ch8dtft dft fs
DESCRIPTION
DTFT and DFTTRANSCRIPT
o Introduction
o DTFS & Properties
FT of periodic signals
ELEC442: DSP
DTFS, DTFT, DFS, DFT, FFT
o FT of periodic signals
o DFT & Properties: Sampling of the DTFT
o DTFT, DTFS, DFT, DFS, FFT, ZT: numerical
o Summary Dr. Aishy Amer
Concordia University
Electrical and Computer Engineering
Figures and examples in these course slides are taken from the following sources:
•A. Oppenheim, A.S. Willsky and S.H. Nawab, Signals and Systems, 2nd Edition, Prentice-Hall, 1997
•A.V. Oppenheim, R.W. Schafer and J.R. Buck, Discrete-Time Signal Processing
•M.J. Roberts, Signals and Systems, McGraw Hill, 2004
•J. McClellan, R. Schafer, M. Yoder, Signal Processing First, Prentice Hall, 2003
•Slides 2-22 are from http://metalab.uniten.edu.my/~zainul/images/Signals&Systems
Periodic DT Signals
A DT signal is periodic with period
where is a positive integer if
The fundamental period of is the
2
The fundamental period of is the
smallest positive value of for which the
equation holds
Example:
is periodic with fundamental period
Fourier representation of
signals The study of signals and systems using sinusoidal
representations is termed Fourier analysis, after Joseph Fourier (1768-1830)
The development of Fourier analysis has a long history involving a great many individuals and the investigation of
3
involving a great many individuals and the investigation of many different physical phenomena, such as the motion of a vibrating string, the phenomenon of heat propagation and diffusion
Fourier methods have widespread application beyond signals and systems, being used in every branch of engineering and science
The theory of integration, point-set topology, and eigenfunction expansions are just a few examples of topics in mathematics that have their roots in the analysis of Fourier series and integrals
Approximation of Signals by
Sinusoids
A signal can be approximated by a sum of many sinusoids at harmonic frequencies of the signal f0with appropriate amplitude and phase
The more harmonic components are added, the
4
The more harmonic components are added, the more accurate the approximation becomes
Instead of using sinusoidal signals, mathematically, we can use the complex exponential functions with both positive and negative harmonic frequencies
A Fourier representation is unique, i.e., no two same signals in time domain give the same function in frequency domain
Overview of Fourier Analysis
Methods
Periodic in Time
Discrete in Frequency
Aperiodic in Time
Continuous in Frequency
Continuous
in Time∫
−=
⇒⊗
T
tjk
k dtetxT
a 0)(1
DTP-CT :SeriesFourier CT
0
T
ω ∫∞
∞−
−
⇒⊗
=
⇒⊗
ω ωdtetxjX
tj
CTCT :TransformFourier CT Inverse
)()(
CTCT :TransformFourier CT
5
Aperiodic in
Frequency
Discrete in
Time
Periodic in
Frequency
∑∞
−∞=
=
⇒⊗
k
tjk
keatx 0)(
P-CTDT :SeriesFourier Inverse CT T
0
ω
∫
∑
=
⇒+⊗
=
+⇒⊗
∞
−∞=
−
π
ωω
π
ωω
π
ωπ
2
2
2
)(2
1][
DT PCT :TransformFourier DT Inverse
][)(
PCTDT :TransformFourier DT
deeXnx
enxeX
njj
n
njj
∑
∑
−
=
−
=
−
=
⇒⊗
=
⇒⊗
1
0
NN
1
0
NN
0
0
][1
][
P-DTP-DT SeriesFourier DT Inverse
][][
P-DTP-DT SeriesFourier DT
N
k
knj
N
n
knj
ekXN
nx
enxkX
ω
ω
∫∞
∞−
=
⇒⊗
ωωπ
ωdejXtx
tj)(2
1)(
CTCT :TransformFourier CT Inverse
Overview of Fourier Analysis
Methods
Variable Period Continuous
Frequency
Discrete
Frequency
DT x[n] n N kNkk /2πω =
ω
6
CT x(t) t T k
Nkk /2πω =
Tkk /2πω =
Ω
• DT-FS: Discrete in time; Periodic in time; Discrete in Frequency; Periodic in Frequency
• DT-FT: Discrete in time; Aperiodic in time; Continous in Frequency; Periodic in Frequency
• CT-FS: Continuous in time; Periodic in time; Discrete in Frequency; Aperiodic in Frequency
• CT-FT: Continuous in time; Aperiodic in time; Continous in Frequency; Aperiodic in Frequency
Negative frequency?
