Digital Communications Lecture Guide Contents: CME 624 May 2016

Lecture Guide Contents
Digital Communications
Department of Communications Engineering
CME 624
May 2016
x (t )
Sampler
x (n)
Quantizer
x(n)
fs 2 B
xk
L 2k
x k
BPSK
QAM, QPSK
OQPSK
MSK
CPFSK - optimal detection
QPR
M-ary PSK
APK
Low
Module 2: Source Encoding & Decoding

Elements of Digital Communication System
Formatting of Analog Information
Sampling, Quantization and Coding
Compounding and Encoding
Speech & Image Coding Techniques
Line Coding Techniques & Pulse Shaping
Inter Symbol Interference (ISI)
High
Complexity
DQPSK
DPSK
CPFSK -discriminator detection
FSK - noncoherent detection
OOK - envelope detection
Federal University of Technology, Minna
Lecture Guide Contents
Equalization
Module 3: Baseband Communication

Digital Baseband Communication Systems
Digital Transmission & Reception
Techniques
Noise in Communication Systems
Detection of Binary Signal in Gaussian
Noise
Optimum Receivers: Maximum Likelihood
Receiver, Matched Filtering, Correlation
Receiver
Correlator
Matched Filter
Coherent & Noncoherent Detection
Probability of Error for Binary Antipodal
Systems
Prof. Okey Ugweje
Digital Communication System
Module 4: Bandpass Communication

Modulation and Demodulation
Why Modulate?, Modulation categories
Basic Binary Modulation Schemes: BPSK,
BFSK, BPSK
Others Modulation Schemes: DPSK,
QPSK, OQPSK, M_ary Signaling
Comparisons of Digital Modulation
Schemes
Detection of Binary Signals
Error Performance (Bit and Symbol Error)
Module 6: Spread Spectrum

What is Spread Spectrum?/Significance of
Spreading
Basic Characteristics of SS System
Classifications of Spread Spectrum
Direct Sequence Spread Spectrum
Summary of Direct Sequence Techniques
Frequency Hopped Spread Spectrum
Direct Sequence vs. Frequency Hopping
Module 1
Introduction and Overview
Module 5: Multiplexing and Multiple Access

Multiplexing techniques
Frequency-Division Multiplexing
Time-Division Multiplexing
Code-Division Multiplexing
Multiple Access
Frequency Division Multiple Access
Time Division Multiple Access
Code Division Multiple Access
Prof. Okey Ugweje
Controling ISI
Lecture Guide
Prof. Okechukwu C. Ugweje

Prof. Okey Ugweje
Module 1: Introduction and Overview

Course Introduction
Review of linear systems
Review of Random Variables
Review of Random Processes:
Autocorrelation, Cross-correlation, Power
spectral density, Energy Spectral Density
Overview of digital communication systems
Why digital communication?, Goals in
communication system design, Digital
signal nomenclature
Review of Linear Systems (Signals and Systems)

Review of Probability and Random Signals
Prof. Okey Ugweje

Note:
Some of the material contained in Module 1 is a review
of prerequisite materials covered in undergraduate
classes such as:
Signals and Systems
Communications and Signal Processing
Random Signals and Processes
Introduction and Handout
Some of the materials are included in this section for

your benefit
It is your responsibility to review most of the material in
this Module
Most materials in this section can be found in Chapter
1 and the Appendix of the recommended textbook
Introductions
Course Outline/Syllabus
Course Calendar
Course Overview
Prof. Okey Ugweje
Signals are used to convey information

Signals and waveforms (voltage, current and intensity)
are central to communication and signal processing
Signals can be viewed either in time or frequency
domain
A signal is any physical quantity that varies with time,
space, or any other independent variables
Often, the independent variables for most signals is
time
Theoretical signals can be described mathematically,
graphically or in tabular form
Real signals are however difficult to describe, and more
often can be described approximately
Signals and Systems

Signals and Systems
Continuous Convolution
Parsevals theorem
Linear Transform
Fourier Transform Techniques
Concept of Bandwidth/ Filtering
Signals - 1
Prof. Okey Ugweje
Prof. Okey Ugweje
Prof. Okey Ugweje
Signals - 2
Signals - 3
Mathematically, a signal is defined as a function of one

or more independent variables, e.g.,
Mathematically, a signal is defined as a function of one or

more independent variables, e.g.,
x(t) = 10t
x(t) = 5t2
s(x,y) = 3x + 2xy + 10y2
x(t) = 10t
x(t) = 5t2
s(x,y) = 3x + 2xy + 10y2
Sometimes the functional dependence on the independent

variable is not precisely known, e.g., speech signal
Sometimes a signal is a combination of other signals
Sometimes the functional dependence on the

independent variable is not precisely known, e.g.,
speech signal
Sometimes a signal is a combination of other signals
e.g., sum of sinusoid of different amplitudes, frequency & phase
s (t ) Ai (t ) sin 2 Fi (t ) i (t )
n
e.g., sum of sinusoid of different amplitudes,

frequency & phase
i 1
Signals are the inputs outputs, and internal functions that

the systems process or produce, such as voltage,
current, pressure, displacements, intensity, etc.
s (t ) Ai (t ) sin 2 Fi (t ) i (t )
n
i 1
Prof. Okey Ugweje
Prof. Okey Ugweje
Signals - 4
Signals - 5
The variable time may be continuous or discrete and the

value of the signal may be represented as
Physical realizable signals must

Have time duration
Occupy finite frequency spectrum
Are continuous (as in analog signal)
Have finite peak value, and
Are real-valued
Continuous-valued x(t)
Discrete-valued x(nts)
Quantized xQ(t), and
Digital x[n]
All real-world signals will have these properties

Sometimes we use mathematical signal models which violate
these conditions
e.g., Dirac delta function (or impulse function)
The most commonly used analog signals are the sinusoidal
signals (sine, cosine, etc.)
In communication systems, we are concerned with info
bearing signals that evolve as a function of the independent
variable, t
These types of signals occur at different stages of the

process
Other variables (distance, angle, etc.) can also be the

independent variable, especially for 2-D signals like
images and video
Prof. Okey Ugweje
10
11
Prof. Okey Ugweje
12
Systems - 1
Systems - 2
When signals are corrupted by noise, they no longer convey

the required information directly, hence they often require
processing
Radio receivers are especially sensitive to noise
Signals are processed by systems, which may modify them
or extract additional information from them
Thus, a system is an entity that processes a set of signals
(inputs) to yield another set of signals (outputs)
A system can also be associated to the signal as in the
source or sink of the signal
A system may be made up of physical components
(hardware realization), as in electrical, mechanical, or
hydraulic systems, or it may be an algorithm (software
realization) that computes an output from an input signal
Many systems have signals that are not wanted (commonly

known as noise or interference)
A system is a device, process, or algorithm that, given an
input x(t), produces an output y(t)
A system is characterized by its input (excitation or forcing
function), its output (response), and the rules of operation
(internal functions)
From a communication engineers viewpoint, a system is a law
that assigns output signals to various input signals
Systems may be realized as an integration of sub-systems or
as a single entity
In practice, systems with feedback is of great importance
Prof. Okey Ugweje
Prof. Okey Ugweje
13
Systems - 3
Systems - 4
Systems may be classified functionally as in
To understand the behavior of systems

(electronic/mechanical), the response to inputs
(usually signals) must be understood
Analyzers, Synthesizers, Transducers, Channels,

Filters, and Equalizers, etc.
or descriptively as in
Terminology of Systems
State:
linear, nonlinear, causal, discrete, continues, time

invariant, etc.
Variables that allow us to determine the energy level

of the system
All physical systems are referenced to zero-energy
state, e.g., ground state, rest state, relaxed state
Examples of Systems
Electronic systems: resistors, inductors, Radio/TV,
phone networks, sonar and radar, guidance &
navigation, satellite, lab instrumentation, biomedical
instrumentation, etc.
Mechanical systems: loudspeakers, microphones,
vibration analyzers, springs, dampers
Prof. Okey Ugweje
14
Initial Conditions
The initial conditions or initial state is the state of the
system before an input is applied
15
Prof. Okey Ugweje
16
Systems - 5
Systems - 6
Broad Classification of Systems
Operation on Linear Systems

An operator, T, is a rule to transform one function to another
Additive
SYSTEMS
Distributed
Parameters
We are
interested only
on the systems
that intersect the
dotted path.
Lumped Parameters
Stochastic
Deterministic
Linear
Time
Varying
Nonlinear
Time
Invariant
Prof. Okey Ugweje
Principle of Superposition
Superposition implies both additive & homogeneous rules
k p
k p
If a system fails either rule, the function is nonlinear

Addition or homogeneity is sufficient condition to test for
linearity
Time
Invariant
17
Prof. Okey Ugweje
18
Why study signals and systems?
Linear Time-Invariant (LTI) Systems

In signals and systems theory we study the definition

and description of signals, and the behavior of systems
under different conditions
Signals form the inputs, outputs and internal
functions of systems
In electrical & computer engineering, the understanding
of signals and the behavior of systems is of immense
importance
Communication engineers are concerned with systems
which transmit, receive, and process signals carrying
information
Hence before one can characterize a system, one must
be able to characterize the system
LTI
x(t)
h(t)
y(t)
x[n]
h[n]
y[n]
x(ejw)
H(ejw)
Y(ejw)
X(f)
H(f)
Y(f)
H(z)
Y(z)
Time Function
Frequency Function
Difference Equation
Pole-Zero Plot
H - Function
Linear systems are characterized by the ability to accept

input and produce output in response to the input
Most communication systems can be modeled as linear
systems with signals forming the input and output functions
Prof. Okey Ugweje
T Ax1(t ) Bx2 (t ) AT x1(t ) BT x2 (t )
X(z)
T x(t ) y(t )
T Kx(t ) KT x(t )
Linear
Time
Varying
p k p k p
Homogeneous
Discrete Time
Continuous Time
Nonlinear
T x1(t ) x2 (t ) T x1(t ) T x2 (t )
19
Prof. Okey Ugweje
20
Size of a Signal - 1
The size of a signal is the value of the strength of the

signal
The signal strength may be measures in its entirety
or in a given interval
Such a measure must consider not only the signal
amplitude, but also its duration
There are two major ways of determining the signal
strength
1. Signal Energy
Prof. Okey Ugweje
Prof. Okey Ugweje
21
A signal is classified as energy-type if its energy Eg is

finite (0<Eg<)
Energy may be computed in either time or frequency
domain, whichever is easier using the following
formula
Eg g (t ) dt
T /2
lim
T T / 2
g (t ) dt G ( f ) df
(unit)2s
where G(f) is the Fourier transform of g(t)

All time-limited signals of finite amplitude are energy
signals
Energy signals have zero power
Since signal energy also depends on the load the actual
signal energy should be normalized by the load R
22
Signals and Spectra - 1
2. Signal Power
Important Signal Classifications

Deterministic and Random Signals
Value of the signal is known or not known at all
times
Periodic and Non-periodic Signals
A signal is power-type if its power Pg is finite (0<Pg<)

The power Pg of a signal can be computed using the
formula
2
Pg Tlim 21T TT g (t ) dt Tlim T1 TT/ 2/ 2 g (t ) dt (unit)2
x(t ) x(t T0 ), t
Notice that the signal power is the time-average

(mean) of the signal amplitude squared
Most periodic signals are power-type signals
For periodic signals Eg & Pg can be computed by
integrating over one period
Prof. Okey Ugweje
Analog (Continuous-Time) and Discrete Signals

Exists for all times t vs. exists at discrete time
only
23
Prof. Okey Ugweje
24
Energy- and Power-Type Signals
ExT .5T x 2 (t )dt

.5T
PxT
1
T
ExT
1
T
Others
Even and Odd Signals
Real and Complex Signals
Causal and Noncausal
.5T
2
.5T x (t )dt
with waveform
EX
Px
.5T
lim .5T
1
T T
lim
x 2 (t )dt x 2 (t )dt
1
2
2
.5T x (t )dt T x (t )dt
.5T
Unit Impulse Function
(t )dt 1,
(t ) 0 for t 0
x(t ) ( to )d x(t0 )
Prof. Okey Ugweje
25
Prof. Okey Ugweje
Examples
Spectral Density
1. Example 1
Signal Power
Energy Spectral Density
26
E X x 2 (t )dt x( f ) df Parseval ' s Theorem
X ( f )
2. Example 2
Signal Energy
X ( f )df ( f ) is defined as energy spectral density
2 0 X ( f )df
Power Spectral Density
3. Example 3
Signal Energy
1
2
Cn power
x (t )dt n
T
For periodic signals, the PSD is given by
PX
T
2
T
2
GX ( f )
Prof. Okey Ugweje
Cn f nf 0
27
Prof. Okey Ugweje
28
Some Important or Common Signals & Functions

Sinusoidal Signal
Complex Exponential (harmonics)
Unit Step Function [denoted by u(t)]
Ramp Function [denoted by r(t)]
Rectangular Pulse Function [denoted by rect(t) or
(t)]
Triangular Pulse Function[denoted by (t)]
Sign (Signum) Function [denoted by sgn(t)]
Sinc Function [denoted by sinc(t)]
Impulse (Delta, Dirac) Function [denoted by (t)]
Operations on Signals
Amplitude Scaling
Amplitude Shifting
Time Shifting
Displaces a signal in time without changing its
shape
Prof. Okey Ugweje
y (t ) x(t )
"+"shifts the signal left by
"-" shifts the signal right by (delayed)
29
Prof. Okey Ugweje
30
Time Scaling
Some useful signal operations and models
Slows down or speeds up time which results in signal

compression or stretching
The expression
t
y (t ) x

Continuous/Discrete Convolution
Parsevals theorem
Hilbert Transform
Concept of Bandwidth and Filtering
Reflection or Folding
Some Important Properties of Signals
A scaling operation with = -1 x(t) = x(-t)

The mirror image of x(t) about the y-axis through t = 0
DC Value
Is the time average of a signal or the time average
over a finite interval [t1, t2]
Operations in Combinations
x(t) delay (shift right) by x(t-)
compress by x(t-)
x(t) compress by x(t)
delay (shift right) by / x(t-)
Prof. Okey Ugweje
Average Power
The ensemble average
RMS Value
31
Prof. Okey Ugweje
32
Examples

Fourier Series and Transform
4. Example 4
Periodicity of Signal
Definition and Properties

Important Fourier transform cases
Energy and power spectral density
g (t ) g t T0
5. Example 5
Different Types of Sampling Techniques

Idea Sampling
Natural Sampling
Sample-and-Hold
Even x(t) = x(-t)

Odd x(t) = -x(-t)
6. Example 6
Prof. Okey Ugweje
33
Prof. Okey Ugweje
Examples
34
Fourier Transform Table
7. Example 7 : Convolution
Convolution is a technique of finding the zero state
response of LTI system
x(t)
y (t ) x(t ) h(t )
h(t)
y(t)
x( )h(t )d x(t )h( )d
8. Example 8: Convolution
Prof. Okey Ugweje
35
Prof. Okey Ugweje
36
Examples
Fourier Transform Pair

9. Example 9: Fourier Transform

X ( f ) F x(t )
z x(t)e j2ft dt
10.Example 10: Fourier Transform

13.Example 13: Inverse Fourier Transform
x(t ) F 1 X ( f )
z X ( f )e j2ftdf
Prof. Okey Ugweje
37
Prof. Okey Ugweje
38
Examples Random Signals
14. Example 14
Random Signals
15. Example 15
Review of Probability and

Random Signals
Probability Theory
Distribution Functions
Density Functions
Expectations
Random Processes, etc
Prof. Okey Ugweje
Random Processes
Please review the course

CME621:Stochastic
Processes
39
Prof. Okey Ugweje
40

Module 2
Source Encoding & Decoding
Elements of Digital
Communication System
Elements of Digital Communication

Formatting of Analog Signal
Sampling and Quantization
Compounding
Encoding and Line Coding Techniques
Intersymbol interference
Prof. Okey Ugweje
41
Prof. Okey Ugweje
Elements of Digital Communication - 1
42
Elements of Digital Communication - 2
Digital
input
lm q
i
Format
Source
Encoder
Source
bits
Channel
bits
Channel
Encoder
From other
sources
Multiplex
Modulate
Spread
Multiple
Access
Tx
Pe
Performance
Measure
Bits or
Symbol
Waveforms
Rx
n s
m$i
Format
Source
Decoder
Digital
output
Channel
Decoder
Source
bits
Demultiplex
Channel
bits
Demodulate
&
Detect
To other
destinations
Despread
Multiple
Access
Carrier & symbol

synchronization
Each of these blocks represents one or more transformations

Each block identifies a major signal processing function which changes or
transforms the signal from one signal space to another
Some of the transformation block overlap in functions
Prof. Okey Ugweje
43
Prof. Okey Ugweje
44
Why Digital Communications? - 1
1. Advantages
Easier and more efficient to multiplex several digital

signals
Two-state signal representation

Hardware is more flexible
Hardware implementation is flexible and permits the use of
microprocessors, mini-processors, LSI or VLSI, etc.
Low cost
Easy to regenerate the distorted signal
Can use packet switching

Encryption and privacy techniques are easier to
implement
Better overall performance
Repeaters can detect a digital signal and retransmit a new,

clean (noise free) signal
Hence, prevent accumulation of noise along the path
Less subject to distortion and interference

Digital system is more immune to channel noise/ distortion
Prof. Okey Ugweje
Can combine different signal types data, voice,

TV, text, etc.
It is possible to combine both format for transmission
through a common medium
With LSI/VLSI, implementation cost is reduced
Digital multiplexing techniques TDMA and CDMA - are

easier to implement than analog techniques such as FDMA
45
Inherently more efficient than analog techniques in

realizing the exchange of SNR for bandwidth
Prof. Okey Ugweje
46
Goals in Communication System Design
2. Disadvantages
To maximize transmission rate, R, e.g., symbols per sec

To minimize bit error rate, Pe, or Pb
To minimize required power, Eb/No (or ~ly required signal
power)
To minimize required systems bandwidth, W
To maximize system utilization, U
To minimize system complexity, Cx
Requires reliable synchronization

Requires A/D conversions at high data rate
Requires larger bandwidth (require BW efficient
MODEM)
Banalog = W Hz
Bdigital = nW Hz
where n is the # of bits used to quantize the amplitude

of the signal
Generally an increase in complexity over analog

system
Prof. Okey Ugweje
In most practical
applications tradeoffs are necessary
47
Prof. Okey Ugweje
Pe
Cx Eb/No
48
Digital Signal Nomenclature - 1
Information Source
Discrete output values, e.g. Keyboard (1~26 (A~Z) symbols)
Symbol
Analog signal source information is continuous valued

Textual Message
A meaningful sequence of character or symbols, e.g.,
Digital Message
Messages constructed from a finite # of symbols (26 letters, 10
numbers, space and punctuation marks).
How are you? I am ok, thank you; I feel like a million dollars!
Character
Member of an alphanumeric/symbol (A ~ Z, 0 ~ 9)
Characters can be mapped into a sequence of binary digits
using one of the standardized codes such as
ASCII: American Standard Code for Information
Interchange
Others: EBCDIC, Hollerith, Baudot, Murray, Morse, etc.
Prof. Okey Ugweje
A digital message made up of groups of k-bits considered as a unit

A member of source alphabet. May or may not be binary, e.g. 2
symbol binary, 4 symbol PSK, 128 symbol ASCII
Hence a text is a digital message with about 50 symbols
Morse-coded telegraph message is a digital message

constructed from 2 symbols Mark and Space
M_ary
A digital message constructed with M symbols
Digital Waveform
Current or voltage waveform that represents a digital symbol
49
Prof. Okey Ugweje
Binary Digit (Bit)

Fundamental unit of info made up of 2 symbols (0 and 1)
Quantity of info carried by a symbol with probability P =
Bit: number with value 0 or 1
n bits: digital representation for 0, 1, , 2n
Byte or Octet, n = 8
Computer word, n = 16, 32, or 64
n bits allows enumeration of 2n possibilities
n-bit field in a header
n-bit representation of a voice sample
Message consisting of n bits
The number of bits required to represent a message is a measure
of its information content
More bits More content
Binary Stream (or bit stream or baseband signal)

A sequence of binary digits, e.g., 10011100101010
Block
Stream
Information that occurs in
Information that is
a single block
produced & transmitted
Text message
continuously
Prof. Okey Ugweje
Prof. Okey Ugweje
50
51
Data file
JPEG image
MPEG file
Real-time voice
Streaming video
Size = Bits / block

or bytes/block
Bit rate = bits / second
1 kbyte = 210 bytes

1 Mbyte = 220 bytes
1 Gbyte = 230 bytes
1 kbps = 103 bps

1 Mbps = 106 bps
1 Gbps =109 bps
52
Examples of Block Information
Transmission Delay
Type
Method
Format
Original
Compressed
(Ratio)
Text
Zip,
compress
ASCII
KbytesMbytes
(2-6)
Fax
CCITT
Group 3
A4 page
200x100
pixels/in2
256
kbytes
5-54 kbytes
(5-50)
JPEG
in2
38.4
Mbytes
Color
Image
Prof. Okey Ugweje
8x10
photo
2
400 pixels/in2
number of bits in message

speed of digital transmission system
time to transmit the information
time for signal to propagate across medium
distance in meters
speed of light (3x108 m/s in vacuum)
L
R bps
L/R
tprop
d
c
Delay = tprop + L/R = d/c + L/R seconds
1-8 Mbytes
(5-30)
Use data compression to reduce L

Use higher speed modem to increase R
Place server closer to reduce d
53
Prof. Okey Ugweje
54
Bit Rate of Digitized Signal

Bit Rate
Bandwidth Ws Hertz: how fast the signal changes

Higher bandwidth more frequent samples
Minimum sampling rate = 2 x Ws
Actual rate at which info is transmitted per second
Baud Rate
The rate at which bits are transmitted, i.e. # of signaling elements per
second
Representation accuracy: range of approximation error
Bit Error Rate
Higher accuracy
smaller spacing between approximation values
more bits per sample
The probability that one bit is in error, Pb, or simply the probability of
error, Pe
Data Rate
The rate at which info is transferred in bits per second
If binary symbols are independent & equiprobable, the bit rate = baud
rate
Character Rate
Characters transmitted per second
Prof. Okey Ugweje
55
Prof. Okey Ugweje
56
Stream Information
Sampling Rate and Bandwidth
A signal that varies faster needs to be sampled more

frequently
Bandwidth measures how fast a signal varies
A real-time voice signal must be digitized &

transmitted as it is produced
Analog signal level varies continuously in time
x1(t)
10 10 1 0 1 0
...
x2(t)
11 1 1 0 000
...
...
...
1 ms
1 ms
Th e s p ee
ch s i g n al l e v el
v a r ie s w i th
m(e)
Prof. Okey Ugweje
57
Bandwidth of General Signals
What is the bandwidth of a signal?

How is bandwidth related to sampling rate?
Prof. Okey Ugweje
speech
(noisy )
Analog
|p
(air stopped)
| ee
(periodic)
Not all signals are periodic

E.g. voice signals varies according X(f)
to sound
Vowels are periodic, s is noiselike
Spectrum of long-term signal
Averages over many sounds, many
speakers
Involves Fourier transform
0
Telephone speech: 4 kHz
CD Audio: 22 kHz
| t (stopped) | sh (noisy)
Digital
Older technology
Newer technology
Used to design mainly for voice
Used to design for data and voice
Inefficient for data
Efficient for data
Noisy and error prone
Noise can be easily filtered out
Lower speeds
Higher speeds
High overhead
Low overhead
Info is precise since recorded,

transmitted or displayed
continuously in time
Digital is accurate since info is displayed in

terms of values; but we don't know if the
precise value is displayed
Interpretation of display is harder
Interpretation of display is easier

More test options
Ws
Discrete-level information
Performance measured with SNR
Prof. Okey Ugweje
58
Analog vs. Digital Communications
59
Prof. Okey Ugweje
Performance measured with BER
60
Analog vs. Digital Transmission
Bandwidth Dilemma
Analog transmission: all details must be reproduced accurately
All bandwidth criteria have in common the attempt to

specify a measure of the width, W, of a nonnegative
real-valued spectral density defined for all frequencies
f <
Distortion
Attenuation Received
Sent
The single-sided power spectral density for a single

heterodyned pulse xc(t) takes the analytical form:
Digital transmission: only discrete levels need to be reproduced

Sent
Prof. Okey Ugweje
Distortion
Attenuation
Received
sin ( f f c )T
Gx ( f ) T
( f f c )T
Simple Receiver:
Was original pulse
positive or
negative?
61
Different Bandwidth Criteria
Prof. Okey Ugweje
(1.73)
62
Digital Communication Transformations
(a) Half-power
bandwidth.
(b) Equivalent
rectangular or noise
equivalent bandwidth.
(c) Null-to-null bandwidth.
(d) Fractional power
containment
bandwidth.
(e) Bounded power
spectral density.
(f) Absolute bandwidth.
Prof. Okey Ugweje
63
Prof. Okey Ugweje
64
Encoding and Decoding of Messages

(Baseband Systems)
Source bits
Formatting of Analog
Signal
lm q
i
Pe
Baseband Systems
Formatting Textual Data (messages, character, symbols)
Formatting Analog Information
Sampling (see prerequisite section)
Quantization
Line Coding
Prof. Okey Ugweje
65
Digital Communication Transformations - 1
Source
Encoder
Format
From other
sources
Channel bits
Channel
Encoder
Multiplex
Pulse
Modulation
Modulate
Spread
Multiple
Access
Digital
input
Performance
Measure
lm q
i
Bits or
Symbol
Waveforms
Digital
output
Format
Source
Decoder
Channel
Decoder
Source bits
Demultiplex
Demodulate &
Detect
Multiple
Access
Channel bits
To other
destinations
Prof. Okey Ugweje
Despread
Carrier and symbol

synchronization
66
Transmit and Receive Formatting

Transition from info source digital symbols info sink
Digital Information
Source
Textual
Information
Analog
Information
Analog
Information
Sink
Sampler
Quantizer
Waveform
Encoder
(Modulator)
Coder
Channel
Format
LPF
Decoder
Textual
Information
Transmitter
Waveform
Detector
Receiver
Digital Information
Prof. Okey Ugweje
67
Prof. Okey Ugweje
68
Character Coding (Textual Info)
Character coding, messages and symbols
A textual info is a sequence of alphanumeric characters

Characters are encoded into bits
Groups of k bits can be combined to form new digits or
symbols of size M
Textual
Message
Encoder
... 01101 ...
Group of k bits
M=2k
M_ary
Waveform
Encoder
(Modulator)
Alphanumeric and symbolic characters are encoded

into digital bits using one of several standard formats
ASCII
EBCDIC
Others Baudot, Hollerith, Morse
M 2k
A symbol set of size M is referred to as M-ary system
Prof. Okey Ugweje
69
Prof. Okey Ugweje
70
Example 16:
In ASCII alphabets, numbers, and symbols are encoded
using a 7-bit code
7-bit ASCII
Most significant
b8
b7
b6
b5
b4
b3
b2
Parity
b1
Least significant
A total of 27 = 128 different characters can be

represented using a 7-bit unique ASCII code
U
7-bit ASCII
16_ary digits
1
(symbols)
Prof. Okey Ugweje
71
Prof. Okey Ugweje
S
1
0
C
72
Digital Representation of Analog Signals

Most practical signal of interest are analog in nature

e.g., speech
biological signals
seismic signals
radar signals
sonar, and
various communication signals (audio, video, text, etc)
Sampling
Conversion to digital form is necessary

Analog
Signal
Prof. Okey Ugweje
Interface
(A/D)
Digital
Signal
73
Prof. Okey Ugweje
Digitization of Analog Signals
Transmitter Side Encoding
(Formatting Analog Information)
1. Sampling: obtain samples of x(t) at uniformly spaced

time intervals
2. Quantization: map each sample into an approximation
value of finite precision
Pulse Code Modulation: telephone speech
CD audio
3. Compression: to lower bit rate further, apply additional
compression method
Differential coding: cellular telephone speech
Subband coding: MP3 audio
Compression discussed in Chapter 12
Structure of Digital Communication Transmitter

Input
Signal
Sampling
Quantization
Digital
Modulation
Transmitted
Signal
Transmitter
Analog-to-Digital (A/D) Conversion

A/D Converter
Analog Input
Signal
xa(t)
Analog signal
Prof. Okey Ugweje
74
75
Prof. Okey Ugweje
x[n]
Sampler
Quantizer
Discrete-time
signal
xq(n)
Quantized
Output Signal
Quantized
signal
76
Sampling - 1
Sampling - 2
A/D conversion involves a 2 step process:
Illustration of sampling:
Sampling (Review 341 course notes)

Converts CT analog signal x(t) to DT continuous value signal
xs(t)
Continuous
Time Analog
Signal
Sampling
Discrete-time
continuous-valued
signal
Obtained by taking the samples of x(t) at DT intervals, Ts

xs(t) is discrete time signal (but still continuous valued)
Proper sampling must satisfy Nyquist theorem
Sampling does not introduce error or distortion
Quantization
Converts DT continuous valued signal to DT discrete valued
signal
Prof. Okey Ugweje
77
Prof. Okey Ugweje
Sampling - 3
Sampling - 4
Sampling Theorem (section 2.4.1)

Let the signal x(t) be bandlimited @ B (or fm), with
Fourier Transform (or spectrum) X(f)
x(t) can be perfectly reconstructed provided Rs
2B (fs 2fm)
2B is called the Nyquist Rate
If Rs < 2B, aliasing (overlapping of spectra) results
If signal is not strictly bandlimited, then it must be
passed through LPF before sampling
The first step in PCM is sampling.

The analog signal is sampled every Ts sec, where Ts is the
sample interval or period.
The inverse of the sampling interval is the sampling rate or
sampling frequency and denoted by fs, where fs = 1/Ts.
Prof. Okey Ugweje
Prof. Okey Ugweje
78
79
80
Sampling - 5
Sampling - 6
There are 3 sampling methods.

Ideal (or Impulse) Sampling
Natural Sampling
Sample-and-Hold
Practical Sampling
Flat-Top Sampling
Covered in 4400:341
Communications and
Signal Processing
In ideal sampling, pulses from the analog signal are sampled.

