CSE311 1stPartMerged

CSE 311:
Data Communication
Instructor:
Dr. Md. Monirul Islam
Course Outline
• Introduction to digital data • Analog Modulation
communication; • Pulse modulation
• Introduction to signals • Inter symbol interference
• Review of Fourier Transform • Pulse shaping
• Frequency Response of • Digital modulation
Linear Systems; • Multiple access techniques:
• Sampling theorem; TDM, FDM; Random processes;
Quantization; • Additive White Gaussian Noise
• Line coding; (AWGN);
• Error rate due to noise
• Introduction to information
theory; Concept of channel
coding and capacity.
Course Objectives
• Learn continuous and discrete-time signals, linear systems and their
properties
• Learn and apply Fourier Transform for analyzing signals and linear
systems,
• Investigate the concepts of modulation, multiplexing and multiple access
• have in-depth knowledge and understanding of Data Comm. Techniques
• identify and compare pros and cons of different CoE techniques
Course Outcomes
CO Statement Domains and Delivery
CO Correspondi
After undergoing this course, students Taxonomy Method(s) and Assessment Tool(s)
No. ng PO(s)*
should be able to: level(s)** Activity(-ies)
Understand and analyze continuous
Class Tests or
and discrete-time signals, linear
Assignments or
CO1 systems, their properties, and the PO1, PO2 C2, C4 Lecture
Projects, and Final
sampling process in time and frequency
Exam
domains
Class Tests or
Apply mathematical concepts for Assignments or
CO2 PO1 C3 Lecture
analysis of signals and linear systems Projects, and Final
Exam
Class Tests or
Understand and analyze underlying
Assignments or
CO3 notions of analog and digital - C2, P7 Lecture
Projects, and Final
communication systems.
Exam
Evaluate and analyze different Class Tests or
modulation, multiplexing and multiple PO2, PO4, Assignments or
CO4 C6, A5 Lecture
access principals, practices and PO5 Projects, and Final
measures. Exam
Text Books
1. Modern Digital and Analog Communication Systems,
International 4th edition, B P Lathi and Zhi Ding
2. An Introduction to Analog and Digital Communications,
2nd edition, S. Haykin and M. Moher
Assessment
• Class Tests: 20%

• Attendance:10 %
• Term final: 70%
Continuous Assessment
• 2 Class Tests in the first part

– 3rd week: Tuesday
– 5th week: Tuesday
in absence of central CT routine

Motivation
• CoE made life easy
• Most of the PhD’s in the Department from

communication background
– Funding is easier
• See: Maths are working

• Application to other fields
– Computer networks
– Digital signal processing
– Image/video processing
– Computer vision
Communication systems
• Continued from ancient times
– Runners
– Carrier pigeons
– Light, mirror, lens
– Smoke and fire
– Runners
– Carrier pigeons
– Smoke and fire
• Slow
• Unreliable
– Runners
– Carrier pigeons
– Smoke and fire
• Slow
• Unreliable
• However, that was enough for the time

• Electrical communication systems
– Reliable
– dependable
– economical
• Electrical communication systems
– Reliable
– dependable
– economical
• Email
• E-commerce
• Intercontinental business meeting, zoom, team, and many more…
Communication system
breakthroughs
• Telegraph (1844, Samuel Morse)
– Transmitted “What hath God wrought” between
Washington, D.C. and Baltimore, Maryland.
breakthroughs
• Telephone (1875, Alexander Graham Bell)
– Made real-time transmission of speech by
electrical
encoding and
replication
of sound
breakthroughs
• Radio (1901,Marconi ) and TV (1925, Jenkins)
broadcasting
• 1906:
– AM radio
• 1935:
– FM radio
• 1953
– Color TV
• Recent innovation: internet radio/TV
breakthroughs
• Satellite Communication
– 1955: John R. Pierce proposed
– 1957: Sputnik I launched by Soviet U transmitted
telemetry signals for 21 days.
– 1962: Telstar I launched by Bell Lab.
• Capable of relaying
TV programs
across
the Atlantic.
breakthroughs
• Cellular Communication
– 1947: cellular concept first proposed at Bell labs.
– 1978: first cellular trial by AT&T.
– 1991:first GSM
cellular service
launched in Finland.
– 1996: 1st commercial
CDMA cellular
service launched
in Finland.
breakthroughs
• Wireless LAN
– 1971: 1st wireless
computer network
1997- to date: defined
a no. of different IEEE
Wireless LAN
standards.
breakthroughs
• Internet: Network of Networks
– 1971: ARPANET put into service.
• Later renamed the Internet, in 1985.
– 1990: Tim Berners-Lee
proposed a Hypermedia
software interface
to
Internet
WWW
Components of
Distortion
and Noise
Components of
1. Sender and Receiver

– Human, computing
devices such as computers
smart phones etc.
2. Message
Distortion
– Information to be communicated. and Noise
5. Switching Devices
3. Signal - Switch, router, hub etc.
– Electrical, Optical, Sound 6. Protocols: Rules of
4. Medium/Channel Communication
– Air, Wire, Fiber, Vacuum – HTTP, SMTP, FTP etc.
Components of
Distortion
and Noise
Main challenges
Challenges of
1. Distortion
–systematic undesirable changes in signals
–Linear or non-linear
2. Noise
– Unwanted signal that interfere with the transmitted
signal
– Random signals from internal or external sources
Components of
Most important Distortion

and Noise
component
Messages/Signals: Definition
- A signal is a set of information or data.
- A signal is a function of independent variables
that carry some information.
- A signal is a physical quantity that varies with
time, space or any other independent variable by
which information can be conveyed.
Example of Signals
• Voice signal
• Telephone or television signal
• Monthly sales figure
• Opening or closing stock prices
• Charge density over a surface
- In this course we deal with signals that are functions
of time.
Signal representation: Time Domain
g(t)
amplitude
t
time
CSE 311:
Data Communication
Instructor:
Messages/Signals: Definition
- A signal is a set of information or data.
- A signal is a function of independent variables
that carry some information.
- A signal is a physical quantity that varies with
time, space or any other independent variable by
which information can be conveyed.
Example of Signals
• Voice signal
• Telephone or television signal
• Monthly sales figure
• Opening or closing stock prices
• Charge density over a surface
- In this course we deal with signals that are functions
of time.
Signal representation: Time Domain
g(t)
amplitude
t
time
Classification of Signals
• Based on continuity in time axis
– Continuous time
– Discrete time
• Based on continuity in amplitude axis
– Continuous amplitude
– Discrete amplitude
Continuous amplitude Discrete amplitude
Continuous time
Discrete time
Analog and Digital Signal
Analog Signal
- Continuous amplitude, i.e., takes any value in a
continuous range.
- May be both continuous and discrete time.
Digital Signal
- Discrete amplitude, i.e., amplitude can take only
a finite number of values.
- Values need not be always integer.
- Not necessarily always binary, rather M-ary.
- May be both continuous and discrete time.
Analog and Digital Signal: Examples
Analog Digital
Thermometer
Clock
Blood Pressure
Monitor
Components of
Distortion
and Noise
Main challenges
Challenges of
1. Distortion
–systematic undesirable changes in signals
–Linear or non-linear
2. Noise
– Unwanted signal that interfere with the transmitted
signal
– Random signals from internal or external sources
Benefits of Digital
Message/Signal
1. Quality, e.g., enhanced noise immunity
2. Economics
Benefits of Digital
Message/Signal
1. noise immunity
• Represented by binary or M-ary pulses
Sent
Benefits of Digital
Message/Signal
1. noise immunity
Sent
Received
w/o noise
Benefits of Digital
Message/Signal
1. noise immunity
Sent
Received
w/o noise
Received
with noise
Benefits of Digital
Message/Signal
1. noise immunity
• Recovered despite small distortion and noises
Sent
Received
w/o noise
Received
with noise
Recovered,
though delayed
Repeater’s Role in Digital
Message/Signal
• Distortion and noise are unavoidable in channel
• Repeaters and nodes regenerates digital pulses
Message/Signal
Channel
Input digital Distorted

signal digital
signal
Message/Signal
Channel repeater
Input digital Distorted Regenerated

signal digital digital signal
signal
Repeater’s Role in Analog
Message/Signal
• Repeaters are filters and amplifiers in analog signals
• Amplifier amplifies both signal and noise
Message/Signal
Channel amplifier
Input Distorted Signal with

analog analog amplified
signal signal noise
Message/Signal
Channel amplifier
Input Distorted Signal with

analog analog amplified
signal signal noise
Noise accumulates along the path! No improvement at all !!

Analog to Digital Conversion
of Message/Signal
• A/D conversion enables digital communication to
convey analog signals
of Message/Signal
• Analog signal characteristics
– values are continuous
– defined over continuous/discrete time
of Message/Signal
• Digital signal characteristic's
– values are a finite discrete set
– defined over preferably discrete time
of Message/Signal
• 2 major steps
– Sampling
– Quantization
of Message/Signal
• Sampling
– Governed by Nyquist ’s Sampling theory
– Selects points for sampling
of Message/Signal
• Sampling
of Message/Signal
• Sampling
of Message/Signal
• Quantization
– Values replaced by a set of L distinct values
– Usually L = 2k
of Message/Signal
• Quantization
– Values replaced by a set of L distinct values
– Usually L = 2k
of Message/Signal
• Increasing sampling rate retains original shape
of Message/Signal
• Increasing sampling rate retains original shape
Remember
Nyquist’s theorem!
of Message/Signal
• Increasing Quantization level L
– increases accuracy
– more noise immunity
– but requires higher channel bandwidth
of Message/Signal
Recall this figure
Sent
Received
w/o noise
Received
with noise
Recovered,
though delayed
of Message/Signal
• Detection is easy when A >> noise
Recall this figure • Usually A >> 5-10 times of noise
Sent
Received
w/o noise
Received
with noise
Recovered,
though delayed
of Message/Signal
• Problem: quantization error is unavoidable
of Message/Signal
of Message/Signal
of Message/Signal
of Message/Signal
of Message/Signal
Quantization
error can be
minimized
increasing L
Representation of Digital
Signal: Pulse Coded Modulation
• Assume, No. of quantization level, L = 2

• Assume, No. of quantization level, L = 2

– Easy to represent or transmit
• If No. of quantization level, L >> 2

• If No. of quantization level, L >> 2 Each sample

is represented
by one of L
mp levels
m(t) Quantized samples m(t)
Allowed quantization levels
2m p
L
-mp
• If No. of quantization level, L >> 2

– Solution is PCM
– Each quantized value is represented by a sequence of
binary pulses.
• Assume, L =16
mp
2m p
L
-mp
• Assume, L =16
– Each quantized value is represented
by a sequence of FOUR binary
pulses.
CSE 311:
Data Communication
Instructor:
Channel & signal
Characteristics
• 2 important properties Distortion

– Bandwidth, B and Noise
– Signal power, Ps
Channel & signal
Characteristics
• Bandwidth
– Channel Bandwidth
– Signal bandwidth
Channel & signal
Characteristics
• Bandwidth
• Range of frequencies that a channel can transmit
• Ex: if transmit 0-5 kHz frequencies, Channel bandwidth B = 5 kHz
Channel & signal
Characteristics
• Bandwidth
• Maximum frequency that is available in a signal
Channel & signal
Characteristics
• Bandwidth
Fourier Transform
f
Frequency
Time domain
domain
Channel & signal
Characteristics
• Bandwidth
maximum
frequency
f
Channel & signal
Characteristics
• Bandwidth
– Signal bandwidth B
maximum
frequency
f
Channel & signal
Characteristics
A complex signal with

multiple frequencies
Channel & signal
Characteristics
A complex signal with

multiple frequencies Individual signals each with
a single frequency
Channel & signal
Characteristics
• Frequency
– Change in signal values
– Faster change in values means higher frequencies
– High frequency signals

• Sports/battle scene
– Low frequency signals

• News/sleeping animal videos
Channel & signal
Characteristics
• Frequency
– Compressing in time, increases frequency, means higher channel
Bandwidth
Channel & signal
Characteristics
• Frequency
Bandwidth
Compressed Expanded in
in time frequency
Channel & signal
Characteristics
• Frequency
Bandwidth
Duration = 0.1 sec, Frequency = 2.5/0.1= 25 Hz

Channel & signal
Characteristics
• Frequency
Bandwidth
Duration = 0.1 sec, Frequency = 2.5/0.1= 25 Hz
Duration = 0.05 sec, Frequency =2.5/0.05=50 Hz

Channel & signal
Characteristics
• More transmission speed requires channel with higher
bandwidth
Channel & signal
Characteristics
bandwidth
N pulses/sec
Channel with B
Channel & signal
Characteristics
bandwidth
N pulses/sec
Channel with B
increase transmission speed K times

Channel & signal
Characteristics
bandwidth
N pulses/sec
Channel with B
NK pulses/sec
Channel & signal
Characteristics
bandwidth
N pulses/sec
Channel with B
NK pulses/sec
New requirement:
Channel with KB
Channel & signal
Characteristics
• Signal Power, Ps
– Dual role
• Higher Quality
• Less channel bandwidth
of Message/Signal
• Detection is easy when A >> noise
• Usually A >> 5-10 times of noise
Sent
Received
w/o noise
Received
with noise
Recovered,
though delayed
Channel & signal
Characteristics
Signal Power, Ps ∞ Amplitude, A

=> Increase in A means increase in PS
SNR = Signal / Noise

Channel & signal
Characteristics
• Signal Power, Ps for higher quality
– Ps++ -> SNR++
– maintains minimum SNR for longer distance
– Higher SNR means
• more noise immunity
• easier detection of pulses
Channel & signal
Characteristics
• Signal Power, Ps needs less channel bandwidth
– Higher signal Ps eases the channel
Channel & signal
Characteristics
– Shannon’s limit on channel capacity
C  B log 2 (1  SNR ) bit s

Channel & signal
Characteristics
– Shannon’s limit on channel capacity
C  B log 2 (1  SNR ) bit s
What happens if SNR = ∞?

Channel & signal
Characteristics
Baseband signal
Channel & signal
Characteristics
Baseband signal 1
Baseband signal 2
Channel & signal
Characteristics
• Assume, these 2 signals need
to be transmitted
Baseband signal 1
Baseband signal 2
Channel & signal
Characteristics
to be transmitted
– 2 different channels?
Baseband signal 1
Baseband signal 2
Channel & signal
Characteristics
to be transmitted
– Even, channel bandwidth and
signal bandwidth may NOT
match!!
Baseband signal 1
Baseband signal 2
Channel & signal
Characteristics
to be transmitted
– 2 different channels?
Baseband signal 1
Solution is
modulation
Baseband signal 2
Channel & signal
Characteristics: modulation
• Use carrier signal to shift these
2 signals to different frequency
positions
Baseband signal 1
Baseband signal 2
Channel & signal
2 signals to different frequency
positions
Baseband Baseband
After signal 1 signal 2
modulation:
Frequency
axis
0 f1 f2
Channel & signal
• other examples of modulation
(shown in time domain)
Baseband
signal
Carrier signal
Modulated
signal
Channel & signal
Baseband
signal
Carrier signal
Frequency
Modulated modulation
signal (FM)
Channel & signal
Baseband
signal
Carrier signal
Amplitude
Modulated
signal
modulation
(AM)
Channel & signal
• Phase modulation (PM)
• Changes phase angle of the signal
Channel & signal
Other reasons for modulation
Higher Generating
Low frequency
wavelength antenna requires
higher dimension
Channel & signal
Higher Generating
Low frequency
higher dimension
• Speech signal characteristics

– Frequency range: 100 – 3000 Hz
– Wavelength 100 to 3000 km
– requires impractically large antenna
Channel & signal
Lower Generating
Modulating a High
frequency carrier
lower dimension
Channel & signal
Lower Generating
Modulating a High
frequency carrier
lower dimension
• 10 MHz carrier signal characteristics

– Wavelength 30 m
– requires antenna of size 3m
Channel & signal
2 signals in different frequency
positions
recall this example!
Baseband Baseband
modulation:
Frequency
axis
0 f1 f2
Channel & signal
positions
also called
Frequency division
multiplexing (FDM) Baseband
Baseband
modulation:
Frequency
axis
0 f1 f2
Channel & signal
• Time division multiplexing
(TDM)
– Interleave pulses from different
signals in time domain signal
Channel & signal
(TDM)
Channel & signal
Characteristics: DeModulation
• Done at the receiving end
– Bandpass filter separates appropriate signal
– Makes necessary corrections for amplitude, frequency
and phase changes
CSE 311:
Data Communication
Instructor:
Channel & signal
positions
also called
Frequency division
multiplexing (FDM) Baseband
Baseband
modulation:
Frequency
f1 f2 axis
0
Channel & signal
(TDM)
Channel & signal
Characteristics: DeModulation
• Done at the receiving end
– Bandpass filter separates appropriate signal
– Makes necessary corrections for amplitude, frequency
and phase changes
Signal Characteristics: digital
source coding and error
correction coding
• 2 opposite procedures
• Digital source coding

– An aggressive measure to reduce redundancy
• Error correction coding

– A systematic addition of redundancy to detect/correct errors
Digital source Coding
• Removes redundancy
• Uses bandwidth as little as possible
• Related to randomness/predictibility in data
Less More Higher Shorter

randomness predictibility redundancy codeword

Higher Less Lower Longer

• Example: Morse code
– Frequently occurring letters e, t, a: shorter codes
– Less frequent letters: longer codes

Higher Less Lower Longer

No NO news Nothing to
randomness at all transmit
• A message with probability p,

its randomness or entropy = log (1/p) = -log (p)
No NO news Nothing to
randomness at all transmit
• A message with probability p,

its randomness or entropy = log (1/p) = -log (p)
What happens when p = 0 and p =1?

Error correction coding
• Add systemic redundancy
• 50% of English text is redundant
• Difficult to recover if error occurs in reduced text
• Difficult to recover if error occurs
• Error Detection: adding parity bit
Assume to transmit 0001

Add this bit to make

even no. of 1’s
Add this bit to make

even no. of 1’s
Detects single bit error; however

cannot detect even no. of errors or
cannot locate or correct single bit
error
Signals and systems
Topics
• Definition, classification and properties of signals

• Examples of some useful signals
• Fourier Series of Periodic Signals
Review
• Based on continuity in time axis
– Continuous time
– Discrete time
• Based on continuity in amplitude axis
– Continuous amplitude (analog)
– Discrete amplitude (digital)
Review
Continuous amplitude Discrete amplitude
(analog) (digital)
Continuous
time
Discrete time
Periodic and Aperiodic signal
• Signal g(t) is periodic for a positive constant T0 so

that
g(t) = g(t + T0) for all t
• Smallest T0 is its period
• A signal is aperiodic if NOT periodic

Periodic
signal
Period, T0 =2
Aperiodic
signal
• Periodic signal starts at t = -α and continues forever
Periodic
signal
Period, T0 =2
• cannot start at an finite time, say, t = 0, otherwise
g(t) = g(t + T0) cannot be satisfied
Zero or Undefined at
this region
Started here
Zero or Undefined at
this region
At t = -T0, g(t) ≠ g(t + T0)

Energy and Power signals
• Energy Signal: finite Eg


2
E g   g (t )dt  

• Power signal: finite Pg
1 T 2
2
0  Pg  lim  g (t )dt  
T  T T 2
Energy and Power signals

2
E g   g (t )dt  

1 T2 2
0  Pg  lim  g (t )dt  
T  T T 2
• Finite Eg signal has zero Pg
• Finite Pg Signal has infinite Eg
• A signal CANNOT be both energy and power signal
• Real life signals are energy signals
• Power signals have infinite duration; impractical to
generate
• Periodic signals are power signals
Deterministic and Random
signals
• Deterministic
– has complete physical description, mathematically or
graphically
• Random
– has only probabilistic description, e.g., mean value, rms,
distribution
• All message signals are random
Signal Properties
Time shifting property
• Whatever happens in g(t) at t second also happens

in (t) T seconds later at instant t + T
Or,
Signal Properties
Time shifting property
Beginning T
seconds later
Beginning T
seconds earlier
Signal Properties
Time scaling property
• Compression or expansion
• Compression:
– Whatever happens in g(t) at t second also happens in (t)
at t/a
,a>1
Signal Properties
• Compression or expansion
• Compression:
at t/a
,a>1
• Expansion:
at at
,a>1
Signal Properties
Compression
Expansion
Signal Properties
Time inversion property
• Mirroring about vertical axis

• Whatever happens in g(t) at t second also happens
in (t) at –t
• Similar to time scaling where a = -1
Signal Properties
Example of Time inversion property
Unit Impulse Signal
• One of the most important signals, (t)
• Also know as Dirac delta function
 if t  0
 (t )  
 0 if t  0
with the constraint,
  (t )dt  1

Unit Impulse Signal
A function with unit area under curve
Unit Impulse Signal
Compress the function leaving
the area unchanged
Unit Impulse Signal
Keep compressing. . . . .
Unit Impulse Signal
Keep compressing. . . . .
Ultimately we got Dirac Delta function
(t)
t
Dirac Delta
Function
Unit Impulse Signal
 if t  0
 (t )  
 0 if t  0 (t)

  (t )dt  1

t
Unit Impulse Signal
 if t  0
 (t )  
 0 if t  0 (t)

  (t )dt  1

t
• Impulse location is at t =0
Multiplication of a Function
by Impulse
 (t ) (t )   (0) (t )
(t)
t
 (t )
Multiplication of a Function
by Impulse
 (t ) (t  T )   (T ) (t  T ) (t-T)
t
T
 (t )
Sampling Property of the Unit
Impulse Function
 
  (t ) (t )dt   (0)   (t )dt

 
(t)
  (0)
t
 (t )
Impulse Function

(t-T)
  (t  T ) (t )dt


t
  (T )   (t  T )dt T
 (t )

  (T )
Impulse Function
b (t-T)
  (t  T ) (t )dt
a
b t
T
  (T )   (t  T )dt  (t )
a
Impulse Function
b
(t-T)
  (t  T ) (t )dt
a
b a b
t
  (T )   (t  T )dt T
 (t )
a
  (T )
Impulse Function
b
  (t  T ) (t )dt
a
(t-T)
b
  (T )   (t  T )dt a b
t
a T
 (t )
 (T ) a  T  b

 0 otherwise
CSE 311:
Data Communication
Instructor:
Signals and systems
Topics

Unit step signal/function, u(t)
1 if t  0
u (t )  
0 if t  0
Unit step signal/function, u(t)
• Causal function, e,g., starts at t = 0
1 if t  0
u (t )  
0 if t  0
• g(t) is causal if
Use of u(t) to start an
everlasting function from t = 0
e-at
a>0
t
e-at
×
t
Alternate representation of u(t)
in terms of (t)
in terms of (t)
()

in terms of (t)
=1 ()

t
in terms of (t)
=0 ()

t
in terms of (t)
()

t
in terms of (t)
()

t
0 if t  0
u (t )  
1 if t  0
in terms of (t)
()

t
0 if t  0
u (t )  
1 if t  0
in terms of (t)
()

t
in terms of (t)
(t)
t
Fourier Series of a periodic signal
‘Any function that periodically repeats itself can be
expressed as the sum of sines and/or cosines of
different frequencies’
- Joseph Fourier
Fourier Series of a periodic signal
‘Any function that periodically repeats itself can be
different frequencies’
- Joseph Fourier
n 
jn0t
g (t )   n
D e
n   with, T0 = period of g(t)
n  f0 = frequency of g(t)
jn 2f 0 t
 D e
n  
n
1  jn 2f 0 t
where, Dn   g (t )e
T0 T0
Fourier Series
Fourier says:
Any function that periodically repeats itself can be

different frequencies
• Doesn’t matter how complicated the function is!