7
Negative Frequency?
8
Negative Frequency?
9
Outline
o Introduction
o DTFS & properties
o DTFT of periodic signals
o DFT: Sampling of the DTFT
10
o DFT: Sampling of the DTFT
o DTFT, DTFS, DFT, DFS, FFT, ZT: numerical (Matlab)
o Summary
Discrete-Time Fourier Series (DTFS)
Given a periodic sequence with period N so that
The FS can be written as
(Recall: the FS of continuous-time periodic signals require infinite many complex exponentials)
Not that for DT periodic signals we have
]n[x~
]rNn[x~
]n[x~
+=
[ ] ( )∑ π=k
knN/2jekX~
N
1]n[x
~
11
Not that for DT periodic signals we have
Due to the periodicity of the complex exponential we only need N exponentials
for DT FS
The FS coefficients can be obtained via
( )( ) ( ) ( ) ( )knN/2jmn2jknN/2jnmNkN/2j eeee πππ+π ==
[ ] ( )∑−
=
π=1N
0k
knN/2jekX~
N
1]n[x
~
[ ] ( )∑−
=
π−=1N
0n
knN/2je]n[x~kX~
DTFS Pair
For convenience we sometimes use
Analysis equation
( )N/2jN eW π−=
[ ] ∑−
=
=1N
0n
knNW]n[x
~kX
~
12
Synthesis equation [ ]∑−
=
−=1N
0k
knNWkX
~
N
1]n[x
~
Concept of DTFS
13
The DTFS
Note: we could divide x[n] or X[k] by N
14
The DTFS
15
The DTFS
16
The DTFS
17
The DTFS
18
The DTFS
19
Example: periodic square
20
Example: periodic square
- We know that
21
Example: periodic square
22
Example: periodic square
23
Example: periodic square
DTFS of an periodic rectangular pulse train
The DTFS coefficients
[ ] ( )( )
( )( ) ( )
( )10/ksin
2/ksine
e1
e1ekX
~ 10/k4j
k10/2j
5k10/2j4
0n
kn10/2j
π
π=
−
−== π−
π−
π−
=
π−∑
24
( )10/ksine10n π−=
Example: periodic impulse train
DFS of a periodic impulse train
Since the period of the signal is N
[ ] =
=−δ= ∑∞
−∞= else0
rNn1rNn]n[x~
r
[ ] ( ) ( ) ( ) 1ee]n[e]n[x~
kX~ 0kN/2j
1NknN/2j
1NknN/2j ==δ== π−
−π−
−π− ∑∑
25
We can represent the signal with the DTFS coefficients
as
[ ] ( ) ( ) ( ) 1ee]n[e]n[x~
kX~ 0kN/2j
0n
knN/2j
0n
knN/2j ==δ== π−
=
π−
=
π− ∑∑
[ ] ( )∑∑−
=
π∞
−∞=
=−δ=1N
0k
knN/2j
r
eN
1rNn]n[x
~
Properties of DTFS
Linearity
Shifting
[ ] [ ][ ] [ ]
[ ] [ ] [ ] [ ]kX~bkX
~anx~bnx~a
kX~
nx~kX
~nx
~
21DFS
21
2DFS
2
1DFS
1
+ →←+
→←
→←
[ ] [ ][ ] [ ]
~~kX
~nx~ DFS →←
26
Shifting
Duality
[ ] [ ][ ] [ ]
[ ] [ ]mkX~
nx~e
kX~
emnx~kXnx
DFSN/nm2j
N/km2jDFS
− →←
→←−
→←
π
π−
[ ] [ ][ ] [ ]kx~NnX
~kX
~nx~
DFS
DFS
− →←
→←
Properties of DTFS
27
Summary of Properties
28
Symmetry Properties
29
Periodic Convolution
Take two periodic sequences
Form the product
[ ] [ ][ ] [ ]kXnx
kXnxDFS
DFS
22
11~~
~~
→←
→←
30
The periodic sequence with given DTFS can be written as
Periodic convolution is commutative
[ ] [ ] [ ]kXkXkX 213
~~~=
[ ] [ ] [ ]∑−
=
−=1
0
213~~~
N
m
mnxmxnx
[ ] [ ] [ ]∑−
=
−=1
0
123~~~
N
m
mnxmxnx
Periodic Convolution
31
Outline
o Introduction
o DTFS & properties
o DTFT of periodic signals
o DFT: Sampling of the DTFT
32
o DFT: Sampling of the DTFT
o DTFT, DTFS, DFT, DFS, FFT, ZT: numerical (Matlab)
o Summary
The DTFT
∑
⊗
=
+⇒
⊗
∞
−∞=
− ωω
π2
:TransformFourier DT Inverse
