Di it l Signal

Digital Si l Processing
P i

Chaotic Systems and Signal Processing Lab.

CSSP Lab.
L b

2010
1

773
x 31538
Office Hour: Wed. 1:30 a.m.-3:30
a.m. 3:30 p.m.
[email protected] & http://cssplab.cn.nctu.edu.tw

A. V. Oppenheim and R. W. Schafer, Discrete-Time
Signal Processing, Prentice Hall: New Jersy, 1999.
:&
917 x 54428
Office Hour : Monday 7:00-9:00
7:00 9:00 p.m.
pm


40 %
30 %
2 30
%
 Important Date (Tentatively)
Homework
k due
d one weekk later
l
Midterm Exam
Exam. Nov.
Nov 11,
11 2010
Final Exam. Jan. 12, 2011
Raw Grade Post Jan. 19, 2011
Final
Fi l Grade
G d Post
P t Jan.
J 20 20, 2011

3
 Contents
Part
P t I:
I Review
R i off Signals
Si l andd Systems
S t ( Chapter
Ch t 2-32 3)
1. DSP Systems and Applications
2. Introduction to Discrete-Time Signals
g and Processingg
3. The Z-transform
Part II: Transform Methods for DSP ( Chapter 4-5,8-9 )
1. Sampling Theory
2. Examples of Digital Filters
3. The Discrete Fourier Transform ( DFT )
4. Fast Fourier Transform
Part III: Digital Filter Design ( Text Chapter 6-7 )
1. Type of Filters --- FIR and IIR
2. Filter Design Techniques

4
 Chapter 1

Introduction

5
 Introduction
Signals
Speech( e.g. sampling frequency 8KHz:64Kbps 2.4Kbps),
Biomedical signals, Audio( e.g., MP3, sampling frequency
44 1KHz) and Sound,
44.1KHz) Sound Video and Image,
Image Radar signals
signals
Digital Signal Processing
Sample an analog signals Signals in digital form.
form
Process by digital processor.
Objectives
j
Remove interference or noise from a signal.
Obtain the spectrum of data.
Transform
T f the
h signal
i l to a suitable
i bl form.
f (e.g.
( Fourier
F i Transform,
T f
Wavelet Transform, Discrete Cosine Transform )

6
 Introduction
Advantages
Guaranteed Accuracy: depend on number of bits need.
Perfect Reproducibility: e.g. digital recording.
Advantages is always taken of the advances in semiconductor
technology.
Flexibility: reprogrammable.
Disadvantages
Speed & cost: fast ADC/DAC are expensive. Bandwidth of 100
MHz increased are processed by analog signal method.
Design time: DSP hardware increasingly complicated.
Finite length effects: limited-bit number of DSP processor.

7
 DSP Applications
pp
Image/Video
g Processingg
Pattern recognition, robot vision, facsimile, animation, data
compression
Controls
Spectrum analysis, position and velocity control of motors, noise
reduction
Speech/Audio
Speech recognition, speech/audio synthesis, text to speech, digital
audio, quantization
Tele-communications
T l i ti
Echo cancellation, adaptive equalization, video conference
Vehicle Electronics
Telematics, Navigation, Car Safety
Biomedical signal processing
Patient monitoring
monitoring, EEG , EKG analysis,
analysis Biomedical image
8 processing
 DSP Applications
pp

Key DSP operations

i
Convolution
Correlation
Filtering
Discrete Transforms
DTFTDFT FFT DCT DWT Z-Transform
Modulation

9
 Overview of Real-Time Signal Processing

f max fs

fs

10
 Sampling Frequency
fs : Sampling frequency
f s 2 f max
fmax: Max frequency of the input signal
Sampling theorem: f s = 2 f max
Sampling rate Nyquist frequency
No distortion (ideally)
e.g.
e g Audio signal BB.W.:
W : 22KHz CD: 44
44.1KHz
1KHz
Speech signal B.W.: 4KHz 8K or 20KHz (recommended)

11
 Sampling
p g Frequency
q y
xt xtpt x t

pt

Band-limited signals

12
 Aliasing
f s f c < f c f s < 2 f c = 2 f max Aliasing

Anti-Aliasing Filter
Remove the undesired input signal or noise.
noise

13
 For B-bit
B bit ADC:
B quantization noise error
SNR[Signal-to-Noise-ratio] 20log[ 1.5 2 B+1 ] in dB

CD quality
B Amin
i (dB)
(d )

8 56
10 68
12 80
CD qqualityy 6
16 104
0
14
 Example
Ex.
E Determine
D t i fs to
t give
i an aliasing
li i error off less
l
than 2% of the signal level in the passband.

15
 Solution:
1
j 2 fC 1 1
| H( f )| = = =
R+
1 1 + j 2 fRC 1
1 + (2 fRC ) 2 2
j 2 fC
1 1
= h fC =
where = 2 KHz
KH
f 2 2 RC
1+ ( )
fC

Assume the analog input has a

wideband spectrum (assured
aliasing).

16
 1
KH X b
P b d @ 2 KHz
Passband = 00.7071
7071
2
2
desired aliasing level < 0.7071
0 7071 = 0.01414
0 01414
100
1
< 0.01414
fa 2
1+ ( )
2
f a > 141.4
141 4 KHz

f s > f C + f a = 2 + 141
141.44 = 143.4
143 4 KHz

say = 150 KHz

f s ,say,
17
 Quantization & Encoding
Q g
For example ADC 0804.
0804 0~5V, 8 bits
Quantization error
V fs
 LSB= .
2 B

1 1 V fs 5V
Range of quantization error = LSB = = 9
2 22 B
2

18
 xt xn

e  x(t ) x(n)
/2
/2 1 1 3 2
= e p (e)de = e
2 2
= Uniformly distributed
3 / 2 12
e / 2

For a sin wave A sin t average signal power A2 / 2.

Signal-to-Noise
Signal to Noise Ratio (SNR)
signal power A2 2 2A
SNR = 10 log10 = 10 log 2 where = B
e2 12 2
= 10log(1.5 22 B ) = 6.02B + 1.76dB
ADC bits 6 dB / bit signal quality
19
 Dynamic Range:
max signal power 2B
D (in dB) = = 20log = 20log2 B = 6.02 B
min signal
g power
p 1
For CD audio quality ( > 90dB) B = 16 16-bit ADC

Ex:
E
1
| H ( f ) |=
1 + ( f fC )6
3rd order 1. Minimum stopband attenuation, Amin, for the anti-
Butterworth aliasing filter.
filter
2. Minimum f S and aliasing error/signal in passband
3. Aliasingg error/signal
g in ppassband
20
 Solution:
2 V fs
RMS quantization noise : = where =
12 2 3 2B
1. For ADC, it will attenuate a signal in stopband.
max passband signal level

stopband signal level
(Half power) 2 B

=( ) /( ) = 1.5 2 B +1
3dB Bandwidth 2 2 3
1 5 2 B +1 )
A min = 20log( 1.5
= 80dB = 6.02 B + 7.78 dB

21
 1
2
2. Amin = 20log
20l , f C = 4 KHz
KH
1 + ( f fC ) 6

fS
f = 86.2 KHz f S 172.4 KHz, say f S = 173 KHz
2

22
 3 Aliasing Level (AL):
3.
1 3 3.64 103
AL = = 3.64 10 = 0.515%
173 4 6 0.7071
1+ ( )
4
sample
p rate AL

23
 DAC: signal
g recover
yn y (t ) yt

24