1

Digital Signal Processing

2
Books.
Text Books:
Digital Signal Processing
Principles, Algorithms and Applications
By:
John G.Proakis & Dimitris G.Manolakis
Reference Books
1. Digital Signal Processing By.
Sen M. Kuo & Woon-Seng Gan
2. Digital Signal Processing A Practical Approach. By
Emmanuel C. Ifeachor & Barrie W. Jervis
[Handouts]
Digital Signal Processing By. Bores [Avail at ww.bores.com/courses/
3
Grading Policy
Term Papers/test/Group Discussion 20 Marks
Mid-Term 30 Marks
Final 50 Marks
Additional Privileges 10%
Trem Paper. Home works, Presentations, Voluntary
assignments managements etc.
Class will be divided different level as per their GPA
Group A- GPA 2.0 to 2.59
Group B- GPA 2.5 to 3.39
Group C GPA 3.4 to 4
Analog and Digital Signals
Analog signal = continuous-time + continuous amplitude
Digital signal = discrete-time + discrete amplitude
analog system = analog signal input + analog signal output
advantages: easy to interface to real world, do not need A/D or D/A converters,
speed not dependent on clock rate
digital system = digital signal input + digital signal output I re-configurability
using software, greater control over accuracy/resolution, predictable and reproducible
behavior
Signal Processing
A.S
.p
D.S
.p
M.
S.p
Signal : f(x1: x2.. ) is function, A function is a dependent variable of independent
variable(s).
X= Time, Distance, Temperature,.
Type of signal Natural Signal [1D,2D,MD]
Continuous? Discrete Signal
Analog Signal = 1-Cont-time--- 2-Discrete Time ---- A/D -----Digital Signal
5
Analog-to-Digital Conversion
Quantize r Sampler Coder
x(t)
0101...
X(n)
Discrete-
time
signal
x
q
(t)
Quanti
zed
Signal
Digital Signal
Sampling:
conversion from cts-time to dst-time by taking \samples" at discrete time instants E.g.,
uniform sampling: x(n) = xa(nT) where T is the sampling period and n Z
Quantization: conversion from dst-time cts-valued signal to a dst-
time dst-valued signal quantization error: e
q
(n) = x
q
(n)- x(n))
Coding: representation of each dst-value xq(n) by a b-bit binary sequence
6
Sampling Theorem
If the highest frequency contained in an analog signal x
a
(t) is F
max
= B and the signal is
sampled at a rate
Fs > 2Fmax=2B
then x
a
(t) can be exactly recovered from its sample values using the interpolation
function
Therefore, given the interpolation relation, x
a
(t) can be written as
where x
a
(nT) = x(n); called band limited interpolation.
Note: F
N
= 2B = 2Fmax
is called the Nyquist rate
7
Digital-to-Analog Conversion
Common interpolation approaches: bandlimited interpolation zero-order hold, linear interpolation,
higher-order interpolation techniques, e.g., using splines
In practice, \cheap" interpolation along with a smoothing filter is employed.
A DSP System ????
8
A DSP System
In practice, a DSP system does not use idealized A/D or D/A
models.
Anti-aliasing Filter: ensures that analog input signal does not
contain frequency components higher than half of the
sampling frequency (to obey the sampling theorem). this
process is irreversible
2Sample and Hold:
holds a sampled analog value for a short time while the A/D
converts and interprets the value as a digital
3 A/D: converts a sampled data signal value into a digital
number, in part, through quantization of the amplitude
4 D/A: converts a digital signal into a \staircase"-like signal
5 Reconstruction Filter: converts a \staircase"-like signal
into an analog signal through low pass filtering similar to the
type used for anti-aliasing
Real-time DSP Considerations ???????
9
Real-time DSP Considerations
What are initial considerations when designing a
DSP system that must run in real-time?
Is a DSP technology suitable for a real-time application?
10
Lecture 1
Week-1st
11
Signal:
A signal is defined as a function of one or more variables
which conveys information on the nature of a physical
phenomenon. The value of the function can be a real
valued scalar quantity, a complex valued quantity, or
perhaps a vector.
System:
A system is defined as an entity that manipulates one or
more signals to accomplish a function, thereby yielding
new signals.
12
Continuos-Time Signal:
A signal x(t) is said to be a continuous time signal if it is
defined for all time t.
Discrete-Time Signal:
A discrete time signal x[nT] has values specified only at
discrete points in time.
Signal Processing:
A system characterized by the type of operation that it
performs on the signal. For example, if the operation is
linear, the system is called linear. If the operation is non-
linear, the system is said to be non-linear, and so forth.
Such operations are usually referred to as Signal
Processing.
13
Basic Elements of a Signal Processing
System
Analog
Signal Processor
Analog input
signal
Analog output
signal
Analog Signal Processing
Digital
Signal Processor
A/D
converter
D/A
converter
Digital Signal Processing
Analog
input
signal
Analog
output
signal
14
Advantages of Digital over Analogue Signal
Processing:
A digital programmable system allows flexibility in
reconfiguring the DSP operations simply by changing the
program. Reconfiguration of an analogue system usually
implies a redesign of hardware, testing and verification
that it operates properly.
DSP provides better control of accuracy requirements.
Digital signals are easily stored on magnetic media (tape
or disk).
The DSP allows for the implementation of more
sophisticated signal processing algorithms.
In some cases a digital implementation of the signal
processing system is cheaper than its analogue
counterpart.
15
DSP Applications
Space
Space photograph enhancement
Data compression
Intelligent sensory analysis
Medical
Diagnostic imaging (MRI, CT,
ultrasound, etc.)
Electrocardiogramanalysis
Medical image storage and retrieval
Image and sound compression for
multimedia presentation.
Movie special effects
Video conference calling
Commercial
Video and data compression
echo reduction
signal multiplexing
filtering
Telephone
16
DSP Applications (cont.)
Military
Radar
Sonar
Ordnance Guidance
Secure communication
Industrial
Oil and mineral prospecting
Process monitoring and control
Non-destructive testing
Scientific
Earth quick recording and analysis
Data acquisition
Spectral Analysis
Simulation and Modeling
17
Classification of Signals
Deterministic Signals
A deterministic signal behaves in a fixed known way with
respect to time. Thus, it can be modeled by a known
function of time t for continuous time signals, or a known
function of a sampler number n, and sampling spacing T
for discrete time signals.
Randomor Stochastic Signals:
In many practical situations, there are signals that either
cannot be described to any reasonable degree of accuracy
by explicit mathematical formulas, or such a description is
too complicated to be of any practical use. The lack of
such a relationship implies that such signals evolve in time
in an unpredictable manner. We refer to these signals as
random.
18
Even and Odd Signals
A continuous time signal x(t) is said to an even signal if it
satisfies the condition
x(-t) = x(t) for all t
The signal x(t) is said to be an odd signal if it satisfies the
condition
x(-t) = -x(t)
In other words, even signals are symmetric about the
vertical axis or time origin, whereas odd signals are
antisymmetric about the time origin. Similar remarks
apply to discrete-time signals.
Example:
even
odd odd
19
Periodic Signals
A continuous signal x(t) is periodic if and only if there
exists a T > 0 such that
x(t + T) = x(t)
where T is the period of the signal in units of time.
f = 1/T is the frequency of the signal in Hz. W= 2t/T is the
angular frequency in radians per second.
The discrete time signal x[nT] is periodic if and only if
there exists an N > 0 such that
x[nT + N] = x[nT]
where N is the period of the signal in number of sample
spacings.
Example:
0
0.2 0.4
Frequency = 5 Hz or 10t rad/s
20
Continuous Time Sinusoidal Signals
A simple harmonic oscillation is mathematically
described as
x(t) = Acos(wt + u)
This signal is completely characterized by three
parameters:
A = amplitude, w = 2tf = frequency in rad/s, and u =
phase in radians.
A T=1/f
21
Discrete Time Sinusoidal Signals
Adiscrete time sinusoidal signal may be expressed as
x[n] = Acos(wn + u) - < n <
Properties:
A discrete time sinusoid is periodic only if its frequency is a
rational number.
Discrete time sinusoids whose frequencies are separated by
an integer multiple of 2t are identical.
0 2 4 6 8 10
-1
0
1
22
Energy and Power Signals
The total energy of a continuous time signal x(t) is
defined as
( )dt t x dt ) t ( x lim E
2
T
T
2
T
x } }



= =
And its average power is
}

=
2 / T
2 / T
2
T
x
dt ) t ( x
T
1
lim P
In the case of a discrete time signal x[nT], the total energy of the
signal is
| |

=
=
n
2
dx
n x T E
And its average power is defined by
| |
2
N
N n
N
dx
nT x
1 N 2
1
lim P

=

|
.
|

\
|
+
=
23
Energy and Power Signals
A signal is referred to as an energy signal, if and only if
the total energy of the signal satisfies the condition
0 < E <
On the other hand, it is referred to as a power signal, if
and only if the average power of the signal satisfies the
condition
0 < P <
An energy signal has zero average power, whereas a power
signal has infinite energy.
Periodic signals and random signals are usually viewed as
power signals, whereas signals that are both deterministic and
non-periodic are energy signals.
des
24
Example1:
Compute the signal energy and signal power for
x[nT] = (-0.5)
n
u(nT), T = 0.01 seconds
Solution:
( ) ( )
2
0 n
n
2
N
N n
N
dx
5 . 0 01 . 0 nT x T lim E


= =

= =
( )


=

=
= =
0 n
n
n 2
0 n
25 . 0 01 . 0 5 . 0 01 . 0
( ) ( ) ....... 25 . 0 25 . 0 25 . 0 1 01 . 0
3 2
+ + + + =
75 / 1
25 . 0 1
01 . 0
=

=
Since E
dx
is finite, the signal power is zero.
25
Example2:
Repeat Example1 for y[nT] = 2e
j3n
u[nT], T = 0.2 second.
Solution:

=

=

|
.
|

\
|
+
= |
.
|

\
|
+
=
N
0 n
2
n 3 j
N
2
N
N n
N
dx
e 2
1 N 2
1
lim ) nT ( y
1 N 2
1
lim P
1 N 2
) 1 N ( 4
lim 1
1 N 2
4
lim 2
1 N 2
1
lim
N
0 n
N N
N
0 n
2
N
+
+
=
+
= |
.
|