][)(
PCTDT
:TransformFourier DT
enxeXn
njj
• DTFT represents a DT aperiodic signal as
a sum of infinitely many complex
exponentials, with the frequency varying
continuously in (-π, π)
• DTFT is periodic
only need to determine it for
33
∫=
⇒+
⊗
π
ωω
π
ωπ
2
2
)(2
1][
DT PCT
:TransformFourier DT Inverse
deeXnxnjj
only need to determine it for
DTFT is continuous in frequency
The DTFT
From the numerical computation viewpoint, the computation of DTFT by computer has several problems:
The summation over n is infinite
nj
n
jenxeX
ωω −∞
−∞=
∑= ][)(
34
The summation over n is infinite
The independent variable is continuous
DTFT and z-transform are not numerically
computable transforms
ω
FS versus FT
Aperiodic signals can be viewed as a periodic signal
with an infinite period
FS: a representation of periodic signals as a linear
combination of complex exponentials
The FS cannot represent an aperiodic signal for all times
35
The FS cannot represent an aperiodic signal for all times
FT: apply to signals that are not periodic
The FT can represent an aperiodic signal for all time
NN
1
0
NN
1
0
P-DTP-DF ][1
][
P-DFP-DT ][][
:DTFS
0
0
⇒=
⇒=
∑
∑−
=
−
=
−
N
k
knj
N
n
knj
ekXN
nx
enxkX
ω
ω
DT PCT )(2
1][
PCTDT ][)(
:DTFT
2
2
2
⇒+=
+⇒=
∫
∑∞
−∞=
−
π
π
ωω
π
ωω
ωπ
deeXnx
enxeX
njj
n
njj
The FT of Periodic Signals
Periodic sequences are not absolute or square summable: no DTFT exist
We can represent them as sums of complex exponentials: DTFS
We can combine DTFS and DTFT
Periodic impulse train with values proportional to DTFS coefficients
36
Periodic impulse train with values proportional to DTFS coefficients
This is periodic with 2π since DTFS is periodic
( ) [ ]∑∞
−∞=
−=
k
j
N
kkX
NeX
πωδ
πω 2~2~
The FT of Periodic Signals
The inverse transform can be written as
( ) [ ]
[ ] [ ]∑∫∑
∫ ∑∫−−
−
∞
−
−
∞
−∞=
−
−
=
−
−=
1 22
0
2
0
2
0
~12~1
2~2
2
1~
2
1
N nN
kj
nj
nj
k
njj
ekXN
deN
kkX
N
deN
kkX
NdeeX
πω
επ
ε
ωεπ
ε
ωεπ
ε
ω
ωπ
ωδ
ωπ
ωδπ
πω
π
37
FT Pair:
Example:
( ) [ ]∑∞
−∞=
−=
k
j
N
kkX
NeX
πωδ
πω 2~2~
∑∫∑=
−−∞= 0
0kk NNN ε
[ ]∑−
=
=1
0
2
~ ~1][
N
k
nN
kj
X ekXN
n
π
Example
Consider the periodic impulse train
The DTFS was calculated previously to be
[ ]∑∞
−∞=
−δ=r
rNn]n[p~
38
Therefore the FT is
[ ] k allfor 1kP~
=
( ) ∑∞
−∞=
ω
π−ωδ
π=
k
j
N
k2
N
2eP
~
Finite-length x[n] & Periodic Signals
Convolve with periodic impulse train
The FT of the periodic sequence is
[ ] [ ]∑∑∞
−∞=
∞
−∞=
−=−δ∗=∗=rr
rNnxrNn]n[x]n[p~
]n[x]n[x~
( ) ( ) ( ) ( )
π
−ωδπ
== ∑∞
ωωωω k22eXeP
~eXeX
~ jjjj
39
This implies that
DFS coefficients of a periodic signal = equally spaced samples of the FT of one period
( ) ( ) ( ) ( )
( )
π−ωδ
π=
−ωδ==
∑
∑
∞
−∞=
π
ω
−∞=
N
k2eX
N
2eX
~
NNeXePeXeX
k
N
k2j
j
k
[ ] ( )N
k2jN
k2j
eXeXkX~
π=ω
ω
π
=
=
Finite-length x[n] & Periodic Signals
40
Example
Consider
≤≤
=else0
4n01]n[x
41
The FT is
The DFS
coefficients
( ) ( )( )2/sin
2/5sineeX 2jj
ω
ω= ω−ω
