0% found this document useful (0 votes)
101 views

DFT

The document discusses the discrete Fourier transform (DFT). It begins by introducing periodic sequences and their representation using the discrete Fourier series (DFS). The DFS decomposes a periodic sequence into a sum of complex exponentials called harmonics. It then describes properties of the DFS such as linearity, how shifts in time or frequency affect the Fourier coefficients, and how periodic convolution can be performed using the DFS. Next, it discusses how the Fourier transform represents periodic signals as a sum of Dirac deltas. Finally, it introduces the DFT and how it provides a sampled representation of the continuous-time Fourier transform of periodic signals.
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
101 views

DFT

The document discusses the discrete Fourier transform (DFT). It begins by introducing periodic sequences and their representation using the discrete Fourier series (DFS). The DFS decomposes a periodic sequence into a sum of complex exponentials called harmonics. It then describes properties of the DFS such as linearity, how shifts in time or frequency affect the Fourier coefficients, and how periodic convolution can be performed using the DFS. Next, it discusses how the Fourier transform represents periodic signals as a sum of Dirac deltas. Finally, it introduces the DFT and how it provides a sampled representation of the continuous-time Fourier transform of periodic signals.
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 69

The Discrete Fourier Transform

Content

Introduction
Representation of Periodic Sequences

Properties of DFS
The Fourier Transform of Periodic Signals
Sampling of Fourier Transform
Representation of Finite-Duration Sequences

DFS (Discrete Fourier Series)

DFT (Discrete Fourier Transform)

Properties of the DFT


Linear Convolution Using the DFT

The Discrete Fourier Transform

Introduction

Signal Processing Methods


Periodicity
Periodic
Aperiodic
Continuous

Fourier
FourierSeries
Series

Continuous-Time
Continuous-Time
Fourier
FourierTransform
Transform

DFS
DFS

Discrete-Time
Discrete-Time
Fourier
FourierTransform
Transform
and
andz-Transform
z-Transform

Time
Discrete

Infinite
Finite
Duration

Frequency-Domain Properties
Periodicity
Periodic
Aperiodic
Continuous
Time
Discrete

Discrete
Discrete
Fourier
&
Fourier
Series
&Series
Aperiodic
Aperiodic

Continuous
Continuous
Continuous-Time
Continuous-Time
&
&
Fourier
Transform
Fourier
Transform
Aperiodic
Aperiodic

Discrete-Time
Discrete-Time
Continuous
Continuous
Fourier
&
FourierTransform
Transform
&
Periodic
(2)
and
z-Transform
Periodic
(2)
and
z-Transform

DFS
DFS

Infinite
Finite
Duration

Frequency-Domain Properties
Periodicity
Periodic
Aperiodic
Continuous
Time

Continuous
Discrete
Continuous
Continuous-Time
Discrete
Continuous-Time
Relation?
Fourier
&
&
Fourier
Series
&
&Series
Fourier
Transform
Fourier
Transform
Aperiodic
Aperiodic
Aperiodic
Aperiodic
Relation?

Discrete

DFT
DFT

Relation?

Discrete-Time
Discrete-Time
Continuous
Continuous
Relation?
Fourier
&
FourierTransform
Transform
&
Periodic
(2)
and
z-Transform
Periodic
(2)
and
z-Transform

Infinite
Finite
Duration

Frequency-Domain Properties
Periodicity
Periodic
Aperiodic
Continuous
Time
Discrete

Discrete
Discrete
Fourier
&
Fourier
Series
&Series
Aperiodic
Aperiodic

Continuous
Continuous
Continuous-Time
Continuous-Time
&
&
Fourier
Transform
Fourier
Transform
Aperiodic
Aperiodic

Discrete
Discrete
DFT
&
DFT
&
Periodic
Periodic

Discrete-Time
Discrete-Time
Continuous
Continuous
Fourier
&
FourierTransform
Transform
&
Periodic
(2)
and
z-Transform
Periodic
(2)
and
z-Transform

Infinite
Finite
Duration

The Discrete Fourier Transform

Representation of
Periodic Sequences --DFS

Periodic Sequences
Notation:

a sequence with period N

~
~
~
~
xx((nn)) xx((nn rN
rN))
where r is any integer.