This method is ideal and cannot be easily implemented.
In natural sampling, a high-speed switch is turned on for only
the small period of time when the sampling occurs. The result is
a sequence of samples that retains the shape of the analog
signal.
The most common sampling method, called sample and hold,
however, creates flat-top samples by using a circuit.
Prof. Okey Ugweje
81
Prof. Okey Ugweje
Sampling - 7
82
Sampling - 8
Ideal Sampling (or Impulse Sampling)
For all sampling techniques

If fs > 2B then we recover x(t) exactly
If fs < 2B) spectral overlapping known as aliasing
will occur
x (t ) x(t ) x (t )
x(t ) (t nTs) x(nTs) (t nTs)

n
n
Natural Sampling (or Gating)

Note
xs (t ) x(t ) x p (t ) x(t ) cn e j2nf st

n
According to the Nyquist theorem, the

sampling rate must be
at least 2 times the highest frequency
contained in the signal.
Sample-and-Hold
x (t ) x '(t ) p(t )
s
x(t ) (t nT s) p(t )
n
Prof. Okey Ugweje
83
Prof. Okey Ugweje
84
Please Note
Examples
First, we can sample a signal only if the signal is band-limited. A

signal with an infinite bandwidth cannot be sampled.
Second, the sampling rate must be at least 2 times the highest
frequency, not the bandwidth.
If the analog signal is low-pass, the bandwidth and the highest
frequency are the same value.
If the analog signal is bandpass, the bandwidth value is lower than
the value of the maximum frequency
17.Example 17
Consider the analog signal x(t) given by
x (t ) 3cos 50 t 100sin 300 t cos 100 t
What is the Nyquist rate for this signal?
Can this signal be reconstructed at the receiver at
the Nyquist rate?
18.Examples 18
Sampling
19.Examples 19
Sampling
Prof. Okey Ugweje
85
Practical Sampling Rates
Prof. Okey Ugweje
86
Speech:
Telephone quality speech has a bandwidth of 4 kHz
Most digital telephone systems are sampled at 8000
samples/sec
Quantization & Pulse

Code Modulation
Audio:
The highest frequency the human ear can hear is
approximately 15 kHz
CD quality audio are sampled at rate of 44,000
samples/sec
Video:
The human eye requires samples at a rate of at least
20 frames/sec to achieve smooth motion
Prof. Okey Ugweje
87
Prof. Okey Ugweje
88
Quantization - 1
Quantization - 2
A/D Converter
Analog
Input
signal
x(t)
Sampler
x[n]
Quantizer
Discrete-time signal
Analog signal
xq(n)
Quantized
output signal
Quantized signal
Sample values require infinite # of bits for perfect

representation since sampler output still continuous in
amplitude
each sample can take on any value, e.g. 4.752, 0.001, etc
the number of possible values is infinite
To transmit as a digital signal we must restrict the # of

possible values to finite bits
Prof. Okey Ugweje
89
Quantization - 3
Prof. Okey Ugweje
90
Quantizer Model and Definitions - 1
e.g., suppose we must round to the nearest tenth,

then:
4.752 4.8
0.001 0
rounds off the sample values to the nearest
discrete value in a set of L quantum levels
quantized samples xq(n) are discrete in time (by
virtues of sampling) and discrete in amplitude (by
virtue of quantization)
Because we are approximating the analog sample
values by using finite # of levels, L, error is
introduced during quantization
Prof. Okey Ugweje
Definition:
Quantization is the process of approximating
continuous-valued samples with a finite number of
bits
Quantizer
device that operates on a discrete-time signal to
produce finite # of amplitudes by approximating the
sampled values
maps each sampled value to one of pre-assigned
output levels
the process of rounding off a sample according to
some rule
Quantizer
X[nT]
Xq(nT)
random process
Definition
number, size, location of its quantizing cell
boundaries, and step size of the quantization process
Quantization Resolution
# of bits, n, used to represent each sample
n log 2 L
where L = number of levels
more bits results in better fidelity

However, the bit rate is higher and more bandwidth is required
91
Prof. Okey Ugweje
92
Quantizer Model and Definitions - 2
Illustration and Description of Quantization - 1
Telephone systems typically use 8 bits of resolution

64 kbps
CD players use 16 bits of resolution/channel
705.6 kbps (mono)
Quantization error = difference of xs(t) and xq(nT)
Unlike sampling quantization is an irreversible
process
It results in signal distortion
Prof. Okey Ugweje
93
input
signal
LPF
Sampler
Quantizer
Encoder
Binary
codes
Operational Description
Process of approximating DT continuous valued samples
with a finite # of bits
the process of rounding off a sample according to some
rule maps each sampled value to one of pre-assigned
output levels, L
quantized samples xq(n) are discrete in time and discrete
in amplitude
the approximation introduces errors
Prof. Okey Ugweje
94
Range over which a quantizer will operate

Vmax, Vmin (Vp, -Vp)
Peak-to-peak voltage range

Vpp = Vp (-Vp) = 2Vp
Dynamic Range
Vmax
Vmin
Vmax
k
Vmax / L L 2
Dynamic Range depends on the

resolution of the converter
min detectable signal variation is
Vmax/L volts =
~ quantization step size, q
Prof. Okey Ugweje
95
Prof. Okey Ugweje
96
Mathematically
Xq ( kTs ) mi , if x k 1 x ( kTs ) x k
Xq (t ) Xq ( kTs ), if kTs t ( k 1)Ts
Sampled values are converted to one of L allowable

levels, m1, m2, , mL, according to some desired rule
Output is a sequence of levels, Xq(t)
Improvement can be achieved by careful selection of xi's
and mi's
Let X be a random variable representing a sample of data
Quantizer
+
x f ( x ) x e(t )
e(t ) x x
Prof. Okey Ugweje
97
Prof. Okey Ugweje
e(t ) x x
98
Then, the quantized value of X is given by
In Tabular Form In Concise Form

k xk 1
1
3
2
2
3
1
4
5
0
6
1
7
2
8
3
X f ( X )
If a quantizer has L quantization levels
X x1 , x 2 , x3 ,, x L
Then, with the endpoints, we have L+1 values
kx , x , x ,, x p,
0
This implies that

x k 1 x x k
Prof. Okey Ugweje
where x 0 , x L
xk
3
2
1
0
1
2
3
xk
3.5
2.5
15
.
0.5
0.5
15
.
2.5
3.5
{-3.5, -2.5, -1.5, -0.5, 0.5, 1.5, 2.5, 3.5}

Why?
We
assume that all points are

quantized to the nearest
quantization level
This determines the position of the
borders of the quantization regions
X f ( X ) X k
99
Prof. Okey Ugweje
100
Transfer Functions
Quantizers input/output characteristics ~ simple staircase

graphs
a f
xq nTs
output
(odd # of levels)
Zero assigned
to a quantization
level
Graphical representation
of the input and output
characteristics of the
quantizer
MIDRISER
MIDTREAD
y7
output
y6
(even # of levels)
x2
input
x4
y3
Zero assigned
to a decision level
y6
y5
y5
x3
x1
a f
xq nTs
x5
x6
a f
x1
x2 x
3
x nTs
input
y3 x4
x5
a f
x nTs
y2
y2
y1
y1
Biased
(Truncation)
Nonuniform
Prof. Okey Ugweje
101
Classification of Quantizers - 1
Prof. Okey Ugweje
Biased
102
Uniform (linear) vs. Nonuniform

Uniform => equally spaced quantization levels
Nonuniform => levels not equally spaced
Differential Pulse-Code Modulation (DPCM)

quantizes the prediction error rather than the actual
signal samples
uses a linear prediction filter
Scalar vs. Vector

Scalar => operates on each output separately
Vector => works on several samples at a time
Many signals exhibit strong correlation between samples
This implies that RX(t) RX(t + TS)
e,.g., in speech correlation b/w adjacent samples =0.9
quantizing 2 or more samples at a time exploits this
correlation
Prof. Okey Ugweje
103
Prof. Okey Ugweje
104
Uniform Quantizer (UQ) - 1
Adaptive DPCM (ADPCM)
A uniform quantizer is a quantizer for which

xk 1 xk q, k 0,1, ..., L 1
Has equal quantization levels
Each sample is approximated within a quantile interval
Optimal when the input pdf is uniform
allows the spacing between quantization levels to be

changed on the fly
used to avoid slope overload
Delta modulation
i.e. all values within the range are equally likely
1-bit DPCM
1
q
Vocoding (Voice Coding)

Transmits a mathematical model of a set of samples
rather than actual samples
q
2
q
2
Most ADCs are implemented using UQ

Error of a UQ is bounded by q e q
2
Prof. Okey Ugweje
105
Uniform Quantizer (UQ) - 1
Prof. Okey Ugweje
106
Nonuniform Quantizer (NQ) - 1
Uniform Quantization Transfer function
NQ have unequally spaced levels

spacing chosen to optimize the SNR
Output
signal
Characterized by:
Xq(t)
Variable step size

Quantizer step size depend on signal pdf
4
2
Input signal
-2
-8 -6 -4
8 X(t)
-2
2V p
Basic principle ~ use variable level sizes at regions

with variable pdf
concentrate q-levels in areas of largest pdf

use small (large) step size for weak (strong) signals
-4
-6
Uniform 3 bit Quantizer

Prof. Okey Ugweje
107
Prof. Okey Ugweje
108
Advantages:
NQ yields a higher average SNR than UQ when the
pdf is nonuniform which is usually the case in
practice
X
XX X X X X X X
The rms value of
Output signal
Xq(t)
Practically, NQ is realized by
sample compression followed

by UQ
Compression transforms the
input variable X to another
variable Y using a nonlinear
transformation
Prof. Okey Ugweje
Input signal
X(t)
109
Mathematical Description of Quantizer - 1
Prof. Okey Ugweje
110
Quantization adds random noise to the true value of

the sample
Process can be interpreted as an additive noise process
x t
the noise power is

proportional to the
sampled values
hence distortion is
minimized
Quantizer
+
x t f ( x) x t e(t )
The variance corresponds to the average quantization

noise power, i.e.,
( x x ) 2 f X ( x)dx
2 E x x 2
See eqn. 13.13
In NQ, we wish to make x x small when fX(x) is large

We can accept larger x x 2 when fX(x) is small
Want to minimize average noise variance
MSE penalizes large errors more than small errors
2
e(t ) x t x t
Let the quantizer error variance be
2
( x x) 2 f X ( x)dx ( x x ) 2 f X ( x)dx
where fX(x) is the probability density function

Prof. Okey Ugweje
111
Prof. Okey Ugweje
112
Signal-to-quantization noise ratio (SQNR) (or

simply SNR)
From above equation, average SNR can be written as
We have assumed
1. e(t) is uniformly distributed
2. {e(t)} is a stationary white noise process, i.e. e(j)
and e(k) are uncorrelated for j = k
3. e(t) is uncorrelated with the input signal x(t), and
4. signal sample xs(t) is zero mean and stationary
Signal Power
S

Noise Power
N avg
E {x 2 }
E e(t )
As a rule of thumb, each bit of quantization increases

the SNR by 6 dB provided that
a) xs(t) has a uniform distribution, and
b) the quantizer is a uniform quantizer
2
E {x 2 }
E {x 2 }
x f X ( x)dx

2
2
D
E x x ( x x ) f X ( x)dx
Prof. Okey Ugweje
113
Prof. Okey Ugweje
SNR for Uniform Quantizer - 1

If the input signal is a sequence, then
Given
1 N 1 2
PS
xs [ n ]
N n 0
N 1
q = step size, max quantization error is q
Signal power
1 N 1 2
PN
e [ n]
N n 0
Noise power
x [ n]
PS n
0
SNR
PN N1 e 2 [n]
L 1
V pp
L
2 2 q (error ) 2 p (e)de 2 q (e) 2 1q de 1q 2 q (e) 2 de

2
Signal-to-noise ratio
q
3 2
q2
1e
q 3 q 12
2
n 0
L/2 1 positive levels

L/2 1 negative levels
1 zero level
Prof. Okey Ugweje
V pp
where L = 2n is the # of quantization levels

The noise variance of the quantization error is given by
q
2
s
114
115
Prof. Okey Ugweje
Equation 13.12
L 1 level
L 2 intervals
This is the MSE

(noise variance)
116
SNR for Uniform Quantizer n- 3
Given
For UQ with nonuniform inputs use the formula
FH S IK E{x 2}2
N avg
E l x x q
q = step size
max quantization error is
V pp
L 1
V pp
Therefore, if a quantizer is (a) uniform with L levels,

(b) input is uniform pdf, then SNR is
Ppeak signal V pp2 12 2

S
SNR
L

Paverage q V pp2 4
N avg
2n
where L = is the # of quantization levels

Peak signal power
V pp
Ppeak signal
D = 2 = MSE
3L
See eqn. 2.20
This is the peak signal power to the average

quantization error power
Assuming Vpp is peak power

centered around zero (Vpp/2)
Average quantization noise power

2
Paverage
V pp
q2
2
12 L 12
Prof. Okey Ugweje
117
Therefore the SNR is
V pp
Ppeak signal V pp2 4 2

S
SNR
2 L L2

V pp 4
Ppeak q
N peak
Hence, there are two SNRs: Peak-to-Average and

Peak-to-Peak
For the peak, since L = 2n, SNR = 22n or in decibels
Peak quantization noise power

The quantization error is at worst half the
distance between quantization levels
The power of this error is therefore
Prof. Okey Ugweje
118
We can also find the peak signal power to the peak

quantization error power
Peak signal power
Ppeak q
Ppeak signal
Prof. Okey Ugweje
2
q V pp

2 2L
FH S IK
N
gain, each additional bit (doubling L) increases SNR

by 6 dB
e NS j a6n f dB RST04,.77,
c h
10 log10 2 2 n 6n dB
dB
dB
average SNR
peak SNR
Same technique is used to compute the SNR of a NQ

119
Prof. Okey Ugweje
120
Non-uniform Quantization - 1
For many classes of signals, UQ is not efficient

E.g., in speech signal smaller amplitudes predominate
and larger amplitudes are relatively rare
UQ will be wasteful for speech signals since many of
the quantizing levels are rarely used
An efficient scheme is to employ a non-uniform

quantizing method
Variable step sizes
smaller steps for small amplitudes
Let x = input
q(x) = quantized version
e(x) = x - q(x) = error
p(x) = pdf of x
Prof. Okey Ugweje
121
Prof. Okey Ugweje
122
NQ operates in 2 regions (linear and saturation)
For NQ, error is amplitude dependent

2
Lin can be formulated into discrete outputs as in UQ
Let Emax = saturation amplitude of the quantizer
The noise variance is given by
q2 E x q ( x) 2
2 0 e 2 ( x) p ( x)dx
2 0 max e 2 ( x) p ( x)dx 2 Emax e 2 ( x) p ( x)dx
2 x n1 e 2 ( x) p ( x)dx
x
n 0
If we consider a quantile interval qn = (xn+1 xn) and

assume e(x) x
see eqn. 13.14
2
2
Lin
sat
Prof. Okey Ugweje
L
1
2
where xn is a quantizer level

Note: In Chapter 13, your textbook uses N instead of L
e 2 ( x) p ( x)dx
E
2
Lin
123
2
Lin
Prof. Okey Ugweje
L
1
2
3
2 x
n 0 3
qn
2
qn
2
1 2
2
qn3
q
p ( xn ) 2 n p ( xn )qn
n 0 12
n 0 12
L
1
2
p ( xn ) 2
124
##Uniform vs. Nonuniform Quantization
Error is the weighted sum of error powers in each

quantile
Let
If the quantizer has uniform quantiles (i.e., UQ), then
Numerical integration will indicate that
L 1
2
2 qn2 p ( xn )qn
12 n0
L 1
2
1
2
2
q
q
12 n0 n qn L 2 n
D 0.1188, E[ x 2 ] 1
F SI
H NK
10 log10
avg
F 1 I 9.25dB
H 01188
K
.
However, NQ will yield a better result
1
q
L
2 1 q 2
q
12 2 q L 2
12
2
The best possible quantizer has
If the Q does not operate in the saturation region, then

2
q2 Lin
Prof. Okey Ugweje
1 x2
e
2
. , x2 0.498, x3 0.498, x4 1494

. q
lx11494
weighted by p(xn)qn
2
Lin
f X ( x)
FH S IK 12.0 dB
N avg
NQ can give better performance for most signals than UQ
125
Types of Noise in Quantizer
Prof. Okey Ugweje
Important/Practical Systems Using Quantization - 1
Overload Noise (Saturation Noise)
Reading Assignment:
Differential Quantization
when input signal > Lmax resulting in clipping of signal
Granularity Noise (Quantization Noise)

when L are not finely spaced apart enough to accurately
approximate input signal
predited (linear interpolation)
Differeence
Value
Actual data
(k+3)T
Truncation or Rounding error
kT
This type of noise is signal dependent
(k+2)T
Is used to reduce the dynamic range

Interpolation from previous value if samples are
correlated
Correlation can be increased by oversampling
Timing Jitter
Error caused by a shift in the sampler position
Easily isolated with stable clock reference and power
supply isolation
Oversampling
Predictor
more samples/sec
Prof. Okey Ugweje
126
127
Prof. Okey Ugweje
Differential
fewer samples/sec
128
Important/Practical Systems Using Quantization - 2

Example 22: (uniform quantization)

Differential PCM (DPCM)

Delta Modulation
Linear Predictive Coding
Adaptive Predictive Coding
x (t )
Sampler
fs 2 B
Quantizer x( n )
x ( n)
xk
n = # of binary bits used to

represent each sample
fs = sampling frequency or
sampling rate
xk x[n] xq [n] = quantized
value of x(t)
20.Example 20
Quantization
2q
3q
Prof. Okey Ugweje
xk
7
2
5
2
3
2
1
2
011
010
001
000
21.Example 21
Uniform Qantrizer
x k
L 2n
111
110
101
100
2q
12 q
3q
xk
32 q
52 q
72 q
Uniform Quantizer
129
Prof. Okey Ugweje
130
Example - Quantization
Let the quantization level be {1,3,5,7}. Assume that

the input signal to a quantizer have the pdf shown
f ( x)
1
4
f ( x)
8
x (t )
RS ,
T 0,
x
32
0 x8
else
a) Compute the signal mean power

b) Compute the mean square error at the quantizer
output
c) Compute the output SNR
d) How would you change the distribution of the
quantization level in order to decrease the
distortion?
Prof. Okey Ugweje
Companding
131
Prof. Okey Ugweje
132
Companding - 1
Companding - 2
Quantization along with sampling is used to generate

a Pulse Code Modulated (PCM) signal.
Using quantization, the instantaneous voltage value of
an analog signal is quantized into 28 (256) discrete
signal levels
With each sample, the signal is instantaneously
measured and adjusted to match one of the 256
discrete voltage levels
The adjustments of the voltage levels (256 discrete
levels), introduces some signal distortion
This distortion (quantizing noise) is greater for lowamplitude signals than for high-amplitude signals.
A technique called companding is used to correct this
problem
a method that compresses and divides the loweramplitude signals into more voltage levels and
provides more signal detail at the lower-voltage
amplitudes
Prof. Okey Ugweje
Prof. Okey Ugweje
133
134
Companding - 4
Companding - 3
Definition: Companding is a process of COMpressing the

signal at the Tx and exPANDING the signal at the Rx
Why Compand?
improve resolution (enhance SQNR) of weak
signals by
enlarging the signal, or
decreasing quantization step size
improves resolution of strong signals by
reducing the signal or
increasing the required quantization step size
reducing the # of bits required in the ADC & DAC
while reducing the dynamic range or improving the
SQNR
Transmitter Side
Signal
Input
Compressor
LPF
law
S/H +
ADC
Transmitter
ADC
Regenerative
Repeater
Receiver Side
Signal
Output
Expander
LPF
law
Signal
Out
DAC
Receiver
Signal
In
DAC
amplitude of one of the

signals is compressed
Prof. Okey Ugweje
135
Prof. Okey Ugweje
136
Companding - 5
Companding - 6
Since NQ are expensive and difficult to make, we

compand the signal and then use UQ
after compression, input of quantizer will have ly
uniform pdf
Companding is important for speech signals and has

been standardized for telephone interconnect around
the world
Two standards of companding techniques
US standard called -law algorithm
Companding introduces nonlinearity into the signal

maps nonuniform pdf into something resembling
uniform pdf
Prof. Okey Ugweje
European standard called A-law algorithm

conversion is required when calls are made between
countries using different algorithms.
137
Prof. Okey Ugweje
Input/Output Relationship
Types of Companding - 1
1.0
1.0
y(t )
ymax
-Law Companding (North & South America,

Japan)
y(t )
ymax
x(t )
xmax
y( x) ymax
R|
L
O
F
I
x
log 1 x
MN H K PQ sgn(x) |Sy
log a1 f
||
|Ty
e
max
1.0
x(t )
xmax
FH IK
FH x x IK 1
log LF x x I O
NM H K PQ , FH x x IK 1
log
xx
max
,
max
loge
max
-1.0
138
max
max
max
where
x and y represent the input and output voltages
is a constant number determined by
experiment
Y = log X is the most commonly used compander

Taking the log of Y = log X reduces the dynamic range since
log e 1 x x if x 0
Prof. Okey Ugweje
139
Prof. Okey Ugweje
140
In U.S., telephone lines uses = 255

Samples 4 kHz speech waveform at 8,000
sample/sec
Encodes each sample with 8 bits, L = 256 quantizer
levels
Hence data rate R = 64 kbit/sec
= 0 corresponds to uniform quantization
A-Law Companding (Europe, China, Russia, Asia,

Africa)
R| A x x
|y 1 A sgn(x),
S| L1 log F A x I O
H x K PQ sgn(x),
||y MN 1 log
A
T
max
y( x )
max
max
max
x
1
xmax A
See eqn. 2.23
1
1
A xmax
x
where
x and y represent the input and output voltages
A is a constant number determined by experiment, A = 87.6
You can find the companding gain by differentiating

the output
d
G
Prof. Okey Ugweje
141
Prof. Okey Ugweje
dx
y( x )
x0
142
Encoding - 1

A/D Converter
xa(t)
Analog
signal
Sampler
x[n]
Quantizer
Discrete-Time
signal
xq[n]
Line
Coder
an
Quantized
signal
Quantizer output is one of L possible signal levels
Encoding
For binary transmission, each quantized sample is

mapped into an n-bit binary word
Encoding is the process of representing each of

the L outputs of the quantizer by an n-bit code
word
one-to-one mapping - no distortion introduced
Prof. Okey Ugweje
143
Prof. Okey Ugweje
144
Encoding - 2
Pulse Modulation Techniques - 1
Pulse Code Modulation (PCM) is commonly used
Recall that analog signals can be represented by a

sequence of discrete samples (output of sampler)
APM results when some characteristic of the pulse
(amplitude, width or position) is varied in
correspondence with the data signal
Can be obtained either by Natural or Flat top Sampling
PCM refers to a digital baseband signal that is

generated directly from the quantizer output
Sometimes PCM is used interchangeably with
quantization
Prof. Okey Ugweje
145
Prof. Okey Ugweje
146
Two Types:
Pulse Amplitude Modulation (PAM)
The amplitude of the periodic pulse train is varied in
proportion to the sample values of the analog signal
Pulse Time Modulation
Encodes the sample values into the time axis of the digital
signal
Pulse Width Modulation (PWM)
Constant amplitude, width varied in proportion to the
signal
Pulse Duration Modulation (PDM)
sample values of the analog waveform are used in
determining the width of the pulse signal
Prof. Okey Ugweje
147
Prof. Okey Ugweje
148
Pulse Code Modulation (PCM) - 1
Analog
PCM
Sample
Quantize
Assign Code #
Convert to Binary #s
See Figure 2.16
Prof. Okey Ugweje
149
150
Quantization and encoding of a sampled signal

Prof. Okey Ugweje
Prof. Okey Ugweje
151
Prof. Okey Ugweje
152
Prof. Okey Ugweje
153
154
Implementation of A/D Converters
Advantages of PCM
Relatively inexpensive
Easily multiplexed
PCM waveforms from different sources can be
transmitted over a common digital channel (TDM)
Easily regenerated:
useful for long-distance communication e.g., telephone
Better noise performance than analog system
Modem is all digital, thus affording reliability, stability and is
readily adaptable to integrated circuits
Signals may be stored and time-scaled efficiently (e.g.,
satellite communication)
Efficient codes are readily available
Disadvantage
Requires wider bandwidth than analog signals
Prof. Okey Ugweje
Prof. Okey Ugweje
155
A/D Converter
xa(t)
Analog signal
x[n]
Sampler
Quantizer
Discrete-Time signal
xq[n]
Coder
Quantized signal
Digital signal
Quantizer
ADC
x f ( x )
SIO
n b its
Serial Input Output (SIO) circuit converts quantization

level to a sequence of bits n = log2 L
Prof. Okey Ugweje
156
Comparison of Practical ADCs
Counting or Ramp ADC

Test value is incremented in equal steps until
it is equal to input sample
Serial or Successive Approximation ADC
Uses binary search to narrow range of input
sample until desired accuracy is reached
Parallel or Flash ADC
Input samples compared with all possible
quantization levels at once
Prof. Okey Ugweje
Speech Coding
157
Prof. Okey Ugweje
Speech Coding - 1
158
Speech Coding - 2
Introduction To Speech Coding

To date, most source encoding techniques is based on
the -law or the A-law companding of A/D and D/A
converters
They are often referred to as CODECS
A CODEC is a device designed to convert analog
signals, such as voice, into PCM-compressed samples
to be sent into digital carries
The process is reversed at the receiver
The term CODEC is an acronym for CODer/DECoder
signifying the pulse coding/decoding function of the
device
Originally, CODEC functions were managed by

separate devices, each performing the function
necessary for PCM communication such as, sampling,
quantization, A/D, D/A, filtering, companding, etc.
Presently, these function are integrated into a single
chip e.g. Intels 2913
CODECS form the digital interface for most telephone
lines all over the world
At the exchange each analog signal from the local
telco is converted using an 8-bit -law or A-law codec,
with a standardized sampling rate of 8000 times per/s
Prof. Okey Ugweje
Prof. Okey Ugweje
159
For max voice frequency 3400 Hz, Nyquist criterion is

satisfied
160
Speech Coding - 3
Characteristics of Speech Signal - 1
This results in a data rate of 64 kbps for each voice link

At the exchange, a number of these 8-bit data words from
different phone sources are multiplexed into a frame (32 for Etype and 24 for A-type systems)
They are then sent using either baseband or bandpass
signaling methods over the national and international exchange
Speech waveform have a number of useful properties

that can be exploited when designing efficient coders
They are then sent using

either baseband or
bandpass signaling
methods
In order to keep pace with
the codec sampling rate, a
new frame must be
constructed and sent
every 1/8000 sec (see fig.)
Prof. Okey Ugweje
1. Nonuniform probability
2.
3.
4.
5.
See Digital Communications by Andy

Bateman
161
6.
distribution of speech amplitude

Nonzero autocorrelation between
successive speech samples
Non-flat nature of the speech
spectra
Existence of voiced and unvoiced
segments in speech
Quasi-periodicity of voice speech
signals
Speech signals are essentially
bandlimited
Prof. Okey Ugweje
(also see Fig. 13.18,

page 836)
162
Hierarchy of Speech Coders
Characteristics of Speech Signal - 2

The most basic property of speech waveform that is

exploited in speech encoders is that they are
essentially bandlimited
A finite bandwidth means that it can be sampled at a
finite rate and reconstructed completely provided that
fs 2fmax (Nyquist criteria)
Speech Coders
Waveform Coders
163
Prof. Okey Ugweje
Vocoders
Frequency
Domain
Time
Domain
PCM
Source Coders
Linear Predictive Coders
Nondifferential
Prof. Okey Ugweje
Power spectrum
Differential
Delta
ADPCM
CVSDM
APC
Subband Coding
Adaptive Transform Coding
164
Coding Techniques for Speech - 1
The goal of all speech coding systems is to

transmit speech with the highest possible quality
using the least possible channel capacity
Speech coders differ widely in their approach to
achieve this objective
They all employ quantization & exploits different
properties of speech signal
Prof. Okey Ugweje
In DPCM, the difference between successive samples are

encoded rather than the samples themselves
Since difference b/w samples are expected to be smaller than the
samples themselves, fewer bits are required to represent the
difference
because most signals sampled at Nyquist rate or faster exhibit
significant correlation between successive samples
165
166
i.e., average change in successive samples is relatively

small
Speech signals fall into this group because samples of
speech signals is very strongly correlated from one sample
instant to the next
Antialiasing
Filter
Sampler
Prediction
Filter
Quantizer
DPCM
Signal
Digital Communication Channel

Regeneration
Circuit
DAC
Analog
Input
Signal
Prediction
Filter
DPCM Block Diagram

Prof. Okey Ugweje
Prof. Okey Ugweje
Analog
Input
Signal
Waveform Coding
A) Time Domain
Designed to represent the time domain characteristics of
speech signal
For high bit rates (16 - 64 kbps) it is sufficient to just sample
and quantize the time domain voice waveform, e.g., Differential
Pulse Code Modulation (DPCM)
Differential Pulse Code Modulation (DPCM)
167
Hence exploiting this redundancy will result in better

performance
This is the concept behind DPCM
A refinement to this general approach is to predict
the current samples based on the previous sample
DPCM quantizes the difference of one sample and
the predicted value of the next sample (this is
usually much less than the absolute value of the
samples)
In practice, DPCM is implemented using a
prediction scheme that exploits the correlation
between successive samples
Prof. Okey Ugweje
168
Instead of quantizing & coding sample values, as in

PCM, an estimate is made (with linear prediction
filter) for the next sample value based on previous
sample
In DPCM, the error at the output of a prediction filter is
quantized, rather than the voice signal itself
It is assumed that the error of the prediction filter is much
smaller than the actual signal itself
DPCM Issues
Linear prediction filter is usually just a feed forward finiteduration impulse response (FIR) filter
The filter coefficients must be periodically transmitted
While DPCM works well on speech, it does not work well
for modem signals
Prof. Okey Ugweje
169
Prof. Okey Ugweje
170
Unlike PCM, APCM and ADPCM however exploit

the redundancies present in the speech signal
because adaptive quantizers vary the step size between
quantization levels depending on whether speech is
loud or soft
Since the speech samples are highly correlated, it

means that the variance of the difference between
adjacent speech amplitude is smaller than the
variance of the signal itself
In ADPCM, the quantization resolution can be
changed on the fly
ADPCM allows speech to be encoded at 32 kb/s
This is used in the DECT
Prof. Okey Ugweje
Adaptive PCM (APCM) and Adaptive DPCM

(ADPCM):
Many sources are quasi-stationary in nature such
that the variance and the ACF of the source vary
slowly with time
The efficiency and performance of PCM can be
improved by exploiting the slowly time-varying
statistics of the source
A simple implementation is to use a uniform
quantizer that varies its step size according to the
past signal samples
Such techniques are known as APCM and ADPCM
171
Delta Modulation (-mod):

In communication systems application, bandwidth is limited
A given transmission channel (wires-pairs, coaxial cables,
optical fibers, microwave links, and others) represents a
finite spectral resource
Hence, developing spectrally efficient (reduced bandwidth)
signaling technique is important
This is the motivation for Delta Modulation (DM)
If a quantizer of a DPCM is restricted to 1 bit (i.e. 2 levels
only q), then the resulting scheme is called DM
In other words, DM is a special case of DPCM where
there are only two quantization levels
Delta modulation can be implemented with an extremely
simple 1 bit quantizer
Prof. Okey Ugweje
172
Adaptive Delta Modulation

In conventional DM, both quantization and slope
overload noise is a problem
The exploitation of signal correlation in DPCM suggest
that oversampling a signal will increase the correlation
between samples
This can be overcome by oversampling (i.e., keeping
the DM size small and sampling at many times the
Nyquist rate)
It is an extreme case of DPCM in which signal is
oversampled and R = 1 bit/sample
Adaptive Delta Modulation at 16 kbits/sec can produce
reasonable quality speech
Prof. Okey Ugweje
173

Prof. Okey Ugweje
174

Subband Coding
Human ear cannot detect quantization distortion at
all frequency equally well
Human perceptions of speech quality depend on
the frequency band
Subband coders filter the speech signal into
multiple bands using Quadrature Mirror Filters
(QMF) or Discrete Fourier Transform (DFT)
That is, the speech is divided into many smaller
bands and then encode each subband separately
according to some perception criteria
Prof. Okey Ugweje
B) Frequency Domain
Spectral Waveform Coders manipulates the spectral
characteristics of speech waveform
Frequency domain samples are represented
according to their perceptual criteria
Subband Coding (SBC) is an example of spectral
waveform coding
175
Band splitting is used to exploit the fact that individual

bands do not all contain signals with the same energy
This permits the accuracy of quantizer to be reduced in
bands with very low energy and very high energy
Higher MSE may be tolerated at very low and very high
frequencies
Band splitting can be done in many ways (equally or

unequally) using a bank of filters
Each subband is sampled at a bandpass Nyquist rate
(lower than the sampling rate) and then encoded with
different accuracy based on perception criteria
Filtered signals are quantized using standard PCM
(different R for each signal)
Prof. Okey Ugweje
176
Adaptive Transform Coding

Signal samples are grouped into frames and
encoded into number of bits proportional to its
perception significance
Correlated time samples are transformed into
(hopefully) uncorrelated frequency domain samples
using FFT or Discrete Cosine Transform
This is a more complex technique which involves
block transformations of input segment of the
speech signal
Source Coding (Model-Based Encoding)

For low bit rate voice encoding it is necessary to
mathematically model the voice and transmit the
parameters associated with the model
This type of coding attempts to replicate a model of
the process by which speech was constructed
Prof. Okey Ugweje
Prof. Okey Ugweje
177
A) Linear Predictive Coding (LPC)

Linear Predictive Coding (LPC) uses a prediction
algorithm for synthesis of the desired signal
Human speech is modeled as noise (air from lungs)
exciting a linear filter (throat, vocal cords, and mouth)
The excitation sequence and filter coefficients are
quantized by a linear prediction speech encoder
Vector quantization is frequently used in this technique

In LPC, speech is divided into frames of approximately 20 ms
Linear predictive coding is similar to DPCM with the following
exceptions:
prediction filter is more complex
Excitted
Sequence
Prediction Filter
Filter Coefficients
Filter Gain
more taps in the FIR filter
filter coefficients are transmitted more frequently

once every 20 milliseconds
The error signal is not transmitted directly

The error signal can be considered as a type of noise
Instead the statistics of the noise are transmitted
Output Speech
Power level
Whether voiced (vowels) or unvoiced (consonants)
LPC quantizes excitation sequence, filter coefficients

and filter gain and transmits them to receiver
Prof. Okey Ugweje
178
This is where big savings (in terms of bit rate) comes from
179
Prof. Okey Ugweje
180

B) Vocoder (voice coders)

Vocoders are coding devices that extract significant
components of a speech waveform, exploiting speech
redundancies, to achieve low bit rate transmission
Most vocoding techniques are based on linear
predictive coding
Vector Sum Excited Linear Prediction (VSELP)
Employed in U.S. Digital Cellular (IS-136) standard
Uses 20 ms frames
Each frame is represented with 159 bits (Total data
rate is 8 kbps)
A two stage vector quantizer is used to quantize the
excitation sequence
Prof. Okey Ugweje
181
Comparison of Speech Coding Standards

Type
PCM
ADPCM
Subband
VSELP
Theory
Some bits (like filter gain) are much more important for
perpetual quality than others. These are protected by
error correction coding
RPE-LTP
Regular Pulse Excited Long Term Prediction
Used in GSM (European Digital Cellular) 13 kbps
QCELP
Qualcomm Code Excited Linear Predictive Coder
Used in IS-95. (US Spread Spectrum Cellular)
Variable bit rate (full, half, quarter, eighth)
Original full rate was 9.6 kbps
Revised standard (QCELP-13) uses 14.4 kbps
Prof. Okey Ugweje
182
Tradeoff: Voice Quality vs. Bit Rate

Rate Complexity
(kb/s)
(MIPS)
64
32
16
8
~1
0.01
0.1
1
~100
?
Delay
(ms)
0
0
25
35
?
Quality
Perceived
Speech
Quality
Excellent (5)
Toll quality
Good (4)
High
High
High
Fair
High
Fair (3)
Communications
quality
Poor (2)
Vocoders
Unsatisfactory (1)
1.2
References for Speech Coding Techniques:

N. S. Jayant, Coding Speech at Low Bit Rates, IEEE
Spectrum, August 1986.
N. S. Jayant, et. al., Coding of Speech and Wideband
Audio, AT&T Technical Journal, October 1990.
this article is more technical than the first, but still very
readable
Prof. Okey Ugweje

2.4
Waveform coders
4.8
9.6
16
Bit Rate (kbps)
24
32
64
The bit rate produced by the voice coder can be

reduced at a price
Increased hardware complexity
Reduced perceived speech quality
183
Prof. Okey Ugweje
184
Image and Video Coding
1000x1000 pixel image with 8 bits for each of 3 colors

requires 24 Mbits to encode
Video requires ~ 20 frames/second
Compression standards vital for any hope of digital
video
JPEG: Image compression of 20:1 or more
MPEG: Video compression of 100:1 or more
Digital-To-Digital Conversion
(Line Coding)
Reference:
P. H. Ang, et. al., Video Compression Makes Big
Gains, IEEE Spectrum, October 1990
Prof. Okey Ugweje
185
Prof. Okey Ugweje
Digital-To-Digital Conversion
186
Line Coding - 1
In this section, we see how we can represent digital

data by using digital signals.
The conversion involves three techniques: line coding,
block coding, and scrambling.
Line coding is always needed; block coding and
scrambling may or may not be needed.
Line coding is the process of converting digital data

to digital signals. We assume that data, in the form of
text, numbers, graphical images, audio, or video, are
stored in computer memory as sequences of bits.
Prof. Okey Ugweje
Prof. Okey Ugweje
187
Line coding and decoding
188
Line Coding - 2
Line Coding - 3
Signal Element Vs Data Element
Let r be the number of data elements carried by each

signal element. Figure below shows several situations
with different values of r.
In data communications, our goal is to send data

elements.
A data element is the smallest entity that can represent
a piece of information: this is the bit.
In digital data communications, a signal element carries
data elements.
A signal element is the shortest unit (timewise) of a
digital signal. In other words, data elements are what
we need to send; signal elements are what we can
send. Data elements are being carried; signal elements
are the carriers.
Prof. Okey Ugweje
189
Signal element versus data element
Prof. Okey Ugweje
Line Coding - 4
190
Example
Data Rate Vs Signal Rate

Data rate defines the number of data elements (bits) sent in
1s. The unit is bits per second (bps).
Signal rate is the number of signal elements sent in 1s. The
unit is the baud.
The data rate is sometimes called the bit rate; the signal rate
is sometimes called the pulse rate, the modulation rate, or
the baud rate.
Relationship of data rate & signal rate (bit rate & baud rate).
A signal is carrying data in which one data element is encoded

as one signal element ( r = 1). If the bit rate is 100 kbps, what is
the average value of the baud rate if c is between 0 and 1?
Solution
We assume that the average value of c is 1/2 . The baud rate is
then
This relationship, of course, depends on the value of r. It also

depends on the data pattern C. If we have a data pattern of all 1s
or all 0s, the signal rate may be different from a data pattern of
alternating 0s and 1s.
Prof. Okey Ugweje
191
Prof. Okey Ugweje
192
Line Coding - 5
Example
Although the actual bandwidth of a digital signal is

infinite, the effective bandwidth is finite.
we can say that the bandwidth (range of frequencies)
is proportional to the signal rate (baud rate). The
minimum bandwidth can be given as
The maximum data rate of a channel (see Chapter 3) is