• The summation is called Fourier Series
Fourier Series
A complicated but periodic function

Complicated
periodic
function
Simpler
sine/cosine
functions
Fourier Series and Light Spectrum
A complicated Simpler
function/signal functions/signals
Representing a Periodic signal using
Fourier Series
Fourier Series
n n 
jn0t j 2 nt
 (t )   n
D e
n  
  n
D e
n  
Here, T0=  and

Fourier Series
Fourier Series
Fourier Series
Fourier Series
n 
j 2 nt
 (t )   n
D e
n  
Fourier Series
• Dn are exponential Fourier spectra
• Dn are complex though (t) is a real periodic
• Dn and D-n are complex conjugates
Fourier Series
–
Fourier Series
–

–
–
Fourier Series
–

–
–
• Therefore,
Fourier Series
• |Dn|are amplitudes and ∠Dn are angles of

exponential spectra
• |Dn|vs. f is an even function
• ∠Dn vs. f is an odd function
Fourier Series
Fourier Series
Note that
ω = ω0n =2f0n
= 2n
Fourier Series
Note that
ω = ω0n =2f0n
= 2n
Negative frequency regions
Meaning of negative frequencies
Consider this real periodic Note that
function with frequency f0 angular
frequency is
ω0 = 2f0
Consider this real periodic
function with frequency f0
Here, angular frequency is |ω0|, a positive

quantity which only indicates rate of sinusoidal
variation
Consider this real periodic
Here, angular frequency is |ω0|, a positive

quantity which only indicates rate of sinusoidal
variation
Cannot explain negative frequency

Consider this complex sinusoid
It both describes the rate variation, |ω0|, and

direction (±)
Consider this complex sinusoid
Im Im
e j 0 t e  j 0 t
Re Re
Fourier series of Some Useful Signals
Periodic
square wave
Period, T0 = 2
Frequency, f0 = 1/T0 = 1/ 2
Periodic
square wave
Period, T0 = 2
Frequency, f0 = 1/T0 = 1/ 2
where, ω0 =2f0
Periodic
square wave
Period, T0 = 2
Frequency, f0 = 1/T0 = 1/2
Periodic
square wave
Period, T0 = 2
Periodic
square wave
Period, T0 = 2
Periodic
square wave
Period, T0 = 2
Periodic
square wave
Period, T0 = 2
D0 is undefined
Periodic
square wave
Period, T0 = 2
1 1
D0   w(t )dt 
T0 T0 2
1
D0 
2
Exponential
spectra of a
Periodic
square wave
f
1
D0 
2
Exponential
spectra of a
Periodic
square wave
CSE 311:
Data Communication
Instructor:
Signals and systems
Topics

Fourier Series of Unit
Impulse Train
(t)
Impulse signal
t
Impulse train
g (t )   T0 (t )
-2T0 -T0 0 T0 2T0 3T0 t

Impulse Train
(t)
Impulse signal
t
Impulse train
g (t )   T0 (t )
-2T0 -T0 0 T0 2T0 3T0 t

n 
Equation of an
Impulse train
g (t )   T0 (t ) = s
T0 (t )    (t  nT )
n  
0
Impulse Train
g (t )   T0 (t )
-2T0 -T0 0 T0 2T0 3T0 t

As it is a periodic signal with period T0
where,
Impulse Train
g (t )   T0 (t )
-2T0 -T0 0 T0 2T0 3T0 t
(t)
t
Impulse Train
g (t )   T0 (t )
-2T0 -T0 0 T0 2T0 3T0 t
(t)
t
Impulse Train
g (t )   T0 (t )
-2T0 -T0 0 T0 2T0 3T0 t
(t)
t
Impulse Train
g (t )   T0 (t )
-2T0 -T0 0 T0 2T0 3T0 t

Impulse Train
g (t )   T0 (t )
-2T0 -T0 0 T0 2T0 3T0 t

Fourier Series is
Fourier’s theorem for
aperiodic Signal
g(t)
t
aperiodic Signal
• aperiodic signal but with finite area under curve can be
represented as integral of sines and cosines

j t
g (t )   G ( f )e df


 jt
where, G ( f )   g (t ) e dt

and ω = 2f
aperiodic Signal
• These two are called Fourier transform pair
 
 j t  j 2ft
G( f )   g (t ) e dt   g (t ) e dt
 
 
jt j 2ft
g (t )   G ( f )e df   G ( f )e df
 
aperiodic Signal
• These two are called Fourier transform pair
 
 j t  j 2ft
G( f )   g (t ) e dt   g (t ) e dt
 
 
jt j 2ft
g (t )   G ( f )e df   G ( f )e df
 
Symbolically shown as,

g (t )  G ( f ) or G(f)  g(t)
Fourier Integral for
aperiodic signal

 j 2ft
G( f )   g (t ) e dt

• G(f) is Fourier spectra and complex even if g(t) is real
• G(f) can be shown as
j g ( f )
G( f )  G( f ) e
aperiodic signal
 
 j t  j 2ft
G( f )   g (t )e dt   g (t )e dt
 
• Replacing f by -f

j 2ft
G ( f )   g (t ) e dt

aperiodic signal
 
 j t  j 2ft
G( f )   g (t )e dt   g (t )e dt
 

j 2ft
G ( f )   g (t ) e dt
Complex 
conjugate 
* j 2ft
G (f)  g (t )e dt provided, g(t) is real

aperiodic signal
 
 j t  j 2ft
G( f )   g (t )e dt   g (t )e dt
 

j 2ft
G ( f )   g (t ) e dt


* j 2ft
G (f)  g (t )e dt  G ( f )

aperiodic signal
• The following is complex conjugate symmetry
G * ( f )  G ( f )
aperiodic signal
G * ( f )  G ( f )
Therefore,
*
G ( f )  G ( f )  G ( f )
aperiodic signal
G * ( f )  G ( f )
Therefore,
*
G ( f )  G ( f )  G ( f )
but
 g ( f )   g ( f )
Conjugate Symmetry Property in
Fourier Series of Periodic signals
–

–
–
• Therefore,
Fourier Transform Example
Find Fourier transform of e-atu(t)
g(t)
e-atu(t)
0 t
  
 j 2ft  at  j 2ft  at  j 2ft
G( f )   g (t ) e dt   e u (t ) e dt   e dt
e
  0
g(t)
e-atu(t)
0 t
  
 j 2ft  at  j 2ft  at  j 2ft
e
  0

g(t)   e ( a  j 2f ) t dt
0
e-atu(t)
0 t
  
 j 2ft  at  j 2ft  at  j 2ft
e
  0

1
g(t)  e  ( a  j 2f ) t
dt 
a  j 2f

e ( a  j 2f )t 

0
0
e-atu(t)
0 t
  
 j 2ft  at  j 2ft  at  j 2ft
e
  0

1
 e  ( a  j 2f ) t
dt 
a  j 2f

e ( a  j 2f )t 

0
0
Now, as t  , e-(a+j2f )t= e-at e-j2ft = 0

  
 j 2ft  at  j 2ft  at  j 2ft
e
  0

1
 e  ( a  j 2f ) t
dt 
a  j 2f

e ( a  j 2f )t 

0
0

  
 j 2ft  at  j 2ft  at  j 2ft
e
  0

1
 e  ( a  j 2f ) t
dt 
a  j 2f

e ( a  j 2f )t 

0
0
Now, as t  , e-(a+j2f )t= e-at e-j2ft = 0 provided a > 0

  
 j 2ft  at  j 2ft  at  j 2ft
G( f )   g (t ) e dt   e u (t ) e dt   e u (t ) e dt
  0

1
 e  ( a  j 2f ) t
dt 
a  j 2f
e  
 ( a  j 2f ) t 
0
0
 1 1 1
 
a  j 2f a  j
1 1
G( f )  
a  j a  j 2f
1 1
G( f )  
a  j a  j 2f
Therefore,
and
Remember, we got this result for a > 0
g(t)
1
e-atu(t) G( f ) 
a  j 2f

e ( a  j 2f )t 

0
1

0 t a  j 2f
Now as t->α, e-(a+j2f )t= e-at e-j2f)t = 0 provided a > 0
What happens if result for a < 0?
g(t)
1
e-atu(t) G( f ) 
a  j 2f

e ( a  j 2f )t 

0
1

0 t a  j 2f
Existence of Fourier Transform
Conditions
• Area under curve must be finite
• The function must be integrable
Dirichlet condition
Conditions
This can be proved from,


 j 2ft
G( f )   g (t ) e dt

Conditions

 
 j 2ft
G( f )   g (t ) e dt   g (t )  e  j 2ft dt
 
Conditions

 
 j 2ft
G( f )   g (t ) e dt   g (t )  e  j 2ft dt
 
Because, 2 + (-3) ≤ |2| + |-3|

Conditions

   
 j 2ft
G( f )   g (t ) e dt   g (t )  e  j 2ft dt   g (t ) 1 dt   g (t ) dt
   
Conditions

   
 j 2ft
G( f )   g (t ) e dt   g (t )  e  j 2ft dt   g (t ) 1 dt   g (t ) dt
   

G( f )   g (t ) dt

Linearity of Fourier Transform
Assume, g1 (t )  G1 ( f ) and g 2 (t )  G2 ( f )
and a1 and a2 are arbitrary constants
Then,
a1 g1 (t )  a2 g 2 (t )  a1G1 ( f )  a2G2 ( f )
Assume, g1 (t )  G1 ( f ) and g 2 (t )  G2 ( f )
and a1 and a2 are arbitrary constants
Then,
a1 g1 (t )  a2 g 2 (t )  a1G1 ( f )  a2G2 ( f )
This means,
a1 g1 (t )  a2 g 2 (t )  a1G1 ( f )  a2G2 ( f )

More general case
a g
k
k k (t )   ak Gk ( f )
k
More general case
a g
k
k k (t )   ak Gk ( f )
k
Superposition Theorem
Area Under Curve of Fourier
Transform G(f)

j 2ft
g (t )   G ( f )e df

Area Under Curve of Fourier
Transform G(f)

j 2ft
g (t )   G ( f )e df

Assume t = 0
 
g (0)   G ( f )e j 2f .0 df   G ( f )df
 
Area Under Curve of Time domain
Function g(t)

 j 2ft
G( f )   g (t )e dt

Area Under Curve of Time domain
Function g(t)

 j 2ft
G( f )   g (t )e dt

Assume f = 0
 
 j 2t .0
G ( 0)   g (t ) e dt   g (t )dt
 
Fourier Transform of Some Useful
Functions
Unit Rectangular Function
Functions
Other names
Box Function
Gate Function
Functions
Expanded
function
Functions
Unit Triangular Function
, ,
, ,
Functions
Sinc Function, sinc (x)
sin x

x
Functions
sin x

x
Functions
• sinc (x) is an even oscillating

function with decreasing
amplitude.
• It has a unit peak at x = 0 and zero
sin x

crossings at integer multiples of . x
Functions
Functions
Unit scaled rectangular function
= g(t)
g(t) =
Functions
= g(t)
g(t) =
Functions
= g(t)
g(t) =
Functions
= g(t)
g(t) =
Functions
= g(t)
Fourier transform pair

CSE 311:
Data Communication
Instructor:
Signals and systems
Topics

Functions
= g(t)

Functions
Unit Impulse function (t)
g(t) = (t)
t
Functions
g(t) = (t)
t
Functions
G(f) = 1
g(t) = (t) 1
f
t 0

Functions
Inverse Fourier Transform of (f)
G(f) = (f)
f
Functions
G(f) = (f)
g (t )  1[ ( f )]
 
  G ( f )e j 2ft df    ( f )e j 2ft df
 
j 2t .0
e 1
Functions
g(t) = 1
1 G(f) = (f)
f
0 t
g (t )  1[ ( f )]
 
  G ( f )e j 2ft df    ( f )e j 2ft df Fourier Transform Pair
 
j 2t .0
e 1
Functions
Inverse Fourier Transform of (f-f0)
G(f) = (f-f0)
f
f0
Functions
G(f) = (f-f0)
g (t )  1[ ( f  f 0 )]
  f
j 2ft j 2ft f0
  G ( f )e df    ( f  f 0 )e df
 
 e j 2tf 0  e j 2f 0t
Functions
G(f) = (f-f0)
g (t )  1[ ( f  f 0 )]
  f
j 2ft j 2ft f0
  G ( f )e df    ( f  f 0 )e df
 
 e j 2tf 0  e j 2f 0t
Fourier Transform Pair
j 2f 0t
e   ( f  f0 )
Functions
G(f) = (f-f0)
j 2f 0t
e   ( f  f0 )
f
f0
Functions
Inverse Fourier Transform of (f-f0) and (f+f0)
G(f) = (f-f0)
j 2f 0t
e   ( f  f0 )
f
f0
G(f) = (f+f0)
 j 2f 0 t
e?   ( f  f0 )
f
-f0
Functions
Inverse Fourier Transform of (f-f0) and (f+f0)
G(f) = (f-f0)
j 2f 0t
e   ( f  f0 )
f
f0
G(f) = (f+f0)
 j 2f 0 t
e   ( f  f0 )
f
-f0
Fourier Transform Pairs

Functions
Fourier Transform of cosine function
Assuming, e j  cos   j sin 

1 j 2f 0t  j 2f 0t
cos 2f 0t  (e e )
2
Functions

1 j 2f 0t  j 2f 0t
cos 2f 0t  (e e )
2

1
cos 2f 0t  [ ( f  f 0 )   ( f  f 0 )]
2
Functions

1 j 2f 0t  j 2f 0t
cos 2f 0t  (e e )
2
cos 2f 0t
t
1
cos 2f 0t  [ ( f  f 0 )   ( f  f 0 )]
2
Functions
Fourier Transform of sine function

1 j 2f 0t  j 2f 0t
sin 2f 0t  (e e )
2j
Functions
Fourier Transform of sine function

1 j 2f 0t  j 2f 0t
sin 2f 0t  (e e )
2j

1
sin 2f 0t  [ ( f  f 0 )   ( f  f 0 )]
2j
g(t)
1
e-atu(t) G( f ) 
a  j 2f

e ( a  j 2f )t 

0
1

0 t a  j 2f
Functions
Fourier Transform of exponential functions
u (t )e  at u (t )e at
Exponential decay fnc Exponential rising fnc

Functions
 
G ( f )   e  at e  j 2ft dt   e ( a  j 2f )t dt
0 0
1
exp( at )u (t ) 
a  j 2f

Exponential decay fnc
Functions
0
G ( f )   e at e  j 2ft dt


  e t ( a  j 2f ) dt exp(at )u (t )
0
1

a  j 2f
Exponential rising fnc
exp(at )u (t )
Functions
Fourier Transform of sgn(t) function
sgn(t)
+1
0 t
-1
Functions
g(t)
+1
t
0
-1
Can be approximated by
Functions
g(t)
+1
t
0
-1
Can be approximated by
Functions
g(t)
+1
t
0
Limiting a to 0
-1
Functions
g(t)
+1
t
0
Limiting a to 0
-1

Functions
Fourier Transform of a periodic signal gT0 (t )
gT0 (t )
-T0/2 +T0/2
Functions
gT0 (t )
-T0/2 +T0/2
g (t )
-T0/2 +T0/2
Functions
gT0 (t )
-T0/2 +T0/2
g (t )
Assume that
g (t )  G ( f )
-T0/2 +T0/2
Functions
gT0 (t )
-T0/2 +T0/2
g (t )
Assume that
g (t )  G ( f )
-T0/2 +T0/2
Functions
gT0 (t )
-T0/2 +T0/2
g (t )
Assume that
g (t )  G ( f )
-T0/2 +T0/2
Functions
gT0 (t )
-T0/2 +T0/2
n 
As it is periodic,
gT0 (t )  c
n  
n exp( j 2nf 0t )
T0 2
1
where, cn  g T0 (t ) exp( j 2nf 0t )dt and
T0 T0 2
Functions
gT0 (t )
-T0/2 +T0/2
T0 2
1
cn  g T0 (t ) exp( j 2nf 0t )dt
T0 T0 2
Functions
gT0 (t )
-T0/2 +T0/2
T0 2
1 g (t )
cn 
T0  gT0 (t ) exp( j 2nf 0t )dt
T0 2
-T0/2 +T0/2
Functions
gT0 (t )
-T0/2 +T0/2
T0 2 T0 2
1 1
cn   g (t ) exp(  j 2nf t ) dt
T0 T0 2
T0 0 
T0  g (t ) exp( j 2nf t )dt
T0 2
0
Functions
g (t )
-T0/2 +T0/2
T0 2 T0 2
1 1
cn   g (t ) exp(  j 2nf t ) dt
T0 T0 2
T0 0 
T0 2
0
Functions
g (t )
 
-T0/2 +T0/2
T0 2 T0 2
1 1
cn   g (t ) exp(  j 2nf t ) dt
T0 T0 2
T0 0 
T0 2
0
Functions
gT0 (t )
-T0/2 +T0/2
T0 2
1
cn 
T0
g
T0 2
T0 (t ) exp( j 2nf 0t )dt
where G(nf0) is FT of g(t) evaluated

at f = nf0
Functions
gT0 (t )
-T0/2 +T0/2
n 
gT0 (t )  c
n  
n exp( j 2nf 0t ) and we found, cn = f0G(nf0)
Functions
gT0 (t )
-T0/2 +T0/2
n 
gT0 (t )  c
n  
n exp( j 2nf 0t ) and we found, cn = f0G(nf0)
Functions
gT0 (t )
-T0/2 +T0/2
 n   n 
 
 gT0 (t )   f 0  G (nf 0 ) exp( j 2nf 0t )  f 0  G (nf 0 )exp( j 2nf 0t )
 n    n  
n 
 f0  G(nf
n  
0 ) ( f  nf 0 )
Functions
n 
 
 gT0 (t )  f 0  G(nf 0 ) ( f  nf 0 ) g (t )  G ( f )
n  
• g(t) is a single period of gT0(t) but its spectrum is continuous

• Spectrum of gT0(t) is discrete
• Spectrum of gT0(t) consists of
• Impulse functions occurring at integer multiples of f0
• each impulse function is weighted by a factor equal to the
corresponding value of G(nf0)
CSE 311:
Data Communication
Instructor:
Signals and systems
Topics

Functions
n 
 
 gT0 (t )  f 0  G(nf 0 ) ( f  nf 0 ) g (t )  G ( f )
n  
• g(t) is a single period of gT0(t) but its spectrum is continuous

• Spectrum of gT0(t) is discrete
• Spectrum of gT0(t) consists of
• Impulse functions occurring at integer multiples of f0
• each impulse function is weighted by a factor equal to the
corresponding value of G(nf0)
Functions
Fourier Transform of an impulse train  T (t )
0
•  T0 (t ) is a periodic signal with period T0

Functions
0
(t)
•  T0 (t ) is a periodic signal with period T0
t
Impulse train  T (t ) Impulse signal
0
-2T0 -T0 0 n  T0 2T0 3T0 t

Equation of an
Impulse train
 T (t ) 
0   (t  nT )
n  
0
Functions
0
gT0 (t )   T0 (t )
-2T0 -T0 0 n  T0 2T0 3T0 t

 T (t ) 
0   (t  nT )
n  
0
Comparing with the previous result

n 
 
 gT0 (t )  f 0  G(nf 0 ) ( f  nf 0 ) g (t )  G ( f )
n  
Functions
0
gT0 (t )   T0 (t )
-2T0 -T0 0 n  T0 2T0 3T0 t

 T (t ) 
0   (t  nT )
n  
0

G ( f )  g (t )   (t )  1
Functions
0
gT0 (t )   T0 (t )
-2T0 -T0 0 n  T0 2T0 3T0 t

 T (t ) 
0   (t  nT )
n  
0

G ( f )  g (t )   (t )  1  G (nf 0 )
Functions
0
gT0 (t )   T0 (t )
-2T0 -T0 0n  T0 2T0 3T0 t

 T0 (t )    (t  nT 0)
n  

n  n 
 
  T0 (t )  f 0  G(nf 0 ) ( f  nf 0 )  f 0   ( f  nf 0 )
n   n  
Functions
n
Impulse train  T (t ) 
0   (t  nT )
n  
0
t
-2T0 -T0 0 T0 2T0 3T0
n 
FT of Impulse train  
  T0 (t )  f 0   ( f  nf 0 )
n  
f
-4f0 -3f0 -2f0 -f0 0 f0 2f0 3f0 4f0 5f0 6f0
Functions
a>0
a>0
a>0
a>0
a>0
Properties of Fourier Transform
Time-Frequency duality
• Difference is minor
• Remarkably similar
• Difference is minor
• Remarkably similar
• For any relationship,

dual is found by
interchanging roles of
g(t) and G(f)
Let, g (t )  G ( f )
Let we have
Let, g (t )  G ( f )
Let we have
This means
Duality property
If g (t )  G ( f )
then G (t )  g ( f )
Duality property Proof
 
If g (t )  G ( f ) g (t )   G ( f )e j 2ft df   G ( x)e j 2xt dx

 
then G (t )  g ( f )
 

 
then G (t )  g ( f )
 

 

then G (t )  g ( f ) g (t )   G ( x)e  j 2xt dx Replaced t by -t

 

 



g ( f )   G ( x)e  j 2xf dx Replaced t by f

 

 





g ( f )   G (t )e  j 2ft dt Replaced x by t

 

 






 G (t )
 

 






This means,g (  f )  G (t )

 

 






This means, G (t )  g ( f )
Use of Duality property G (t )  g ( f )
t
g (t )   ( ) G(f) =  sinc (f)

t
g (t )   ( ) G(f) =  sinc (f)

G(t) =  sinc (t)
f f
g ( f )   ( )  ( )
 
t
g (t )   ( ) G(f) =  sinc (f)

G(t) =  sinc (t)
f f
g ( f )   ( )  ( )
 
t
g (t )   ( ) G(f) =  sinc (f)

f
g ( f )   ( )
G(t) =  sinc (t) 
f
t
1 f
Time scaling or dilation g ( at )  G( )
property a a
1 f
property a a
1 f
property a a
assuming a > 0, and at = 

1 f
property a a

1 f
property a a
1 f
F g (at )  G( ) assuming a < 0
a a
1 f
property a a
1 f
F g (at )  G( ) assuming a < 0
a a
1 f
Therefore, g (at )  G ( )
a a
1 f
Significance of Time g (at )  G ( )
scaling property a a
• If a > 1, g(at) means compression in time which increases

the signal frequency
• Compression in time results in spectral expansion by same
factor
1 f
scaling property a a
• If a > 1, g(at) means compression in time which increases

the signal frequency
• Compression in time results in spectral expansion by same
factor
• Expansion in time (a < 1) results in spectral compression
• If g(t) is wider, spectrum is narrower
• Bandwidth (spectrum) is inversely proportional to signal
duration (width)
1 f
scaling property t a a
g (t )   ( )
 G(f) =  sinc (f)
1 f
scaling property t a a
g (t )   ( )
 G(f) =  sinc (f)
t
g (t )   ( ) G(f) = 2 sinc (2f)
2
Reflection property g (t )  G ( f )
Directly results assuming a = -1 in scaling property
1 f
g (at )  G ( )
a a
Time shifting property g (t  t0 )  G ( f ) exp( j 2ft0 )

