4/29/2020

Lecture I
DSP Basics
Dr. Twe Ta Oo
Lecturer
Faculty of Computer Systems and Technologies
University of Computer Studies, Yangon
1

Acknowledgement:
Some of the contents are prepared according to online materials.
Lecture Contents
Part I:
 What is Digital Signal Processing (DSP)?
 Signal basics

4/29/2020
 Systems and their properties
 Sound creation and its intensity
 Sound intensity in dB scale
Part II:
 Analog to digital conversion
 Sampling
 Quantization
 Encoding 2

 Application areas of DSP

Digital Signal Processing (DSP)
 DSP is concerned with
 the representation of analog signals by sequences of

4/29/2020
numbers (ADC),
 the processing of these sequences by numerical
computation techniques (digital signal processor), and
 the conversion of such sequences into analog signals
(DAC).

3
Signal Basics
 Something that conveys information
 e.g. speech signal, image signal, video signal, etc.
 Represented by a function of independent variables such

4/29/2020
as time, distance, space, etc.
 e.g. Speech and music signals represent air pressure
as a function of time.
 Basic element – sample

Fig. A speech /music signal s(t) (one-dimension).

Signal Basics
 A black-and-white picture is a representation of light
intensity as a function of two spatial coordinates.
 Basic element – pixel

4/29/2020
Fig. An image signal i(x,y) (two dimension).

 A video signal consists of a sequence of images, called

frames, and is a function of three variables: two spatial
coordinates and time.

5
Fig. A video signal (multi-dimension).
Sound Creation and Propagation
 Sound originates from the disturbance of the air by any object.
 E.g. Two hands clapping cause a disturbance of the air
around the hands.

4/29/2020
 The hands are the source of the sound.
 The local region of air has gained energy caused by the motion
of the air molecules. This energy spreads outwards in sound
waves.
 Sound travels through the air at about 340 metres per second.

6
7

4/29/2020
Sound − Representation
 Sound is often represented diagrammatically as a sine wave.
 The wave crests can be considered as the pressure maxima,
whilst the troughs represent the pressure minima. 

4/29/2020
crest through

8
Sound − Physical Variables
 Two physical measurements describe the sound wave
completely: the amplitude (A) and the period (T).
 We usually speak, not of the period, but of the frequency (f) of

4/29/2020
vibrations, which is simply the inverse of the period (i.e. 1/T).

 The frequency (f) measures the

Amplitud
number of waves that travel by
in each unit of time.

e
 It is usually measured in Hertz T
(Hz, cycle per second) (1 kHz =
1000 Hz).
 The range of human hearing is from 20 Hz
to 20 kHz. 9
 2kHz − 4kHz are easiest to perceive (detectable at a
relatively low volume);
Sound − Physical Variables (cont.)
 The distance per cycle is named as wavelength of the
sound wave (λ):

4/29/2020
λ =where
c /f, c = sound velocity and f = frequency.

 E.g. The source is vibrating at 100 Hz, then it will vibrate

once per one hundredth of a second. The sound will then
travel 340/100 = 3.4 m/cycle.

10
Sound Intensity
 Sound intensity is a measure of power for a sound, as it
contacts an area such as the eardrum and directly
proportional to the square of the amplitude of the waveform.

4/29/2020
 The intensity as a function of the pressure is:

I(t) = x2(t)/ρc.

 In MKS units, the intensity I(t) is measured in Watts/m2 and

the pressure x(t) is measured in Newtons/m2 (or) Pascals.
 The ρc is the impedance of air, which is 415 Rayls in MKS
unit.
11
Sound Measurement in Decibel (dB) Scale
 It is common to refer the power of a signal or intensity of a
sound as a “sound level” measured in Decibel (dB). It is a
logarithmic measure.

4/29/2020
Acousticians use the dB scale for the following reasons: 
1. Quantities of interest often exhibit such huge ranges of
variation that a dB scale is more convenient than a linear
scale.
2. The human ear interprets loudness more easily represented
with a logarithmic scale than with a linear scale. 

e.g. power ratio is 100 >> 20 dB

12
100000000 >> 80 dB
 Acoustical Measures in DB Scale

dB SPL  20 log10 ( X 2 / X 1 ), dB SIL  10 log10 ( I 2 / I1 ),

4/29/2020
where X1 is reference pressure level and I1 is reference intensity
level.
 An absolute acoustical decibel, dB scale is referenced to the
minimal threshold of hearing.
 The reference pressure is 20 μPa and it is the minimal
threshold of hearing at 1,000 Hz for most people.
 The reference intensity is derived as follows.
12 12
I1  (20 ) /(415)  0.964 10
2 2
I1  X 12 /( c),  10 W / m
13 4/29/2020
dB SIL & dB SPL

 I2   CX 2 
2
 CX 2 

dB SIL  10 log10    10 log10  
2 
 20 log10  ,
 I1   CX 1   CX 1 

4/29/2020
where

I2: sound intensity

Therefore:
I1: reference intensity dB IL = dB SPL
X2: sound pressure

X1: reference pressure 14

C : constant (1/ρc)
Limits of Sound Perception
 Intensity/pressure of audible sound is enormous, generally 0 dB
SPL – 120 dB SPL;

4/29/2020
15
Limits of Sound Perception (cont.)
 Threshold of Audibility is the acoustic intensity level of a
pure tone that can barely be heard at a particular frequency.
In other words, it is the minimum detectable level of a

4/29/2020
sound in the absence of any external sounds.
 Threshold of audibility ≈ 0 dB at 1000 Hz
 Threshold of feeling ≈ 120 dB
 Threshold of pain ≈ 140 dB
 Immediate damage ≈ 160 dB
 Thresholds vary with frequency and from person-to-person.
 Maximum sensitivity is at about 3000 Hz. 16
Adding Decibels
 Cannot directly add dB
 E.g. 72 dB + 72 dB is NOT 144 dB.