Nmax = 2 B log2 L (defined by the Nyquist formula).
Does this agree with the previous formula for Nmax?
Solution
A signal with L levels actually can carry log2L bits per level. If each
level corresponds to one signal element and we assume the average
case (c = 1/2), then we have
We can solve for the maximum data rate if the bandwidth of the
channel is given.
Prof. Okey Ugweje
193
Prof. Okey Ugweje
Transmitter - 1
Transmitter - 2
Output of the A/D converter is a set of binary bits
A line coder or baseband binary transmitter

transforms a stream of bits into a physical waveform
suitable for transmission over a channel
There are many types of waveforms. Why?
performance criteria!
Each line code type have merits and demerits
The choice of waveform depends on operating
characteristics of a system such as
which are abstract entities that have no physical definition
We use pulses to convey a bit of information, e.g.,

f(t)
f(t)
-1
1
T
To transmit over a physical channel, bits must be

transformed into a physical waveform
Baseband systems transmit data using many kinds of
pulses
Before signals are applied to the modulator, it may be
put into several different waveforms
Prof. Okey Ugweje
194
195
Modulation-demodulation requirements
Bandwidth requirement
Synchronization requirement
Receiver complexity, etc.,
Prof. Okey Ugweje
196
Goals of Line Coding (qualities to look for) - 1
Baseline Wandering
In decoding a digital signal, the receiver calculates a running
average of the received signal power. This average is called
the baseline.
The incoming signal power is evaluated against this baseline
to determine the value of the data element.
A long string of 0s or 1s can cause a drift in the baseline
(baseline wandering) and make it difficult for the receiver to
decode correctly.
A good line coding scheme needs to prevent baseline
wandering.
DC Components
When the voltage level in a digital signal is constant for a
while, the spectrum creates very low frequencies.
These frequencies around zero, called DC (direct-current)
components, present problems for a system that cannot
pass low frequencies or a system that uses electrical
coupling (via a transformer).
For example, a telephone line cannot pass frequencies
below 200 Hz. Also a long-distance link may use one or
more transformers to isolate different parts of the line
electrically.
For these systems, we need a scheme with no DC
component.
Prof. Okey Ugweje
Prof. Okey Ugweje
197
Self-synchronization
Low probability of bit error
To correctly interpret the signals received from the

sender, the receiver's bit intervals must correspond
exactly to the sender's bit intervals. If the receiver clock
is faster or slower, the bit intervals are not matched and
the receiver might misinterpret the signals.
The ability to recover timing from the signal itself
i.e., self-clocking (self-synchronization)
- ease of clock lock or signal recovery for symbol synch.
Receiver needs to be able to distinguish the waveform

associated with a mark (or 1) from a space (or 0)
BER performance
relative immunity to noise
Error detection capability

enhances low probability of error
Transparency
Long series of ones and zeros could cause a problem
Prof. Okey Ugweje
198
property that any arbitrary symbol or bit pattern can be

transmitted and received, i.e., all possible data
sequence should be faithfully reproducible
199
Prof. Okey Ugweje
200
Summary of Major Line Codes - 1
Spectrum suitable for the channel
Categories of Line Codes

1. Polar - send pulse or negative of pulse
2. Unipolar - send pulse or a 0
3. Bipolar (a.k.a. Alternate Mark Inversion (AMI), pseudoternary)
Spectrum matching of the channel

e.g. presence or absence of DC level
In some cases DC components should be avoided

The transmission bandwidth should be minimized
Represent 1 by alternating signed pulses
Power Spectral Density (PSD)
Generalized Pulse Shapes

1. NRZ - pulse lasts entire bit period
2. RZ - pulse lasts just half of bit period
3. Manchester Line Code
Particularly its value at zero

PSD of code should be negligible at the frequency near zero
Transmission bandwidth
Should be as small as possible
Send a 2- pulse for either 1 (highlow) or 0 (lowhigh)
4. HS ( Half Sine)
Prof. Okey Ugweje
201
Prof. Okey Ugweje
202
Combined category and generalized pulse shapes
Bipolar RZ
Alternate between positive and negative pulses to send a 1
This alternation eliminates the DC component
Polar NRZ
Wireless, radio, satellite applications (bandwidth efficient)
Unipolar NRZ
Turn the pulse ON for a 1, leave the pulse OFF for a 0 in
entire bit period
For noncoherent communication where receiver cant decide
the sign of a pulse
fiber optic communication often use this signaling format
Unipolar RZ
RZ signaling has both a rising and falling edge of the pulse
This can be useful for timing and synchronization purposes
There are many other variations of line codes (see Fig. 2.22,
page 87 for more)
Prof. Okey Ugweje
Prof. Okey Ugweje
203
desirable for many channels that cannot transmit DC components
Generalized Grouping
Non-Return-to-Zero:
Return-to-Zero:
Phase-Coded:
Multilevel Binary:
NRZ-L, NRZ-M NRZ-S

Unipolar, Bipolar, AMI
bi--L, bi--M, bi--S, Miller, Delay Mod.
dicode, doubinary
204
Line Coder
Line Coder
Xn
s( t )
Input Xn is the output of the A/D converter

or a sequence of values that is a function of the data bit
Output is given by
where
These values are determined by the type of line code

that is being used
205
Prof. Okey Ugweje
1. Unipolar NRZ
Unipolar NRZ is defined by unipolar mapping
Unipolar NRZ
RS A,
T 0,
FG t IJ ,
HT K
Compared with its polar counterpart, this scheme is

very costly
The normalized power (power needed to send 1 bit
per unit line resistance) is double that for polar NRZ
For this reason, this scheme is normally not used in
data communications today
NRZ Pulse Shape
where Tb is the bit period

A
Prof. Okey Ugweje
206
when X n 1
when X n 0
The pulse shape for unipolar NRZ is:

f (t )
Commonly Used Line Codes - 2
an
an = symbol mapping function

f(t) = pulse shape function
Tb = bit period (Tb=Ts/n for n bit quantizer)
NRZ = Non-Return-to-Zero
RZ = Return-to-Zero
AMI = Alternate Mark Inversion
Prof. Okey Ugweje
s(t ) an f (t nTb )
Tb
2Tb
3Tb
4Tb
5Tb
207
Prof. Okey Ugweje
208
2. Polar Line Codes

A Polar line code uses the antipodal mapping
Polar NRZ-L and NRZ-I
an
RS A,
T A,
when Xn 1
when Xn 0
where Xn is the nth data bit
Polar NRZ uses NRZ pulse shape

Polar RZ uses RZ pulse shape
1
Tb
0
1
nonreturn to zero- level; nonreturn to zero- invert
4Tb
2Tb
3Tb
2Tb
3Tb
5Tb
Polar NRZ
In NRZ-L the level of the voltage determines the value of the bit.
In NRZ-I the inversion or the lack of inversion determines the value of the bit.
-A
A
4Tb
Tb
0
-A
Prof. Okey Ugweje
5Tb
Polar RZ
NRZ-L and NRZ-I both have an average signal rate of N/2 Bd.
NRZ-L and NRZ-I both have a DC component problem.
209
Prof. Okey Ugweje
210
Polar NRZ-L and NRZ-I
3. Bipolar Line Codes

A space is mapped to '0' & a mark is alternately
mapped to -A and +A
A, when Xn 1 and last mark -A
an A, when Xn 1 and last mark +A
0, when Xn 0
Baseline Wandering is a problem for both variations, it is

twice as severe in NRZ-L. If there is a long sequence of 0s
or ls in NRZ-L, the average signal power becomes skewed.
The receiver might have difficulty discerning the bit value. In
NRZ-I this problem occurs only for a long sequence of 0s. If
somehow we can eliminate the long sequence of 0s, we can
avoid baseline wandering. We will see shortly how this can
be done.
The synchronization problem (sender and receiver clocks
are not synchronized) also exists in both schemes. Again,
this problem is more serious in NRZ-L than in NRZ-I. While a
long sequence of 0s can cause a problem in both schemes,
a long sequence of ls affects only NRZ-L.
R|
S|
T
Also called pseudoternary or AMI

Either RZ or NRZ pulse shape can be used
A
2Tb
0
3Tb
Tb
5Tb
4Tb
Bipolar
RZ
-A
Prof. Okey Ugweje
211
Prof. Okey Ugweje
212

Polar Biphase: Manchester Line Codes

Uses antipodal mapping and split-phase pulse shape
Ft T I Ft T I
f (t) G T 4 J G T 4 J
GH JK GH JK
b
21
-A
In Manchester and differential Manchester encoding, the transition at

the middle of the bit is used for synchronization.
The minimum bandwidth of Manchester and differential Manchester is
2 times that of NRZ.
Prof. Okey Ugweje
213
Prof. Okey Ugweje
214
Bipolar schemes: AMI and pseudoternary

Bipolar encoding (a.k.a multilevel binary), three levels are used: positive,
zero, and negative.
The bipolar scheme was developed as an alternative to NRZ. It has the

same signal rate as NRZ, but there is no DC component.
The NRZ scheme has most of its energy concentrated near zero frequency,
which makes it unsuitable for transmission over channels with poor
performance around this frequency. The concentration of the energy in
bipolar encoding is around frequency N/2.
Prof. Okey Ugweje
The Manchester scheme overcomes several problems associated with NRZ-L, and
differential Manchester overcomes several problems associated with NRZ-I.
First, there is no baseline wandering. There is no DC component because each bit
has a positive and negative voltage contribution.
The only drawback is the signal rate. The signal rate for Manchester and differential
Manchester is double that for NRZ. The reason is that there is always one transition
at the middle of the bit and maybe one transition at the end of each bit.
215
mBnL Multilevel Scheme:

In the schemes, a pattern of m data elements is encoded as
a pattern of n signal elements in which 2m Ln.
E.g., Multilevel: 2B1Q scheme (two binary, one
quaternary).
It uses data patterns of size 2 and encodes the 2-bit patterns
as one signal element belonging to a four-level signal. In this
type of encoding m = 2, n = 1, and L = 4 (quaternary).
The average signal rate of 2B1Q is S = N/4. This means that
using 2B1Q, we can send data 2 times faster than by using
NRZ-L. However, 2B 1Q uses four different signal levels,
which means the receiver has to discern four different
thresholds.
Prof. Okey Ugweje
216
Multilevel: 8B6T scheme eight binary, six ternary

The idea is to encode a pattern of 8 bits as a pattern of 6 signal
elements, where the signal has 3 levels (ternary).
The 3 possible signal levels are represented as -, 0, and +.

The first 8-bit pattern 00010001 is encoded as the signal pattern -00++ with weight 0; the second 8-bit pattern 01010011 is encoded as + - + + 0 with weight +1. The third bit pattern should be encoded as +
- - + 0 + with weight +1.
To create DC balance, the sender inverts the actual signal. The
receiver can easily recognize that this is an inverted pattern because
the weight is -1. The pattern is inverted before decoding.
Prof. Okey Ugweje
217
Prof. Okey Ugweje
218
Multilevel: 8B6T scheme eight binary, six ternary

28
Multitransition: MLT-3 scheme

36
In this scheme, we can have = 256 different data patterns and =

478 different signal patterns. There are 478 - 256 = 222 redundant
signal elements that provide synchronization and error detection.
Part of the redundancy is also used to provide DC balance. Each
signal pattern has a weight of 0 or +1 DC values.
That is, there is no pattern with the weight -1.
To make the whole stream DC-balanced, the sender keeps track of
the weight. If two groups of weight 1 are encountered one after
another, the first one is sent as is, while the next one is totally
inverted to give a weight of -1.
1. If next bit is 0, there is no transition.

2. If next bit is 1 and the current level is not 0, the next level is 0.
3. If the next bit is 1 and the current level is 0, the next level is the
opposite of the last nonzero level.
The average signal rate of the scheme is theoretically

The minimum bandwidth is very close to 6N/8.
Prof. Okey Ugweje
219
Prof. Okey Ugweje
220

Multitransition: MLT-3 scheme

One scheme that maps one bit to one signal element.
The signal rate is the same as that for NRZ-I, but with greater
complexity (three levels and complex transition rules).
It turns out that the shape of the signal in this scheme helps to
reduce the required bandwidth.
Let us look at the worst-case scenario, a sequence of 1 s. In
this case, the signal element pattern +V0 -V0 is repeated every
4 bits.
A nonperiodic signal has changed to a periodic signal with the
period equal to 4 times the bit duration.
This worst-case situation can be simulated as an analog signal
with a frequency one-fourth of the bit rate. In other words, the
signal rate for MLT-3 is one-fourth the bit rate.
Prof. Okey Ugweje
221
Summary of Line Codes

NRZ-M
+V
1
NRZ level (or change)
"1" represented by one level
"0" represented by other level
-V
+V
Bipolar
RZ
RZ-AMI
Bi-o-L
-V
+V
NRZ Space
"1" represented by no change in level
"0" represented by a change in level
-V
+V
0
+V
Bipolar RZ
"0's" & "1's" represented by opposite level polar
pulses that are half-bit wide
-V
+V
-V
+V
Bi-phase Level (Manchester II + 180)

"1" represented by a "10"
"0" represented by a "01"
+V
Bi-phase Space
A transition at beginning of every bit period
"1" represented by a no 2nd transition
"0" represented by a 2nd transition one-half bit period
later
-V
+V
Delay
Modulation -V
+V
Decode
NRZ
-V
Decode
RZ
Decode NRZ
A "1" to "0" or "0" to "1" transition produces a half
duration polarity change; otherwise a zero is sent.
-V
0
Prof. Okey Ugweje
2T
3T
4T
5T
6T
7T
8T
9T
10T
11T
s(t )
an f t nTs
X(f )
R (k )e j 2 fkTs
Ts k
where R(k) is the Autocorrelation (AC) Function of the data

sequence at the encoder output
For Autocorrelation please see Section 1.4
Delay Modulation
A "1" to "0" or "0" to "1" changes polarity;
otherwise a zero is sent.
+V
s( t )
Gs ( f )
Bi-phase Mark (Manchester I)

A transition at beginning of every bit period
"1" represented by a 2nd transition 1/2 bit period later
"0" represented by no 2nd transition
+V
Line Coder
Average PSD of a line code is given by
RZ AMI
"0" represented by no signal; successive "1's"
represented by equal amplitude alternating pulses
-V
Bi-o-S
222
Ts = symbol duration (Ts= Tb for binary, Ts= kTb for M-ary)

f(t) = symbol pulse shape
an = a set of Random Variables representing data bits (voltage level of data)
Unipolar RZ
"1" represented by a 1/2-bit wide pulse
"0" represented by no pulse condition
-V
Bi-o-M
Xn
NRZ Mark
"1" represented by a change in level
"0" represented by no change in level
NRZ-S
Unipolar
RZ
Prof. Okey Ugweje
Power Spectral Density (PSD) of Line Codes - 1
NRZ-L
Summary of line coding schemes
Decode RZ
A "1" represented by a transition at the midpoint of a
bit interval; a "0" is represented by no transition
unless it is followed by another zero; In this case, a
transition is placed at the end of the bit period.
Correlation is a matching process

AC is the matching of a signal with the delayed version of itself
223
Prof. Okey Ugweje
224
AC is denoted as RXX(t1,t2) or RXX(t, t+) or RXX()

AC of a random process X(t) is given by
Average PSD of a line code (contd)

R( k ) E an an k , k 0,1, 2,
RXX t1 , t2 E X t1 X * t2

x1 x2 f x1 , x2 ; t1 , t2 dx1dx2
It follows that
a f
an an* k i Pi
i 1
where
af af
RXX t1, t2 E X t2 X t1
*
Value of RX(t1, t2) when t1 = t2 = t is the average power

of X(t), i.e.,
Pi = probability of getting (anan+k)i

M = # of positive values of anan+k
a f
RXX t, t E X 2 t 0
Reading Assignment: Section 1.4
Prof. Okey Ugweje
225
How to Compute PSD of Line Codes - 1
226
How to Compute PSD of Line Codes - 2
Unipolar line codes
Brute-force Method
Model s(t) as a Wide Sense Stationary (WSS) random
process
Find Autocorrelation Function (ACF) of s(t)
Gs ( f )
Gs ( f )
Apply Wiener-Khintchine theorem to get PSD

PSD of a RP X(t), GX(f), is the Fourier transform of the ACF
Shortcut Method for Finding PSD of a Line Code
A2
2
X ( f ) sin 2 fTb
Tb
Unipolar Line Codes with NRZ Pulse Shapes

If the pulse shape is NRZ, then
n
X ( f ) 0for f when n 0
Tb
Thus
A2
2
Gs ( f )
X ( f ) 1 ( f )
4Tb
Assume equiprobable & independent data symbols

Polar line codes
2
A
2
X(f )
Tb
X(f) = Fourier Transform of the pulse shape
Gs ( f )
A2
1
n
2
X ( f ) 1
f
4Tb
Tb
Tb n
Bipolar line codes
this step can be tricky & cumbersome!
Prof. Okey Ugweje
Prof. Okey Ugweje
227
Prof. Okey Ugweje
228
Example 24
Example 24 Solution
For k 0:
4
R(k ) an ank i pi
Find the PSD of x(t) Unipolar NRZ

1
i 1
2Tb
Tb
3Tb
4Tb
an ank 1 p1 an an k 2 p2 an ank 3 p3 an ank 4 p4
5Tb
x1 (t ) A, 0 t Tb binary 1
0 0 14 0 A 14 A 0 14 A A 14
x0 (t ) 0, 0 t Tb binary 0
A2
Possible levels = A, 0
Assume that values are equally likely to occur with probability Pi = 0.5
For k=0:
k=0
k0
anan
anan+k
00
00
11
01
Hence,
A2
2 , k 0
R(k ) 2
A , k 0
4
R(0) an an i pi
i 1
an an 1 p1 an an 2 p2
10
0 0 12 A A 12
11
Prof. Okey Ugweje
A2
2
229
Prof. Okey Ugweje
Example 24 Solution
A2Tb sin fTb

Gs f
4 fTb
sin fTb
t
Tb
fTb
Tb
j 2 fkTb
sin fTb
j 2 fkTb
Tb
R (k )e
k
fTb
A Tb sin fTb
4 fTb
A2Tb sin fTb
4 fTb
Prof. Okey Ugweje
1
e j 2 fkTb
k
1
f Tkb Fourier Series
Tb k
we can write
Using the fact that
Applying the formula

2
1
Gs f
X f R (k )e j 2 fkTb
k
Tb
230
Example 24 Solution
But
A2
2
A42
Since
2 e j 2 fkTb
k ,k 0
f Tkb
1 T k
b
sin fTb
0 @ f Tkb , k 0
fTb
we have
A2Tb sin fTb

Gs f
4 fTb
2
e j 2 fkTb e j 2 fkTb
k
k 0
A2Tb sin fTb

Gs f
4 fTb
231
Prof. Okey Ugweje
1 T f
b
232
Example 25
Example 26
Find the PSD of x(t) Unipolar RZ

A,
x2 (t ) 1
1
(
)
x
t
0
1
0
1
A
0,
0
Tb
2Tb
3Tb
0 t T2b
binary 1
Tb
2 t Tb
4Tb
Find the PSD of x(t) NRZ-L
x 3 (t )
A 1
x0 (t ) 0, 0 t Tb binary 0
-A
This is the same as Unipolar NRZ except for pulse duration

of Tb/2 instead of Tb
Hence
T sin f T2b
Xf b
2 f T2b
Gs f
A2Tb sin f T2b
16 f T2b
Prof. Okey Ugweje
(Left as an exercise. Please do)
1 1
f Tnb
Tb n
233
Comparison of Line Codes - 1
Prof. Okey Ugweje
Self-synchronization (SS)
Power Spectral Density comparison:
SS codes are good for error detection and correction
Different pulse shapes are used
Manchester codes have built in timing info because they

always have a zero crossing in the center of the pulse
Polar RZ codes tend to be good because the signal level
always goes to zero for the 2nd half of the pulse
NRZ signals do not have good SS capabilities
Error probability
Polar codes perform better (more energy efficient) than
Unipolar or Bipolar codes
Requires PSD of the line codes to determine channel
matching characteristics
to control the spectrum of the transmitted signal
(no DC value, bandwidth, etc.)
guarantee transitions every symbol interval to assist in

symbol timing recovery
After line coding, the pulses may be filtered or shaped

to further improve there properties such as
Spectral efficiency
Immunity to Inter-symbol Interference (ISI)
Distinction between Line Coding and Pulse Shaping is

not easy
Channel characteristics
Prof. Okey Ugweje
234
235
Prof. Okey Ugweje
236
Prof. Okey Ugweje
237
Prof. Okey Ugweje
DC Components
Summary
Unipolar NRZ, polar NRZ, and unipolar RZ all have DC

components
Bipolar RZ and Manchester NRZ do not have DC
components
First Null Bandwidth

Unipolar NRZ, polar NRZ, and bipolar all have 1st null
bandwidths of Rb = 1/Tb
Unipolar RZ has 1st null BW of 2Rb
Manchester NRZ also has 1st null BW of 2Rb, although
the spectrum becomes very low at 1.6Rb
Prof. Okey Ugweje
238
239
Timing
Error
Detection
Average
Power
Peak
Power
First Null
Bandwidth
AC
coupled
Transparent
Unipolar NRZ
Difficult
No
f0
No
No
Unipolar RZ
Simple
No
2f0
No
No
Polar NRZ
Difficult
No
f0
No
No
Polar RZ
Rectify
No
1/2
2f0
No
No
Bipolar NRZ
Difficult
No
2f0
Yes
No
Bipolar RZ
Simple
No
2f0
Yes
Yes
Dipolar NRZ
Rectify
Yes
f0
Yes
No
Dipolar RZ
Difficult
Yes
f0/2
Yes
No
HDB3
Rectify
Yes
f0
Yes
Yes
CMI
Simple
Yes
2f0
Yes
Yes
Prof. Okey Ugweje
240
Generation of Line Codes
Transmitter:
1
N
2N
ROM
0
1
3N
an
-1
+1
4N
binary
bits
Make
Impulse
5N
an [n]
impulse train which

represents the data
2N
3N
4N
5N
anp[n]
h[n] = p[n]
s[n]
Pulse Shaping
Inter-symbol Interference
pulse shape defined by impulse

response of FIR filter
The FIR filter realizes the different pulse shapes

Baseband modulation with arbitrary pulse shapes can be detected by
correlation detector
matched filter detector (this is the most common detector)
Prof. Okey Ugweje
241
Baseband Communication System
Prof. Okey Ugweje
Problems with Line Codes - 1
Baseband Communication System:
A) Line codes are not bandlimited
Input
A/D
Converter
an
Line
Coder
s(t )
Channel
Decoder
Transmiter
A/D
Converter
Output
We have been considering the transmitter side

Analog Input
an
Line
Coder
s(t )
Channel
absolute bandwidth, B, is infinite

power outside the 1st null bandwidth is not negligible
i.e., power in the sidelobes can be quite high
This can cause Adjacent Channel Interference (ACI)
Receiver
A/D
Converter
242
If transmission channel is bandlimited, then high freq

components will be cut off
To Receiver
High freq components correspond to sharp transition in

pulses
Hence, the pulse will spread out
If pulse spreads out into adjacent symbol period, then
inter-symbol interference (ISI) occurred
Transmiter
Transmitted signal is created by the line coder

according to
s(t ) an g (t nTb )
n
where an is the information sequence & g(t) is pulse shape

Prof. Okey Ugweje
243
Prof. Okey Ugweje
244
Illustration of ISI
B) Inter-symbol Interference (ISI)

ISI occurs when a pulse spreads out in such a way that
it interferes with adjacent pulses at the sample instant
Causes
Assume polar NRZ line code

data 1
data 0
1. Channel induced distortion which spreads or disperses

the pulses
2. Multipath effects (echo)
245
Tb
Tb
Tb
output
Tb
2Tb
3Tb
4Tb
5Tb
Tb
2Tb
3Tb
4Tb
5Tb
Prof. Okey Ugweje
246
Nyquist in the 1940s, studied the problem of ISI

He suggested that by carefully manipulating the filtering
characteristics of the channel (Tx and/or Rx), ISI can be control
Recall filter Characteristics
Channel output is the sum of the contributions from

each bit
0
2Tb
Tb
4Tb
3Tb
5Tb
Tb
2Tb
3Tb
4Tb
5Tb
A filter is a freq selective device used to limit the spectrum of signal to

some band of interest
Filters take an input waveform and modify the freq spectrum to produce
an output waveform
Filters are energy storing elements used as frequency discriminator
Some Notes on ISI

ISI can occur whenever a non-bandlimited line code is
used over a bandlimited channel
ISI can occur only at the sampling instants
Overlapping pulses will not cause ISI if they have zero
amplitude at the time the signal is sampled
Prof. Okey Ugweje
Tb
Tb
Strategies for Eliminating ISI - 1
Tb
Input data stream and bit superposition
3. Due to improper filtering (@ Tx and/or Rx), the received

pulses overlap one another thus making detection
difficult
Tb
input
Prof. Okey Ugweje
Tb
Filter Classifications
Ideal Filter:
A
X(f)
-B
Has a constant passband

Perfect rejection
No transition region
Filter is not physically realizable, only used for problem solving

247
Prof. Okey Ugweje
248
H( f )
LPF
-f1
f1
Ideal Filters
Low-Pass Filter (LPF)

High-Pass Filter (HPF)
Band-Pass Filter (BPF)
Band-Stop Filter (BSF)
For the ideal low-pass filter transfer function with bandwidth Wf =

fu hertz can be written as:
H( f )
HPF
f1
-f1
H ( f ) H ( f ) e j ( f )
(1.58)
H( f )
Where
BPF
-f2
-f1
f1
f2
f2
1
H( f )
0
H( f )
BSF
-f2
-f1
f1
for | f | fu
for | f | f u
(1.59)
Filter functions are implied in their respective names,
j ( f )
j 2 ft0
e.g., a LPF passes all freqs in the neighborhood of zero

Prof. Okey Ugweje
249
Ideal low-pass filter

Prof. Okey Ugweje
Ideal Filters
The impulse response of the ideal low-pass filter:
For the ideal band-pass filter

transfer function
h ( t ) 1 { H ( f )}
250
Ideal Filters
(1.60)
For the ideal high-pass filter

transfer function
H ( f ) e j 2 ft df
fu
e j 2 ft0 e j 2 ft df
fu
fu
e j 2 f ( t t0 ) df
fu
2 fu
sin 2 f u ( t t 0 )
2 f u ( t t 0 )
Ideal band-pass filter
2 f u sin nc 2 f u ( t t 0 )
Prof. Okey Ugweje
251
Prof. Okey Ugweje
Ideal high-pass filter
252

Frequency response of a typical filter is shown below:
Realizable Filters
H( f )
The simplest example of a realizable low-pass filter; an RC filter
H(f)
1 j 2 f
1 (2 f ) 2
H ( f ) max
j ( f )
0.707 H ( f ) max
Skirt of the filter
1/2-power
bandwidth, B
Stop Band
Transition
Band
f1
f2
Passband
f
Transition
Band
Stop Band
Such a filter is characterized by three regions:

1.Passband:
freqs in this band are transmitted with little or no attenuation
2.Stopband:
the freqs in this band are completely rejected
3.Transition band (roll off):

the gain of the freqs gradually falls off
Prof. Okey Ugweje
253
Realizable Filters
Realizable Filters
There are several useful approximations to the ideal low-pass filter

characteristic and one of these is the Butterworth filter
characteristic of RC filter
Hn ( f )
1 ( f / fu )2 n
Prof. Okey Ugweje
254
Phase
Prof. Okey Ugweje
255
Prof. Okey Ugweje
n 1
Butterworth filters are

popular because they
are the best
approximation to the
ideal, in the sense of
maximal flatness in the
filter passband.
256
Avoiding ISI
Nyquist suggested that the overall channel filter

transfer function (TF) must have a transition region
Use line code that is absolutely bandlimited

Cant actually do this (but can approximate)
Would require Sa(.) or sinc(.) pulse shape
Nyquist frequency response
Use a line code that is zero during adjacent sample

instants
Attn
Point of symmetry
It is ok for pulses to overlap somewhat, as long as there is no

overlap at the sample instants
Question: Is there pulse shapes that dont overlap during
adjacent sample instants?
Answer: Yes, e.g., Raised-Cosine Rolloff pulse
Frequency
1 1
fs
2 Ts
This TF should have a transition band between

passband & stopband and symmetric about a freq
equal to 0.5 x 1/Ts
Prof. Okey Ugweje
Use a filter at the receiver to undo the distortion

introduced by the channel
This is known as Equalization
257
Baseband Communication System Model - 1

x(t)
Transmitter s(t)
HT(f)
258

Channel
HC(f)
r(t)
Receiver y(t)
HR(f)
Note that he(t) is the equivalent impulse response of the

receiving filter
To recover the information sequence {an}, the output y(t)
is sampled at t = kT, k = 0, 1, 2,
The sampled sequence is
t = kT
n(t)
hT(t) = Impulse response of the transmitter

hC(t) = Impulse response of the channel
hR(t) = Impulse response of the receiver
s(t ) anhT t nT , where Ts n

Tb
n
y (kT ) an he kT nT ne (kT )
n
T = k/Tb
yk an hk n nk ho a k
r (t ) an gT t nT n(t ) h(t ),
where g (t )=hT (t ) hC (t ), Ts
y ( t ) an he t nT ne (t )
1
fs
an hk n nk
n ,n k
Desired symbol scaled

by gain parameters ho
h0 is an arbitrary constant
ISI terms - effect of other symbols at

the sampling instants t = kT
where hk ho (kT ), nk no (kT ), k 0, 1, 2,
where he (t ) hT ( t ) hC (t ) hR (t ), ne (t ) n( t ) hR ( t ) hC ( t )
noise term
or equivalently
Prof. Okey Ugweje
Prof. Okey Ugweje
259
Prof. Okey Ugweje
260
Signal Design for Bandlimited Channel
Generally, the optimum filter at the Rx is matched to

the received pulse he(t)
If the received signal is matched, then
Zero ISI
y (kT ) h a
o k
n ,n k
a h kT nT n (kT )
n e
e
To remove ISI, it is necessary and sufficient to make

the term he(kT - nT) = 0 for n k and h0 0
This means that
ho h 2 (t )dt
HR (f ) df HC (f ) HT (f ) df
By proper design of transmitting and receiving filters, it

is possible to satisfy the condition that he(kT - nT) = 0
for n k
This will eliminate the ISI term
RS
T
1, n 0
h (nT )
e
0, n 0
A pulse will produce zero ISI if it satisfies the following

condition:
h (t ) 0 at t kT k 0
e
Nyquist studied this problem many years ago

Prof. Okey Ugweje
261
Nyquist first method for zero ISI - 1
Prof. Okey Ugweje
262
A pulse will produce zero ISI at sampling instants if
Case I: Sampling at above Nyquist rate:
RS
T
1, n 0
h (nT )
e
0, n 0
H( f )
IK
n
H( f ) H f
T
e
n
T
2 fs
For channel bandwidth B, HC(f) 0, |f| > B and He(f) =

0 for |f| > B
Prof. Okey Ugweje
I
K
H(f) consist of non-overlapping replicas separated by fs = 1/T
provided that its Fourier Transform satisfy
FH
F
H
1
1
or 2 B
2B
T
263
fs
B
B fs
fs
B fs
2 fs
In this case, elimination of ISI is not possible. Why?

we cannot design He(f) to ensure that H(f) T
Prof. Okey Ugweje
264
The smallest value of T for which transmission with zero

1
ISI is possible is T
2B
Case II: Sampling at Nyquist rate:
FH
IK
1
1
or 2 B
2B
T
In this case, the pulses touch and almost begin to overlap
T
Problems with Sa(.) or Sinc(.) function

It is not possible to create Sinc pulses due to
H( f )
2 fs
fs
1.Infinite time duration

2.Sharp transition band in the frequency domain
2 fs
fs
There exist one He(f) for which H(f) T

1
1
t
f
f B
H ( f ) 2B
h (t ) sin c
e
e
2B
T
2B
0,
f B
R|
S|
T
Sa(.) pulse shape can cause ISI in the presence of timing

errors
FH IK
FH IK
signal is not sampled at exactly the bit instant, then ISI will occur
We seek a pulse shape that
Pulse shape that satisfy this criteria is Sa(.) or Sinc(.)

function, e.g.,
t
h (t ) sin c
sin c 2 Bt
e
T
Has a more gradual transition in the frequency domain

Is more robust to timing errors
Yet still satisfies Nyquists first condition for zero ISI
FH IK
Prof. Okey Ugweje
265
Prof. Okey Ugweje
Raised Cosine Pulse - 1

Case III: Sampling at below Nyquist rate
For fs > 2B, a particular pulse spectrum that has a desirable

spectral properties is the Raised Cosine (RC) spectrum
The following pulse shape satisfies Nyquists method for zero
ISI
t
sin t cos t
t cos T
T
T
h (t )
sin c
2 2
t
e
4 2 t 2
T 1 4 t
1
T
T2
T2
The Fourier Transform of this pulse shape is
F
H
1
1
or 2 B
2B
T
I
K
e j e j
In this case, pulses touch and overlap

H( f )
2 fs
fs
fs
2 fs
266
There are many He(f) for which H(f) T
R|T ,
|T L T 1 I O,
H ( f ) S M1 cos F f
e
|| 2 N H 2T K PQ
|T0,
FH IK e j
1
2T
1
1
f
2T
2T
1
f
2T
0 f
where is the roll-off factor that determines the bandwidth

(0 1)
Prof. Okey Ugweje
267
Prof. Okey Ugweje
268
BW occupied beyond 1/2T is called excess bandwidth (EB)

EB is usually expressed as a %tage of the Nyquist frequency,
e.g.,
= 1/2 ===> excess bandwidth is 50 %
= 1 ===> excess bandwidth is 100 %
RC filter is used to realized Nyquist filter since the transition
band can be changed using the roll-off factor
The sharpness of the filter is controlled by the parameter
When = 0 this corresponds to an ideal rectangular pulse
B occupied by a RC filtered signal is increased from its min
value
1
Bmin
2Ts
The Nyquist pulse shape can now be written as
to actual modulation bandwidth

Prof. Okey Ugweje
R|1,
1L
F a f f fIJ OP,
H ( f ) S M1 cosG
e
||T02,N H 2 f K Q
1
R|1,
F a f W 2W fIJ ,
H ( f ) Scos G
|T0, H 4 W W K
2
W - Wo
269
f1 f B
f1 2 f 0 B
f B f0
f B
Prof. Okey Ugweje
f 2W0 W
2W0 W f W
f W
270

Note that the bandwidth of a RC pulse shape is a
function of the bit rate and the rolloff factor
Rs
1
theoretical minimum BW
2 2T
W - Wo
= excess bandwidth, r =
rolloff factor, 0 r 1
Wo
absolute bandwidth, Wo
f1 2W0 W ,
f B f 0 , B f 0 f f 0 1 f 0 ( 1)
f
0
or solving for bit rate yields the expression
f W W0
A RC rolloff pulse shape is defined in this case by the rolloff

factor
W W0
f
r=
f0
W0
Rb
1.0
0.5
He ( f )
This means that to achieve zero ISI, it is necessary

sometimes to reduce the symbol rate below the Nyquist rate,
for practically realizable filters
f B f0 , f1 f0 f
Also see Fig. 3.17
f1
fo
B
2B
1
This is the max transmitted bit rate when an RC pulse

shape with rolloff factor is used over a baseband
channel with bandwidth B
where fo is the 6 dB bandwidth of the pulse

f1 and f are related to the pulse bandwidth B (or W) as follows
f1
f f1
Comparatively
fo
B
Prof. Okey Ugweje
Rb
1
2 2T
This is equivalent to equation 3.78, p. 139 in your text
where W
where f0
with Fourier Transform
B Bmin 1
f LMN a f OPQ
cos(2f t )
h (t ) 2 f0 Sa 2f0 t
e
1 4 f t 2
271
Prof. Okey Ugweje
272
Root RC rolloff Pulse Shaping

Later, we will show that the noise is minimized at the

receiver by using a matched filter
If the transmit filter is H(f), then the receive filter should
be H*(f)
The combination of transmit and receive filters must

satisfy Nyquists first method for zero ISI
H ( f ) H( f )H( f ) H( f ) H ( f )
e
e
Transmit filter with the above response is called the

root raised cosine-rolloff filter
Root RC rolloff pulse shapes are used in many
applications such as IS-54 and IS-136
Prof. Okey Ugweje
273
Practical Issues with Pulse Shaping - 1
Prof. Okey Ugweje
274
Practical Issues with Pulse Shaping - 2
Like the Sa(.) pulse, RC rolloff pulses extend infinitely in time