Harmonics
eekk ((nn)) ee

jj((22//NN))kn
kn

k 0,1,2,

Facts:

ek (n) ek (n rN )

Each has Periodic N.

e0 (n) eN (n), e1 (n) eN 1 (n)


N distinct harmonics e0(n), e1(n),, eN1(n).

~
~
DFS
~
DFS
~
xx((nn))

XX((kk))

Synthesis and Analysis


jj((22//NN))

Notation W
WNN ee

NN11

~
~
kn
~
kn
~
x
(
n
)

X
(
k
)
W
Synthesis x (n)
X (k )WNN
kk00

NN11

1
1
kn
kn
%
%
%
X
(
k
)

x
(
n
)
W
x%
(n)WNN
Analysis X (k )

N
N nn00

Both have
Period N

Example

A periodic impulse train


~
~
xx((nn))
rN)) with period N.
((nnrN
rr

N 1
~
X (k ) ~
x (n)WNkn 1
n 0

1
~
x ( n)
N

N 1

W
k 0

kn
N

N 1

e
k 0

j ( 2 / N ) kn

Example
0 1 2 3 4 5 6 7 8 9
5k
1

W
~
10
X (k ) W10kn
k
1 W10
n 0
j ( 4 k / 10 ) sin( k / 2)
e
sin( k / 10)
4

Example
0 1 2 3 4 5 6 7 8 9
5k
1

W
~
10
X (k ) W10kn
k
1 W10
n 0
j ( 4 k / 10 ) sin( k / 2)
e
sin( k / 10)
4

Example
0 1 2 3 4 5 6 7 8 9
5k
1

W
~
10
X (k ) W10kn
k
1 W10
n 0
j ( 4 k / 10 ) sin( k / 2)
e
sin( k / 10)
4

DFS vs. FT
~
x~
x((nn))
N

xx(n
(n))

X (e ) x ( n )e
j

jn

n 0

N 1

x ( n )e
n 0

jn

N 1
~
X (k ) ~
x (n)e j ( 2 / N ) kn
n 0

~
j
X ( k ) X (e )

2 k / N

Example
~
x~
x((nn))
0 1 2 3 4 5 6 7 8 9

0 1 2 3 4 5 6 7 8 9

xx(n
(n))
4

X ( e j ) e j n
n 0

~
X ( k ) X ( e j )

1 e j 5
j 2 sin(5 / 2)

e
1 e j
sin( / 2)

2 k / 10

j ( 4 k / 10 )

sin( k / 2)
sin( k / 10)

Example
~
x~
x((nn))
0 1 2 3 4 5 6 7 8 9

0 1 2 3 4 5 6 7 8 9

xx(n
(n))
4

X ( e j ) e j n
n 0

~
X ( k ) X ( e j )

1 e j 5
j 2 sin(5 / 2)

e
1 e j
sin( / 2)

2 k / 10

j ( 4 k / 10 )

sin( k / 2)
sin( k / 10)

Example
~
x~
x((nn))
0 1 2 3 4 5 6 7 8 9

0 1 2 3 4 5 6 7 8 9

xx(n
(n))
4

X ( e j ) e j n
n 0

~
X ( k ) X ( e j )

1 e j 5
j 2 sin(5 / 2)

e
1 e j
sin( / 2)

2 k / 10

j ( 4 k / 10 )

sin( k / 2)
sin( k / 10)

The Discrete Fourier Transform

Properties of
DFS

Linearity
~~
DFS
~
DFS
~
xx11((nn))

XX11((kk))
~~
DFS
~
DFS
~
xx22((nn))

XX22((kk))
~
~
DFS
~
~
~
~
DFS
~
~
aaxx11((nn))bbxx22((nn))

aaXX11((kk))bbXX22((kk))