\
|
+
=

=

=

2
2
1
4
1 N 2
1
1 N 2
N
4 lim
N
= = |
.
|

\
|
+
+
+
=

What is energy of this signal?
26
Tutorial 1: Q3
Determine the signal energy and signal power for each
of the given signals and indicate whether it is an energy
signal or a power signal?
], 3 n [ u ) 2 . 0 ( 3 ] nT [ y
n
= (a) T = 2 ms
(b)
( ) ] 1 n [ u 1 . 1 4 ] nT [ z
n
+ =
T = 0.02 s
(c)
27
Time Shifting, Time Reversal,Time Scaling
Suppose we have a signal x(t) and we say we want to
shift a signal such as x(t-2) or x(t+2) so - values
indicate the past values while the + values indicate
the future value
Time reversal is the mirror image of the given signal
as x(t) = x(-t)
Time Scaling is the scaled time according to input for
e.g x(2t) will be a compact signal as compared to x(t).
28
Basic Operations on Signals
(a) Operations performed on dependent
variables
1. Amplitude Scaling:
let x(t) denote a continuous time signal. The signal y(t)
resulting from amplitude scaling applied to x(t) is
defined by
y(t) = cx(t)
where c is the scale factor.
In a similar manner to the above equation, for discrete
time signals we write
y[nT] = cx[nT]
x(t)
2x(t)
29
(b) Operations performed on independent
variable
Time Scaling:
Let y(t) is a compressed version of x(t). The signal y(t)
obtained by scaling the independent variable, time t, by
a factor k is defined by
y(t) = x(kt)
if k > 1, the signal y(t) is a compressed version of
x(t).
If, on the other hand, 0 < k < 1, the signal y(t) is an
expanded (stretched) version of x(t).
30
Example of time scaling
0 5 10 15
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
exp(-2t)
exp(-t)
exp(-0.5t)
Expansion and compression of the signal e
-t
.
31
-3 -2 -1 0 1 2 3
0
5
10
x
[
n
]
-1.5 -1 -0.5 0 0.5 1 1.5
0
5
10
x
[
0
.
5
n
]
-6 -4 -2 0 2 4 6
0
5
x
[
2
n
]
n
Time scaling of discrete time systems
32
Time Reversal
This operation reflects the signal about t = 0
and thus reverses the signal on the time scale.
0 1 2 3 4 5
0
5
x
[
n
]
n
0 1 2 3 4 5
-5
0
x
[
-
n
]
n
33
Time Shift
A signal may be shifted in time by replacing the
independent variable n by n-k, where k is an
integer. If k is a positive integer, the time shift
results in a delay of the signal by k units of time. If
k is a negative integer, the time shift results in an
advance of the signal by |k| units in time.
-2 0 2 4 6 8 10
0
0.5
1
x
[
n
]
-2 0 2 4 6 8 10
0
0.5
1
x
[
n
+
3
]
-2 0 2 4 6 8 10
0
0.5
1
x
[
n
-
3
]
n
34
2. Addition:
Let x
1
[n] and x
2
[n] denote a pair of discrete time signals.
The signal y[n] obtained by the addition of x
1
[n] + x
2
[n]
is defined as
y[n] = x
1
[n] + x
2
[n]
Example: audio mixer
3. Multiplication:
Let x
1
[n] and x
2
[n] denote a pair of discrete-time signals.
The signal y[n] resulting from the multiplication of the
x
1
[n] and x
2
[n] is defined by
y[n] = x
1
[n].x
2
[n]
Example: AMRadio Signal
35
Analog to Digital and Digital to Analog
Conversion
A/D conversion can be viewed as a three
step process
1. Sampling: This is the conversion of a continuous time
signal into a discrete time signal obtained by taking
samples of the continuous time signal at discrete time
instants. Thus, if x(t) is the input to the sampler, the
output is x(nT), where T is called the Sampling interval.
2. Quantization: This is the conversion of discrete time
continuous valued signal into a discrete-time discrete-
value (digital) signal. The value of each signal sample is
represented by a value selected from a finite set of
possible values. The difference between unquantized
sample and the quantized output is called the
Quantization error.
36
Analog to Digital and Digital to Analog
Conversion (cont.)
3. Coding: In the coding process, each discrete value is
represented by a b-bit binary sequence.
Quantize r Sampler Coder
x(t)
0101...
A/D Converter
37
Digital Signal Processing
(DSP)
Fundamentals
38
Overview
What is DSP?
Converting Analog into Digital
Electronically
Computationally
How Does It Work?
Faithful Duplication
Resolution Trade-offs
39
What is DSP?
Converting a continuously changing waveform
(analog) into a series of discrete levels (digital)
40
What is DSP?
The analog waveform is sliced into equal
segments and the waveform amplitude is
measured in the middle of each segment
The collection of measurements make up
the digital representation of the waveform
41
What is DSP?
0
0
.
2
20
.
4
40
.
6
4
0
.
8
2
0
.
9
8
1
.
1
1
1
.
2
1
.
2
4
1
.
2
7
1
.
2
4
1
.
2
1
.
1
1
0
.
9
8
0
.
8
2
0
.
6
4
0
.
4
4
0
.
2
2
0
-
0
.
2
2
-
0
.
4
4
-
0
.
6
4
-
0
.
8
2
-
0
.
9
8
-
1
.
1
1
-
1
.
2
-
1
.
2
6
-
1
.
2
8
-
1
.
2
6
-
1
.
2
-
1
.
1
1
-
0
.
9
8
-
0
.
8
2
-
0
.
6
4
-
0
.
4
4
-
0
.
2
2
0
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
13579
1
1
1
3
1
5
1
7
1
9
2
1
2
3
2
5
2
7
2
9
3
1
3
3
3
5
3
7
42
Converting Analog into Digital
Electronically(1/3)
The device that does the conversion is
called an Analog to Digital Converter
(ADC)
There is a device that converts digital to
analog that is called a Digital to Analog
Converter (DAC)
43
Converting Analog into Digital
Electronically(2/3)
The simplest form of
ADC uses a resistance
ladder to switch in the
appropriate number of
resistors in series to
create the desired
voltage that is
compared to the input
(unknown) voltage
V-7
V-6
V-low
V-1
V-2
V-3
V-4
V-5
V-high
SW-8
SW-7
SW-6
SW-5
SW-4
SW-3
SW-2
SW-1
Output
44
Converting Analog into Digital
Electronically(3/3)
The output of the
resistance ladder is
compared to the
analog voltage in a
comparator
When there is a match,
the digital equivalent
(switch configuration)
is captured
Analog Voltage
Resistance
Ladder Voltage
Comparator
Output Higher
Equal
Lower
45
Converting Analog into Digital
Computationally(1/2)
The analog voltage can now be compared with the
digitally generated voltage in the comparator
Through a technique called binary search, the
digitally generated voltage is adjusted in steps
until it is equal (within tolerances) to the analog
voltage
When the two are equal, the digital value of the
voltage is the outcome
46
Converting Analog into Digital
Computationally(2/2)
The binary search is a mathematical technique that
uses an initial guess, the expected high, and the
expected low in a simple computation to refine a
new guess
The computation continues until the refined guess
matches the actual value (or until the maximum
number of calculations is reached)
The following sequence takes you through a
binary search computation
47
Binary Search
Initial conditions
Expected high 5-volts
Expected low 0-volts
5-volts 256-binary
0-volts 0-binary
Voltage to be converted
3.42-volts
Equates to 175 binary
Analog Digital
5-volts
256
0-volts
0
2.5-volts 128
3.42-volts
Unknown
(175)
48
Binary Search
Binary search algorithm:
First Guess:
Analog Digital
5-volts
256
0-volts
0
128
3.42-volts
unknown
NewGuess Low
Low High
= +

2
128 0
2
0 256
= +

Guess is Low
49
Binary Search
New Guess (2):
Analog Digital
5-volts
256
0-volts
0
192
3.42-volts
unknown
192 128
2
128 256
= +

Guess is High
50
Binary Search
New Guess (3):
Analog Digital
5-volts
256
0-volts
0
160
3.42-volts
unknown
160 128
2
128 192
= +

Guess is Low
51
Binary Search
New Guess (4):
Analog Digital
5-volts
256
0-volts
0
176
3.42-volts
unknown
176 160
2
160 192
= +

Guess is High
52
Binary Search
New Guess (5):
Analog Digital
5-volts
256
0-volts
0
168
3.42-volts
unknown
168 160
2
160 176
= +

Guess is Low
53
Binary Search
New Guess (6):
Analog Digital
5-volts
256
0-volts
0
172
3.42-volts
unknown
172 168
2
168 176
= +

Guess is Low
(but getting close)
54
Binary Search
New Guess (7):
Analog Digital
5-volts
256
0-volts
0
174
3.42-volts
unknown
174 172
2
172 176
= +

Guess is Low
(but getting really,
really, close)
55
Binary Search
New Guess (8):
Analog Digital
5-volts
256
0-volts
0
3.42-volts
175!
175 174
2
174 176
= +

Guess is Right On
56
Binary Search
The speed the binary search is
accomplished depends on:
The clock speed of the ADC
The number of bits resolution
Can be shortened by a good guess (but usually
is not worth the effort)
57
How Does It Work?
Faithful Duplication
Now that we can slice up a waveform and
convert it into digital form, lets take a look
at how it is used in DSP
Draw a simple waveform on graph paper
Scale appropriately
Gather digital data points to represent the
waveform
58
Starting Waveform Used to
Create Digital Data
59
How Does It Work?
Faithful Duplication
Swap your waveform data with a partner
Using the data, recreate the waveform on a
sheet of graph paper
60
Waveform Created from Digital Data
61
How Does It Work?
Faithful Duplication
Compare the original with the recreating,
note similarities and differences
62
How Does It Work?
Faithful Duplication
Once the waveform is in digital form, the
real power of DSP can be realized by
mathematical manipulation of the data
Using EXCEL spreadsheet software can
assist in manipulating the data and making
graphs quickly
Lets first do a little filtering of noise
63
How Does It Work?
Faithful Duplication
Using your raw digital data, create a new
table of data that averages three data points
Average the point before and the point after
with the point in the middle
Enter all data in EXCEL to help with graphing
64
Noise Filtering Using Averaging
Raw
-150
-100
-50
0
50
100
150
0 10 20 30 40
Time
A
m
p
l
i
t
u
d
e
Ave before/after
-150
-100
-50
0
50
100
150
0 10 20 30 40
Time
A
m
p
l
i
t
u
d
e
65
How Does It Work?
Faithful Duplication
Lets take care of some static crashes that
cause some interference
Using your raw digital data, create a new
table of data that replaces extreme high and
low values:
Replace values greater than 100 with 100
Replace values less than -100 with -100
66
Clipping of Static Crashes
Raw
-150
-100
-50
0
50
100
150
0 10 20 30 40
Time
A
m
p
l
i
t
u
d
e
eliminate extremes (100/-100)
-150
-100
-50
0
50
100
150
0 10 20 30 40
Time
A
m
p
l
i
t
u
d
e
67
How Does It Work?
Resolution Trade-offs
Now lets take a look at how sampling rates
affect the faithful duplication of the
waveform
Using your raw digital data, create a new
table of data and delete every other data
point
This is the same as sampling at half the rate
68
Half Sample Rate
Raw
-150
-100
-50
0
50
100
150
0 10 20 30 40
Time
A
m
p
l
i
t
u
d
e
every 2nd
-150
-100
-50
0
50
100
150
0 10 20 30 40
Time
A
m
p
l
i
t
u
d
e
69
How Does It Work?
Resolution Trade-offs
Using your raw digital data, create a new
table of data and delete every second and
third data point
This is the same as sampling at one-third
the rate
70
1/2 Sample Rate
Raw
-150
-100
-50
0
50
100
150
0 10 20 30 40
Time
A
m
p
l
i
t
u
d
e
every 3rd
-150
-100
-50
0
50
100
150
0 10 20 30 40
Time
A
m
p
l
i
t
u
d
e
71
How Does It Work?
Resolution Trade-offs
Using your raw digital data, create a new
table of data and delete all but every sixth
data point
This is the same as sampling at one-sixth
the rate
72
1/6 Sample Rate
Raw
-150
-100
-50
0
50
100
150
0 10 20 30 40
Time
A
m
p
l
i
t
u
d
e
every 6th
-150
-100
-50
0
50
100
150
0 10 20 30 40
Time
A
m
p
l
i
t
u
d
e
73
How Does It Work?
Resolution Trade-offs
Using your raw digital data, create a new
table of data and delete all but every twelfth
data point
This is the same as sampling at one-twelfth
the rate
74
1/12 Sample Rate
Raw
-150
-100
-50
0
50
100
150
0 10 20 30 40
Time
A
m
p
l
i
t
u
d
e
every 12th
-150
-100
-50
0
50
100
150
0 10 20 30 40
Time
A
m
p
l
i
t
u
d
e
75
How Does It Work?
Resolution Trade-offs
What conclusions can you draw from the
changes in sampling rate?
At what point does the waveform get too
corrupted by the reduced number of
samples?
Is there a point where more samples does
not appear to improve the quality of the
duplication?
76
How Does It Work?
Resolution Trade-offs
Bit
Resolution
High Bit
Count
Good
Duplication
Slow
Low Bit
Count
Poor
Duplication
Fast
Sample Rate High Sample
Rate
Good
Duplication
Slow
Low Sample
Rate
Poor
Duplication
Fast
77
Digital Signal Processing
Lecture -2
78
Sampling of Analog Signals
0 2 4 6
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
t
a
n
a
l
o
g

s
i
g
n
a
l
0 2 4 6
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
s
a
m
p
l
e
d

s
i
g
n
a
l
n
Uniform Sampling:
x[n] = x[nT]
79
Uniform sampling
Uniform sampling is the most widely used sampling scheme.
This is described by the relation
x[n] = x[nT] - < n <
where x(n) is the discrete time signal obtained by taking samples
of the analogue signal x(t) every T seconds.
The time interval T between successive symbols is called the
Sampling Period or Sampling interval and its reciprocal 1/T = F
s
is
called the Sampling Rate (samples per second) or the Sampling
Frequency (Hertz).
A relationship between the time variables t and n of continuous time
and discrete time signals respectively, can be obtained as
s
F
n
nT t = =
(1)
80
A relationship between the analog frequency F and the
discrete frequency f may be established as follows.
Consider an analog sinusoidal signal
x(t) = Acos(2tFt + u)
which, when sampled periodically at a rate F
s
= 1/T samples
per second, yields
( )
|
|
.
|

