kHz.
rad/s
• The Butterworth BPF is used for the noncoherent receiver.

The coherent receiver uses the integrator as a virtual
BPF:
MS Figure 3.33
DPSK
MS Figure 3.12
PSK

Function of Eb / No in a Binary DPSK Digital
Communication System with Noncoherent Correlation
Receiver Statistical variation
due to small sample
Eb / No dB BER Pb size
∞ 0 0
12 0 ≈ 6.6 × 10-8
10 2 × 10-4 2.3 × 10-5
8 5.1 × 10-3 1.8 × 10-3
6 2.61 × 10-2 9.3 × 10-3
4 7.91 × 10-2 4.06 × 10-2
2 1.559 × 10-2 1.025 × 10-1
0 2.393 × 10-1 1.839 × 10-1
• BER and Pb comparison between noncoherent, source

coded DPSK and coherent BPSK:
Eb / No dB BER Pb
12 0 6.6 × 10-8 DPSK
10 2 × 10-4 2.3 × 10-5
8 5.1 × 10-3 1.8 × 10-3
Eb / No dB BER Pb
10 0 4.05 × 10-6 BPSK
8 2 × 10-4 2.06 × 10-4
6 2.5 × 10-3 2.41 × 10-3
BPSK performs better than DPSK but requires a coherent
reference signal. DPSK performs nearly as well as BPSK
at high SNR.
Chapter 5
and Demodulation
Techniques
• M-ary Bandpass Techniques:
Quaternary Phase Shift Keying
• Pages 274-286
• Quaternary phase shift keying (M-ary, M = 2n = 4 or QPSK)

source codes dibits bi-1bi as a symbol with one possible
protocol as:
bi-1bi = 11 A sin(2π fC t + 45°)

bi-1bi = 10 A sin(2π fC t + 135°)
−
bi-1bi = 00 A sin(2π fC t + 225°)
bi-1bi = 01 A sin(2π fC t + 315°)
The Gray code is used as

for M-ary PAM to improve +
the BER performance by
mitigating adjacent symbol
error. The symbols are best
displayed as a constellation plot S&M Figure 5-36 modified
• Quaternary phase shift

keying (QPSK) displayed
as a constellation plot
−
Note the signs on the
sine reference axes.
cos
constellation points sin

Constellation
Gemini S&M Figure 5-36 modified
Chapter 3
Demodulation
• Multilevel (M-ary) Phase Shift
Keying
• Pages 117-123
• The QPSK signal can be simulated in Simulink using

Subsystems to simplify the design.
QPSK I-Q correlation
receiver
4-level Gray coded 4-level Gray coded

bit to symbol symbol to bit
MS Figure 3.22
• The random binary data source is converted to an M = 4

level Gray encoded symbol by a Simulink Subsystem.
MS Figure 3.22
• The M = 4 level symbol (0, 1, 2 and 3) is Gray encoded by

a Lookup Table Block from the Simulink Blockset.
MS Figure 3.22

(0, 1, 2 and 3) is Gray
encoded by a Lookup
Table Block from the
Simulink Blockset.
• The M = 4 level symbol (0, 1, 2, and 3) is Gray encoded by

01 → 01
10 → 11
11 → 10

is inputted to the Phase
Modulator block with a
carrier frequency fC =
20 kHz, initial phase
φo = π/4 and a phase
deviation factor
kp = π/2 / V
• The M = 4 level symbol (0, 1, 2, 3) and the carrier

frequency fC = 20 kHz, initial phase φo = π/4 and a phase
deviation factor kp = π/2 / V produces the phase shifts:
di = 0 φ = π/4 + 0(π/2) = π/4

di = 1 φ = π/4 + 1(π/2) = 3π/4
di = 2 φ = π/4 + 2(π/2) = 5π/4
di = 3 φ = π/4 + 3(π/2) = 7π/4
MS Figure 3.22
• The modulation phase shifts are the phase angle φ of the

sinusoidal carrier A sin (2π fC t + φ) in QPSK.
di = 0 φ = π/4 + 0(π/2) = π/4 45°

di = 1 φ = π/4 + 1(π/2) = 3π/4 135°
di = 2 φ = π/4 + 2(π/2) = 5π/4 225°
di = 3 φ = π/4 + 3(π/2) = 7π/4 315°
MS Figure 3.22
• The QPSK signal can be resolved into In-phase (I, cosine)

and Quadrature (Q, sine)
constellation
components. For example,
Plot Quadrature
if φ = π/4 = 45°:
−
s1(t) = A sin(2π fC t + 45°) =

cos
A /√2 [ I sin (2π fC t)

+ Q cos (2π fC t) ] = +
A /√2 [ sin (2π fC t)

sin In-phase
+ cos (2π fC t) ]
• The QPSK signal is derived from Gray coded dibits with

00 → 00 (0), 01 → 01 (1), 10 → 11 (3) and 11 → 10 (2).
10 11 11 01 01 11 rb = 1 kb/sec
00 00
M=4 3 2 0 2 0 1 1 2
Delay
± 5 V, fC = 2 kHz, rS = 500 Hz QPSK signal
TS = 2 msec
• The QPSK signal can be decomposed into I and Q BPSK

signals which are orthogonal to each other.
±5V QPSK signal, fC = 2 kHz, rS = 500 b/sec
± 5 / √2 = 3.536 V Binary PSK signal, sine carrier (I)
± 5 / √2 = 3.536 V Binary PSK signal, cosine carrier (Q)
TS = 2 msec
• The orthogonality of the I and Q components of the QPSK

signal can be exploited by the universal coherent receiver.
The orthogonal I and Q components actually occupy the
same spectrum without interference. The coherent
reference signals are:
Quadrature In-phase
s1(t) = cos (2π fC t + θ) s2(t) = sin (2π fC t + θ)
S&M Figure 5-40

• The orthogonality of the QPSK signals can be shown by

observing the output of the quadrature correlator to the I
and Q signal.
iTS
γA
z1(nTS ) = ∫
2 (i-1)TS
dI sin(2π fC t) + dQ cos(2π fC t) cos (2π fC t) dt
0
iTS
γA
z1(nTS ) = ∫
2 (i-1)TS
dI sin(2π fC t) cos (2π fC t) dt +
iTS
S&M Eq. 5.109
γA
∫
2
dQ cos (2π fC t) dt
2 (i-1)TS
z1(nTS)
γ A TS
z1(nTS ) = dQ
2 2
• The probability of bit error Pb and the energy per bit Eb for a
QPSK signal is the same as that as for a BPSK signal but
with a I and Q carrier amplitude of A / √2 .
 2 Eb, PSK   γ 2 A 2Tb 
Pb, BPSK = Q  = Q 
 No   No 
    note TS
 2 Eb, QPSK   γ 2 A 2TS 
Pb, QPSK = Q  = Q  S&M Eq. 5.117
 No   2 No 
   
γ 2 A 2Tb z1(nTS)
Eb, BPSK =
2 note TS
γ 2 A 2TS
Eb, QPSK =
4
• Since TS = 2 Tb BPSK and QPSK have the same Pb but

QPSK can have twice the data rate rb = 2 rS within the same
bandwidth because of the orthogonal I and Q components.
 2 Eb   γ 2 A 2Tb 
Pb, BPSK = Pb, QPSK = Q
  = Q  
No  No 
   
γ 2 A 2Tb
Eb, BPSK = Eb, QPSK =
2
z1(nTS)
• QPSK coherent digital communication system with BER

analysis:
4-Level Gray coded 4-Level Gray coded

bit to symbol symbol to bit
MS Figure 3.22
• QPSK coherent digital communication system uses a

4-level Gray coded bit to symbol converter Simulink
Subsystem.
MS Figure 2.43
• QPSK coherent receiver uses an I-Q correlator Simulink