Shift of A Sequence
~
~
DFS
~
DFS
~
xx((nn))

XX ((kk))
~
~
DFS
km
~
DFS
km
~
xx((nn m
m))

W
WNN XX ((kk))
Change Phase
(delay)
DFS
km
DFS
km
NN

~
~
~
~
xx[[nn ((m
m lN
lN)]
)]

W
W XX ((kk))

Shift of Fourier Coefficient


~
~
DFS
~
DFS
~
xx((nn))

XX ((kk))
~
~
DFS
~
DFS
~
xx((nn))

XX ((kk ll))

nl
nl
NN

W
W

Modulation

Duality
~
~
DFS
~
DFS
~
xx((nn))

XX ((kk))
~~
DFS
~
DFS
~
XX ((nn))

N
x
Nx((kk))

Periodic Convolution
~
DFS
~
~
DFS
~
xx11((nn))

XX11((kk))
~
DFS
~
~
DFS
~
xx22((nn))

XX22((kk))
NN11

Both have
Period N

~~
~~
DFS
~
~
DFS
~
~

XX11((kk))XX22((kk))

xx11((nn))xx22((nnmm))

mm00

Periodic Convolution
~
DFS
~
~
DFS
~
xx11((nn))

XX11((kk))
~
DFS
~
~
DFS
~
xx22((nn))

XX22((kk))

Both have
Period N

NN11

11
~~ ~~
DFS
~
~
DFS
~
~
xx11((nn))xx22((nn))
XX11((ll))XX22((kk ll))

N
N ll00

The Discrete Fourier Transform

The Fourier Transform of


Periodic Signals

Fourier Transforms of Periodic Signals


Periodicity
Periodic
Aperiodic
Continuous
Time
Discrete

Continuous
Discrete
Continuous
Continuous-Time
Discrete
Continuous-Time
Fourier
&
&
Fourier
SeriesSampling
&
&Series
Fourier
Transform
Fourier
Transform
Aperiodic
Aperiodic
Aperiodic
Aperiodic
Discrete-Time
Discrete-Time
Continuous
Continuous
Sampling
Fourier
&
FourierTransform
Transform
&
Periodic
(2)
and
z-Transform
Periodic
(2)
and
z-Transform

DFS
DFT
DFT

Infinite
Finite
Duration

~
DFS
~
~
DFS
~
xx((nn))

XX((kk))

~~
j
XX((kk)) XX((ee j))
2 k / N

2 k / N

Fourier Transforms of Periodic Signals


~
x~
x((nn))
N

xx(n
(n))
n

~ j
FT
~
x (n) X (e )

2 ~
2k

X (k )

k N

The Discrete Fourier Transform

Sampling of
Fourier Transform

~~
j
XX((kk)) XX((ee j))
2 k / N

2 k / N

Equal Space Sampling of Fourier Transform

xx(n
(n))
0

N1

jj

xx((nn))

XX((ee )) X ( z ) z e j
FT
FT

z-plane

z-plane

~
X (k ) X ( z ) j ( 2 / N ) k
z e
~
x ( n) ?

N >
= N

~~
j
XX((kk)) XX((ee j))
2 k / N

11 NN11 kk( n( nmm) )


W

((nnmmrN

WNN
rN))

NN kk00
r r

2 k / N

Equal Space Sampling of Fourier Transform

~
X (k ) X ( z )
1
~
x ( n)
N

z e j ( 2 / N ) k

N 1

X (e j ( 2 / N ) k )

1
~
kn
X (k )WN

N
k 0
N 1

x(m)e
k 0

N 1

j (2 / N )k
kn
X
(
e
)
W

N
k 0

j ( 2 / N ) km

x(m) (n m rN )

kn
N

1
x(m)
m
N

N 1

W
k 0

x(n) * ( n rN )
r

k ( nm)
N

Equal Space Sampling of Fourier Transform

~~
XX((kk)) XX((zz)) j j((22/ /NN))kk
z e
z e

x%
((nn)) xx((nn))**
x%
rN))
((nnrN
rr

z-plane

Example
xx(n
(n))
~
x~
x((nn))
~
x~
x((nn))

N=9
n

N=12
0

12

N=7
0

12

Example
xx(n
(n))
~
x~
x((nn))
~
x~
x((nn))

N=9
n

N=12

Time-Domain
Aliasing
Time-Domain
Aliasing
0
8
12

N=7
0

12

Time-Domain Aliasing vs.