However, a very good approximation can be obtained by
truncating the pulse
Can make h(t) extend from -3Tb to +3Tb
RC rolloff pulses are less sensitive to timing errors than Sa(.)
pulses
Larger values of are more robust against timing errors
Sample Applications:
US Digital Cellular (IS-54/136) uses root RC rolloff pulse
shaping with = 0.35
IS-95 uses pulse shape that is slightly different from RC
rolloff shape
European GSM uses Gaussian shaped pulses
Implementation of Raised Cosine Pulse:

Can be digitally implemented with an FIR filter
Analog filters such as Butterworth filters may also
be used
Practical pulses must be truncated in time
Truncation leads to sidelobes - even in RC
pulses
Sometimes a square-root raised cosine spectrum
is used at Tx and Rx
This has to do with matched filtering
Prof. Okey Ugweje
275
Prof. Okey Ugweje
276
EYE Diagram - 1
EYE Diagram - 2
Effect of ISI and noise in digital communication can be

viewed on an oscilloscope from an eye diagram
ISI causes:
the eye to close thereby reducing the margin of error
distorts the position of the zero crossing, thereby causing the
system to be more sensitive to synchronization error
Effect of timing error is seen as a skewing of the eye diagram

and a closing of the eye due to the received symbol stream no
longer being sampled at the point of zero ISI
Width = time interval over which received signal can be sampled

Height = defines the noise margin of the system
Sensitivity to timing error = rate of closure of the eye
Diagram displays y(t) on vertical with horizontal sweep rate set to fs = 1/Ts
Prof. Okey Ugweje
The addition of noise affects the timing recovery circuitry and also causes a
general closing of the eye
Noise may occasionally causes full 'eye-closure'
277
EYE Diagram - 3
Prof. Okey Ugweje
278
Eye Diagrams for Raised Cosine Filtered Data - 1
With no
bandwidth
limitation
Small :
As is reduced, the eye opening narrows, requiring the
accuracy of symbol timing to be even more exact
overshoot caused by filtering is greater for small
With
bandwidth
limitation
This increases the peak-to-mean ratio of the data energy

Increases peak signal handling requirement of the
modulator/demodulator
A benefits of small is greater bandwidth efficiency

Prof. Okey Ugweje
279
Prof. Okey Ugweje
280
Eye Diagrams for Raised Cosine Filtered Data - 2


Large :
Simpler filter
fewer stages (or taps), hence easier to implement with
less processing delay
Less signal overshoot, resulting in lower peak to mean

excursions of the transmitted signal
Less sensitivity to symbol timing accuracy wider eye
opening
= 0 corresponds to Sa(.) function
Prof. Okey Ugweje
281
Controlled ISI
Controlling ISI
Partial Response Signaling
Duobinary Signaling
Prof. Okey Ugweje
282
Partial Response Signaling (PRS) - 1
To achieve zero ISI, we have seen that it is necessary to

transmit at below the Nyquist rate
Is it possible to relax condition on zero ISI and allow for
some amount of ISI in order to achieve fs > 2B?
The idea behind this is to introduce some controlled
amount of ISI instead of trying to eliminate it
ISI that we introduce is deterministic (or controlled) and
hence we can take care of it at the receiver
How do we do this?
A.k.a Doubinary signaling, Correlative coding, Polybinary

PRS is a technique that deliberately introduces some
amounts of ISI into the transmitted signal in order to ease the
burden on the pulse-shaping filters
It removes the need to strive at achieving Nyquist filtering
conditions, and high rolloff factors
This strategy involves two key operation
Correlative filtering
Digital precoding
Correlated filtering purposely introduces some ISI, resulting
in a pulse train with higher & correlated amplitude sequences
Nyquist rate no longer applies since the correlated
symbols are no longer independent
Hence higher signaling rate can be used
Controlled amount of ISI is introduced by combining a

number of successive binary pulses prior to transmission
Since the combination is done in a known way, the receiver
can be designed to correctly recover the signal
Prof. Okey Ugweje
283
Prof. Okey Ugweje
284
Digital
Precoding
ak
a' k
Impulse
Generator
x(t )
H(f)
y(t )
Regenerator
Since h(t) = sinc(t/T) and R=1/T, the overall impulse

response is
ak
FH
x (t ) ak t kT
y(t ) ak h t kT , where h(t ) F 1 H ( f )
and
RS
T
The transfer function H(f) is equivalent to the Tap

Delay line
T
x(t )
C0
C1
T
Cn-2
FH
N
t
y(t ) a c sin c n k
k k n0 n
T
IK UV ak sin cFH t kIK

T
W
k
where
Cn
Cn-1
a k co ak c1ak 1 c N ak N cn ak n
N
n0
+
LPF @
B = R/2
Prof. Okey Ugweje
IK
N
t
h(t ) c sin c n
n
n 0
T
y(t )
285
Prof. Okey Ugweje
286
Duobinary Signaling - 1

Partial Response signaling changes the amplitude

sequence ak a+k
a+k has a correlated amplitude span of N symbols
since each a+k depends on the previous N values of ak
Also, when ak has M levels, a+k sequence has M+ > M
levels
A whole family of Partial Response Signaling (PRS)
methods exists
Lets look at a few specific cases of PRS
Simplest form of PRS with M = 2, N = 1, Co = C1 = 1

Input sequence is combined with a 1-bit delayed
version of itself and then pulse-shaped
Duobinary Encoder
lx k q
yk
+
Delay
T
xk
He ( f )
H1
RS1,
T0,
xk 1
1
2T
1
2T
t kT
H2
1
2T
1
2T
if symbol a k = 1
if symbol a k = 0
yk xk xk 1
Prof. Okey Ugweje
287
Prof. Okey Ugweje
288
Each incoming pulse is added to the previous pulse

The bit or data sequence {yk} are not independent
Each yk digit caries with it the memory of the prior
digit
It is this correlation between digit that is considered
the controlled ISI which can be easily removed at
the receiver
Impulse Response of Duobinary Signal:
RST,
T0,
RS
T
it can be shown that (exercise show this)
289
Prof. Okey Ugweje
290
Impulse response h(t) for the duobinary scheme is

simply the sum of two sinc waveforms, delayed by one
bit period w.r.t each other:
He ( f )
arg H e ( f )
1
2T
Amplitude Response
1
2T
The role of the receiver is to recover xk from yk
1
2T
Duobinary Decoding:
yk
yk
2
xk
a A, 0, Af
+
xk1 -
Delay
T
Duobinary Decoder
Transmitted signal (assuming no noise) is yk xk xk 1

xk can assume one of 2 values A, depending on
whether the k-th bit is 1 or 0
Since yk depends on xk and xk-1, yk can have 3 values
(no noise)
Phase Response
Decision
Circuit
t kT
Duobinary signaling can be interpreted as adjacent

pulse summation followed by rectangular low pass
filtering
Encoder takes a 2 level waveform and produces a 3
level waveform
Prof. Okey Ugweje
e j
He ( f ) T e jfT e jfT e jfT
ej
1
2T
he (t )
From
f 21T
otherwise
Prof. Okey Ugweje
sin(t / T ) sin( (t T ) / T ) sin(t / T ) sin(t / T )
t / T
(t T ) / T
t / T
(t T ) / T
t
t T T 2 sin(t / T )
sin c
sin c
T
T
t(T t)
H1 ( f ) 1 e j 2fT
H2 ( f )
He ( f ) H1( f )H2 ( f ) 1 e j 2fT T T e jfT e jfT e jfT

jfT
, f 21T
2T cos(fT )e
else
0,
291
Prof. Okey Ugweje
292
In general, (M-ary transmission), PRS results in 2M1 output levels
R|2 A,
y S0,
k
|T2 A,
if the kth and (k -1)th bits are 1's
The detection process is the reverse of the transmitter

process
Major drawback
once errors are made, they tend to propagate through the
system
if the kth and (k -1)th bits are different

Detection involves subtracting xk-1 decisions from yk

digits such that
Decision rules is
xk yk xk 1
H1
lx k q
+
Delay
T
Decision
Circuit
xk 1
xk
xk 1
Delay
T
Duobinary Decoder
A Duo-binary Baseband System
293
Prof. Okey Ugweje
294
Advantage:
Duobinary signaling permits transmission at the
Nyquist rate without the need for linear phase,
rectangular shaped LPF
Disadvantages:
There is no one-to-one mapping between the
original binary digits and detected ternary symbol
(2 3)
Require more power
Ternary nature of duobinary signal requires about
3 dB greater SNR compared to ideal signaling
(i.e, binary) for a given PB
Prof. Okey Ugweje
y k
yk
t kT
Duobinary Encoder
0, decide that xk opposite of previous xk

y k
2, decide that xk 1
Prof. Okey Ugweje
LPF
295
Decoding process, xk = yk-xk-1, results in errors

propagation, Why?
output data bits are decoded using previous data
bit. If previous bit is in error, then the new output
will be in error, and so on
i.e., errors will propagate through the system
It is ineffective for AC coupled signal
AC coupling means that zero and low fred. data
are rejected
The PSD has substantial values at zero making it
unsuitable for AC coupled transmission
Prof. Okey Ugweje
296
Note:
Problem 3 can be solved with a technique known
as precoding
Problem 4 can be solved with a technique known
as modified duobinary
Duobinary Transfer
Function and pulse
shape
(a) Cosine Filter
(b) Impulse
response of the
cosine filter
Prof. Okey Ugweje
297
Prof. Okey Ugweje
298
Composite pulses arising from like and unlike

combinations of input impulse pair
Prof. Okey Ugweje
Duobinary waveform arising from an example binary

sequence
299
Prof. Okey Ugweje
300
Duobinary Precoding - 1
Duobinary Precoding - 2
A precoder consist of an exclusive-OR gate &

feedback through a one unit delay
The basic idea of precoding is that from the data

sequence {xk}, a new sequence {wk} (precoded
sequence) is generated
Unlike basic duobinary, precoding is nonlinear
The transmitted signal amplitude
wk
xk
wk1
Duo-binary
Encoder
yk
wk xk wk 1
Delay
T
1, if either xk or wk 1 is 1
w
k 0, otherwise
Conversion
rule
xk
wk-1
wk
wk+wk-1
RS
T
ak 1, if wk 0
y
ak 2 w k 1
k
ak 1,
if wk 1
At the receiver, the decoding decision rule is:
The binary stream wk is applied to the input of the

duobinary filter with output yk
yk wk wk 1
wk
Delay
T
yk wk wk 1
x w
wk1
Prof. Okey Ugweje
k 1
0, decide that xk 1
yk
2, decide that xk 0
0, if yk 2
i.e, x
xk 1 yk 1 mod 2
k 1, if yk 0
2
hw
k 1
301
Summary of Duobinary Baseband System - 1

+
Delay
T
LPF
y k
yk
Decision
Circuit
xk 1
Duobinary Encoder
Summary of Duobinary Baseband System - 2
Delay
T
Duobinary Decoder

if the kth and (k -1)th bits are different
k 1
0, decide that xk xk 1 opposite of prior decoded value

y k
2, decide that xk 1
Decision rules if precoding is used

y k
Decision rules is
In general, (M-ary transmission), PRS results in 2M-1 output

levels
Prof. Okey Ugweje
302
Detection involves subtracting xk-1 decisions from yk

digits such that x y x
xk
t kT
xk 1
R|2 A,
y S0,
k
|T2 A,
H1
lx k q
Prof. Okey Ugweje
303
Prof. Okey Ugweje
RS 0,
T2,
decide that xk 0
decide that xk 1
304
Modified Duobinary Signaling - 1


H ( f ) H ( f )H ( f ) b1 e
gT Tbe e ge
R2 jT sin(2fT )e , f
S
else
T0,
Also called class 4 signaling

Problem #4 (i.e, large DC value of duobinary PSD) can
be addressed by this signaling techniques
The encoder involves a two-bit delay, causing the ISI to
spread over two symbols (correlation span of 2 binary
H1
T
digits)
y
lxk q
Delay
2T
+
-
1
2T
xk2
From
Prof. Okey Ugweje
|ST ,
|T 0,
j 2fT
j 2fT
1
2T
it can be shown that (exercise show this)

sin(t / T ) sin( (t 2T ) / T )
t / T
(t 2 T ) / T
sin(t / T )
sin(t / T )
t / T
(t 2 T ) / T
he (t )
H2
H2 ( f )
j 2fT
He ( f ) T e j 2fT e j 2fT e j 2fT
Here again, we find that a 3 level signal is generated

Similarly
1
R
H1 ( f ) 1 e j 4fT
j 4fT
j 2fT
yk xk xk 2
1
2T
2T
otherwise
2T 2 sin(t / T )
t (2T t )
Spectrum shows a null @ zero but is still strictly bandlimited to 1/2T
305
Prof. Okey Ugweje
306
Characterization of PRS Systems - 1
Similar to basic duobinary, error propagation

necessitates the use of a precoding which is
implemented in a similar manner
For consistency, lets characterize the PRS systems
xk
wk 2
Delay
2T
wk
xk
xk
H1
yk
Delay
2T
wk
yk
duo
Mod.duo
wk
y k
x k
xk
1
2T
1
2T
xk 2
yk
wk
1, Duobinary
2, Modified Duobinary
H2
T , Duobinary
D
2T , Modified Duobinary
Prof. Okey Ugweje
307
Prof. Okey Ugweje
308
Duobinary:
Modified Duobinary:
a) Without Precoding: (wk = xk)
a) Without Precoding: (wk = xk)
yk xk xk 1
Decode : xk y k xk 1
Code :
1,
Output sequence : xk
0,
b) With Precoding:
Code : wk xk wk 1
Code : wk xk wk 2
yk wk wk 2 xk wk 2 wk 2
1, if y k 0
Decode xk
0, if y k 2
y k 1
else
b) With Precoding:
yk wk wk 1 xk wk 1 wk 1
Prof. Okey Ugweje
yk xk xk 2
xk yk xk 2
Code :
Decode :
Decode :
309
1, if y k 2
xk
0, if y k 0
Prof. Okey Ugweje
Examples
310
Multipath Channels - 1
Example: (Duobinary Coding)
Have already seen that bandlimited channel induce ISI

A good strategy was to pick a pulse shape that was
bandlimited and thus was not distorted by the
channel
It is also possible for a channel that is not bandlimited
to cause ISI, e.g., the multipath channel

Find the output sequence of duobinary signaling
system if the input data sequence is 1 1 0 0 0 1 0 1
00111
a) without precoding, b) with precoding
Antenna
Gain Pattern
Transmitter

Prof. Okey Ugweje
Direct Ray
Difused
Component
Specular
Component
Receiver
hc (t) (t 1 ) (t 2 )
311
Prof. Okey Ugweje
312
Multipath Channels - 2
If the direct path has time delay 1 and the reflected

path has time delay 2 (2 > 1) then the impulse
response of the channel is
H c ( f ) F (t 1 ) (t 2 ) e j 2 f 1 e j 2 f 2
The channels frequency response
Equalization
A plot of the magnitude response will not be flat!

Because the magnitude response is not flat, the signal
will undergo distortion, possibly resulting in ISI
It is therefore possible to encounter ISI even when the
channel itself has an infinite bandwidth
So, how do we handle this problem?

Prof. Okey Ugweje
313
Prof. Okey Ugweje
Equalization - 1
314
Equalization - 2
Nyquist filtering and pulse shaping schemes assumes

that the channel is precisely known and its
characteristics do not change with time
However, in practice we encounter channels whose
frequency response are either unknown or change
with time
Examples of time-varying channels are radio channels

These channels are characterized by time-varying
frequency response characteristics
To compensate for channel induced ISI and other
distortions, we use a process known as Equalization
a technique of correcting the frequency response of
the channel
The filter used to perform such a process is called an
equalizer
e.g., each time we dial a phone #, the communication

channel will be different because the communication
route will be different
But, when connection is made, the channel becomes
time-invariant (land line only)
The characteristics of such channels are not known a
priori
Prof. Okey Ugweje
Transmiter
hT(t)
Channel
hC(t)
Receiver
hR(t)
Equalizer
hEQ(t)
Noise
n(t)
315
Prof. Okey Ugweje
316
Equalization - 3
Problems with Equalization
Since HR(f) is matched to HT(f), we usually worry

about HC(f)
Goal is to pick the frequency response Heq(f) of the
equalizer such that
1. It can be difficult to determine the inverse of the channel

response
If the channel response is zero at any frequency, then the
inverse is not defined at that frequency
Rx generally does not know what the channel response is
H c ( f ) H eq ( f ) 1 H eq ( f )
with amplitude
H eq ( f )
1
j ( f )
e c
Hc ( f )
Channel changes in real time, so realistic equalization must be

adaptive
2. The equalizer can have an infinite impulse response even if

the channel has a finite impulse response
The impulse response of the equalizer must usually be
truncated
3. The equalizer can actually enhance the noise in the channel
Nonlinear equalization techniques are available that
minimize the amount of noise enhancement
1
Hc ( f )
and phase
eq ( f ) c ( f )
Prof. Okey Ugweje
317
Prof. Okey Ugweje
318
Linear Transversal Equalizer - 1
Equalization Techniques or Structures

Three Basic Equalization Structures

Linear Transversal Filter
This is simply a linear filter with adjustable parameters

Parameters are adjusted on the basis of the measurement of
channel characteristics
A common choice for implementation is the transversal filter
(Tap Delay Line (TDL)) or the FIR filter with adjustable tap
coefficient
xk
Simple implementation using Tap Delay Line or FIR filters

FIR filter has guaranteed stability (although adaptive
algorithm which determines coefficients may still be
unstable)
Decision Feedback Equalizer
k 1
Extra step in subtracting estimated residual error from

signal
C-N
C-N+1
CN-2
CN-1
Maximal Likelihood Sequence Estimator (Viterbi)

Optimal performance
High complexity and implementation problem (not heavily
used)
Prof. Okey Ugweje
CN
319
yk
Algorithm for
coefficient adjustment
Total number of taps = 2N+1

Total delay = 2NT = 2N
Prof. Okey Ugweje
320
N is chosen sufficiently large so that equalizer spans

length of the ISI
Assuming the ISI is limited to a finite # of samples, say
L, then 2N+1 > L
Output yk of the equalizer in response to the input
sequence {xk} is
Ideally, we would like the equalizer to eliminate ISI

resulting in
yk cn xk n ,
yk
Prof. Okey Ugweje
k 2 N , , 2 N
yk
RS1,
T0,
k0
k 1, 2,, N
There are two types of such equalizer (i.e., linear

equalizers)
321
Prof. Okey Ugweje
322
Preset Equalizer:
Transmits a training sequence that is compared at the
receiver with a locally generated sequence
Requires an initial training sequence
Differences between sequences are used to update the
coefficient cn
Time varying channel can change the sequence, since
the coefficients are fixed
Adaptive Equalizer:
Equalizer adjust itself periodically during transmission of
data
The tap weights constitute the adaptive filter coefficient
Prof. Okey Ugweje
k0
k0
But this cannot be achieved

However, the tap gains can be chosen such that
n N
where cn is the weight of the nth tap
RS1,
T0,
323
The 2 techniques can be combined into a robust

equalizer
In this case, there are two modes of operation
Training Mode
For the training mode, a known sequence is
transmitted and a synchronized version is
generated at the receiver
Decision-Directed Mode
When training mode is complete, the adaptive
algorithm is switched on
The tap weights are then adjusted with info from
training mode
Prof. Okey Ugweje
324
The impulse response of the transversal filter is
Nyquist zero ISI condition implies that

yk y (kT )
heq (t ) cn (t n )
N
1, k 0
cn x kT n
n N
0, k 1, 2, , N
n N
H eq ( f ) cn e j 2 fn
N
n N
If x(t) is the signal pulse corresponding to
Since there are 2N+1 coefficients, we may express in

matrix form as
X ( f ) = HT ( f ) HC ( f ) H R ( f )
y xc
then the equalized output signal is
where
y(t ) c x (t n )
n N n
Prof. Okey Ugweje
x = (2N+1) (2N+1) matrix with elements x(kT - n)

c = (2N+1) column coefficient vector
y = (2N+1) column vector
325
Prof. Okey Ugweje
Survey of Equalizers
Since this design forces the ISI to be zero at sampling

instants t = kT, the equalizer is called Zero-Forcing
Equalizer (ZFE)
Thus we obtain a set of (2N+1) linear equations for
ZFE
In the figure, is chosen as high as T
= T Symbol-spaced equalizer;
< T Fractional-spaced equalizer
Equalizer
Types
327
Linear
Nonlinear
ML Symbol
Detector
DFE
Structures
Algorithms
Prof. Okey Ugweje
326
Prof. Okey Ugweje
Transversal
Lattice
Zero Forcing
LMS
RLS
Fast RLS
Square Root RLS
Transversal
Gradient RLS
Lattice
LMS
Gradient RLS
RLS
Fast RLS
Square Root RLS
MLSE
Transversal
Channel Estimator
LMS
RLS
Fast RLS
Square Root RLS
328
Examples: (Equalizer/Equalization)
Decision Feedback Equalizer
Example: Equalization
A Decision-Feedback Equalizer (DFE) is a nonlinear equalizer

that employs previous decisions to eliminate the ISI caused by
previously detected symbol
It consists of a feed forward section a feedback section and a
detector connected together as shown
Problem
Example: Equalizer/Equalization
Input from
matched
filter
Example: Equalization
Feedforward
Filter
zm
+
-
Feedback
Filter
Example: Equalizer/Equalization
Problem
output
data
Detector
z m
The filters are usually fractionally spaced FIR with adjustable tap coefficients
The detector is a symbol-by-symbol detector
Note
self study
Prof. Okey Ugweje
329
Maximum Likelihood Sequence Detector (MLSD)

Prof. Okey Ugweje
330

This technique provides an algorithm for searching through the trellis for the
ML signal path
A trellis is a schematic used to represent signal waveforms with memory,
e.g., the trellis for duobinary PRS is given by
1/2
-1
1/2
1
t0
0
1
tT
1/2
new data bit/received signal level
Module 3
0
1
t 2T
t 3T
For binary, this trellis contains 2 states corresponding to 2 possible input values
Since the duobinary have memory of length L = 1, the number of states is S = 2L
In general, for M-ary, the number of trellis states is S = ML
Maximum Likelihood Sequence Detector selects the most probable path through
the trellis upon observing the received sequence y(kT)
In general each node in the trellis will have M incoming paths and M metrics
Search through the trellis for the minimum distance may be performed sequentially
using Viterbi algorithm - beyond the scope of this class!
Prof. Okey Ugweje
331
Baseband Communication System
Prof. Okey Ugweje
332
Noise on Communication Systems

In the process of communication, noise arises in

various forms
m(t) is corrupted in the transmitter by thermal noise due
to the presence of electronic devices (e.g., Audio
Amplifier)
c(t) is not a pure sine wave - in fact, it contains
harmonic distortions
s(t) experiences multiplicative noise in the process of
being transmitted thru the channel due to turbulence in
the air, reflection, refractions, multipath etc.
s(t) also suffers from additive noise during transmission
(passing automobiles, static electricity, lightning, power
lines, sunspots, etc)
thermal and short noise at the receiver
Noise in Communication System

n(t)
x(t)
yi(t)
s(t)
Transmitter
input
m(t)
Prof. Okey Ugweje
PT
Channel
y0(t)
Receiver
Si, Ni
output
S0, N0
333
Prof. Okey Ugweje
Noise Modeling - 1
334
Noise Modeling - 2
All these different noise components degrade the

performance of communications system
Among these types of noise, the additive noise is the
most annoying
usually contains most power and is of most interest
in many applications
In the channel, the signal experience attenuation, time

delay (precisely known) and additive noise
Transmitter
Channel
(modulated signal ) s(t)
Receiver
r(t) (received signal )
Most disturbances, interference, attenuation, etc., are

usually classified as noise
The most important type of noise that occur in
communications system is said to be white noise,
n(t)
Usually n(t) is assumed to be Additive, White and a
Gaussian Noise (AWGN) with power spectral density
Gn(f)
n(t)
(noise)
Prof. Okey Ugweje
335
Prof. Okey Ugweje
336
White Noise and Filtered Noise - 1
White Noise and Filtered Noise - 2
White Noise is a random process having a flat

(constant) power spectral density Gn(f), over the entire
frequency range
This type of noise is wideband and cannot be expressed

in terms of quardrature components
However, in most communications systems operating at
carrier frequency fc, the bandwidth of the channel B (or
W), is small compared to fc narrowband systems
In such situations, it is mathematically convenient to
represent the white noise process in terms of the
quadrature components
white because it is analogous to white light

assumed to be a Gaussian random process
Gn (f)
usually additive in nature
N0
2
2-sided power spectral density of noise
Hence this type of noise is commonly called Additive,

White and Gaussian (AWGN) with power spectral
density such that G ( f ) N 0
n
Prof. Okey Ugweje
Accomplished by passing signal plus noise at the receiving

terminal through an ideal BPF having a passband as
Gn (f)
-fc
337
Prof. Okey Ugweje
338
y0 (t ) s0 (t ) no (t )
Assume that:
n(t)
x(t)
yi(t)
s(t)
Transmitter
PT
Channel
Si, Ni
Noise n(t) is zero-mean Gaussian with PSD Gn(f) = N0/2 or

/2
y0(t)
Receiver
output
S0, N0
A certain signal m(t) (or x(t)) is transmitted with power PT

s(t) is corrupted by additive noise n(t) during transmission
Channel may also attenuate (and/or distort) the signal
At receiver, we have a signal mixed with noise
Signal and noise power at the receiver input are Si and Ni
Receiver processes the signal (filters, demodulation, etc.)
to yield the desired signal power So, plus noise power No
Prof. Okey Ugweje
fc
SNR is the figure of merit for evaluating the performance

of analog communications systems
m(t)
Signal-to-Noise Ratio (SNR) - 2
Signal-to-Noise Ratio (SNR) - 1

input
N0
2
339
Noise is uncorrelated with s(t)

Hence output power is
E y02 (t ) E s02 (t ) E n02 (t )
S0 N 0
The output signal-to-noise ratio (SNR) is

SNR0
Prof. Okey Ugweje
E s 2 (t )
S0 S
02
N 0 N 0 E n0 (t )
340
Baseband System Model - 1
Baseband System Model - 2
In baseband systems, signal is transmitted w/o modulation

and we also assume that channel is distortionless, hence
Assumes:
x0 t x0 t td
This mode of communication is used in short-haul links

over a pair of wires or coaxial cable
Although this mode of communication is not widely used,
their study is important because many of the basic
concepts can be carried over to modulated systems
Also, baseband systems are used as benchmark for
comparing the performance of analog systems
LPF
m(t)
ST
Hp(f )
input
Channel
HC ( f )
Si
LPF
So
Ni
Hd ( f )
N0
n(t ) Noise
limits m(t)
Si PT N , where N 2 0 Gn ( f )df
B
Signal-to-noise ratio is then given as

SNR Mean Signal Power So
No
Noise Power
Therefore, for a baseband system,
y D (t )
S
S
SNR i
N b N 0 B b
eliminates outof-band noise
A baseband Communication System Model

Prof. Okey Ugweje
m(t) is zero-mean, wide sense stationary random

process bandlimited to B Hz
Assume that the channel is distortionless with unit gain,
341
Prof. Okey Ugweje
With Gain - 1
S 0 Si
n(t )
gT
ST
SX
Channel
L
xR (t )
SR
gR
LPF
x0 (t ) n0 (t )
S0 N 0
Receiver
ST gT x 2 gT S X
S R xR2
ST
L
S0 x02 g R S R
N 0 output g R N 0 B
S SR

N o N 0 B
Receiver output SNR does not depend on the gain, gR

However, channel gain or losses will affect the output
N 0 E n02 (t ) B Gn ( f )df B N20 df N 0 B

B
Therefore
S S 0 Si

N 0 N 0 B
Larger value of SNR is desirable

This can be achieved by simply increasing PT
However, this is usually not possible since in practice,
(PT)max is limited by other considerations such as
FCC (NCC) rule; transmitter cost; channel capacity;
interference with other channels, and so on
In practice, it is more convenient to deal with received

signal power Si instead of PT
S ST

N o LN 0 B
Prof. Okey Ugweje
342
With Gain - 2
x(t )
This is used as a
standard for making
comparisons of the
various analog
modulation schemes
343
Prof. Okey Ugweje
344
Binary Signal Transmission - 1
In a binary commun. system, binary data (0s, 1s) are

transmitted by means of 2 signal waveform s0(t) & s1(t)
0 s0(t), 0 t Tb
where Tb = 1/Rb
1 s1(t), 0 t Tb
Assumptions:
data bits 0 & 1 are equally probable (each has
probability 0.5)
0 and 1 are mutually independent
The channel corrupts the signal by adding noise,
denoted by n(t)
n(t) is assumed to be Additive White Gaussian
Noise with PSD N0/2 W/Hz
Detection of Binary Signal in

Gaussian Noise
Prof. Okey Ugweje
345
Binary Signal Transmission - 2
Prof. Okey Ugweje
346
Detection of Binary Signal in Gaussian Noise - 1
The received signal waveform is expressed as

r(t) = si(t) + n(t), i = 0, 1; 0 t Tb
Receiver is to determine whether a 0 or a 1 was
transmitted
si (t )
r (t )
n(t )
h(t)
z(t )
z(T )
t T
H1
si (t )
H0
(AWGN)
Recovery of signal at the receiver consist of 2 parts

Signal correlator or Matched filter
reduces received signal to a single variable z(T)
z(T) is called the test statistics
Analysis that follow will assume that the filtering

operation is linear
linear input linear output
Gaussian output
Gaussian input
Detector (or decision circuit)

compares the z(T) to some threshold level 0, i.e.,
z (T )
H
1
H
0
where H1 and H0 are the two possible binary hypothesis

Prof. Okey Ugweje
347
Prof. Okey Ugweje
348
Detection of Binary Signal in Gaussian Noise - 2
Maximum Likelihood Detector (MLD) - 1
Signal correlator and detector processes are

independent
Once r(t) is transformed to z(T), the shape of the
waveform is no longer important
The concept of maximum likelihood detector is based

on Statistical Decision Theory
It allows us to
This means that any kind of transmitter waveform

transforms to z(T) for detection purposes
Hence, detection for baseband and bandpass are the

same
A particular detector that minimizes the probability of
error is known as the maximum likelihood detector
That is, it minimizes the cost of making an error
Prof. Okey Ugweje
formulate hypothesis that characterizes the transmission

test the hypothesis
formulate the decision rule that operates on the data
optimize the detection criterion
The formulation of this topic requires the knowledge of

probability (in particular Bayes rules) and random
variables
For a binary data stream there are two types of decision
Soft decision (multi-level)
Hard decision (2 level)
349
Prof. Okey Ugweje
350
Hard decision is more common than soft decision
Each soft decision contains

Information about the most likely transmitted signal
000 to 011 0
100 to 111 1
Information about the likelihood of a decision
Soft decisions are converted to hard decisions by
some algorithm
Let T be the length of time it takes to transmit one bit
of data
Decides immediately whether the signal is 0 or 1

Uses either Bayes decision criterion or Newman-Pearson
criterion
Matched
Filter
S&H
8-level 3-bit
quantization
Combined Soft decision/

error control decoding
soft decision
hard decision
a) Soft decision Receiver

Matched
Filter
S&H
Binary
quantization
Error control
hard decision
b) Hard decision Receiver

Digital 0
Digital 1
000 010 100 110

000 010 100 110
0
Prof. Okey Ugweje
s (t ), 0 t T for a binary 0
s (t ) 0
s1 (t ), 0 t T for a binary 1
soft decision
hard decision
351
Prof. Okey Ugweje
352
At the output of the demodulator
z(T) is known as decision variable or test statistics

and it is a random process corrupted by noise
Assume that pdf of z0(T) and z1(T) are Gaussian with
equal likelihood, and with 0 = a0, 1 = a1
z (t ) a0 (t ) n(t ), 0 t T for a binary 0

z (t ) 0
z1 (t ) a1 (t ) n(t ), 0 t T for a binary 1
where ai(t) is the signal component & noise n is zero mean
Gaussian
Region 0
Likelihood of s0
Decision
Line
P[z|s0 sent]
At the sampling instant t = T

z (T ) a0 (T ) n(T ), 0 t T for a binary 0
z (T ) 0
z1 (T ) a1 (T ) n(T ), 0 t T for a binary 1
p(z| s0 )
353
Prof. Okey Ugweje
P[z|s1 sent]
a0
Pe(s0)
L F IO
expM 1 G z a J P
2
N 2H K Q
2
p(z| s1)
Minimum error criterion

a0
For simplicity we will drop the index such that z = ai + n
Prof. Okey Ugweje
Region 1
Likelihood of s1
0 a0 a1 0
354

This is an averaging operation

It makes sense because the logical point is halfway
between the two voltage levels representing each
symbol
Questions:
How do we implement this averaging operation?
How do we choose the threshold, 0?
Hypothesis:
H0: r(t) = s0(t) + n(t) 0 sent
H1: r(t) = s1(t) + n(t) 1 sent
Definitions of Probabilities:
P[s0], P[s1] a priori probabilities
These probabilities are known before transmission
P[z]
probability of the received sample
p(z|s0), p(z|s1)
conditional pdf of received signal z, conditioned on the class si
P[s0|z], P[s1|z] a posteriori probabilities

After examining the sample, we make a refinement of our previous
knowledge
P[s1|s0], P[s0|s1]
wrong decision (error)
P[s1|s1], P[s0|s0]
IJ OP
KQ
Prof. Okey Ugweje
LM F
N GH
1 exp 1 z a1
1 2
2 1
355
Prof. Okey Ugweje
correct decision
356
Types of Decision Rules - 1

Decision Rule:
H1
z(T )
H0
zi (t )
Acquiring information at the receiver about the transmitted

signal involves making decisions
We must decide which of the set of hypothesis best
describes the received signal
This involves uncertain (error in judgment)
If the signals we are trying to detect do not overlap, we
can make a decision without error
On the contrary, we need some rules to help classify the
received signal once they fall in the overlap region
A set of rules known as decision rules allow us to decide
Prof. Okey Ugweje
357
1. Bayes decision criterion:

It formulates the problem of making a decision
under conditions of uncertainty by selecting the
hypothesis with the greatest a posteriori probability
This scheme assumes that some errors are more
costly than others
Hence, it assigns cost (weighting factors) that
reflect the risk involved
This is the most widely applied decision rule in
communications
Prof. Okey Ugweje
358
2. Maximum a posteriori (MAP) criterion:

Decide that the received signal belongs to the class
with the maximum a posteriori probabilities, i.e.,
maximize P(si|z)
It equivalently examines the pdf conditioned on
each signal class (p(z|s0), p(z|s1)) and choose the
maximum
For the received signal za, the likelihood that za
belongs to s1 or s2 corresponds to the circled point
on the pdf
The decision criterion is based on the likelihood of
P[z|si], i = 0,
3. Newman-Pearson (N-P) criterion

Makes no assumption on the a priori source
statistics (requires only a posteriori probabilities)
Widely used in pulse detection in Gaussian noise as
in Radar applications where the source probabilities
(presence or absence of a target) is unknown
fix probability of false alarm
minimize probability of error
maximize probability of correct decision
Prof. Okey Ugweje
Prof. Okey Ugweje
359
360
Bayes Decision Criterion - 1
4. Min-max criterion
Also in this criterion, the a priori probability is not
known
Since P(H1) is unknown, the rule maximizes the risk
with respect to P(H1) and minimizes the risk with
respect to P(H0)
Recall that the Bayes equation is given by