\
|
O +
t
= O + t =
s
F
nF 2
cos A FnT 2 cos A ] nT [ x
(2)
But a discrete sinusoid is generally represented as
( ) O + t = fn 2 cos A ] n [ x
(3)
Comparing (2) and (3) we get
s
F
F
f =
(4)
81
Since the highest frequency in a discrete time signal is f = .
Therefore, from (4) we have
T 2
1
2
F
F
s
max
= =
(5)
or
F
s
= 2 F
max
(6)
Sampling Theorem:
If x(t) is bandlimited with no components of frequencies greater
than F
max
Hz, then it is completely specified by samples taken at
the uniform rate F
s
> 2F
max
Hz.
The minimum sampling rate or minimum sampling frequency,
F
s
= 2Fmax, is referred to as the Nyquist Rate or Nyquist
Frequency. The corresponding time interval is called the Nyquist
Interval.
82
Sampling Theorem (cont.)
Signal sampling at a rate less than the Nyquist rate is
referred to as undersampling.
Signal sampling at a rate greater than the Nyquist rate is
known as the oversampling.
Example 1:
The following analogue signals are sampled at a sampling frequency of 40
Hz. Find the corresponding discrete time Signals.
(i) x(t) = cos2t(10)t (ii) y(t) = cos2t(50)t
Solution:
(i)
n n n x
|
.
|

\
|
=
|
.
|

\
|
=
2
cos
40
10
2 cos ] [
1
t
t
(ii)
n n n n n n x
2
cos ) 2 / 2 cos(
2
5
cos
40
50
2 cos ] [
2
t
t t
t
t = + = =
|
.
|

\
|
=
Note: The frequency F
2
= 50 Hz is an alias of F
1
= 10 Hz. All of the
sinusoids cos2t(F
1
+ 40k)t, t = 1,2,3, are aliases.
As, Shows identical in [ x
1
(n) & x
2
(n)] sinusoidal signals & indistinguishable. Ambiguity
is there for samples values. x(t) yield same values as y(t) when two are sampled at Fs=40,
then
83
Example 2
Consider the analog signal
x(t) = 3cos100tt
(a) Determine the minimum required sampling rate to
avoid aliasing.
(b) Suppose that the signal is sampled at the rate F
s
=
200 Hz. What is the discrete time signal obtained
after sampling?
Solution:
(a) The frequency of the analog signal is F = 50 Hz.
Hence the minimum sampling rate to avoid aliasing is
100Hz.
(b)
n
2
cos 3 n
200
100
cos 3 ] n [ x
t
=
t
=
84
Example 3
Consider the analog signal
x(t) = 3cos50tt + 10sin300tt - cos100tt
What is the Nyquist rate for this signal.
Solution:
The frequencies present in the signal above are
F
1
= 25 Hz, F
2
= 150 Hz F
3
= 50 Hz.
Thus F
max
= 150 Hz.
Nyquist rate = 2.F
max
= 300 Hz.
Note: It should be observed that the signal component
10sin300tt, sampled at 300 Hz results in the samples
10sintn, which are identically zero, hence we miss the signal
component completely.
What should we do to avoid this situation????
85
Tutorial
Q1: Find the minimum sampling rate that can be used to obtain samples that
completely specify the signals:
(a) x(t) = 10cos(20tt) 5cos(100tt) + 20cos(400tt)
(b) y(t) = 2cos(20tt) + 4sin(20tt - t/4) + 5cos(8tt)
Q2: Consider the analog signal
x(t) = 3cos2000tt + 5sin6000tt + 10cos12000tt
(a) What is the Nyquist rate for this signal?
(b) Assume now that we sample this signal using a sampling rate F
s
=
5000 samples/s. What is the discrete time signal obtained after sampling?
86
Some Elementary Discrete Time signals
Unit Impulse or unit sample sequence:
It is defined as
In words, the unit sample sequence is a signal that is zero
everywhere, except at t = 0.
| |

=
=
= o
0 n 0
0 n , 1
n
Unit impulse function
- 3 - 2 - 1 0 1 2 3
0
0 . 2
0 . 4
0 . 6
0 . 8
1
87
Some Elementary Discrete Time signals
Unit step signal
It is defined as

<
>
=
0 n 0
0 n , 1
] n [ u
0 1 2 3 4 5 6 7
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
88
Some Elementary Discrete Time signals
Unit Ramp signal
It is defined as

<
>
=
0 n 0
0 n , n
] n [ r
0 1 2 3 4 5 6
0
1
2
3
4
5
6
89
Some Elementary Discrete Time signals
Exponential Signal
The exponential signal is a sequence of the form
x[n] = a
n
, for all n
If the parameter a is real, then x[n] is a real signal.
The following figure illustrates x[n] for various
values of a.
a>1
-1<a<0
a<-1
0<a<1
90
Some Elementary Discrete Time signals
Exponential Signal (cont)
when the parameter a is complex valued, it can be expressed
as
where r and u are now the parameters. Hence we may
express x[n] as
Since x[n] is now complex valued, it can be represented
graphically by plotting the real part
as a function of n, and separately plotting the imaginary part
as a function of n. (see plots on the next slide)
u
=
j
re a
( ) n sin j n cos r e r ] n [ x
n j n
u + u = =
u
n cos r ] n [ x
n
R
u =
n sin r ] n [ x
n
I
u =
91
0 10 20 30 40 50 60
-0.5
0
0.5
1
0 10 20 30 40 50 60
-0.5
0
0.5
1
x
I
[n] = (0.9)
n
sin(tn/10)
x
R
[n] = (0.9)
n
cos(tn/10)
92
Exponential Signal (cont.)
Alternatively, the signal x[n] may be graphically represented by the
amplitude or magnitude function
|x[n]| = r
n
and the phase function
u[n] = un
The following figure illustrates |x[n| and u[n] for r = 0.9 and u = t/10.
-
0 5 10
0
2
|
x
[
n
]
|
- 0 5 10
-
0

n
93
Discrete Time Systems
A discrete time system is a device or algorithm that operates
on a discrete time signal x[n], called the input or excitation,
according to some well defined rule, to produce another
discrete time signal y[n] called the output or response of the
system.
We express the general relationship between x[n] and y[n] as
y[n] = H{x[n]}
where the symbol H denotes the transformation (also called
an operator), or processing performed by the system on x[n]
to produce y[n].
Discrete Time System
H
x[n]
y[n]
94
Example 4
Determine the response of the following
systems to the input signal:
(a) y[n] = x[n]
(b) y[n] = x[n-1]
(c) y[n] = x[n+1]
(d) y[n] = (1/3)[x[n+1] + x[n] + x[n-1]]
(e) y[n] = max[x[n+1],x[n],x[n-1]]

s s
=
otherwise , 0
3 n 3 |, n |
] n [ x

=
=
n
k
] k [ x ] n [ y
(f)
95
Solution:
(a) In this case the output is exactly the same as the input
signal. Such a systemis known as the identity System.
(b) y[n] = [,3, 2, 1, 0, 1, 2, 3,]
(c) y[n] = [.,3, 2, 1, 0, 1, 2, 3,.]
(d) y[n] = [., 5/3, 2, 1, 2/3, 1, 2, 5/3, 1, 0,]
(e) y[n] = [0, 3, 3, 3, 2, 1, 2, 3, 3, 3, 0, .]
(f) y[n] = [,0, 3, 5, 6, 6, 7, 9, 12, 0, ]
96
Classification of Discrete Time Systems
Static versus Dynamic Systems
A discrete time system is called static or memory-less if its
output at any instant n depends at most on the input sample
at the same time, but not on the past or future samples of the
input. In any other case, the system is said to be dynamic or
to have memory.
Examples: y[n] = x
2
[n] is a memory-less system, whereas the
following are the dynamic systems:
(a) y[n] = x[n] + x[n-1] + x[n-2]
(b) y[n] = 2x[n] + 3x[n-4]
97
Time Invariant versus Time Variant Systems
A system is said to be time invariant if a time delay or time
advance of the input signal leads to an identical time shift in
the output signal. This implies that a time-invariant system
responds identically no matter when the input is applied.
Stated in another way, the characteristics of a time invariant
system do not change with time. Otherwise the system is said
to be time variant.
Example1: Determine if the system shown in the figure is
time invariant or time variant.
Z
-1
x[n]
y[n]
-
+
Solution: y[n] = x[n] x[n-1]
Now if the input is delayed by k units
in time and applied to the system, the
Output is
y[n,k] = n[n-k] x[n-k-1] (1)
On the other hand, if we delay y[n] by k units in time, we obtain
y[n-k] = x[n-k] x[n-k-1] (2)
(1) and (2) show that the system is time invariant.
98
Time Invariant versus Time Variant Systems
Example 2: Determine if the following systems are time invariant or
time variant.
(a) y[n] = nx[n] (b) y[n] = x[n]cosw
0
n
Solution:
(a) The response to this systemto x[n-k] is
y[n,k] = nx[n-k] (3)
Now if we delay y[n] by k units in time, we obtain
y[n-k] = (n-k)x[n-k]
= nx[n-k] kx[n-k] (4)
which is different from(3). This means the systemis time-variant.
(b) The response of this systemto x[n-k] is
y[n,k] = x[n-k]cosw
0
n (5)
If we delay the output y[n] by k units in time, then
y[n-k] = x[n-k]cosw
0
[n-k]
which is different from that given in (5), hence the system is time
variant.
99
Linear versus Non-linear Systems
A system H is linear if and only if
H[a
1
x
1
[n] + a
2
x
2
[n]] = a
1
H[x
1
[n]] + a
2
H[x
2
[n]]
for any arbitrary input sequences x
1
[n] and x
2
[n], and any
arbitrary constants a
1
and a
2
.
+
H
H
H
+
x
1
[n]
x
2
[n]
a
1
a
2
x
1
[n]
x
2
[n]
a
1
a
2
y
1
[n]
y
2
[n]
If y
1
[n] = y
2
[n], then H is linear.
100
Examples
Determine if the following systems are linear or nonlinear.
(a) y[n] = nx[n]
Solution:
For two input sequences x
1
[n] and x
2
[n], the corresponding
outputs are
y
1
[n] = nx
1
[n] and y
2
[n] = nx
2
[n]
A linear combination of the two input sequences results in the
output
H[a
1
x
1
[n] + a
2
x
2
[n]] = n[a
1
x
1
[n] + a
2
x
2
[n]] = na
1
x
1
[n] + na
2
x
2
[n] (1)
On the other hand, a linear combination of the two outputs results in
the out
a
1
y1[n] + a
2
y
2
[n] = a
1
nx
1
[n] + a
2
nx
2
[n] (2)
Since the right hand sides of (1) and (2) are identical, the system is
linear.
101
(b) y[n] = Ax[n] + B
Solution:
Assuming that the system is excited by x
1
[n] and x
2
[n]
separately, we obtain the corresponding outputs
y
1
[n] = Ax
1
[n] + B and y
2
= Ax
2
[n] + B
A linear combination of x
1
[n] and x
2
[n] produces the
output
y
3
[n] = H[a
1
x
1
[n] + a
2
x
2
[n]] = A[a
1
x
1
[n] + a
2
x
2
[n]] + B
= Aa
1
x
1
[n] + Aa
2
x
2
[n] + B (3)
On the other hand, if the system were linear, its output to
the linear combination of x
1
[n] and x
2
[n] would be a linear
combination of y
1
[n] and y
2
[n], that is,
a
1
y
1
[n] + a
2
y
2
[n] = a
1
Ax
1
[n] + a
1
B + a
2
Ax
2
[n] + a
2
B (4)
Clearly, (3) and (4) are different and hence the system is
nonlinear. Under what conditions would it be linear?
102
Causal versus Noncausal Systems
A system is said to be causal if the output of the system at
any time n [i.e. y[n]) depends only on present and past
inputs but does not depend on future inputs.
Example: Determine if the systems described by the
following input-output equations are causal or
noncausal.
(a) y[n] = x[n] x[n-1] (b) y[n] = ax[n] (c)
(d) y[n] = x[n] + 3x[n+4] (e) y[n] = x[n
2
]
(f) y[n] = x[-n]
Solution: The systems (a), (b) and (c) are causal,
others are non-causal.