Frequency-Domain Aliasing
To avoid

Signal is bandlimited
Sampling rate in time-domain is high enough

To avoid

frequency-domain aliasing

time-domain aliasing

Sequence is finite
Sampling interval (2/N) in frequencydomain is small enough

DFT vs. DFS


Use

DFS to represent a finite-length sequence


is call the DFT (Discrete Fourier Transform).
So, we represent the finite-duration sequence
by a periodic sequence. One period of which is
the finite-duration sequence that we wish to
represent.

xx((nn))

~
x~
x((nn))
00

00 nn NN 11
otherwise
otherwise

The Discrete Fourier Transform

Representation of
Finite-Duration Sequences --DFT

Definition of DFT
~
DFS
~
~
DFS
~
xx((nn))

XX((kk))
NN11

~~
kn
~
kn
~
xx((nn))
X
(
k
)
W
X
(
k
)
W
N

N
kk00

NN11

~~
kn
~
kn
~
XX((kk))
x
(
n
)
W
x
(
n
)
W
N

N
nn00

xx((nn))

XX((kk))
DFT
DFT

NN11
knkn
XX((kk)W
00nnNN11
NN
)
W

x
(
n
)

Synthesis x(n) kk00


00
otherwise
otherwise

NN11
knkn

x
(
n
)
W
00nnNN11
NN
x
(
n
)
W

Analysis XX((kk)) nn00


00
otherwise
otherwise

Example
xx(n
(n))
~
x~
x((nn),),NN 55
~
XX~((kk))

XX(k
(k))

Example
xx(n
(n))
~
x~
x((nn),),NN 10
10
| |XX((kk))| |

XX(k
(k))

The Discrete Fourier Transform

Properties of the DFT

Linearity
xx11((nn))

Duration N1
0

N11

xx22((nn))

Duration N2
0

DFT
x
(
n
)


NN max(
XX11((kk))
max(NN11,,NN22)) x11 (n) DFT

N21

DFT
xx22((nn))


DFT
XX22((kk))
DFT
ax
(
n
)

bx
(
n
)


ax11 (n) bx22 (n) DFT
aX
aX11((kk))bX
bX22((kk))

~
x~
x11((nn)) xx((((nnmm))))NN 00nn NN 11
xx11((nn))
otherwise
otherwise
00

Circular Shift of a Sequence


xx(n
(n))
0

~
x~
x((nn))

~
~ ( n m)
x~
x11((nn))x~
x ( n m)

~
x~
x11((nn)) xx((((nnmm))))NN 00nn NN 11
xx11((nn))
otherwise
otherwise
00

Circular Shift of a Sequence


DFT
DFT
xx((nn))


XX((kk))

j j((22/ /NN))mm

xx((((nnm
m))))NN,, 00 nn NN 11

ee
DFT
DFT

XX((kk))

Duality
DFT
DFT
xx((nn))


XX((kk))

XX((nn))

Nx
Nx((((kk))))NN,, 00 kk NN 11
DFT
DFT

Example
Choose
ChooseN=10
N=10

Re[x
Re[X(n)]
Re[x1(n)]=
1(n)]= Re[X(n)]

Re[X(k)]
Re[X(k)]

Im[x
Im[X(n)]
Im[x1(n)]=
1(n)]= Im[X(n)]

Im[X(k)]
Im[X(k)]