[ ] ( ) ( )( )10/ksin
2/ksinekX
~ 10/k4j
π
π= π−
Outline
o Introduction to frequency analysis
o DTFS & properties
o DTFT of periodic signals
o DFT: Sampling of the DTFT
42
o DFT: Sampling of the DTFT
o DTFT, DTFS, DFT, DFS, FFT, ZT: numerical (matlab)
o Summary
Sampling the DTFT:Sampling in frequency domain
In the DTFT
The summation over n is infinite
nj
n
jenxeX
ωω −∞
−∞=
∑= ][)(
43
The independent variable is continuous
DTFT is not numerically computable transform
To numerically represent the continuous frequency DTFT, we must take samples of it DFT
ω
Sampling the DTFT:
Review to sampling
Sampling is converting x(t) to x[n]
T : sampling period in second; fs = 1/T : sampling frequency in Hz
Ωs=2πfs : Sampling frequency in radian-per-second
In frequency domain: convolution of X(jw) with an impulse train
[ ] ( ) ∞<<∞−= nnTxnx c
( ) ( )( )∑∞
−∞=
Ω−Ω=Ωk
scs kjXT
jX1
44
Creates replica of the FT of x(t); Replica are periodic with Ωs
If Ωs< ΩN sampling maybe irreversible due to aliasing of images
−∞=kT
( )ΩjX c
( )ΩjX s
( )ΩjX s
ΩN-ΩN
ΩN-ΩN Ωs 2Ωs 3Ωs
-
2ΩsΩs3Ωs
ΩN-ΩN Ωs 2Ωs 3Ωs
-
2ΩsΩs
3Ωs
Ωs<2ΩN
Ωs>2ΩN
Sampling the DTFT:Sampling in frequency domain
Consider an aperiodic x[n] with a DTFT
Assume a sequence is obtained by sampling the DTFT
Since the DTFT is periodic, the resulting sequence is also
[ ] ( )( )
( )( ) 10 ;~ /2
/2−≤≤==
=LkeXeXkX
kNj
kN
j π
πω
ω
( )ωjDTFTeXnx →←][
45
Since the DTFT is periodic, the resulting sequence is also
periodic
could be the DFS of a sequence
The corresponding sequence is
[ ]kX~
[ ] ( ) 10 and 10 ;~1
][~1
0
/2 −≤≤−≤≤= ∑−
=
LkNnekXN
nxN
k
knNj π
Sampling the DTFT
We can also write it in terms of the z-transform
[ ] ( )( )
( )( )kNj
kN
jeXeXkX
/2
/2
~ π
πω
ω ===
[ ] ( ) ( )( )kNjeXzXkX
/2~ π==
46
The sampling points are shown in figure
[ ] ( ) ( )( )( )kNj
ezeXzXkX kN
/2/2
~ ππ ==
=
Sampling the DTFT
The only assumption made on x[n]: its DTFT exist
Combine the equations gives
( ) [ ]∑∞
−∞=
−=m
mjjemxeX
ωω [ ] ( )∑−
=
=1
0
/2~1][~
N
k
knNjekX
Nnx
π[ ] ( )( )kNjeXkX
/2~ π=
[ ] ( ) ( )∑ ∑− ∞
−
=N
knNjkmNjeemxnx
1][~
1/2/2 ππ
47
Term in the parenthesis [] is
[ ] ( ) ( )
[ ] ( ) ( ) [ ] [ ]∑∑ ∑
∑ ∑∞
−∞=
∞
−∞=
−
=
−
= −∞=
−
−=
=
=
mm
N
k
mnkNj
k
knNj
m
kmNj
mnpmxeN
mx
eemxN
nx
~1
1][~
1
0
/2
0
/2/2
π
ππ
[ ] ( ) ( ) [ ]∑∑∞
−∞=
−
=
− −−==−r
N
k
mnkNjrNmne
Nmnp δπ
1
0
/21~
[ ] [ ] [ ]∑∑∞
−∞=
∞
−∞=
−=−∗=rr
rNnxrNnnxnx δ][~
Sampling the DTFT
48
FS are samples of the FT of one period
FS are still samples of the FT; But, one period is no longer identical to x[n]
Sampling the DTFT
DFS coefficients of a periodic sequence obtained through summing periodic replicas of aperiodic original sequence x[n]
If x[n] is of finite length & we take sufficient number of samples of its DTFT, x[n] can be recovered by
49
samples of its DTFT, x[n] can be recovered by
No need to know the DTFT at all frequencies, to recover
x[n]