=
=
n
k
] k [ x ] n [ y
103
Stable versus Nonstable Systems
A system is said to be bonded input
bounded output (BIBO) stable if and only
if every bounded input produces a
bounded output.
DSP Slide 104
z-transform
Transform techniques are an important role in the analysis of
signals and LTI system.
Z- transform plays the same role in the analysis of discrete time
signals and LTI system as Laplace transform does in the
analysis of continuous time signals and LTI system.
For example, we shall see that in the Z-domain (complex Z-
plan) the convolution of two time domain signals is equivalent
to multiplication of their corresponding Z-transform.
This property greatly simplifies the analysis of the response of
LTI system to various signals.
DSP Slide 105
1-The Direct Z- Transform
The z-transform of a sequence x[n] is
Where z is a complex variable. For convenience, the z-transform of a
signal x[n] is denoted by X(z) = Z{x[n]}

=
n
n
z n x z X ] [ ) (
e j
e z X = ) (
We may obtain the Fourier transform from the z transform by
making the substitution . This corresponds to
restricting Also with ,
1 = z
e
j
r z
e
=

=
n
n
j j
e e
r n x r X ) ]( [ ) (
e e
That is, the z-transform is the Fourier transform of the sequence x[n]r
- n
. for r=1
this becomes the Fourier transform of x[n].
The Fourier transform therefore corresponds to the z-transform evaluated on the
unit circle:
DSP Slide 106
z-transform(cont:
The inherent periodicity in frequency of the Fourier transform
is captured naturally under this interpretation.
The Fourier transform does not converge for all sequences - the infinite
sum may not always be finite. Similarly, the z-transform does not
converge for all sequences or for all values of z.
For any Given sequence the set of values of z for which the z-transform
converges is called the region of convergence (ROC).
DSP Slide 107
z-transform(cont:
The Fourier transform of x[n] exists if the sum
converges. However, the z-transform of x[n] is just the Fourier
transform of the sequence x[n]r
-n
. The z-transform therefore exists
(or converge) if
This leads to the condition
for the existence of the z-transform. The ROC therefore consists of a
ring in the z-plane:

= n
n x ] [
< =

=

n
n
r n x z X ] [ ) (
<

=

] [
n
n
z n x
In specific cases the inner radius of this ring may include the origin, and the outer radius
may extend to infinity. If the ROC includes the unit circle , then the
Fourier transform will converge.
1 = z
DSP Slide 108
z-transform(cont:
Most useful z-transforms can be expressed in the form
where P(z) and Q(z) are polynomials in z. The values of z for
which P(z) = 0 are called the zeros of X(z), and the values with
Q(z) = 0 are called the poles. The zeros and poles completely
specify X(z) to within a multiplicative constant.
,
) (
) (
) (
z Q
z P
z X =
In specific cases the inner radius
of this ring may include the
origin, and the outer radius may
extend to infinity. If the ROC
includes the unit circle
, then the Fourier transform will
converge.
1 = z
DSP Slide 109
Example: right-sided exponential sequence

=

= =
n n
n n n
az z n u a z X
0
1
) ( ] [ ) (

<
n
az
1
Consider the signal x[n] = a
n
u[n]. This has the z-transform
Convergence requires that
. a z > . 1
1
<

az which is only the case if

or equivalently
In the ROC, the series converges to

>

= =
0
1
1
, ,
1
1
) ( ) (
n
n
a z
a z
z
az
az z X
since it is just a geometric series.
DSP Slide 110
Example: right-sided exponential sequence
The z-transform has a region of convergence for any finite
value of a.
The Fourier transform of x[n] only exists if the ROC
includes the unit circle, which requires that On
the other hand, if then the ROC does not include
the unit circle, and Fourier transform does not exist. This
is consistent with the fact that for these values of a the
sequence a
n
u[n] is exponentially growing, and the sum
therefore does not converge.
. 1 < a
1 > a
DSP Slide 111
Example: left-sided exponential sequence
Now consider the sequence
]. 1 [ ) ( = n u a n x
n
. 1 s n
This sequence is left-sided because it is nonzero only for
The z-transform is

=

= =
= =
1 0
1
1
) ( 1
] 1 [ ) (
n n
n n n
n n
n n n n
z a z a
z a z n u a z X
For or the series converges to
, 1
1
<

z a
, a z <
Note that the expression for the
z-transform (and the pole zero
plot) is exactly the same as for
the right-handed exponential
sequence - only the region of
convergence is different.
Specifying the ROC is therefore
critical when dealing with the z-
transform.
DSP Slide 112
Example: Sum of two exponentials
ls exponentia real two of sum the is u[n]
3
1
u[n]
2
1
x[n] signal The
n n
|
.
|

\
|
+
|
.
|

\
|
=

)
`

|
.
|

\
|
+
|
.
|

\
|
=
n
n
n n
z n u n u z X ] [
3
1
] [
2
1
) (


=

|
.
|

\
|
+
|
.
|

\
|
=
n
n
n
n
n
n
z n u z n u ] [
3
1
] [
2
1


=

|
.
|

\
|
+
|
.
|

\
|
=
0
1
0
1
3
1
2
1
n
n
n
n
z z
The z transform is
From the example for the right-handed exponential sequence, the first term in this
sum converges for and the second for The combined
transform X(z) therefore converges in the intersection of these regions, namely when
.
In this case
2 / 1 > z 3 / 1 > z
2 / 1 > z
|
.
|

\
|
+
|
.
|

\
|

|
.
|

\
|

=
+
+

=

3
1
2
1
12
1
2
3
1
1
1
2
1
1
1
) (
1 1
z z
z z
z z
z X
DSP Slide 113
Example: Sum of two exponentials
The pole-zero plot and region of convergence of the signal is
DSP Slide 114
Example: finite length sequence
The pole-zero plot and region of convergence of the signal is
The signal
1
1
1
0
1
1
0
1
) ( 1
) ( ) (

= = =

az
az
az z a z X
n
N
n
n
N
n
n n
.
1
1
a z
a z
z
N N
N

=

has z transform
Since there are only a finite number of nonzero terms the sum always converges when
is finite. There are no restrictions on
,
and the ROC is the
entire z-plane with the exception of the origin z = 0 (where the terms in the sum are
infinite). The N roots of the numerator polynomial are at
1
az ) ( < a
1 ,...... 1 , 0 ,
) / 2 (
= = N k ae Z
N k j
k
t
*since these values satisfy the equation Z
N
= a
N
The zero at k = 0 cancels the pole at z = a, so there are no poles except
at the origin, and the zeros are at zk = ae
j(2k/N)
k = 1; : : : ;N -1 The zero at k = 0 cancels the pole at z = a, so there
are no poles except at the origin, and the zeros are at zk = ae
j(2k/N)
k = 1; : : : ;N -1
DSP Slide 115
2-Properties of the region of convergence

). 0 ( s < < s
L R
z t t
). (
2 1
< s s < N n N
The ROC is a ring or disk in the z-plane, centered on the origin
The Fourier transform of x[n] converges absolutely if and only if the ROC of
the z-transform includes the unit circle.
The ROC cannot contain any poles.
If x[n] is finite duration (ie. zero except on finite interval
), then the ROC is the entire Z-plan except perhaps at z=0 or z= .
If x[n] is a right-sided sequence then the ROC extends outward from the
outermost finite pole to infinity.
If x[n] is left-sided then the ROC extends inward from the innermost nonzero
pole to z = 0.
A two-sided sequence (neither left nor right-sided) has a ROC consisting of a
ring in the z-plane, bounded on the interior and exterior by a pole (and not
containing any poles).
The ROC is a connected region.
The properties of the ROC depend on the nature of the signal. Assuming that the
signal has a finite amplitude and that the z-transform is a rational function:
DSP Slide 116
3 - The inverse z-transform
, 2
1
,
2
1
1
1
) (
1
>
|
|
|
|
.
|

\
|

z
z
z X
Formally, the inverse z-transform can be performed by evaluating a
Cauchy integral. However, for discrete LTI systems simpler methods
are often sufficient.
A-Inspection method: If one is familiar with (or has a table
of) common z-transform pairs, the inverse can be found by
inspection. For example, one can invert the z-transform
], [
2
1
x[n]
that recognise we inspection By
. z for ,........
1
1
] [ a
1
n
n u
a
az
n u
n
z
|
.
|

\
|
=
>

Also, if X(z) is a sum of terms then one may be able to do a term-by-
term inversion by inspection, yielding x[n] as a sum of terms.
Using Z-transform pair
DSP Slide 117
3 - The inverse z-transform
( )
( )
X(z). of poles and nonzero the are s ck' the where
1
1
b
X(z)
as factorX(z) to possible always is It
, ) (
1
1
1
1
0
0
0
0
[
[

=

=

=

=

=
=
N
k
k
M
k
k
N
k
k
k
M
k
k
k
z d
z c
a
z a
z b
z X
B-Partial fraction expansion:
For any rational function we can obtain a partial fraction expansion,
and identify the z-transform of each term. Assume that X(z) is
expressed as a ratio of polynomials in z
-1
:
DSP Slide 118
The inverse z-transform
( )
k
d z
k
k
k
z X z d
z d
=

=

=

) ( 1 A
by given are A ts coefficien the case in this
,
1
A
X(z)
1
k
N
1 k
1
k
Partial fraction expansion (Continue:)
N M using obtained be can s ' A The
1
k
0 1
1
<

+ =

= =

M-N
r
N
k
k
k
r
r
,
z d
A
z B X(z)
If M>N and the poles are first order, then an expression of the form
cab be used, and B
r
s be obtained by long division of the numerator.
If M<N and the poles are all first order, then X(z) can be expressed
as
DSP Slide 119
3 - The inverse z-transform Partial fraction expansion
( )

= =

= =

+ =
M-N
r
s
m
m
i
m
N
i k k
k
k r
r
z d
C
z d
A
z B X(z)
0 1
1
, 1
1
.
1
1
sided. - right or sided - left are
sequences the whether decide to used be must properties ROC
the terms For these sequences. l exponentia to correspond
is poles, order - multiple with deal also can which
expansion, fraction partial for form general most The
s term fractional The r]. - [n B form
the of terms to invert and sequences, impulse scaled
and shifted to correspond B terms The texts. DSP
standard most in found be can s ' C the finding of Ways
r
r
m
o
r
z

1
1

z d
A
k
k
DSP Slide 120
Example: inverse by Partial fractions
( )
( )
. 1 z ,
1
2
1
1
1
2
1
2
3
1
1
transform - z with x[n] sequence the Consider
1 1
2
1
2 1
2 1
2
>

+
=
+
+ +
=

z z
z
z z
z z
X(z)
,
1
2
1
1
as expressed be can this 2 N M Since
1
2
1
1
0
z
A
z
A
B
X(z)

+ =
= =
( )
z z
z z z z
1 1
1
1
1 2
1 2 1 2
0
1
2
1
1
5 1 -
2 X(z)

1 5
2 3

2
1 2 ) 1
2
3
2
1
: division long by found be can B value The
z
z
z z

|
.
|

\
|

+
+ =

+
+ + +
DSP Slide 121
Example: inverse by Partial fractions
( ) . ) ( 1
using found be can A and A coecients The
1
2 1
d
z X
k
z k k
z d A
=

=
z
z
z
z z
A
z
z z
A
z
z
1
1
1
1
2 1
2
1
1
2 1
1
1
8
2
1
1
9
- 2 X(z) fore There
9
2 / 1
1 2 1
2
1
1
2 1
and
9
2 1
4 4 1
1
2 1
So
1
1

=
=
+ +
=

+ +
=
=

+ +
=

+ +
=

Using the fact that the ROC , the terms can be inverted one at a time
by inspection to give
. 1 z >
| | | | ]. [ ) 2 / 1 ( 9 2 n u n n x
n
= o
DSP Slide 122
C- Power Series Expansion
( )
...... ] 2 [ ] 1 [ ] 0 [ ] 1 [ ] 2 [ ........ ..........
] [ z
form in series power as given is transform Z If
2
1 1
2
z z
z
z x x z x
n x X
n
n
+ + + + + =
=

then any value in the sequence can be found by identifying the

coefficient of the appropriate power of z
-1
.
DSP Slide 123
Example; Power Series Expansion
( ) ( ) a az X > + =

z , 1 log z
transform Z he Consider t
1
Using the power series expansion for log(1 + x), with /x/< 1, gives
( )

=
+

=
1
1
,
) 1 (
z
n
n n n
n
z a
X
DSP Slide 124
Example; Power Series Expansion by long division
( ) a
az
X >

=

z ,
1
1
z
transform the Consider
1
Since the ROC is the exterior of a circle, the sequence is right-sided. We therefore
divide to get a power series in powers of z
-1
:
( )
]. [ ] [ .... .......... ........ z a az 1
1
1
.....