XX1(k)
= 10x((k))10
1(k) = 10x((k))10

Linear Convolution (Review)

xx3 ((nn))
xx11((mm))xx((nnmm))xx11((nn))**xx22((nn))
3
mm

FT
FT X (e j
j)
xx1 ((nn))

X
(
e
)
1
1

FT
FT X (e j
j)
xx2 ((nn))

X
(
e
)
2
2

FT
FT X (e j
j) X (e j
j) X (e j
j)
xx3 ((nn))xx1 ((nn))**xx2 ((nn))

3
1
2

X
(
e
)

X
(
e
)
X
(
e
)
3
1
2
3
1
2

Circular Convolution
DFT
DFT
xx1 ((nn))

XX11((kk))
1

DFT
DFT
xx2 ((nn))

XX22((kk))
2

both of length N
NN11

xx3 ((nn))
xx11((mm))xx((((nnmm))))NN xx11((nn))xx22((nn))
3
mm00

DFT
DFT
xx3 ((nn))xx1 ((nn))
x
(
n
)

x
(
n
)

XX33((kk)) XX11((kk))XX22((kk))
3
1
2

nn00=2,
=2, N=5
N=5

Example
xx1 ((nn))((nnnn0 ))
1
0

n0

xx2 ((nn))
2
0

xx1 ((nn))**xx2 ((nn))


1
2
xx1 ((nn))
xx22((nn))
1

nn00=2,
=2, N=7
N=7

Example
xx1 ((nn))((nnnn0 ))
1
0

n0

xx2 ((nn))
2
0

xx1 ((nn))**xx2 ((nn))


1
2
xx1 ((nn))
xx22((nn))
1

L=N=6
L=N=6

Example
11 00nnLL11
xx1 ((nn))xx2 ((nn))
1
2
00 otherwise
otherwise

xx1 ((nn))
1
0

N 1
N 1

kn
kn
XX1 ((kk))XX2 ((kk)) W
N
W
N
1
2

xx2 ((nn))
2

n 0
n 0

NN kk00

00 otherwise
otherwise

xx3 ((nn))xx1 ((nn))


xx22((nn))
3
1
NN22 kk00
XX3 ((kk))XX1 ((kk))XX2 ((kk))
3
1
2
00 otherwise
otherwise

xx3 ((nn))
3
0

L=2N=12
L=2N=12

Example
11 00nnLL11
xx1 ((nn))xx2 ((nn))
1
2
00 otherwise
otherwise
L 1
L 1

kn
kn
XX1 ((kk))XX2 ((kk)) W
2L
W

1
2
2L

xx1 ((nn))
1
0

xx2 ((nn))
2

n 0
n 0

k ( L 1) 1
k ( L 1) 1
11W
N
W
N
00kk22LL11

1 W2kLk
1 W2 L
0
otherwise
otherwise
0

xx3 ((nn))xx1 ((nn))


xx22((nn))
3
1

N
xx3 ((nn))
3
0

1 W k k( L( L1)1)11 22
1 W22LL k 00kk22LL11
XX3 ((kk))XX1 ((kk))XX2 ((kk)) 1 W k

3
1
2
1 W22LL
0
otherwise
0
otherwise

The Discrete Fourier Transform

Linear Convolution
Using the DFT

Why Using DFT for Linear Convolution?


FFT

But.,

(Fast Fourier Transform) exists.

we have to ensure that circular convolving


nature of DFT gives the linear convolving
result.

The Procedure
1.

Let x1(n) and x2(n) be two sequences of length L


and P, respectively.
Compute N-point (N = ?) DFTs X1(k) and X2(k).

2.

Let X3(k) = X1(k) X2(k), 0 k N1.

3.