F g (t  t0 )   g (t  t 0 ) e  j 2ft
dt


F g (t  t0 )   g (t  t 0 ) e  j 2ft
dt


 j 2f ( t 0  ) assuming t =t0 +
  g ( ) e d


F g (t  t0 )   g (t  t 0 ) e  j 2ft
dt


 j 2f ( t 0  ) assuming t =t0 +
  g ( ) e d


 e  j 2ft0  g ( ) e  j 2f
d

 e  j 2ft0 G ( f )

F g (t  t0 )   g (t  t 0 ) e  j 2ft
dt


 j 2f ( t 0  ) assuming t =t0 +
  g ( ) e d


 e  j 2ft0  g ( ) e  j 2f
d

 e  j 2ft0 G ( f )
• Delaying the signal by t0 seconds DOES NOT change the

amplitude spectrum, but changes phase spectrum by -2ft0

• If g(t) is synthesized by Fourier components (sinusoids of
certain amplitudes and phases), g(t-t0) can be synthesized by
same sinusoids delayed by t0.

g (t )   G ( f )e j 2ft df


  [G ( f ) cos 2ft  jG ( f ) sin 2ft ]df


• If g(t) is synthesized by Fourier components (sinusoids of
certain amplitudes and phases), g(t-t0) can be synthesized by
same sinusoids delayed by t0.
Assume the cosine signal cos 2ft delayed by t0
cos 2f (t  t0 )  cos(2ft  2ft0 )

cos 2f (t  t0 )  cos(2ft  2ft0 )

• Phase changes is linear function of f
• Higher frequency components undergoes proportionately
higher phase shift
Lower frequency
signal
Higher frequency
signal
Lower frequency
signal
Same time delay, different

t0 frequency signals Higher frequency
signal
Lower frequency
signal, phase shift =/2
Same time delay, different

t0 frequency signals Higher frequency
signal, phase shift = 
Frequency shifting g (t ) exp( j 2f 0t )  G ( f  f 0 )
property
property


F g (t )e j 2f 0t
   g (t )e j 2f 0t  j 2ft
e dt


 j 2 ( f  f 0 ) t
  g (t )e

dt
 G( f  f0 )
property


F g (t )e j 2f 0t
   g (t )e j 2f 0t  j 2ft
e dt


 j 2 ( f  f 0 ) t
  g (t )e

dt
 G( f  f0 )
property


F g (t )e j 2f 0t
   g (t )e j 2f 0t  j 2ft
e dt


 j 2 ( f  f 0 ) t
  g (t )e

dt
 G( f  f0 )
property


F g (t )e j 2f 0t
   g (t )e j 2f 0t  j 2ft
e dt


 j 2 ( f  f 0 ) t
  g (t )e

dt

because,  j 2ft
 G( f  f0 ) G( f )   g (t ) e dt

property
Replacing f0 by –f0
g (t ) exp( j 2f 0t )  G ( f  f 0 )
property
Significance:
Multiplying g(t) by exp(j2f0t) shifts the spectrum by f0 to a new
location
property
Significance:
location
Recall FDM!
property
Significance:
location
Significance:
As exp(j2f0t) is complex, g(t) is multiplied by a real sinusoid
cos(2f0t)
property
Significance:
location
Significance:
As exp(j2f0t) is complex, g(t) is multiplied by a real sinusoid
cos(2f0t)
1 j 2f 0t
But, cos( 2f 0t )  (e  e  j 2f 0t )
2
property
1 j 2f 0t  j 2f 0t
Therefore, g (t ) cos(2f 0t )  ( g (t )e  g (t )e )
2
property
1 j 2f 0t  j 2f 0t
Therefore, g (t ) cos(2f 0t )  ( g (t )e  g (t )e )
2
1
This results, g (t ) cos(2f 0t )  [G ( f  f 0 )  G ( f  f 0 )]
2
property
1
g (t ) cos(2f 0t )  [G ( f  f 0 )  G ( f  f 0 )]
2
Significance:
• Multiplying a sinusoid cos(2f0t) by g(t) means modulation of

sinusoidal amplitude
• cos(2f0t) is the carrier
• g(t) is modulating signal
• This is amplitude modulation
property
1
g (t ) cos(2f 0t )  [G ( f  f 0 )  G ( f  f 0 )]
2
g(t)cos(2πf0t)
CSE 311:
Data Communication
Instructor:
Signals and systems
Topics

property
1
g (t ) cos(2f 0t )  [G ( f  f 0 )  G ( f  f 0 )]
2
g(t)cos(2πf0t)
property
1
g (t ) cos(2f 0t )  [G ( f  f 0 )  G ( f  f 0 )]
2
Further phase change of modulated signal by 0:
• Easily done by multiplying cos(2f0t +0) instead of cos(2f0t)

by g(t)
property
1
g (t ) cos(2f 0t )  [G ( f  f 0 )  G ( f  f 0 )]
2

by g(t)
1
g (t ) cos(2f 0t   0 )  [ g (t )e j 2f 0t  j 0  g (t )e  j 2f 0t  j 0 ]
2
1
 [ g (t )e j 2f 0t e j 0  g (t )e  j 2f 0t e  j 0 ]
2
property
1
g (t ) cos(2f 0t )  [G ( f  f 0 )  G ( f  f 0 )]
2

by g(t)
1
g (t ) cos(2f 0t   0 )  [ g (t )e j 2f 0t e j 0  g (t )e  j 2f 0t e  j 0 ]
2
This means,
1
g (t ) cos(2f 0t   0 )  [G ( f  f 0 )e j 0  G ( f  f 0 )e  j 0 ]
2
property
1
g (t ) cos(2f 0t   0 )  [G ( f  f 0 )e j 0  G ( f  f 0 )e  j 0 ]
2
0 = 0 means
1
g (t ) cos(2f 0t )  [G ( f  f 0 )  G ( f  f 0 )]
2
0 = -/2 means
 
 1 j j
g (t ) cos(2f 0t  )  [G ( f  f 0 )e 2
 G ( f  f 0 )e 2 ]
2 2
property
property in Modulation
1
g (t ) cos( j 2f 0t )  [G ( f  f 0 )  G ( f  f 0 )]
2
Significance:
• A single common channel is used for multiple low band signals

(e.g., multiple radio stations)
• Carriers of different frequencies are assigned to different
transmission stations
• Each station shifts its signal to its own allocated band
• Simultaneous transmission is NO longer a problem
Convolution and
Modulation Property
Time Convolution
g1 (t )  g 2 (t )  G1 ( f )G2 ( f )
Modulation or frequency Convolution
g1 (t ) g 2 (t )  G1 ( f )  G2 ( f )
What is Convolution?

g1 (t )  g 2 (t )   g ( ) g

1 2 (t   )d

g1 (t )  g 2 (t )   g ( ) g

1 2 (t   )d
g1 (t )  g 2 (t )  g 2 (t )  g1 (t )

g1 (t )  g 2 (t )   g ( ) g

1 2 (t   )d
g1 (t )  g 2 (t )  g 2 (t )  g1 (t )
 
g1 (t )  g 2 (t )   g ( ) g

1 2 (t   )d  g

2 ( ) g1 (t   )d
Time Convolution g1 (t )  g 2 (t )  G1 ( f )G2 ( f )
Property

   j 2ft
g1 (t )  g 2 (t )     g1 ( ) g 2 (t   )d  e dt
    

  j 2ft

  g1 ( )   g 2 (t   )e dt  d
  
Property

   j 2ft
g1 (t )  g 2 (t )     g1 ( ) g 2 (t   )d  e dt
    

  j 2ft

  g1 ( )   g 2 (t   )e dt  d
  
Property

   j 2ft
g1 (t )  g 2 (t )     g1 ( ) g 2 (t   )d  e dt
    

  j 2ft

  g1 ( )   g 2 (t   )e dt  d
  

 j 2ft  j 2f
g
 2 ( t   ) e dt is FT of  - time shifted g 2 (t ), e. g ., G 2 ( f ) e

Property

   j 2ft
g1 (t )  g 2 (t )     g1 ( ) g 2 (t   )d  e dt
    

  j 2ft

  g1 ( )   g 2 (t   )e dt  d
  
 
 j 2f  j 2f
 g (
 1 2 )G ( f ) e d   G 2 ( f ) g
 1 ( ) e d
 

 G2 ( f )  g1 (t )e  j 2ft dt  G2 ( f )G1 ( f )

Property

   j 2ft
g1 (t )  g 2 (t )     g1 ( ) g 2 (t   )d  e dt
    

  j 2ft

  g1 ( )   g 2 (t   )e dt  d
  
 
 j 2f  j 2f
 g (
 1 2 )G ( f ) e d   G 2 ( f ) g
 1 ( ) e d
 

 G2 ( f )  g1 (t )e  j 2ft dt  G2 ( f )G1 ( f )

Property

   j 2ft
g1 (t )  g 2 (t )     g1 ( ) g 2 (t   )d  e dt
    

  j 2ft

  g1 ( )   g 2 (t   )e dt  d
  
 
 j 2f  j 2f
 g (
 1 2 )G ( f ) e d   G 2 ( f ) g
 1 ( ) e d
 

 G2 ( f )  g1 (t )e  j 2ft dt  G2 ( f )G1 ( f )

Modulation or frequency g1 (t ) g 2 (t )  G1 ( f )  G2 ( f )
Convolution Property
 
  j 2ft
 G1 ( f )  G2 ( f )     G1 ( )G2 ( f   )d  e df
1
    

 j 2ft

  G1 ( )   G2 ( f   )e df  d
  
 
  G1 ( ) g 2 (t )e j 2t d  g 2 (t )  G1 ( )e j 2t d
 

 g 2 (t )  G1 ( f )e j 2ft df  g 2 (t ) g1 (t )

 
  j 2ft
 G1 ( f )  G2 ( f )     G1 ( )G2 ( f   )d  e df
1
    

 j 2ft

  G1 ( )   G2 ( f   )e df  d
  
 
  G1 ( ) g 2 (t )e j 2t d  g 2 (t )  G1 ( )e j 2t d
 

 g 2 (t )  G1 ( f )e j 2ft df  g 2 (t ) g1 (t )

 
  j 2ft
 G1 ( f )  G2 ( f )     G1 ( )G2 ( f   )d  e df
1
    

 j 2ft

  G1 ( )   G2 ( f   )e df  d
  
 
  G1 ( ) g 2 (t )e j 2t d  g 2 (t )  G1 ( )e j 2t d
 

 g 2 (t )  G1 ( f )e j 2ft df  g 2 (t ) g1 (t )

 
  j 2ft
 G1 ( f )  G2 ( f )     G1 ( )G2 ( f   )d  e df
1
    

 j 2ft

  G1 ( )   G2 ( f   )e df  d
  
 
  G1 ( ) g 2 (t )e j 2t d  g 2 (t )  G1 ( )e j 2t d
 

 g 2 (t )  G1 ( f )e j 2ft df  g 2 (t ) g1 (t )

t
Time Integration G( f ) 1
Property g ( )d  j 2f  2 G(0) ( f )
t
Property g ( )d  j 2f  2 G(0) ( f )
Recall 1 if t  0
u (t )  
0 if t  0
t
Property g ( )d  j 2f  2 G(0) ( f )
Recall 1 if t  0
u (t )  
0 if t  0
Therefore,
1 if t    0 or   t
u (t   )  
0 if t    0 or   t
t
Property g ( )d  j 2f  2 G(0) ( f )
u(t-) u(t-)
1 1
0 t 
 - t 0
1 if t    0
u (t   )  
1 if   t 0 if t    0
u (t   )  
0 if   t
t
Property g ( )d  j 2f  2 G(0) ( f )

u(t-)
g (t )  u (t )   g ( )u (t   )d

1
t 
  g ( )u (t   )d   g ( )u (t   )d
 t 
t  t - t 0
  g ( ).1d   g ( ).0d   g ( )d
 t 
1 if t    0
u (t   )  
0 if t    0
t
Property g ( )d  j 2f  2 G(0) ( f )

u(t-)
g (t )  u (t )   g ( )u (t   )d

1
t 
  g ( )u (t   )d   g ( )u (t   )d
 t 
t  t - t 0
  g ( ).1d   g ( ).0d   g ( )d
 t 
1 if t    0
u (t   )  
0 if t    0
t
Property g ( )d  j 2f  2 G(0) ( f )

u(t-)
g (t )  u (t )   g ( )u (t   )d

1
t 
  g ( )u (t   )d   g ( )u (t   )d
 t 
t  t - t 0
  g ( ).1d   g ( ).0d   g ( )d
 t 
1 if t    0
u (t   )  
0 if t    0
t
Property g ( )d  j 2f  2 G(0) ( f )
t
g (t )  u (t )   g ( )d

t
Property g ( )d  j 2f  2 G(0) ( f )
t
g (t )  u (t )   g ( )d

t
 g ( )d  g (t )  u (t )  G( f )U ( f )

u(t) and sgn(t) function
sgn(t)
u(t)
+1
1
0 t
0 t
-1
1 if t  0
u (t )  
0 if t  0
sgn(t)
u(t)
+1
1
0 t
0
t
-1
1 if t  0
u (t )  
0 if t  0
1
Therefore, u (t )  sgn(t )  1
2
1
u (t )  sgn(t )  1
2
1 1  1 1
Therefore, U ( f )  
   ( f ) 
   (f )
2  jf  j 2f 2
t
Property g ( )d  j 2f  2 G(0) ( f )
t
g (t )  u (t )   g ( )d

t
 g ( )d  g (t )  u (t )  G( f )U ( f )

1 1  1 1
U ( f )     ( f )    (f )
2  jf  j 2f 2
t
Property g ( )d  j 2f  2 G(0) ( f )
t
g (t )  u (t )   g ( )d

t
 g ( )d  g (t )  u (t )  G( f )U ( f )

G( f ) 1
  G ( f ) ( f )
j 2f 2
G( f ) 1
  G (0) ( f )
j 2f 2
Time Differentiation dg (t )
Property  j 2fG ( f )
dt
dt

g (t )   G ( f )e j 2ft df

dt

g (t )   G ( f )e j 2ft df

 
dg (t ) d  
   G ( f )e j 2ft df    j 2fG ( f )e j 2ft df
dt dt    
dt

g (t )   G ( f )e j 2ft df

 
dg (t ) d  
   G ( f )e j 2ft df    j 2fG ( f )e j 2ft df
dt dt    
dg (t )
 j 2fG ( f )
dt
dt
d n g (t ) n
n
  j 2f  G( f )
dt
Time Differentiation
d n g (t ) n
Property to find FT of n
  j 2f  G( f )
complex signal dt
d n g (t ) n
  j 2f  G( f )
complex signal dt
d n g (t ) n
  j 2f  G( f )
complex signal dt
d n g (t ) n
  j 2f  G( f )
complex signal dt
Let, (g(t))  G ( f )
d n g (t ) n
  j 2f  G( f )
complex signal dt
d n g (t ) n
  j 2f  G( f )
complex signal dt
d n g (t ) n
  j 2f  G( f )
complex signal dt
d n g (t ) n
  j 2f  G( f )
complex signal dt
Time Domain and Frequency Domain
Representation
Limited in time Indefinite spread in frequency

Representation
t
g (t )   ( ) G(f) =  sinc (f)


Representation
t
g (t )   ( ) G(f) =  sinc (f)


f
g ( f )   ( )
G(t) =  sinc (t) 
f
t
Indefinite spread in time Limited in frequency
Representation
• A signal cannot be BOTH band limited and time limited
• Relation is inverse
• Relation changes in inverse manner
• Any change in time specification changes spectrum
speciation inversely
• We can specify EITHER of them, NOT BOTH of them
CSE 311:
Data Communication
Instructor:
Signals and systems
Topics

Representation
t
g (t )   ( ) G(f) =  sinc (f)


f
g ( f )   ( )
G(t) =  sinc (t) 
f
t
Indefinite spread in time Limited in frequency
Representation
• A signal cannot be BOTH band limited and time limited
• Relation is inverse
• Relation changes in inverse manner
• Any change in time specification changes spectrum
speciation inversely
• We can specify EITHER of them, NOT BOTH of them
Bandwidth of a Signal
• Definition: extent of SIGNIFICANT spectral content of a

signal for POSITIVE frequencies
• Alternate definitions exist

Lowpass signal: B
Frequency spectrum is centered

around f = 0
Lowpass signal: B

around f = 0
B
Bandpass signal:
around ±f0
Lowpass signal: B

around f = 0
Shifting a low pass signal to f0

location DOUBLES its
bandwidth
Lowpass signal: B

around f = 0
B
Shifting a low pass signal to f0
location DOUBLES its
bandwidth
3-dB Bandwidth of a Signal
B
Lowpass signal: 1
G (0f )
Separation betn 2
peak and the

position where it
1
is 2 of its peak
3-dB Bandwidth of a Signal
B
Lowpass signal: 1
G (0f )
Separation betn 2
peak and the

position where it
1
is 2 of its peak
Bandpass signal: 1
B G ( f0)
Separation betn two positions 2
where amplitudes are 1 of its
2
peak
rms Bandwidth of a Signal
Root mean square bandwidth:
the square root of the second moment of a properly
normalized form of the squared amplitude spectrum of the
signal about a suitably chosen point

k 2
kth-moment
origin
of squared amplitude about  f

G ( f ) df

2 2
2nd moment of squared amplitude about m2  f G ( f ) df
origin 

2
0-th moment of squared amplitude about m0   G( f ) df
origin 

2 2
2nd moment of normalized squared

f

G ( f ) df
m2
amplitude about origin  

2 m0

 G( f ) df

2 2
f

G ( f ) df
m2
rms bandwidth, Wrms  

2 m0
 G ( f ) df

Time-Bandwidth Product
Duration × bandwidth = constant
• Relationship is true for every definition of

bandwidth
• Relationship is true for every definition of

bandwidth
• Compression in time by a factor a expands the

spectrum by the same factor and vice versa
• No change in the constant
t B
g (t )   ( )  sinc (f)=

Duration × bandwidth = × (1/ ) =1 = constant

 
2 2 2 2
t

g (t ) dt f

G ( f ) df
Trms  
Wrms  
2 2
 g (t )

dt  G( f )

df
1
Trms  wrms 
4
1
Trms  wrms 
4
• Equality holds for Gaussian pulse

Signal Transmission through
Linear Time Invariant System
System
x(t) System, S y(t)
y(t) = S(x(t))
System
x(t) System, S y(t)
y(t) = S(x(t))
y1(t) = S(x1(t))
y2(t) = S(x2(t))
Linear System
x(t) Linear System, LS y(t)
y(t) = LS(x(t))
y1(t) = LS(x1(t)) y2(t) = LS(x2(t))
Linear System
a1x1(t) + a2x2(t) Linear System, LS a1y1(t) + a2y2(t)
y(t) = LS(x(t))
y1(t) = LS(x1(t)) y2(t) = LS(x2(t))
a1y1(t) + a2y2(t) = LS(a1x1(t) + a2x2(t))

Linear System
Principle of Superposition
the response of a linear system to a number of excitations applied
simultaneously is equal to the sum of the responses of the system when
each excitation is applied individually.
Linear System
Principle of Superposition
the response of a linear system to a number of excitations applied
simultaneously is equal to the sum of the responses of the system when
each excitation is applied individually.
Examples
Filters and Communication Channels
Time Invariant System
Time Invariant
x(t) System, TIS y(t)
y(t) = TIS(x(t))
Time Invariant
x(t) System, TIS y(t)
y(t) = TIS(x(t))
y1(t) = TIS(x1(t))
y2(t) = TIS(x2(t))
Time Invariant
x1(t-ta) System, TIS y1(t-ta)
y(t) = TIS(x(t))
y1(t-ta) = TIS(x1(t-ta)) y2(t-tb) = TIS(x2(t-tb))
Linear Time
x(t) Invariant System, y(t)
LTI
y(t) = LTI(x(t))
Linear Time
x(t) Invariant System, y(t)
LTI
y(t) = LTI(x(t))
y1(t) = LTIS(x1(t))
y2(t) = LTIS(x2(t))
Linear Time
a1x1(t-ta) + a2x2(t-tb) Invariant System, a1y1(t-ta) + a2y2(t-tb)
LTI
y(t) = LTI(x(t))
a1y1(t-ta) + a2y2(t-tb) = LTI(a1x1(t-ta) + a2x2(t-tb))
Linear Time
Invariant System,
x(t) y(t)
h(t)
LTI system is described by impulse response, h(t)

CSE 311:
Data Communication
Instructor:
Signals and systems
Topics

Linear System
y(t) = LS(x(t))
y1(t) = LS(x1(t)) y2(t) = LS(x2(t))
a1y1(t) + a2y2(t) = LS(a1x1(t) + a2x2(t))

Time Invariant
x1(t-ta) System, TIS y1(t-ta)
y(t) = TIS(x(t))
y1(t-ta) = TIS(x1(t-ta)) y2(t-tb) = TIS(x2(t-tb))
Linear Time
a1x1(t-ta) + a2x2(t-tb) Invariant System, a1y1(t-ta) + a2y2(t-tb)
LTI
y(t) = LTI(x(t))
a1y1(t-ta) + a2y2(t-tb) = LTI(a1x1(t-ta) + a2x2(t-tb))
Linear Time
Invariant System,
x(t) y(t)
h(t)

Linear Time
Invariant System,
x(t) = (t) y(t) = h(t)
h(t)

Linear Time
Invariant System,
x(t-t0) y(t-t0)
h(t)

Linear Time
Invariant System,
(t-t0) h(t-t0)
h(t)

Linear Time
Invariant System,
a(t-t0) ah(t-t0)
h(t)

Linear Time
Invariant System,
a1(t-t1) + a2(t-t2) a1h(t-t1) + a2h(t-t2)
h(t)

Linear Time
Invariant System,
 a  (t  t
k k )  a h(t  t
k k )
k h(t) k

Linear Time
Invariant System,
x(t) y(t)
h(t)
We will find relation betn x(t) and y(t)

x(t)
Let the input signal be x(t)

t

Stair case approximation of x(t) using narrow rectangular pulses

t

As  approaches to 0, pulse approaches to (.) ×height× 

x()
 t

At t = , pulse approaches to (t- )x() 

x()
0 1 2 . . . n
 t

At t = , pulse approaches to (t- )x() 

x()
0 1 2 . . . n
 t

At t =  = n, pulse approaches to (t- n)x(n) 

x(t)   δ(t  nτ)x(nτ)τ
n
x()
 t

At t = , pulse approaches to (t- n)x(n) 

x(t)   δ(t  nτ)x(nτ)τ
n
x() and  are constants
x()
 t

At t = , pulse approaches to (t- n)x(n) 