P[ si | z ]
where
P[ z | si ]P[ si ]
, i 0, 2, , M 1
P[ z ]
M
P[ z ] P[ z | si ]P[ si ]
i 1
Recall from probability theory that
P[ si | z ]P[ z ] p ( z | si ) P[ si ]
In communications, we can interpret the Bayes

equation as a description of an experiment involving
a received sample, and
a statistical knowledge of the signal classes to
which the received sample may belong
Prof. Okey Ugweje
361
Prof. Okey Ugweje
That is,
si denote the ith transmitted signal class from a set
of M classes
zj denotes the jth sample of the received signal
Hence, we can write the Bayes equation in terms of
the pdf
By examining a particular received sample zj, it is

possible to find likelihood that zj belongs to class si
P[ si | z ]
362
p ( z | si ) P[ si ]
, i 0, 2, , M 1
p( z )
This means that after the experiment, we will refine our

knowledge by computing the a posteriori probability
Note that the terms a priori and a posteriori imply

cause to effect and effect to cause, respectively
Assume that
Pdf of z0(T) and z1(T) are Gaussian with equal
likelihood, having mean values of a0 and a1 respectively
a0 and a1 are mutually independent
Noise n0 is independent zero mean AWGN with PSD No
where
M
p ( z ) p ( z | si ) P[ si ]
i 1
Prof. Okey Ugweje
363
Prof. Okey Ugweje
364
In this case, for binary signal
L( z )
P[ s1 | z ] P[ s0 | z ] decision rule
The right-hand side (RHS) is called the likelihood ratio

When the two signals, s0(t) and s1(t), are equally likely,
i.e., P[s0] = P[s1] = 0.5, then the decision rule becomes
p ( z | s1 ) P[ s1 ] p ( z | s0 ) P[ s0 ]
p ( z | s1 ) P[ s1 ] p ( z | s0 ) P[ s0 ]
p( z )
p( z )
L( z )
p ( z | s1 ) P[ s0 ]
likelihood ratio test (LRT)
p ( z | s0 ) P[ s1 ]
L( z )
365
Prof. Okey Ugweje
366
Substituting the pdfs
Taking the log of both sides will give
1 z a
0
exp

0 2
2 0
1 z a 2
1
1
H1 : p ( z | s1 )
exp

1 2
2
1
1
1
2
exp 2 ( z a1 )
p ( z | s1 )
2
2 1

L( z )
1 1
1
1
p ( z | s0 )
1 ( z a )2
exp 2 0
0 2
2 0
ln{L( z )}
z (a1 a0 ) a12 a02 (a1 a0 )(a1 a0 )
2 02
02
2 02
2 (a a )(a a )
z 0 12 0 1 0
2 0 (a1 a0 )
(a1 a0 )
z
0
z (a a ) (a 2 a 2 )
exp 1 2 0 1 2 0 1
2 0
0
z (a1 a0 ) (a12 a02 )

0
2 02
02
Hence
02 12
Prof. Okey Ugweje
max likelihood ratio test
H 0 : p ( z | s0 )
p ( z | s1 )
1
p ( z | s0 )
In terms of the Bayes criterion, it implies that the cost of

both types of error is the same
This type of decision rule is called the maximum a
posteriori (MAP) criterion (or minimum error
criterion)
The last equation corresponds to making a decision

based on the comparison of received signal to some
threshold level
Prof. Okey Ugweje
p ( z | s1 ) P[ s0 ]
likelihood ratio test (LRT)
p ( z | s0 ) P[ s1 ]
where z is minimum error criterion and 0 is optimum threshold

367
Prof. Okey Ugweje
368
Probability of Error - 1
For antipodal signal, s1(t) = - s0(t) a1 = - a0
Error will occur if

s1 is sent s0 is received
z 0
ao
P[H0|s1] = P[e|s1]
a1
P[e | s1 ] 0 p ( z | s1 ) dz
This means that if received signal was positive, s1(t)

was sent, else s0(t) is sent
s0 is sent s1 is received
ao
a1
P[H1|s0] = P[e|s0]
P[e | s0 ] p ( z | s0 )dz
0
See pp. 121~122 & section

B.2
The total probability of error is the sum of the errors

2
PB P (e, si ) P[e | s1 ]P[ s1 ] P[e | s0 ]P[ s0 ]

i 1
P[ H 0 | s1 ]P[ s1 ] P[ H1 | s0 ]P[ s0 ]
Prof. Okey Ugweje
369
Prof. Okey Ugweje
1 z a 2
0
exp
PB
dz
0
2
0
0
( z a0 )
u
, 0 du dz
If signals are equally probable
P[ H 0 | s1 ] P[ H1 | s0 ]
by symmetry
PB 12 P[ H 0 | s1 ] P[ H1 | s0 ] P[ H1 | s0 ]
( a a ) / 2
Hence, PB, is probability that an incorrect hypothesis is

made
Think of PB as the area under the tail of either of the
conditional distributions, p(z|s1) or p(z|s2), i.e.,

Prof. Okey Ugweje
a a
PB Q 1 0
2 0
1 z a 2
0
exp
dz
2
2
0
1
2
exp u du A *
2
2
This equation cannot be evaluated in closed form

This is the famous Q-function or complementary error
function
Hence,
PB p ( H1 | s0 )dz p ( z | s0 )dz
PB P[ H 0 | s1 ]P[ s1 ] P[ H1 | s0 ]P[ s0 ]
1
2
370
371
Prof. Okey Ugweje
372
Pe is minimized by choosing h(t) or H(f) such that optimum

threshold 0 is minimized
That is
a (t ) a1 (t )
[a (T ) a1 (T )]2
or 02 0
0 0
2
A note on the Q(x) - complementary (co) error function

Equivalent Definitions
Correlator
1
x
e rfc
2
2
e rfc x 2Q x 2
Q( x)
For large arguments (x large), Q function

Q( x)
1
2
exp x
x 2
2
Prof. Okey Ugweje
373
Prof. Okey Ugweje
Correlator-Type Receiver - 1
374
The correlator cross-correlates r(t), with the 2

possible transmitted symbols s0(t) and s1(t)
The detector compares z1 and z0 and decides that

1 was transmitted if z1 > z0
0 was transmitted if z1 < z0
So when s1(t) is transmitted,
PB = P[z0 > z1] = P[n0 > E + n1] = P[n0 -n1> E]
Let x = n0 - n1
z ()dt
T
r (t )
z0 (t )
tT
z0 ( T )
Threshold
Detector
s0 ( t )
z ()dt
T
z1(t )
si ( t )
z1 ( T )
s1 ( t )
Output for either z0 or z1 is given by

z0 (T ) 0 r (t ) s0 (t )dt
T
E x 0 zero mean
z1 (T ) 0 r (t ) s1 (t ) dt
T
This cross-correlation process basically computes the

projection of r(t) into 2 basis functions s0(t) and s1(t)
The outputs z0 and z1 are then feed to the Threshold Detector
Prof. Okey Ugweje
375
0, orthogonal
E x E n0 n1 E n E n 2 E n0 n1
N N
Noise Variance
2 E n 2 (t ) 2 0 0 n2
4 2
2
Prof. Okey Ugweje
2
0
2
1
376
Hence PB is
Other Forms of the Correlator
x
1
exp 2 dx
P
E
B 2
2 x
x
2
Form 1:
x
x2
1
exp
dx
E
2
2
E
Q
N0
z ()dt
T
s0 ( t )
r (t )
z0 (t )
z ()dt
T
z(T )
si ( t )
tT
z1 (t )
s1 ( t )
A similar procedure can be used to derive the PB
Form 2:
r (t )
z () dt
si (t )
s1(t ) s0 (t )
t T
observe the correlating signal given by s1(t)-s0(t)

Prof. Okey Ugweje
377
Prof. Okey Ugweje
378
Suppose there are many signals si(t), i = 0, 2, , M-1,

the received signal can be correlated using a bank of
correlators
x
z ()dt
T
tT
z0 ( T )
s0 ( t )
r (t )
z ()dt
T
s1 ( t )
z1 ( T )
z ()dt
T
Selects
si(t) with
the
max zi(t)
Matched Filter
si ( t )
z (T )
M 1
s (t )
M 1
Prof. Okey Ugweje
379
Prof. Okey Ugweje
380
Matched Filter Receivers - 1
A matched filter is a linear filter that optimizes the

SNR for a symbol
i.e., maximizes the SNR at the output for a given
transmitted symbol waveform
Given r(t) = s(t) + n(t) at the input, we want to find the
filter characteristics h(t) or H(f) that maximizes the
output SNR
A filter that is matched to the waveform s(t), has an

impulse response
h(t) = s(Tb-t), 0 t Tb
s(t )
r (t )
h(t ) s(Tb t )
z(T )
z(t )
Notice that h(t) is a delayed version of the mirror

image (rotated on the t = 0 axis) of the original signal
waveform
E.g.,
s(t )
si (t )
t T
n(t )
Tb
signal
Prof. Okey Ugweje
h(t ) s(Tb t )
s(t )
381
Prof. Okey Ugweje
Tb
image
Tb
signal delayed by Tb
382
This is a causal system

a system is causal if before an excitation is applied
at time t = T, the response is zero for - < t < T
Signal waveform at the output of the matched filter is
z (t ) 0 r ( )h(t )d
convolution
0 r ( ) s (Tb t )d
t
If we sample z(t) at t = Tb, we obtain
z (Tb ) z (t ) t T 0 b r ( ) s ( )d
T
Important Property of Matched Filter:

If s(t) is corrupted by an AWGN, the filter with impulse
response h(t) maximizes the SNR
To prove this let
r(t) = si(t) + n(t), t0 t t0+Tb , i= 0,1
S(f) = Fourier Transform of s(t)
H(f) = Transfer function of the filter h(t)
For MF, we want to determine h(t) or H(f) that
maximizes output SNR
Hence the sampled output of the filter at time t = T is

exactly the same as the output of the correlator
Prof. Okey Ugweje
383
Prof. Okey Ugweje
384
But
Time Domain Analysis:

h(t)
r(t)
2
E yn2 (T ) E 0T n( )h(T )d
y(t)
y (t ) 0 r ( )h(t )d
t
E yn2 (T ) 0 0 E n( )n(t ) h(T )h(T t )dtd

N T T
0 0 0 t h(T )h(T t )dtd
2
N0 T 2
h (T t )dt
2 0
T T
0 s ( )h(t )d 0 n( )h(t )d
t
sample at t T
0 s ( )h(T )d 0 n( )h(T )d
ys (T ) yn (T )
T
2
S ys (T )

N T E yn2 (T )
Prof. Okey Ugweje
noise variance
385
386
The noise variance depends on the PSD of the

noise and the energy in the impulse response, h(t)
2
T
T
S 0 s ( )h(T )d 0 h( ) s (T )d

N0 T 2
N0 T 2
N T
0 h (T t )dt
h (T t )dt
2
2 0
Hence by replacing x(t) = h(t), y(t) = s(T-t)

2
2
S 0 h ( )d 0 s (T )d

N T 2
0
N T
0 h (T t )dt
T
From Cauchy-Schwarz inequality, we know that

2

x(t ) y (t )dt x(t ) dt y (t ) dt
with equality when x(t) = ky(t), k = constant

We can maximize this expression by holding the

denominator constant and then optimizing the numerator
Prof. Okey Ugweje
Prof. Okey Ugweje
387
It is clear here that SNR is maximum when h(t) =

ks(T-t)
2 2
2
2 T 2
0 k s (T t )dt 0 s (T )d
S
s (t )dt

N0 T 2 2
N0 0
N T
k
s
(
T
t
)
dt
2 0
2E
N0
Prof. Okey Ugweje
388
Substituting
Frequency Domain Analysis:
The numerator is of the form
H ( f )Y ( f )df
S a (t )

N T E n 2 (t )
E n (t ) Rn (0) Gny ( f ) df
N0
2
Gny ( f ) H ( f ) Gnx ( f )
H ( f ) df
Prof. Okey Ugweje
H ( f )Y ( f )df
H ( f ) Gnx ( f )df
If written with Cauchy-Schwartz inequality we have
eqn. 1.53
389
Prof. Okey Ugweje
max at 0
390

Hence
j 2 ft
df
S 2 H ( f ) S ( f )e

2
N T N 0
H ( f ) df
When the signal is matched, it means that the transfer

function achieves the equality condition, i.e.,
2E
S
max
0 max
N T N 0
Parsavals theorem
This also means that the optimum choice of H(f) is
S ( f ) df s 2 (t ) dt E Energy of the signal

S 2 S ( f ) 2 df 2 E

N0
N T N 0
This is the maximum SNR

It depends on signal energy E and noise PSD
Does not depend on signal waveform
2 H ( f ) df S ( f ) df
N0
H ( f ) df
2
2
S ( f ) df
N0
Prof. Okey Ugweje
Hence
Equality holds iff H(f) = KY*(f)
H ( f ) df Y ( f ) df
H ( f ) df S ( f ) df
eqn. 1.42
But
where Y(f) = S(f)ej2ft
but the denominator is the noise variance

2
ai (t ) FT 1 H ( f ) Si ( f ) H ( f ) Si ( f )e j 2 ft df
Therefore
j 2 ft
dt
S H ( f ) S ( f )e

N
2
0
N T
H ( f ) df
Since z(t) = a(t) + n0(t), where a(t) is the signal

component, we can write
H 0 ( f ) H ( f ) kS ( f )e j 2 fT
391
Prof. Okey Ugweje
392
This implies that
Similarly, bank of Matched filters is used to receive

several signals
h(t ) F 1 H ( f ) kS ( f )e j 2 fT e j 2 ft df
kS ( f ) e j 2 f (T t ) df
h(t ) s0 (Tb t )
z0 (t )
z0 ( T )
kS ( f )e j 2 f (T t ) df
h(t ) s1(Tb t )
z1(t )
z1 ( T )
r (t )
ks T t ks T t
M 1
tT
The impulse response of the M matched filters are

given by
hk (t ) ksk (Tb t ), o t T
h(t ) ks T t
Prof. Okey Ugweje
si ( t )
z (T )
M 1
M 1
ks T t ks T t
Thus
h(t ) s (Tb t ) z (t )
If the signal is real, then
Selects
si(t)
with
the
max zi(t)
where sk(t) are the set of basis function
393
Summary of Matched Filters
Prof. Okey Ugweje
394
Correlator vs. Matched Filter - 1
A Matched filter is a detection filter that optimizes the

output SNR
The functions of the correlator and matched filter

are the same
s(t )
r (t )
h(t ) s(Tb t )
r (t )
si (t )
h(t ) s0 (Tb t )
z0 (t )
z0 ( T )
h(t ) s1(Tb t )
z1(t )
z1 ( T )
h(t ) s (Tb t ) z (t )
M 1
M 1
z () dt
T
t T
n(t )
r (t )
z(T )
z(t )
Selects
si(t)
with
the
max zi(t)
r (t )
t T
h(t ) s(Tb t )
n(t )
si ( t )
si (t )
(a)
s(t )
s(t )
z(t )
z(T )
z(t )
si (t )
t T
(b)
Comparing (a) and (b) have

From (a)
T
z (T )
z (t ) 0 r (t ) s (t )dt
M 1
tT
z (t ) t T z (T ) 0 s ( )r ( )d
T
Prof. Okey Ugweje
395
Prof. Okey Ugweje
396
Correlator vs. Matched Filter - 2
Examples
From (b):
z' (t) r(t)h(t)
But
Example Signal to Noise Ratio
r( )h(t )d 0 r( )h(t )d
t
h(t) s(T t) h(t ) s[T (t )] s(T t)

z' (t)
Example Correlator Output
z r( )s( T t)d
t
At sampling instant t = T, we have
Example Matched Filter
z ' (t ) t T z ' (T ) 0T r ( )s( T T )d

0T r ( )s( )d
This is the same result obtained in (a)
Hence
z(T ) z' (T )
Prof. Okey Ugweje
397
#Generalized One Dimensional Signals - 1
-A
Binary Baseband Orthogonal Signals

Binary Antipodal Signals
2
2
Eavg A A A2
2
s1
0
398
One Dimensional Signal Constellation

so
M=2
Prof. Okey Ugweje
+A
so
s1
0
-A
2
2
Eavg A A A2
2
+A
Binary Orthogonal Signals
M=4
so
s1
-3A
-A
s2
s3
+A
+3A
2
+A s1
s0
Eavg 9 A A A 9 A 5 A2
4
M=8
Eavg
Prof. Okey Ugweje
so
s1
s2
s3
-7A
-5A
-3A
-A
+A
2
2
Eavg A A A2
2
2-Dimensional Signal Constellation
s4
s5
s6
s7
+A
+3A
+5A
+7A
An example:
2
2
2
2
2
2
2
2
49 A 25 A 9 A A A 9 A 25 A 49 A 21A2
8
1 (t )
2 (t )
1
T
1
T
T
2
399
Prof. Okey Ugweje
T
2
zoT 1(t) 2(t)dt 0
1
T
400
Generalization to M-ary Orthogonal Signals
M=4
M=8
TimeDomain
s0 (t ) A1(t )
s1(t ) A 2 (t )
s2 (t ) A 3(t )
s3(t ) A 4 (t )
TimeDomain
s0 (t ) A1(t )
s1(t ) A 2 (t )
s2 (t ) A 3(t )
s3(t ) A 4 (t )
s4 (t ) A 5(t )
s5(t ) A 6 (t )
s6 (t ) A 7(t )
s7(t ) A 8(t )
Signal Space
s0 ( A, 0, 0, 0)
s1 (0, A, 0, 0)
s2 (0, 0, A, 0)
s3 (0, 0, 0, A)
where {1(t), 2(t), 3(t) 4(t)}

are a set of orthonormal basis
functions
Signal Space
s0 ( A, 0, 0, 0, 0, 0, 0, 0)
s1 (0, A, 0, 0, 0, 0, 0, 0)
s2 (0, 0, A, 0, 0, 0, 0, 0)
s3 (0, 0, 0, A, 0, 0, 0, 0)
s4 (0, 0, 0, 0, A, 0, 0, 0)
s5 (0, 0, 0, 0, 0, A, 0, 0)
s6 (0, 0, 0, 0, 0, 0, A, 0)
s7 (0, 0, 0, 0, 0, 0, 0, A)
Prof. Okey Ugweje
where {1(t), 2(t), 3(t)

4(t), 5(t), 6(t), 7(t)
8(t)} are a set of
orthonormal basis
functions
General M
(M is a power of 2)
401
Most Common Signal Constellations - 1
s2 (t ) A3 (t )
s2 (0, 0, A, 0, 0, 0, 0, 0)
s3 (t ) A4 (t )
s3 (0, 0, 0, A, 0, 0, 0, 0)
sM 1 (t ) A M (t )
sM 1 (0, 0, 0, 0, 0, 0, 0, , A)
402

On-Off Keying
Are one dimensional signals either ON or OFF
with signaling points falling on the real line
so
Three common types of binary signals:

Antipodal
Two signals are said to be antipodal if one signal is the

negative of the other s1(t) = - s0(t)
Signal have equal energy with signal point on the real line
Prof. Okey Ugweje
s1 (0, A, 0, 0, 0, 0, 0, 0)
Prof. Okey Ugweje
That is, a geometric representation of signals
s1
0
s1 (t ) A2 (t )
Constellation is a method of representing the symbol

states of modulated bandpass signals in terms of their
amplitude and phase
Signal Space
s0 ( A, 0, 0, 0, 0, 0, 0, 0)
where {1(t), 2(t), 3(t) M-1(t)} are a set of orthonormal basis functions
so
Time Domain
s0 (t ) A1 (t )
Eavg E E E
2
403
s1
0
Eavg 0 E E
2
2
With OOK, there are just 2 symbol states to map

onto the constellation space
a(t) = 0 (no carrier amplitude, giving a point at
the origin)
a(t) = A cosct (giving a point on the positive
horizontal axis at a distance A from the origin)
Prof. Okey Ugweje
404

Orthogonal
Maximum Likelihood Receiver
Requires a 2 dimensional geometric

representation since there are 2 linearly
independent functions s1(t) and s0(t)
(derivation will be given in class)
E so
s1
0

Eavg E E E
2
Typically, the horizontal axis is taken as a

reference for symbols that are In-phase with the
carrier cosct, and the vertical axis represents
the Quadrature carrier component, sinct
Prof. Okey Ugweje
405
Prof. Okey Ugweje
406
Probability of Error for Binary Signals - 1

Unipolar Baseband Signaling

s1 (t ) A, 0 t T , for binary 1
1 0 0 1 1
T
s0 (t ) 0, 0 t T , for binary 0
r (t )
z () dt
T
z(t )
t T
z(T )
z(T ) o
3T
5T
si (t )
s1(t ) s0 (t )
Probability of Error
r(t) = s(t) + n(t)
For s1(t):
a1 (T ) E z (T ) | s1 (t ) E 0 r ( ) s1 ( )d 0 r ( ) s0 ( )d
T
E 0 s1 ( ) n s1 ( )d 0 s1 ( ) n s0 ( )d
T
T
E 0 s12 ( ) d 0 0 0
A2T
Prof. Okey Ugweje
407
Prof. Okey Ugweje
408
For s0(t):
Also:
a0 (T ) E z (T ) | s0 (t )
Ed 0 s1 (t ) s0 (t ) dt A2T
E 0 r ( ) s1 ( )d 0 r ( ) s0 ( )d
T
A2T

P Q 0 Q
B
N0
2 N0
E 0 s0 ( ) n s1 ( )d 0 s0 ( ) n s0 ( )d
T
T
T
E 0 s0 ( ) s1 ( ) d 0 0 s02 ( ) d 0
0
Ed
P Q
B
2 N0
a1 a0 A T
2
2
2
Prof. Okey Ugweje
409

0 t T , for binary 1
z ()dt
T
s0 ( t )
x
-A
3T
5T
FG E IJ
H NK
Pb Q
z(T )
tT
z1 (t )
z(T ) o
Prof. Okey Ugweje
IJ FG
K H
FG
H
Pb Q
Ed
10
si ( t )
T
0
Bipolar signals require a

factor of 2 increase in
energy compared to
Unipolar
Since 10log102 = 3 dB, we
say that bipolar signaling
offers a 3 dB better
performance than Unipolar
A A dt 2 A T
z (t ) z1 (t ) z0 (t ) a1 a0 0 0 0
Ed
Q
2 No
Bipolar (antipodal)
s1 ( t )
F
P QG
H
410
2 Eb
No
IJ
K
z0 (t )
z ()dt
Unipolar (orthogonal)
1 0 0 1 1
T
s0 (t ) A, 0 t T , for binary 0
r (t )
Bipolar Signaling (antipodal)
Prof. Okey Ugweje
s1 (t ) A,
Eb
P Q
B
N0
IJ FG
K H
4 A2T Q
2 No
2 Eb
No
IJ
K
411
Prof. Okey Ugweje
Othogonal
Antipodal
10
FG E IJ
H NK
-2
Q
P robability of Bit Error
10
10
10
10
-4
-6
FG
H
-8
2 Eb
No
IJ
K
3-dB
-10
10
12
Eb/No (dB)
14
16
18
20
412
Examples

Comparing BER Performance
Example
10
Othogonal
Antipodal
7.810
For Eb/No = 10 dB
Pb,orthogonal = 9.2x10-2
Pb, antipodal = 7.8x10-4
-2
10
Probability of Bit Error
9.2 10
-4
10
Example
-6
10
-8
10
-10
10
10
12
Eb/No (dB)
14
16
18
20
For the same received signal to noise ratio, antipodal

provides lower bit error rate than orthogonal
Prof. Okey Ugweje
413
Prof. Okey Ugweje
414
Baseband Communication Systems
A baseband signal x(t) with bandwidth B is a signal for

which X(f) is non-zero for |f| B and for which X(f) = 0 for
|f| > B (PSD concentrated near DC)
X(f)
-B
Digital Baseband
Communication System
A baseband communication system transmits

information using a baseband signal
Analog Input
A/D
Converter
an
Line
Coder
s(t )
Channel
To Receiver
Transmiter
Here the transmitter is simply a line coder (w/pulse

shaping function) that maps the sequence of bits an onto a
line code signal s(t)
Prof. Okey Ugweje
415
Prof. Okey Ugweje
416
Problems with Baseband Communication - 1
Most channels require that the baseband signal be

shifted to a higher frequency
Since antenna size is inversely proportional to the
center frequency fc, this is difficult to realize
Most channels are shared by several transmitters at

the same time
Shifting each user to different freq the channel can
be divided into freq slots
Frequency Division Multiple Access (FDMA)
Thus we must look at the process of shifting a
baseband signal to higher frequency
This process is called Carrier Wave Modulation
Problems:
Higher frequencies allow for the use of smaller antennas size versus
For speech signal f = 3 kHz = 105

Antenna size w/o modulation = 105 m = 60 miles practically unrealizable
This is evident that efficient antenna of realistic physical
size is needed for radio communication system
Prof. Okey Ugweje
417
Prof. Okey Ugweje
418
Solution is to use Bandpass Communication Systems
In a bandpass digital communication system, the

bit stream an is first converted to a baseband line code
m(t) by a line coder and is then converted to a
bandpass signal s(t) by a modulator
A bandpass signal has non-negligible spectrum only about

some carrier frequency fc >> 0
i.e., x(t) with bandwidth B is a signal for which X(f) is nonzero at some region about fc and for which X(f) = 0
elsewhere
X(f)
B
Analog
Input
-fc
fc
an
Line
Coder
m(t )
Modulator
s(t )
Channel
To
Receiver
Transmiter
Note: the bandwidth of a bandpass signal is the range of

positive frequencies for which the spectrum is non-zero
Usually, the bandwidth of bandpass signal is twice the
bandwidth of the baseband signal used to create it
Effective transmission of baseband information signal
usually requires the use of a bandpass signal
Prof. Okey Ugweje
A/D
Converter
419
Baseband signals m(t) may be transformed into

bandpass signals s(t) through the process of
modulation
Prof. Okey Ugweje
420
Representation of Bandpass Signals
We need some additional analytical tools to handle

bandpass signals
1. Magnitude and Phase (M & P)

Any bandpass signal can be represented as:
s (t ) R (t ) cos c t (t )
3 Major ways of Representing Bandpass Signals
R(t) 0 is real valued signal representing the magnitude

(t) is a real valued signal representing the phase
Magnitude and Phase (M&P) Representation

In-phase and Quadrature (I&Q) Representation
Complex Envelope Representation
This representation is easy to interpret physically, but

often is not mathematically convenient
In this form, modulated signal can represent information
through changing three parameters of the signal namely:
Amplitude R(t): as in Amplitude Shift Keying (ASK)
Phase
(t): as in Phase Shift Keying (PSK)
Frequency d(t)/dt: Frequency Shift Keying (FSK)
Prof. Okey Ugweje
421
422
2. In-phase and Quadrature (I & Q) Representation

Any bandpass signal can also be represented as
Relationship Between M & P and I & Q Forms:

To transform from M&P to I&Q
x(t) = R(t)cos(t), y(t) = R(t)sin(t)
To transform from I&Q to M&P
s (t ) x(t ) cos( ct ) y (t )sin( ct )

x(t) is a real-valued signal called In-phase (I)
y(t) is a real-valued signal called Quadrature (Q)
This is often a convenient form which
Emphasizes the fact that two signals may be
transmitted within the same bandwidth
Closely parallels the physical implementation of
the Tx/Rx
Prof. Okey Ugweje
Prof. Okey Ugweje
R (t ) x 2 (t ) y 2 (t )
(t) tan 1
LM y(t)OP
N x(t)Q
I and Q portions of the signal are orthogonal

Look
at the correlation between I & Q portions
T
x(t ) cos ct y (t )sin ctdt
1T
x(t ) y (t ) sin ct ct sin ct ct dt
20
1T
x(t ) y (t ) sin 0 sin 2 ct dt 0
20
423
Prof. Okey Ugweje
424
3. Complex Envelope (CE) Representation
Relationship: CE and I & Q Forms
Any bandpass signal can also be represented as
To transform from CE to I&Q:
s (t ) Re g (t ) exp( j c t )
where g(t) = complex envelope - complex-valued signal
S(t) is convenient in many instances for analysis. Why?
Compact
Easy to manipulate without recourse to trig. identities
x(t) = Re[g(t)], y(t) = Im[g(t)]

s(t) = Re[g(t)ejt] = Re[(x(t)+jy(t)).(cosct+jsinct)]
= x(t)cosct - y(t)sinct
Relationship between Spectral Representations

Assume that
j t
s (t ) Re g (t )e c
Relationship: Complex Envelope and M&P Forms

To transform from CE to M&P:
R(t) = |g(t)|, (t) = g(t)
To transform from M&P to CE:
Fourier Transform (Deterministic Signals):

S( f ) 1 G( f fc ) G( f fc )
2
g(t) = R(t)ej(t)
Prof. Okey Ugweje
425

Prof. Okey Ugweje
426
Bandpass Modulation & Demodulation - 1

Power Spectral Density (Random Signals):
Gs ( f ) 1 Gg ( f fc ) Gg ( f fc )
4
Relationship: Power and Envelope of Bandpass
Power of bandpass signal is one half of power in
complex envelope:
Source
bits
lm q
i
Pe
Performance
Measure
Gs Rs (0)
Source
Encoder
Format
Channel
Encoder
Modulate
Spread
Multiple
Access
Bits or
Symbol
Waveforms
Digital
output
lm q
1
1
1
2
g (t ) Rg (0) Gg
2
2
2
Multiplex
Digital
input
Format
Source
Decoder
Channel
Decoder
Source
bits
Prof. Okey Ugweje
From other
sources
Channel
bits
427
Prof. Okey Ugweje
Demultiplex
Demodulate
&
Detect
Despread
Multiple
Access
Channel
bits
To other
destinations
Carrier and symbol

synchronization
428
Aspects of Conversion
Bandpass Modulation & Demodulation - 1

Transmitter
Message
source
m(t )
Signal
transmission
encoder
si
Receiver
Modulator
si (t )
Channel
x(t )
Decoder
Signal
transmission
decoder
Carrier Wave
r = bits/signal = log2( L )
L = number of levels (signal elements)
N = bps
S = signals/sec (baud)
c = 1 for broadband (WAN digital-to-analog)
c = for baseband (LAN digital-to-digital)
Bandpass Modulation shifts the spectrum of a baseband

signal so that it becomes a bandpass signal
Why Modulate? (a review)
signals propagate well through the atmosphere
allows many signals w/different carrier freqs to share the
spectrum
is used to place signals at desired freq band for signal
processing
Info signal must conform to limitation of its channel
is used to map digital data sequence into waveform
Prof. Okey Ugweje
cN
r
429
Digital Modulation
Prof. Okey Ugweje
430
Digital Modulation Schemes
Basic Digital Modulation Schemes:

Amplitude Shift Keying (ASK) not commonly used
Frequency Shift Keying (FSK) very useful
Phase Shift Keying (PSK) very useful
For Binary signals (M = 2), we obtain BASK, BPSK,

BFSK, BAPK
For M > 2, many variations of the above techniques
exit usually classified as M-ary Modulation/detection,
e.g., MPSK
Prof. Okey Ugweje
431
Prof. Okey Ugweje
432
Most Common Digital Nodulation
MOdulation and DEModulation - 1
MODEM
(Phase info
required)
COHERENT
BINARY
ASK
(OOK)
M-ary
ASK
NONCOHERENT
HYBRID
APK(QAM)
BINARY
M-ary
ASK
(OOK)
ASK
FSK
FSK
(MSK)
FSK
PSK
(QPSK,
OQPSK)
PSK
DPSK
(No Phase info

required)
HYBRID
FSK
DPSK
CPM
CPM
Prof. Okey Ugweje
433
MOdulation and DEModulation - 1
Prof. Okey Ugweje
434
Analysis or Method of Approach:

Modulation Process
Mathematical Signal Representation
Power Spectral Density of the modulated signal

Bandwidth of the System
Detection Processes
Performance of the system
Amplitude Shift Keying
Error Probability
Prof. Okey Ugweje
435
Prof. Okey Ugweje
436
Amplitude Shift Keying - 2

In amplitude shift keying, the amplitude of the carrier

signal is varied to create signal elements.
Both frequency and phase remain constant while the
amplitude changes.
Modulation Process
Also called ON-OFF Keying (OOK))
In ASK, amplitude of carrier is switched between 2 (or more)
levels according to the digital data
1s & 0s are represented by two amplitude levels A1 & A0
Product modulator or
ON-OFF switch
m(t )
0
s(t )
3T
A cos( ot )
Baseband Data
Prof. Okey Ugweje
437
Prof. Okey Ugweje
OOK Modulator
Modulated bandpass Signal
438

Implementation of binary ASK
Analytical Expression:
Ai cos(0t ), 0 t T binary 1
s (t )
0,
0 t T binary 0
where Ai = peak amplitude

2
s (t ) A cos(0t ) 2 Arms cos(0t ) 2 Arms
cos(0t )
V2
2 P cos( t ) P
R
0
2E
T
Hence,
Prof. Okey Ugweje
439
Prof. Okey Ugweje
cos( t )
0
2 E cos( t ),
0
s (t ) T
0,
0t T , binary 1
0t T , binary 0
440
Generally, we can write

si (t )
2 E (t )
i cos( t ),
0
T
0 t T,
Power Spectral Density (PSD)

From the given signal
i 0,1, 2,..., M 1
s (t ) Ac m(t ) cos c t
where
T
Ei 0 si2 (t )dt , i 0,1, 2, ..., M 1
The PSD can be found using

Gs ( f )
We may also write

s1 (t ) Ac m(t ) cos(0t ), 0 t T binary 1
s0 (t ) 0,
To evaluate this we must first find the PSD of the

complex envelope m(t)
Using the fact that m(t) is a unipolar NRZ line code
given by
0 t T binary 0
This can be used to derive the transmitter for ASK:

Xn
line coder
A2
GM ( f f c ) GM ( f f c )
2
2,
an
0,
m(t )
m(t ) an f (t nT ),
Ac
cos( ct)
Prof. Okey Ugweje
441
Prof. Okey Ugweje
442
With A Ac 2 and using the general expression for

PSD of a unipolar line code, we obtain
Gg
A 2
(f) c
for binary 1
for binary 0
impulse
4T
T
A2
1
c TSa ( fT ) 1 ( f )
2T
T
F ( f ) 2 1 ( fc )
2
A
c ( f ) TSa 2 ( fT )
2
Tb
Note:
fc 2 Rb
It can be seen that the bandwidth B of ASK modulated

signal is twice that occupied by the source baseband
stream
The spectrum of a digitally modulated signal depends

on the baseband data format used to represent the
digital data
Prof. Okey Ugweje
fc Rb
443
Prof. Okey Ugweje
444
Receivers - Demodulators & Detectors

Coherent Receiver - 1
Bandwidth of ASK
Bandwidth B, of ASK can be found from its power spectral
density
B is twice that of unipolar NRZ line code used to create it, i.e.,
B 2 Rb 2
Tb
This is the null-to-null bandwidth of ASK
If raised cosine rolloff pulse shaping is used, then
1) Low Pass Filter Receiver

t T
r (t )
LPF
z(T )
si (t )
cos(t )
Coherent detection requires the phase information

