Lecture 8

Digital Signal & Image Processing
Lecture-8
Dr Muhammad Arif
m.arif@faculty.muet.edu.pk
https://sites.google.com/site/mdotarif/teaching/dsip
Overview
Discrete Time Fourier Transform (DTFT)

Properties of DTFT
Frequency Response & Difference Equation
Frequency Response & Transfer Function
Frequency Response & Impulse Response
System Output in the Frequency Domain
Frequency Response & Filter Shape
Properties of Magnitude & Phase Reponses
Filters Shape from Poles & Zeros
Discrete Time Fourier Transform
The discrete time Fourier transform (DTFT) is a frequency

domain tool used to find the spectrum X() of a signal x[n] (or
the frequency response H() of a system.
The frequency response H() of a system is composed of two

parts, the magnitude response and phase response.
This is a consequence of the fact that the frequency response

is a complex number and therefore can be written in polar
form, using magnitude and phase.
The magnitude response gives the shape of a filter and the

greatest insight into the way the filter behaves. 3
The DTFT of the impulse response h[n] of the system is

defined as
Where H() is the frequency response of a system.
4
The discrete time Fourier transform (DTFT) of a signal [] is
where is the digital frequency (in radians) defined as
The DTFT of a signal [] is denoted as
5
Example-1: Find the DTFT of the signal.
Solution
Only the four nonzero samples
at = 0, 1, 2 and 4 are contributing
to the transform.
= []
=
= 2 + 3 2 + 4
6
Example-2: Find the DTFT of the signal.

x[n] = 4(u[n] u[n-3])
Solution
The signal [] can be written as
= []
=
= + +
7
Since =
= []

Where = cos n sin n

Therefore

= [] cos n sin n
=
The transform X() is different at each different digital frequency .
It is largest when the signal x[n] resonates with the sinusoids cos(n)
or sin(n).
That is., the X() is largest when signal x[n] varies at nearly the same
frequency .
The product of the cosine signal x[n] with other cosines is largest when
the signals are most similar as shown in the next slides. 8
Signal Resonance for the Discrete Time Fourier Transform
9
10
11
12
Properties of DTFT
The two important properties of DTFT are
1. Time delay or Time shifting

2. Periodicity
13
Properties of DTFT
1. Time delay or Time shifting
14
Properties of DTFT
A delay of M steps in the time domain appear in the DTFT

domain as a complex exponential. That is,
For example, when delay M = 1
15
Properties of DTFT
This delay property of DTFT also makes conversion from a

difference equation to a frequency response H() very
straightforward.
In practice, the process is accomplished most easily by using
the transfer function H(z) as an intermediate step.
y[n] + a1y[n-1] + a2y[n-2] + = b0x[n] + b1x[n-1] + b2x[n-2] +
16
Properties of DTFT
2. Periodicity
Consider (+2)
17
Properties of DTFT
The DTFT is periodic with a period of 2.
So the magnitude and phase responses also repeat with the

period of 2.
The DTFT tests a signal against an infinite number of

frequencies and measures how well an impulse response (or
signal) resonates with each frequency.
When the test frequency is close to the frequency of the

impulse response (or signal), the DTFT produces a strong
response. Otherwise it produces a weak response.
18
The general form of a difference equation is
or
It can be transformed into the frequency domain by taking the

DTFT of each term.
19
20
Example-3: Find the frequency response that corresponds to the

difference equation
Solution
Taking term by term DTFT
21
22
Frequency Response vs. Transfer Function
The frequency response H() is a ratio of output to input in

the Discrete Fourier frequency domain.
The transfer function H(z) is a ratio of output to input in the z

frequency domain.
23
The frequency response () and the impulse response []

are related as
So the frequency response () is the DTFT of the impulse

response [].
24
Example-4: Find an expression for the frequency response of the

filter with transfer function
Solution
we get the frequency response by replacing z with ej
25
Example-5: The impulse response for a digital filter is
Find an expression for the frequency response of the filter.
Solution
The frequency response is the DTFT of the impulse response and
can be computed as
26
27
The DTFT of the systems output Y() may be calculated from the
frequency response H() and the DTFT of the input X() as
follows:
The frequency response analyzes a systems behavior in terms of

