Aarti Bang,
VIIT, Pune
DFT
ond
SpecfroI AnoIysis of SiqnoIs
Mrs.Aarti Bang,
VIIT, Pune
Confinuous Time Cose Fourier Trunsform
dt e t h F H
Ft j
2
) ( ) ( dF e F H t h
Ft j
2
) ( ) (
Mrs.Aarti Bang,
VIIT, Pune
Discrefe Time Cose Discrefe Time Fourier
Tronsform
n
n j
e n h H
) ( ) (
d e H n h
n j
) (
2
1
) (
Mrs.Aarti Bang,
VIIT, Pune
Discrefe Fourier Tronsform
1
0
/ 2
) ( ) (
N
n
N nk j
e n h k H
1 ..... 2 , 1 , 0 N k
1
0
/ 2
) (
1
) (
N
k
N nk j
e k H
N
n h
DFT
IDFT
1 ..... 2 , 1 , 0 N n
Mrs.Aarti Bang,
VIIT, Pune
e.q.
F~I0 H; Fs ~ o4 H; M ~ 3Z
0 5 1 0 1 5 2 0 2 5 3 0 3 5
0
2
4
G
8
1 0
1 2
1 4
1 G


X
[



x(n) ~ sin(Z*pi*n*I0/o4)
Mrs.Aarti Bang,
VIIT, Pune
n
n j
e n h H
) ( ) (
1
0
/ 2
) ( ) (
N
n
N nk j
e n h k H
N
k
N
k
Fs
F
k
2 2
N
kF
F
s
k
Mrs.Aarti Bang,
VIIT, Pune
F~II H; Fs ~ o4 H; M ~ 3Z
0 5 10 15 20 25 30 35
0
2
4
G
8
10
12


X
[



This phenomenon is coIIed
Mrs.Aarti Bang,
VIIT, Pune
The phenomenon of fhe spreod of enerqy
from o sinqIe frequency fo mony
frequency Iocofions is coIIed Ieokoqe.
Mrs.Aarti Bang,
VIIT, Pune
F~II H; Fs ~ o4 H; M ~ o4
0 10 20 30 40 50 G0 70
0
5
10
15
20
25
30
35


X
[



Mrs.Aarti Bang,
VIIT, Pune
Windowinq
x(n) ~ x(n) x w(n)
where
)
1 2
2
cos( 46 . 0 54 . 0 ) (
N
n
n w
n ~ 0I, MI
Mrs.Aarti Bang,
VIIT, Pune
0 10 20 30 40 50 G0 70
1
0.8
0.G
0.4
0.2
0
0.2
0.4
0.G
0.8
1
input signal
0 5 10 15 20 25 30 35
1
0.8
0.G
0.4
0.2
0
0.2
0.4
0.G
0.8
1
input signal
Mrs.Aarti Bang,
VIIT, Pune
0 10 20 30 40 50 G0 70
0
0.1
0.2
0.3
0.4
0.5
0.G
0.7
0.8
0.9
1
hamming window
0 5 10 15 20 25 30 35
0
0.1
0.2
0.3
0.4
0.5
0.G
0.7
0.8
0.9
1
hamming window
Mrs.Aarti Bang,
VIIT, Pune
0 10 20 30 40 50 G0 70
1
0.8
0.G
0.4
0.2
0
0.2
0.4
0.G
0.8
1
windowed signal
0 5 10 15 20 25 30 35
1
0.8
0.G
0.4
0.2
0
0.2
0.4
0.G
0.8
1
windowed signal
Mrs.Aarti Bang,
VIIT, Pune
0 5 10 15 20 25 30 35
0
1
2
3
4
5
G
7


X
[



32 pt. DFT of windowed signal
Mrs.Aarti Bang,
VIIT, Pune
0 5 10 15 20 25 30
0
2
4
G
8
10
12
14
1G
F