Recall Linear Time Invariant
System
Linear Time
Invariant System,
 a  (t  t
k k )  a h(t  t
k k )
k h(t) k
Linear Time
Invariant System,
x(t)   δ(t  nτ)x(nτ)τ
n h(t)
Linear Time
Invariant System,
x(t)   δ(t  nτ)x(nτ)τ y(t)   h(t  nτ)x(nτ)τ
n h(t) n
Linear Time
Invariant System,
n h(t) n
As  approaches zero,

y (t )   h(t  nτ)x(nτ)τ   h(t  τ ) x( τ )τ   h(t  τ ) x( τ )dτ
n τ 
Linear Time
Invariant System,
n h(t) n
As  approaches zero,

y (t )   h(t  nτ)x(nτ)τ   h(t  τ ) x( τ )τ   h(t  τ ) x( τ )dτ
n τ 
Linear Time
Invariant System,
x(t)   δ(t  nτ)x(nτ)τ y(t)   h(t  τ)x(nτ)τ
n h(t) n

y (t )   h(t  τ ) x( τ )dτ is convolution integral and can be written as

y (t )  h(t )  x(t )
Linear Time
Invariant System,
x(t ) y (t )  h(t )  x(t )
h(t)
Linear Time
Invariant System,
x(t ) y (t )  h(t )  x(t )
h(t)
Now Recall
g1 (t )  g 2 (t )  G1 ( f )G2 ( f )
Linear Time
x(t ) Invariant System, y (t )  h(t )  x(t )
X(f)
h(t)
Y(f) = H(f)X(f)
H(f)
H(f) is
• Fourier transform of h(t)
• the frequency response of LTI system
• a transfer function, because it transfers or passes desired frequencies of X(f)
• a complex function
Linear Time
X(f)
h(t)
Y(f) = H(f)X(f)
H(f)
H ( f )  H ( f ) e j h ( f )
• |H(f)| is amplitude response of LTI system
• h(f) is phase response of LTI system, measured in radians
Linear Time
X(f)
h(t)
Y(f) = H(f)X(f)
H(f)
H ( f )  H ( f ) e j h ( f )
ln H ( f )  ln H ( f )  j h ( f )   ( f )  j h ( f )
Linear Time
X(f)
h(t)
Y(f) = H(f)X(f)
H(f)
H ( f )  H ( f ) e j h ( f )
ln H ( f )  ln H ( f )  j h ( f )   ( f )  j h ( f )
• ln|H(f)| or α(f) is the gain of LTI system, measured in nepers

Linear Time
X(f)
h(t)
Y(f) = H(f)X(f)
H(f)
H ( f )  H ( f ) e j h ( f )
 ( f )  20 log10 ln H ( f ) dB
• α´(f) is an alternate measurement of gain of LTI system, measured in dB

Causal LTI System
• does not respond before the excitation is applied
• physically realizable system
• impulse response must vanish for negative time
h(t)=0, t<0
Causal LTI System
Necessary and sufficient condition for causal system
where  ( f )  ln H ( f )
Causal LTI System
where  ( f )  ln H ( f )
Paley-Wiener Criterion
Causal LTI System
where  ( f )  ln H ( f )
Significance
• If α(f) satisfy P-W criterion, α(f) can be combined with h(f) to find h(t) for a
causal system
• Infinite attenuation (α(f) = - or |H(f)| =0 ) for a band of frequencies DOES NOT
satisfy P-W criterion
• However, α(f) = - allowed for a discrete set of frequencies
Stable LTI System
• output signal is bounded for all bounded input signals.
• The criteria is also known as bounded input–bounded output (BIBO)
stability criterion
Stable LTI System
For bounded input signal, x(t)
|x(t)| < M for all t

Stable LTI System
Output from a LTI system

 
y (t )  h(t )  y (t )   h(t  τ ) x( τ ) dτ   x(t  τ )h( τ )dτ
 
Stable LTI System

 
 
 
y (t )   x(t  τ )h(τ )dτ   x(t  τ ) h( τ ) dτ
 
Stable LTI System

 
 
 
y (t )   x(t  τ )h(τ )dτ   x(t  τ ) h( τ ) dτ
 
Stable LTI System

 
 
  
y (t )   x(t  τ )h(τ )dτ   x(t  τ ) h(τ ) dτ  M  h(τ ) dτ
  
Stable LTI System

Output from a LTI system 
y (t )  M  h(τ ) dτ

Stable LTI System

Output from a LTI system 
y (t )  M  h(τ ) dτ

ouput signal y(t) to be bounded, |h(t)| must be integrable, that is,

 h(τ ) dτ  

Stable LTI System
Necessary and sufficient condition for BIBO stability
 h(τ ) dτ  

System Response in System Response in
Frequency domain Time domain
|H(f)|
1 h(t)
f
-B B
td
System Response in System Response in
Frequency domain Time domain
|H(f)|
1 h(t)
f
-B B
td
Is this a causal system?

CSE 311:
Data Communication
Instructor:
CT2
Tomorrow, Wednesday
Signals and systems
Topics

Linear Time
X(f)
h(t)
Y(f) = H(f)X(f)
H(f)
H ( f )  H ( f ) e j h ( f )
Linear Time
X(f)
h(t)
Y(f) = H(f)X(f)
H(f)
H ( f )  H ( f ) e j h ( f )
Signal Distortion during
Transmission
Linear Time
X(f)
h(t)
Y(f) = H(f)X(f)
H(f)
Y ( f ) e jy ( f )  H ( f ) X ( f ) e j  h ( f ) x ( f ) 
H ( f )  H ( f ) e j h ( f )
Signal Distortion during
Transmission
Linear Time
X(f)
h(t)
Y(f) = H(f)X(f)
H(f)
Y ( f ) e jy ( f )  H ( f ) X ( f ) e j  h ( f ) x ( f ) 
• System distorts the input signal
• Input signal amplitude |X(f)| is changed to |X(f)| |H(f)|
• Input signal phase is shifted by h(f)
• Plot of |H(f)| and h(f) vs. f shows how the system changes diffrerent
frequency components differently (attenuation/boosting)
Distortionless Transmission
Linear Time
x(t ) Invariant System, y (t )  x(t )
X(f)
h(t)
H(f) Y( f )  X( f )
• Input and output signals have identical shape

Linear Time
x(t ) Invariant System, y (t )  kx(t )
X(f)
h(t)
H(f) Y ( f )  kX ( f )

• Boosting/amplification is allowed
Linear Time
x(t ) Invariant System, y (t )  kx(t  t d )
X(f)
h(t)
H(f) Y ( f )  kX ( f )e  j 2ft d

• Boosting/amplification is allowed
• Delaying is allowed
Linear Time
X(f)
h(t)
H(f) Y ( f )  kX ( f )e  j 2ft d

• Delaying is allowed
Y ( f )  kX ( f )e  j 2ftd  ke  j 2ftd X ( f )  H ( f ) X ( f )
 j 2ft d
H ( f )  ke
Linear Time
X(f)
h(t)
H(f) Y ( f )  kX ( f )e  j 2ft d
 j 2ft d
H ( f )  ke
H ( f )  k and  h ( f )  2ft d
• |H(f)| is constant
• h(f) is a linear function of f
|H(f)|
h(f)
H ( f )  k and  h ( f )  2ft d  t d

• |H(f)| is constant
• h(f) is a linear function of f
• Slope of h(ω) is –td, where td is delay of the output
All Pass Vs. Distortionless
System
|H(f)|
H( f )  k
Y( f )  H( f ) X ( f )
• |H(f)| = k means that every frequency amplitude of |X(f)| is
passed to |Y(f)| with same gain
• Everything is present |Y(f)|, nothing is missing!
System
|H(f)|
H( f )  k
Y( f )  H( f ) X ( f )
• |H(f)| = k means that every frequency amplitude of |X(f)| is
passed to |Y(f)| with same gain
• Everything is present |Y(f)|, nothing is missing!
• However, this does NOT guarantee distortionless system
System
All pass system:
H( f )  k
Consider a real composite signal consisting of multiple sinusoids
of different frequencies
• |H(f)| = k guarantees only presence of all frequency components
• Signals with different frequencies may get different delays.
• Synchronization of multiple frequency components is NOT

guaranteed
• Output composite signal will be out of sync

System
Distortionless system:
H ( f )  k and  h ( f )  2ft d
• Linear phase h(f) = -2ftd guarantees a constant time delay td

for every sinusoid
• Synchronization is guaranteed
System
How to Check Distortionless transmission?
Find the slope (derivative) of h(f)
1 d h ( f )
td ( f )   
2 df
System
1 d h ( f )
td ( f )   
2 df
Hint: Rearrange and differentiate h(f) = -2ftd

System
1 d h ( f )
td ( f )   
2 df
• If the slope of h(f) is constant, it guarantees a constant time delay
td for every sinusoid
• If the slope is a function of f, different frequency components will

have different time delay
Nature of Distortion in Audio
Signals
Human ear is
• sensitive to amplitude distortion which results if |H(f)| ≠ k
• Insensitive (??) to phase distortion which results if the slope of

h(f) is NOT constant
Nature of Distortion in Audio
Signals
• Duration of each spoken syllable : 0.01 to 0.1 second
• Variation of the slope of h(f) is ONLY fraction of a millisecond,

meaning a nearly constant slope, e.g., time delay
Nature of Distortion in Video
Signals
Human eye is
• insensitive to amplitude distortion
• sensitive to phase distortion
Nature of Distortion in Video
Signals
Insensitive amplitude distortion

• amplitude distortion changes only half-tone values of picture
Sensitive phase distortion

• Nonlinear phase distortion means different time delay for different
picture (signal) element
• This results in smeared picture
Nature of Distortion in
Communication Systems
Sensitive phase distortion

• Nonlinear phase distortion causes pulse dispersion (spreading out)
• Pulses are interfered with near by pulses
• This results in misinterpretation in the receiver

– Binary 1 may read as 0 and vice versa
Ideal Vs. Practical Filters
Ideal filters
• allow distortionless transmission for certain band of frequencies
• suppress all other frequencies
This Ideal Lowpass filter
• allows transmission for frequencies f  |B|
• suppresses all other frequencies f  |B|
|H(f)|
1
f
-B B
h(f) = -2ftd
This Ideal Highpass filter
• allows transmission for frequencies f  |B|
• suppresses all other frequencies f  |B|
|H(f)|
f
-B B
h(f) = -2ftd
This Ideal Bandpass filter
• allows transmission for the band centered at ±f0
• suppresses all other frequencies
|H(f)|
f
-f0 f0
h(f) = -2ftd
|H(f)|
1
f
-B B
h(f) = -2ftd
• Slope of h(f) is constant –td, means time delay is td
|H(f)|
1
f
-B B
h(f) = -2ftd
y(t) = g(t -td)

|H(f)|
1
f
-B B
h(f) = -2ftd
f
y(t) = g(t -td) H ( f )  ( ) and
2B
|H(f)|  h ( f )  2ft d
1
f
-B B
h(f) = -2ftd
f
y(t) = g(t -td) H ( f )  ( ) and
2B
|H(f)|  h ( f )  2ft d
1 f  j 2ftd
H ( f )   ( )e
2B
f
-B B
h(f) = -2ftd
Impulse response of Ideal Lowpass filter
Frequency Transform of
Impulse response
f  j 2ft d
H ( f )   ( )e
2B
|H(f)|
1
f
-B B
h(f) = -2ftd
Frequency Transform of Impulse response
1  f  j 2ftd 
Impulse response h(t )   H ( f )    ( )e
1

f  j 2ft d
 2B 
H ( f )   ( )e  2 Bsinc(2B (t  t d ))
2B
|H(f)|
1
f
-B B
h(f) = -2ftd
Frequency Transform of Impulse response
1  f  j 2ftd 
1

f  j 2ft d
 2B 
H ( f )   ( )e  2 Bsinc(2B (t  t d ))
2B
|H(f)|
1 h(t)
f
-B B
td
h(f) = -2ftd
CSE 311:
Data Communication
Instructor:
Signals and systems
Topics

Ideal Low Pass (ILP) Filter

in Frequency domain
f  j 2ft d
H ( f )   ( )e
2B
|H(f)|
1
f
-B B
h(f) = -2ftd
ILP Filter in Frequency In Time domain

1  f  j 2ftd 
domain h(t )   H ( f )    ( )e
1

f  j 2ft d
 2B 
H ( f )   ( )e  2 Bsinc(2B (t  t d ))
2B
|H(f)|
1
f
-B B
h(f) = -2ftd
ILP Filter in Frequency In Time domain

1  f  j 2ftd 
domain h(t )   H ( f )    ( )e
1

f  j 2ft d
 2B 
H ( f )   ( )e  2 Bsinc(2B (t  t d ))
2B
|H(f)|
1 h(t)
f
-B B
td
h(f) = -2ftd
ILP Filter in Frequency Impulse Response
1  f  j 2ftd 
domain h(t )   H ( f )    ( )e
1

f  j 2ft d
 2B 
H ( f )   ( )e  2 Bsinc(2B (t  t d ))
2B
|H(f)|
1 h(t)
f
-B B
td
h(f) = -2ftd
Frequency Transform of Impulse Response
1  f  j 2ftd 
1

f  j 2ft d
 2B 
H ( f )   ( )e  2 Bsinc(2B (t  t d ))
2B
|H(f)|
1 h(t)
f
-B B
td
h(f) = -2ftd
Impulse response
h(t) h(t )  2 Bsinc(2B(t  t d ))
td
Response begins
before t = 0
Impulse response
h(t) h(t )  2 Bsinc(2B(t  t d ))
td
Response begins
before t = 0
Noncausal and
unrealizable filter
Impulse response
h(t) h(t )  2 Bsinc(2B(t  t d ))
Condition to be causal
td h(t) = 0 for all t < 0
Response begins
before t = 0
Noncausal and
unrealizable filter
Causal and realizable Lowpass filter
h(t) Condition in time domain

h(t) = 0 for all t < 0
td
Response begins
on or after t = 0
h(t) Condition in time domain

h(t) = 0 for all t < 0
Condition in frequency
td
domain
Response begins
on or after t = 0
where  ( f )  ln H ( f )
Causal and realizable Lowpass filter??
|H(f)|
1
domain
f
-B B
where  ( f )  ln H ( f )
h(f) = -2ftd Paley-Wiener Criterion
|H(f)|
1
domain
f
-B B
where  ( f )  ln H ( f )
• If for certain band of frequencies |H(f)| = 0 means |α(f)|=| ln |H(f)| |=,

which does NOT satisfy P-W criterion
|H(f)|
1
domain
f
-B B
where  ( f )  ln H ( f )
• If for discrete of frequencies |H(f)| = 0, P-W criterion may be satisfied

hˆ(t ) domain
where  ( f )  ln H ( f )
td
hˆ(t )  h(t )u (t )
Other Causal and realizable Lowpass filters from ILP filter
|H(f)|
1
domain
f
-B B
where  ( f )  ln H ( f )
• Instead of sharp change at ≠B, change smoothly and approach

asymptotically to zero
|H(f)| |H(f)| Butterworth
1 Lowpass filters
f
-B B
h(f) = -2ftd
f
B

|H(f)| |H(f)| Butterworth
1 Lowpass filters
1
H( f ) 
1   f / B
2n
f
-B B
h(f) = -2ftd
f
B

Parseval’s Theorem: Signal
Energy in Frequency Domain
From Chapter 2,
Energy for real signal,

2
E g   g (t )dt

From Chapter 2,

2
E g   g (t )dt

 2
For complex valued E g   g (t ) dt
signal, 
From Chapter 2,

2
E g   g (t )dt

For complex valued  2

signal, Eg   g (t ) dt



  g (t ) g (t ) dt

 2
E g   g (t ) dt


  g (t ) g (t ) dt

 2
E g   g (t ) dt
 
j 2ft

  g (t ) g (t ) dt g (t )   G ( f )e df



  j 2ft 
g (t )   G ( f ) e df


  G ( f ) e  j 2ft df

 2
E g   g (t ) dt


  g (t ) g (t ) dt

    j 2ft

  g (t )   G ( f ) e df  dt

  
 2
E g   g (t ) dt


  g (t ) g (t ) dt




  j 2ft

  G ( f )   g (t )e dt  df

 
 2
E g   g (t ) dt


  g (t ) g (t ) dt




  j 2ft

  G ( f )   g (t )e dt  df

 

  G ( f ) G ( f )df

 2
E g   g (t ) dt


  g (t ) g (t ) dt




  j 2ft

  G ( f )   g (t )e dt  df

 

  G ( f ) G ( f )df

 2
  G ( f ) df

 2  2
E g   g (t ) dt   G ( f ) df
 
Application of Parseval’s
Theorem
 2  2
E   g (t ) dt   G ( f ) df
g
 
Find
|G(f)|2
Energy
G ( f )  Tsinc(fT )
Application of Parseval’s
Theorem
 2  2
E   g (t ) dt   G ( f ) df
g
 
Find
|G(f)|2
Energy
f
0
G ( f )  Tsinc(fT )
Sampling
and
Analog-to-Digital Conversion
Analog to Digital Conversion of
Message Signal
• 2 major steps
– Sampling
– Quantization
mp
2m p
L
-mp
Message Signal
• Sampling
4
Sampling Theorem
Let a message signal g(t) band limited to B Hz
g(t)
G(f) = 0 for all |f| > B

Sampling Theorem
g(t)

The signal g(t) can be reconstructed exactly from its discrete time
samples taken at a rate of fs Hz, where fs ≥ 2B
Sampling Theorem
g(t)

For recovery, the minimum sampling rate, fs = 2B Hz
Sampling Theorem
Uniform sampling g(t) at a rate fs

g(t)
t
Sampling Theorem

g(t)
Ts
Sampling Theorem

g(t)
Multiply g(t) by impulse train  Ts (t )
t where sampling period Ts = 1/fs
The result is g (t )
 T (t )    (t  nTs )
s
 T (t ) n
s
Ts
Sampling Theorem

g(t) g (t )
 T (t )    (t  nTs )
s
n
Ts
Sampling Theorem

g(t) g (t )
 T (t )    (t  nTs )
s
n
Ts
Sampling Theorem

g (t )
• g (t ) consists of impulses uniformly spaced Ts seconds apart

• nth impulse is weighted by g(nTs), the value of g(t) at t = nTs
Sampling Theorem

g (t )
g (t ) is multiplication of g(t) and impulse train  Ts (t ) and can be

written as
g (t )  g (t ) Ts (t )   g (t ) (t  nTs )   g (nTs ) (t  nTs )
n n
Sampling Theorem

g (t )
 (t ) (t )   (0) (t )
 (t ) (t  T )   (T ) (t  T )
 (t ) (t  nTs )   (nTs ) (t  nTs )
g (t ) is multiplication of g(t) and impulse train  Ts (t ) and can be

written as
n n
Sampling Theorem
Uniform sampling g(t) at a rate fs = 1/ Ts

g(t) g (t )
 T (t )    (t  nTs )
s
n Alternate Representation
1 jn 2f s t
 Ts (t ) 
Ts
n
e
Ts
Sampling Theorem

g (t ) 1 jn 2f s t
g (t )  g (t ) Ts (t ) 
Ts
n
g (t ) e
1 j 2nf s t

Ts
n
g (t ) e

n n
Sampling Theorem

g (t ) 1 jn 2f s t
g (t )  g (t ) Ts (t ) 
Ts
n
g (t ) e
1 j 2nf s t

Ts
n
g (t ) e
FT of g (t ) is defined by
1
G ( f )  g (t )   G ( f  nf )s
Ts n
Sampling Theorem

g (t ) 1 jn 2f s t
g (t )  g (t ) Ts (t ) 
Ts
n
g (t ) e
1 j 2nf s t

Ts
n
g (t ) e
FT of g (t ) is defined by G(f)
G ( f ) consists of G(f)
scaled by fs repeated
1 periodically with
G ( f )  g (t )   G ( f  nf )s
Ts n period fs
Sampling Theorem
G( f )
g(t)
g (t ) G( f )
-2fs -fs fs 2fs

Sampling Theorem
G( f )
g(t)
Can we reconstruct
g(t) from g (t ) ?
g (t ) G( f )
-2fs -fs fs 2fs

Sampling Theorem
G( f )
g(t)
Can we reconstruct Possible, if we can we

g(t) from g (t ) ? reconstruct G(f) from G ( f )
g (t ) G( f )
-2fs -fs fs 2fs

Sampling Theorem
G( f )
• We have to isolate one cycle from G ( f )
• This is possible when cycles are non-

overlapping, this means
Possible, if we can
fs> 2B
we reconstruct G(f)
from G ( f ) . G ( f )
-2fs -fs fs 2fs

Sampling Theorem
G( f )

Possible, if we can
fs> 2B
we reconstruct G(f)
Alternately, from G ( f ) . G ( f )
1
Sampling interval of g(t), Ts 
2B
-2fs -fs fs 2fs

Sampling Theorem
G( f )

fs> 2B
Low pass filter of
Alternately, bandwidth B G( f )
1
Sampling interval of g(t), Ts 
2B
Use an Ideal lowpass filter of bandwidth
B to isolate one cycle.
-2fs -fs fs 2fs

Sampling Theorem
G( f )

Low pass filter of

Minimum sampling rate can be bandwidth B G( f )
fs = 2B
This is Nyquist sampling rate
-3fs -2fs -fs fs 2fs 3fs

Sampling Theorem
G( f )

Low pass filter of

Maximum sampling interval is bandwidth B G( f )
1
Ts 
2B
This is Nyquist sampling interval

CSE 311:
Data Communication
Instructor:
Sampling
and
Sampling Theorem
G( f )
g(t)
Can we reconstruct Possible, if we can we

g(t) from g (t ) ? reconstruct G(f) from G ( f )
g (t ) G( f )
-2fs -fs fs 2fs

Sampling Theorem
G( f )

Low pass filter of

Minimum sampling rate can be bandwidth B G( f )
fs = 2B
This is Nyquist sampling rate

Sampling Theorem
G( f )

Low pass filter of

Maximum sampling interval is bandwidth B G( f )
1
Ts 
2B
This is Nyquist sampling interval

Sampling Theorem
G( f )
• For any signal g(t)
fs > 2B
Low pass filter of

bandwidth B G( f )
-2fs -fs fs 2fs

Sampling Theorem
G( f )
fs > 2B
• If there is no impulse at ±B, we can Low pass filter of

use fs = 2B. Therefore, bandwidth B G( f )
fs  2B

Sampling Theorem
G( f )
fs > 2B
• If there is no impulse at ±B, we can Low pass filter of

use fs = 2B. Therefore, bandwidth B G( f )
fs  2B
• If there is IMPULSE at ±B, we can

NOT use fs = 2B, because overlap will
occur. Therefore, fs > 2B
Sampling Theorem
G( f )
• In this case we cannot isolate one

cycle from G ( f ) if fs = 2B
Low pass filter of

bandwidth B G( f )

Sampling Theorem
• In this case we cannot isolate one cycle fromG ( f ) if fs = 2B

• Example of sine wave of frequency B
sin 2Bt jG ( f )
-B B
Sampling Theorem
• In this case we cannot isolate one cycle fromG ( f ) if fs = 2B

• Example of sine wave of frequency B
sin 2Bt jG ( f )
-B B
Sampling at the rate of 2B at these points

will give constant values (zeros)
Signal Reconstruction from
Uniform Samples
• Reconstruction 4
objective
Need to
interpolate
intermediate
values
Uniform Samples
G( f )
• We proved theoretically:
exact reconstruction is
possible provide
– The signal is sampled at rate
fs > 2B
– The sampled signal is
passed though an ILPF with Low pass filter of
bandwidth B bandwidth B G( f )
-2fs -fs fs 2fs

Uniform Samples
G( f )
• We proved theoretically
possible provide
– The signal is sampled at rate
fs > 2B
1
G( f ) 
Ts
 G( f  nf )
n
s
-2fs -fs fs 2fs

Uniform Samples
G( f )
• We proved theoretically
possible provide
– the signal is sampled at rate
fs > 2B
1
G( f ) 
Ts
 G( f  nf )
n
s
f
H ( f )  Ts ( )
2B -2fs -fs fs 2fs
Uniform Samples
f
• To recover g(t), the transfer function will be H ( f )  Ts ( )
2B
H(f)
Ts
f
-B B
f
H ( f )  Ts ( )
2B
Uniform Samples
f
2B
H(f) h(t)
Ts 2BTs
f
-B B
f h(t )  2 BTs sinc(2Bt )