its individual frequencies, this equation is normally only used
when the input x[n] is sinusoidal.
x[n] = Acos(nW0 + qX)
with the short form notation AqX.
28
At the frequency W= W0, the system has a gain H and a phase

difference q, that is
H(W0) = Hq
then the output from the filter is
Y(W) = (Hq AqX
Y(W) = q + qX)
In the time domain, this equates to

y[n] = HAcos(nW0 + q + qX)
Thus, when the input to a linear filter is a sinusoid, the output is

a sinusoid at the same digital frequency, but with a different
amplitude and phase. 29
30
31
The magnitude response of a filter is a collection of the gains of the

filter over all frequencies.
It is a plot of filter gain |H(W)| versus digital frequency W in radians.
The magnitude response gives the shape of the filter.
The phase response is a collection of the phase differences over all

frequencies.
It is a plot of phase difference q(W) versus digital frequency W in

radians.
Both the magnitude response and the phase response are smooth
functions of frequency.
32
The magnitude response is an even function, that is,

|H(-W)| = |H(W)|
33
The phase response is a odd function, that is,

q(-W) = -q(W)
34
The gains are expressed either in linear form, |H(W)|, or in logarithmic form
in decibels, 20log|H(W)| dB.
The phases q(W) are expressed either in degrees or radians.
It is often useful to re-express the values along the horizontal frequency

axis as analog frequencies in Hz instead of digital frequencies in radians.
The equation
can be re-arranged to give the formula needed for the conversion:
35
Sampling Frequency Unknown
Magnitude Response Phase Response
36
Sampling Frequency of 12 kHz
37
Example-6: Frequency response of a system is
Find and sketch the magnitude and phase response for the
system. Cover a range of digital frequencies from to 3.
Solution
According to the Euler Form
ej = cos +jsin
1
=
1 0.4 cos W sin W
38
1
=
1 0.4 cos W sin W
1
=
1 0.4 cos W + 0.4 sin W
When W = 0
1 1
0 = = = 1.666
1 0.4 cos 0 + 0.4 sin 0 1 0.4 1 + 0
When W = /4
1
/4 =
1 0.4 cos /4 + 0.4 sin /4
1 0.71 0.28
= = 1.24 0.49 = 1.2972 0.3757
0.71 + 0.28 0.71 0.28
39
40
Magnitude Response
41
Phase Response
42
Magnitude & Phase Reponses
(Linear & Logarithmic Scales)
43
The magnitude and phase responses are periodic, and they repeat
every 2 radians.
This periodicity holds for all DTFTs.
Both magnitude and phase responses are continuous functions

(they have values for every value of frequency).
The magnitude response is an even function. The portion to the

left of the zero is a perfect mirror image of what lies to the right of
zero, that is,
This is true for all magnitude responses. 44

The phase response is an odd function.
This is true for all phase responses.
Normally, there is no need to record the part of the magnitude and

phase responses to the left of = 0: This potion can be deduced
from the plot for > 0.
Furthermore, there is no need to consider values of above.
The frequency responses of the filters are usually plotted for values
0 .
The advantage of using decibels is that a that an extremely large

range of gains , from very large to very small, can be conveniently
plotted on same graph. 45
The filter shape for a digital filter is furnished by its magnitude response.
LOW PASS FILTER
46
HIGH PASS FILTER
47
BAND PASS FILTER
48
BAND STOP FILTER
49
Example-7: Sketch the frequency response of the system.
. W
W =
+ . W
Solution
Computing the frequency response with a step size of W = /4
50
51
52
Example-8: Find and sketch the frequency response of the filter

whose impulse response is
h[n] = [n] 1.25 [n-2]
Plot phases in degrees and linear gains.
Solution
The frequency response is given by
W = 1 0.25 2W
Magnitude and phase responses are calculated in the table.
53
54
55
Example-9: the difference equation describing a digital filter is

y[n] = 1.5y[n-1] 0.85y[n-2] + x[n]
Find and sketch the frequency response. Plot phases in degrees
and gains in decibels.
Solution
The frequency response is given by
1
W =
1 1.5 W + 0.85 W
Magnitude and phase responses are calculated in the table.
56
57
58
A filter is considered to pass signals where the gain is at least