X
[
F


DFT
e.q.
F~I0 H; Fs ~ o4 H; M ~ 3Z
Mrs.Aarti Bang,
VIIT, Pune
PeIofionship of DFT wifh Z  Tronsform
n
n
z n x z X ) ( ) (
Sample X(z) at N equally spaced points
on unit circle
N nk j
k
e z
/ 2
k = 0,1,..,N
j
re z
On unit circle
j
e z
n
n j
e n x z X
) ( ) (
1
0
/ 2
) ( ) (
N
n
N nk j
e n x z X
Mrs.Aarti Bang,
VIIT, Pune
Mrs.Aarti Bang,
VIIT, Pune
Controlling Gain by PoleZero Placement
Frequency response can be
controlled by assigning poles and
zeros at specific locations.
To enhance the gain at a
frequency, we assign a pole at that
frequency.
To decrease the gain at a
frequency, we assign a zero at that
frequency
Mrs.Aarti Bang,
VIIT, Pune
Spectral Analysis of Sinusoidal signals
or
Deterministic Signals
or
Stationary Signal
Signal length or Window length N
Effect of:
DFT length R
Window function
x(n) = 0.5sin(2*pi*n*f1) + sin(2*pi*n*f2)
f1 = 0.22 f2 = 0.34
Mrs.Aarti Bang,
VIIT, Pune
Effect of increasing length of DFT R by zero padding
Window function : Rectangular
Difficult to determine
whether there is one or
more sinusoids and their
exact locations.
0 5 10 15
0
1
2
3
4
5
G


X
[



N = 1G, R=1G
Mrs.Aarti Bang,
VIIT, Pune
0 5 10 15 20 25 30 35
0
1
2
3
4
5
G
7
8


X
[



N= 1G, R=32
Size of DFT is increased
(by zero padding)
There is some concentration
Around k=7, & k=11
Mrs.Aarti Bang,
VIIT, Pune
0 20 40 G0 80 100 120 140
0
1
2
3
4
5
G
7
8


X
[



N = 1G, R=128
Peaks around k=27 & k=45
No. of minor peaks
Not clear whether additional
sinusoids of lesser strengths
are present
I
Increasing R improves sampling accuracy of
DTFT reducing spectral separation of adjacent
DTFT samples
Mrs.Aarti Bang,
VIIT, Pune
Frequency Resolution Property
Minimization of leakage using a tapered window
x(n) = 0.85sin(2*pi*n*f1) + sin(2*pi*n*f2)
f1 = 0.22 f2 = 0.26
Mrs.Aarti Bang,
VIIT, Pune
0 2 4 G 8 1 0 1 2 1 4 1 G
0
0 . 5
1
1 . 5
2
2 . 5
3
3 . 5
4
4 . 5


X
[



N = 1 G , R= 1 G
Leakage is minimized
due to windowing
Mrs.Aarti Bang,
VIIT, Pune
0 1 0 2 0 3 0 4 0 5 0 G 0 7 0
0
0 . 5
1
1 . 5
2
2 . 5
3
3 . 5
4
4 . 5


X
[



N = 1 G , R= G 4
Increasing R,
No change in resolution
Mrs.Aarti Bang,
VIIT, Pune
0 20 40 G0 80 100 120 140
0
2
4
G
8
10
12
14
1G
18


X
[



N = G4, R=128
Increasing the window length improves the frequency resolution
i.e. it can resolve two closer frequencies.
Mrs.Aarti Bang,
VIIT, Pune
Hamming window:
Width of main lobe : 8*pi/N
We can resolve two frequencies, if their difference is
close to (greater than) half the main lobe width
i.e. 8*pi/2*N
Frequency difference is 0.04
For N=16, it is 0.125 (8*pi)/(2*pi*N*2)
For N=64, it is 0.03125
Mrs.Aarti Bang,
VIIT, Pune
0 0.5 1 1.5 2 2.5 3 3.5
80
G0
40
20
0
20
40
Frequency response of Hanning Window
Main lobe
Side lobes
Mrs.Aarti Bang,
VIIT, Pune
NonStationary signals
Signals that change with time.
Mrs.Aarti Bang,
VIIT, Pune
Stationary NonStationary
0 50 100 150 200 250 300
4
3
2
1
0
1
2
3
4
x = 1 +4 +8 +1G
0 50 100 150 200 250 300
1
0.8
0.G
0.4
0.2
0
0.2
0.4
0.G
0.8
1
1 4 8 1G
Mrs.Aarti Bang,
VIIT, Pune
0 50 100 150 200 250 300
0
20
40
G0
80
100
120
140
dft of stationary signal
0 50 100 150 200 250 300
0
5
10
15
20
25
30
35
dft of nonstationary signal
Mrs.Aarti Bang,
VIIT, Pune
Spectral Analysis of Random Signals
Estimate Power Density Spectrum
Periodogram Method
Signal x(n) sectioned into smaller blocks
Window w(n)
Modified blocks g(n) = x(n) x w(n)
Compute DFT of each modified block
Take square of magnitudes ( Periodogram)
Take average of all periodograms
Much more than documents.
Discover everything Scribd has to offer, including books and audiobooks from major publishers.
Cancel anytime.