Subsystem
MS Figure 3.24
• The I-Q correlation receiver is the universal structure with

an integration time equal to the symbol time TS.
MS Figure 3.24
• The output of the I-Q correlation receiver is a dibit and

converted to an M = 2n = 4 level symbols (0, 1, 2, and 3).
dibits M-ary scaling

MS Figure 3.24
• The BER and Pb comparison for 4-PSK (QPSK):
Table 3.11 Observed BER and Theoretical Upper-Bound

of Pb as a Function of Eb / No in a Gray coded 4-PSK
(QPSK) Digital Communication System with Optimum
Receiver
Ed/No dB BER Pb
∞ 0 0
12 0 ≈10-8
10 0 ≈10-6
8 2 × 10-4 ≈10-4
6 2.3 × 10-3 2.4 × 10-3
4 1.20 × 10-2 1.25 × 10-2
2 3.62 × 10-2 3.75 × 10-2
0 7.65 × 10-2 7.85 × 10-2
• The single-sided power spectral density PSD of the QPSK

signal uses rs = rb /2 and is:
GQPSK(f) = 1/2 GPAM(f - fC)

GPAM(f) = A2 / rs sinc2 (π f / rs)
No carrier rs = 500 s/sec, rb = 1 kb/sec

Sinc2
MS Figure 3.25
• The single-sided power spectral density PSD of BPSK has

double the bandwidth than that for QPSK for the same bit
rate rb = 1/Tb,
No carrier rb = 1 kHz
Sinc2
MS Figure 3.14
• The bandwidth of a QPSK signal as a percentage of total

power is half that for the same bit rate rb = 1/Tb BPSK
signal since rs = rb /2 or Ts = 2Tb (MS Table 3.9).
2/Ts 1/Tb 90%

3/Ts 1.5/Tb 93%
4/Ts 2/Tb 95%
6/Ts 3/Tb 96.5%
8/Ts 4/Tb 97.5%
10/Ts 5/Tb 98%
Chapter 5
and Demodulation
Techniques
8-Phase Shift Keying
• Pages 286-292
• M-ary phase shift keying (M = 8 or 8PSK) source codes

tribits bi-2bi-1bi as a symbol with one possible protocol as:
bi-2bi-1bi = 000 A sin(2π fC t + 0°) Constellation Plot

bi-2bi-1bi = 001 A sin(2π fC t + 45°)
bi-2bi-1bi = 011 A sin(2π fC t + 90°)
Quadrature
bi-2bi-1bi = 010 A sin(2π fC t + 135°)
bi-2bi-1bi = 110 A sin(2π fC t + 180°)
bi-2bi-1bi = 111 A sin(2π fC t + 225°) cos
bi-2bi-1bi = 101 A sin(2π fC t + 270°)
bi-2bi-1bi = 100 A sin(2π fC t + 315°)
θ
I, Q = 0, ± 1/√2, ± 1
s(t) = A [ I sin (2π fC t) + sin In-phase
Q cos (2π fC t) ] S&M Figure 5-43

• The correlation receiver

for 8-PSK uses four
reference signals:
sref n(t) =
sin (2π fC t +
n 45° + 22.5°)
φ
n = 0, 1, 2, 3
φ = 22.5°, 67.5°,
112.5°, 157.5°
S&M Eq. 5.124

• The output from any one of the four correlators is:

iTS
z1(nTS ) = γ A ∫
(i-1)TS
sin (2π fC t + θ) sin (2π fC t + φ)dt
0
iT iT
γA S S
z1(nTS ) = ∫
2 (i-1)TS
cos (θ − φ) dt − ∫ cos (4π fC t + θ + φ) dt
(i-1)TS
γ A TS
z1(nTS ) = cos (θ − φ)
2
S&M Eq. 5.125
z
1
(
n
T
S
)
• The correlator output is > 0 if | θ – φ | < 90° and < 0 if not

because of the cos (θ – φ) term. For example, if s6(t) is
received, the ABCD correlator sign output is: – – – +. The
patterns of signs are unique and can be decoded to bi-2bi-1bi
(S&M Tables 5-7 and 5-8)
D: sref 4(t)
C: sref 3(t)
z
1
(
B: sref 2(t) n
T
A: sref 1(t) S
)
• The probability of symbol error PS for coherently

demodulated M-ary PSK is:
 A 2T 2 π

PS coherent M -ary PSK ≈ 2Q  S
sin  M ≥ 4
 No M 

  Eb  2 π 
PS coherent M -ary PSK ≈ 2Q  2 log2 M   sin  M ≥ 4
  No  M
Ps  
S&M Eq. 5.126
S&M Figure 5-46
Eb / No dB

demodulated M-ary PSK is:
Ps S&M Figure 5-46
Eb/No dB
• The probability of symbol error PS must be related to

probability of bit error Pb for consistency. If Gray coding is
used, assume that errors will only be due to adjacent
symbols. Thus each symbol error produces only one bit in
error and log2 (M – 1) correct bits or:
1
Pb errors due to adjacent symbols = PS S&M Eq. 5.127
log2 M
However for M-ary PSK with M > 4 the assumption of errors
being due to only adjacent symbols is invalid. For the
worst case there are M – 1 incorrect symbols and in M / 2 of
these a bit will different from the correct bit so that:
1 M
PS ≤ Pb ≤ PS S&M Eq. 5.129
log2 M 2 (M − 1)
Chapter 5
and Demodulation
Techniques
Quaternary Frequency Shift Keying
• Pages 292-298
• The analytical signal for quaternary (M-ary, M = 2n = 4)

frequency shift keying (QFSK or 4-FSK) is:
s4-FSK(t) = A sin (2π (fC + 3∆f) t + θ) if bi-1bi = 11

s4-FSK(t) = A sin (2π (fC + ∆f) t + θ) if bi-1bi = 10
s4-FSK(t) = A sin (2π (fC – ∆f) t + θ) if bi-1bi = 00
s4-FSK(t) = A sin (2π (fC – 3∆f) t + θ) if bi-1bi = 01
MS Figure 3.19
fC + 3∆f fC – ∆f fC – 3∆f fC + ∆f fC + 3∆f

11 00 01 10 11
• Chose fC and ∆f so that if there are a whole number of half

cycles of a sinusoid within a symbol time TS for M = 4 for
orthogonality of the signals so that a correlation receiver
can be utilized.
s4-FSK(t) = A sin (2π (fC + 3∆f) t + θ) if bi-1bi = 11

s4-FSK(t) = A sin (2π (fC + ∆f) t + θ) if bi-1bi = 10
s4-FSK(t) = A sin (2π (fC – ∆f) t + θ) if bi-1bi = 00
s4-FSK(t) = A sin (2π (fC – 3∆f) t + θ) if bi-1bi = 01
MS Figure 3.19
• The correlation receiver

for 4-FSK uses four
reference signals:
sref n(t) =
sin (2π (fC + n ∆f) t)
n = ±1, ±3
S&M Figure 5-49


demodulated M-ary FSK is:
 A 2T 
PS coherent M-ary FSK ≤ (M − 1) Q  s
 M ≥ 4
 2 No 
Ps  
PS coherent M-ary FSK =
  Eb  
(M − 1) Q  log2 M    M ≥ 4
  No  