A coherent detector mixes the incoming signal with a
locally generated carrier reference
Multiplying r(t) by the receiver LO (say cos(ct)) yields a
signal with a baseband component plus a component at
2fc
B (1 r)Rb W 1 (1 r)Rb
2
Spectral efficiency of ASK is half that of a baseband unipolar

NRZ line code
This is because the quadrature component is wasted
95% energy bandwidth B 3 3Rb
Tb
Prof. Okey Ugweje
445
Prof. Okey Ugweje
446
2) Matched Filter Receiver
Passing this signal through a low pass filter

eliminates the high frequency component
An integrator can be used in place of the LPF
The output of the LPF is sampled once per bit
period
This sample z(T) is applied to a decision rule
z(T) is called the decision statistic
r (t )
h(t ) s(Tb t )
z(T )
z(t )
si (t )
t T
MF receivers are very common approach in

signal detection in most bandpass data
modems
Prof. Okey Ugweje
447
Prof. Okey Ugweje
448

Non-Coherent Receiver - 1
3) Correlator Receiver
r (t )
z(t )
() dt
Does not require a phase reference info at the receiver

If we do not know the phase and frequency of the carrier,
we can use a non-coherent technique to recover signal
1) Envelope Detector:
t T
si (t )
z(T )
s1(t ) s0 (t )
r (t )
4) Quasi-coherent Square-law Receiver

r (t )
( )2
Prof. Okey Ugweje
z () dt
T
z(t )
449
t T
z(t )
z(T )
si (t )
If quadrature versions of the modulated carrier signal are

available then we may use the following receiver
I
Noncoherent reception of OOK is popular in fiber optics

Prof. Okey Ugweje
si (t )
Prof. Okey Ugweje
450
2) Square-law Detector:
T (n 1/ 2)
T (n 1/ 2) ()dt
z(T )
Probability of Error (Bit Error Rate)

Non-Coherent Receiver - 2
( )2
LPF
t T
The simplest implementation

of an envelope detector
comprises a diode rectifier
and smoothing filter
si (t )
z(T )
r (t )
Rectifier
z(t )
Envelope Detector
t T
BPF
@ fo
BASK effectively uses unipolar signal source and the

performance depends on whether coherent or noncoherent detection is used
Error analysis is similar for both cases
For both cases,
r (t ) s (t ) n(t )
s(t) is exactly the same as in both cases

For coherent detection n(t) is Gaussian, however for
noncoherent detection n(t) is no longer Gaussian due to
the squaring operation
Because of this squaring, the optimal threshold is not
necessarily halfway between the 2 possible values of s(t)
Derivation given in class

451
Prof. Okey Ugweje
452
Example 39: ASK
Probability of Error (Bit Error Rate) - ASK

Derivation
A binary ASK communication system employs

rectangular pulses of duration Tb and amplitude A to
transmit digital information at a rate R = 105 bps. If the
PSD of the AWGN is N0/2, where N0 = 10-2 W/Hz,
determine the value of A that is required to achieve the
probability of error of PB = 10-6
Prof. Okey Ugweje
453
Prof. Okey Ugweje
454
Frequency Shift Keying (FSK) - 1

In frequency shift keying (FSK), the frequency of the

carrier signal is varied to represent data.
The frequency of the modulated signal is constant for
the duration of one signal element, but changes for the
next signal element if the data element changes.
Both peak amplitude and phase remain constant for all
signal elements.
Frequency Shift Keying
Prof. Okey Ugweje
455
Prof. Okey Ugweje
456
Modulation Process:
The instantaneous carrier freq is switched b/w 2 or
more levels according to the baseband digital data
data bits select a carrier at one or more freqs
the data is encoded in the freq
FSK conveys the data using distinct carrier freqs to
represent symbol states
Important property = amplitude of the modulated
wave is constant
Prof. Okey Ugweje
457
Prof. Okey Ugweje
458
Analytical Expression
Generally, MFSK may be used to transmit k = log2M bps

waveforms
f determines the degree to which we can discriminate among
M possible signals
As a measure of similarity (or dissimilarity) between a pair of
signal waveforms, a correlation coefficient ij, is used
s (t )
i
2E
cos
T
i t ,
i 0,1, , M 1
t
i (t ) 0t d
m( )d )
fi
d
i (t ) f 0 f d m(t )
dt
Can also be expressed as

si (t )
where
Analog form
E1 oT si (t ) s j (t )dt
ij
freq offset
2E
cos 2 f 0t 2 ift , i 0,1, , M 1
T
f fi fi 1 ,
fi f o if
Prof. Okey Ugweje
459
s
1 T 2 Es
cos 2 f ot 2 ift cos 2 f o t 2 j ft
E o T
s
1 T
1 T
cos 2 (i j )ft dt o cos 4 f o t 2 (i
T o
T
sin 2 (i j )fT
2 (i j )fT
Prof. Okey Ugweje
dt
j )ft dt
0 since fo >> 1/T
460
Binary FSK - 1
ij
2 different freqs, f1 and f2 = f1 + f are used to transmit

binary data
sin 2 (i j )fT
2 (i j )fT
ij
f1
so (t ) Ac cos( 1t 1)
0.715
Tb
1
2Tb
1
Tb
-0.217
3
2Tb
2
Tb
2E
cos(2 f 0t 0 ), 0 t T
T
2E
cos(2 f1t 1 ), 0 t T
s1 (t )
T
s0 (t )
ij, is orthogonal when f is a multiple of 1/2T

Minimum of ij, = - 0.217 @ f = 0.715/T
461
Prof. Okey Ugweje
Binary FSK - 2
Binary Orthogonal Phase FSK
For NRZ Pulse Shape:
s1
so 0
0
1(t )
1 (t )
2E
cos( t 1 )
T
2 (t )
2E
cos( 2t 2 )
T

1 (t )2 (t )dt
ij
2E
cos(1 t 1 ) cos( 2t 2 )dt
T
sin 2Tb Tb 1 2 sin Tb 1 2
ij
2sin Tb
When 0 an 1 are chosen so that 1(t) and 2(t) are

orthogonal, i.e.,
1 (t )2 (t )
462
Binary FSK - 3
2 (t )
s1(t ) Ac cos( 2t 2 )
Data is encoded in the freqs

That is, m(t) is used to select between 2 freqs
f1 is the mark freq, and f2 is the space freq
Note:
Prof. Okey Ugweje
f2
form a set of k = 2 basis orthonormal basis functions
sin 4fTb 2fTb 1 2 sin 2fTb 1 2

2sin 2Tb
We need to look at two cases

1.Continuous Phase: 1 = 2
2.Non-continuous Phase: 1 2
Prof. Okey Ugweje
463
Prof. Okey Ugweje
464
Binary FSK - 5
Binary FSK - 4
Requiring 2 oscillators adds to the system

complexity and cost
Because there are 2 different fc it is difficult to use
complex envelope notation
This makes analysis difficult
Discontinuities in phase of s(t) at switching instants
result in undesirable spectral characteristics
Corresponds to high sidelobe levels which could
cause adjacent channel interference
Discontinuous-phase FSK is not used much in
practice
Discontinuous Phase FSK

1
1 2
Phase Discontinuities
Phase discontinuities occur at symbol boundaries
Prof. Okey Ugweje
465
Prof. Okey Ugweje
Binary FSK - 6
Binary FSK - 7
Continuous Phase FSK
Implementation of BFSK
466
0 1
No Phase Discontinuities
m(t)
BFSK
Frequency Modulator @ fo
1(t )
2
Tb
cos 2f1t
m(t)
ON-OFF Level
Encoder
+
x
Prof. Okey Ugweje
2 (t)
BFSK
2 cos 2f t
2
Tb
467
Prof. Okey Ugweje
468
Binary FSK - 8
Binary FSK - 9
A continuous-phase FSK (CPFSK) signal is

represented by:
si (t ) Ac cos c (t )
Peak Frequency Deviation

f
Df is the frequency deviation constant

m(t) is a digital line code
Usually polar, either with or without pulse shaping
CPFSK is an FM signal with digital line code modulating
signal
CPFSK is much more common than discontinuous phase
FSK
Unless otherwise specified, FSK will usually mean CPFSK
Thus:
Ac cos( ot D f m( )d )
Prof. Okey Ugweje
Df
469
f1 f c f , f 2 f c f , f1 f 2 2f
Modulation Index
h
2f
2 fT
R
minimum value of h for which the 2 possible signals do not

interfere with one another is h = 0.5
CPFSK with h = 0.5 is called minimum shift keying (MSK)
GSM uses MSK with Gaussian pulse shapes (GMSK)
Prof. Okey Ugweje
470
Binary FSK - 11
Binary FSK - 10
Other FSK Modulation Methods

Vector or Quadrature
Representation of Continuous Phase FSK

Magnitude and Phase
R(t ) Ac
t
m( )d
(t ) D f
Complex Envelope Notation

See Fig. 4.24
FSK requires the generation of 2 symbols, one at a frequency

(c + 1) and one at a frequency (c 1)
To generate a freq. shift of 1 at modulator output , the I and
Q inputs need to be fed with cos1 and 1sin respectively
This approach is now frequently used to generate some of the
more elaborate filtered CPFSK formats in cellular handsets
Prof. Okey Ugweje
471
Quadrature Notation
g(t ) Ac exp( jDf
t
m( )d )
x(t ) Ac cos( D f m( )d )
t
y (t ) Ac sin( D f m( )d )
t
Alternate Representation for CPFSK

Because frequency is the time rate of change of the
phase, we can represent a bandpass signal as
Prof. Okey Ugweje
472
Binary FSK - 12
Binary FSK - 13
PSD of CPFSK
Because complex envelope g(t) is a nonlinear function of
m(t), an exact expression for the PSD is difficult to obtain
A good approximation for s(t) can be found by considering
FSK to be the sum of 2 OOK signals
d (t )
d (t )

s (t ) R (t ) cos c
t R (t ) cos 2 f c
t
dt
2 dt
Hence for CPFSK

d (t )
R(t ) Ac
and
dt
D f m(t )
1 m(t )
cos(2 ( f c f )t )
2
1 m(t )
Ac
cos(2 ( f c f )t )
2
s (t ) Ac
D m(t )

f
t A cos(2 f t )
s (t ) Ac cos 2 f c
c
i

2 dt
If m(t) is polar NRZ (and A = 1)

Df
, when m(t ) 1
f1 f c
2
fi
f f D f , when m(t ) 1
c
2
2
Prof. Okey Ugweje
This approximation can be used to find the PSD

Result is that the null-to-null bandwidth is
B 2 f 1 r
473
Prof. Okey Ugweje
Binary FSK - 14
2Tb
474
Binary FSK - 15
Sunde's FSK
Clearly, the overall bandwidth occupied by the FSK

signal depends f
An FSK system using continuous phase transitions will
have much lower side-lobe energy than the
discontinuous case
Sunde's FSK arises when the spacing between the 2

symbol frequencies is made exactly equal to the symbol
rate
The spectrum uniquely contains 2 discrete spectral lines
at the two symbol frequencies in addition to a broad
spectral spread
These spectral lines may be used in coherent FSK
detector as the source of carrier references, often
extracted using a PLL
Minimum Shift Keying (MSK)

MSK employs symbol spacing of one half the
symbol rate
Prof. Okey Ugweje
475
Prof. Okey Ugweje
476
Binary FSK - 16
Binary FSK - 17
It produces a smooth spectrum with narrow main lobe

and reduced side-lobe energy
This narrow symbol spacing means that MSK is spectrally
efficient (more than BASK and BPSK, and about QPSK)
The price to be paid for this excellent performance is
more complexity in the generation and detection process
compared with Sunde's FSK
3
B r Rb
Detection of FSK:
Coherent
Coherent detection of FSK is similar to that for ASK but in this
case there are 2 detectors tuned to the 2 carrier frequencies
Recovery of fc in receiver
is made simple if the
frequency spacing
between symbols is made
equal to the symbol rate
(Sundes FSK)
Drawback of using Sunde's FSK
The bandwidth of the FSK signal is approximately 1.5 to 2
times that of an optimally filtered ASK or PSK binary signal
Bandwidth is minimized when h = 0.5 (i.e. for MSK)

Prof. Okey Ugweje
477
Prof. Okey Ugweje
Binary FSK - 18
The following configurations can be used for detecting FSK

signal
z () dt
z(t )
z ()dt
z0 (t )
t T
si (t )
z(T )
r (t )
tT
BPF/Envelope Detector:
z0 ( T )
Threshold
Detector
s0 ( t )
x
Noncoherent
Pass the signal through 2 BPF tuned to the 2 frequencies
and detect which has the larger output averaged over a Ts
s1(t ) s0 (t )
x
z ()dt z (t)
T
si ( t )
z1 ( T )
s1 ( t )
h(t) = s(Tb-t)
t T
h(t) = s(Tb-t)
BPF
Tuned @ f1
z (T )
y (t )
r (t )
r(t)
si (t )
z (T )
y (t )
Envelope
Detector
BPF
Tuned @ f2
Envelope
Detector
t T
Prof. Okey Ugweje
478
Binary FSK - 19
r (t )
479
Prof. Okey Ugweje
Sampler
Time
Sync
480
Binary FSK - 20
Binary FSK - 21
Phase Locked Loop (PLL)
Probability of Error Performance for FSK

(see derivation in class handout)
Coherent
FG E IJ
H NK
Pb Q
Zero-Crossing:
One simple digital method involves counting the zerocrossings of the carrier during a symbol and hence
directly estimating the frequency on a symbol-by-symbol
basis
Noncoherent
F
H
E
Pb 1 exp b
2
2 No
I
K
Coherent orthogonal BFSK performance is identical to coherent ASK
Quadrature Receiver
Eb/N0 penalty of noncoh. detection is only about 1 dB lower
Alternate BFSK demodulator is shown in Fig. 4.16

Prof. Okey Ugweje
Note:noncoherent FSK performance is not nearly as bad as ASK

481
Prof. Okey Ugweje
482
Example 40: FSK
Probability of Error (Bit Error Rate) - FSK

Derivation
If a system's main performance criterion is bit error

probability, which of the following two modulation
schemes would be selected for an AWGN channel?
Show computations.
Binary noncoherent orthogonal FSK with Eb/NO = 13 dB
Binary coherent PSK with Eb/NO = 8 dB
Prof. Okey Ugweje
483
Prof. Okey Ugweje
484
Phase Shift Keying (PSK) - 1

In PSK, the phase of the carrier signal is switched

between 2 or more phases in response to the
baseband digital data
The info is contained in the instantaneous phase of
the carrier
For binary PSK, phase states of 0o and 180o are used
Phase Shift Keying
Waveform:
Prof. Okey Ugweje
485
Prof. Okey Ugweje
486
Analytical expression can be written as
si (t ) Ag (t ) cos ot i , 0 t T , i 1, 2,..., M
where
g(t) = transmitting signal pulse shape
A = amplitude of the signal
= carrier phase
Range of the carrier phase can be determined using

i
2 (i 1)
2 i
or i
M
M
For a rectangular pulse, we obtain

g (t )
Prof. Okey Ugweje
487
Prof. Okey Ugweje
2
, 0 t T ; and assume A E
T
488
We can now write the analytical expression as
Also can be written as
si (t )
2 Es
cos
T
0t
2 i 1
,
M
s (t )
i
0 t T , i 1, 2,..., M
Constant envelope
carrier phase changes abruptly at

the beginning of each signal interval
180-phase
shift
0-phase
shift
-90-phase
shift
2T
3T
4T
the carrier phase changes abruptly at the beginning of each

signal interval while the amplitude remains constant
Prof. Okey Ugweje
489
490
01
00
11
M_ary Constellations
01
00
11
10
011
3 5 7
4
01
There is no non-coherent detection equivalent for PSK
491
M 2k
MPSK
BPSK
4
8
QPSK
8 PSK
16
16 PSK
Prof. Okey Ugweje
001
010
E
11
This technique is known as Differential PSK (DPSK)
100
110
111
10
M=4
01
010
M=8
00
10
001
M=4
110
111
101
011
11
000
It is also possible to transmit data encoded as the

phase change (phase difference) between
consecutive symbols
00
10
Prof. Okey Ugweje
3
0, , ,
2
2
2 i 1
M
2 (i 1)
2 (i 1)
2E
cos
cos c t sin
sin c t
M
M
T
Prof. Okey Ugweje

cos c t
For M-ary phase modulation M = 2k, where k is the # of

info bits per transmitted symbol
In an M-ary system, one of M 2 possible symbols, s1(t),
, sm(t), is transmitted during each Ts-second signaling
interval
The mapping or assignment of k info bits into M = 2k
possible phases may be done in many ways, e.g. for M = 4
2E
T
000
100
101
M=8
492
Binary Phase Shift Keying (BPSK) - 2

Is also called Phase Reversal Keying (PRK)

For BPSK, M = 2 and o = 0, 1 =
In your text, BPSK modulated signal is also written as

s (t ) m(t )
i.e.,2 carrier phases at o= 0 and 1 = rad are used to transmit data
There is a 1800 ( radian) phase shift

the two phases are separated by 180o
2 Es
T
s0 (t )
s1 (t )
v(t ) Re g (t )e
cos c t
2 Es
T
cos c t for binary 1
493
where g(t )
RS 1,
T1,
binary 1
binary 0
Prof. Okey Ugweje
494
The entire quadrature component is not used
The magnitude and phase of an OOK signal are:
This means that half the bandwidth is wasted

BPSK requires twice as much bandwidth as the polar line
code used to create it
si (t ) R(t ) cos( 0t i (t ))
where R (t ) 1, constant envelope
0,
binary1
(t )
, binary 0
If y(t) can be used, then loss in spectral efficiency is

recovered
1, binary1
x(t )
1, binary 0
The in-phase and quadrature components are:
s (t ) x(t ) cos( 0t ) y (t ) sin( 0t )
I-component is just the polar NRZ signal

If the second BPSK is transmitted as the Q-component,
then we have QPSK (quadrature PSK) signal
where y(t) = 0 and no Q component
Prof. Okey Ugweje
j c t
Complex envelope is entirely real

Complex envelope is equivalent to polar NRZ signaling
Imaginary portion of corresponds to Q component
Thus, binary phase modulated signal may be viewed as 2

quadrature carrier with amplitude depending on transmitted
phase of each signal
Prof. Okey Ugweje
cos 2 f c t c
where m(t) is the message waveform

Representation of BPSK
The complex envelope of an OOK signal is:
cos c t , for binary 0
2 Es
T
2 Es
T
495
Prof. Okey Ugweje
496

PSK Generation (Modulators)
Transmitters for PSK (Modulators)
The simplest means of realizing BPSK is to switch the sign

of fc with data signal, causing a 0 or 180 phase shift
Product modulators
This method is not too good because of the difficulty in

implementing bandpass high frequency, high Q filters
Differential encoding
Switching modulators
Receivers for PSK (Demodulators)
Coherent Receiver
Maximum Likelihood Detector
Square Law Detector
Correlator Detector or Costas Loop
Data stream may be pre-shaped at baseband prior to

modulation
Because the modulation process
is linear, the baseband filter
shape is imposed directly onto
the bandpass modulating signal
Prof. Okey Ugweje
Noncoherent Receiver
497
Differential PSK
Prof. Okey Ugweje
Modulation/Transmitter Process
Power Spectral Density of PSK
s (t )
i
2 Es
T
cos

c t 2 Mi 1
498
a f LMMFGH sin aaffffffTT IJK FGH sin aaffffffTT IJK OPP

N
Q
P f Eb
2
2 (i 1)
2 (i 1)
2 Es
cos
cos c t sin
sin c t
M
M
T
or
P( f )
e insin c ( f f )T 0.25 s e insin c ( f f )T 0.25 s

A2 T
b
2
A2 T
b
2
Bandwidth 2 R
Product modulators
2
T
Bbpsk signal is identical to Bbask

assuming the same degree of
pulse shaping
Switching modulators
Differential encoding
In fact, a BPSK signal can be viewed as an ASK signal with

the carrier amplitudes as +A and A (rather than +A and 0)
Prof. Okey Ugweje
499
Prof. Okey Ugweje
500
Receiver for PSK (Demodulators)

Coherent Receiver
1. Low Pass Filtering
2. Maximum Likelihood Detector (matched filter &
correlator)
3. Square Law Detector
4. Correlator Detector/Costas Loop
Prof. Okey Ugweje
1) Low Pass Filtering

Incoming data signal is mixed with a locally generated
carrier reference, and the difference component is selected
at the output
t T
r (t )
z(T )
si (t )
cos(t )
501
Prof. Okey Ugweje
502

2. Matched Filter
Noncoherent Receiver
h(t ) s(Tb t )
z(T )
z(t )
There is no noncoherent PSK because noncoherency implies no phase information

With no phase, there is no PSK
Instead, we use a pseudo noncoherent technique
known as Differential PSK (DPSK)
si (t )
t T
3. Correlator receiver
r (t )
LPF
Multiplying r(t) by receiver LO (say Accos(ct)) yields 2

components: a baseband component & a component at
2fc
LPF eliminates the high frequency component (@ 2fc )
The output of the LPF is sampled once per bit period
The sampled value z(T) is applied to a decision rule
r (t )
z () dt
T
z(t )
t T
si (t )
z(T )
s1(t ) s0 (t )
4. Quasi-coherent square-law receiver

r (t )
Prof. Okey Ugweje
( )2
z () dt
T
z(t )
t T
z(T )
si (t )
503
Prof. Okey Ugweje
504
Examples
Probability of Error (Bit Error Rate) - FSK

Probability of Error for BPSK
Example
Suppose that the binary PSK is used in transmitting info over
AWGN channel with power spectral density of N0/2 = 10-10
watts/Hz and Eb=A2T/2. Determine the signal amplitude
required to achieve an error probability of 10-6 if the data rate is
(a) 10 kbps, (b) 1Mbps
(see derivation in class notes or see class handout)

2 Eb
PB Q
N0
Example
Find the expected number of bit errors made in one day by the
following continuously operating coherent BPSK receiver. The
data rate is 5000 bits/s. The input digital waveforms are s1(t) =
Acos(w0t) and s2(t) = -Acos(w0t) where A = 1 mV and the singlesided noise power spectral density is N0 = 10-11 W/Hz. Assume
that signal power and energy per bit are normalized relative to a
1 ohm resistive load.
Prof. Okey Ugweje
505
Prof. Okey Ugweje
Binary Differential PSK - 1
506
Binary DPSK is regarded as the noncoherent version of BPSK

Data is encoded in phase shift between successive symbols
rather than the actual value of the phase
The Basic Idea:
This requires differential encoding of the data

The idea is to come up with an encoding/decoding
scheme that will give the same decoded output
regardless of whether the received data is inverted
In DPSK, the carrier phase of the previous data bit can
be used as a reference
If ak = 0 then shift carrier phase by 180o

If ak = 1 then no shift in carrier phase
ak 1
0
0
1
1
D-BPSK
BPSK
Differential BPSK looks just like BPSK except that the phase shift are in
a different place
Prof. Okey Ugweje
507
Prof. Okey Ugweje
508
If ak = 1, leave dk unchanged w.r.t. the previous bit

If ak = 0, change dk w.r.t. the previous bit
The encoded sequence {dk} is used to phase-shift a
carrier with phase angle 0 and representing symbols 1
and 0 respectively
Differential Data Encoding:

ak
dk1
dk
dk
RSd
Td
k 1,
k 1,
ak 0
ak 1
ak
0
0
1
1
Delay
Ts
ak
dk
dk 1 dk
0
1
1
0
0
0
1
1
This encoding process is efficient since it does not

introduce any extra data bits and hence does not
affect the throughput
D-BPSK
The 1-bit delay can be realized very simply using a clocked shift register
Prof. Okey Ugweje
509
Prof. Okey Ugweje
510
Differential Data Decoding:

The differential decoding process is equally simple to
implement using a 2nd exclusive-nor gate and a 1-bit
delay
Drawback of Differential Encoding/Decoding:

When single bit errors occur in the received data
sequence due to noise, they tend to propagate as
double bit errors
Error
dk
Delay
Ts
EX-NOR
ak
01101100
ak
dk1
dk1
dk
Delay
Ts
01111100
dk
Delay
Ts
EX-NOR
ak
dk1
Since the decoder is comparing the logic state of

current bit with previous bit, and if the previous bit is
in error, the next decoded bit will also be in error
Prof. Okey Ugweje
511
Prof. Okey Ugweje
512

DPSK Modulation:
DPSK combines two basic operations at the
transmitter
Differential encoding of the binary data, and
modulation
Demodulation of DPSK
r (t )
r
k
z () dt
T
x
Delay
T
r
k 1
t T
z(t )
si (t )
z(T )
Suboptimum Detector
See Fig. 4.17 (b)

x
z(t )
z0T ()dt
t T
r (t )
cos 0t
x
z(t )
z0T ()dt
z(T )
si (t )
x
T
sin 0t
Optimum Detector
Exercise:
Draw a matched filter implementation of the optimum
detector
Prof. Okey Ugweje
513
Prof. Okey Ugweje
BER Performance for DPSK

r (t )
x
Delay
T
x(t )
LPF
y(t )
In general, DPSK performs less than BPSK because the errors

tend to propagate due to correlation between bit waveforms
BPSK performs about 3 dB better than DPSK
The difference decreases with increasing Eb/No
t T
z(T )
si (t )
r
k 1
x(t ) A cos ot n(t ) A cos o (t Tb ) n(t Tb )
Differentially Encoded PSK (DEPSK)
y(t ) const A nc (t ) A nc (t Tb ) ns ns (t Tb )
514

r(t ) s(t ) n(t )
Sometimes, differentially encoded PSK is coherently

detected (see section 4.7.2)
In this case, the probability of error is
PB Pr Z(Tb ) 0
1
E
PB exp b
2
N0
Theoretical performance for CPSK & DPSK is shown for an AWGN channel
2 Eb
PB 2Q
N0
2 Eb
1 Q
N0
BER for CPSK is exactly the same as that derived for bipolar baseband
transmission
Prof. Okey Ugweje
515
Prof. Okey Ugweje
516
Example 43 - DPSK
a) The bit stream 11011100101 is to be

transmitted using DPSK. Determine the
encoded sequence, the transmitted phase
sequence and the detected sequence.
M-ary Modulation
Prof. Okey Ugweje
517
Prof. Okey Ugweje
518
M-ary Modulation Types Partial List
M-ary Digital Communications

In M-ary signaling scheme, we may send one of M = 2k possible

symbols, s1(t), s2(t), , sM(t) during each interval Ts
We refer to each M-ary message sequence as a character or
symbol
The rate at which M-ary symbols are transmitted through the channel
is called the Baud Rate
M-ary signals may be generated by changing the Amplitude,
Frequency or Phase of the carrier in M discrete steps resulting to the
following:
M-ary PSK
M-ary ASK
M-ary FSK
Another way of generating M-ary signals is to combine different
methods of modulation into a hybrid form e.g.,
Amplitude Phase Keying (APK) ASK + PSK
Abbreviation
MASK
MQAM
MFSK
MPSK
M=4
QPSK
/4 QPSK
OQPSK
DQPSK
/4 DQPSK
M>4
DMPSK
MSK
DMSK (GMSK)
MAPK
Prof. Okey Ugweje
Prof. Okey Ugweje
519
Descriptive Names
M-ary Amplitude Shift Keying
M-ary Quadrature Amplitude Modulation
M-ary Frequency Shift Keying
M-ary Phase Shift Keying
Quadrature Phase Shift Keying
/4 Quadrature Phase Shift Keying
Offset Quadrature Phase Shift Keying
Differential QPSK
/4 Differential QPSK
MPSK (e.g, 8-PSK, 16-PSK, 64-PSK, etc., )
Differential MPSK
Minimum Shift Keying
Differential MSK (Gaussian MSK)
M-ary Amplitude Phase Keying
520
M-ary vs. Binary
Practical Modulation Schemes

Modulation Type
Applications
FM (analog) AMPS
MSK
CT2
GMSK
GSM, DCS 1800, CT3, DECT, HIPERLAN-1
QPSK
NADC (CDMA) - base transmitter
OQPSK
NADC (CDMA) - mobile transmitter
4-DQPSK
NADC (TDMA), PDC, PHP (Japan)
/4-DQPSK
N. A. TDMA, PHS
QPSK/OQPSK CDMA One
QAM
IEEE 802.11 (5.7 GHz), HIPERLAN-2
GFSK
Bluetooth, IEEE 802.11-FHSS)
DPSK
IEEE 802.11-DSSS
CCK
IEEE 802.11-DSSS
Each symbol in an M-ary alphabet can be related to a

unique sequence of k-bits
Prof. Okey Ugweje
Prof. Okey Ugweje
521
M 2k k log 2 M
where M is the size of the alphabet

Rs
Rb bits
k log 2 M
Ts
Ts
log 2 M s
Tb
1 Ts
1

Rb k kRs
Any digital system that transmits k bits in Ts seconds using

bandwidth efficiency of
Rb log 2 M
1
bits / s / Hz
B
BTs
BTb
Any digital system will become bandwidth efficient if its BTb is
increase
522
Quadrature PSK (QPSK) - 1

QPSK (4PSK) is just 2 BPSK

arranged in phasequadrature, each operating
at half the bit rate of the
original bit stream
It transmits 2-bit of info using
4 states of phases
Quadrature PSK (QPSK)
2 bits are transmitted per

modulation symbol 2Tb=Ts)
The I and Q channels are

aligned and phase transition
occur once every Ts = 2Tb
seconds with a maximum at
180 degrees
Prof. Okey Ugweje
523
Prof. Okey Ugweje
524
Example QPSK encoding
The signals are:
2-bit
information
00
01
/2
10
11
3/2
Each symbol
corresponds to two bits
General expression:
sQPSK (t )
2E
s
T
s
cos 2 f ot
2 (i 1)
,
M
i 1, 2,3, 4
0 t Ts
Also can be written as

si (t )
2 Eb
Tb
2 (i 1)
2 (i 1)
sin c t sin
cos c t cos M
M
Prof. Okey Ugweje
525
s1
2 Es
Ts
cos c t
s2
2 Es
Ts
cos c t
s3
2 Es
Ts
cos c t
s4
2 Es
Ts
cos c t
2E
sin c t
2 Es
Ts
cos c t
s
T
s
(see next slide for

illustration)
3
2
s1,3 (t )
2 Es
Ts
cos 2 f ot ,
shift of 0o and 180o
s2,4 (t )
2 Es
Ts
sin 2 f ot ,
shift of 90o and 270o
Prof. Okey Ugweje
2 Es
Ts
sin c t
526
In terms of basis functions

1 (t )
2
Ts
cos 2 f o t
and 1 (t )
2
Ts
sin 2 f ot
we can write sQPSK(t) as

01 s
1
11
s2
sQPSK (t )
00
s0
s3
Es
cos
2 (i 1)
(t ) Es
M 1
sin
2 (i 1)
(t )
M 2
With this expression, the constellation diagram can easily be

drawn
10 Q
Q
For example:
Es
10
00
10
00
11
01
11
0,
Prof. Okey Ugweje
527
Prof. Okey Ugweje
01
, ,
3
2
2 Es
3 5 7
, , ,
4 4 4 4
528
QPSK Modulator:
I
A
2
RS
T
1
m2
1
A
2
R
Rs b
2
m2 (t )cos ot
cos ot
~ A cos ot
Serial-toParrallel
Converter
m(t)
Rb T1
b
R
Rs b
2
90
R
Rs b
2
m1
RS11
T
A sin t
o
2
A
2
SystemView
m1(t )sin ot
529
2.5e-3
5.e-3
7.5e-3
10.e-3
12.5e-3
2.5e-3
5.e-3
7.5e-3
10.e-3
12.5e-3
1.5
Amplitude
Source data is first split into 2 data streams (often by allocating alternate bits
to the upper and lower modulator)
with each data stream runs at half the rate of the input data stream
Think of m1 & m2 as bit stream that modulates the quadrature carriers
In QPSK the Tx is 2 BPSK Transmitters arranged in phase-quadrature, each
operating at half the bit rate of the original bit stream
Prof. Okey Ugweje
Modulated QPSK (t22)

0
500.e-3
-500.e-3
-1.5
Time in Seconds
Prof. Okey Ugweje
530
QPSK Demodulator:
QPSK receiver is composed of 2 BPSK receivers
one that locks on to the sine carrier and
the other that locks onto the cosine carrier
x
z ()dt
T
z1(t ) t T z1(T )
1(t)
r (t )
1 (t ) A cos o t
()dt
z0 (t )
z0 (T )
si ( t )
Compare
Z1 and Z0
2 (t ) A sin o t
2 (t)
z1(t )
z0T s1(t) 1(t)dt z0T a A cos otfa A cos otfdt A2Ts
zo ( t )
s (t ) 2 (t )dt
Ts
0 1
Prof. Okey Ugweje
Ts
0
Lo
a A cos t fa A sin t fdt 0

o
531
Prof. Okey Ugweje
532

Implementation of QPSK
Prof. Okey Ugweje
533
Prof. Okey Ugweje
534
Phase Diagrams:
s
45o
s3
Offset QPSK
Phase changes: 0, 90o, 180o
In QPSK phase transition between all the states are

possible
Since transition through the origin is possible (phase shift
of p), the signal envelope can pass through zero
momentarily
This could lead to errors or signal loss during transmission
Prof. Okey Ugweje
535
Prof. Okey Ugweje
536
Offset QPSK - 1
Offset QPSK - 2
Offset Quadrature Phase Shift Keying (OQPSK), also called

staggered QPSK (SQPSK) is a modified version of QPSK
Recall that in QPSK, the bit transition in I- & Q-channels occur
simultaneously
However, in OQPSK, I-channel (or Q-channel) bit stream is
offset by one bit period relative to Q-channel (or I-channel) prior
to modulation
Notice that the I and Q channels
are not aligned
This misalignment implies that
only one phase transition can
occur once every Ts = Tb sec with
a maximum at 90o
Q-channel: even bits, mI(t)
I-channel: odd bits, mQ(t)
Prof. Okey Ugweje
537
Offset QPSK - 3
Offset between I and Q means that transition is potentially

possible every Tb sec
OQPSK can be used to achieve a non-zero envelope in the
modulated signal
Prof. Okey Ugweje
538
For OQPSK, symbol transition across the origin (phase

changes of 180o) is prohibited (Compare this to QPSK)
Unlike QPSK, signal transition do not

pass through the origin
Differential QPSK (DQPSK)

QPSK
OQPSK
OQPSK is a constant envelope modulation scheme that is

attractive for systems using nonlinear transponders, e.g.,
satellite communication
Prof. Okey Ugweje
539
Prof. Okey Ugweje
540
/4 QPSK - 1
Differential QPSK (DQPSK)

For M = 4, the PSK signal can be considered as 2 BPSK

signals using sint and cost as carriers
Another variant of QPSK which is now widely used in

majority of digital radio modems is the /4 QPSK format
The 4-phases can then be differentially encoded by encoding 2

BPSK signals differentially as discussed
i.e., DQPSK modulator uses same differential data encoder for
each parallel data stream as binary DPSK counterpart
It is so called because the 4 symbol set is rotated by /4 or

45o at every new symbol transition
450
450
It employs the same principle of using a 1 symbol delayed

version of the received symbol stream to act as the
reference for demodulation
Symbol 1
Symbol 2
Symbol 3
/4 rotating symbol set
Time
The reason for this rotation is to ensure that the

modulation envelope of the QPSK signal never passes
through zero
Prof. Okey Ugweje
541
Prof. Okey Ugweje
542
/4 QPSK - 3
/4 QPSK - 2
The fact that the modulation envelope does not pass

through zero is important for the design of radio power
amplifiers
/4-QPSK is a compromise between QPSK and QPSK