H ( f )  Ts ( )
2B Impulse response
Uniform Samples
f
2B
H(f) h(t)
Ts
f
-B B
h(t )  2 BTs sinc(2Bt )

f
H ( f )  Ts ( )  sinc( 2Bt )
2B 1
Assuming Nyquist sampling interval Ts 
2B
Uniform Samples
f
2B
h(t)
Note: h(t) = 0 at Nyquist
sampling instant  n
2B
h(t )  sinc(2Bt )
Impulse response
Uniform Samples
f
2B
If we pass g (t ) through an h(t)
LTI with this impulse
response, the output will be
g(t).
Impulse response
Uniform Samples
f
2B
If we pass g (t ) through an LTI h(t)
with this impulse response, the
output will be g(t).
g (t )   g (nTs ) (t  nTs )
n
Uniform Samples
f
2B
g (t )   g (nTs ) (t  nTs )
n
Linear Time
Invariant System,
 a  (t  t
k k )  a h(t  t
k k )
k h(t) k
Uniform Samples
f
2B
g (t )   g (nTs ) (t  nTs )
n
Linear Time
Invariant System,
x(t)   δ(t  nτ)x(nτ)τ y(t)   h(t  τ)x(nτ)τ
n h(t) n
Uniform Samples
f
2B
g (t )   g (nTs ) (t  nTs )
n
Linear Time
Invariant System,
g (t )   g (nTs ) (t  nTs ) g (t )   g (nTs )h(t  nTs )
n h(t) n
Uniform Samples
Linear Time
Invariant System,
n h(t) n
 sinc(2Bt )
Uniform Samples
Linear Time
Invariant System,
n h(t) n
 sinc(2Bt )
 g (t )  h(t )
Uniform Samples
Linear Time
Invariant System,
n h(t) n
 sinc(2Bt )
 g (t )  h(t )
g (t )   g (nTs )h(t  nTs )
n
  g (nTs )sinc2B (t  nTs )

n
  g (nTs )sinc(2Bt  n )
n
Uniform Samples
Linear Time
Invariant System,
n h(t) n
 sinc(2Bt )
 g (t )  h(t )
g (t )   g (nTs )sinc(2Bt  n )
n
Interpolation formula to find the value g(t)
at arbitrary time t.
Uniform Samples
Significance of Interpolation
formula
g (t )   g (nTs )sinc(2Bt  n )
n
Sampled signal Reconstructed signal

g (t ) g(t)
Practical Signal Reconstruction
(Interpolation)
H(f)
Ts
generates g (t )   g (nTs )sinc(2Bt  n )
n
f
-B B Interpolation formula
ILPF
(Interpolation)
H(f)
Ts
n
f
ILPF:
Noncausal
and
Unrealizable
(Interpolation)
H(f)
Ts
n
f
ILPF: h(t)
Noncausal
and
Unrealizable
Also seen form long nature of sinc

reconstruction pulse
(Interpolation)
H(f)
Ts
n
f
h(t)
Sharp transition of
ILPF at ±B causes
this problem
Also seen form long nature of sinc

(Interpolation)
Alternate solution is to use

realizable pulse for interpolation
p(t)
t
(Interpolation)
Alternate solution is to use realizable pulse for interpolation

g(t) g~ (t )
Sampled signal
Original signal
(Interpolation)

g(t) g~ (t )
Sampled signal
Original signal
g~ (t )   g (nTs ) p (t  nTs )
n
(Interpolation)

g(t) g~ (t )
Sampled signal
Original signal
g~ (t )   g (nTs ) p (t  nTs )  g (t )  p (t )
n
(Interpolation)
New interpolation formula:
n
(Interpolation)
n
In frequency domain, it becomes

~ 1
G ( f )  P( f )
Ts
 G ( f  nf )
n
s
(Interpolation)
n

~ 1
G ( f )  P( f )
Ts
 G ( f  nf )
n
s
1
Remember, G ( f ) 
Ts
 G ( f  nf )
n
s
(Interpolation)
n

~ 1
G ( f )  P( f )
Ts
 G ( f  nf )
n
s
The reconstructed signal g~ (t ) also contains multiples replicas

of G(f) at every fs distance apart and filtered by P(f)
(Interpolation)
n

~ 1
G ( f )  P( f )
Ts
 G ( f  nf )
n
s
To fully recover g(t) from g~ (t ) , we need another filter or

Equalizer E(f)
(Interpolation)
~ 1
G ( f )  P( f )
Ts
 G ( f  nf )
n
s
To fully recover g(t) from g~ (t ) , we need another filter or

Equalizer E(f)
That means, ~
G ( f )  E ( f )G ( f )
1
 E ( f ) P( f )
Ts
 G ( f  nf )
n
s
(Interpolation)
~ 1
G ( f )  P( f )
Ts
 G ( f  nf )
n
s
~
G ( f )  E ( f )G ( f )
E(f)P(f) must
1
 E ( f ) P( f )
Ts
 G ( f  nf )
s REMOVE this
n portion
~
G( f )
-2fs -fs fs 2fs

(Interpolation)
~ 1
G ( f )  P( f )
Ts
 G ( f  nf )
n
s
~
G ( f )  E ( f )G ( f )
E(f)P(f) must
1
 E ( f ) P( f )
Ts
 G ( f  nf )
s REMOVE this
n portion
~
G( f )
E(f)P(f) = 0 |f| > fs- B
-2fs -fs fs 2fs

(Interpolation)
~ 1
G ( f )  P( f )
Ts
 G ( f  nf )
n
s
~
G ( f )  E ( f )G ( f )
E(f)P(f) must
1
 E ( f ) P( f )
Ts
 G ( f  nf )
s PRESERE this
n portion
~
G( f )
-2fs -fs fs 2fs

(Interpolation)
~ 1
G ( f )  P( f )
Ts
 G ( f  nf )
n
s
~
G ( f )  E ( f )G ( f )
E(f)P(f) must
1
 E ( f ) P( f )
Ts
 G ( f  nf )s PRESERE this
n portion
~
G( f )
E(f)P(f) = Ts |f| < B
-2fs -fs -B B fs 2fs

(Interpolation)
~ 1
G ( f )  P( f )
Ts
 G ( f  nf )
n
s
E(f)P(f) is flexible
~ in this portion
G ( f )  E ( f )G ( f )
1 B < |f| < fs- B
 E ( f ) P( f )
Ts
 G ( f  nf )
n
s
~
G( f )
-2fs -fs fs 2fs

(Interpolation)
~ 1
G ( f )  P( f )
Ts
 G ( f  nf )
n
s
~
G ( f )  E ( f )G ( f )
1
 E ( f ) P( f )
Ts
 G ( f  nf )
n
s
~
G( f )
3 conditions together
Ts P ( f ) f  B

E ( f )  Flexible B  f  f s  B
0 f  fs  B

-2fs -fs fs 2fs
(Interpolation)
Samples
Sampling at g(nTs) g~ (t ) Equalizer
g(t)
frequency fs × E(f)
g(t)
Pulse
generator
(Interpolation)
Samples
g(t)
g(t)
Pulse
generator
Let this pulse generator generates p(t)

this short pulse 1
t
0 Tp
(Interpolation)
Samples
g(t)
g(t)
Pulse
generator
Let this pulse generator generates p(t)

this short pulse 1  t  0.5T p 
p (t )    
 T 
t  p 
0 Tp
(Interpolation)
Samples
g(t)
g(t)
Pulse
generator
0
(Interpolation)
Samples
g(t)
g(t)
Pulse
generator
0
(Interpolation)
Samples
g(t)
g(t)
Pulse
generator
0
(Interpolation)
Samples
g(t)
g(t)
Pulse
generator
0
Ts
(Interpolation)
Samples
g(t)
g(t)
Pulse
generator Generates
this pulse
p(t)
1
0 t
Ts 0 Tp
(Interpolation)
Samples
g(t)
g(t)
Pulse
generator
p(t)
1
t
Ts 0 Tp
Tp
(Interpolation)
Samples
g(t)
g(t)
Pulse
generator What happens
if Tp→0?
p(t)
1
t
Ts 0 Tp
Tp
(Interpolation)
Samples
g(t)
g(t)
Pulse
After
generator
equalizing
p(t)
1
t
Ts 0 Tp
Tp
(Interpolation)
Samples
g(t)
g(t)
Pulse Why are these two

generator different?
Why do we need
equalizer?
p(t)
1
t
Ts 0 Tp
Tp
(Interpolation)
Let’s find the FT
response of the
rectangular
reconstruction p(t)  t  0.5T p 
1 p (t )    
pulse  T 
 p 
t
0 Tp
(Interpolation)
Let’s find the FT
response of the
rectangular
1 p (t )    
pulse  T 
 p 
t
0 Tp
 jfT p
P ( f )  T p sinc(fT p )e
(Interpolation)
Let’s find the FT
response of the
rectangular
1 p (t )    
pulse  T 
 p 
t
0 Tp
 jfT p
P ( f )  T p sinc(fT p )e
t 
    sinc (f )
 
g (t  t0 )  G ( f ) exp( j 2ft0 )
(Interpolation)
rectangular |P(f)|
p(t)
1
t

0 Tp
3 2 1 0 1 2 3
 t  0.5T p  Tp Tp Tp Tp Tp Tp
p (t )    
 T 
 p  P ( f )  T p sinc(fT p )e
 jfT p
(Interpolation)
rectangular |P(f)|
FT of Ideal
p(t) Reconstruction
pulse
1
t

0 Tp
3 2 1 0 1 2 3
 t  0.5T p  Tp Tp Tp Tp Tp Tp
p (t )    
 T 
 p  P ( f )  T p sinc(fT p )e
 jfT p
(Interpolation)
Samples
g(t)
g(t)
Pulse
After
generator
equalizing
p(t)
1
t
Ts 0 Tp
Tp
(Interpolation)
~ 1
G ( f )  P( f )
Ts
 G ( f  nf )
n
s
~
G ( f )  E ( f )G ( f )
1
 E ( f ) P( f )
Ts
 G ( f  nf )
n
s
~
G( f )
Equalizer
Ts P ( f ) f  B

0 f  fs  B

-2fs -fs fs 2fs
(Interpolation)
~ 1
G ( f )  P( f )
Ts
 G ( f  nf ) s p(t)
n 1
~
G ( f )  E ( f )G ( f ) t
0 Tp
1
 E ( f ) P( f )
Ts
 G ( f  nf )
n
s
~
G( f )
Ts P ( f ) f  B

0 f  fs  B

-2fs -fs fs 2fs
(Interpolation)
~ 1
G ( f )  P( f )
Ts
 G ( f  nf ) s p(t)  t  0.5T p 
p (t )    
n 1  T 
~  p 
G ( f )  E ( f )G ( f ) t
0 Tp
1
 E ( f ) P( f )
Ts
 G ( f  nf )
n
s
~
G( f )
Ts P ( f ) f  B

0 f  fs  B

-2fs -fs fs 2fs
(Interpolation)
~ 1
G ( f )  P( f )
Ts
 G ( f  nf ) s p(t)  t  0.5T p 
p (t )    
n 1  T 
~  p 
G ( f )  E ( f )G ( f ) t
0 Tp
1
 E ( f ) P( f )
Ts
 G ( f  nf )
n
s
 jfT p
P( f )  T p sinc(fT p )e
Ts P ( f ) f  B

0 f  fs  B

(Interpolation)
~ 1
G ( f )  P( f )
Ts
 G ( f  nf ) s p(t)  t  0.5T p 
p (t )    
n 1  T 
~  p 
G ( f )  E ( f )G ( f ) t
0 Tp
1
 E ( f ) P( f )
Ts
 G ( f  nf )
n
s
 jfT p
P( f )  T p sinc(fT p )e
Ts P ( f ) f  B
 f
E ( f )  Ts  f B
0 f  fs  B sin(fT p )

(Interpolation)
p(t)
1
t
Ts 0 Tp
Tp
Tp must be < 1/B,

f
E ( f )  Ts  f B otherwise signal will be
sin(fT p )
distorted
(Interpolation)
p(t)
1
t
Ts 0 Tp
Tp
Tp must be < 1/B,

f
E ( f )  Ts  f B otherwise signal will be
sin(fT p )
distorted
Tp > 0, otherwise E(f) will
have infinite gain
(Interpolation)
p(t)
1
t
Ts 0 Tp
Tp
in special case when Tp is very small,

f Ts
E ( f )  Ts   f B
sin(fT p ) T p
CSE 311:
Data Communication
Instructor:
Sampling
and
Sampling Theorem
G( f )
g(t)
g (t ) G( f )
-2fs -fs fs 2fs

Realizability of Reconstruction
Filters
Spectrum G ( f ) when signal g(t) is
sampled at the Nyquist rate fs = 2B
G( f )
f
B fs
Filters
G( f )
f
B fs
Repetitions of G(f) without any gap
between successive cycles
Filters
G( f )
f
B fs
Ideal reconstruction
filter (Box filter)
Filters
G( f )
f
B fs
Ideal reconstruction
filter (Box filter)
But box filter is unrealizable; can be
approximated only with infinite time delay
Filters
Alternate: Sample at a rate higher than
Nyquist rate fs > 2B
G( f )
f
B fs
Filters
G( f )
f
B fs
Repetitions of G(f) with finite gap

between successive cycles
Filters
G( f )
f
B fs
Low pass filter with

gradual cutoff
characteristics
Filters
G( f )
f
B fs
Low pass filter with

gradual cutoff
characteristics Disadvantage: filter gain is zero beyond first cycle
of G(f) → fails to satisfy Paley-Wiener criteria
Filters
Findings:
• Exact recovery of g(t) from samples is impossible, even fs > 2B
• As fs increases, recovered signal approaches g(t)

Wrong Assumption on Bandwidth:
Treachery of Aliasing
Definition of aliasing
• A false or assumed identity: same person, different names
• an alternative name or label that refers to a file, command,

address, or other item, and can be used to locate or access it.
in signal processing
• each of a group of signal frequencies that, when sampled

at a given uniform rate, would give the same set of
sampled values, and thus might be incorrectly substituted
for one another when reconstructing the original signal
A signal of 60 Hz frequency
Sampled at 400 Hz frequency

Sampled Signal
Another signal with 340 Hz frequency


We got the same sampled signal

The third signal: 460 Hz frequency


No difference: the same sampled signal is obtained

from 3 different analog signals
While reconstructing, these samples may generate all three

signals
While reconstructing, these samples may generate all three

signals
60, 340, and 460, are aliased frequencies for 400 Hz

sampling rate
Aliasing: Cause and Result
• Aliasing occurs when sampling rate is < Nyquist rate

• Spectral cycles overlaps
G(f)
Spectrum of
0 f
unsampled g(t) -B B
G( f )
Spectrum of over
f
sampled g(t) -2fs -fs -B 0 B fs 2fs
G( f )
Spectrum of critically
f
G( f )
Spectrum of under
f
sampled g(t) -2fs -fs 0 fs 2fs
G(f)
Spectrum of
0 f
unsampled g(t) -B B
G( f )
Spectrum of over
f
G( f )
f
G( f )
Overlapping cycles
f
of G(f) -2fs -fs 0 fs 2fs
G(f)
Spectrum of
0 f
unsampled g(t) -B B
G( f )
Spectrum of over
f
G( f )
f
G( f )
a complete cycle of G(f)

f
cannot be recovered -2fs -fs -B 0 B fs 2fs
G(f)
Spectrum of
0 f
unsampled g(t) -B B
G( f )
Spectrum of over
f
G( f )
f
G( f )
Original g(t) may not be

f
recovered, too -2fs -fs 0 fs 2fs
G(f)
Spectrum of
0 f
unsampled g(t) -B B
G( f )
Spectrum of over
f
G( f )
f
G( f )
Rather aliased signals may

f
be reconstructed -2fs -fs 0 fs 2fs
While sampling and reconstructing,
• Signal g(t) is assumed to be band-limited to B
However,
• all practical signals are time-limited
However,
• time limited signal CANNOT be band-limited
However,
• time limited signal CANNOT be band-limited
• practical signals are therefore NON-band-limited
G(f)
f
0
A practical signal with
infinite bandwidth (B)
G(f)
f
0
A practical signal with
infinite bandwidth (B)
Sampling will result G ( f ) consisting of overlapping G(f) at every fs

Spectrum of
G( f ) Sampled signal
G(f)
-fs 0 fs

Spectrum of
G(f)
-fs 0 fs
Overlapping
tails

Spectrum of
G(f)
-fs 0 fs
Spectral overlap is unavoidable as bandwidth is infinite

Spectrum of
G(f)
-fs 0 fs
Higher sampling rate ONLY reduces overlap; but cannot eliminate

Spectrum of
G(f)
-fs 0 fs
NO complete information of G(f)

G ( f ) has
g(t) cannot be fully recovered from g (t )
Ideal reconstruction Spectrum of
filter H(f) G( f ) Sampled signal
-fs 0 fs
-fs/2 fs/2
Ideal reconstruction filter CANNOT separate G(f), rather generates

Ga(f) which is the spectrum of aliased signal ga(t)
H(f) Spectrum of
-fs 0 fs
-fs/2 fs/2
Reconstructed
spectrum Ga(f)
H(f) Spectrum of
-fs 0 fs
-fs/2 fs/2
Lost tail is
folded back Lost tail
Reconstructed
spectrum Ga(f)
H(f) Spectrum of
-fs 0 fs
-fs/2 fs/2
Lost tail is
Folding
frequency
H(f) Spectrum of
-fs 0 fs
-fs/2 fs/2
Lost tail is
Lost tail is folded back onto Folding

itself at folding frequency frequency
H(f) Spectrum of
-fs 0 fs
-fs/2 fs/2
fs/2 - fz fs/2 +fz
Component at fs/2 +fz is folded

back to fs/2 -fz
H(f) Spectrum of
-fs 0 fs
-fs/2 fs/2
fs/2 - fz fs/2 +fz
Tail folding is known as

Component at fs/2 +fz is folded
spectral folding or aliasing
back to fs/2 -fz
H(f) Spectrum of
-fs 0 fs
-fs/2 fs/2
2. Folded back tail
distorts lower
Reconstructed frequencies
spectrum Ga(f)
1. Lost tail results in
loss of higher
TWO frequencies
Problems! -fs/2 fs/2
Combatting Aliasing
H(f) Spectrum of
-fs 0 fs
-fs/2 fs/2
this portion is the
main betrayer
Combatting Aliasing
H(f) Spectrum of
-fs 0 fs
-fs/2 fs/2
REMOVE this
portion before
sampling using an
Antialiasing filter
(ILPF)
Combatting Aliasing
H(f) Spectrum of
-fs 0 fs
-fs/2 fs/2
Antialiasing filter
gaa(t) g aa (t )
g(t) Haa(f) Sampler
Gaa ( f )
T(t)
Combatting Aliasing
H(f) Spectrum of
-fs 0 fs
-fs/2 fs/2
H(f)
Gaa ( f )
Sample signal
spectrum
Reconstructed spectrum
no distortion of low frequency
-fs/2 fs/2
Lost tail results in loss of high
frequencies
Combatting Aliasing
H(f)
Gaa ( f )
Sample signal
spectrum
-fs/2 fs/2
frequencies
Advantage: suppressed
component cannot reappear
Gaa ( f )  G ( f ) | f | f s / 2 in low frequency
components
Combatting Aliasing
H(f)
Gaa ( f )
Sample signal
spectrum
-fs/2 fs/2
frequencies
Other characteristics:
• Antialiasing filter band-limits g(t) to fs/2
• Reduces noise too, otherwise noise might be folded back to low
frequency components
• Noises beyond fs/2 are suppressed
Sampling Effect: Non-band-limited
signal to band-limited signal
G(f)
f
0
G ( f ) = Spectrum of
G( f ) Sampled signal, g (t )
G(f)
-fs 0 fs
-fs/2 fs/2
Bandwidth of G(f) is practically infinite
G(f)
f
0
-fs 0 fs
-fs/2 fs/2
fs << Nyquist rate
G(f)
f
0
-fs 0 fs
-fs/2 fs/2
G ( f ) , that means, spectrum of g (t ) consists of overlapping cycles of G(f)
G(f)
f
0
-fs 0 fs
-fs/2 fs/2
g (t ) are sub-Nyquist samples of g(t)!
-fs 0 fs
-fs/2 fs/2
The above bold line spectrum can be viewed as repetition of the following Ga(f)
band-limited to fs/2
Ga(f)
-fs/2 0 fs/2
-fs 0 fs
-fs/2 fs/2
The above bold line spectrum can be viewed as repetition of the following Ga(f)
band-limited to fs/2 Ga(f) = spectrum of ga(t)
-fs/2 0 fs/2
-fs 0 fs
-fs/2 fs/2
g (t ) are Nyquist samples of ga(t)

Ga(f) = spectrum of ga(t)
-fs/2 0 fs/2
-fs 0 fs
-fs/2 fs/2
Sampling a non-band-limited signal g(t) at fs is equivalent to Nyquist sampling of

some signal ga(t) band-limited to fs/2
CSE 311:
Data Communication
Instructor:
Sampling
and
G(f) = spectrum of g(t)
f
0
-fs 0 fs
-fs/2 fs/2
G(f)
-fs/2 0 fs/2
-fs 0 fs
-fs/2 fs/2
G(f)
Sampling a non-band-limited signal g(t) at fs is equivalent to Nyquist sampling of

some signal ga(t) band-limited to fs/2
f
0
Let
sub-Nyquist Sampling of g(t) at fs generates samples g(0), g(Ts), g(2Ts), g(3Ts),, . . .
-fs/2 0 fs/2
Let
Nyquist Sampling of ga(t) at fs generates samples ga(0), ga(Ts), ga (2Ts), ga (3Ts),, . . .
G(f) = spectrum of g(t) Ga(f) = spectrum of ga(t)
f
0 -fs/2 0 fs/2
sub-Nyquist Samples g(0), Nyquist Samples ga(0), ga(Ts), ga(2Ts),
g(Ts), g(2Ts), g(3Ts),, . . . ga(3Ts),, . . .
f
0 -fs/2 0 fs/2
g(Ts), g(2Ts), g(3Ts),, . . . ga(3Ts),, . . .
According to sampling effect that we saw,
g(0) = ga(0), g(Ts) = ga(Ts), g (2Ts) = ga(2Ts), g (3Ts) = ga(3Ts), and so on

f
0 -fs/2 0 fs/2
g(Ts), g(2Ts), g(3Ts),, . . . ga(3Ts),, . . .
Therefore,
g(nTs) = ga(nTs) = gn
f
0 -fs/2 0 fs/2
g(Ts), g(2Ts), g(3Ts),, . . . ga(3Ts),, . . .
In other words, sampling g(t) and ga(t) at

Therefore,
the rate of fs =1/Ts will generate the same
g(nTs) = ga(nTs) = gn data sequence, gn
f
0 -fs/2 0 fs/2
g(Ts), g(2Ts), g(3Ts),, . . . ga(3Ts),, . . .