Let x3(n) = DFT1[X3(k)] = x1(n) x2(n).

xx11(n)
(n) ** xx22(n)
(n) =
= xx11(n)
(n)
xx22(n)?
(n)?

xx3 ((nn))xx1 ((nn))**xx2 ((nn)) xx1 ((mm))xx2 ((mmnn))


1 2
3
1
2
mm

Linear Convolution of
Two Finite-Length Sequences
x2(m)
0

x1(m)
0

x3(1) = 0
P

1 0

m
x2(1 m)
m

x3(n) 0
n = 0, 1, , L+P 2

x2(n m)
0

nP+1

L n

x3(L+P1) = 0
L

m
x2(L+P1 m)
n+P1

xx3 ((nn))xx1 ((nn))**xx2 ((nn)) xx1 ((mm))xx2 ((mmnn))


1 2
3
1
2
mm

Linear Convolution of
Two
Finite-Length
Sequences
Length
Length of
of xx33(n)
(n) == xx11(n)*x
(n)*x22(n)
(n) == L+P1
L+P1
x2(m)
0

x1(m)
0

x3(1) = 0
P

1 0

m
x2(1 m)
m

x3(n) 0
n = 0, 1, , L+P 2

x2(n m)
0

nP+1

L n

x3(L+P1) = 0
L

m
x2(L+P1 m)
n+P1

Circular Convolution as Linear


Convolution with Time Aliasing
x1(n) Length L
x2(n) Length P

FT
j
FT X (e j)
xx1 ((nn))

X11 (e )
1
FT
j
FT X (e j)
xx2 ((nn))

2
X 2 (e )
2

x(n) = x1(n)*x2(n) Length L+P1


X (e j ( 2 k / N ) ) 0 k N 1

Let X (k )

otherwise

x ( n) 0 n N 1
~
Let x p (n)
otherwise
0

FT
j
j
j
FT X (e j) X (e j) X (e j)
xx((nn))

1
2
X (e ) X 1 (e ) X 2 (e )

~
x ( n)

x(n rN )

xxpp((nn)) xx((nn))??

Circular Convolution as Linear


Convolution with Time Aliasing
x1(n)
0

x2(n)
0

n
n

x (n)

L+P1

N = L+P1
x (n)

L+P1

~
x ( n)

L+P1

~
x~
xpp((nn)) xx((nn))
0

L+P1

For Finite Sequences


x1(n)
0

x2(n)

Zero padding to length L+P1


n

Zero padding to length L+P1


0

x p(n) = x1(n)*x2(n)
= x1(n)x2(n), 0 n L+P2

L+P2

N=L

xx((nn))xx((nnLL))

~
~
xxpp((nn)) xx((nn))
0
0

00nn PP22
PP11nn LL
otherwise
otherwise

x (n)

P1

P1

L+P1

P1

L+P1

L+P1

~
x ( n)

N=L

xx((nn))xx((nnLL))

~
~
xxpp((nn)) xx((nn))
0
0

00nn PP22
PP11nn LL
otherwise
otherwise

x (n)

P1

P1

Corrupted
(P1) points

L+P1

~
x ( n)

L+P1

Uncorrupted
(LP+1) points
0

P1

L+P1

FIR Filter for Indefinite-Length Signals


h (n)
x (n)

Overlap-Add

Method
Overlap-Save Method

Block
Convolution

Overlay-Add Method
h (n)
x (n)

x0 (n)

x1 (n)

xx((nnrL
rL)) 00nn LL11
xxrr((nn))
00
otherwise
otherwise

x2 ( n )

xx((nn))
xxrr((nnrLrL))
rr00

Overlay-Add Method

yy((nn)) xx((nn))**hh((nn))
yyrr((nnrLrL))
rr00

yyrr((nn)) xxrr((nn))**hh((nn))

xx((nnrL
rL)) 00nn LL11
xxrr((nn))
00
otherwise
otherwise

xx((nn))
xxrr((nnrLrL))
rr00

Overlay-Add Method
Set N = L+P1
for each block convolution

Overlay-Save Method

Each block is of length L.


P1 samples are overlaid
btw. adjacent blocks.
Set N = L+P1 for each block
convolution.
Save the last LP+1 values
for each block convolution.

You might also like