0.707 of its maximum value.
This is equivalent to a gain of 3 dB below maximum.
In other words, when the maximum gain of a filter is 1, or 0

dB, the edge of the pass band occurs where the gain is 0.707,
or -3 dB.
The shape of a digital filter can be designed independent of

sampling frequency.
The sampling frequency chosen will, however, determine the

range of input frequency over which filter will act.
59
When sampling rate is known, the frequency axis can be

plotted in terms of analog frequency in Hertz instead of
digital frequency in radians.
It is often easier to understand filter behavior in this way.
As we know that
With this conversion, nothing in frequency response diagram

changes except labelling of the horizontal axis.
But the choice of sampling frequency has an enormous

impact on the effect of the filter.
60
Example-10: A filter with frequency response shown in the

following figure is used in a digital filter with a 12 kHz sampling
frequency.
a) Plot the frequency response against analog frequency in Hz

instead of digital frequency in radians.
b) What is the bandwidth of the low pass filter in Hz.
c) What is the bandwidth if the sampling frequency is changed to

30 kHz.
61
62
a) Plot the frequency response against analog frequency in Hz instead

of digital frequency in radians.
Digital Frequencies: 0 to radians
Analog Frequencies: 0 to Fs/2 Hz
b)
The bandwidth is the range of frequencies passed by the filter.
For a low pass filter, this corresponds to the frequency where the
gain drops by 3 dB from its low frequency value.
For the magnitude response in figure (on previous slide), this occurs
at a frequency of 1800 Hz.
c) If the sampling frequency were changed to 30khz, frequencies of

interest would range from 0 to 15 kHz, and bandwidth would
become 4500 Hz.
63
When the poles and zeros of a filter are known, the filters
shape can be estimated without computing the frequency
response first.
For a digital frequency , the smaller the distance between

ej and a zero, the smaller the gain at that frequency.
Similarly, the smaller the distance between ej and a pole, the

greater the gain at that frequency.
64
Thus a filter with a single pole at z = 0.99 has its largest gains
at low frequencies , which describes a low pass filter.
Conversely, a filter with a single pole at z = -0.99 has its largest

gains at high frequencies , which describes a high pass filter.
The closer the zeros and poles are to the unit circle, the more
selective the filter, that is, the steeper the transition between
the pass band and stop band regions.
65
Point which is moving on unit circle according to the different

values of frequency is
ej = cos +jsin
Or in ordered pair form (cos, sin)
As the frequency increases from 0 to radians, the point

(cos, sin) will travel through unit circle from (1,0) to (-1,0).
The pole P is defined as a complex number

+ j
Or in ordered pair form ( , )
66
The filters shape can be computed as
0 1
=
0 1
Where the distance between the point ej and pole P and can
be computed as
ej P = (cos +jsin) ( + j)
67
For a digital frequency between 0 and radians, the closer

ej is to a pole of a filter, and further ej is to a zero, the larger
the magnitude tends to be.
A pole close to the unit circle leads to a filter shape with an

extremely large magnitude at some frequency.
A zero close to the unit circle leads to a filter shape with a

small magnitude at some frequency.
Extremely large and small magnitudes tend to signal filters

that are very selective.
68
Example-11: Deduce the filter shape describe by the transfer function.