Therefore,
This means, data sequence gn can generate ga(t) by interpolation
Maximum Information Rate of a
Channel with BW = B Hz
Assume
• Error free, noise less channel
• Channel bandwidth is B
Assume
We will prove
• Maximum 2B pieces of information can be sent per second
Assume
Previous Knowledge
• Channel can send a low pass signal of B Hz
• This signal can be recovered from samples uniformly taken at 2B samples per
second
Assume
Previous Knowledge
second
• This means, 2B samples/second can be sent through the channel
Assume
Previous Knowledge
second
• This means, 2B samples/second can be sent through the channel
We have to prove that

• A sequence of data at the rate of 2B Hz can come from uniform sampling of a
signal of bandwidth B Hz
• The signal can be recovered from this data sequence
Assume a sequence of samples g0, g1, g2, g3, . . . denoted as {gn} at the rate of 2B s/s
We will always find a signal g(t) whose samples g(0), g(Ts), g(2Ts), g(3Ts), . .
matches with {gn}.
f
0
We will always find a signal g(t) whose samples g(0), g(Ts), g(2Ts), g(3Ts), . .
matches with {gn}.
G(f) = spectrum of g(t) This means,

gn = g(nTs)
f
0
f
0 -fs/2 0 fs/2
g(Ts), g(2Ts), g(3Ts),, . . . ga(3Ts),, . . .

Therefore,
This means, data sequence gn can generate ga(t) by interpolation
f
0 -fs/2 0 fs/2
g(Ts), g(2Ts), g(3Ts),, . . . ga(3Ts),, . . .
g (t )   g (nTs ) (t  nTs )
n
f
0 -fs/2 0 fs/2
g(Ts), g(2Ts), g(3Ts),, . . . ga(3Ts),, . . .
g (t )   g (nTs ) (t  nTs )
n
  g a (nTs ) (t  nTs )
n
f
0 -fs/2 0 fs/2
g(Ts), g(2Ts), g(3Ts),, . . . ga(3Ts),, . . .
g (t )   g (nTs ) (t  nTs )
n
  g a (nTs ) (t  nTs )
n
  g n (t  nTs )
n
f
0 -fs/2 0 fs/2
g(Ts), g(2Ts), g(3Ts),, . . . ga(3Ts),, . . .
To recover ga(t), we can use {gn} using,

g (t )   g (nTs ) (t  nTs )
n
  g a (nTs ) (t  nTs ) g a (t )   g n sinc(2Bt  n )
n n
  g n (t  nTs )
n
Application of Sampling Theorem
Continuous time signal
sampling
Discrete time signal

Continuous time signal
can be represented by pulse train and transmitted thereafter

Pulse modulations: different ways to transmit sampled signal

modifying pulse trains
PAM: Pulse amplitude is modulated

PWM: Pulse width is modulated
PPM: Pulse position is modulated

Pulse code modulation:

• most widely used pulse modulation
• each sample value is converted to a set of pulses.
TDM: Pulses from multiple signals are interweaved
on the same channel
TDM Example
TDM: dual of FDM where different signals share channel bandwidth
TDM Example
G1(f) G2(f)
FDM Example
f
0 f1 f2
Pulse Code Modulation (PCM)
LPF Bit
Sampler Quantizer
Encoder 1 0 1 1 0 1
PCM system: basically an ADC
Two major Steps:

• Sampling and
• quantizing
Message Signal
• 2 major steps
– Sampling
The range (–mp, mp) is
– Quantizing
divided into L sub-
mp intervals, each of
m(t) Quantized samples m(t) magnitude v
2m p
v 
L
2m p
L
-mp
Message Signal
• 2 major steps
– Sampling
– Quantizing
divided into L sub-
2m p
v 
L
L is known as
2m p quantization level
L
-mp
Message Signal
• 2 major steps
– Sampling
A sampled value is
– Quantizing
placed into one of
mp these L sub-intervals,
m(t) Quantized samples m(t) thus gets ONE of the
L values
2m p
L
-mp
Message Signal
• 2 major steps
– Sampling
A sampled value is
– Quantizing
placed into one of
mp these L sub-intervals,
m(t) Quantized samples m(t) thus gets ONE of the
L values
Signal is known as
L-ary digital signal
2m p
L
-mp
Message Signal
L-ary digital signal is converted to binary

digital signal using pulse coding
Each of L values is encoded as a group of

binary digits
Message Signal
L-ary digital signal is converted to binary

digital signal using pulse coding
Each of L values is encoded as a group of

binary digits
Each bit is transmitted using a distinct pulse

shape
Message Signal
Analog signal bandwidth to digital data rate
Audio signal b/w = 15 KHz
However, up to 3400 Hz is sufficient for articulation (intelligibility).
Fidelity is compromised!
Message Signal
However, up to 3400 Hz is sufficient for articulation (intelligibility)
B = 3400 Hz
fs = 8000 > 2B
Message Signal
B = 3400 Hz
fs = 8000 > 2B
Quantization level, L = 256 (8 bits)

Message Signal
B = 3400 Hz
fs = 8000 > 2B
Quantization level, L = 256 (8 bits)
Data rate = 8000*8 =64000 pulse/second = 64 Kbps

Message Signal
Example 2: data rate for compact disc
Fidelity is required!
B = 20000 Hz
fs = 441000 Hz > 2B
Quantization level, L = 65,536 (16 bits)
Data rate = 44100*16 = 1.4 Mbps

Advantages of Digital
Communication
Self Study
Quantization

divided into L sub-
2m p
v 
L
L is known as
L
-mp
Quantization
mp is NOT the signal PEAK, rather is it’s the LIMIT of the quantizer

divided into L sub-
2m p
v 
L
L is known as
L
-mp
Quantization
k-th sample value m(kTs) is replaced by the midpoint of an interval where it lies

divided into L sub-
2m p
v 
L
L is known as
L
-mp
Quantization
Replaced by
m(kTs ) mˆ (kTs )
divided into L sub-
2m p
v 
L
L is known as
L
-mp
Quantization
Replaced by
m(kTs ) mˆ (kTs )
quantization error is unavoidable which is lies in (-v/2, v/2)
mp
2m p
v 
L
L is known as
L
-mp
Quantization
Replaced by
m(kTs ) mˆ (kTs )
mp
mˆ (kTs )  v / 2
mˆ (kTs )
mˆ (kTs )  v / 2
2m p
L
2m p
-mp v 
L
Quantization
Replaced by
m(kTs ) mˆ (kTs )
mp
mˆ (kTs )  v / 2
m(kTs)
mˆ (kTs )
mˆ (kTs )  v / 2
2m p
L
2m p
-mp v 
L
Quantization
Replaced by
m(kTs ) mˆ (kTs )
mp
mˆ (kTs )  v / 2
m(kTs)
mˆ (kTs )
mˆ (kTs )  v / 2
2m p
L
2m p
-mp v 
L
Quantization
Replaced by
m(kTs ) mˆ (kTs )
mp
error
v/2
q(kTs) = q
0
-v/2
2m p
L
2m p
-mp v 
L
Quantization
Replaced by
m(kTs ) mˆ (kTs )
mp
mˆ (kTs )  v / 2
mˆ (kTs )
m(kTs)
mˆ (kTs )  v / 2
2m p
L
2m p
-mp v 
L
Quantization
Replaced by
m(kTs ) mˆ (kTs )
mp
error
v/2
0
q(kTs) = q
-v/2
2m p
L
2m p
-mp v 
L
Quantization
Binary
quantization coding
m(kTs ) mˆ (kTs ) 101101 . . . Transmission
pulse
Pulse detection
at receiving end
Quantization
Binary
quantization coding
m(kTs ) mˆ (kTs ) 101101 . . . Transmission
pulse
quantization Pulse detection

error at receiving end
Pulse
detection
error
Quantization
If there were no quantization error,
m(t )   m(kTs )sinc(2Bt  k )

k
Quantization
m(t )   m(kTs )sinc(2Bt  k )

k
Due to quantization error,
mˆ (t )   mˆ (kTs )sinc( 2Bt  k )

k
Quantization
m(t )   m(kTs )sinc(2Bt  k )

k
Due to quantization error,
mˆ (t )   mˆ (kTs )sinc( 2Bt  k )

k
Quantization error q(t),

q (t )  mˆ (t )  m(t )
Quantization
Quantization error or quantization noise or undesired signal,
q (t )   mˆ (kTs )  m(kTs )sinc( 2Bt  k )

k
  q (kTs )sinc( 2Bt  k )

k
Quantization
Quantization noise,
q (t )   mˆ (kTs )  m(kTs )sinc( 2Bt  k )

k
  q (kTs )sinc( 2Bt  k )

k
q(kTs) = Quantization
error for kth sample
Quantization
Power or Mean square of Quantization noise,

~~~~~~
2 lim 1 T / 2 2
q (t )   q (t ) dt
T   T T / 2
T /2 2
lim 1  
   
T   T T / 2  k
q ( kTs )sinc( 2Bt  k ) 

dt
Quantization

~~~~~~
2 lim 1 T / 2 2
q (t )   q (t ) dt
T   T T / 2
T /2 2
lim 1  
   
T   T T / 2  k
q ( kTs )sinc( 2Bt  k ) 

dt
lim 1 T / 2
  a1  a2  a3   dt
2
T   T T / 2
Quantization

~~~~~~
2 lim 1 T / 2 2
q (t )   q (t ) dt
T   T T / 2
T /2 2
lim 1  
   
T   T T / 2  k
q ( kTs )sinc( 2Bt  k ) 

dt
lim 1 T / 2
  a1  a2  a3   dt
2
T   T T / 2
lim 1 T / 2 2
 
T   T T / 2
a1 
 a 2
2  a 2

3    2 a1a 2  2 a1a3  2 a1a 4   dt
Quantization

~~~~~~
2 lim 1 T / 2 2
q (t )   q (t ) dt
T   T T / 2
T /2 2
lim 1  
   
T   T T / 2  k
q ( kTs )sinc( 2Bt  k ) 

dt
lim 1 T / 2 2
 
T   T T / 2
a1 a 2
2  a 2

3    2 a1a 2  2 a1a3  2 a1a 4   dt
lim 1 T / 2  2 
   
T   T T / 2  k
ak  2  am an  dt
m n 
Quantization
2
~~~~~~
2 lim 1 T / 2  
q (t )    
T   T T / 2  k
q (kTs )sinc(2Bt  k ) dt

lim 1 T / 2  2 
   
T   T T / 2  k
ak  2  am an  dt
mn 
lim 1 T / 2  2 2 
   
T   T T / 2  k
q (kTs )sinc (2Bt  k ) dt

lim 1 T / 2  
   2 
T   T T / 2  m  n
q ( mTs ) q ( nTs )sinc( 2Bt  m  )sinc( 2Bt  n  ) 

dt
Quantization
~~~~~~ lim 1 T / 2 
2 2 2 
q (t )    
T   T T / 2  k
q ( kTs )sinc ( 2Bt  k ) 

dt
lim 1 T / 2  
   2 
T   T T / 2  m  n
q ( mTs ) q ( nTs ) sinc( 2Bt  m ) sinc( 2Bt  n  ) 

dt
T /2
lim 1
  q 2 (kTs )  sinc 2 (2Bt  k )dt
T  T k T / 2
T /2
lim 1
 
T   T mn
q(mTs )q (nTs )  sinc( 2Bt  m )sinc( 2Bt  n )dt
T / 2
Quantization
~~~~~~ T /2
2 lim 1 2 2
q (t ) 
T  T k
 q ( kTs )  sinc (2Bt  k )dt
T / 2
T /2
lim 1
 
T   T mn
T / 2
Quantization
~~~~~~ T /2
2 lim 1 2 2
q (t ) 
T  T k
 q ( kTs )  sinc (2Bt  k )dt
T / 2
T /2
lim 1
 
T   T mn
T / 2
We can prove that,
  0 mn
sinc(2Bt  m )sinc(2Bt  n )dt   1 mn
 2B
Quantization
~~~~~~
2 lim 1 2
q (t )  
T   2BT k
q (kTs )
Quantization
~~~~~~
2 lim 1 2
q (t )  
T   2BT k
q (kTs )
As sampling frequency fs = 2B,

2BT = total no. of samples over averaging time T
RHS is the mean of the square of quantization error

Quantization
~~~~~~
2 lim 1 2
q (t )  
T   2BT k
q (kTs )
As sampling frequency fs = 2B,

2BT = total no. of samples over averaging time T
RHS is the mean of the square of quantization error
Therefore, power of quantization noise = mean square quantization error

Quantization
We know, quantization error q lies in (-v/2, v/2)
mp
error
v/2
0
q(kTs) = q
-v/2
2m p
L
2m p
-mp v 
L
Quantization
Mean square quantization error is given by

~~~ v / 2
2 1 2
q   q dq
v  v / 2
mp
error
v/2
0
q(kTs) = q
-v/2
2m p
L
2m p
-mp v 
L
Quantization

~~~ v / 2 2
2
q 
1
q 2
dq
v 

v  v / 2

12
Quantization

~~~ v / 2 2
2
q 
1
q 2
dq
v  m 2p

v  v / 2

12

3L2
2m p
where, v 
L
Quantization

~~~ v / 2 2
2
q 
1
q 2
dq
v  m 2p

v  v / 2

12

3L2
We proved, power of quantization noise (N0) = mean square quantization error

Quantization

~~~ v / 2 2
2
q 
1
q 2
dq
v  m 2p

v  v / 2

12

3L2
We proved, power of quantization noise (N0) = mean square quantization error
~~~~~~
2
~~~
2
m 2p
N 0  q (t )  q 
3L2
Quantization
power of quantization noise (N0) = mean square quantization error
~~~~~~
2
~~~
2
m 2p
N 0  q (t )  q 
3L2
~~~~~~
2
Assume, power of message signal (S0) is given by S 0  m (t )
Quantization
~~~~~~
2
~~~
2
m 2p
N 0  q (t )  q 
3L2
~~~~~~
2
Signal-to-noise ratio (SNR) is

~~~~~~
2
S0 2 m (t )
SNR   3L
N0 m 2p
Quantization
~~~~~~
2
S0 2 m (t )
SNR   3L
N0 m 2p
• Higher SNR means higher quality of received signal

• L increases SNR
• Higher limit of quantizer (mp) decreases SNR
Quantization
~~~~~~
2
S0 2 m (t )
SNR   3L
N0 m 2p
• Higher SNR means higher quality of received signal

• L increases SNR
• Higher limit of quantizer (mp) decreases SNR
~~~~~~
2
• SNR is linear function of signal power, S 0  m (t )
Quantization
~~~~~~
2
S0 2 m (t )
SNR   3L
N0 m 2p
~~~~~~
2
S0 varies
• from speaker to speaker
• due to different length of connecting circuits
Quantization
~~~~~~
2
S0 2 m (t )
SNR   3L
N0 m 2p
~~~~~~
2
S0 varies
For these reasons,

• SNR varies widely
• Quality of received signal deteriorates remarkably for soft speakers
Quantization
~~~~~~
2
S0 2 m (t )
SNR   3L
N0 m 2p
~~~~~~
2
S0 varies
For these reasons,

However statistically,
• Small amplitudes (soft speakers) predominate in speech
• Larger amplitudes (loud speakers) are less frequent
CSE 311:
Data Communication
Instructor:
Sampling
and
Quantization
Replaced by
m(kTs ) mˆ (kTs )
mp
error
v/2
0
q(kTs) = q
-v/2
2m p
L
2m p
-mp v 
L
Quantization
~~~~~~
2
~~~
2
m 2p
N 0  q (t )  q 
3L2
~~~~~~
2
Signal-to-noise ratio (SNR) is

~~~~~~
2
S0 2 m (t )
SNR   3L
N0 m 2p
Quantization
~~~~~~
2
S0 2 m (t )
SNR   3L
N0 m 2p
~~~~~~
2
S0 varies
For these reasons,

However statistically,
• Small amplitudes (soft speakers) predominate in speech
• Larger amplitudes (loud speakers) are less frequent
Quantization
~~~~~~
2
S0 2 m (t )
SNR   3L
N0 m 2p
Causes of the problem
2m p
• All amplitudes are uniformly quantized with equal step size v 
L
• Variation in amplitudes is NOT considered
Quantization
~~~~~~
2
S0 2 m (t )
SNR   3L
N0 m 2p
Causes of the problem
2m p
• All amplitudes are uniformly quantized with equal step size v 
L
• Variation in amplitudes is NOT considered
Quantization noise N0 = (v)2/12 is directly proportional to square of step size v
2
N 0  v 
Quantization
Solution to the problem: Non-uniform quantization
• Smaller step size for smaller amplitudes
• Lager step size for larger amplitudes
mp
2
1
-1
-2
-mp
Non-uniform quantization
Quantization
mp
Same result is obtained
by compressing signal
samples non-linearly and
then using a uniform 2
quantizaton 1
-1
-2
Non-linear Uniform
Compression Quantization
-mp
Non-uniform quantization Non-uniform quantization
Quantization
mp
output
2
1
Uniform
-1
-2
input
-mp
Non-uniform Non-uniform quantization
Quantization
output
• Input signal increment m is

mapped to larger output
increment y for smaller m
Uniform
• The opposite happens for larger

m
input
Non-uniform
Quantization
output
106
6
105
5
Uniform
2
input 102 1
10
output
Non-uniform input
Quantization
output
106
6
105
5
Uniform
2
input 102 1
10
output
Non-uniform input
Quantization
output
106
6
105
5
Logarithmic
Uniform
transform 2
input 102 1
10
output
Non-uniform input
Quantization
output
• Loud talkers are penalized by

larger step size v to
compensate soft talkers
Uniform
input
Non-uniform
Quantization
output
• Quantization noise becomes

lower for smaller input signal
m(t)
Uniform
• The opposite happens for larger

m(t)
2
input
N 0  v 
Non-uniform
Quantization
output
• Logarithmic compression
makes quantization noise
proportional to signal power
Uniform
~~~~~~
2
N 0  m (t )  S 0
input
Non-uniform
Quantization
output
• Logarithmic compression
makes quantization noise
proportional to signal power
Uniform
~~~~~~
2
N 0  m (t )  S 0
input
S0
SNR   constant
Non-uniform N0
Quantization: standard
Compression Characteristic
μ-law characteristic A-law characteristic
y y
m
m 
 mp
mp
 A  m  m 1
 0 
 
 1  ln A  m p  mp A
1  m  m y
y ln1  
 
0 1  1 1  ln Am  1  m  1
ln(1   )  mp  mp 1  ln A   A m
  m p  p
Restoring the Compressed Signal
Non-linear Uniform Expander

PCM channel
Compression Quantization Non-linearly
Non-uniform quantization
SNR improvement in μ-law compandor
S0 3L2

S0 N 0 1  ln  2
dB
N0
~~~~~~
2
Relative Signal power m (t ), dB
SNR improvement in μ-law compandor
Nearly constant SNR for S0 > 40dB
S0 3L2

S0 N 0 1  ln  2
dB
N0
~~~~~~
2
Relative Signal power m (t ), dB
Pulse Code Modulation (PCM)
Encoder
LPF Bit
Sampler Quantizer
Encoder 1 0 1 1 0 1
PCM system: basically an ADC

The PCM Encoder
LPF Bit
Sampler Quantizer
Encoder 1 0 1 1 0 1
m(kTs ) mˆ (kTs ) 1 0 1 1 0 1 ...

The PCM Encoder
LPF Bit
Sampler Quantizer
Encoder 1 0 1 1 0 1
m(kTs ) mˆ (kTs ) 1 0 1 1 0 1 ...
Analog-to-digital conversion:
taught at Electronics course
Self Study
Transmission Bandwidth and
Output SNR
Sampling at Encoding in
Nyquist rate Quantization n = log2L bits
m(t ) m(kTs ) mˆ (kTs ) 1 0 1 1 0 1 ...
Level: L
B Hz 2B Hz 2nB bits/s
Output SNR
m(t ) m(kTs ) mˆ (kTs ) 1 0 1 1 0 1 ...
Level: L
Transmission media has to send 2nB bits/sec

Output SNR
m(t ) m(kTs ) mˆ (kTs ) 1 0 1 1 0 1 ...
Level: L
Alternatively, transmission media has to send 2nB PIECES of information/sec

Output SNR
m(t ) m(kTs ) mˆ (kTs ) 1 0 1 1 0 1 ...
Level: L
Alternatively, transmission media has to send 2nB PIECES of information/sec
As 1 unit transmission (channel) bw (1 Hz) can transmit 2 pieces of information/sec,
Therefore, channel bandwidth, BT = nB Hz

SNR vs Channel Bandwidth, BT
~~~~~~
2
S0 2 m (t )
SNR   3L Without compression
N0 m 2p
S0 3L2
SNR   With non-linear compression
N 0 1  ln  2
~~~~~~
2
S0 2 m (t )
SNR   3L Without compression
N0 m 2p
S0 3L2
SNR   With non-linear compression
N 0 1  ln  2
S0
can be combined and written as, SNR   c  (2) 2 n = c.L2
N0
~~~~~~
2
S0 2 m (t )
SNR   3L
N0 m 2p
S0 3L2
SNR  
N 0 1  ln  2
S0
can be combined and written as, SNR   c  (2) 2 n = c.L2
N0
 ~~~~~~2
 3 m (t )
 m2 uncompressed case
where, c p
and, L = 2n
 3
 1  ln  2 compressed case

S0
SNR   c  ( 2) 2 n
N0
Replace n by BT/B where,
S0
 c  (2) 2 BT / B
N0
SNR exponentially increases with channel bandwidth BT

S0
SNR   c  ( 2) 2 n
N0
 S0   S0 
   10 log10  
  10 log10 c  (2) 2 n 
 N 0  dB  N0 
S0
SNR   c  ( 2) 2 n
N0
 S0   S0 
   10 log10  
  10 log10 c  (2) 2 n 
 N 0  dB  N0 
 10 log10 c  20n log10 2

 (  6n) dB
S0
SNR   c  ( 2) 2 n
N0
 S0   S0 
   10 log10  
  10 log10 c  (2) 2 n 
 N 0  dB  N0 
 10 log10 c  20n log10 2

 (  6n) dB
1 bit increase in L quadruples SNR ( ≡ 6 dB increase)

Application of PCM
Time Division Multiplexing (TDM): Pulses from multiple signals are
interweaved on the same channel
TDM Example
PCM in TDM: T1 Carrier System
- A Case Study
T1 Carrier
• physically a pair of copper wires
• replaces earlier technology that transmitted a single analog audio signal at a time
• now transmits 24 time-division-multiplexed PCM telephone signals
• can transmit 24 telephone calls simultaneously
• Current total transmission speed is 1.544 Mbps compared to previous 4 KHz

analog bandwidth
- A Case Study
24 input
channels
24 output
channels
- A Case Study
24 input
channels
• 24 input channels each carrying voice signal of 4KHz

- A Case Study
24 input
channels Commutators
• work as samplers
• are high speed electronic

switching circuits
Commutators
• are opened periodically with
narrow pulses of 2 μs
• sample 24 channels in turns at the

rate of 8 KHz
• generates 8K ×24 samples/second

- A Case Study
24 input
channels
8K ×24 samples/sec
- A Case Study
24 input
channels
8K ×24 s/s
Samples from different channels

- A Case Study
24 input
channels
8K ×24 s/s
One complete cycle

- A Case Study
24 input
channels
8K ×24 s/s
8 bit representation of
each sample
- A Case Study
24 input
channels
8K ×24 s/s
Information bits Information bits Information bits

- A Case Study
24 input
channels
8K ×24 s/s
Framing bit

- A Case Study
24 input
channels
8K ×24 s/s
1 complete cycle = 1 Frame = 1+8*24 bits =193 bits

Framing bit

- A Case Study
24 input
channels
8K ×24 s/s
193 × 8K bits/s =1.544 Mbps
1 complete cycle = 1 Frame = 1+8×24 bits =193 bits

Framing bit

- A Case Study
24 input
channels
8K ×24 s/s
193 × 8K bits/s =1.544 Mbps
8K complete cycle/s = 8K Frame/s

Therefore, 1 frame in every 125 μs
Framing bit

- A Case Study
24 input
channels
8K ×24 s/s
193 × 8K bits/s =1.544 Mbps

Speed = 8Kbps
8K complete cycle/s = 8K Frame/s
Therefore, 1 frame in every 125 μs
Framing bit

- A Case Study
24 input
channels
8K ×24 s/s
1.544 Mbps
Digital Signal Level 1 (DS1)
- A Case Study
24 input
channels
8K ×24 s/s
1.544 Mbps
Digital Signal Level 1 (DS1)
• DS1 signals can be progressively multiplexed to generate DS2, DS3, . . .
• T1 carrier system is used in US and Japan
• 30-channel multiplexed signal (2.048 Mbps) is used elsewhere in the World

- A Case Study
24 input
channels
DS1 (1.544 Mbps)

Commutators
24 output
channels
with regenerative repeaters,
each one is 6K feet apart
Decommutators
Synchronizing and Signaling in
T1 Carrier
Frame No.s

T1 Carrier
Frame No.s
We need to know where each frame begins

T1 Carrier
Frame No.s
Framing bit
A special bit (framing bit) added at the beginning of each frame

Framing bit’s speed is 8 kbps
T1 Carrier
Frame No.s
Framing bit

Framing bit’s speed is 8 kbps
as if an extra channel (Framing channel) generates these framing bits
T1 Carrier
Frame No.s
Framing bit

Sequence of framing bits, one from each frame, form a special pattern
T1 Carrier
Frame No.s
Framing bit
This special pattern is searched for synchronization.