It performs better in multipath environment
It is possible to differentially encode /4 QPSK /4DQPSK
/4-QPSK is widely used because it can be noncoherently
detected
/4-QPSK Mapping:
Qk
Ik
Comparing the vector diagrams for QPSK and /4 QPSK,

this property is clearly evident
Data bits
mI, mQ
Phase shift
T=2mTs
Phase shift
T=(2m+1)Ts
Since envelope never goes through zero, /4 QPSK

mitigates spectral spreading caused by system nonlinearity
00
-3/4
01
3/4
/2
/4-QPSK differs from QPSK in that I-Q phases of 0 & /2 &

those of /4 & /4 are alternatively changed every Ts sec
10
-/4
-/2
11
/4
Prof. Okey Ugweje
543
Prof. Okey Ugweje
Modulation
Max pahsechange
OQPSK
180o
90o
/ 4 QPSK
135o
QPSK
544
Generalized M-ary Differential PSK - 1
Generalized M-ary Differential PSK - 2
For the case of M > 2, the signal can also be differentially

encoded using the phase comparisons
For differentially coherent detection of DPSK it is given

by
FG
H
Increasing M > 4 allows further improvements in

bandwidth efficiency, but the additional symbol states are
no longer orthogonal
PE ( M ) 2Q
2 Es
sin
No
2M
IJ
K
they do not lie on the sine or cosine axis of constellation diagram
Error Probability Performance:

BER is difficult to compute
Symbol error probability for general M-ary PSK is given
by
FG
H
PE ( M ) 2Q
Prof. Okey Ugweje
2 Es
sin
No
M
IJ
K
545
Prof. Okey Ugweje
M-ary ASK
Orthogonal M-ary FSK - 1
Generation and detection process is scaled up, requiring multi-level symbol

mapping and comparison
Recall that in M-ary FSK we have M transmitted

signals si(t), i = 1,2, , M having waveforms
si (t )
s
T
s
cos 2 f o t 2 ift , i 1, 2, , M , 0 t Ts
A minimum frequency separation is required
MASK is not practically useful because of
i i 1
its relatively poor BER performance
Ts
or f i fi 1
1
, where Ts Tb log 2 M
2Ts
Modulator is same as BFSK and individual

frequencies are separated by 1/2Ts
its sensitivity to any gain variations in the

channel
its need for reasonable linearity in the
transceiver processing
For coherent MFSK, the Rx consist of bank of Mcorrelators or MF
Only BASK is usually used in practice

2E
f f i fi 1 , fi f o if
Detection of MASK is performed with the same methods employed with

binary ASK for either coherent or non-coherent detection
Prof. Okey Ugweje
546
547
Prof. Okey Ugweje
548
Probability of Error Performance:

Unlike M-ary ASK, M-ary FSK is important because of
its increased noise immunity compared to binary FSK
PE ( M ) M 1 Q
E
N
M 1 Q
o
kE
b
N
o
(eqn.
3.122)
coherent
k 1
549
FH
M 1
kEs
exp
k
k 1
(k 1) No
M
E M
E
1 exp s (1)k
exp s
k
M
No k 2
(k ) No
E
M 1 exp s
2
2 No
PE ( M )
noncoherent
Prof. Okey Ugweje
IK F
I
H
K
FH IK F
F I
I
H K
H
K
F I
H K
M 1 (1)k 1
MFSK is good for reliable data transmission in the

presence of high levels of noise
Prof. Okey Ugweje
550
As the number of symbol states

increases, the symbol averaging time
becomes very large, reducing the
effect of noise to almost zero
Quadrature Amplitude
Modulation
The Eb/N0 required for error-free transmission will thus
approach the Shannon-Hartley limit of -1.6 dB
M-ary FSK is a very effective modulation technique in
applications where the optimum performance in noise is
required
for example in deep space missions where the path loss is so great
Prof. Okey Ugweje
551
Prof. Okey Ugweje
552
Quadrature Amplitude Modulation - 1
The most commonly used combination of amplitude

and phase signaling is the Quadrature Amplitude
Modulation (QAM)
It is sometimes regarded as M-ary APK with

constraints put on the amplitude and phase
si (t )
Some books regard it as an extension of the QPSK

since it consist of two independent amplitudemodulated carrier in quadrature. i.e.,
si (t )
where
2E
[ai cos ot bi sin ot ]
T
Rb
m n
In this case, both the amplitude and phase can be

varied
Any combination of M1-level amplitude and M2-level

phase can be used in the construction of QAM
si (t ) g (t )[ai cos o t bi sin o t ]

where g(t) is the signal pulse shaping function
Rs
M 1 2m , M 2 2n , m n log 2 M 1M 2 ,
where ai and bi are amplitude levels obtained by

mapping k-bit sequence into amplitudes, or
Prof. Okey Ugweje
2E
Vi cos[ ot j ], i 1, 2, , M 1 , j 1, 2, , M 2
T
553
Prof. Okey Ugweje
554
Any combination of M1-level amplitude and M2-level

phase can be used in the construction of QAM
QAM waveform can be represented as a linear
combination of 2 orthogonal signals 1(t) and 2(t)
Using the vector representation, we can realize an Lby-L matrix representing the coordinates of (ai, bi)
( L 3, L 1)
( L 1, L 1)
( L 1, L 3) ( L 3, L 3)
{a , b }
i i
( L 1, L 1) ( L 3, L 1)
si (t ) Ai1 (t ) Bi2 (t )
where
1 (t )
2
T
s
cos[ ot ], 2 (t )
2
T
s
sin[ ot ]
where
L M
In vector notation:
si si1 , si 2 Ai E , Bi E
Prof. Okey Ugweje
( L 1, L 1)
( L 1, L 3)
( L 1, L 1)
ai , bi
555
Prof. Okey Ugweje
556
The 4-QAM and 8-QAM constellations
16-QAM constellations
Prof. Okey Ugweje
557
Prof. Okey Ugweje
558

MQAM Modulator:
Conventional M-ary QAM Modulation

I
A serial-to-parallel
converter divides the
incoming data stream
into two bit stream each
at one-half the rate
fb
2
fb
fb
2
Q
559
Prof. Okey Ugweje
Premod
LPF
DSB-SC
AM Mod
X
0o
fb 1
2 log2 L
Data
Slicer
Then each branch is applied to a DSB-SC AM modulator

The output of both quadrature is added to yield an MQAM
signal
Although the modulator above is for 16QAM, it is good for any
M-ary QAM by changing the level shifter
Prof. Okey Ugweje
2-to-L-level
converter
LO
Phase
split
BPF
IF AMP
90o
2-to-L-level
converter
Premod
LPF
DSB-SC
AM Mod
560
Correlator Receiver Structure:
M-ary QAM Demodulation:
r (t )
z0T ()dt
2
cos o t
T
Threshold
and
Decision
Logic
2
sin o t
T
z0T ()dt
si ( t )
tT
With this receiver, any QAM signal can be recovered

with only two correlators
The output of the correlators give a point on the signal
constellation
Prof. Okey Ugweje
561
This demodulator uses I & Q

remodulation of the received signal
It can be used to demodulate any MQAM signal by changing
the level shifter
Level shifter can be implemented by A/D flash decoder
consisting of M-1 comparators each which is set at various
M-threshold levels
Their output are sampled and applied to parallel-to-serial
converter
Prof. Okey Ugweje
QAM Signal Constellation
16-QAM Constellations
Signal space diagram (constellation) is very important

in QAM
This is because any combination of M1-level amplitude
and M2-level phase (or amplitude) can be used to
construct M=M1M2 QAM signal
QAM allows the signal vectors to be placed anywhere
on the constellation plane
Usually, signal points are placed at equally spaced
distance
A particular constellation gives rise to different
probability of error
Prof. Okey Ugweje
562
563
Type I QAM (Star Constellation)

C. R. Cahn, 1960
Type II QAM Constellation
J. C. Hancock and R. W. Lucky
Type III QAM Constellation
Compopiano & Glazer, 1962; J. Salz, J. R. Sheenhan, & D.J. Paris 1971
Q
Type I
Type II
Type III
16 QAM (8, 8)
16 QAM (4, 12)
16 QAM (4, 8, 4)
Prof. Okey Ugweje
564

Since we desire min radial distance, but max

separation between points the square constellation is
easier to implement and has a slightly better
probability of error performance
Type I and Type II constellations are not preferred for
Gaussian channels
need higher energy to achieve the same min
distance compared to Type III
QAM is used by high-speed wireline modems

Allows data rates of 9,600 bps and above over
ordinary telephone lines
9,600 bps modem uses 16-QAM or 32-QAM (V.22
and V.32)
14.4 kbps uses 128-QAM
28.8 kbps uses 512-QAM
Prof. Okey Ugweje
Prof. Okey Ugweje
565
566
General Decision Rule for M-ary - 1

Comparing the constellation diagrams of M-ary QAM

with M-ary PSK we can see that the spacing between
symbol states for QAM is greater than that for PSK
Once a point on the signal constellation plane is determined for

the received signal, a decision can be made
The decision rule is to pick the signal point that is closest to
the received point
The distance between the signal point and the received point is
a function of the noise in the environment during the symbol
interval
If the noise has moved the received point closer to a different
signal point, then the receiver will make an error
This is because PSK constellation are restricted to

symbol states of equal amplitude and thus on a circle
equidistant from the origin
The larger spacing between symbols for QAM means
that the detection process should be less susceptible
to noise
Prof. Okey Ugweje
567
Prof. Okey Ugweje
RSIf the receiver

Tcalculates this point
RSThen, it will pick the symbol

Tcorresponding to this signal point
568
Thus, for decision purposes, we partition the signal

constellation diagram into decision regions
min Euclidean distance amongst phasors gives rise to noise
immunity
the min distance between any pair of signal vectors is
In a special case where amplitudes take discrete values (2i-1M)d, constellation is rectangular
nb
g b
dij si s j 1 E ai a j 2 bi bj
2
s1
d min d 2 E
gs
1 (t )
s2
569
BER Performance for QAM - 1
1b
a
IK
Min separation between signal points determines PE(M)

Energy of signal depends on the radial distance from origin to
signal point
desire minimum radial distance, but max separation between
points
Prof. Okey Ugweje
570
The exact performance of QAM depends on the shape of a

particular signal constellation diagram
For a rectangular constellation, the probability of correct
detection is
2
PC ( M ) 1 P M
where P M 2 1
1
M
3 Eav
( M 1) N
o
PE ( M ) 1 PC ( M ) 1 1 P M
For this odd-bit constellation root of

M is not an integer
It is not possible to gray encode
I
d
5-bit QAM Constellation
Hence the probability of error is given by
A general performance of coherent QAM (even or odd), the

symbol error probability can be bounded as (M > 4)
3 Eav
3 kEb
PE ( M ) 1 1 2Q
4Q
( M 1) N o
( M 1) N o
Eav is the average energy per bit

k is the number of bits per symbol
This probability of error is exact for M =

k is even
That is, a rectangular QAM (Type III) can only be
implemented when k = 2M (even)
Odd-bit constellations add complexities to the CODEC
2k,
Prof. Okey Ugweje
FH
(a, b) a b cos o t tan
Minimum phase rotation amongst constellation points

determines the phase jitter immunity
resilience against clock recovery imperfections & channel
phase rotations
Peak-to-average phase power ratio
robustness against nonlinear distortion of power amplifier
Prof. Okey Ugweje
2 (t ) Q
571
Prof. Okey Ugweje
, k 1
572
Variants of QAM - 1
Offset QAM Modulation

I
2-to-L-level
converter
Premod
LPF
fb
2
fb
PB ( M )
Data
Slicer
Prof. Okey Ugweje
573
Phase
split
LO
2-to-L-level
converter
Premod
LPF
Prof. Okey Ugweje
Superposed M-ary QAM Modulation
Variable Rate QAM Modulations
Data
input
QPSK
Modulator
Serial-to-2x2 bit
paralel converter
574
Variants of QAM - 3
BPSK
BPF
IF AMP
s (t ) a2 k h t 2kT cos c t a2 k 1h t 2k 1 T sin c t

k
Variants of QAM - 2
BPF
DSB-SC
AM Mod
LO
90o
The improvement of 16-QAM over 16-PSK comes from the

noise immunity capability of QAM
However, the design requirements of QAM is more complicated
needing to handle both amplitude and phase
DSB-SC
AM Mod
0o
fb
2
2(1 M 1 ) 3log 2 M 2 Eb
Q
2 1) N
log M
M
(
2
o
Half
Symbol
Delay
QPSK
2-level QPSK
32-level Star
QAM
16 Star QAM
Type 1
IF AMP
64-level Star
QAM
QPSK
Modulator
QAM transmission over Rayleigh Fading Channel

Burst error due to deep fades
Varying the modulation levels in response to fading
conditions
Be able to use built-in PSK MODEMs in realization

Less efficient than the conventional QAM
implementation
Suitable for data transmission

Variable QAM constellation
Prof. Okey Ugweje
575
Prof. Okey Ugweje
576
Trellis-Coded Modulation - 1
Trellis-Coded Modulation - 2
Combined coding and modulation scheme

TCM achieves coding gain without BW expansion and reduction
of effective information rate
Both power and bandwidth efficient
In power limited environment:
Use error correcting code increases power efficiency
Requires higher rate higher bandwidth
In bandwidth limited environment:
Choose higher-order modulation increases spectral
efficiency
Larger signal power is needed for the same signal separation
TCM combines the choice of higher-order modulation with
convolutional code
TCM achieves coding gain without BW expansion and reduction
of effective information rate
TCM is classified into two basic types

Lattice type MPAM and MQAM
Better power efficiency
Constant amplitude type MPSK
Lower power efficiency, better over satellite channel
Observations:
We can use coding gain without BW expansion
Coding and modulation are not separate entities
Demodulation and decoding in single step
Performance is governed by free Euclidean distance not
free hamming distance of the code
Optimization of TCM is based on the free Euclidean
distance
Detection is based on soft decision
Prof. Okey Ugweje
Prof. Okey Ugweje
577
Summary list of Digital MODEM - 1
Binary Modulation Schemes

Amplitude Shift Keying or ON-OFF Keying
Coherent and Noncoherent
Frequency Shift Keying (FSK) or Continuous-Phase FSK
Coherent and Noncoherent
Phase Shift Keying (PSK)
Coherent and Differential PSK
M-ary (multi-level) Modulation Schemes
M-ary Amplitude Shift Keying (MASK)
M-ary Frequency Shift Keying (MFSK)
M-ary Phase Shift Keying (MPSK)
QPSK, Differential QPSK, OQPSK, /4 PSK and /4 QPSK
M-ary Amplitude Phase Keying (MAPK)

Quadrature Amplitude Modulation (MQAM)
578
Prof. Okey Ugweje
Federal University
of Technology,
Minna
Part 5: Digial Bandpass
Communication
579
Minimum Shift Keying (MSK) or Fast Frequency Shift Keying

Differential MPSK (MPSK)
Differential Encoded MPSK (DEMPSK)
Differential MSK (DMSK)
Gaussian MSK (GMSK)
Superposed QAM (SQAM)
/4 Differential PSK
Quadrature Partial Response (QPR)
Sinusoidal Frequency Shift Keying (SFSK)
Comparison of Modulation Schemes
For practical application, the choice of digital MODEM depends on:
bandwidth efficiency,
power efficiency,
error performance,
Complexity of implementation, and
Cost
Prof. Okey Ugweje
580
Power Efficiency - 1

Probability of symbol error or Probability of bit error is related to:

Power efficiency
Bandwidth efficiency (spectral efficiency)
The performance of modulation schemes is summarized based on BER and
complexity
Usually transmitted power and complexity increases with increase in
bandwidth efficiency
The linear or nonlinear nature of the channel also affect the choice of
digital MODEM
Lastly, but not the least, government regulations also affect the choice of
digital MODEM
A desirable characteristics of any modulation scheme is the simultaneous
conservation of bandwidth and power
This has lead to the combination of coding and modulation (also known
as Trellis Coded Modulation)
Definition:
Power Efficiency (), is a measure of how much received
power is needed to achieve a specified bit error rate
Power efficient modulation schemes requires less power for
satisfactory BER
is a function of signal-to-noise ratio (SNR)
In the computation of , it is assumed that:
All modulation levels occur with equal probability, 1/M
Gray encoding is used to map the information bits into levels
Differential encoding may be employed
Power efficient modems are not bandwidth efficient (next 2
slides)
Power efficient schemes are more appropriate for satellite &
mobile communications
Prof. Okey Ugweje
Prof. Okey Ugweje
581
Power efficient modulation schemes include:

BPSK (or equivalently DSB-SC-AM in analog
system)
QPSK and 4-QAM
Assuming both I- and Q-channel is an unfiltered
balanced NRZ bit stream
BPSK and QPSK
is 2 b/s/Hz theoretical (1.5 ~ 1.8 b/s/Hz
practical)
Low Eb/No for good error probability performance
Relatively simple hardware design
Summary of Power efficient modulation

More appropriate for satellite communications
systems
BPSK and QPSK
Requires less power for satisfactory BER
They are not bandwidth efficient modulation
Expressed in terms of SNR for required BER
Power efficient:
If a Pe= 10-8 requires an Eb / No < 14 dB
Prof. Okey Ugweje
Prof. Okey Ugweje
582
583
584
Bandwidth Efficiency - 1
Definition:
Bandwidth efficiency () is the ratio of the bit rate to
channel bandwidth expressed in bit per second per
hertz (b/s/Hz)
It is also called Spectral Efficiency
The primary objective of spectrally efficient modulation
is to maximize the bandwidth efficiency
With data rate denoted as R, and the channel
bandwidth by B, then Bandwidth Efficiency is given
as
Rb
1
log 2 M 2 bits / s / Hz
B BTb
Prof. Okey Ugweje
In theory, BT 1 (for role-off-factor, = 0)

In practice, > 0, out-of-band emission constraints
imposed by FCC spectrum regulation
T is well defined, but B is not - hence of a digitally
modulated signal depends on the definition adopted for B
Capacity of a digital communication system is directly

related to
The max possible bandwidth efficiency is
max
C
S bps
log 2 1
B
N Hz
Note that binary systems are more power efficient, but

less spectral efficient than M-ary systems
585
Prof. Okey Ugweje
586
Note that bandwidth efficient modem are not power

efficient
Spectrally efficient modems include:
Summary of Bandwidth efficient modulation

More appropriate for microwave radio
M-ary level schemes (MPSK, MQAM) (M > 4)
Can transmit more information bit / BW
They are not power efficient modulation
Expressed in terms of Rb/B (b/s/Hz)
Spectral Efficiency:
If spectral efficiency > 2 b/s/Hz
M-ary QAM
In theory, = 4, 6, & 8 b/s/Hz for 16-, 64-, and 256QAM, respectively
But in practice we have, 2.5-3.5, 4.5-5, & 5-6, respectively
Available Eb/No > 30 dB
Usually, in spectral efficient modulation, the common

carrier band is subdivided into channels of width B
4-, 6-, 11-GHz bands in the USA have channel
bandwidths of 20, 30, and 40 MHz, respectively
More appropriate for digital microwave radio

Prof. Okey Ugweje
587
Prof. Okey Ugweje
588
Comparison of Digital MODEM - 1
Spectral Efficiency Plane

How do I compare one modulation format to another?

Bandwidth of Coherent Binary Modulation Schemes
Rectangular Pulses
Raised Cosine
ASK
2/T
(1+r)/T
FSK
4/T
2(1+r)/T
PSK
2/T
(1+r)/T
Comparison of some PSK Modulation Schemes

Modulation
Scheme
Prof. Okey Ugweje
589

Bandwidth Efficiency,
Immunity to
Nonlinearity
Implementation
Complexity
0.5
D (worst)
a (simple)
2.0
1.0
9.6 dB
2.0
1.0
9.6 dB
N/A
2/3
A (best)
d (complex)
EbNo
(dB)
Nyquist
Null-to-Null
BPSK
9.6 dB
1.0
QPSK
9.6 dB
PSK, QAM
Coherent FSK
Noncoherent FSK
Prof. Okey Ugweje
Required
CNR
BPSK
10.6 dB
Rb
10.6 dB
QPSK
10.6 dB
0.5Rb
13.6 dB
8-PSK
14.0 dB
0.33Rb
18.8 dB
16-PSK
18.3 dB
0.25Rb
24.3 dB
Prof. Okey Ugweje
Pb = 10-6
Pb = 10-5
590
Bandwidth Efficiency of M-ary PSK
16
32
64
(bits/s/Hz)
0.5
1.0
1.5
2.0
2.5
3.0
Bandwidth Efficiency of M-ary FSK
Bandwidth Efficiency of some Modulation Schemes
M-ary System
Max
(bits/s/Hz)
Modulation
Scheme
MSK
Min Channel B for

ISI free signaling
OQPSK
Required
Eb/No
Bandwidth Efficiency bits/s/Hz
16
32
64
(bits/s/Hz)
1.0
1.0
0.75
0.5
0.3125
0.1875
1 log M
2 2
2log 2 M
M 3
log M
2
2M
Assuming frequency separation of Rs/2

Assuming frequency separation of 2Rs/2
591
Prof. Okey Ugweje
592
Assuming PB 10
Modulation
Scheme
Complexity of Modulation Schemes
Bandwidth,
B
Eb/No
(dB)
Equipment
Complexity
coh. ASK
2Rb
14.45
Moderate
noncoh. ASK
2Rb
18.33
Major
Rarely used;
o A/2
o A2Tb / 4
Peo Pe1
coh. FSK
2Rb
10.60
Minor
Seldom used
Performance does not justify complexity
0 0
noncoh. FSK
2Rb
15.33
Minor
Used for slow speed data transmission

Poor utilization of power and bandwidth
0 0
coh. PSK
2Rb
8.45
Major
Used for high speed data transmission

Better overall performance but requires
complex equipment
0 0
Differential
PSK
2Rb
9.30
Major
Most commonly used in medium speed

data transmission
Error tend to occur in pairs
0 0
Prof. Okey Ugweje
BPSK
QAM, QPSK
OQPSK
MSK
CPFSK - optimal detection
QPR
M-ary PSK
APK
Comments
Low
High
Complexity
DQPSK
DPSK
CPFSK -discriminator detection
FSK - noncoherent detection
OOK - envelope detection
IEEE 1979
593
Prof. Okey Ugweje
594
Probability of Error Calculation - 1
What is the difference between symbol error and bit error?

Probability of symbol error vs. Probability of bit error?
One important parameter of communication systems is the
SNR or the Eb/No defined as:
b
Probability of Error Calculations
2
Eb A2Tb
A2
A
No
No
No (1 / Tb ) NoW
Also, p. 158 of your textbook defines

b
e j
e j
Eb STb
S SW S W S W
No No RNo RNoW NoW R
N R
where S = Average message signal power

N = Noise variance NoW
W = Bandwidth
R = Rate
1
Eb
Eavg
log 2 M
Generally,
Prof. Okey Ugweje
595
Prof. Okey Ugweje
596
Symbol Energy and SNR per Symbol:
Bit Energy and SNR per Bit
Consider the signals sm, m = 1, 2, , M

Assume symbols are equiprobable
Bit rate is R
P( sm )
1
M
Bit Rate
RT log2 M
Symbol Rate
, m 1, 2,, M
Energy of signal m is Em: (called Es if Em is equal for all m)

Average energy per symbol
Eav
1
M
Em , ( Es if Em equalfor all m)
Average Power
m1
Pav Eav R ( Eb R if Em is equal for all m)
Average power
Pav
Eav
,
T
1
where the symbol rate is
T
Average SNR per bit
Average SNR per symbol

S Eav

N No
Es
No
Prof. Okey Ugweje
Eb av
if Em equal for all m)

Average Energy per bit

Eav
(called Eb Es if Em is equal for all m)
Eb av
log 2 M
N0
597
Prof. Okey Ugweje
598
Error Probability:
Probability of bit error is Pb
Probability of symbol error is Pe or PM or P(M)
We can compare modulation schemes in terms of the
Eb/No required to achieve a specified or Pe or Pb
Generally,
1
Pe ( M )
log 2 M
Relation between Pe and Pb for Orthogonal Signals
The number of symbols conveying an incorrect bit in one

of the log2M positions is M/2
The probability of having an incorrect bit in any one of the
log2M positions is
M
1
2 M 1
The probability of bit error is

Pb
Pb
Since the Euclidean distance between any 2 signals is

the same, there is no benefit to Gray coding
When a symbol error occurs, each of the (M-1) remaining
symbols is chosen with probability (1/M-1)
Prof. Okey Ugweje
Eav
E
Es
( b
if Em is equal for all m)
log 2 M N o
N o log 2 M N o
599
M
Pe
2( M 1)
Relation between Pe and Pb and for PAM, QAM, PSK

Assume that Gray coding is used, then the most probable
symbol errors cause exactly one bit error each, since
each symbol encodes bits:
Hence Pb 1 Pe ( M )
log 2 M
Prof. Okey Ugweje
600
Probability of Error of Modulation Schemes - 1
Modulation
Probability of Symbol Error for M-ary Orthogonal

Signals
Coherent Exact
1
2
PM
c z
y
1
2
x2
2
M 1
exp 21 y
2 Es
No
FG
H
Es
No
PM
n 1
( )
n 1
Orthogonal
IJ
K
Bandpass Systems
Es
Q
BASK (OOK)
N0
FG M 1IJ 1 expFG nE IJ
H n K n 1 H (n 1) N K
s
BFSK
All Cases
1
Pb
PM
2( M 1)
Prof. Okey Ugweje
BPSK
Es
Eb
log 2 M
601
Prof. Okey Ugweje
Pb (noncoherent)
2 Eb
Q
No
dy
Noncoherent Exact
M 1
Pb (coherent)
Baseband Systems
Antipodal
Coherent Union Bound

PM M 1Q
PM (coherent)

Es
N0

2Es
No
Eb
N 0
Eb
N 0
Eb
N 0
2
1 exp A
2
8N0
1 exp Eb
2
2 N0
2 Eb
N 0
1 exp Eb
N
2
0
602
Modulation
QPSK
OQPSK
PM (coherent)
2 Es
2Q
N0
2 Es
2Q
sin
M
N0
DPSK
2Q
F
H
6 log2 MEb
( M 2 1) No
MASK
2( M 1)
M Q
MFSK
( M 1)Q
MPSK
Prof. Okey Ugweje
2Q
I
K

kEb
N0
2 Eb log 2 M
N0
Modulation
Pb (coherent) Pb (noncoherent)
sin M

2 Eb
N0
2Es
Q
N0
1 Q
2( M 1)
M
1 exp Eb
No
2
2 Eb
N0
Eb
N0
Eb
( M 2 1) No

Q 2Q
kEb
N0
( M 1)Q
2 Eb
N0
Eb
M 1 e 2 No
2
4 Es
N0
sin M
603
2 Es
N0
Es
N0
3kEs
( M 1) N0
MDPSK
2Q
/4QPSK
2Q
MQAM
PM (coherent)
4Q
sin M
sin
MSK
2 Es
N0
GMSK
0.68
2 Eb
N0
Prof. Okey Ugweje
Pb (coherent) Pb (noncoherent)
2M
j
1
FH
1 M )
2(log
M Q
2
2 Eb
N0
3 log2 M 2 Eb
( M 2 1) N0
IK
604
Resource Sharing Techniques - 1
Digital Communications
Duplexing (Review Read Section 9.1)

Multiplexing Techniques (self study)
Module 5
Multiplexing & Multiple Access
Resource Sharing Techniques
Duplexing
Multiplexing Techniques
Frequency Division Multiplexing
(FDM)
Time Division Multiplexing (TDM)
Code Division Multiplexing (CDM)
Wavelength Division Multiplexing
(WDM)
What is Multiple Access?

Prof. Okey Ugweje
Multiple Access Techniques

Frequency Division Multiple
Access (FDMA)
Time Division Multiple Access
(TDMA)
Practical TDMA Systems
Code Division Multiple Access
(CDMA)
How CDMA Works
Practical CDMA Systems
Hybrid Multiple Access
Techniques
605
Frequency Division Multiplexing (FDM)

Wavelength Division Multiplexing (WDM)

Frequency Division Multiple Access (FDMA)
Time Division Multiple Access (TDMA)
Code Division Multiple Access (CDMA)
Direct Sequence CDMA
Other Multiple Access Techniques

Prof. Okey Ugweje
606
Duplexing Techniques - 1
Resource Sharing Techniques - 2

Since the RF spectrum is a finite and limited resource,

it is necessary to share the available resources
between users
A technique commonly used in many

Terminal
Terminal
radio and telecommunication between a
A
B
pair of users Tx and Rx
Simplex
Simplex
Terminal
Terminal
Info is transmitted in one and only
A
B
one pre-assigned direction
Half-duplex
Half Duplex
Transmission of information in only
Terminal
Terminal
one direction at a time
A
B
Full-duplex
Uses simplex operation both end
Full Duplex
Simultaneous transmission and reception of info in both directions
In general, duplex operation require 2 frequencies
May be achieved by simplex operation of 2 or more simplex at both ends
Digital
input
lm q
i
Forma
t
Source
Encoder
Source
bits
Channel
bits
Channel
Encoder
From other
sources
Multiplex
Modulate
Spread
Multiple
Access
Tx
Pe
Performance
Measure
Bits or
Symbol
Waveforms
Rx
lm q
Format
Source
Decoder
Channel
Decoder
Demultiplex
Digital
output
Prof. Okey Ugweje
Source
bits
Channel
bits
Demodulate
&
Detect
To other
destinations
Despread
Multiple
Access
Duplexing can be implemented in either Frequency or Time domain

Frequency Division Duplexing (FDD) & Time Division Duplexing (TDD)
Carrier & symbol

synchronization
607
Prof. Okey Ugweje
608
Frequency Division Duplexing (FDD)

Multiplexes the Tx and Rx in one time slot in which
transmission and reception is on 2 different frequencies
It provides simultaneous transmission channels for
mobile/base station
i.e. each channel has a Forward and a Reverse
frequency
At the base station, separate transmit and receive antennas
are used to accommodate the two separate channels
At the mobile unit, a single antenna (with duplexer) is used
to enable transmission and reception
To facilitate FDD, sufficient frequency isolation of the
transmit and receive frequencies is necessary
FDD is used exclusively in analog mobile radio systems
Time Division Duplexing (TDD)
Prof. Okey Ugweje
Prof. Okey Ugweje
609
Multiplexing Techniques
Amplitude
Multiplexes the Tx & Rx in one frequency at different time slots

Time Division Duplexing
Time
A portion of the time is used to transmit and a portion is used to

receive
TDD is used, for example, in a simple 2-way radio where a button
is pressed to talk and released to listen
If the data rate from the base station >> the end-users data rate,
it is possible to use buffer-and-burst transmission (giving the
appearance of full duplex)
TDD is only possible for digital transmission
610
Frequency Division Multiplexing (FDM)
Multiplexing (sometimes called channelization) is the process of

simultaneously transmitting several information signals using a
single communication channel
Commonly used to separate different users such that they share the
same resource without interference
In Frequency Division Multiplexing (FDM), the available

bandwidth is divided into non-overlapping frequency slots
Each message is assigned a frequency slot within the available
band
Signals are translated to different frequency band using
modulation and then added together to form a baseband signal
The signals are narrowband and frequency limited
Communication recourses are allocated a priori and allocated resources are

fixed
Only one pair of transceivers are required
Prof. Okey Ugweje
611
fN-1
fN-2
Frequency
Three major kinds

Frequency Division
Multiplexing
Time Division Multiplexing
Code Division Multiplexing
f3
f2
f1
Frequency Band N
Frequency Band 2
Frequency Band 1
f0
Time
FDM can be used for either digital or analog transmission

Prof. Okey Ugweje
612

Digitized info from several sources are multiplexed in time and transmitted
over a single communication channel
The communication channel is divided into frames of length Tf
Each frame is further segmented into N subinterval called slots, each with
duration Ts = Tf/N, where N is the number of users
CDM is a multiplexing method

where multiple users are
permitted to transmit
simultaneously on the same
time and same frequency
In CDM system, users time
share a higher-rate digital
channel by overlaying a higherrate digital sequence on their
transmission
Each user is assigned distinct
code sequence (or waveform)
This technique may be viewed
as a combination of FDM and
TDM using some sort of code
...
Slot
N
s1
s2
...
Sync word
sk
Slot
1
Slot
2
Slot
N
...
s1
s2
...
Information or data word

FRAME
Each user is assigned a slot (or channel) within each time frame
TDM is used to combine several low bit rate signals to form a high-rate
signal to be transmitted over a high bit rate medium
Individual message signals need not have the same rate, or same type of
signal since each channel is independent of one another
TDM is usually used for digital communication and cannot be used in analog
communication
Different combining techniques are shown below
Prof. Okey Ugweje
613
Prof. Okey Ugweje
Frequency
Band 3
Signal 3
Signal 1
Signal 3
Band 2
Signal 2
Signal 3
Signal 2
Band 1
Signal 1
Signal 2
Signal 1
Slot 1
Slot 2
Slot 3
In optics, the process of using laser source, repeater

amplifier, and optical detector to independently
modulated light carriers to be sent over a single fiber is
known as WDM
Definition:
Multiple Access (MA) techniques
are multiplexing protocols that allow
more than a pair of transceivers
to share a common medium
i.e., the simultaneous use of a
channel by more than one user
Allocation of resources
not defined a priori
not necessarily fixed
Each users signal must be kept
uniquely distinguishable from other
users signals, to allow private
communications on demand
Users can be separated many ways:
physically: on separate wires by
arbitrarily defined channels
established in frequency, time, or
any other variable imaginable
This process has been very

difficult until recently
fc of light with sufficient
spectral stability is required
and was not available until
recently
Each individual light carrier could support data rates of up to

10 Gbps with users time multiplexed onto the channel
WDM thus offers the possibility of several hundreds of gigabits
transmission over a single fiber and also bi-direction
transmission over the same fiber
614
What is Multiple Access?
Wavelength Division Multiplexing (WDM)

Prof. Okey Ugweje
Time
Code Division Multiplexing
615
Prof. Okey Ugweje
616

Multiple Access can be implemented in:

Frequency Division Multiple Access
A users channel is a private frequency uses different frequencies for different users
Time Division Multiple Access (TDMA)
A users channel is a specific frequency, but
it only belongs to the user during certain
time slots in a repeating sequence
That is, same frequency is used but
different time for different users
Code Division Multiple Access (CDMA)
Each users signal is a continuous unique
code pattern buried within a shared signal,
mingled with other users code patterns
If a users code pattern is known, the
presence or absence of their signal can be
detected, thus conveying information
Uses same frequencies and time but
different codes (3G wireless systems)
Prof. Okey Ugweje
Space Division Multiple Access (SDMA)

Uses spot beam antennas to separate radio signals by
pointing at different users with different spot beam, e.g.,
ACTS
Demand Access Multiple
Multiple Access Protocol
Contentionless
(Scheduling Access)
Contention
(Random Access)
CDMA
617
Fixed
Assigned
Demand
Assigned
Repeated Random
Access
Random Access
w/reservation
FDMA
TDMA
Polling
Token Passing
ALOHA
Slotted ALOHA
Implicit
Explicit
Prof. Okey Ugweje
Access (DAMA)
Uses dynamic
assignment protocol
(allocates resources on
request)
Random Access Multiple
Access (RAMA)
Hybrid Multiple
Accesses
Time Division CDMA, Time