=
.
Solution
This filter has a single pole at z = 0.45, with no zeros, and a unity gain.
1
() =
(0.45,0)
The distance between the point ej and

pole is smaller near = 0, and largest at = radians.
Since the magnitude of the filter is reciprocal to the this

distance, the magnitude is largest near =0, and smallest at = rad. 69
70
Filter Shape
71
Example-12: Deduce the filter shape describe by the difference

equation.
= + [ ]
Solution
The transfer function of the filter is
2 + 1
=
3
This filter has three poles at z = 0

Zeros at z = j and z = -j
72
Since all poles are at origin the distances between the point
ej and the poles are the same for all values of .
Thus only the zero positions affect the overall filter shape.
A magnitude of zero will be obtained for = /2 radians.
The filter have a band stop effect as shown in the figure.
73
Example-13: Compare the filter shape for the following filters.

a) = +. +.

b) = +. +.

c) = +. +.
Solution
None of the filters have zeros.

The first filter has the poles . . , with magnitude of
0.95.
The second filter has the poles . . , magnitude of 0.9.
The third filter has the poles . . , magnitude of 0.8. 74
As the figure shows, the value of for which the poles are nearest the unit
circle is the same for all three filters.
Therefore the magnitude response should all give a spike at this value of .
The poles of first filter (a) are closest to the unit circle, therefore the spike for
this filter should be largest.
Three Pole-Zero Plots Superimposed
75

Lecture 8

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Lecture 8

Hochgeladen von

Copyright:

Verfügbare Formate

Digital Signal & Image Processing

Discrete Time Fourier Transform (DTFT)

The discrete time Fourier transform (DTFT) is a frequency

The frequency response H() of a system is composed of two

This is a consequence of the fact that the frequency response

The magnitude response gives the shape of a filter and the

The DTFT of the impulse response h[n] of the system is

Where H() is the frequency response of a system.

The discrete time Fourier transform (DTFT) of a signal [] is

where is the digital frequency (in radians) defined as

The DTFT of a signal [] is denoted as

Example-1: Find the DTFT of the signal.

Example-2: Find the DTFT of the signal.

Where = cos n sin n

Signal Resonance for the Discrete Time Fourier Transform

Signal Resonance for the Discrete Time Fourier Transform

Signal Resonance for the Discrete Time Fourier Transform

Signal Resonance for the Discrete Time Fourier Transform

The two important properties of DTFT are

1. Time delay or Time shifting

1. Time delay or Time shifting

A delay of M steps in the time domain appear in the DTFT

For example, when delay M = 1

This delay property of DTFT also makes conversion from a

y[n] + a1y[n-1] + a2y[n-2] + = b0x[n] + b1x[n-1] + b2x[n-2] +

The DTFT is periodic with a period of 2.

So the magnitude and phase responses also repeat with the

The DTFT tests a signal against an infinite number of

When the test frequency is close to the frequency of the

The general form of a difference equation is

It can be transformed into the frequency domain by taking the

Example-3: Find the frequency response that corresponds to the

Taking term by term DTFT

The frequency response H() is a ratio of output to input in

The transfer function H(z) is a ratio of output to input in the z

The frequency response () and the impulse response []

So the frequency response () is the DTFT of the impulse

Example-4: Find an expression for the frequency response of the

we get the frequency response by replacing z with ej

Example-5: The impulse response for a digital filter is

Find an expression for the frequency response of the filter.

The frequency response analyzes a systems behavior in terms of

x[n] = Acos(nW0 + qX)

with the short form notation AqX.

At the frequency W= W0, the system has a gain H and a phase

In the time domain, this equates to

Thus, when the input to a linear filter is a sinusoid, the output is

The magnitude response of a filter is a collection of the gains of the

It is a plot of filter gain |H(W)| versus digital frequency W in radians.

The magnitude response gives the shape of the filter.

The phase response is a collection of the phase differences over all

It is a plot of phase difference q(W) versus digital frequency W in

The magnitude response is an even function, that is,

The phase response is a odd function, that is,

The phases q(W) are expressed either in degrees or radians.

It is often useful to re-express the values along the horizontal frequency

can be re-arranged to give the formula needed for the conversion:

Sampling Frequency Unknown

Magnitude Response Phase Response

Sampling Frequency of 12 kHz

Magnitude Response Phase Response

Example-6: Frequency response of a system is

This periodicity holds for all DTFTs.