Detection time: 0.4 – 6 ms
Reframe time: 50 ms
T1 Carrier
Frame No.s
Framing bit
T1 requires to also transmit signaling bits

• Dialing pulses
• On-hook/off-hook signals
T1 Carrier
Frame No.s
Framing bit
T1 requires to also transmit signaling bits All 8 slots are occupied for
• Dialing pulses information bits
• On-hook/off-hook signals
T1 Carrier
signaling bit
Frame No.s
Framing bit

T1 Carrier
signaling bit
Frame No.s
ONLY every 6th sample contains signaling bit in its LSB position
Therefore, Frame No.s 1, 7, 13 … contains signaling bits
Signaling bit rate = 8K/6 bps = 1333 bps
T1 Carrier
signaling bit
Frame No.s
In each of Frame No.s 1, 7, 13 …

7×24 = 168 information bits, 24 signaling bits, 1 framing bit
Total 168+24+1 = 193 bits
T1 Carrier
signaling bit
Frame No.s
In each of Frame No.s 1, 7, 13 … Robbed-bit-signaling

7×24 = 168 information bits, 24 signaling bits, 1 framing bit
Encoding is 7 5 encoding
Total 168+24+1 = 193 bits 6
T1 Carrier
Frame No.s
Framing bit
In all other frames except Frame No.s 1, 7, 13 …

8×24 = 192 information bits, 1 framing bit
Total 192+1 = 193 bits
How to Differentiate Signaling
and Non-Signaling Frames?
Signaling
frame
signaling bit
Frame No.s
Non-Signaling
frame

Concept of Superframe
Superframe
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Frame No.s
• 12 frames make a Superframe

• Framing bits form a special pattern 100111011100
Superframe
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Frame No.s
• The special pattern 100111011100 repeats in every 12 frames

• Identification of 6th and 12th frame is easy, and so is signaling
frames.
Superframe
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Frame No.s
• 2 signaling frames in a superframe

• 2 signaling bits in a superframe
• 4 state signaling is possible
Extended Superframe (ESF)
Structure of ESF
ESF
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Frame No.s
• 24 frames make an ESF

• Reduces framing bits
• Signaling bits in the LSB of 6th, 12th, 18th and 24th frames
• 4 signaling bits makes possibility of 16 state-signaling
Structure of ESF
ESF
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Frame No.s
• Reduces framing bits: every 4th frame has a framing bit

• 8 Kbps framing capacity is distributed to
– 2 Kbps for framing
– 2 Kbps for CRC-6
– 4 Kbps for data
Structure of ESF
ESF
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Frame No.s

framing bits make a pattern 001011: repeats in every 24
frames
Further improvement: Common
Channel Interoffice Signaling
CCIS
• A separate network for signaling

• decreases the necessity of robbed-bit-signaling
• All 8 bits can be used for transmitting information bits
CSE 311:
Data Communication
Instructor:
Sampling
and
T1 System Review
24 input
channels
8K ×24 s/s
4 KHz
analog 1.544 Mbps
signals Digital Signal Level 1 (DS1)
• Converts 24 4-KHz analog signals to digital signal of 1.544

Mbps
European Counterpart of T1
System
Conference on European Postal and Telegraph Administration
(CEPT)
• standardizes PCM with 256 time slots per frame
• capable to transmit 32 channel audio signals simultaneously
• uses 30 channels with 30×8 =240 information bits per frame
• uses remaining 16 bits for sync and signaling

Digital Multiplexing:
Generalization of T1 System
24 input
channels
8K ×24 s/s
4 KHz
analog 1.544 Mbps
• Converts 24 4-KHz analog signals to digital signal of 1.544

Mbps
24 input
channels
8K ×24 s/s
4 KHz
analog 1.544 Mbps
• Several slow-bit-rate digital signals can be combined to get high-

bit-rate signal and can time-share same medium (TDM)
• Input bit rates may be different

TDM can be
• Word-by-word basis interleaving (like T1 system)
• Bit-by-bit basis interleaving
8K ×24 s/s
Word-by-word
interleaving in T1
system
Digital Multiplexing: Word
Interleaving
Channels
Identical bit rates for
all channels
Used in: T1 system to generate DS1

Digital Multiplexing: Bit/Digit
Interleaving
Channels
Identical bit rates for
all channels
Used in: North American Digital

Hierarchy
Interleaving
Channels
When bit rates are not
identical
Bit rate R for Channel B, C and D

Bit Rate 3R for Channel A
Interleaving
Channels
identical

Higher rate channels

are allocated more
slots
Interleaving
Channels
identical

Frame size will be multiple of the

lowest common multiple of all rates
Interleaving
Channels
A different architecture when bit
rates are not identical
Digital Multiplexing: Receiver
Side
• Data has to be divided and distributed in different output

channels
• Requires correct identification of bits which is correct identification

of
• beginning of each frame
• slots in each frame
• bit in a slot
Digital Multiplexing: Receiver
Side
• Data has to be divided and distributed in different
channels
• Requires correct identification of bits which is correct

identification of
• beginning of each frame
• slots in each frame
• bit in a slot
All these are done by adding overhead bits

which consist of
• Framing and synchronizing bits
Structure of ESF
ESF
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Frame Numbers

framing bits make a pattern 001011: repeats in every 24
frames
DM1/2 Multiplexer Frame Format
M0 [48] CA [48] F0 [48] CA [48] CA [48] F1 [48]
M1 [48] CB [48] F0 [48] CB [48] CB [48] F1 [48]
M1 [48] CC [48] F0 [48] CC [48] CC [48] F1 [48]
M1 [48] CD [48] F0 [48] CD [48] CD [48] F1 [48]
• DM1/2: converts 4 × DS1 (1.544 Mbps) to 1 × DS2 (6.312 Mbps)

• Bit-by-bit interleaving of 4 channels of 1.544 Mbps
• The frame has 4 sub-frames
• Each sub-frame has 6 overhead bits
M0 [48] CA [48] F0 [48] CA [48] CA [48] F1 [48]
M1 [48] CB [48] F0 [48] CB [48] CB [48] F1 [48]
M1 [48] CC [48] F0 [48] CC [48] CC [48] F1 [48]
M1 [48] CD [48] F0 [48] CD [48] CD [48] F1 [48]
All [48]’s : Data bits, interleaved CA, CB, CC, CD, : Stuffing bits
from 4 channels
F0, F1 : Main framing bits
M0, M1 : Sub-framing bits

M0 [48] CA [48] F0 [48] CA [48] CA [48] F1 [48]
M1 [48] CB [48] F0 [48] CB [48] CB [48] F1 [48]
M1 [48] CC [48] F0 [48] CC [48] CC [48] F1 [48]
M1 [48] CD [48] F0 [48] CD [48] CD [48] F1 [48]
Bit wise
interleaved from
4 channels
M0 [48] CA [48] F0 [48] CA [48] CA [48] F1 [48]
M1 [48] CB [48] F0 [48] CB [48] CB [48] F1 [48]
M1 [48] CC [48] F0 [48] CC [48] CC [48] F1 [48]
M1 [48] CD [48] F0 [48] CD [48] CD [48] F1 [48]
Bit wise
interleaved from
4 channels
bitA bitB bitC bitD bitA bitB bitC bitD bitA bitB bitC bitD . . . . so on
M0 [48] CA [48] F0 [48] CA [48] CA [48] F1 [48]
M1 [48] CB [48] F0 [48] CB [48] CB [48] F1 [48]
M1 [48] CC [48] F0 [48] CC [48] CC [48] F1 [48]
M1 [48] CD [48] F0 [48] CD [48] CD [48] F1 [48]
Total Data bits = 48×6×4 = 1152 bits
Total overhead bits = 6×4 = 24 bits
Efficiency = 1152/1176 ≈ 98%

M0 [48] CA [48] F0 [48] CA [48] CA [48] F1 [48]
M1 [48] CB [48] F0 [48] CB [48] CB [48] F1 [48]
M1 [48] CC [48] F0 [48] CC [48] CC [48] F1 [48]
M1 [48] CD [48] F0 [48] CD [48] CD [48] F1 [48]
M0, = F0 = 0 Values of CA, CB, CC, CD, depend

on stuffing types
M1, = F1 = 1
M0 [48] CA [48] F0 [48] CA [48] CA [48] F1 [48]
M1 [48] CB [48] F0 [48] CB [48] CB [48] F1 [48]
M1 [48] CC [48] F0 [48] CC [48] CC [48] F1 [48]
M1 [48] CD [48] F0 [48] CD [48] CD [48] F1 [48]
Framing bits make a periodic pattern 010101… : used to synchronize on the

frame
M0 [48] CA [48] F0 [48] CA [48] CA [48] F1 [48]
M1 [48] CB [48] F0 [48] CB [48] CB [48] F1 [48]
M1 [48] CC [48] F0 [48] CC [48] CC [48] F1 [48]
M1 [48] CD [48] F0 [48] CD [48] CD [48] F1 [48]

frame
M0M1M1M1 = 0111 is used to identify sub-frames.

M0 [48] CA [48] F0 [48] CA [48] CA [48] F1 [48]
M1 [48] CB [48] F0 [48] CB [48] CB [48] F1 [48]
M1 [48] CC [48] F0 [48] CC [48] CC [48] F1 [48]
M1 [48] CD [48] F0 [48] CD [48] CD [48] F1 [48]

frame
M0M1M1M1 = 0111 is used to identify sub-frames.
Pattern M0M1M1M1 = 0111 also confirms genuine F0F1F0F1 pattern

Complication in Switching &
Signaling
• All channels are NOT active always: some transmits, some are idle
• multiplexer’s slots may be underutilized
• Additional input channels are allowed

Signaling
Complexity
• NO guaranty to hold: No. of active channels  No. of available slots
• Complicated switching operations

Signaling
Complexity
• NO guaranty to hold: No. of active channels  No. of available slots
• Complicated switching operations
Management
• Use previous statistics to minimize the overload
Asynchronous Channels & Bit
Stuffing
• Synchronization is lost due to different reasons
• variation in oscillation of oscillators
• variation in temperature of carrying cables

Stuffing
Cable length = 1000 km
2×108 Propagation Speed = 2×108 m/s

pulses/s
Stuffing

pulses/s
Propagation time for a pulse through the cable = 1000 km / (2×108 m/s)
=1/200 second
Stuffing

pulses/s
=1/200 second
No. of Pulses in transit within this time = 1/200 second * (2×108 pulses/s)
=106 pulses
Stuffing

pulses/s
=1/200 second
=106 pulses
10 F increase in cable temperature => 0.01% increase in propagation speed

Stuffing

pulses/s
=1/200 second --
++
No. of Pulses in transit within this time = 1/200 second * (2×108 pulses/s) ++
=106 pulses ++

Stuffing

pulses/s
=1/200 second --
No. of Pulses in transit within this time = 1/200 second * (2×108 pulses/s) ++
=106 pulses ++
Extra pulses must be stored somewhere

Stuffing

pulses/s
=1/200 second
=106 pulses
drop in cable temperature => rate of received pulses drops => vacant slots in
multiplexer => vacant slots to be stuffed with dummy bits
Stuffing

pulses/s
Findings:
• Even synchronous multiplexer system may receive bits asynchronously
• We need
• storage for / removal of extra bits
• Stuffing for vacant slots
Stuffing
Stuffing classification
• Positive
• Negative
• Positive/negative
Stuffing
Positive Stuffing
• Multiplexer's rate > total max rates of all incoming channels
• Time slots in multiplexer remain vacant at some stage
• Pulse stuffing becomes necessary
• Stuffed bit and position are encoded using C-bits

Stuffing
Positive Stuffing
Input signal to multiplexer
Transmitted signal with stuffing
Un-stuffed signal
Output signal with smoothing

Stuffing
Negative Stuffing
• Multiplexer's rate < total rates of all incoming channels
• Time slots in multiplexer becomes overloaded at some stage
• Multiplexer cannot accommodate all incoming signals
• Information of left-out pulses and position is transmitted through overhead

bits
Stuffing
Positive/Negative Stuffing
• Combination of positive and negative stuffing
• Nominal rate of multiplexer = total nominal rate of incoming channels
• positive stuffing at sometimes and negative at others
• All information is transmitted through overhead bits

Transmitting Stuffing
Information through C-bits
M0 [48] CA [48] F0 [48] CA [48] CA [48] F1 [48]
M1 [48] CB [48] F0 [48] CB [48] CB [48] F1 [48]
M1 [48] CC [48] F0 [48] CC [48] CC [48] F1 [48]
M1 [48] CD [48] F0 [48] CD [48] CD [48] F1 [48]
• Four incoming channels: A, B, C, D
• Only 1 bit stuffing is allowed per channel per frame

M0 [48] CA [48] F0 [48] CA [48] CA [48] F1 [48]
M1 [48] CB [48] F0 [48] CB [48] CB [48] F1 [48]
M1 [48] CC [48] F0 [48] CC [48] CC [48] F1 [48]
M1 [48] CD [48] F0 [48] CD [48] CD [48] F1 [48]
Channel Stuffing No-Stuffing

A CACACA = 111 CACACA = 000
B CBCBCB = 111 CBCBCB = 000
C CCCCCC = 111 CCCCCC = 000
D CDCDCD = 111 CDCDCD = 000
M0 [48] CA [48] F0 [48] CA [48] CA [48] F1 [48]
M1 [48] CB [48] F0 [48] CB [48] CB [48] F1 [48]
M1 [48] CC [48] F0 [48] CC [48] CC [48] F1 [48]
M1 [48] CD [48] F0 [48] CD [48] CD [48] F1 [48]
• The stuffed bit is the first bit following F1 in the corresponding sub-frame
Plesiochronous (almost
Synchronous) Digital Hierarchy
1 T1 multiplexer/
64 Kbps 2 Channel Bank/
DS0 Signal
24 digital switch
1
1.544 Mbps
2
DS1 Signal 3 DM 1/2
4
1
6.312 Mbps
2
DS2 Signal DM 2/3
7
• Hierarchy of multiplexers
to produce digital signals 1
44.736 Mbps
2 DM 3/4 NA
of progressively higher bit DS3 Signal
rates 3
139.264 Mbps: DS4NA

1 T1 multiplexer/
DS0 Signal
24 digital switch
1
1.544 Mbps
2
DS1 Signal 3 DM 1/2
4
1
6.312 Mbps
2
DS2 Signal DM 2/3
7
• Two major types of
multiplexers 1
44.736 Mbps
– One combines low–data- 2 DM 3/4 NA
DS3 Signal
rate channels 3
– Other combines high-data-
rate channels 139.264 Mbps: DS4NA
1 T1 multiplexer/
DS0 Signal
24 digital switch
1
1.544 Mbps
2
DS1 Signal 3 DM 1/2
4
1
6.312 Mbps
2
DS2 Signal DM 2/3
• Low–data-rate 7
multiplexer, Digital Data 1
44.736 Mbps
system (DDS) 2 DM 3/4 NA
DS3 Signal
– Generates 64 Kbps signal 3
– Combines channels of
rates from 2.4 Kbps up to
9.6 Kbps 139.264 Mbps: DS4NA
1 T1 multiplexer/
DS0 Signal
24 digital switch
1
1.544 Mbps
2
DS1 Signal 3 DM 1/2
4
1
6.312 Mbps
2
DS2 Signal DM 2/3
7
• high–data-rate
multiplexers, all others 1
44.736 Mbps
2 DM 3/4 NA
DS3 Signal
3
139.264 Mbps: DS4NA

1 T1 multiplexer/
DS0 Signal
24 digital switch
1
1.544 Mbps
2
DS1 Signal 3 DM 1/2
4
1
6.312 Mbps
2
DS2 Signal DM 2/3
7
• 24 channels in T1: not
necessarily all are voice 1
44.736 Mbps
2 DM 3/4 NA
channels DS3 Signal
• The same thing is true for 3
other higher levels
139.264 Mbps: DS4NA
Other Plesiochronous Digital
Hierarchies
PDH T-Carrier
worldwide US and Canada Japan
E4 139.264 Mbps 97.728 Mbps J4

×3
×4
E3 34.368 Mbps T3 44.736 Mbps 32.064 Mbps J3
×4 ×7
×5
E2 8.448 Mbps T2 6.312 Mbps J2
×4 ×3 ×4
E1 2.048 Mbps T1 1.544 Mbps J1

×30
×24
Single user line 64 Kbps
CSE 311:
Data Communication
Instructor:
Differential Pulse Code
Modulation (DPCM)
PCM and Transmission Channel
Bandwidth: Example-1
Analog signal bandwidth => digital data rate => Channel Bandwidth
requirement
Signal bandwidth, B = 3400 Hz

Sampling frequency, fs = 8000 > 2B
PCM Quantization level, L = 256 (8 bits)
Data rate = 8000*8 =64000 pulse/second = 64 Kbps

Bandwidth: Example-2
Example 2: data rate for compact disc
High Fidelity is required!
B = 20000 Hz
fs = 44100 Hz > 2B
Quantization level, L = 65,536 (16 bits)
Data rate = 44100*16 = 1.4 Mbps =1400 Kbps

Bandwidth
Findings
• PCM generates too many bits
• Requires high bandwidth to transmit them
Bandwidth
Causes:
Every sample is represented by n = log2L bits
mp
No. of quantization levels = L
m(t) Quantized samples of m(t)
2m p
L
-mp
Bandwidth
Causes:
mp

decreasing L
increasing v
2m p
v 
L
increasing NOISE
-mp
Bandwidth
Causes:
mp

decreasing L
increasing v
2m p
v 
L
increasing NOISE
-mp
Because quantization noise N0 = (v)2/12

Bandwidth
WAY OUT
• Successive samples are NOT independent, can be predicted
• Redundancy (CORRELATION) exists among neighboring samples
• Difference can be transmitted instead of actual sample
mp
2m p
v 
L
-mp
Bandwidth
WAY OUT
mp
m(t) Quantized samples of m(t) d[k] = m[k] - m[k-1]
Due t o redundancy,
d[k] <<< m[k]

2m p
v 
L
-mp
Bandwidth
WAY OUT
mp
decreases peak dp
2m p decreases vd
v 
L
-mp decreases NOISE
Bandwidth
Alternatively,
For same NOISE or SNR level, samples can be transmitted by fewer n = log2L bits
mp
decreases peak dp
v 
L
-mp decreases NOISE
Bandwidth
d[k] effect
• For same NOISE or SNR, fewer n = log2L bits
• For same n, reduced noise or higher SNR
mp
decreases peak dp
v 
L
-mp decreases NOISE
Bandwidth
At the receiver
• m[k] is reconstructed from d[k] and previous samples m[k-1]
m[k] = d[k] + m[k-1]

Bandwidth
Findings
• d[k] reduces the no. of transmitted bits
• Smaller is d[k], so is n
Bandwidth
Findings
Effective prediction of m[k] => further reduction of d[k]

Bandwidth
Findings
Effective prediction of m[k] => further reduction of d[k]
Assuming, mˆ [k ] is the predicted (estimated) value of m[k]

Calculate d[k] as d [k ]  m[k ]  mˆ [k ]
√
X instead of d[k] = m[k] - m[k-1]
Bandwidth
Better is the prediction, closer is mˆ [k ] to m[k]
Smaller is d[k]
Smaller is n
Bandwidth
Steps
Predict mˆ [k ]
Calculate the difference, d [k ]  m[k ]  mˆ [k ]
Represent d[k] in fewer bits and transmit

Bandwidth
Differential pulse code modulation (DPCM)
Predict mˆ [k ]
Calculate the difference, d [k ]  m[k ]  mˆ [k ]
Represent d[k] in fewer bits and transmit

Bandwidth
Differential pulse code modulation (DPCM)
difference betn successive samples is a special case of DPCM
In DPCM: d [k ]  m[k ]  mˆ [k ]
In simple difference: d[k] = m[k] - m[k-1]
which assumes mˆ [k ]  m[k  1]

Prediction in DPCM
The key step: predicting the future, a kind of fortune telling
Predict mˆ [k ] from past samples m[k-1], m[k-2], . . .