S&M Eq. 5.132
S&M Figure 5-51

Eb/No dB

demodulated M-ary FSK is:
Ps S&M Figure 5-51
Eb/NodB
Chapter 3
Demodulation
• Multilevel (M-ary) Frequency
Shift Keying
• Pages 110-116
• 4-FSK coherent digital communication system with BER

analysis:
MS Figure 3.18
• The dibits are converted to a symbol and scaled. The data

is not Gray encoded. For M-ary FSK symbol errors are
equally likely among the M – 1 correlators and there is no
advantage to Gray encoding.
MS Figure 3.18

analysis:
4-FSK correlation receiver
MS Figure 3.18

analysis: 4-FSK correlation receiver
MS Figure 3.20
• The 4-FSK correlation

receiver has four
correlators with an
integration time equal
to the symbol time TS.
• The symbols are converted to dibits. The original data is

not Gray encoded and is therefore not Gray decoded.
MS Figure 3.18
• The BER and Pb comparison for 4-FSK:
Table 3.10 Observed BER and Theoretical Upper Bound

of Pb as a Function of Eb / No in 4-level FSK Digital
Ed/No dB BER Pb
∞ 0 0
12 0 ≈10-8
10 0 ≈10-6
8 1 × 10-4 ≈10-4
6 5.1 × 10-3 4.8 × 10-3
4 2.26 × 10-2 2.52 × 10-2
2 5.97 × 10-2 7.54 × 10-2
0 1.209 × 10-1 1.586 × 10-1
• The single-sided power spectral density PSD with a

minimum carrier frequency deviation (MFSK) for M-ary
FSK is ∆f = 1/2TS = rS/2. For MFSK the carriers should be
spaced at multiples of 2∆f = 1/TS = rS (S&M Eq. 5.131 is
incorrect). Here ∆f = 2 rS = 1 kHz
M=4 rs = 500 s/sec, rb = 1 kb/sec

∆f = 1 kHz Sinc2
MS Figure 3.21
• The bandwidth of a M-ary FSK signal as a percentage of

total power (MS Table 3.9).
2( M – 1) ∆f + 4/Ts 95%
2 (M – 1) ∆f + 6/Ts 96.5%
2 (M – 1) ∆f + 8/Ts 97.5%
2 (M – 1) ∆f + 10/Ts 98%
For MFSK: ∆f = 1/2TS = rS/2

M = 2n and rS = rb/n
Chapter 5
and Demodulation
Techniques
Quadrature Amplitude Modulation
• Pages 298-301
• The analytical signal for quadrature amplitude modulation

(QAM) has I-Q components:
sQAM(t) = I sin (2π fCt) + Q cos (2π fCt)
A QAM signal has both

16-ary
amplitude and phase QAM
components which can be
shown in the constellation Q
plot.
S&M Figure 5-53

I
• An M-ary PSK signal also has I-Q components but the

amplitude is constant and only the phase varies:
sQAM(t) = I sin (2π fCt) + Q cos (2π fCt)
S&M Figure 5-54

16-ary 16-ary
PSK QAM
Q Q
I I
• The orthogonality of the I and Q components of the QAM

signal can be exploited by the universal coherent receiver.
The orthogonal I and Q components actually occupy the
same spectrum without interference. The coherent
reference signals are:
Quadrature In-phase
s1(t) = cos (2π fCt) s2(t) = sin (2π fCt)
S&M Figure 5-55

• An upper-bound for the probability of symbol error PS for

coherently demodulated M-ary QAM is:
 3 Es 
PS coherent M-ary QAM ≤ 4Q   S&M Eq. 5.135
 (M − 1) No
 
QAM BER curve M = 256
M=4
• An M-ary QAM constellation plot shows the stability of the

signaling and the transition from one signal to another:
256-ary QAM
16-ary QAM
Chapter 3
Demodulation
• Quadrature Amplitude
Modulation
• Pages 123-130
• QAM coherent digital communication system with BER

analysis:
4 bit to I,Q symbol
16-QAM correlation
QAM receiver
16-level symbol to bit

MS Figure 3.26

analysis: 4 bit to I-Q symbol Simulink subsystem.
MS Figure 3.27

analysis: Table 3.12 I and Q output amplitudes
Input I Q Input I Q Input I Q

0 –1 1 5 1 3 10 –3 –1
1 –3 1 6 3 1 11 –3 –3
2 –1 –3 7 3 3 12 1 –1
3 –3 3 8 –1 –1 13 3 –1
4 1 1 9 –1 –3 14 1 –3
15 3 –2
I LUT ± 1 to ± 3
Q LUT
MS Figure 3.27
symbol 0 to 15

analysis: QAM modulator Simulink subsystem.
MS Figure 3.27

analysis: 16-QAM correlation receiver Simulink subsystem.
MS Figure 3.30
• QAM coherent digital

communication system
with BER analysis:
Table 3.14 I, Q Symbol LUT
16-level output amplitudes
I Q Output I Q Output
1 1 11 3 1 1
1 2 9 3 2 0
1 3 14 3 3 4
1 4 15 3 4 6
2 1 10 4 1 3
2 2 8 4 2 2
2 3 12 4 3 5
2 4 13 4 4 7

analysis: 16 level symbol to 4 bit Simulink subsystem.
MS Figure 3.31

analysis: 16 level symbol to 4 bit Simulink subsystem.
MS Figure 3.31
• The single-sided power spectral density PSD of the 16-ary

QAM has a bandwidth of 1/M that of a PSK signal with the
same data rate rb.
rs = 250 s/sec, rb = 1 kb/sec
Sinc2
no discrete MS Figure 3.32

component at fC =
20 kHz
1 kHz BPSK PSD

rb = 1 kb/sec
16-ary QAM PSD
250 Hz rb = 1 kb/sec M = 4
rS = 250 s/sec
• The bandwidth of an M-ary QAM signal as a percentage of

total power is 1/n that for the same bit rate rb = 1/Tb BPSK
signal since rs = rb/n or Ts = nTb where M = 2n (MS Table
3.14).
2/Ts 2/nTb 90%

3/Ts 3/nTb 93%
4/Ts 4/nTb 95%
6/Ts 6/nTb 96.5%
8/Ts 8/nTb 97.5%
10/Ts 10/nTb 98%
• 16-QAM coherent digital communication system received

I-Q components can be displayed on as a signal trajectory
or constellation plot .
real-imaginary (a + b j)
conversion to complex
polar (M exp(jθ)) conversion
Figure 3.42
• The Real-Imaginary to
Complex conversion block
is in the Math Operations,
Simulink Blockset
Figure 3.42
• The constellation plot

(scatter plot) and
signal trajectory are
Comm Sinks blocks
from the
Communications
Blockset
Figure 3.42
• The 16-ary QAM I-Q component constellation plot with

Eb/No → ∞ (MS Figures 3.43, 3.45). signal transitions
signal points decision boundaries


Eb/No = 12 dB, Pb ≈ 10-4 (MS Figures 3.44, 3.46 (top))

Eb/No = 6 dB, Pb = 3.67 x 10-4 (MS Figures 3.44, 3.46 (bot))
End of Chapter 5
and Demodulation
Techniques
Chapter 6
Analog Modulation
and Demodulation
Chapter 6
Analog Modulation
and Demodulation
• Amplitude Modulation
• Pages 306-309
• The analytical signal for double sideband, large carrier

amplitude modulation (DSB-LC AM) is:
sDSB-LC AM(t) = AC (c + s(t)) cos (2π fC t)
where c is the DC bias or offset and AC is the carrier

amplitude. The continuous analog signal s(t) is a baseband
signal with the information content (voice or music) to be
transmitted.
• The baseband power spectral density (PSD) spectrum of

the information signal s(t) or S(f) for voice has significant
components below 500 Hz and a bandwidth of < 8 kHz:
S(f) = F(s(t))
The single-sided spectrum of the modulated signal is:
F(AC (c + s(t)) cos (2π fC t)) = S(f – fC)
500 Hz Power Spectral Density of s(t) 8 kHz
dB
• The single-sided (positive frequency axis) spectrum of the

modulated signal replicates the baseband spectrum as a
double-sided spectrum about the carrier frequency.
Double-sided spectrum Carrier 25 kHz
Baseband spectrum
• The double-sided modulated spectrum about the carrier

frequency has an lower (LSB) and upper (USB) sideband.
• The modulated DSB-LC AM signal shows an outer envelope

that follows the polar baseband signal s(t).
• The analytical signal for double sideband, suppressed

carrier amplitude modulation (DSB-SC AM) is:
sDSB-SC AM(t) = AC s(t) cos (2π fC t)
where AC is the carrier amplitude. The single-sided

spectrum of the modulated signal replicates the baseband
spectrum as a double-sided spectrum about the carrier
frequency but without a carrier component.
• The analytical signal for double sideband, suppressed