1

z a az 1
1 1 z
2 - 2 1
1
2 2
2 2 1
1
1
-2 2 1
1
n u a n x Therefore
az
z a
z a az
az
az
az X
n
= + + + =

+ +
=

DSP Slide 125

Example; Power Series Expansion for left-side Sequence
( ) a
az
X <

=

z ,
1
1
z
orm transf - Z the Consider
1
Because of the ROC, the sequence is now a left-sided one. Thus we
divide to obtain a series in powers of z:
]. 1 [ ] [ .... ..........

z a a -
z
1
2 1
2.. -2 1
=

n u a n x Thus
az
z a z
z
z a
n
DSP Slide 126
4- Properties of the z-transform
R ROC ), ( ] [
x
z X n x
z

if X(z) denotes the z-transform of a sequence x[n] and the ROC of X(z) is indicated
by R
x
, then this relationship is indicated as
Furthermore, with regard to nomenclature, we have two sequences such that
R ROC ), ( ] [
R ROC ), ( ] [
x2 2 2
x1 1 1
z X n x
z X n x
z
z


ALinearity: The linearity property is as follows:
. R R contains ROC ), ( ] [ ) ( ] [
x1 x1 2 1 2 1
+ + z bX z aX n bX n ax
z
BTime Shifting: The time shifting property is as follows:
x 0
R ROC ), ( ] [
0
z X z n n x
n
z


(The ROC may change by the possible addition or deletion of z =0 or z = .)
This is easily shown:
). ( ] [
] [ ] [ ) (
0 0
0
) (
0
z X z z m x z
z m x z n n x z Y
n
n
m
n
n
n m
n
n

=
+

= =
= =
DSP Slide 127
Example: shifted exponential sequence
4
1
z ,
4
1
1
) ( >

=
z
z X
Consider the z-transform
From the ROC, this is a right-sided sequence. Rewriting,
. ] 1 [ ) 4 / 1 ( ] [
1
=

n u n x
n
The term in brackets corresponds to an exponential sequence (1/4)
n
u[n]. The
factor z
-1
shifts this sequence one sample to the right.
The inverse z-transform is therefore
4
1
z
4
1
- 1
1
,
4
1
1
) (
1
1
1
1
>
|
|
|
|
.
|

\
|
=

z
z
z
z
z X
DSP Slide 128
C - Multiplication by an exponential sequence
The exponential multiplication property is
where the notation indicates that the ROC is scaled by (that is,
inner and outer radii of the ROC scale by ). All pole-zero locations are
similarly scaled by a factor z
0
: if X(z) had a pole at then X(z/z
0
)
will have a pole at z=z
0
z
1
.
( ). ] [
) (
0 0
e e
j
F
n j
X n x
e e e

If z
0
is positive and real, this operation can be interpreted as a shrinking or
expanding of the z-plane | poles and zeros change along radial lines in the z-
plane.
If z
0
is complex with unit magnitude (z0 = e
jw0
) then the scaling operation
corresponds to a rotation in the z-plane by and angle w
0
, That is, the poles and
zeros rotate along circles centered on the origin. This can be interpreted as a
shift in the frequency domain, associated with modulation in the time domain
by e
jw0n
. If the Fourier transform exists, this becomes
1
z z =
, R ROC ], / [ ] [
x
0
0
0 z
z z X n x
z
n
z

, 0 x
R z
0
z
0
z
DSP Slide 129
Example: exponential multiplication
The z-transform pair
can be used to determine the z-transform of x[n] = r
n
cos(w
0
n)u[n].
Since cos (w
0
n) = 1/2e
jw0n
+ 1/2e
jw0n
. The signal can be written as
( ) ( ) ]. [
2
1
] [
2
1
] [
0 0
n u re n u r n x
n
j
n
j
e
e e
+ =
1 z ,
1
1
] [
1
>

z
n u
z
From the exponential multiplication property,
( ) . z ,
1
2 / 1
] [
2
1
1
0
0
r
z r
n u r
e
e
j
z
n
j
>

e
e
( ) . z ,
1
2 / 1
] [
2
1
1
0
0
r
z r
n u r
e
e
j
z
n
j
>

e
e
. z
1
2 / 1

1
2 / 1
X(z)
So
1 1
0 0
r
z r z r
e e
j j
>

=
e e
. z ,
cos 2 1
cos 1
2 2 1
0
1
0
r
z r z r
z r
>
+

e
e
DSP Slide 130
D- Differentiation
The differentiation property states that
This can be seen as follows: since
. R ROC ,
) (
] [
x
=
dz
z dX
z n nx
z
. ]} [ { ] [ ] [ ) (
) (
1


=

= = =
n
n n
n nx z z n nx z n x n z
dz
z dX
z
have We
, ] [ ) (
- n

=
n
z n x z X
Example: second order pole
The z-transform of the sequence
] [ ] [ n u na n x
n
=
Can be found
( )
. z ,
1 1
1
dz
d
- X(z)
be to
, z
1
1
] [
2
1
1
1
1
a
az
az
az
a
z
n u a
z n
>

=
|
.
|

\
|

=
>

DSP Slide 131

E- Conjugation
This property is
F- Time reversal.
. R ROC *), ( * ] [ *
x
= z X n x
z
Example: Time-reversed exponential sequence
The Signal is a time-reversed version of a
n
u[n]. The
z-transform is therefore
] [ ] [ n u a n x
n
=

.
R
1
ROC *), / 1 ( * ] [ * Here
x
= z X n x
z
, z
L R
r r < <
The notation 1/R
x
means that the ROC is inverted, so if R
x
is the set
of values such that then the ROC is the set of values of z su
that . 1/r / 1
R
< z r
l
. R z ,
1 1
1
) (
x
1
1 1
1 1
= <

=



a
z a
z a
az
z X
DSP Slide 132
G- Convolution
This property state that
. R R contains ROC ), ( ) ( ] [ * ] [
x2 x1 2 1 2 1

z X z X n x n x
z

Example: evaluating a convolution using the z-transform
The z-transforms of the signal x
1
[n] =a
n
u[n] and x
2
[n] = u[n] are
.
R
1
ROC *), / 1 ( * ] [ * Here
x
= z X n x
z
, 1 a <
1 z ,
1
1
) ( .
and
z ,
1
1
) (
1
0
2
1
0
1
>

= =
>

= =

az
z z X
a
az
z a z X
n
n
n
n n
For The z-transforms of the convolution y[n] = x
1
[n] *x
2
[n] is
( )( ) ( )( )
1 z
1 1 1
1
) (
2
1 1
>

=

=

z a z
z
az az
z Y
( )( ) ( )( )
1 z
1 1 1
1
) (
2
1 1
>

=

=

z a z
z
az az
z Y
DSP Slide 133
Using a partial fraction expansion,
( )
( ). ] [ ] [
1
1
) (
1 z ,
1
-
1
1
1
1
) (
1
1 1
n u a n u
a
n y
So
az
a
z a
z Y
n +

=
>
|
.
|

\
|

=
Example: evaluating a convolution using the z-transform
H- Initial Value theorem
If x[n] is zero for n<0, then
). ( ] 0 [
lim
z X x
z
=
DSP Slide 134
Some common z-transform pairs are:
DSP Slide 135
I- Relationship with the Laplace transform:
.
have we
,
) ( T j dT T j d
e e e z
j d s
e e
e
= =
= =
+
Continuous-time systems and signals are usually described by the Laplace
transform. Letting z = e
sT
, where s is the complex Laplace variable
, / 2 f/f 2 T z and
Therefore
s s
dT
e z e te t e = = = =
where w
s
is the sampling frequency. As varies from to , the s-plane is
mapped to the z-plane:
The j axis in the s-plane is mapped to the unit circle in the z-plane.
The left-hand s-plane is mapped to the inside of the unit circle.
The right-hand s-plane maps to the outside of the unit circle.
e

e
DSP Slide 136
137
Lecture -4
Frequency Analysis
Voltage Vs Time Representation That
become
Magnitude Vs Frequency
,
Phase Vs Frequency Representation
And
Vice Versa
138
Frequency Analysis of Signals
Fourier transform and Fourier series basically involve the
decomposition of the signal in terms of sinusoidal components.
With such a decomposition ,a signal is said to be represented in
the frequency domain.
These decompositions are very important in the analysis of LTI
systems because response of a system to a sinusoidal input signal
is a sinusoid of the same frequency but of different amplitude
and phase.
Many other decompositions of signals are possible, only the
class of sinusoidal signals possess this desirable property in
passing through a LTI system.
139
The Fourier Series for Continuous-Time Periodic Signals
The Fourier Series of a periodic analogue signal x(t) is given by
Tp t dt
Tp to
to
Tp to
to
= =
+
+
}
1 . .......... .......... .......... ) (
0
2

=
=
k
t kF j
k
e c t x
t
is a periodic signal with fundamental period T
p
=1/F
o
and k = 0,1, 2,
We can construct periodic signals of various types by proper choice of
fundamental frequency and the coefficients C
K
.F
O
determines the fundamental
period of x(t) and coefficient C
k
specify the shape of waveform. We determine
the expression for C
k
2 ........ ) (
1
0
2
}

=
p
T
t kF j
p
k
dt e t x
T
c
t
dt e c e dt e t x
k
kFot j
k
lFot j
Tp to
to
lFot j
Tp to
to
|
.
|

\
|
=

} }

=
+
+

+
t t t 2 2 2
) (
p l
lFot j
Tp to
to
T C dt e t x =

+
}
t 2
) (
140
In general, the Fourier Coefficients c
k
are complex valued.
If the periodic signal is real, c
k
and c
-k
are complex conjugates. As a
result, if
k
k
j
k k
j
k k
e c c
then
e c c
u

u
=
=
141
Other forms of Fourier Series
Representation
As we have just mentioned

=
t
=
k
t kF 2 j
k
0
e c ) t ( x
The above equation can be re-written as
| |

+ + =
1
) ( 2 2
0
0 0
) (
k
t F k j
k
t kF j
k
e c e c c t x
t t
since
k
j
k k
e c c
u
=
and
k
j
k k
e c c
u

=
( ) ( )
| |
k k
t kF j t kF j
k
k
e e c c t x
u t u t + +

=
+ + =

0 0
2 2
1
0
) (
( )
k
k
k
t kF c c u t + + =

=
0
1
0
2 cos 2
This is called the Cosine
Fourier Series.
142
Other forms of Fourier Series
Representation
Yet another form for the Fourier Series can be obtained by
expanding the cosine Fourier series as
| |

=
u t u t + =
1 k
k 0 k 0 k 0
sin t kF 2 sin cos t kF 2 cos c 2 c ) t ( x
Consequently, we may rewrite the above equation in
the form
( )

=
+ =
1
0 0 0
2 sin 2 cos ) (
k
k k
t kF b t kF a a t x t t
This is called the Trigonometric form of the FS,
where a
0
= c
o
, a
k
= 2|c
k
|cosu
k
and b
k
= 2|c
k
|sinu
k
.
143
Power Density Spectrum of Periodic
Signals
A periodic signal has infinite energy and a
finite average power, which is given as
dt ) t ( x
T
1
P
2
T
p
x
p
}
=
If we take the complex conjugate of (1) and
substitute for x
*
(t), we obtain

} }


=
t

=
t
(
(

= =
k
t kF 2 j
T
p
*
k
T
k
t kF 2 j *
k
p
x
dt e ) t ( x
T
1
c dt e c ) t ( x
T
1
P
0
p p
0
2
k
k
c

=
=
144
Therefore, we have established the relation
2
k
k
2
T
p
x
c dt ) t ( x
T
1
P
p