DFT: Representing a finite length sequence by samples of DTFT
[ ][ ]
−≤≤
=else
Nnnxnx
0
10~
Sampling in the frequency domain
The relationship between and one period of in the under-sampled case is considered a form of time domain aliasing
Time domain aliasing can be avoided only if has finite length
just as frequency domain aliasing can be avoided only for
][nx ][~ nx
][nx
50
just as frequency domain aliasing can be avoided only for signals being band-limited
If has finite length N and we take a sufficient number Lof equally spaced samples of its FT, then
the FT is recoverable from these samples
equivalently is recoverable from
Sufficient number L means: L>=N We must have at least as many frequency samples as the
signal’s length
][nx
][nx ][~ nx
The DFT
Consider a finite length sequence x[n] of length N
For x[n] associate a periodic sequence
The DFS coefficients of the periodic sequence are samples of the DTFT of x[n]
[ ] 10 of outside 0 −≤≤= Nnnx
[ ] [ ]∑∞
−∞=
−= rNnxnx~
51
Since x[n] is of length N there is no overlap between terms of x[n-rN] and we can write the periodic sequence as
To maintain duality between time and frequency
We choose one period of as the DFT of x[n]
∑−∞=r
[ ] ( )[ ] ( )( )[ ]NkXkXkX == N mod ~
[ ]kX~
[ ][ ]
−≤≤
=else
NkkXkX
0
10~
[ ] ( )[ ] ( )( )[ ]N
nxnxnx == N mod ~
The DFT
Consider the DFS pair
The equations involve only one period so we can
write
[ ] ( )∑−
=
=1
0
/2~1][~
N
k
knNjekX
Nnx
π
[ ] ( )∑−
=
−=1
0
/2][~~ N
n
knNjenxkX
π
−N ~1 1
52
write
The DFT pair
[ ]( )
−≤≤
= ∑−
=
−
else
NkenxkX
N
n
knNj
0
10][~1
0
/2π [ ] ( )
−≤≤
= ∑−
=
else
NkekXNnx
N
k
knNj
0
10~1
][
1
0
/2π
[ ] ( )
N, LLk
enxkXN
n
knNj
>=−≤≤
=∑−
=
−
10
][1
0
/2π [ ] ( )
NLwhereLk
ekXN
nxN
k
knNj
>=−≤≤
= ∑−
=
,10
1][
1
0
/2π
[ ] ][nxkXDFT →←
DFT: x[n] finite duration
53
DFT: Example 1
DFT of a rect. pulse x[n], N=5
Consider x[n] of any length L>5
Let L=N=5
Calculate the DFS of the
periodic form of x[n]
54
periodic form of x[n]
[ ] ( )
( )
±±=
=
−
−=
=
π−
π−
=
π−∑
else0
,...10,5,0k5
e1
e1
ekX~
5/k2j
k2j
4
0n
n5/k2j
DFT: Example 1
Let L=2N=10
We get a different set
of DFT coefficients
Still samples of the
55
Still samples of the
DTFT but in different
places
x[n] = Inverse X[k]
depends on relation L
& N
DFT: Example 1summary
56
The larger the DFT size K, the more details of the INVERSE DFT, i.e., x[n ] can be seen
DFT: example 2
57
NLwhereLk >=−≤≤ ,10
DFT: example 3
58
NLwhereLk >=−≤≤ ,10
DFT: example 3
59
Properties of DFT (very similar to that of DTFS)
Linearity [ ] [ ][ ] [ ]
[ ] [ ] [ ] [ ]kbXkaXnbxnax
kXnx
kXnx
DFT
DFT
DFT
2121
22
11
+ →←+
→←
→←
60
Duality[ ] [ ][ ] ( )( )[ ]N
DFT
DFT
kNxnX
kXnx
− →←
→←
Example: Duality
61
Circular Shift property
1-Nn0 range over the defined belonger nomay
m],-x[ny[n] shifted them,arbitrary an For -
Nn and 0nfor 0x[n]-
1-Nn0for defined ]length x[n-NConsider -
≤≤
=
>=<=
≤≤
62
( )( )[ ] ( )
( )N
NN
mN
Nnnmnxny
