Signals And Systems

(Spring 2024)

Fourier Applications

Geng-Shi Jeng (鄭耿璽), Ph.D.

Institute of Electronics
National Yang Ming Chiao Tung University
 Content
• Introduction
• Fourier Transform of Periodic Signals
• Convolution/Multiplication with Mixed Signals
• Fourier Transform of Discrete-Time Signals
• Sampling
• Reconstruction of Continuous-Time Signals
• Discrete-Time Processing of Continuous-Time Signals
• Fourier Series of Finite-Duration Nonperiodic Signals
 3

Generalization of FT on all signals

• Original
• Periodic time signals →FS, DTFS
• Nonperiodic time signals →FT, DTFT
• Why applying FT to periodic signals?
• When a system involves “MIXED” signals (e.g., nonperiodic for impulse
response + periodic for input signals), a bridge between them is needed.
• We need to know what EXACT the frequency is
• X[k] at which frequency?
• We need “FT” of all different classes signals.
• CT: FS→FT
• DT: DTFS → DTFT , DTFS→FT, DTFT → FT
• DTFT is related to FT:
 4

Practical application of DTFS to FT

• Why DTFS to represent FT/FS/DTFT?
• IT is DFT or FFT, commonly used for computational applications.
 5
FT representations
for ALL transform
 6

Relating FT to FS
 7

Relating FT to FS

Look identical
but with exact
frequency bins

• FT of PERIODIC signal is a series of impulses spaced by ωo = 2π/T

• X(jω) = 2πX[k]
 8

Impulse train (T) → Impulse train (F)

 9

Periodic signals = Nonperiodic * Impulse train

前幾張投影片教我
們的
 10

Periodic signals = Nonperiodic * Impulse train

• How to relate the periodic signals to FT coefficients?

 11

Example 1
 12

Relating DTFT to DTFS

 13

Relating DTFT to DTFS

 14

Example 1
 15

Convolution with Mixed Signals (CT)

 16

Convolution with Mixed Signals

Convolution property for mixture of periodic (x(t))

and nonperiodic signals (h(t)).
 17

Example 1

Next slide
 18

Recall FS: Example 5

2 1 T /2 1 T0 − jk0t − 1 − jk0t T0
T −T / 2 T −T0
0 = X [k ] = x (t ) e − jk0t
dt = e dt = e |−T0
T Tjk 0
e jk0T0 − e − jk0T0
=
2
Tk0
( 2j
) = 2 sin(Tkk T ) = 2 sin( kk22T / T ) .
0
0 0 0
 19

Convolution with Mixed Signals (DT)

X[n] y[n] = x[n]*h[n]

h[n]

h[n]
 20

Multiplication with Mixed Signals (CT)

 21

Multiplication with Mixed Signals (CT)

 22

Example 1
 23

Example 1
 24

Example 2
• Example 4.6 AM Radio
• Emitted signal r(t) = m(t)cos(wct), where m(t) is the wanted signal.
• Received signal r(t) = emitted signal (assuming perfect communication)
• What is y(t)?

-w w
 25

Example 2
 26

Demodulation + LPF
 27

Multiplication with Mixed Signals (DT)

 28

Example 1
 29

Example 1
 30

Recall: Example 2

M
1 − e − j (2 M +1)
j
X (e ) = e
n =− M
− j n
=e j M

1 − e − j
− j (2 M +1) / 2 j (2 M +1) / 2 − j (2 M +1) / 2
j M e ( e − e ) where we may use L’Hopital’s
=e
e − j / 2 ( e j / 2 − e − j / 2 ) rule to obtain the value of the
RHS at Ω= 0, ±2π, ±4π, …. (It
sin((2 M + 1) / 2) is 2M + 1)
= ,
sin( / 2)
 31

Example 1
Effect of windowing a data record. Y(ej) for different values of M, assuming
that 1 = 7/16 and 2 = 9/16. (a) M = 80, (b) M = 12, (c) M = 8.
 32
Fourier Transform of Discrete-Time
Signals
Q: Is it DTFT? A: No! We try to represent a DT signal with CT, and then do FT.
Q: Why? A: We deal with mixture of CT and DT signals.

Relation between CT and DT

frequency

Relating FT to DTFT
 33
Fourier Transform of Discrete-Time
Signals
 34

Example 1
 35

Recall: Example 1

x[n] is absolutely summable

when |α| <1
 36

Relating FT to DTFS
 37

Recall: Time scaling of impulse

1
 (at ) =  (t )
|a|
Consider a>0;
 
s 1

−
x (t ) ( at )dt = −  a  a ds
x  ( s )

 
1 s 1 1
=  x    ( s )ds = x (0) =  x (t )  (t )dt
a −  a  a −
a

If a<0, then s = at = -|a|t.