}

=
= =
Which is called Parse Val's relation for power signals.
This relation states that the total average power in the periodic
signal is simply the sum of the average powers in all the
harmonics.
If we plot the |c
k
| as a function of the frequencies kF
o
,k=0,1,2,.
the diagram we obtain shows how the power of the periodic signal
is distributed among the various frequency components. This diagram
is called the Power Density Spectrum of the periodic signal x(t). A
typical PSD is shown in the next slide.
145
-2F
0
-F
0
0 F
0
2F
0
|c
k
|
2
F
Power density spectrum of a continuous time periodic signal
146
Example1: Determine the Fourier Series and the Power
Density Spectrum of the rectangular pulse train signal
illustrated in the following figure.
Solution:
-t/2 t/2
T
p
T
p
A
x(t)
}
t
t
t
= =
2 /
2 /
p p
0
T
A
Adt
T
1
c
t t
t t t
= =
}
t
t
t
0
0
p
2 /
2 /
t kF 2 j
p
k
kF
kF sin
T
A
dt Ae
T
1
c
0
and
where k = 1, 2, ..
Figure (a), (b) and (c) illustrate the Fourier
coefficients when T
p
is fixed and the pulse width
t is allowed to vary.
147
-60 -40 -20 0 20 40 60
-0.05
0
0.05
0.1
0.15
0.2
t = 0.2T
p
-60 -40 -20 0 20 40 60
-0.04
-0.02
0
0.02
0.04
0.06
0.08
0.1
t = 0.1T
p
Fig.(a)
Fig. (b)
148
-60 -40 -20 0 20 40 60
-0.02
-0.01
0
0.01
0.02
0.03
0.04
0.05
t = 0.05T
p
Fig. (c)
From these three figures we observe that the
effect of decreasing t while keeping T
p
fixed is to
spread out the signal power over the frequency
range. The Spacing between the adjacent lines is
independent of the value of the width t.
149
The following figures demonstrate the effect
of varying T
p
when t is fixed.
150
The figures on the previous slide ()
show that the spacing between adjacent spectral lines
decreases as T
p
increases. In the limit as T
p
, the
Fourier coefficients c
k
approach zero. This behavior is
consistent with the fact that as T
p
and t remains
fixed, the resulting signal is no longer a power signal.
Indeed it becomes an energy signal and its average
power is zero.
The Power Density Spectrum for the rectangular pulse train is

=
|
|
.
|

\
|
t t
t t
|
|
.
|

\
|
t
=
|
|
.
|

\
|
t
=
,... 2 , 1 k ,
kF
kF sin
T
A
0 k ,
T
A
c
2
0
0
2
p
2
p
2
k
151
Lecture -4
Frequency Analysis of Discrete-Time Signals
152
Frequency Analysis of Discrete-Time Signals
We have already discussed the Fourier series representation for continuous-
time periodic (power) signals and the Fourier transform for finite energy
aperiodic signals.
The frequency range for continuous-time periodic signals extends
from - to ,that contain infinite number of frequency
components with frequency spacing (1/T
p
)
.
The frequency range for discrete-time signals is unique over the
interval (-,) or (0,2).
A discrete-time signal of fundamental period N can consist of
frequency components separated by 2/N radians or f= 1/N
cycles.
Consequently, the Fourier series representation of the discrete-
time periodic signal will contain N frequency components (the
basic difference b/w Fourier series representation for continuous-
time and discrete-time periodic signals).
153
The Fourier Series for Discrete-Time Signals
Suppose that we are given a periodic sequence with
period N. The Fourier series representation for x[n]
consists of N harmonically related exponential
functions
e
j2tkn/N
, k = 0, 1,2,.,N-1
and is expressed as

=
t
=
1 N
0 k
N / kn 2 j
k
e c ] n [ x
where the coefficients c
k
can be computed as:

=
t
=
0 n
N / kn 2 j
k
e ] n [ x
N
1
c
154
Example: Determine the spectra of the following signals:
(a) x[n] = [1, 1, 0, 0], x[n] is periodic with period 4 (b) x[n] = costn/3
(c) x[n] = cos(\2)tn
Solution: (a) x[n] = [1, 1, 0, 0]
Now
| | | |
3
0
0
1 1 1 1
[ ] [0] [1] [2] [3] 1 1 0 0
4 4 4 2
n
c x n x x x x
=
= = + + + = + + + =

3 3
2 / 4 / 2 / 2
1
0 0
1 1 1
[ ] [ ] [0] [1] 0 0
4 4 4
j n j n j
n n
c x n e x n e x x e
t t t
= =
( = = = + + +


( ) ( ) ( )
2 2
1 1 1
1 1 cos sin 1 0 1
4 4 4
j j j
t t
( = + = + = (


=
t

=
t
= =
3
0 n
N / kn 2 j
1 N
0 n
N / kn 2 j
k
e ] n [ x
4
1
e ] n [ x
N
1
c

155
The magnitude spectra are:
and the phase spectra are:
0
0
= u
4
1
t
= u
undefined = u
2 4
3
t
= u
| | 0 sin j cos 1
4
1
= t t + =
( ) | | | | | | j 1
4
1
j 0 1
4
1
) 2 / 3 sin( j 2 / 3 cos 1
4
1
e ] n [ x
4
1
c
3
0 n
4 / 3 n 2 j
3
+ = + + = t t + = =

=
t
| |
t
=
t
=
t
+ = = =

j
3
0 n
n j
3
0 n
4 / n 2 2 j
2
e . 1 1
4
1
e ] n [ x
4
1
e ] n [ x
4
1
c
4
2
c
1
= 0 c
2
=
4
2
c
3
=
2
1
c
0
=
156
(b) x[n] = costn/3
Solution: In this case, f
0
= 1/6 and hence x[n] is periodic with
fundamental period N = 6.
Now

=

= =

= = =
5
0
3 / 6 / 2
5
0
5
0
/ 2
3
cos
6
1
3
cos
6
1
] [
6
1
n
kn j kn j
n n
N kn j
k
e
n
e
n
e n x c
t t t
t t
| |
( ) ( )
| |

=
+

=
+ = + =
5
0
1 1
3 / 3 / 3 /
5
0
3 3
12
1
2
1
6
1
n
k j k j
kn j n j n j
n
n n
e e e e e
t t
t t t
| | 0 cos cos cos cos cos 0 cos
6
1
3
cos
6
1
3
cos 2
12
1
3
5
3
4
3
3
3
2
3
5
0
5
0
0
= + + + + + =
= =

= =
t t t t t
t t
n n
n n
c
Similarly, c
2
= c
3
= c
4
= 0, c
1
= c
5
= .
157
(c) Cos(\2)tn
Solution: The frequency f
0
of the signal is 1/\2
Hz. Since f
0
is not a rational number, the signal
is not periodic. Consequently, this signal cannot
be expanded in a Fourier series.
158
Power density Spectrum of Periodic Signals
The average power of a discrete time periodic signal with period N is

=
=
1
0
2
) (
1
N
n
x
n x
N
P
The above relation may also be written as


=

=
t

=
|
|
.
|

\
|
= =
1 N
0 n
1 N
0 n
N / kn 2 *
k
1 N
0 n
*
x
e c ] n [ x
N
1
] n [ x ] n [ x
N
1
P
or

=
t

=
= =
(

=
1 N
0 n
2
2
1 N
0 k
k
1 N
0 n
N / kn 2 j
1 N
0 n
*
k x
] n [ x
N
1
c
e ] n [ x
N
1
c P
This is Parse Val's Theorem for Discrete-Time Power Signals.
159
The Fourier Transform of Discrete-Time Aperiodic
Signals
The Fourier Transform of a finite energy discrete time signal x[n] is defined as

=
n
jwn
e ] n [ x ) w ( X
X(w) may be regarded as a decomposition of x[n] into its Frequency
components. It is not difficult to verify that X(w) is periodic with frequency
2t.Indeed,X() is periodic with period 2,that is,
( ) ( )
( )
( )
( ) ) ( . .......... ..........
......... ..........
2
2
2
e
t e
e
t e
t e
X e n x
e e n x
e n x k X
n j
n
kn j n j
n
n
n k j
= =
=
= +

=
+

160
We observe two basic differences b/w the Fourier transform of a discrete-
time finite-energy signal and the Fourier transform of a finite-energy
analog signal .
First, for continuous time signals, the spectrum of the signal have a
frequency range of (-,). In contrast, the frequency range for a discrete -
time signal is unique over the frequency interval of (-,).
The second one is also a consequence of the discrete-time nature of the
signal. Since the signal is discrete in time , the Fourier transform of the
signal involves the summation of terms instead of an integral, as in the
case of continuous time signals.
Let us evaluate the sequence x(n) from X().we multiply both sides of
X() by e
jm
and integrate over the interval (-,).
e e e
e
t
t
e e
t
t
d e e n x d e X
m j n j
n
m j
}

}

= ) ( ) (
}

=
=
=
t
t
e
t
e
n m
n m
d e
n m j
......... 0
........ 2
) (
161
Energy Density Spectrum of Aperiodic Signals
Energy of a discrete time signal x[n] is defined as
2
n
x
] n [ x E

=
=
Let us now express the energy E
x
in terms of the spectral characteristic
X(w). First we have

}

=
t
t
-
(

t
= =
n n
jwn *
x
dw e ) w ( X
2
1
] n [ x ] n [ x ] n [ x E
If we interchange the order of integration and summation in the above
equation, we obtain
dw ) w ( X
2
1
dw e ] n [ x ) w ( X
2
1
E
2
n
jwn
x
} }

t
t
t
t

=
-
t
=
(

t
=
e e
t
t
e
e
t
t
t
t
e
d e X n x
n m
n m n x
d e n x
n j
n m j
n
}
}

=
=

=
=
=
) (
2
1
) (
......... 0
)........ ( 2
) (
) (
162
Therefore, the energy relation between x[n] and X(w) is
dw ) w ( X
2
1
] n [ x E
2
2
n
x
}

t
t

=
t
= =
This is Parse Val's relation for discrete-time aperiodic signals.
163
Example: Determine and sketch the energy density
spectrum of the signal x[n] = a
n
u[n],
-1<a<1
Solution:
( )


=

= = = =
0 n
jw
n
jw jwn
0 n
n
n
jwn
ae 1
1
ae e a e ] n [ x ) w ( X
The energy density spectrum (ESD) is given by
( )( )
jw jw
2
xx
ae 1 ae 1
1
) w ( X ) w ( X ) w ( X ) w ( S
+
= = =

-
2
a w cos a 2 1
1
+
=
0
t t
w
X(w)
a = 0.5
a= -0.5
164
Example: Determine the Fourier Transform and the energy
density spectrum of the sequence
Solution:

s s
=
otherwise , 0
1 L n 0 , A
] n [ x
) 2 / w sin(
) 2 / wL sin(
Ae
e 1
e 1
A Ae e ] n [ x ) w ( X
) 1 L )( 2 / w ( j
jw
jwL
1 L
0
jwn
n
jwn

= = =

The magnitude of x[n] is

=
=
otherwise , A
0 w , L A
) w ( X
) 2 / w sin(
) 2 / wL sin(
and the phase spectrum is
) 2 / w sin(
) 2 / wL sin(
2
w
) 2 L ( A ) w ( X Z + Z Z = Z
The signal x[n] and its magnitude is plotted on the next slide. The
Phase spectrum is left as an exercise.
165
x[n]
|X(w)|
166
Properties of DTFT
Symmetry Properties:
Suppose that both the signal x[n] and its transform X(w) are complex valued.
Then they can be expressed as
x[n] = x
R
[n] + j x
I
[n] (1)
X() = X
R
() + j X
I
() (2)
The DTFT of the signal x[n] is defined as

=
n
n j
e n x X
e
e ] [ ) (
(3)
Substituting (1) and (2) in (3) we get
| |
n j
n
I R I R
e n x n x jX X
e
e e

+ = + ] [ ] [ ) ( ) (
bu
t
n j n e
n j
e e
e
sin cos =

167
| || | n j n n x n x jX X
n
I R I R
e e e e sin cos . ] [ ] [ ) ( ) ( + = +

=
separating the real and imaginary parts, we have
| |

=
+ =
n
I R R
n n x n n x X e e e sin ] [ cos ] [ ) (
(4)
| |

=
=
n
I R I
n n x n x X e e e e cos ] [ sin ) ( ) (
(5)
In a similar manner, one can easily prove that
| |
| | e e e e e
t
e e e e e
t
t
t
d n X n X n x
d n X n X n x
I R I
I R R
}
}
+ =
=
2
2
cos ) ( sin ) (
2
1
] [
sin ) ( cos ) (
2
1
] [
168
DTFT Theorems and Properties
Linearity
If x
1
[n] X
1
(w) and x
2
[n] X
2
(w), then
a
1
x
1
[n] + a
2
x
2
[n] a
1
X
1
(w) + a
2
X
2
(w)
Example 1: Determine the DTFT of the signal
x[n] = a
|n| , -1< a <1
Solution: First, we observe that x[n] can be expressed as
x[n] = x
1
[n] + x
2
[n]
where
169