carrier amplitude modulation (DSB-SC AM) is:
where AC is the carrier amplitude. The modulated signal

sDSB-SC AM(t) looks similar to s(t) but has a temporal but not
spectral carrier component.
• The DSB-LC AM and the DSB-SC AM modulated signals

have the same sidebands.
DSB-LC AM Carrier 25 kHz
DSB-SC AM No carrier
• The modulated DSB-LC AM and the DSC-SC AM signals

are different.
• The modulated DSB-SC AM signal has an envelope that

follows the polar baseband signal s(t) but not an outer
envelope.
Chapter 6
Analog Modulation
and Demodulation
• Coherent Demodulation
of AM Signals
• Pages 309-315
• The DSB-SC AM coherent receiver has a bandpass filter

centered at fC and with a bandwidth of twice the bandwidth
of s(t) because of the LSB and USB. The output of the
multiplier is lowpass filtered with a bandwidth equal to
the bandwidth of s(t).
r(t) = γ sDSB-SC(t) + n(t)
• The DSB-SC AM received signal is r(t) = γ sDSB-SC(t) + n(t).
The bandpass filter passes the modulated signal but filters
the noise:
z(t) = γ sDSB-SC(t) + no(t) S&M Eq. 6.3
no(t) has a Gaussian distribution. The bandpass filter has a

center frequency of fC = 25 kHz and a -3 dB bandwidth of 8
kHz (25 ± 4 kHz).
• The filter noise no(t) has a flat power spectral density

within the bandwidth of the bandpass filter:
no(t)
PSD 21 kHz 29 kHz
fC = 25 kHz
• The filter noise no(t) can be described as a quadrature

representation:
no(t) = W(t) cos (2π fCt) + Z(t) sin (2π fCt) S&M Eq. 5.62R
In the coherent receiver the noise is processed:
no(t) cos (2π fCt) = W(t) cos2 (2π fCt) + S&M Eq. 6.5
Z(t) cos (2π fCt) sin (2π fCt)
PSD 21 kHz 29 kHz
fC = 25 kHz
• Applying the trignometric identity the filter noise no(t) is:
no(t) cos (2π fCt) = ½ W(t) + ½ W(t) cos (4π fCt) +

½ Z(t) sin (4π fCt) S&M Eq. 6.5
After the lowpass filter in the receiver the demodulated

signal is:
sdemod(t) = ½ γ AC s(t) + ½ W(t) S&M Eq. 6.7
PSD 21 kHz 29 kHz
fC = 25 kHz
• The transmitted DSB-SC AM signal is:
The average normalized bi-sided power of sDSB-SC(t) is

found in the spectral domain with S(f) = F (s(t)):
2
1
Ptrans = A 2 ∫ [S(f − fC ) + S(f + fC )] df S&M Eq. 6.8
2
• The dual-sided spectral do not overlap (at zero frequency)

and the cross terms are zero so that:
2
1
Ptrans = A 2 ∫ [SDSB-SC (f − fC ) + SDSB-SC (f + fC )] df
2
A2 S&M Eq. 6.9
Ptrans = Ps
2
where Ps is the average normalized power of s(t).
• The average normalized power of s(t) is found in the

spectral domain:
2 2
Ps = ∫ S(f) df = ∫ S(f + fC ) df
S&M Eq. 6.10
In a noiseless channel the power in the demodulated
DSB-SC AM signal is:
1 2 2 γ2
Pdemod, noiseless = γ A Ps = Ptrans S&M Eq. 6.11
4 2
• The average normalized power of the processed noise is:

1
Pprocessed noise = No (2 B)
4
The signal-to-noise power ratio then is:
γ2
Ptrans 2
2 γ Ptrans
SNRcoherent DSB-SC = = S&M Eq. 6.12
1 No B
No (2 B)
4
2B
• The DSB-SC AM coherent receiver requires a phase

and frequency synchronous reference signal. If the
reference signal has a phase error φ then:
SNRcoherent DSB-SC phase error =
S&M Eq. 6.17
γ 2 cos2 ϕ Ptrans
No B
• The DSB-SC AM coherent receiver requires a phase

and frequency synchronous reference signal. If the
reference signal has a frequency error ∆f then:
Sdemod frequency error(t) = ½ γ AC s(t) cos (2π ∆f t)

+ ½ X(t) cos (2π ∆f t)
+ ½ Y(t) sin (2π ∆f t) S&M Eq. 6.18
• Although the noise component remains the same, the

amplitude of the demodulated signal varies with ∆f:
Chapter 6
Analog Modulation
and Demodulation
• Non-coherent Demodulation
of AM Signals
• Pages 315-326
• The non-coherent AM (DSB-LC) receiver uses an envelope

detector implemented as a semiconductor diode and a low-
pass filter:
The DSB-LC AM analytical signal is:
sDSB-LC AM(t) = AC (c + s(t)) cos (2π fC t)
where c is the DC bias (offset).

• The envelope detector is a half-wave rectifier and

provides a DC bias (c) to the processed DSB-LC AM
signal :
c = DC bias
• The output of the half-wave diode rectifier is low-pass

filtered to remove the carrier frequency and outputs the
envelope which is the information:
• The DSB-LC AM signal can be decomposed as:
sDSB-LC AM(t) = s(t) cos (2π fC t) + AC c cos (2π fC t)

S&M Eq. 6.20R
The average normalized power of the information term:

A C2
Pinfo term = PS S&M Eq. 6.23
2
• The average normalized transmitted power is:

T
1
∫ [A c cos(2πfC t)] dt
2
Pcarrier term = C
T 0
A C2 c 2
Pcarrier term = S&M Eq. 6.24
2
Since s(t) + c must be >= 0 to avoid distortion in the
DSB-LC AM signal: c ≥ | min [s(t)] | or c2 ≥ s2(t) for all t.
• Therefore c2 ≥ Ps and for DSB-LC AM:

Pcarrier term ≥ Pinfo term S&M Eq. 6.28
The power efficiency η of a DSB-LC AM signal is:
Pinfo term Pinfo term
η = = ≤ 0.5
Pcarrier term + Pinfo term Ptrans DSB-LC AM term
S&M Eq. 6.29

• The DSB-LC AM signal wastes at least half the

transmitted power because the power in the carrier term
has no information:
Pcarrier term ≥ Pinfo term η ≤ 0.5
The modulation index m is defined as:
max [ s(t) + c ] − min [ s(t) + c ]
m = S&M Eq. 6.30
max [ s(t) + c ] + min [ s(t) + c ]
• The modulation index m defines the power efficiency but

m must be less than 1. If m > 1 then min [s(t) + c] < 0 and
distortion occurs.
max [ s(t) + c ] − min [ s(t) + c ]
m = S&M Eq. 6.30
max [ s(t) + c ] + min [ s(t) + c ]
• The average normalized power of the demodulation

noiseless DSB-LC AM signal is:
Pdemod, noiseless = 2 γ 2 Pinfo term S&M Eq. 6.39
Then the signal-to-noise power ratio for the DSB-LC AM

signal is:
S&M Eq. 6.40
2 γ 2 Pinfo term γ 2 Ptrans DSB-LC
SNRnoncoherent DSB-LC = = η
No (2 B) No B
2B
• The non-coherent AM (DSB-LC) receiver is the crystal

radio which needs no batteries! Power for the high-
impedance ceramic earphone is obtained directly from the
transmitted signal. For simplicity, the RF BPF is omitted
and the audio frequency filter is a simple RC network.
Chapter 6
Analog Modulation
and Demodulation
• Frequency Modulation and
Phase Modulation
• Pages 334-343
• The analytical signal for an analog phase modulated (PM)

signal is:
sPM(t) = AC cos [2π fC t + α s(t)] S&M Eq. 6.53
where α is the phase modulation constant rad/V and AC is

the carrier amplitude. The continuous analog signal s(t) is a
baseband signal with the information content (voice or
music) to be transmitted.
• The analytical signal for an analog frequency modulated

(FM) signal is:
sFM(t) = AC cos{ 2π [fC + k s(t)] t + φ] S&M Eq. 6.53
where k is the frequency modulation constant Hz / V, AC

is the carrier amplitude and φ is the initial phase angle at
t = 0. The continuous analog signal s(t) is a baseband
signal with the information content.
• The instantaneous phase of the PM signal is:
ΨPM(t) = 2π fC t + α s(t) S&M Eq. 6.56
The instantaneous phase of the FM signal is:
ΨFM(t) = 2π [fC + k s(t)] t + φ] S&M Eq. 6.57
The instantaneous phase is also call the angle of the signal.