by shift circular toequivalent is mshift circular A -
m)-(Nby shift circular left a toequivalent is mby shift Circular -
modulo where, ][
1-Nn0 range in the be always bemust y[n] :shift"Circular "
m]-n[ :shiftlinear apply cannot We
>
=−=
≤≤−
∞≤≤∞−==>
Circular Shift property[ ] [ ]
( )( )[ ] [ ] ( )mNkjDFT
N
DFT
ekXmnx
kXnx/21-Nn0 π− →←≤≤−
→←
63
64
Circular Shift property
65
Circular Convolution Property
][~ nx
][ N ][][ 213 nxnxnx =
66
10 −≤≤ Nn
• Linear convolution: one sequence is multiplied by a time–
reversed and linearly-shifted version of the other
•Circular convolution: the second sequence is circularly time-reversed and circularly-shifted it is called an N-point circular
convolution
Circular Convolution Property
( ) ( ) ( )
( ) ( )( ) )()()( so, is from DFT -
so, :nconvolutioCircular -
so h[n],* x[n] y[n] :nconvolutioLinear
/2 ==
==
==−
kHkXkYeYY(k) eY
X(k)H(k)W(k) n] x[n] N h[w[n]
eHeXeY
kNjj
jjj
πω
ωωω
67 aliasing en with timConvolutioLinear n ConvolutioCircular
0
10][][
thenN, period of sequence periodic a ][~y form :y[n] of DFTget To
)12 oflength max. has BUT length of
then,length of and If
===>
−≤≤−
=
−
−
∑∞
−∞=
else
NnrNnynw
nx
N-(y[n]Nw[n]
Nh[n]x[n]
r
Circular Convolution:
example 1
Circular convolution of two finite length sequences
][][ 01 nnnx −= δ
][10WkX
kn
N=
][][ 01 nnnx −= δ
68
[ ] [ ] ( )( )[ ]∑−
=
−=1
0
213
N
m
Nmnxmxnx
[ ] [ ] ( )( )[ ]∑−
=
−=1
0
123
N
m
Nmnxmxnx
][][
][
23
1
0 kXWkX
WkX
kn
N
N
=
=
Example 2: L=N
Two rect. X[n]: L=N=6
DFT of each sequence
[ ] [ ] −≤≤
==else
Knnxnx
0
10121
[ ] [ ] =
=== ∑− − kN
ekXkXN knj 01 2π
69
Multiplication of DFTs
Inverse DFT
[ ] [ ] =
=== ∑=
−
else
kNekXkX
n
knN
j
0
0
0
21
[ ] [ ] [ ] =
==else
kNkXkXkX
0
02
213
[ ] −≤≤
=else
NnNnx
0
103
Example 2: L=2N
Augment zeros to each sequence L=2N=12
The DFT of each
sequence
[ ] [ ]N
Lk2j
e1kXkX
π−
−==
70
Multiplication of DFTs
[ ] [ ]N
k2j
N
21
e1
e1kXkX
π−
−
−==
[ ]
2
N
k2j
N
Lk2j
3
e1
e1kX
−
−=
π−
π−
x[n] = Inverse DFT X[k] is not unique; depends on L and N
Circular convolution example
71
72
Symmetry Property
73
Symmetry Properties
74
Outline
o Introduction to frequency analysis
o DTFS & properties
o DTFT of periodic signals
o DFT: sampling of the DTFT
75
o DFT: sampling of the DTFT
o DTFT, DTFS, DFT, DFS, FFT, ZT: numerical (matlab)
o Summary
Discrete-time signal transforms
7676
Numerical Calculation of FT
1. The original signal is digitized
2. A Fast Fourier Transform (FFT) algorithm is applied, which yields samples of the FT at equally spaced intervals
77
at equally spaced intervals
For a signal that is very long, e.g., a speech signal or a music piece, spectrogram is used
FT over successive overlapping short intervals
Matlab examples: DTFT
Suppose that:
Analytically, the DTFT is
X(ejω): continuous function of ω
X(ejω): periodic with period 2π
78
X(e ): periodic with period 2π
Plot it using
78
Matlab examples: DTFT