<
>
=
0 n , 0
0 n , a
] n [ x
n
1
and

>
<
=

0 n , 0
0 n , a
] n [ x
n
2
Now
( )


=

=
= = =
0 0
1 1
] [ ) (
n
n
j
n
n j n n j
n
ae e a e n x X
e e e
e
( ) ( )
e
e e e
j
j j j
ae
ae ae ae

= + + + + =
1
1
.... 1
3 2
and
( ) ( )


=

= = =
1 1
2 2
] [
n
n
j n j
n
n
n
n j
ae e a e n x X
e e e
e
( )
e
e
e e e
j
j
j j
k
k
j
ae
ae
ae ae ae

= + + = =

=
1
... ) (
2
0
Now
2
2
2 1
cos 2 1
1
1 1
1
) ( ) ( ) (
a a
a
ae
ae
ae
X X X
j
j
j
+

=

= + =

e
e e e
e
e
e
170
Time Shifting
If x[n] X() then x[n-k] = e
-jk
X()
Proof:
| |

=
n
n j
e k n x k n x F
e
] [ ] [
Let n k = m or n = m+k
| | ) ( ] [ ] [ ] [
) (
e
e e e
X e e m x e e m x k n x F
k j
m
m j jwk
m
k m j

=
+
= = =

Time Reversal property
If x[n] X() then x[-n] X(-)
Proof:
| |


=

= = = =
m m
m j m j
n
n j
X e m x e m x e n x n x F ) ( ] [ ] [ ] [ ] [
) (
e
e e e
171
Lecture -7
172
Convolution Theorem
If x
1
[n] X
1
() and x
2
[n] X
2
()
then
x[n] = x
1
[n]*x
2
[n] X () = X
1
()X
2
()
Proof: As we know[convolution formula]
] k n [ x ] k [ x ] n [ x * ] n [ x ] n [ x
2
k
1 2 1
= =

=
n j
n k n
n j
e k n x k x e n x X
e e
e

= = ] [ ] [ ] [ ) (
2 1
Therefore
Interchanging the order of summation and making a substitution
n-k = m, we get
) ( ) ( ] [ ] [
] [ ] [ ) (
2 1 2 1
) (
2 1
e e
e
e e
e
X X e m x e k x
e m x k x X
m
m j k j
k
k m j
k m
=
(

=
(

=
+

=
If we convolve two signal in time domain, then this is equivalent to
multiplying their spectra in frequency domain.
173
Example 2: Determine the convolution of the sequences
x
1
[n] = x
2
[n] = [1, 1, 1]
Solution:

=

= = =
1
1
1 1 2 1
] [ ] [ ) ( ) (
n
n j
n
n j
e n x e n x X X
e e
e e
| | e
e e e e
cos 2 1 ] 1 [ ] 1 [ ] 0 [ ] 1 [
2 1 1
+ = + + = + + =
j j j j
e e e x x e x
Then X() = X
1
()X
2
() = (1 + 2cos)
2
=1 + 4cos+ 4(cos)
2
.
= 1 + 4cos+ 4(1+cos2/2) = 1 + 4cos+ 2(1+cos2).
= 1 + 4cos+ 2+2cos2).
= 3 + 4cos + 2cos2
= 3 + 2(e
j
+ e
-j
) + (e
j2
+ e
-j2
)
Hence the convolution of x
1
[n] and x
2
[n] is
x[n] = [1 2 3 2 1]
As known
1 2
( ) ( ) 1 2cos( ) X X e e e = = +
174
The Wiener-Khintchin Theorem:
Let x[n] be a real signal. Then r
xx
[k] S
xx
(w)
In other words, the DTFT of autocorrelation function is
equal to its energy density function.*
Proof: The autocorrelation of x[n] is defined as

=
=
k
xx
n k x k x n r ] [ ] [ ] [
Now
jwn
k n
xx
e n k x k x n r F

=
(

=

] [ ] [ ]] [ [
Re-arranging the order of summations and making
Substitution m = k-n we get
) (
] [ ] [ ]] [ [
m k jw
m k
xx
e m x k x n r F

=
(

=

175
( ) e e e e
e e
xx
m
m j
k
k j
S X X X e m x e k x = = =
(

=


=

=
2
| ) ( | ) ( ) ( ] [ ] [
Frequency Shifting:
) ( ] [
) ( ] [
0
0
e e
e

X n x e
then
X n x
If
n jw
As from above property, multiplication of a sequence x(n) by
is equivalent to a frequency translation of the spectrum X(w)
by w
o
. So it be periodic, The shift o applies to the spectrum of
the signal in every period
o
j n
e
e
Displacement in frequency multiplies the time/space function by a
unit phasor which has angle proportional to time/space and to the
amount of displacement.
176
The Modulation Theorem:
If x[n] X(w) then
x[n] cos
0
[n] X( +
0
) + X( -
0
)
Proof: Multiplication of a time/space function by a
cosine wave splits the frequency spectrum of the
function.
Half of the spectrum shifts left and half shifts right.
This is simply a variant of the shift theorem
which makes use of Euler's relationship
| |
n j
n
n j n j
n
n j
e
e e
n x ne n x n n x F
e
e e
e
e e

+
= =
2
] [ cos ] [ cos ] [
0 0
0 0
Use
frequency
shift
property
2
) cos(
jx jx
e e
x

+
=
177
The Modulation Theorem: If x[n] X(w) then
x[n] cos
0
[n] X( +
0
) + X( -
0
)
Proof:
| |
) ( ) (
0 0
] [
2
1
e e e e +

=
+ =

j n j
n
e e n x
| |
n j
n
n j n j
n
n j
e
e e
n x ne n x n n x F
e
e e
e
e e

=

=

+
= =
2
] [ cos ] [ cos ] [
0 0
0 0
( ) ( )
0 0
) ( ) (
2
1
2
1
] [
2
1
] [
2
1
0 0
e e e e
e e e e
+ + = + =


=

=
+
X e n x e n x
n
n j
n
n j
Use frequency shift property
178
Parsevals Theorem:
If x
1
[n] X
1
(w) and x
2
[n] X
2
(w) then
dw ) w ( X ) w ( X
2
1
] n [ x ] n [ x
*
2 1
n
*
2 1
}

t
t

=
t
=
Proof:
( ) ( ) | | e e
t
e e e
t
t
t
e
t
t
d X e n x d X X S H R
n j
n
) (
2
1
2
1
. . .
*
2 1
*
2 1
}

}

= =
S H L n x n x d e X n x
n
n j
n
. . ] [ ] [ ) (
2
1
] [
*
2 1
*
2 1
= = =

}

=
e e
t
e
t
t
In the special case where x
1
[n] = x
2
[n] = x[n], the Parsevals
Theorem reduces to
e e
t
t
t
d X n x
n
2
2
) (
2
1
) (
}

=
=
We observe that the LHS of the above equation is energy E
x
of the Signal and the R.H.S is equal to the energy
density spectrum.
179
Thus we can re-write the above equation as
} }

=
= = =
t
t
t
t
e e
t
e e
t
d S d X n x E
xx
n
x
) (
2
1
) (
2
1
] [
2
2
Multiplication of two sequences: [Windowing Theorem]
If x
1
[n] X
1
() and x
2
[n] X
2
() then
( ) ( ) e
t
t
t
d X X n x n x
}

2 1 2 1
2
1
] [ ] [
Windowing is the process of taking a small subset of a larger
dataset, for processing and analysis. A naive approach, the
rectangular window, involves simply truncating the dataset before
and after the window, while not modifying the contents of the
window at all. However, as we will see, this is a poor method of
windowing and causes power leakage.
Application of a window to a dataset will alter the spectral
properties of that dataset. In a rectangular window, for instance, all
the data points outside the window are truncated and therefore
assumed to be zero. The cut-off points at the ends of the sample will
introduce high-frequency components
180
Multiplication of two sequences: [Windowing Theorem](cont:)
If x
1
[n] X
1
() and x
2
[n] X
2
() then
( ) ( ) e
t
t
t
d X X n x n x
}

2 1 2 1
2
1
] [ ] [
Proof:
| | ( )
n j
n
n j n j
n
e n x d e X e n x n x n x n x F
e
t
t
e

t

=

}
(

= = ] [
2
1
] [ ] [ ] [ ] [
2 1 2 1 2 1
( ) ( ) e
t

t
t
t
e
t
t
d X X e n x d X
n
n j
) (
2
1
] [
2
1
2 1
) (
2 1
=
(

=
}

Show periodic Convolution

Technique use for FIR filter design
181
Differentiation in the Frequency Domain:
If x[n] X(w) then Fnx[n] jdX(w)/dw
Proof:
n j
n n
n j
e
d
d
n x e n x
d
d
d
dX
e e
e e e
e

=
(

= ] [ ] [
) (

=
n
n j
e n nx j
d
dX
e
e
e
] [
) (
Multiplying both sides by j we have

=
n
n j
e n nx
d
dX
j
e
e
e
] [
) (
OR
]] [ [
) (
n nx F
d
dX
j =
e
e
Differentiation of a function induces a 90 phase shift in the
spectrum and scales the magnitude of the spectrum in proportion
to frequency. Repeated differentiation leads to the general result:
This theorem explains why differentiation of a signal has the reputation
for being a noisy operation. Even if the signal is band-limited, noise will
introduce high frequency signals which are greatly amplified by
differentiation.
182
The response of any LTI system to an arbitrary input
signal x[n] is given by convolution sum Formula
The Frequency Response Function:

=
=
k
] k n [ x ] k [ h ] n [ y
(6)
In this I/O relationship, the systemis characterized
in the time domain by its unit impulse response
h[k]. To develop a frequency domain
characterization of the system, let us excite the
systemwith the complex exponential
x[n] = Ae
jwn
. - < n < (7)
where Ais the amplitude and w is an arbitrary frequency
confined to the frequency interval [-t, t]. By substituting
(7) into (6), we obtain the response
183

=
k
) k n ( jw
Ae ] k [ h ] n [ y
jwn
k
jwk
e e ] k [ h A
(

or
jwn
e ) w ( AH ] n [ y =
where

=
k
jwk
e ] k [ h ) w ( H
(8)
(9)
The exponential Ae
jwn
is called an Eigen-function of
the system. An Eigen function of a system is an input
signal that produces an output that differs from the
input by a constant multiplicative factor. The
multiplicative factor is called an Eigen-value of the
System.
Response is also in form of complex exponential with same frequency as input, but altered by the multiplicative factor H(w).
184
Example: Determine the magnitude and phase of H(w)
for the three point moving average(MA) system
y[n] = 1/3[x[n+1] + x[n] + x[n-1]]
Solution: since h[n] = [1/3, 1/3, 1/3]
It follows that
H(w) = 1/3(e
jw
+1 + e
-jw
) = 1/3(1 + 2cosw)
Hence
|H(w)| = 1/3|1+2cosw| and

t < s t t
t s s
= u
w 3 / 2 ,
3 / 2 w 0 , 0
) w (
185
1
2t/3
w
0
1
0
2t/3
t
w
186
Example:An LTI system is described by the
following difference equation:
y[n] = ay[n-1] + bx[n], 0 < a < 1
(a) Determine the magnitude and phase of
the frequency response H(w) of the system.
(b) Choose the parameter b so that the
maximum value of |H(w)| is unity.
(c) Determine the output of the system to
the input signal
x[n] = 5 + 12sin(t/2)n 20cos (tn + t/4)
187
Solution:
(a) The frequency response is
) w ( bX ) w ( Y ae ) w ( Y
jw
+ =

) w ( bX ) w ( Y ) ae 1 (
jw
=

( ) w sin ja w cos a 1
b
) w sin j w (cos a 1
b
ae 1
b
) w ( X
) w ( Y
) w ( H
jw