The instantaneous frequency is the time rate of change of
the angle:
f(t) = (1/2π) dΨ(t) / dt S&M Eq. 6.58

• The instantaneous frequency of the unmodulated carrier

signal is:
fcarrier(t) = dΨcarrier(t) / dt = d/dt {2π fCt + φ} S&M Eq. 6.59
The instantaneous phase is also:

t t
Ψ(t) = ∫ f (λ) dλ = ∫ f (λ) dλ + φ S&M Eq. 6.60
-∞ 0
There are practical limits on instantaneous frequency and

instantaneous phase. To avoid ambiguity and distortion in
FM signals due to phase wrapping:
k s(t) ≤ fC for all t S&M Eq. 6.61

• To avoid ambiguity and distortion in PM signals due to

phase wrapping:
-π < α s(t) ≤ π radians for all t S&M Eq. 6.61
Since FM and PM are both change the angle of the carrier

signal as a function of the analog information signal s(t), FM
and PM are called angle modulation.
For example, is this signal FM, PM or neither:

t
x(t) = AC cos { 2π fCt + ∫ k s(λ) dλ + φ} S&M Eq. 6.60
-∞
• The instantaneous phase of the signal is:

t
Ψx(t) = 2π fCt + ∫ k s(λ) dλ + φ S&M Eq. p. 336
-∞
which is not a linear function of s(t) so the signal is not PM.
The instantaneous frequency of the signal is:
fx(t) = (1/2π) dΨx(t) / dt = fC + k s(t) / 2π
and the frequency difference fx – fC is a linear function of s(t)

so the signal is FM.
The maximum phase deviation of a PM signal is

max | αs(t) |. The maximum frequency deviation of a FM
signal is ∆f = max | k s(t) |.
• The spectrum of a PM or FM signal can be developed as

follows: S&M Eqs. 6.64 through 6.71
v(t) = A C sin(2π fC t + β sin 2π fm t)
v(t) = Re { exp(j 2π fC t + j β sin 2π fm t) }
now exp(j 2π fC t + j β sin 2π fm t) =
cos (2π fC t + β sin 2π fm t) + j sin (2π fC t + β sin 2π fm t)
v(t) = Im { A C exp(2π fC t + jβ sin 2π fm t) }
∞
now exp(j β sin 2π fm t) = ∑ c exp(j 2π n f
n = -∞
n m t)
after further development Bessel function of the first kind

∞
exp(j β sin 2π fm t) = ∑ J (β) exp(j 2π n f
n = -∞
n m t)
• Bessel functions of the first kind Jn(β) are tabulated for FM

with single tone fm angle modulation (S&M Table 6.1):
β
n
• For single tone fm angle modulation the spectrum is periodic

and infinite in extent:
∞
v(t) = A C ∑ J (β) sin[2π (n f
n = -∞
n m + fC ) t] S&M Eq. 6.72
β
• The complexity of the Bessel function solution for the

spectrum of a single tone angle modulation can be
simplified by the Carson’s Rule approximation for the
bandwidth B. Since β = ∆f / fm:
B = 2 (β + 1) fm = 2 (∆f + fm) Hz S&M Eq. 6.74
β
• Carson’s Rule for the approximate bandwidth of an angle

modulated signal was developed by John R. Carson in
1922 while he worked at AT&T. Prior to this in 1915 he
presaged the concept of bandwidth
efficiency in AM by proposing
the suppression of a sideband
(see S&M p. 326-333):
B = 2 (β + 1) fm = 2 (∆f + fm) Hz
1886-1940
• The normalized power within the Carson’s Rule bandwidth

for a single tone angle modulated signals is:
A C2 β+1 2
Pin-band, sinusoid = ∑
2 n = -(β+1)
Jn (β) S&M Eq. 6.75
Note that J-n(β) = ± Jn(β) so that J-n2(β) = Jn2(β) and for the
normalized power calculation the sign of J(β) is not used.
Spectrum of
single tone FM
modulation
• The analog FM power spectral density PSD of the voice

signal has a bandwidth predicted only by Carson’s Rule
since it is not a single tone.
Voice
PSD
• Here fmax = 4 kHz, k = 25 Hz/V and ∆fmax = 40(25) = 1 kHz.

The Carson’s Rule approximate maximum bandwidth
B = 2 (∆f + fm) = 10 kHz or ± 5 kHz (but seems wrong!)
40
Voice
fC
PSD
Bandwidth
• A 200 Hz single tone FM signal has a PSD with periodic

terms at fC ± n fm = 25 ± 0.2 n kHz.
fC
200 Hz
PSD
• Here fm = 200 Hz, k = 25 Hz/V and ∆fmax = 40(25) = 1

kHz. The Carson’s Rule approximate maximum bandwidth
B = 2 (∆f + fm) = 2.4 kHz or ± 1.2 kHz:
fC
200 Hz
PSD
Bandwidth
• Since β = ∆f / fm = 1 kHz / 0.2 kHz = 5 and the Bessel

function predicts a bandwidth of 2 n fm = 2(12)(200) =
4.8 kHz (since n = 12 for β = 5 from Table 6.1):
fC
200 Hz
PSD
Bandwidth
Chapter 6
Analog Modulation
and Demodulation
• Noise in FM and PM Systems
• Pages 347-355
• A general angle modulated transmitted signal, where Ψ(t)

is the instantaneous phase, is:
sangle-modulated(t) = AC cos [Ψ(t)] S&M Eq. 6.86
The received signals is:
rangle-modulated(t) = γ AC cos [Ψ(t)] + n(t) S&M Eq. 6.87

• The analytical signal for PM is:
sPM(t) = AC cos [Ψ(t)] = AC cos [2π fC t + α s(t)]

S&M Eq. 6.53
After development the SNR for demodulated PM is:
SNRPM = (αγ AC)2 PS / (2 No fmax) S&M Eq. 6.98
where −π < α s(t) ≤ π for all t.

• The analytical signal for FM is:
sFM(t) = AC cos [Ψ(t)] = AC cos [2π fC t + ∫ k s(λ) dλ]

S&M Eq. 6.53
After development the SNR for demodulated FM is:
SNRFM = 1.5 (k γ AC /(2π) )2 PS / (No fmax3) S&M Eq. 6.98
where k s(t) ≤ fC for all t.