Signal x[n] DTFT
7979
Matlab examples: DFT
Close form X(ejω) not always easy
To plot |X(ejω)|, we sampled from 0 to 2π
In code: w and X are vectors
Small step size 0.001 to simulate continuous frequency
80
Workaround: DFT
Uniform L-samples from DTFT from 0 to 2π
Takes discrete values and returns discrete values
No need to find |X(ejω)| analytically
Fast implementation using the fast Fourier transform (FFT)
Matlab: fft(x,L)• L: number of samples to take
• More L more resolution
• Default L is N=length(x)
80
Matlab examples: DFT
Calculating the DFT
81
Plotting the DFT against k
81
Matlab examples: DFT
Notes: Default L=32 gives bad
resolution
information lost
82
information lost
x-axis not useful
Cannot find fundamental frequency 3π/8
82
Matlab examples: DFT
Effect of increasing L (better resolution)
• L=64
83
• L-128
83
Matlab examples: DFT
Obtaining the frequency (x-axis)
84
Spike at 3π/8=1.17
Spike at 2π-3π/8 = 5.11
FFT calculates from 0 to 2π
More familiar to shift using fftshift
84
Matlab examples: DFT
85
Spikes at 3π/8 and -3π/8
85
Matlab examples: DFT
Sometimes we want
frequency in Hz
8686
Matlab examples: DFT
87
|X[k]| vs. ωk
Discrete
DFT
87
|X(ejω)| vs. ω
Continuous
By interpolating DFT
|X(f)| vs. f
Continuous
f = (ω/ 2π) fs fs : sampling frequency
fft values divided by N
Peak at 0.5 (half our
amplitude of 1)
Matlab examples: DFS
No special function
Same as DFT
Provided signal corresponds to 1 period
8888
Matlab examples: z-Transform
Suppose that:
8989
Matlab examples: z-Transform
9090
Matlab examples: z-Transform
9191
Matlab examples: z-Transform
Evaluate H2(ejω) directly from z-Transform
9292
Matlab examples: z-Transform
Finding z-Transform analytically
9393
Outline
o Introduction to frequency analysis
o DTFS & properties
o DTFT & properties
o FT of periodic signals
94
o FT of periodic signals
o DTFT, DTFS, DFT, DFS, FFT, ZT: numerical (matlab)
o FTT
o Summary
FFT: Fast Fourier transform
FFT is a direct computation of the DFT
FFT is a set of algorithms for the efficient and digital computation of the N-point DFT, rather than a new transform
95
rather than a new transform
Use the number of arithmetic multiplications and additions as a measure of computational complexity
FFT
The DFT pair was given as
Baseline for computational complexity:
Each DFT coefficient requires
• N complex multiplications
[ ] ( )∑−
=
π=1N
0k
knN/2jekXN
1]n[x
[ ] ( )∑−
=
π−=1N
0n
knN/2je]n[xkX
96
• N-1 complex additions
All N DFT coefficients require
• N2 complex multiplications
• N(N-1) complex additions
Complexity in terms of real operations
• 4N2 real multiplications
• 2N(N-1) real additions
FFT
Most fast methods are based on symmetry
properties
Conjugate symmetry( ) ( ) ( ) ( ) ( ) ( )knN/2jnkN/2jkNN/2jnNkN/2j eeee π−π−π−−π− ==
97
Periodicity in n and k
The Second Order Goertzel Filter
• Approximately N2 real multiplications and 2N2 real additions