=

=

= =

Now
( ) ( )
w cos a 2 a 1
b
w sin a w cos a 1
b
) w ( H
2 2 2
+
=
+
=
and
|
.
|

\
|

= u

w cos a 1
w sin a
tan ) w (
1
These responses are sketched on the next slide.
188
(b) It is easy to find that |H(w)| attains its
maximum value at w = 0. At this frequency,
we have
1
a 1
b
) 0 ( H =

=
Which implies that b = (1-a).
At b = (1-a), we have
w cos a 2 a 1
a 1
) w ( H
2
+

=
and
w cos a 1
w sin a
tan ) w (
1

= u

(c) The input signal consists of components of frequencies
w = 0, t/2 and t radians.
For w = 0, |H(0)| = 1 and u(0) = 0.
For w = t/2,
. 074 . 0
) 9 . 0 ( 1
9 . 0 1
a 1
a 1
2
H
2 2
=
+

=
+

=
|
.
|

\
|
t
0 1 1
2
42 ) 9 . 0 ( tan a tan = = =
|
|
.
|

\
|
t
O

189
For w = t,
0 ) (
053 . 0
a 1
a 1
| ) w ( H |
= t O
=
+

=
Therefore, the output of the system is
( )
(

t O +
t
+ t t
(

|
.
|

\
|
t
O +
t
|
.
|

\
|
t
+ =
4
n cos ) ( H 20
2
n
2
sin
2
H 12 ) 0 ( H 5 ] n [ y
) n cos( 06 . 1 ) 42 n sin( 888 . 0 5
4
0
2
t t
+ t + =
190
Response to A-periodic input signals
Consider the LTI system of the following figure
where x[n] is the input, and y[n] is the output.
LTI System
h[n], H(w)
x[n] y[n]
If h[n] is the impulse response of the system, then
y[n] = h[n]*x[n]
The corresponding frequency domain representation is
Y(w) = H(w)X(w) [Corresponding Fourier transform of the y(n),x(n), & h(n) respectively]
Now the squared magnitude of both sides is given by
|Y(w)|
2
= |H(w)|
2
|X(w)|
2
Or
S
yy
(w) = |H(w)|
2
S
xx
(w)
where S
xx
(w) and S
yy
(w) are the energy density spectra of the
input and output signals, respectively.
191
The energy of the output signal is
} }
t
t
t
t
t
=
t
= dw ) w ( S | ) w ( H |
2
1
dw ) w ( S
2
1
E
xx
2
yy y
Example: An LTI system is characterized by its
impulse response h[n] = (1/2)
n
u[n]. Determine the
spectrum and the energy density spectrum of
the output signal when the systemis excited by the
signal x[n] = (1/4)
n
u[n].
Solution:
( )
jw
2
1
jwn
n
0 n
2
1
e 1
1
e ) w ( H

=

= =

Similarly,
jw
4
1
e 1
1
) w ( X

=
Hence the spectrum of the signal at the output of the system
is
192
( )( )
jw
4
1
jw
2
1
e 1 e 1
1
) w ( X ) w ( H ) w ( Y


= =
The corresponding energy density spectrum is
2 2 2
yy
) w ( X ) w ( H ) w ( Y ) w ( S = =
( )( ) w cos w cos
1
2
1
16
17
4
5

=
193
DTFT and DFT
The DTFT of an aperiodic discrete time
signal is defined as
The DFT of a signal is defined as
Inverse DFT is defined as

=
n
jwn
e n x w X ] [ ] [

=
t
=
1 N
0 n
N / n 2 jk
e ] n [ x ) k ( X

=
=
1
0
/ 2
] [
1
] [
N
k
N kn j
e k X
N
n x
t
What is difference between DTFT and DFT?
(1)
(2)
(3)
194
The DFT is periodic with period N.
Proof:

=
t
=
1 N
0 n
N / n 2 jk
e ] n [ x ] k [ X

=
t +
= +
1 N
0 n
N / n 2 ) N k ( j
e ] n [ x ] N k [ X

=
t t
=
1 N
0 n
n 2 j N / n 2 jk
e e ] n [ x
Since e
-j2tn
= 1

=
t
= +
1 N
0 n
N / n 2 jk
e ] n [ x ] N k [ X
] k [ X =
proved
195
Example 1: Find the DFT of the following sequence
[1 0 0 1]

=
t
=
t

=
t
= = =
3
0 n
2 / n jk
3
0 n
4 / n 2 jk
1 N
0 n
N / n 2 jk
e ] n [ x e ] n [ x e ] n [ x ] k [ X
2 1 0 0 1 ] 3 [ x ] 2 [ x ] 1 [ x ] 0 [ x ] n [ x ] 0 [ X
3
0 i
= + + + = + + + = =

=
2 / 3 j
3
0 n
2 / n jk
e ] 3 [ x 0 0 ] 0 [ x e ] n [ x ] 1 [ X
t
=
t
+ + + = =

j 1 ) sin( j ) cos( 1 e . 1 1
2
3
2
3
2 / 3 j
+ = + = + =
t t
t
( ) 0 ] 3 sin ) 3 .[cos( 1 1 ] 3 [ ] 0 [ ] [ ] 2 [
3
0
3
= + = + = =

=

t t
t t
j e x x e n x X
n
j n j
196
Example 2: Find the IDFT of the sequence
[2 1+j 0 1-j]
Solution:
Now

=
t
=
1 N
0 k
N / n 2 jk
e ] k [ X
N
1
] n [ x
| | ] 3 [ X ] 2 [ X ] 1 [ X ] 0 [ X
4
1
] k [ X
4
1
] 0 [ x
1 N
0 k
+ + + = =

=

=
t
=
t
= =
3
0 k
2 / jk
3
0 k
4 / 2 jk
e ] k [ X
4
1
e ] k [ X
4
1
] 1 [ x
0 e ] 3 [ X e ] 2 [ X e ] 1 [ X ] 0 [ X
2 / 3 j j 2 / j
= + + + =
t t t
Similarly,
X[2] = 0 and
X[3] = 1
197
Computational Complexity of the DFT
A large number of multiplications and
additions are required for the calculation
of the DFT.
Consider an 8-point DFT as given by
Let k2t/8 = K

=
t
=
7
0 n
8 / n 2 jk
e ] n [ x ] k [ X

=
7
0 n
jKn
e ] n [ x ] n [ x
7 jK 6 jK 5 jK
4 jK 3 jK 2 jK 1 jK 0 jK
e ] 7 [ x e ] 6 [ x e ] 5 [ x
e ] 4 [ x e ] 3 [ x e ] 2 [ x e ] 1 [ x e ] 0 [ x


+ +
+ + + + +
198
There are eight complex multiplications and
seven complex additions. There are also
eight harmonic components to be evaluated.
Therefore, for an 8-point DFT:
Number of complex multiplications = 88
Number of complex additions = 87
For an N-point DFT
complex multiplications = N
2
complex additions = N(N-1)
Clearly some means of reducing these is
required.
199
Decimation-in-time fast fourier transform
algorithm (Cooley-Tuckey Algorithm):
Notations: Equation (2) can be re-written as
Let
N / nk 2 j
1 n
0 N
n 1
e x ] k [ X
t

=
(4)
N / 2 j
N
e W
t
=
(5)
Also note that
2 /
) 2 / /( 2 2 ) / 2 ( 2
] [
N
N j N j
N
W e e W = = =
t t
(6)
and
) 2 / N )( N / 2 ( j k
N
2 / N
N
k
N
) 2 / N k (
N
e W W W W
t +
= =
( )
k
N
k
N
j k
N
W sin j cos W e W = t t = =
t
(7)
200
Summary:
N / 2 j
N
e W
t
=
2 / N
2
N
W W =
k
N
) 2 / N k (
N
W W =
+
DFT:

=
=
1 N
0 n
kn
N n 1
W x ] k [ X
(8)
201
Consider n data samples as:
x
0
x
1
x
2
x
3
x
n
Divide these samples into an even
numbered and odd numbered sequenes x
2n
and x
2n+1
respectively.
That is,
x
2n
= x
0
x
2
x
4
..,x
N-2
x
2n+1
= x
1
x
3
x
5
.x
N-1
Both of the above sequences contain N/2
points.
202
Now equation (8) can be re-written as
follows:
k ) 1 n 2 (
N
1 2 / N
0 n
1 n 2
nk 2
N
1 2 / N
0 n
n 2 1
W x W x ] k [ X
+

=
+

=

+ =


=
+

=
+ =
1 2 / N
0 n
nk 2
N 1 n 2
1 2 / N
0 n
k
N
nk 2
n 2
W x W W x
N
since
nk
2 / N
nk 2
n
W W =
Therefore,


=
+

=
+ =
1 2 / N
0 n
nk
2 / N 1 n 2
k
N
1 2 / N
0 n
nk
2 / N n 2 1
W x W W x ] k [ X
The above equation can be re-written as
] k [ X W ] k [ X ] k [ X
12
k
N 11 1
+ =
(9)
203
Considering line 6 of the table it is seen that
4
k
4 / N 0 21
x w x ] k [ X + =
k = 0,1
Thus
4 0 21
x x ] 0 [ X + =
while
4 0 4
2 / 2 j
0 4 4 / N 0 21
x x x e x x W x ] 1 [ X = + = + =
t
similarly
7 3 24
5 1 23
6 2 22
x x ] 0 [ X
x x ] 0 [ X
x x ] 0 [ X
+ =
+ =
+ =
7 3 24
5 1 23
6 2 22
x x ] 1 [ X
x x ] 1 [ X
x x ] 1 [ X
=
=
=
We observe that the values with k = 1 differ only by a sign from
those with k = 0.
204
Now
] k [ X W ] k [ X ] k [ X
22
k
2 / N 21 11
+ =
(10)
So,
] 0 [ X ] 0 [ X ] 0 [ X W ] 0 [ X ] 0 [ X
22 21 22
0
2 / N 21 11
+ = + =
(11)
] 1 [ jX ] 1 [ X e ] 1 [ X ] 1 [ X W ] 1 [ X ] 1 [ X
22 21
2 / j
21 22
1
2 / N 21 11
= + = + =
t
(12)
] 2 [ X ] 2 [ X ] 2 [ X e ] 2 [ X ] 2 [ X W ] 2 [ X ] 2 [ X
22 21 22
2 2 ) 8 / 2 ( j
21 22
2
2 / N 21 11
= + = + =
t
Now
] 0 [ X x x x W x x W x ] 2 [ X
21 4 0 4
2
2 0 4
2
2 / N 0 21
= + = + = + =
and
] 0 [ X x x x W x ] 2 [ X
22 6 2 6
2
4 / N 2 22
= + = + =
(13)
Hence equation (13) is equivalent to
] 0 [ X ] 0 [ X ] 2 [ X
22 21 11
+ =
] 3 [ X W ] 3 [ X ] 3 [ X
22
3
2 / N 21 11
+ =
(14)
(15)
205
Now
] 1 [ X x x x e x x e x x W x ] 3 [ X
21 4 0 4
3 j
0 4
3 ) 2 / 2 ( j
0 4
3
4 / N 0 21
= = + = + = + =
t t
and ] 1 [ X x x ] 3 [ X
22 6 2 22
= =
Hence equation (15) is equivalent to
] 1 [ jX ] 1 [ X ] 1 [ X e ] 1 [ X ] 3 [ X
22 21 22
3 ) 4 / 2 ( j
21 11
+ = + =
t
(16)
Drawing these results together gives
] 1 [ X W ] 1 [ X ] 1 [ jX ] 1 [ X ] 3 [ X
] 1 [ X W ] 1 [ X ] 1 [ jX ] 1 [ X ] 1 [ X
] 0 [ X W ] 0 [ X ] 0 [ X ] 0 [ X ] 2 [ X
] 0 [ X W ] 0 [ X ] 0 [ X ] 0 [ X ] 0 [ X
22
2
8 21 22 21 11
22
2
8 21 22 21 11
22
0
8 21 22 21 11
22
0
8 21 22 21 11
= + =
+ = =
= =
+ = + =
(17)
The above equations are known as recomposition equations.
206
The number of complex additions and
multiplications involved is reduced in this way
because:
(i) the recomposition equations are expressed in
terms of powers of the recurring factor W
N
.
(ii) use is also made of relationships of the type
X
21
[2] = X
21
[0] and X
21
[3] = X
21
[1] and
(iii) the presence of only sign differences in the
pairs of expressions is exploited.
The algorithm is known as the Cooley-Tukey
algorithm.
It can be shown that
Number of complex multiplications = (N/2)log
2
N