End of Chapter 6
Analog Modulation
and Demodulation
Chapter 7
Multiplexing Techniques
Chapter 7
• Time Division Multiplexing
• Pages 364-368
• Time division multiplexing (TDM) combines several low,

fixed and predefined bit rate sources into a single high
speed bit stream for transmission over a single digital
communication channel:
Time slots
Multiplexer Demultiplexer
• The TDM time slots have to be chosen properly. If the time

slots are too small (for example, 1 bit) then the multiplexer
and demultiplexer must switch rapidly. If the time slots are
too large (for example, 64 Kb) then the data must be
buffered and delay would be produced. TDM is used for
baseband (not bandpass) data transmission.
Time slots
Multiplexer Demultiplexer
• The T1 TDM system for telephone networks uses a 193 bit

frame where each frame has 24 8-bit slots and 1-bit in each
frame for signaling. The T1 bit rate rb = 1.544 Mb/sec.
S&M Figure 7-3
The duration of each frame Tf = 193 b / 1.544 Mb/sec =

125 µsec or 8 k samples/sec. Here all the data sources
have the same data rate.
• If the data sources have different rates a multiplexer

scheme must reconcile the disparate rates.
S&M Figure 7-4
The data rates are in the lowest possible ratio of 8:16:24 or

1:2:3 for a total of 1 + 2 + 3 = 6 slots. The slots are
apportioned to the channels as: a b b c c c and the channel
data rate is 48 kb/sec.
Another example is data rates of 10, 15, 20, and 30 Kb/sec

which reduces to 2:3:4:6 (the LCD is 5) for 2 + 3 + 4 + 6 =
15 slots and the channel data rate is 45 kb/sec.
• If data arrives asynchronously from variable rate sources, a

statistical multiplexer with input buffers is used. The design
is performed by observation and tested in simulation.
S&M Figure 7-5

• The statistical TDM packet consists of a start flag, address

field, control field, information bits, error control, and an end
flag. S&M Figure 7-6
TCP/IP data packet

• TDM with four equal data rate sources rb = 250 b/sec and a
transmission rate rTDM = 1 kb/sec
0110
• TDM with unequal data rate sources rb = 250, 250 and 500
b/sec and a transmission rate rTDM = 1 kb/sec
rb = 500 b/sec (input)
rTDM = 1 kb/sec
1 msec delay rb = 500 b/sec (output)

Chapter 7
• Frequency Division Multiplexing
• Pages 368-370
• Frequency division multiplexing (FDM) divides the total

bandwidth available to the system into non-overlapping
frequency sub-bands for transmission over a single digital
communication channel.
• FDM usually utilizes guard bands to separate the digital

data transmissions. FDM is used for bandpass (not
baseband) data transmission
• FDM for FM broadcasting over a cable system:
WPTS
Pittsburgh
WQAN
Scranton
WXTU
S&M
Figure 7-7a
• FDM employs a common shared bandwidth but here with

no apparent guard bands.
S&M Figure 7-7b
• Optical FDM is usually called wavelength division

multiplexing (WDM) and utilizes separate wavelengths
(λ) of light.
• TDM is used for baseband and FDM for bandpass data

transmission. FDM requires the assignment and
coordination of carrier frequencies which can be
problematic. Code division multiple access (CDMA) utilizes
spread spectrum modulation over the same frequency
band. CDMA is considered in EE4542 Telecommunications
Engineering.
• FDM PSD with three equal data rate sources rb = 1 kb/sec

and three carrier frequencies of 15.5, 20 and 24.5 kHz.
BFSK (MFSK)
BASK (OOK) BPSK

End of Chapter 7
Chapter 8
Analog-to-Digital and
Digital to Analog
Conversion
Chapter 8
Digital-to-Analog
Conversion
• Sampling and Quantization
• Pages 390-391
• Traditional analog transmission (AM, FM and PM) are less

complex than digital data transmission have been the basis
of broadcasting and communication for 100 years.
S&M Figure 8-1a
Analog television signal Analog television spectrum

• Digital data transmission (PAM, ASK, PSK, FSK and QAM)

is more complex but (perhaps) offers higher performance
with control of accuracy and easier storage, simpler signal
processing for noise reduction, error detection and
correction and encryption.
S&M Figure 8-1b

• Digital data transmission requires analog-to-digital (ADC)

and digital-to-analog (DAC) converters. The ADC process
utilizes sampling and quantization of the continuous analog
signal.
ADC
DAC
S&M Figure 8-1b

• ADC sampling occurs at a uniform rate (the sampling rate)

and has a continuous amplitude.
S&M Figure 8-2a,b
Analog signal
Uniform
sampling Continuous
rate amplitude
• The continuous amplitude sample is then quantized to n

bits or resolution for the full scale input or 2n levels.
Uniform Continuous
sampling amplitude
rate
Quantized
Quantized
amplitude
S&M Figure 8-2b,c

• Here n = 4 and there are 24 = 16 levels for a full scale input

of 2 V (± 1 V). The step size = 2 V / 16 = 0.125 V and the
quantized value
is the midpoint
of the voltage
range.
S&M Table 8.1

Chapter 8
Digital-to-Analog
Conversion
• Sampling Baseband
Analog Signals
• Pages 392-399
• The analog signal x(t) which is continuously, uniformly

sampled is represented by:
∞
x s (t) = x(t) ∑ δ(t − kTS ) S&M Eq. 8.1
k = −∞
Multiplication in the temporal domain is convolution in the

frequency domain and the frequency domain representation
is:
 ∞ 
F
Xs (f) = X(f) ∗  ∑ δ(t − k TS )
k = −∞ 
 ∞

Xs (f) = X(f) ∗ fS

∑
k = −∞
δ(f − k fS )

S&M Eq. 8.2
∞
Xs (f) = fS ∑
k = −∞
X(f − k fS )
• Temporal and spectral representation of the continuous

sampling process for a sum of three sinusoids.
∑
∞
δ(t − kTS ) ∑ δ(f − k f
k = −∞
S )
k = −∞
∞
∞
x s (t) = x(t) ∑ δ(t − kTS )
Xs (f) = fS ∑
k = −∞
X(f − k fS )
k = −∞
S&M Figure 8-3

• 2 V, 20° initial
phase, 500 Hz
sinusoid
sampled at
5 k samples/sec
S&M Figure 8-4a,b

• Aliased samples
can be
reconstructed
for a 4500 Hz
and a 5500 Hz
sinusoid that
appears to be
a 500 Hz
sinusoid
S&M
Figure 8-4a,c,d
• The aliasing of
the signal can
be predicted by
the magnitude
spectrum of the
original 500 Hz
sampled signal.
If the 4500 Hz
and 5500 Hz
signals are then
sampled at
S&M Figure 8-4a,b
5 k samples/sec
aliasing at occurs at | 4500 – 5000 | and (5500 – 5000) Hz
• The sum of
three sinusoids
does not have
any aliased
frequencies
since the
S&M Figure 8-4a,c
sampling
frequency fS
is greater than
twice the
highest
frequency fmax
fS > 2 fmax
S&M Figure 8-5
• The frequency
2 fmax is called
the Nyquist
frequency.
Harry Nyquist,
who contributed S&M Figure 8-4a
to the understanding of thermal noise
while at Bell Labs, is also remembered
in electrotechnology for his analysis of
sampled data signals.
Harry Nyquist
1889-1976
• The analog signal is reconstructed from the quantized

samples by a DAC and a low pass filter (LFP).
S&M Figure 8-6

• For practical signals fS > 2 fmax using a guard band for LPFs
fS = 2 fmax
guard band
fS > 2 fmax
S&M Figure 8-7

• With out-of-band noise and sample signals, aliases of the

noise now appear in-band and should be filtered before the
sampling process.
S&M Figure 8-8
Chapter 8
Digital-to-Analog
Conversion
• Sampling Baseband
Analog Signals
• Pages 149-182
• The periodic baseband

signal consisting of
three sinusoids is
impulse sampled,
sampled-and-held,
processed by an
8-bit ADC-DAC
and a quantizer
in Simulink.
MS Figure 4.1
• The periodic baseband signal is the sum of a 1 V 500 Hz, a

0.5 V 1.5 kHz and a 0.2 V 2.5 kHz sinusoid.
MS Figure 4.2
• The power spectral density (PSD) of the periodic baseband

signal has the expected peaks at 0.5, 1.5 and 2.5 kHz.
Three sinusoids MS Figure 4.3

• The periodic baseband signal is overlaid with the continuous

amplitude sample-and-hold signal with fS = 8 kHz.
MS Figure 4.4
0.125 msec
• The analog signal x(t) here is sampled and held rather than
impulse sampled:
y s-h (t) = ∑ x(nTS ) h(t − nTS ) h(t) = 1 0 ≤ t ≤ TS
n
h(t) = 0 otherwise MS Eq. 4.3