 −
s −1

−
x (t ) ( at )dt = 

x    ( s)
a |a |
ds


1 s 1
=  x  
| a | −  a 
 ( s ) ds =
|a |
x (0)
 38

Relating FT to DTFT
 39

Relating FT to DTFS
 40

Sampling

CT DT
Sampling

Closed every Ts CT format

•
 41

Sampling (Graphic derivation)

 42

Sampling (Nyquist Criterion)

Oversampling
Original CT spectrum

Sampling

To avoid aliasing (distortion)

Sampling rate > 2* maximum Undersampling → Aliasing
frequency (Nyquist Criterion)
 43

Example 1
 44

Example 1

Aliasing
 45

Remarks
 46

Example 2: Aliasing of wheel in a movie

Wagon Wheel effect

https://www.youtube.com/watch?v=6XwgbHjRo30
Example 2: Aliasing of wheel in a movie 

 = r= /4
Assume frame rate = 30Hz

→Aliasing

→Aliasing
→ DC
 48

Remarks: Aliasing in a movie

 49

Downsampling

• Where q is a positive integer.

• What is the relation between q and maximum frequency of x[n]
to avoid aliasing?
• We don’t use textbook’s contents because they are pretty confusing.
• We use graphic method to explain.
 50

Reconstruction of Continuous-Time Signals

CT DT DT CT
Sampling Recon.

Closed
every Ts

• Two continuous-time signals x1(t) (dashed line) and x2(t) (solid

line) that have the same set of samples.
• The samples of a signal do not always UNIQUELY determine the CT
signals.
 51

Reconstruction of Continuous-Time Signals

 52

Recall: Sampling
 53

Reconstruction: Frequency domain

 54

Recon: Time domain

 55

Example 1
 56

Recon: Zero-order hold

 57

Recon: Zero-order hold

 58

Recon: Zero-order hold

 59

Example 1

Effective Filter

0.99~1.01
<10^(-3)
-fs -20k 20k fs Hz
 60

Example 1
• We use the frequency (i.e., H(jf)) instead of angular frequency
(i.e., H(jw)) in this case.
 61

Example 1

Figure 4.42 (p. 382)

Anti-imaging filter design with and without
oversampling. (a) Magnitude of Ho(jf) for 44.1-kHz
sampling rate. Dashed lines denote signal passband and
images. (b) Magnitude of Ho(jf) for eight-times
oversampling (352.8-kHz sampling rate. Dashed lines
denote signal passband and images. (c) Normalized
constraints on passband response of anti-imaging filter.
Solid lines assume a 44.1-kHz sampling rate; dashed
lines assume eight-times oversampling. The normalized
filter response must lie between each pair of lines.
 62

Example 1
 63
Discrete-Time Processing of Continuous -Time
Signals
 64

Discrete-Time Processing of Continuous -Time Signals

 65

Discrete-Time Processing of Continuous -Time Signals

 66

Oversampling (Practical)
 67

Decimation = downsampling
 68

Decimation filter

Decimation filter = LPF

is required to avoid aliasing -2π 2π
before downsampling

-2π 2π

-2π 2π

-2π 2π
 69

Interpolation = upsampling
 70

Interpolation filter
Interpolation = LPF
is required to reject
-2π -π π 2π
“images” after upsampling

-2π 2π

-2π 2π

-2π 2π
 71
Discrete-Time Processing of Continuous -Time
Signals (w/ decimation & interpolation)
 72

Fourier Series of Finite -Duration Nonperiodic Signals

 73

Relating DTFS to DTFT

 74

Relating FS to FT
 75

Another recap
Nonperiodic Periodic

Periodic Periodic

• DFT: Discrete-time Fourier

Transform
• DTFS: Discrete-time Fourier
Series
 76

What we have learned

• FT of Periodic Signals
• Convolution/Multiplication with Mixed Signals
• FT of Discrete-Time Signals
• Sampling & Reconstruction
• Discrete-Time Processing of Continuous-Time Signals
• Fourier Series of Finite-Duration Nonperiodic Signals
• FFT (Optional, see appendix)
 7
7

Appendix
(Fast Fourier Transform)

Optional reading: Will address in the

DSP course
 78

Fast Fourier Transform (FFT)

 79

FFT
 80

FFT
 81

FFT
 82

FFT
 83

FFT