Prediction in DPCM
Assuming that, a signal m(t) has derivatives of all

orders at t,
we can predict m(t+Ts) from m(t)
where, Ts is sampling interval

Prediction in DPCM
According to Taylor’s series,
Ts2 Ts3
m(t  Ts )  m(t )  Ts m (t )   (t ) 
m (t )  
m
2! 3!
Prediction in DPCM
Ts2 Ts3
m(t  Ts )  m(t )  Ts m (t )   (t ) 
m (t )  
m
2! 3!
This formula predicts a future sample m(t+Ts) from past samples of

m(t) and its derivatives
Prediction in DPCM
Ts2 Ts3
m(t  Ts )  m(t )  Ts m (t )  m (t )  (t )  
m
2! 3!
 m(t )  Ts m (t ) for small Ts
Prediction in DPCM
Ts2 Ts3
m(t  Ts )  m(t )  Ts m (t )  m (t )  (t )  
m
2! 3!
To replace with discrete time samples,

Assume, t = kTs
k-th sample m[k] = m(t) = m(kTs)
(k+1)-th sample m[k+1] = m(t+Ts) =m(kTs+Ts)
Prediction in DPCM
Ts2 Ts3
m(t  Ts )  m(t )  Ts m (t )  m (t )  (t )  
m
2! 3!
And, the derivate is
 m[k ]  m[k  1] 
m (t )   
 Ts 
Prediction in DPCM
Ts2 Ts3
m(t  Ts )  m(t )  Ts m (t )  m (t )  (t )  
m
2! 3!
Finally,
 m[k ]  m[k  1] 
m[k  1]  m[k ]  Ts  
 Ts 
 2m[k ]  m[k  1]
Prediction in DPCM
m[k  1]  2m[k ]  m[k  1]

Prediction in DPCM
m[k  1]  2m[k ]  m[k  1]
m[k ]  2m[k  1]  m[k  2]

Prediction in DPCM
m[k  1]  2m[k ]  m[k  1]
m[k ]  2m[k  1]  m[k  2]
Generalization,
m[k ]  a1m[k  1]  a2 m[k  2]    a N m[k  N ]

Prediction in DPCM
Generalized prediction,
m[k ]  a1m[k  1]  a2 m[k  2]    a N m[k  N ]
Prediction in DPCM
m[k ]  a1m[k  1]  a2 m[k  2]    a N m[k  N ]
ˆ [k ]
The RHS is the predicted value of m[k], that is, m
Prediction in DPCM
m[k ]  a1m[k  1]  a2 m[k  2]    a N m[k  N ]
ˆ [k ]
The RHS is the predicted value of m[k], that is, m
Therefore,
mˆ [k ]  a1m[k  1]  a2 m[k  2]    a N m[k  N ]

Prediction in DPCM
Predicted output,
mˆ [k ]  a1m[k  1]  a2 m[k  2]    a N m[k  N ]

Prediction in DPCM
Predicted output,
mˆ [k ]  a1m[k  1]  a2 m[k  2]    a N m[k  N ]
The most simplified case:
mˆ [k ]  m[k  1]
The previous sample is the current sample
Prediction in DPCM
Predicted output,
mˆ [k ]  a1m[k  1]  a2 m[k  2]    a N m[k  N ]
Input m[k] Delay Delay Delay Delay

Ts Ts Ts Ts
a1 a2 aN-1 aN
Tapped Delay Implementation 

of Linear Predictor ˆ [k ]
Output, m
Prediction in DPCM
Predicted output,
mˆ [k ]  a1m[k  1]  a2 m[k  2]    a N m[k  N ]
Input m[k] Delay Delay Delay Delay

Ts Ts Ts Ts
a1 a2 aN-1 aN
Predictor 
ˆ [k ]
Output, m
Prediction in DPCM
Predicted output,
mˆ [k ]  a1m[k  1]  a2 m[k  2]    a N m[k  N ]
• We DO NOT transmit m[k]

• We transmit difference betn m[k] and mˆ [k ]
d [k ]  m[k ]  mˆ [k ]
• At receiver, we recalculate m[k] from d[k] and mˆ [k ]
m[k ]  d [k ]  mˆ [k ]
Prediction in DPCM
Transmitter m[k] + d[k]
 To channel
-
mˆ [k ]
Receiver
d[k] + m[k]
Input  Output
+
mˆ [k ] Predictor
Prediction in DPCM
 To channel
-
mˆ [k ]
At Receiver:
m[k ]  mˆ [k ]  d [k ]
 a1m[k  1]  a2 m[k  2]    a N m[k  N ]  d [k ]
d[k] + m[k]
Input  Output
+
mˆ [k ] Predictor
Prediction in DPCM
 To channel
-
mˆ [k ]
Difficulty at Transmitter:
We DO NOT transmit d[k], rather its quantized version dq[k]
d[k] + m[k]
Input  Output
+
mˆ [k ] Predictor
Prediction in DPCM
Transmitter m[k] + d[k] dq[k]
 Quantizer To channel
-
Difficulty at Transmitter:
We DO NOT transmit d[k], rather its quantized version dq[k]
d[k] + m[k]
Input  Output
+
mˆ [k ] Predictor
Prediction in DPCM
-
Receiver
dq[k] + mq[k]
Input  Output
+
Predictor
Prediction in DPCM
-
Predictor will generate

mˆ q [k ]  a1mq [k  1]  a2 mq [k  2]    a N mq [k  N ]
Receiver
dq[k] + mq[k]
Input  Output
+
Predictor
Prediction in DPCM
-
mˆ q [k ]
Predictor will generate
mˆ q [k ]  a1mq [k  1]  a2 mq [k  2]    a N mq [k  N ]
Receiver
dq[k] + mq[k]
Input  Output
+
mˆ q [k ] Predictor
Prediction in DPCM
-
mˆ q [k ]
The difference d [k ]  m[k ]  mˆ q [k ] is quantized to yield dq[k]:
dq[k]= d[k] +q[k], where q[k] is quantization error
Receiver
dq[k] + mq[k]
Input  Output
+
Prediction in DPCM
-
+
mˆ q [k ] + 
Predictor mq[k]
Receiver
dq[k] + mq[k]
Input  Output
+
Prediction in DPCM
-
+
mˆ q [k ] + 
Predictor mq[k]
Predictor output mˆ q [k ] is feedbacked to its input, that is,

mq [k ]  mˆ q [k ]  d q [k ]
 m[k ]  d [k ]  d q [k ]
 m[k ]  q[k ]
Prediction in DPCM
-
+
mˆ q [k ] + 
Predictor mq[k]

mq [k ]  mˆ q [k ]  d q [k ]
 m[k ]  d [k ]  d q [k ] because, d [k ]  m[k ]  mˆ q [k ]
 m[k ]  q[k ]
Prediction in DPCM
-
+
mˆ q [k ] + 
Predictor mq[k]

mq [k ]  mˆ q [k ]  d q [k ]
 m[k ]  d [k ]  d q [k ] because, d [k ]  m[k ]  mˆ q [k ]
 m[k ]  q[k ] because, dq[k]= d[k] +q[k]
SNR Improvement in DPCM
Let,
mp = peak value of message signal m(t) mp
dp = peak value of difference signal d(t) m(t)

Quantized samples
of m(t)
We know, 2m p
vm = quantization step of m(t) =
L
2d p 2mp
v 
vd = quantization step of d(t) = L
L
-mp
dp
Therefore, quantization step reduction by a factor of
mp
2
 dp 
As quantization noise N0 = (v) /12, it reduces by a factor of 
2 

 m p 
Improvement of DPCM: Delta
Modulation
Modulation
DPCM Motivation:
mp
2m p
v 
L
-mp
Modulation
DPCM Motivation:
mp

CORRELATION
further increases
if samples are
taken too closely
2m p
v 
L
-mp
Modulation
DPCM Motivation:
fs + +
mp

CORRELATION ++
d[k] --
2m p
v 
L
-mp 1-bit coding
Modulation
Delta Modulation:
• 1- bit DPCM
• Difference m[k ]  mˆ q [k ] is quantized in L = 2 levels
• Very inexpensive A/D conversion
• mˆ q [k ] is assumed to be the previous sample mq[k-1]
mˆ q [k ]  mq [k  1]
Transmitters: DPCM Vs. DM
DPCM m[k] + d[k] dq[k]
-
+
mˆ q [k ] + 
Predictor mq[k]
DM m[k] + d[k] dq[k]

-
mˆ q [k ]  mq[k-1] +
+ 
Delay Ts mq[k]
Receivers: DPCM Vs. DM
DPCM
dq[k] + mq[k]
Input  Output
+
DM
dq[k] + mq[k]
Input  Output
+
mq[k-1] Delay Ts
Delta Modulation
-
+
mq[k-1]
+ 
Delay Ts mq[k]
Receiver
dq[k] + mq[k]
Input  Output
+
mq[k-1] Delay Ts
Delta Modulation
-
+
mq[k-1]
+ 
Delay Ts mq[k]
mq[k] =mq[k-1] +dq[k]

Receiver
dq[k] + mq[k]
Input  Output
+
mq[k-1] Delay Ts
Delta Modulation
mq[k-1] =mq[k-2] +dq[k-1]

Receiver
dq[k] + mq[k]
Input  Output
+
mq[k-1] Delay Ts
Delta Modulation
mq[k] =mq[k-2] + dq[k] + dq[k-1]
mq[k-1] =mq[k-2] +dq[k-1]

Receiver
dq[k] + mq[k]
Input  Output
+
mq[k-1] Delay Ts
Delta Modulation
mq[k] =mq[k-2] + dq[k] + dq[k-1]
Finally, we get k
mq [k ]   d q [m]
m 0
Receiver
dq[k] + mq[k]
Input  Output
+
mq[k-1] Delay Ts
Delta Modulation
mq[k] =mq[k-2] + dq[k] + dq[k-1]
Finally, we get k
mq [k ]   d q [m]
m 0
That means, mq[k] is generated accumulating dq[k]’s
Receiver
dq[k] + mq[k]
Input  Output
+
mq[k-1] Delay Ts
Delta Modulation
Alternate Receiver
mˆ q (t ) ~ (t )
√ Input dq[k] Low Pass m
filter
Integrator-
amplifier
Receiver X dq[k] + mq[k]

Input  Output
+
mq[k-1] Delay Ts
Delta Modulation
Sampler
Alternate Transmitter Comparator frequency fs
E dq[k]
m(t) dq(t)
+  d(t)
mˆ q (t ) - -E
Integrator-
amplifier
Alternate Receiver
dq[k] mˆ q (t ) ~ (t )
m
Input Low Pass
filter
Integrator-
amplifier
Delta Modulation
Sampler
E dq[k]
m(t) dq(t)
+  d(t)
mˆ q (t ) - -E
E: Threshold
of coding
Integrator-
amplifier
• Input m(t) is analog

• d(t) is compared in comparator: d(t) > 0? dq(t) = E : dq(t) = -E
• dq(t) is sampled @ fs to generate dq[k]
• dq[k]’s are integrated and amplified to get mˆ q (t )
Delta Modulation
Sampler
E dq[k]
m(t) dq(t)
+  d(t)
mˆ q (t ) - -E
Integrator-
amplifier
• The sampler generates pulse train dq[k] as follows
positive pulse if m(t )  mˆ q (t )

negative pulse if m(t )  mˆ q (t )
Delta Modulation
Sampler
E dq[k]
m(t) dq(t)
+  d(t)
mˆ q (t ) - -E
Integrator-
amplifier
mˆ q (t )
m(t)
dq[k]
Modulated signal to
be transmitted
Delta Modulation
Sampler
E dq[k]
m(t) dq(t)
+  d(t)
mˆ q (t ) - -E
Integrator-
amplifier
mˆ q (t )
• dq[k] gives rise to a step
function in mˆ q (t ) m(t)
• dq[k] equalizes mˆ q (t )to m(t)
in small step, E
dq[k]
Delta Modulation
Alternate Receiver
dq[k] mˆ q (t ) ~ (t )
m
Input Low Pass
filter
Integrator-
amplifier
mˆ q (t )
• Coarseness of mˆ q (t ) is
removed when passed m(t)
through low pass filter at
receiver
dq[k]
Delta Modulation
Sampler
E dq[k]
m(t) dq(t)
+  d(t)
mˆ q (t ) - -E
Integrator-
amplifier
mˆ q (t )
• DM encodes and transmits
NOT the actual signal but m(t)
the differences (derivatives)
• Integrator assures this dq[k]

Delta Modulation
Sampler
E dq[k]
m(t) dq(t)
+  d(t)
mˆ q (t ) - -E
Integrator-
amplifier
mˆ q (t )
• DM encodes and transmits
NOT the actual signal but m(t)
the differences (derivatives)
• Difference is encoded as E dq[k]

or -E
Delta Modulation
Sampler
E dq[k]
m(t) dq(t)
+  d(t)
mˆ q (t ) - -E
Integrator-
amplifier slope-overload
mˆ q (t )
• If m(t) changes too fast, mˆ q (t )
cannot follow m(t): slope- m(t)
overload occurs
• Reason: E is not high
enough dq[k]
Delta Modulation
Sampler
E dq[k]
m(t) dq(t)
+  d(t)
mˆ q (t ) - -E
Integrator-
amplifier slope-overload
Granular noise
mˆ q (t )
• If E is too high, granular
noise occurs : cannot follow m(t)
small changes of m(t)
dq[k]
Digital-to-Digital Conversion:
Line Coding, Block Coding and
Scrambling
From
Data Communications and Networking, 5th Edition
By Behrouz A. Forouzan
Line Coding and Decoding
Line Coding
• converts digital data elements to digital signal elements
• Example:
1 → +V
0 → -V
Data Element and Signal
Element
• data element
– smallest entity to represent a piece of information
– 0, 1, etc,
• signal element
– shortest unit of a digital signal
– Each represents one or more data element
Element
• data element
– 0, 1, etc,
• signal element
• signal element is carrier, data element is carried

Element
• data element
– 0, 1, etc,
• signal element
• signal element is carrier, data element is carried
• r = No. of data element is carried by each signal element

Element
Data Rate and Signal Rate
• data rate, N
– No. of data elements sent in 1 second
– Other names: bit rate
• signal rate, S
– No. of signal elements sent in 1 second
– Other names: baud rate, pulse rate, modulation rate
• r = N/S alternately, S = N/r

Data Rate and Signal Rate
• r = N/S alternately, S = N/r
• Average Signal rate
S ave  c  N  (1 r ) baud
Line Coding Guidelines
• r should be higher, S = N/r should be lower
• Minimize Baseline wandering:
– Running average of incoming signal power is baseline
– Baseline is compared with incoming signal for decoding
– Long runs of 0 or 1 drifts the baseline

• Minimize DC components:
– Constant voltage for long period
– increaes low frequency components
– bandpass channels do NOT support low frequencies

• Self synchronization
– Sender’s and receiver’s clock must have same bit interval
– Different intervals may misinterpret incoming signal

Illustration of lack of synchronization
• Built-in Error Detection
– Error occurs during transmission
– Must be able to detect and/or correct them
– Example: certain type of signal transition (change) is NOT part of the

coding
• Higher Noise immunity
– certain line coding prevents noises to be added to the signal
– It is better than error detection

• Complexity
– Simpler is better
– A coding with FOUR signal levels is more difficult to interpret than one
that uses TWO signal levels
Line Coding Schemes
Unipolar
• All signal levels are on one side of the time axis - either above
or below
• Example: Non Return to Zero (NRZ)
– The signal level does not return to zero at middle of the bit
Unipolar: NRZ
• Pros
– Simple
• Cons
– Baseline wandering, DC components

– No synchronization or error detection
r=1
Polar
• signal levels are on BOTH sides of the time axis
• Polar NRZ
– Uses two different voltage: +V and –V

– Two versions: NRZ-Level (NRZ-L) and NRZ-Inversion (NRZ-I)
Polar: NRZ-L and NRZ-I
• Polar NRZ-L
+V: 0 and –V: 1
• Polar NRZ-I
Inversion: next bit is 1, No Inversion: next bit 0

Polar: NRZ-L and NRZ-I
• Pros
– Simple
• Cons
– Baseline wandering, DC components, worse in NRZ-L

– No synchronization or error detection
r=1
Polar: Return to Zero (RZ)
• 3 signal levels: +V, -V, 0
• Signal transition in the middle of the bit
– bit 0: -V to 0 bit 1: +V to 0
Polar: Return to Zero (RZ)
• Pros
– NO Baseline wandering, NO DC components
– Self synchronization: transition synchronizes both sender and receiver
• Cons
– More complex: 3 voltage levels to identify

– No error detection
r = ⅟2
Polar: Biphase-Manchester and
Differential Manchester
• 2 signal levels: +V, -V
• Signal transition always in the middle of the bit, similar to RZ
• Manchester
– Combines NRZ-L and RZ
• Differential Manchester
– Combines NRZ-I and RZ
• Manchester
– Level determines bits
• Differential Manchester Manchester

– Inversion/no inversion
determines bits
• Pros
– 2 signal elements
– NO Baseline wandering, NO DC components
– Self synchronization
Manchester
• Cons
– r = 1/2
– No error detection
Bipolar
• 3 signal levels: 0, +V, -V
• 2 variations: alternate mark inversion (AMI) and Pseudoternary
• AMI
– bit 0: signal level 0

– bit 1: successively alternates between +V and -V
• Pseudoternary

Bipolar
• AMI

• Pseudoternary

Bipolar
• Pros
– r=1
• Cons
– 3 signal elements
– NO synchronization
– NO error detection
Multilevel Schemes
• Target: to increase r or bit rate
• Individual bit is NOT coded
• rather a sequence of m bits is coded by a sequence of n signal

elements
• Assuming, No. of signal levels is L
– Possible data patterns: 2m

– Possible signal patterns: Ln
2m data patterns → Ln signal patterns

Multilevel Schemes
2m data patterns → Ln signal patterns
• 2m >Ln : mapping is NOT possible
• 2m =Ln : exact mapping is possible
• 2m <Ln :
– Redundant signal patterns
– Very flexible mapping is possible
– Better noise immunity and error detection
Multilevel Schemes
• Coding symbol: short representation is mBnL
• L is replace by character code
– L = 2 : B meaning Binary
– L = 3 : T meaning Ternary
– L = 4 : Q meaning Quaternary
11 00 01 11 10
2B1Q Scheme +3
Rules +1
00: -3 -1
01: -1
10: +3
-3
11: +1
Multilevel Schemes
• Pros
– r=2
– Used in DSL to provide high speed connection to the Internet
• Cons
– 4 signal elements,
– Baseline wandering, DC components are possible
– NO redundancy, NO error detection
11 00 01 11 10
+3
2B1Q Scheme
Rules +1
00: -3 -1
01: -1
10: +3 -3
11: +1
Multilevel Schemes: 8B6T
• 28 = 256 different data patterns
• 36 = 729 different signal patterns of 3 levels, +V, -V and 0
• No. of redundant signal patterns 729-256 = 473
• Flexible pattern mapping
• Synchronization and error detection by 473 redundant signal

patterns
• All mapped signal patterns have weight +1 or 0 dc values
Mapping Table
Data patterns Signal patterns weight
00010001 –0–0++ 0
01010011 –+–++0 +1
01010000 +––+0+ +1
• NO signal patterns have weight -1
• DC balance is maintained by inverting polarity of signal pattern
if necessary + – – + 0 + is replaced by - + + - 0 - (WEIGHT: -1)

Let, we have to send data sequence
000100010101001101010000
00010001 01010011 01010000
-0-0++ -+-++0 +--+0+
Weight: 0 Weight: +1 Weight was +1

Now it is -1
to ensure DC balance
Multilevel Schemes: 4D-PAM5
• 4 Dimensional: 4 signal elements, each sent through separate wire
• Pulse Amplitude Modulation 5: 5 different levels -2, -1, 0, +1, +2
• Used signal levels: -2, -1, +1, +2
• Total signal patterns: 44 = 256, which support 256 binary data

patterns
• 4D-PAM5 is equivalent to linear 8B4Q

Multilevel Schemes: 4D-PAM5
• 1 Gbps Gigabit LAN through 4 copper wires of 125 Mbaud each

• Self synchronization, NO DC components
Multitransition: MLT-3
• Similar to differential coding like NRZ-I and differential Manchester
• Uses more than 2 signal levels, MLT-3 uses 3 levels: +V, 0, -V
• Complex transition rules
– If next bit = 0, no transition

– If next bit = 1 and current level = ±V, transition to level 0
– If next bit = 1 and current level = 0, transition to opposite of last nonzero level
• Complex transition rules

• r=1
• No synchronization for long sequence bit 0’s
Nonperiodic to periodic signal

with a period of 4× bit duration
Repeating pattern: +V0-V0

Line Coding summary
Polar
Block Coding: mB/nB
• replaces m bits by n bits, where m < n
• increaes redundancy and synchronization

• Done before line coding
using 3 steps
– division
– substitution
– combination
Block Coding: 4B/5B
• Done in combination with NRZ-I which has sync problem for long
sequence of 0’s
• 4B/5B ensures no more than 3 consecutive 0’s

Block Coding: 4B/5B
• 16 data sequences
• 32 available encoding sequences
• redundant sequences are for error
detection, overhead control
• Maximum 1 leading 0 and 2
trailing 0’s
Block Coding: 8B/10B
• More redundancy and better error detection than 4B/5B
• Actually done by combination of 5B/6B and 3B/4B to simplify

mapping table
Search for Best Line Coding:
Scrambling
• Characteristics of a good line coding
– No increase in b/w for self sync
– NO DC
• Biphase scheme
– High bandwidth (r < 1)
• NRZ + block coding

– DC still exists
• Bipolar AMI
– NO sync for long sequence of 0’s
– Other characteristics are OK
Search for best line coding:
Scrambling
• Bipolar AMI
• Scrambling uses AMI but inserts nonzero pulses in long

sequence of 0’s
Search for best line coding:
Scrambling
• Bipolar AMI
• Scrambling uses AMI but inserts nonzero pulses in long

sequence of 0’s
• Thus AMI’s rules are violated on the fly
decoding
Scrambling: B8ZS
• Bipolar 8 Zeros (B8ZS)
– Replaces 8 zeros by 0 0 0 V B 0 V B
– V means Violation to AMI rules
– B means normal polarity according to AMI rules

Scrambling: B8ZS

Scrambling: B8ZS

Scrambling: B8ZS

Scrambling: HDB3
• High Definition Bipolar 3-Zero (HDB3)
• Replaces 4 zeros by
– 0 0 0 V if No. of nonzero pulses after last substitution is ODD
– B 0 0 V if No. of nonzero pulses after last substitution is EVEN
Scrambling: HDB3
Scrambling: HDB3
Scrambling: HDB3
Scrambling: HDB3
Odd
Scrambling: HDB3
Even
Scrambling: HDB3
Even

CSE311 1stPartMerged

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

CSE311 1stPartMerged

Uploaded by

Copyright:

Available Formats

CSE 311:

• Class Tests: 20%

• 2 Class Tests in the first part

in absence of central CT routine

• Most of the PhD’s in the Department from

• See: Maths are working

• However, that was enough for the time

1. Sender and Receiver

Most important Distortion

Input digital Distorted

Input digital Distorted Regenerated

Input Distorted Signal with

Input Distorted Signal with

Noise accumulates along the path! No improvement at all !!

Recall this figure

• Assume, No. of quantization level, L = 2

• Assume, No. of quantization level, L = 2

• If No. of quantization level, L >> 2

• If No. of quantization level, L >> 2 Each sample

• If No. of quantization level, L >> 2

• 2 important properties Distortion

A complex signal with

A complex signal with

– High frequency signals

– Low frequency signals

Duration = 0.1 sec, Frequency = 2.5/0.1= 25 Hz

Duration = 0.1 sec, Frequency = 2.5/0.1= 25 Hz

Duration = 0.05 sec, Frequency =2.5/0.05=50 Hz

increase transmission speed K times

increase transmission speed K times

increase transmission speed K times

Signal Power, Ps ∞ Amplitude, A

SNR = Signal / Noise

– Shannon’s limit on channel capacity

C  B log 2 (1  SNR ) bit s

– Shannon’s limit on channel capacity

C  B log 2 (1  SNR ) bit s

What happens if SNR = ∞?

• Speech signal characteristics

• 10 MHz carrier signal characteristics

recall this example!

• Digital source coding

• Error correction coding

Less More Higher Shorter

Less More Higher Shorter

Higher Less Lower Longer

Less More Higher Shorter

Higher Less Lower Longer

• A message with probability p,

• A message with probability p,

What happens when p = 0 and p =1?

Assume to transmit 0001

Assume to transmit 00011

Add this bit to make

Assume to transmit 00011

Add this bit to make

Detects single bit error; however

• Definition, classification and properties of signals

• Signal g(t) is periodic for a positive constant T0 so

• Smallest T0 is its period

• A signal is aperiodic if NOT periodic

At t = -T0, g(t) ≠ g(t + T0)

• Energy Signal: finite Eg