Division Frequency
Hopping, FDMA/CDMA,
etc.
618
FDMA - 2
FDMA - 1
FDMA is the oldest and most familiar method of radio

communication
Distant users are far enough that they cause no interference

When the call is finished, the channel is released and available for a
new call
If the transmission path deteriorates, the controller switches the
system to another channel
FDMA is the method used in the original cellular systems
AMPS Advanced Mobile Phone System
Although technically simple to implement, FDMA is wasteful of BW
Channel is assigned to a single conversation whether or not
somebody is speaking
It cannot handle alternate forms of data, only voice is permissible
Used extensively in the early telephone and wireless multi-user
communication systems
FDMA is the most commonly used access protocol especially for
satellite communication
used since 1890 in broadcasting, two-way radio, and

cellular systems
Individual frequencies (private frequencies) are

assigned to individual users on demand for the
duration of their call
Guard band (at the edges & between) to
minimize crosstalk
B
FRAME
Prof. Okey Ugweje
619
Prof. Okey Ugweje
620
FDMA - 3
FDMA - 4
In a cluster, each user is assigned a portion of the

available bandwidth
Channel
1
Channel
2
MHz
Bs
Let
Ndata = number of data channel
Nctl = number of control channel
Total Bandwidth
Bs N data Bc N ctl Bc 2 Bg
N s , or N N data N ctl
N Ns
N data Bc Bs
Bs 2 Bg
621
N ch / cluster
t number of calls per hour
N
Prof. Okey Ugweje
N user
number of calls/hour/cell
average # of calls/user/hour
BW available for data transmission N data Bc
1
Bs
sytem bandwidth
N data / cluster
# of data channel/cluster
chls/MHz/km 2
Bs N Acell
sytem BW Acluster
Guard Bands
Bg
Bs
N s Bc Bg
Prof. Okey Ugweje
Channel
1
622
FDMA
FDMA Capacity
TDMA - 1
FDMA - 5
Average number of
users per hour per cell
Spectral Efficiency
N data / cell
Number of calls per hour per cell (where t is the trunk

efficiency)
N calls
Bc
Prof. Okey Ugweje
Bs 2 Bg
N data / cluster
N
We can also determine the # of control channels per cluster of
cell in a similar manner
Number of data channels/cell
Bs N s Bc 2 Bg
Number of Channels
N ch / cluster
Bc
N
Number of channels/cell
N ch / cell ch / cluster
N
Number of data channels/cluster N data / cluster N ch / cluster N ctl / cluster
Channel
Ns
......
Bc
Bg
Number of channels/cluster
Channel
2
......
Bc
Bs
Channel
Ns
MHz
623
In TDMA, each user has a specific

frequency but only during an
assigned time slot
The freq is used by other users
during other time slots
Available time is divided into frames
of equal duration
In each time slot, only one user is allowed to either transmit or
receive
Number of time slots/ frame is a design parameter depending
on requirements (e.g., modulation, bandwidth, data rate, etc.)
In TDMA, bitstream are broken into frames, frames broken into
slots and slots are assigned to users
Prof. Okey Ugweje
624
TDMA - 2
TDMA - 3
One TDMA Frame
TDMA/FDD
Control Bits
Information Data
Slot 1 Slot 2 Slot 3
Also see Fig. 9.4
Trail Bits
Sync. Bits
Illustration of TDMA Transmission
Trail Bits
Slot N
Information Data
Guard Bits
Forward and Reverse channels are duplexed within time domain (TDD) or
frequency domain (FDD)
Slots contain data, error check, guard, synchronization training, and control
bits
TDMA transmits data in a buffer-and-burst technique and hence
transmission is not continuous
low battery consumption is achieved, and simplification of handoff
process is achievable
Transmission from users are interlaced into cyclic time structure
TDMA requires very high data rate compared to FDMA and hence
equalization is not required
Prof. Okey Ugweje
625
Each earth station

is assigned a time
slot in a repetitive
time frame
Over the length
of the time slotthe earth station
occupies the
entire bandwidth
of the
transponder
Prof. Okey Ugweje
TDMA - 4
626
TDMA - 5
TDMA Operation
TDMA Systems
TDMA can operate in wideband or narrowband
Wideband TDMA (W-TDMA)
entire freq spectrum is available to any individual user
Narrowband TDMA (N-TDMA)

total available freq spectrum is divided into subbands, with each
subband operating as a TDMA system
A user only uses the allocated subband
Both frequency and time are partitioned
Basic Frame Structure

Let
Bs = Bt = total spectrum allocation
Bg = guard band
Prof. Okey Ugweje
627
Prof. Okey Ugweje
628
TDMA - 6
TDMA - 7
Bc = Channel bandwidth of individual user

N = frequency reuse factor
Nu = number of subbands
Ld = number of information data symbols in
each slot
Ls, = the total number of symbols in each slot
Nslot = m in your textbook

N s N u N slot
for W-TDMA
1,
N u Bs 2 Bg
, for N-TDMA
B
c
Tf
Preamble
T1
T2
......
(N-1)slot
Nslot
TNslot
sec
N cell
Prof. Okey Ugweje
N u N slot
N
N N slot
u
sf N
N cell
Trailer
629
Prof. Okey Ugweje
For voice communication with talk spurt

(on) state and silence (off) state
TDMA - 8
630
TDMA - 9
Overhead bits per frame
Total number of bits per frame b bT b0 H
bOH N r br N t bp N t bg N r bg
Information bit burst rate, Rb+ Rb
where
bOH =overhead bits per frame

Nr = # of reference burst per frame
br = # of overhead bits per frame
bp = # of overhead bits per preamble in each time slot
bg = # of equivalent bits in each guard time interval
Spectral Efficiency of TDMA

Tf

Tf
Total number of traffic bits per frame

bT T f R
where R = channel bit rate
Prof. Okey Ugweje
100%
p t
Tf
p t
Tf
Ld
,
Ls
Ld Bs 2 Bg
, for N-TDMA
Ls
Bs
for W-TDMA
TDMA Capacity
Frame efficiency
b
f 1 OH
bT
T frame
Rb
Tslot
631
Prof. Okey Ugweje
Ttraffic
T
R
f frame f b
Tslot
Tslot
Rb
632
TDMA - 10
TDMA - 11
Advantages:
No inter-modulation impairment
Since TDMA uses one carrier at a time
No interference from other simultaneous transmissions
TDMAs technology separates users in time ensuring that they will not
experience interference from other simultaneous transmissions
Flexibility
TDMA can be easily adapted for the transmission of data or voice
Variable rates
TDMA offers the ability to carry data rates of 64 kbps to 120 Mbps
(expandable in multiples of 64 kbps)
This enables operators to offer PCS (fax, voice-band data, and SMS,
etc.), as well as bandwidth-intensive applications multimedia and
videoconferencing
Bandwidth efficient protocol
TDMA uses bandwidth more effectively because no frequency guard
bands are required between channels
Low power consumption
since transmission is bursty and non-continuous
i.e, TDMA provides the user with extended battery life and talk time since
the mobile is only transmitting a portion of the time (from 1/3 to 1/10)
during conversations
Guard time between time slots may be used to accommodate
clock instability
delay spread
transmission (or propagation) delays and pulse spreading
Prof. Okey Ugweje
Achieves selectivity in time domain, and selectivity is simpler than FDMA

TDMA devices can be mass produced by VLSI giving rise to low cost
TDMA offers the possibility of a frame monitoring of signal strength (or BER)
to enable better handoff strategies
Ideal for digital communications
TDMA is also the most cost-effective technology for upgrading a current
AMPS analog system to digital
633
Prof. Okey Ugweje
Practical TDMA Systems
TDMA - 12
Ideal for satellite on-board processing

TDMA is the only technology that offers an efficient utilization of hierarchical cell
structures offering pico-, micro-, and macro-cells
Hierarchical cell structures allow coverage for the system to be tailored to support
specific traffic and service needs
By using this approach, system capacities of more than 40-times AMPS can be
achieved in a cost-efficient way
Because of its inherent compatibility with FDMA analog systems, TDMA allows
service compatibility with the use of dual-mode handsets
IS-54 and IS-136 (TDMA)
Disadvantage
In TDMA, each user has a predefined time slot. However, users roaming
from one cell to another are not allotted a time slot
Thus, if all the time slots in the next cell are already occupied, a call
might well be disconnected
Likewise, if all the time slots in the cell in which a user happens to be in are
already occupied, a user will not receive a dial tone
TDMA is subjected to multipath distortion because of its sensitivity to timing
Even at thousandths of seconds, these multipath signals cause problems
Overall TDMA is more complex and costly compared to FDMA
Prof. Okey Ugweje
634
635
IS-54: The original TDMA format, intended for use

within existing AMPS systems
These systems use TDMA by dividing a 30-kHz channel into 3 time
slots, enabling 3 different users to occupy it at same time
IS-54 provides a 3-fold increase in traffic capacity relative to AMPS,
given the same bandwidth allocation
This effectively triples the capacity of the system (freq reuse)
A second phase of the IS-54 standard provides for 6 (instead of 3)
TDMA user channels in each 30 kHz radio channel
IS-136: Enhanced TDMA with special control channels to

allow short message service, battery life extension, other
features 6 timeslots, three users occupy in rotation
Prof. Okey Ugweje
636
CDMA
GSM (Groupe Special Mobile)
Each users signal is a continuous unique code pattern

buried within a shared signal, mingled with other users
code patterns
If a users code pattern is known, the presence or absence
of their signal can be detected, thus conveying information
All CDMA users occupy same frequency at the same time!
Time and frequency are not used as discriminators
CDMA operates by using coding to discriminate between
users - instead of using freq or time slots
Each user is assigned a unique PN code sequence
GSM standard was developed as a Pan-European

digital cellular standard to replace six incompatible
analog cellular systems then in use in different
geographic areas
GSM standard is similar to IS-54, employing TDMA, but
with 8 timeslots (7 or 8 users occupy in rotation), and
with RF carriers spaced 200 kHz apart
Japanese Digital Cellular

Please note that TDMA is well understood, commonly
employed, and is an efficient media access technique
Prof. Okey Ugweje
637
Prof. Okey Ugweje
638
How CDMA Works An Analogy

The assigned code is uncorrelated with the data

Because the signals are distinguished by digital codes,
many users can share the same bandwidth simultaneously
i.e., signals transmitted in same frequency & same time
The PN code used for spreading must have
low cross-correlation values and
be unique to every user
Each user is a small voice in a roaring crowd - but with a
uniquely recoverable code
CDMA technology focuses primarily on the DSSS
technique
4 speakers are simultaneously giving presentation, each

with different language -- Arabic, Chinese, English & Hindu
Chinese
Principles
English
OF
CDMA
Arabic
English
Major
Hindu
You are in the audience, and English is your native language

Prof. Okey Ugweje
639
Prof. Okey Ugweje
640
Characteristic of DS/CDMA
How CDMA Works

You only understand the words of the English speaker

and tune out the Arabic, Chinese, and Hindu speakers
You hear only what you know and recognize
This is the general idea of CDMA systems

Multiple users share the same frequency band at the
same time, yet each user can only recognize his or her
own code
This technique allows numerous phone calls to be
simultaneously transmitted in one radio frequency band
Coded conversations are encoded/decoded for each user
A signal correlated with a given PN code and

decorrelated with the same PN code returns the original
signal
Prof. Okey Ugweje
641
Universal Frequency Reuse

Uses one universal cell frequency reuse pattern
improves the capacity of the system
Ease of freq management is also found in
DS/CDMA
Power Control
Reverse Link (from mobile unit to base station)
link is designed to be asynchronous and is
susceptible to the near-far problem
In order to remedy this, the use of power control is
employed
Prof. Okey Ugweje
Characteristic of DS/CDMA
642
In Summary
Effective use of the power control will ensure that

power control must be accurate and fast enough to
compensate for fading
Forward Link (from base station to mobile unit)
Link does not suffer much from near-far problem
since all cell signals can be received at the mobile
with equal power
When at excessive intercell interference, the power
control can be applied by increasing the power to the
mobile
1. In CDMA, a signal is spread into a larger freq band

than is needed to represent it - the redundancy gives
error resilience, and the wideband frequency combats
multipath effects because of frequency diversity
2.Cell-reuse patterns are no longer strictly necessary
3.CDMA is described as having a universal one-cell
reuse pattern
Prof. Okey Ugweje
Prof. Okey Ugweje
643
644
Advantages of CDMA
Advantages of CDMA
1.Voice Activities Cycles

CDMA is the only technique that succeeds in taking
advantage of the nature of human conversation
In CDMA, all the users are sharing one radio channel
The human voice activity cycle is 35%, the rest of the time
we are listening
Because each channel user is active just 35% of the entire
cycle, all others benefit with less interference in a single
CDMA radio channel
2.Improved call quality, with better and more consistent sound
as compared to other systems
3.No Equalizer Needed
When the transmission rate is much higher than 10 kbps in
both FDMA and TDMA, an equalizer is required
On the other hand, CDMA only needs a correlator, which is
cheaper than the equalizer
Prof. Okey Ugweje
4.No Hard Handoff

In CDMA, every cell uses the same radio
This feature avoids the process of handoff from one freq to another while
moving from one cell to another
5.No Guard Time in CDMA
TDMA requires the use of guard time between time slots
guard time does occupy the time interval for some info bits
This waste of bits does not exists in CDMA, because guard time is not
needed in CDMA technique
6.Less Fading
Less fading is observed in the wide-band signal while propagating in a
mobile ratio environment
7.Capacity Advantage
Given correct parameters, CDMA can have as much as four times the
TDMA capacity; and twenty times FDMA capacity per channel/cell
8.No frequency management or assignment needed
In both, TDMA and FDMA, the frequency management is always a critical
Since there is only one channel in CDMA, no frequency management is
needed
645
Prof. Okey Ugweje
Advantages of CDMA
646
Disadvantages of CDMA
9.Enhanced privacy
CDMA signals resistant to interception or jamming
10.Soft Capacity
Because in CDMA all the traffic channels share a single radio channel,
we can add one additional user so the voice quality is just slightly
degraded
11.Coexistence
Both systems, analog and CDMA can operate in two different spectra,
with no interference at all
12.Simplified system planning through the use of the same frequency in
every sector of every cell
Improved coverage characteristics, allowing for the possibility of fewer
cell sites
13.Increased talk time for portables
14.Bandwidth on demand
1. Capacity not well defined

The capacity of CDMA systems is not well defined. The
effective (Eb/No) formula demonstrates the interferencelimited nature of the system, but more than one factor in that
formula is affected by the number of users, making it hard to
gauge how performance degrades as a function of users
Prof. Okey Ugweje
647
2. The Near-Far Problem

Effect is present when an
interfering Tx is much closer
to Rx than the intended Tx
Assume there are 2 users,
one near the base and one
far from the base as shown
CDMA interference comes
mainly from nearby users
Prof. Okey Ugweje
Near-Far effect illustrated
648
Disadvantages of CDMA
Practical CDMA Systems
Although the cross-correlation between codes A and B is low, the correlation between
the received signal from the interfering Rx and code A can be higher than the
correlation between the received signal from the intended Rx and code A
In CDMA, stronger received signal levels raise the noise floor at the base station
demodulators for the weaker signals, thereby decreasing the probability that weaker
signals will be received
The result is that proper data detection is not possible
IS-95 (cdmaOne)
To help eliminate the Near-Far effect, power control is used

Base Station (BS) rapidly samples the signal strength of each mobile and
then sends a power change command over the forward link
This sampling is done 800 times per second and can be adjusted in 84
steps of 1 dB
The purpose of this is so that the received powers from all users are roughly
equal
That is, when a mobile unit is close to a BS, its power output is lower
the mobile unit transmits only at the power necessary to maintain
connection
This solves the problem of a nearby subscriber overpowering the BS receiver
and drowning out the signals of far away subscribers
An extra benefit of power control is extended battery life
Prof. Okey Ugweje
649
After the development of the IS-54 standard,

Qualcomm, a San Diego-based company, developed a
new digital cellular system design utilizing Code Division
Multiple Access (CDMA)
This is known as IS-95
Unlike IS-54, which utilizes the same 30-kHz (same as
AMPS), IS-95 uses a SS signal with 1.2288 MHz
spreading bandwidth
a frequency span equivalent to 41 AMPS channels
IS-95 has been shown to theoretically offer greater

traffic capacity than TDMA
CDMA2000
Prof. Okey Ugweje
CDMA Performance - 1
CDMA System Analysis
CDMA Transmitter
Data signal
dk(t)
Users are identified by unique code sequence

Let
t iTb
d k t ski
ski PTb t iTb
i
Tb i
ak(t)dk(t)
Baseband
BPF
Modulator
Transmitted Signal
xk(t)
PN Code
Generator
Acos ct
Chip Clock
First the data symbols dk(t) are spread into ak(t)dk(t)

Then spread signal is modulated (usually by PSK)
T
Notice that f 1
T NT N b
c
t lTc
ak t akl
akl PTc t lTc
l
Tc l
Tc
Tc
N=PG = Gp = number of chips per data symbol = processing gain

Hence, resulting spread spectrum signal can be written as
Please note that ak(t) and dk(t) are completely independent

ak(t)
K = number of users
dk = kth users baseband data sequence with amplitude 1
ak = kth users spreading code sequence with amplitude 1
Prof. Okey Ugweje
650
651
t iTb lTc
sk (t ) Ac ski akl
cos 2 f c t
Tc
l
i
Prof. Okey Ugweje
652
yk (t ) h s t k d
where fc = carrier frequency, = carrier phase
We can simplify the expression above and use

sk (t ) 2 Pk ak t d k t cos 2 f c t k
2E
where Pk b , Pk = k-th user power
Tb
2 Pk kl ak t - k d k t - k cos c t k t - kl e j kl d
L
l 1
L
2 Pk kl ak t - kl d k t - kl cos c t kl
2E
or Pk s
Ts
l 1
where
kl k kl c kl Asynchronism
Let L be the number of resolvable paths which is
assumed to satisfy the condition
The Channel Model
T
L m 1
Tc
hk (t ) kl t kl e j kl
L
channel output is
Tm = maximum delay spread

Tc = chip period
l 1
Prof. Okey Ugweje
653
Prof. Okey Ugweje
CDMA Receiver
CDMA system model (k-th user)
Signal is first demodulated and then despread

The signal is despread by the same amount through a
cross-correlation by locally generated PN sequence
i.e., demodulation accomplished by remodulating w/spreading
code
involves correlation of the received signal with the delayed
version of the spreading signal (despreading operation)
In other words, the received signal is multiplied again by a
synchronized version of the PN code
Channel
Transmitter
dk t
ak t
Acos (c t+ k )
n (t)
PN signal
Generator
r(t)
T
X
()dt
a k (t-)
rt
s kl (t)
Acos (c t+ k )
Receiver
Demodulator
y t
2 Pk cos ( c t k )
Prof. Okey Ugweje
654
T
0 b dt
Decision
Device
skl
Notice that the despreading operation is similar to the

spreading operation
ak (t Td )
655
Prof. Okey Ugweje
656
CDMA system model (K active users)

d1 (t)
a1 (t) cos (c t+1 )
( i 1)Tb
z1 iT
n( t )
d K (t)
a K (t) cos (c t+ K )
Assuming user #1 is our reference user.
+
n( t )
r (t )
T
X
()dt
Assume that bit zero is transmitted and is being detected

(i.e., i = 0)
sk (t)
z1 0 b r t a1 t cos 2 f c t dt
a (t-)
A c cos (c t+ k )
Substituting
K
z1i 2 Pk kl 0 b ak t - kl a1 t d k t - kl cos c t cos c t kl
Using a simplified diagram, can determine the received signal
k 1 l 1
rk (t ) yk t n(t )
0 b n t a1 t cos c t dt
k 1
K
r t a1 t cos 2 f c t dt
2 Pk kl ak t - kl d k t - kl cos 2 f c t kl n(t )
k 1 l 1
Prof. Okey Ugweje
657
658
Spread Spectrum (SS) is a modulation technique where the

bandwidth of the transmitted signal is made to be greater than
the Bmin required for transmission
The data is scattered (spread) across the available frequency
band in a pseudo random pattern
The idea behind SS is to transform a signal with bandwidth B
into a noise-like signal of much larger bandwidth Bss
Module 6
Spread Spectrum (SS)
What is Spread Spectrum?
Significance of Spreading
Classifications/ Benefits/Applications of Spread Spectrum
Direct Sequence Spread Spectrum
Frequency Hopped Spread Spectrum
Direct Sequence vs. Frequency Hopping
Prof. Okey Ugweje
What is Spread Spectrum? - 1

Prof. Okey Ugweje
659
Prof. Okey Ugweje
660
Spreading Action
At the transmitter, the baseband signal m(t), is usually
spread by a pseudo-noise (PN) code sequence p(t)
Spreading is achieved by modulating the original
signal with a pseudo-random code sequence p(t)
The code sequence p(t) is independent of the data
sequence m(t)
In Spreading the signal

The original signal is embedded in noise (see fig.)
Power of spread signal = Power of original signal
Total power is the area under the spectral density
curve (see fig.)
Prof. Okey Ugweje
661
signals with equivalent total power may have

either a large signal power concentrated in a
small area or a small signal power spread over a
large area
Typically, power of SS signal is spread between
10-30 dB
i.e., power is spread over 10-1000 times original
power
Make signal resistant to noise, interference, and
snooping
Increases the probability of correct reception
Prof. Okey Ugweje
Significance of Spreading - 1
Despreading
At the receiver, the received signal r(t) is despread by the
same amount
Processing gain (PG or Gp) or spreading factor is

defined as
PG G p
by cross-correlating r(t) by a locally generated version of the PN

sequence p(t)
Cross-correlating with the correct sequence recovers the

original data
Is evident from Shannon's capacity equation
C B log 2 1
N
C = channel capacity in bits

B = bandwidth in hertz
S = signal power
N = noise power
B
Spread Bandwidth
ss
Information Bandwidth B
Gp is the improvement gained by spreading the BW

Gp determines the # of users that can be allowed in a system
Gp determines the amount of multipath effect reduction
Gp determines the difficulty of jamming or detecting a signal
Gp may be viewed as performance increase achieved by
spreading
It can be used to describe the signal fidelity gained at the
cost of bandwidth expansion
Observe the effect of increasing the bandwidth B

If B is increased, we may decrease SNR without decreasing
capacity
Prof. Okey Ugweje
662
663
Prof. Okey Ugweje
664
Significance of Spreading - 2
It is through Gp that increased system performance

is achieved without requiring a high SNR
Gp (# of chips per data symbol ) can also be written
as
T R
B
G s c ss
p T
R
Rs
2
s
c
For SS systems, it is advantageous to have Gp as

high as possible
Prof. Okey Ugweje
665
Classifications of Spread Spectrum - 1
Carrier is unpredictable (pseudo-random noise) and

is wideband
BW of the transmitted signal must be greater then the
BW of the data signal
BW of transmitted signal must be determined by some
function that is independent of the message and is
known to the receiver
Despreading involves cross correlation of the received
signal with a synchronously generated replica of the
wideband carrier
Prof. Okey Ugweje
666
Classifications of Spread Spectrum - 2
Direct Sequence Spread Spectrum (DS-SS)

Signal is modulated a 2nd (or 1st) time using a
wideband spreading signal/code
Frequency Hopping Spread Spectrum (FH-SS)
fc is randomly switched from one band to another
during radio transmission according to some specified
algorithm
Time Hopping Spread Spectrum (TH-SS)
The signal hope within a particular time frame
Only one time slot in a frame is modulated
Multi-Carrier Spread Spectrum (MC-SS)
Different carriers are used to transmit the signal
Hybrid Forms of Spread Spectrum

These techniques implement SS in different ways, but
implementations requires:
Signal spreading by means of a code
Synchronization between pairs of users is required
Ensure that some signals do not overwhelm others
(power control)
Uses source and channel coding to optimize
performance
Direct Sequence and Frequency Hopping techniques
are the two most popular SS techniques
Prof. Okey Ugweje
Prof. Okey Ugweje
667
668
Benefits and Applications of SS - 1
Anti-jam (AJ) capability (especially narrow-band (NB)

jamming)
AJ capability is due to the unpredictable nature of the carrier
signal
Since NB interference affects only a small portion of the
spectrum, it is difficult to jam the entire spectrum
Because of the difficulty to jam or detect SS signals, the first
applications were in the military
Covert operation or low probability of intercept (LPI)
LPI can be achieved with high Gp and unpredictable fc
When power is spread thinly and uniformly in freq domain,
detection by surveillance receiver is difficult
Multiple-access capability
SS systems are used for random and multiple
access systems
Users can start their transmission at an arbitrary
time without worrying about channel saturation
Multipath protection
SS implies a reduction of multipath effects, hence a
reduction in fading
i.e., high time resolution is attained by the correlation
detection of wide-band signals
Prof. Okey Ugweje
Prof. Okey Ugweje
669
670
Secure communications
SS systems achieves privacy due to unknown
random codes
Since code is unknown to a hostile user,
detection is difficult
Cryptographic capabilities result when the data
cannot be distinguished from the carrier to an
unauthorized observer
In this case, SS carrier is like a key in a cipher
system
A system using indistinguishable data and SS
carrier modulation is a form of privacy system
Low power spectral density (PSD)

Spreading over a large frequency-band reduces the
PSD, while Gaussian Noise level increases
This improved the spectral efficiency in some
special circumstances
Interference limited operation
Performance is limited by interference rather than
noise
Transmitter-receiver pairs using independent
random carriers can operate in the same BW with
minimal co-channel interference
Prof. Okey Ugweje
671
Prof. Okey Ugweje
672
Direct Sequence Spread Spectrum - 1
Definition:
K = number of users, k = 1, 2, , K
m(t) = user data signal with bit duration, Tb
p(t) = spreading code sequence (pulse or symbol of the PN
code) or chip with duration Tc
Note that Tc << Tb
DS-SS Transmitter
Narrowband
Data In
DS-SS
Modulation
Multipath
Channel
Narrowband
Data Out
m(t)
Spreading
Process
Data Bits
m(t)
Both m(t) and p(t) has amplitude 1 (anti-podal or polar)

In DS-SS, m(t) is directly multiplied by p(t)
B= bandwidth of data signal m(t)
Bss = bandwidth of spread signal s(t)
Spreading
Process
Data Bits
m(t)
Tb
+1
Spread Signal
s(t)
Code Sequence, p(t)
each Tb is coded into a spreading sequence of Gp chip durations
-1
p(t)
+1
Spread Signal
s(t)
chip
-1
PN Code Sequence
p(t)
1 1 0 1 1 0 1 0 0 0 1 0 1 0 0 1 0
m(t) x p(t)
+1
Note that Bss >> B
Gp
Please note that m(t) and p(t) are completely independent

Prof. Okey Ugweje
Diversity
Receiver
673
Tb Rc Bss
Tc Rb 2 Rb
Prof. Okey Ugweje
-1
1 1 0 1 1 0 1 0 1 1 0 1 0 1 1 0 0
674
For example in IS-95, we have
Each Tb is coded into a sequence of Gp chips

T
Gp b
Tc
m(t)
-V
+1
Bandwidth Expansion Factor
Tc
0 0
m(t) x p(t)
1
1 0
Code Length
+V
-V
Baseband
BPF
Sss(t)
Transmitted
Signal
PN Code
Generator
IS-95
LO @ fc
Chip Clock
T 12288
106
.
Gp b =
Tc
9.6x103
128
Tc
Prof. Okey Ugweje
Message
Processing Gain
chip
p(t)
-1
This increases the rate by a factor of Gp

Each binary chip can change with probability 0.5 in Tc sec.
Tb
+V
First the data symbols m(t) are spread into p(t)m(t)

Then spread signal is modulated (usually by MPSK)
We must have
Tb
Gp
sss (t )
675
Prof. Okey Ugweje
2 Es
m(t ) p (t ) cos 2 f c t
Ts
Tb G pTc
f c
1
;
Tc
Tb
Tc
676
DS-SS Receiver
fc of DS is fixed, but m(t) is spread out into a much larger BW (at

least 10 times) by using PN code sequence
Both m(t) and Sss(t) signal use same amount of transmit power
However, the PSD of Sss(t) is much lower than that of m(t)
As a result, it is more difficult to detect the presence of
Sss(t)
In this case, the power density of m(t) is 10 times higher
than Sss(t), assuming the spreading ratio is 10
If there is an interference or jammer in the same band, it will be
spread out during the spreading operation
Hence, its impact is greatly reduced
i.e, the offending jammer's power is reduced by at least 90%
At the Rx the spread signal Sss(t), is despread in a similar
manner to recover m(t)
r(t) is first demodulated and then despread

Demodulation is accomplished in part by re-modulation
with a PN spreading (coherent detection)
The correlation of r(t) with the delayed version of the p(t)
(despreading operation)
rt
Demodulator
y t
2 P cos (c t )
Prof. Okey Ugweje
Tb
0 dt
Decision
Device
p(t Td )
677
Frequency Hopped Spread Spectrum - 1
Prof. Okey Ugweje
678
FH is the repeated switching of fc from one band to

another during transmission
Radio signal hops from one fc to another at a specific
hopping rate and sequence that appears to be random
(see animated)
Typically, each fc is chosen from a set of 2k frequencies

spaced Tb
The # of discrete frequency determines the BW of the
system
Gp is directly dependent on # of available freq choices for
a data rate
PN code does not directly modulate the data, but is used
to control the hopping sequence of fc
The fi(t) output of the Tx

jumps from one value to
another based on the
pseudo-random input
from the code generator
m(t )
Overall BW required for FH is much wider than that required to

transmit the same info using only one fc
Each fc and its associated sidebands must stay within a defined BW
Prof. Okey Ugweje
679
p( t )
2 P cos( c t )
Prof. Okey Ugweje
680
Minimum time required to change the frequency is

dependent on
the chip rate
the amount of redundancy used,
the distance to the nearest interference source
To successfully jam a hopper, either the entire band

must be saturated with noise or jamming source must
be able to track the hopping sequence
Neither of these scenarios is likely to occur
naturally, and they are quite difficult to achieve
intentionally
FH-SS enjoys jamming & multipath immunity, as in
DS-SS
If data cannot be received on a particular channel due
to fading, hopper moves to an unfaded channel and
retransmits the data
FH is less effected by the Near-Far problem
Other FH transmitters will be

using different patterns, which
usually will be on non-interfering
freqs
At Rx, FH is removed by mixing
with a local oscillator signal
which is hopping synchronously
with received signal
Prof. Okey Ugweje
r (t )
m t
681
Prof. Okey Ugweje
682
FH sequences have only a limited number of hits

with each other
This means that if a near interferer is present, only
a number of frequency-hops will be blocked
instead of the whole signal
Usually FH is accomplished by multiple frequency
code selected FSK
Obtaining a high Gp is hard because of the
requirement that a frequency synthesizer be able
perform fast-hopping over fc
The faster the hopping-rate the higher the Gp
FH may be classified as either fast or slow

Slow FH is when the hopping rate is less than the
data rate
single hop per symbol bit
Fast FH is the converse
multiple hops per symbol bit
Hopping sequence is designed for allowing
orthogonality in cells and
minimum correlation with respect to intercell
interference
The motivation and advantages of FH is similar to that
of DS system
Prof. Okey Ugweje
Prof. Okey Ugweje
683
684
Direct Sequence vs. Frequency Hopping - 1
Processing Gain
FH does not spread the signal no processing gain from spreading
Power Usage
FH requires more power to achieve same SNR compared to DS
Synchronization
Communication in FH is more difficult to synchronize compared to the DS
since both time and fc need to be in tune
In DS, only the timing of the chips needs to be synchronized since the
carrier fc is fixed
Latency Time
FH spend more time to search the signal to lock to it (longer latency time)
DS radio can lock-in the chip sequence in just a few bits
Usually, to make the initial synchronization possible, the hopper will park
at a fixed fc before hopping. If the jammer happens to locate at the same fc
as the parking fc, the hopper will not be able to hop at all!
And once it hops, it will be very difficult, if not impossible to re-synchronize
if the Rx ever lost sync
Complexity and Cost

FH is usually more costly and more complicated than the DS because it
needs extra circuits for hopping and synchronizing
Performance in Multipath
FH performs better than DS in multipath fading environment
FH does not stay at the same fc and a null at one fc is usually not a null
at other fc (survives multipath environment better)
Capacity
FH can usually carry more data than the DS since FH is completely
narrowband at all times
Interference Rejection Capability
FH reduces its impact by avoiding the jammer and DS reduces its impact
by spreading or diluting the effect of the jammer (net effect is the same)
Application
Hence FH is more popular for voice than data communication because of
their higher error tolerance
Prof. Okey Ugweje
Prof. Okey Ugweje
685

Potable Comparison
Direct Sequence
Easy and Simple
Use Lower Power
Short Latency Time
Quick Lock-In
Short Indoor Range
Low Data Rate
Prof. Okey Ugweje
Frequency Hopper
Complicated
Use Higher Power
Long Latency Time
Slow Lock-In
Long Indoor Range
High Data Rate
Better for multipath channel
Less susceptible to
jamming
687
686

Digital Communications Lecture Guide Contents: CME 624 May 2016

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Digital Communications Lecture Guide Contents: CME 624 May 2016

Caricato da

Copyright:

Formati disponibili

Lecture Guide Contents

Department of Communications Engineering

Department of Communications Engineering

Module 2: Source Encoding & Decoding

Federal University of Technology, Minna

Lecture Guide Contents

Module 3: Baseband Communication

Prof. Okey Ugweje

Federal University of Technology, Minna

Digital Communication System

Department of Communications Engineering

Department of Communications Engineering

Module 4: Bandpass Communication

Module 6: Spread Spectrum

Module 5: Multiplexing and Multiple Access

Prof. Okechukwu C. Ugweje

Module 1: Introduction and Overview

Federal University of Technology, Minna

Review of Linear Systems (Signals and Systems)

Prof. Okey Ugweje

Federal University of Technology, Minna

Digital Communication System

Digital Communication System

Department of Communications Engineering

Introduction and Handout

Some of the materials are included in this section for

Federal University of Technology, Minna

Department of Communications Engineering

Signals are used to convey information

Signals and Systems

Federal University of Technology, Minna

Digital Communication System

Department of Communications Engineering

Prof. Okey Ugweje

Prof. Okey Ugweje

Prof. Okey Ugweje

Federal University of Technology, Minna

Department of Communications Engineering

Department of Communications Engineering

Mathematically, a signal is defined as a function of one

Mathematically, a signal is defined as a function of one or

Sometimes the functional dependence on the independent

Sometimes the functional dependence on the

e.g., sum of sinusoid of different amplitudes, frequency & phase

e.g., sum of sinusoid of different amplitudes,

Signals are the inputs outputs, and internal functions that

Prof. Okey Ugweje

Federal University of Technology, Minna

Prof. Okey Ugweje

Federal University of Technology, Minna

Department of Communications Engineering

Department of Communications Engineering

The variable time may be continuous or discrete and the

Physical realizable signals must

All real-world signals will have these properties

These types of signals occur at different stages of the

Other variables (distance, angle, etc.) can also be the

Federal University of Technology, Minna

Prof. Okey Ugweje

Federal University of Technology, Minna

Department of Communications Engineering

Department of Communications Engineering

When signals are corrupted by noise, they no longer convey