• Do not need to evaluate all N DFT coefficients
Decimation-In-Time FFT Algorithms
(N/2)log2N complex multiplications and additions
eeee ==
( ) ( ) ( ) ( )( )nNkN/2jNnkN/2jknN/2j eee +π+π−π− ==
Symmetry and periodicity of complex exponential
Complex conjugate symmetry
Periodicity in n and k
ImRe)( *][ kn
N
kn
N
kn
N
kn
N
nNk
N WjWWWW −=== −−
98
For example
The number of multiplications is reduced by a factor of 2
nNk
N
Nnk
N
kn
N WWW)()( ++ ==
Re])[Re][(Re
Re][ReRe][Re ][
kn
N
nNk
N
kn
N
WnNxnx
WnNxWnx
−+=
−+ −
Outline
o Introduction to frequency analysis
o DTFS & properties
o DTFT & properties
o FT of periodic signals
99
o FT of periodic signals
o DTFT, DTFS, DFT, DFS, FFT, ZT: numerical (matlab)
o FTT
o Summary
Overview of signal transforms
Variable Period Continuous
Frequency
Discrete
Frequency
DT x[n] n N kNkk /2πω =
ω
100
CT x(t) t T kTkk /2πω =Ω
• DT-FS: Discrete in time; Periodic in time; Discrete in Frequency; Periodic in Frequency
• DT-FT: Discrete in time; Aperiodic in time; Continous in Frequency; Periodic in Frequency
• CT-FS: Continuous in time; Periodic in time; Discrete in Frequency; Aperiodic in Frequency
• CT-FT: Continuous in time; Aperiodic in time; Continous in Frequency; Aperiodic in Frequency
• DFT: Discrete in time; Aperiodic in time; Discrete in Frequency; Periodic in Frequency;
finite-duration x[n]
• DFS: Discrete in time; Periodic in time (make finite-duration x[n] periodic);
Discrete in Frequency; Periodic in Frequency;
Relationships between signal transforms
Continuous-time
analog signal
x(t)
Discrete-time
analog sequence
x [n]
Sample in time
Sampling period = Ts
ContinuousFourier Transform
X(f)
Discrete Fourier Transform
X(k)
Discrete-Time
Fourier Transform
Ω)
Laplace
Transformz-Transform
X(z)C
C D
101
Ω= j ez
ω=2πfΩ = ω Ts,
scale
amplitude
by 1/Ts
Sample in
frequency,Ω = 2πn/N,
N = Length
of sequence
X(f)
∞≤≤∞
∫∞
∞−
f-
dt e x(t) ft2 j- π
X(k)
10
e [n]x 1
0 =n
N
nk2j-
−≤≤
∑−
Nk
N π
X(Ω)
π20
e [n]x - =n
j-
≤Ω≤
∑∞
∞
Ωn
Transform
X(s)s = σ+jω
∞≤≤∞
∫∞
∞−
−
s-
dt e x(t) st
X(z)
∞≤≤∞−
∑∞
∞−
z
=n
n- z [n]x
s = jω
ω=2πf
C CC
C
D
DC Continuous-variable Discrete-variable
Ω= j ez r
Fourier versus Cosine Transform
Recall: the cosine wave starts out 1/4th later in its period
It has an offset Common to measure this offset in degree or radians
One complete period equals 360°or 2π radian
102
One complete period equals 360°or 2π radian
The cosine wave thus has an offset of 90°or π/2 This offset is called the phase of a sinusoid We cannot restrict a signal x(t) to start out at zero
phase or 90°phase all the time Must determine its frequency, amplitude, and phase to
uniquely describe it at any one time instant With the sine or cosine transform, we are restricted to
zero phase or 90°phase
DCT: One Dimensional
∑
+=
−
=
1
0 2
)12(cos
2
1 n
t n
ftxCX tff
π
103
>
==
0,1
0,2
1
f
fC f
where
n = size
x = signal
X = transform coefficients