The power spectral density (PSDs-h) of the sample and hold
operation is:
∞
PSDs-h = fS2 ∑
k = −∞
| X(f − k fS ) | 2 TS2 sinc 2 ( 2π f TS )
∞
PSDs-h = ∑
k = −∞
| X(f − k fS ) | 2 sinc 2 ( 2π f TS )
MS Eq. 4.4
• However, if the analog signal x(t) is impulse sampled:
x(nTS ) = ∑ x(t) δ(t − nTS ) MS Eq. 4.1

n
Then the power spectral density (PSD) does not have a

sinc2 term:
∞
PSD = fS2 ∑
k = −∞
| X(f − k fS ) | 2 MS Eq. 4.2
The PSDs-h does have the sinc2 term:

∞
PSDs-h = ∑
k = −∞
| X(f − k fS ) | 2 sinc 2 ( 2π f TS )
MS Eq. 4.4
• The PSD of the impulse sampled sum of three sinusoid

signal with fS = 8 kHz is:
8 kHz 16 kHz MS Figure 4.5
No sinc2 term
∞
PSD = fS2 ∑
k = −∞
| X(f − k fS ) | 2 MS Eq. 4.2
• The PSD of the continuous amplitude sample and hold

sum of three sinusoid signal with fS = 8 kHz is:
8 kHz 16 kHz MS Figure 4.6
sinc2 term
∞
PSDs-h = fS2 ∑
k = −∞
| X(f − k fS ) | 2 TS2 sinc 2 ( 2π f TS )
∞ MS Eq. 4.4
PSDs-h = ∑
k = −∞
| X(f − k fS ) | 2 sinc 2 ( 2π f TS )
Chapter 8
Digital-to-Analog
Conversion
• Sampling Bandpass
Analog Signals
• Pages 399-400
• A bandpass signal does not need to be sampled at 2 f2.

Nyquist’s bandpass sampling theory states that the
sampling rate fS > 2(f2 − f1) which is substantially less than
2 f2
S&M Figure 8-9
8 10 kHz
f1 f2
LPF 10 kHz fS = 20 ksamples/sec
BPF 8-10 kHz fS = 7 ksamples/sec

Chapter 8
Digital-to-Analog
Conversion
• Sampling Bandpass
Analog Signals
• Pages 180-181
• The Simulink simulation uses the DSB AM modulation

block and the sum of three sinusoids source.
MS Figure 4-32
MS Figure 4-33
• The Simulink simulation initially uses a sampling rate of

5 MHz and results in 4 194 304 = 222 sampling points. The
PSD shows the DSB-LC AM signal with the LSB and USB.
fC
LSB USB
Scaled PSD fmax = 50 kHz MS Figure 4-33

• The bandwidth of the bandpass signal is f2 − f1 = 22.5 −

17.5 = 5 kHz and the Simulink sampling rate is set to
50 kHz and results in only 32 768 = 215 sampling points.
fC
LSB USB
Aliased frequency range > 25 kHz

Scaled PSD fmax = 50 kHz MS Figure 4-34
Chapter 8
Digital-to-Analog
Conversion
• Quantizing Process:
Uniform Quantization
• Pages 400-404
• The quantizing process divides the range (± full scale) into

2n (n = 4 here) regions which are assigned an n-bit binary
code.
S&M Figure 8-10

• The error associated with the quantizing process is

assumed to have a uniform probability density function.
The maximum error for uniform quantization is:
 2 Vmax  Vmax
q = ± 0.5  n  = ± 2n
 2 
The quantizer range is ± Vmax
and the uniform quantizer
voltage step size is:
2 Vmax Vmax
∆= = n-1 MS Eq. 4.6
2n
2 S&M Figure 8-11
The mean square quantizing Eq is the normalized
quantizing noise power:
∆/2 2 2
1 ∆2 Vmax Vmax
Eq = ∫ q dq = = =
2
MS Eq. 4.7
∆ −∆ / 2 12 3 2 n
2
( ) ( )
3 22n
• The signal to quantizing noise power (SNRq) is:

12 PS PS
SNRq =
∆2
= 3( )
2 2n
2
Vmax
MS Eq. 4.8
PS is the normalized power of the signal that is quantized.
For the ADC here ∆ = 10 mV and n = 8. The sum of three

sinusoids as the input signal has a peak amplitude of 1.1 V
and the quantizing noise has a peak amplitude of 10 mV.
10 mV
MS Figure 4.7
Chapter 8
Digital-to-Analog
Conversion
• Quantizing Process:
Nonuniform Quantization
• Pages 400-404
• Nonuniform quantization divides the dynamic range of an

analog signal into nonuniform quantization regions. Lower
magnitudes have smaller quantization regions than high
magnitudes.
The quantization of speech benefits from nonuniform

quantization since the perception of hearing is logarithmical
rather than linear.
• Uniform Uniform quantization

quantization
(top) results in
a large amount
of error for
small sample
amplitude.
Nonuniform quantization
Non-uniform
quantization
(bottom)
reduces the
error for small
sample
amplitudes. S&M Figure 8-13
• Uniform quantization is simpler to implement so a

compressor (a non-linear transfer function) is used before
the quantizer.  Vin 
The µ-Law ln  1+µ 
 Vmax  Vin
compressor Vout = Vmax 0 ≤ ≤1
is used in ln (1+µ) Vmax
telephony
with MS Eq. 4.9
µ = 255. At the receiver
an expander has the
inverse non-linear
transfer function and
results in companding
(COMpressing and
exPANDING).
Chapter 8
Digital-to-Analog
Conversion
• Companding
• Pages 157-159
• The µ-Law compander concept can be simulated in

Simulink with the µ-Law Compressor and µ-Law Expander
blocks. The A-Law Compressor and A-Law Expander
blocks are included for comparison.
MS Figure 4.13
• The µ-Law compressor voltage transfer function is

sigmoidal (S-shaped).
MS Figure 4.14
Chapter 8
Digital-to-Analog
Conversion
• Pulse Code Modulation
• Pages 171-175
• The pulse code modulator (PCM) transmitter utilizes a

Simulink µ-Law compressor block, an 8-bit ADC subsystem,
an 8-bit DAC subsystem and a µ-Law expander block.
MS Figure 4.21
• The Simulink 8-bit ADC subsystem has a sample-and-hold

block controlled by a sampling pulse generator, an 8-bit
encoder block, an integer-to-bit converter block which
provides an 8-bit vector to a demultiplexer block and a
multiport switch. An 8-level staircase subsystem
sequences the multiport switch to select 1 of the 8 inputs for
bit serial output.
MS Figure 4.22
• The 8-level staircase Simulink subsystem sequences the

multiport switch with a 3-bit counter and a 3-bit DAC for the
output.
3-bit counter
3-bit DAC
MS Figure 4.22
• The 8-bit DAC Simulink subsystem for the PCM system

uses a 8-bit shift register and an 8-bit DAC.
8-bit shift
register
8-bit DAC
MS Figure 4.24
• The 8-bit DAC Simulink subsystem for the PCM system

uses a 8-bit shift register and an 8-bit DAC.
8-bit shift
register
8-bit DAC
MS Figure 4.24
• Analog input signal to the PCM system
MS Figure 4.25
MS Figure 4.21
• 8-bit DAC output after 8-bit ADC and µ-Law Compressor
MS Figure 4.25
MS Figure 4.21
• µ-Law Expander block output of the PCM system
MS Figure 4.25
MS Figure 4.21
• LPF output of the PCM system
MS Figure 4.25
MS Figure 4.21
• Analog input signal to the PCM system
• LPF output of the PCM system MS Figure 4.25
startup
Chapter 8
Digital-to-Analog
Conversion
• Differential Pulse Code
Modulation
• Pages 407-411
• Sampled speech data are highly correlated and differential

pulse code modulation (DPCM) exploits this to lower the
overall data
rate.
DPCM
uses a
predictor
to subtract
a predicted S&M Figure 8-15
value from
the input.
The error
difference
is sent.
• The predictor is a recursive equation, for example:

S(n) = 0.75 s(n−1) + 0.2 s(n−2) +0.05 s(n−3)
where S(n)
is the
predicted
value of the
n th sample
and s(n-i) S&M Figure 8-15
is the n-i th
sample.
The error
signal is
s(n) − S(n)
• A typical continuous analog signal is sampled and results in

a discrete signal s(n), The discrete predicted signal S(n) is
recursively computed. The discrete error signal is
transmitted and has less quantizing bits than the actual
discrete signal.
S&M Figure 8-16b

• A DPCM example of actual discrete values, predicted

values and the error terms:
S&M Table 8-3

Chapter 8
Digital-to-Analog
Conversion
• Differential Pulse Code
Modulation
• Pages 175-180
• A 4-bit first order differential pulse code modulator (DPCM)

can be simulated in Simulink.
MS Figure 4.26
• The first order linear predictor MetaSystem determines the

error signal: e(n) = s(n+1) − 2 s(n) + s(n-1)
error signal
input
MS Figure 4.27
ADC conversion command
• The Simulink 4-bit ADC subsystem of the DPCM system is

similar to the 8-bit ADC of the PCM system and illustrates
design reuse.
MS Figure 4.28
• The Simulink 4-bit DAC subsystem of the DPCM system is

also similar to the 8-bit DAC of the PCM system and
again illustrates design reuse.
4-bit shift
register
MS Figure 4.29 4-bit DAC

• The first order linear predictor Simulink subsystem

reconstructs an estimate of the signal se(n) from the error
signal e(n) received and past estimates:
se(n+1) = e(n+1) + 2 se(n) − se(n−1)
input
reconstructed
signal
MS Figure 4.30
• Analog input signal of the DPCM system
MS Figure 4.31
MS Figure 4.31
• Output of the 4-bit first order DPCM system
MS Figure 4.31
MS Figure 4.31
• Analog input signal to the DPCM system
• Output of the 4-bit first order DPCM system MS Figure 4.31
startup
• Output of the 8-bit PCM system MS Figure 4.25
startup
• Output of the 4-bit first order DPCM system MS Figure 4.31
startup
Chapter 8
Digital-to-Analog
Conversion
• Delta Modulation
• Pages 411-415
• Delta modulation is an extreme example of DPCM using

1-bit data representing ± ∆:
S(n) = S(n−1) + ∆ bi = 1 if S(n−1) ≤ s(n−1)
S(n) = S(n−1) − ∆ bi = 0 if S(n−1) > s(n−1)
S&M Eq. 8.10
DM transmitter
DM receiver
S&M Figure 8-18

• The reconstructed signal increments ± ∆ on each

transmitted bit.
bi = 1 S(n) = S(n−1) + ∆ bi = 0 S(n) = S(n−1) − ∆
4 1s 4 0s S&M Figure 8-19

Chapter 8
Digital-to-Analog
Conversion
• Delta Modulation
• Pages 72-75
• Delta modulation (DM) can be simulated in Simulink. The

DM receiver utilizes a sample and hold token as an
accumulator and the step size ∆ = 20 mV.
fS = 2 kHz
TS = 0.5 msec DM transmitter DM receiver
f = 2 Hz
A=1V MS Figure 2.61
• DM can be subject to slope overload which occurs when:

∆ / TS < max | d m(t) / dt | SVU Eq. 2.61 modified
Here the sinusoid has A = 1 but f = 10 Hz and:
∆ / TS = 20 mV / 0.5 msec = 40 < max | d m(t) / dt | = 80π
and slope overload occurs.
sinusoidal MS Figure 2.63

input signal
• Granular noise occurs in DM because if the input m(t) is

constant the received signal oscillates by ± ∆ because there
is no 0 possible. Clocking occurs at the DM symbol interval
TS = 0.5 msec.
± ∆ = ± 20 mV
∆ = 20 mV MS Figure 2.64
TS = 0.5 msec
• The tradeoff between slope overload and granular noise is

that a large value of ∆ (to avoid slope overload) would
increase granular noise. A decrease in Ts (again to avoid
slope overload) would increase the data rate rS.
• The step size ∆ = 20 mV and TS = 0.5 msec (rS = 2 kb/sec)

here.
• For a 10 Hz sinusoidal input signal to a DM:

m(t) = sin (2π 10t)
max | d m(t) / dt | = 20π
• If step size ∆ = 20 mV and TS = 0.5 msec then

∆ / TS = 40 < max | d m(t) / dt | = 20π so slope overload is
predicted to occur.
• For the 10 Hz sinusoidal input signal to a DM

m(t) = sin (2π 10t)
max | d m(t) / dt | = 20π
and the step size remains ∆ = 20 mV but TS = 0.25 msec

then ∆ / TS = 80 > max | d m(t) / dt | and slope overload is
mitigated but rS = 4 kb/sec.
• In comparison, an 8-bit PCM system sampling a 10 Hz

sinusoid at a reasonable sampling rate of 500 Hz (50
sampling points/period) has rb = 8(500) = 4 kb/sec or rb = rS
but PCM is more complicated than DM.
End of Chapter 8
Digital-to-Analog
Conversion

Eetop - CN - Digital Communication Systems Using MATLAB and Simulink

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Eetop - CN - Digital Communication Systems Using MATLAB and Simulink

Uploaded by

Copyright:

Available Formats

Contents

Embedded Design Using Programmable Gate Arrays

Digital Communication Systems Using MATLAB and Simulink

Digital Communication Systems Using SystemVue (Hardcover)

MATLAB/Simulink for Digital Communication (Paperback)

(Author), Kyung W. Park (Author), Kyung S. Woo (Author)

• In-class problems and discussions form the basis for the

• SNAP quizzes (weekly, 15 minutes, 25%) and

• Blackboard is used for the

• MATLAB and Simulink modeling of communication

• MATLAB and Simulink modeling in EE4513 Analog and

Temple Technical Institute

• This course is the application of electrotechnology

• Professional career path in digital data communications

• Course sequence continues in EE4542

• Read the assigned text

• SNAP stands for Students’ Note-taking Achieves

• Your organized notes are a compilation of reading the

• The SNAP concept relates to the learning process

I hear and I forget,

• I hear and I forget, ← Listening to the lecture

The notes are to be

Leonardo Da Vinci and

• The 15 minute SNAP quizzes are

• The SNAP quiz is the assessment

• Modulation (partial definition):

• Carrier (partial definition):

Source Simulink subsystems

Analog amplitude modulation (AM)

Analog amplitude modulation (AM)

Source Simulink subsystems

Analog amplitude modulation (AM)

Sinusoidal input and DSB AM

Sinusoidal input and DSB AM

DSB AM temporal signal and power spectrum

Demodulated DSB AM spectrum

Random data source Bit error rate

Digital binary phase shift keying (BPSK)

• Input binary data, rate rb = 1 kb/sec, bit time Tb = 1 msec

• Output binary data with no noise, note 2 bit delay in output

• Input binary data, rate rb = 1 kb/sec, bit time Tb = 1 msec

• Received signal with significant additive noise

Where’s the signal?

• Input binary data, rate rb = 1 kb/sec, bit time Tb = 1 msec

• Output binary data with bit errors due to additive noise

• Simulink Library Browser and the model window

• Source parameter block for the Random Integer

• Configuration parameter window for the Simulink

• Sinusoid and half wave rectified sinusoid Simulink

• Sinusoid and half wave rectified sinusoid temporal display

• Half wave rectified sinusoid Simulink spectral display

• Scope and axis properties parameter windows of the

• Scaled power spectral display of a half wave rectified

• Gaussian noise source and lowpass filtered power

• Gaussian noise source and LPF PSD MS Figure 1.12

LPF PSD, fcutoff = 11.25 kHz

• Gaussian noise source and lowpass filtered

• Gaussian noise source and lowpass filtered power

Why frequency domain

• Allows simple algebra

• Transfer functions can 500, 1500 and 2500 Hz

Why frequency domain

MS Figure 4.2 and Figure 4.3