4/29/2020
 Can directly add intensities (W/m2) or pressures (Pa)
 E.g. 5 Pa + 5 Pa is 10 Pa and
 10-7 W/m2 + 10-7 W/m2 is 2 x 10-7 W/m2.
 Therefore, when adding, convert dB into W/m2 or Pa, and
add, and then convert the result back to dB.

17
Adding Decibels (cont.)
Q. The JMU music school has a brass ensemble that consists of
a trumpet (83.4 dB IL), French horn (78.1 dB IL), trombone
(85.0 dB IL), and a tuba (86.3 dB IL). What is the level of the
music when the entire ensemble is playing together?

4/29/2020
Trumpet
 I 
83.4 dB IL  10 log  12 
10 
83.4  I 
 log  12 
10 10 
 I  8.34
10 12   10

I  2.19 10  4W / m 2 18
 Adding Decibels (cont.)
Trumpet All together
83.4 dB IL = 2.19 x 10-4 W/m2 1.03 x 10-3 W/m2

4/29/2020
French horn  (1.03 10 3 ) 
dB IL  10 log  12 
78.1 dB IL = 6.46 x 10-5 W/m2  10 
 90.13 dB IL
Trombone
85.0 dB IL = 3.16 x 10-4 W/m2

Tuba
86.3 dB IL = 4.27 x 10-4 W/m2

19
Adding Decibels (cont.)
Q. You are attending an outdoor lecture held in a temporary
shelter with a tin roof. Of course, you take your sound level
meter with you. You measure the speaker at 77.3 dB IL. It

4/29/2020
then starts to rain, which makes a terrible racket. You
measure the speaker together with the rain at 78.9 dB IL.
What was the level of the rain?

speaker rain alone

77.3 dB IL = 5.37 x 10-5 W/m2 dB IL = 2.39 x 10-5 W/m2
speaker + rain = 73.8 dB IL
78.9 dB IL = 7.76 x 10-5 W/m2
20
 Inverse Square Law
 The law of decreasing power per unit area (intensity) of a
wavefront with increasing distance from the source is known
as the inverse square law, because intensity drops in

4/29/2020
proportion to the inverse square of the distance from the
source. 2
I  1/ r ,

where I is the sound intensity and r is the distance.

21 4/29/2020
 4/29/2020
22
Inverse Square Law (cont.)
Q. Marv is standing 8 feet from a sound source and measures
a sound to be 80 dB SPL. Del is standing 32 feet from the
sound source. How many dB SPL would the sound be

4/29/2020
when it arrives at Del?
dB SPL  20 log d

Del
32 feet ? dB SPL

Marv
80 dB SPL

dB SPL  20 log 32  12 dB

8 23
 80  12  68 dB SPL
Inverse Square Law (cont.)
Q.

4/29/2020
200 feet

2 mil
es

80 dB SPL
? dB SPL

dB SPL  20log d
200
  20 log  34 dB
10560
 80  34  114 dB SPL
24
Assignment I
1. Given four machines producing sound pressure levels of
100 dB, 91 dB, 90 dB, and 89 dB respectively, what is the
total sound pressure level?

4/29/2020
2. If the sound pressure from a rifle shot is measured as 134
dB at 1.25 ft away, how does the sound pressure level
change at 80 ft away?
3. If the sound intensity at 3.0m away from a sound source is
4.0×10-6 W/m2, what is the intensity at (a) 1.5m and (b) 12 m
from the source?
4. Karen measures the sound intensity at a distance of 5.0m
from a lawn-mower to be 3.0×10-2 W/m2. Assuming that the
lawn-mower acts as a point sound source and ignoring the
25
effects of reflection and absorption, what is the total power
of the sound emitted by the mower?
Assignment I (cont.)

5. Each of four people talking, when speaking individually

produce an unknown sound level L1 corresponding to an
unknown intensity I1. When all four talk together, the

4/29/2020
sound level is 70 dB and the total intensity is 4×I1. What
are the values of I1 and L1?
6. Given a sound source 10 ft away from the receiver has an
IL of 90 dB,
(a) what is the sound intensity at the receiver?
(b) what is the sound power of the source?
(c) what is the sound IL at 80 ft from the source?

26
Lecture Contents
Part I:
 What is Digital Signal Processing (DSP)?
 Signal basics

4/29/2020
 Systems and their properties
 Sound creation and its intensity
 Sound intensity in dB scale
Part II:
 Analog to digital conversion
 Sampling
 Quantization
 Encoding 27

 Application areas of DSP

Basic Types of Signals
 Analog (or) continuous-time signal (x(t))
 Continuous values defined along a continuum of times;
 Biological signals are almost always continuous.
 E.g. temperature, pressure, speech, etc.

4/29/2020
Fig. Typical annual vertical wind profile.

28
Basic Types of Signals
 Discrete-time signal (x[n])
 Defined only at discrete times, and consequently, the
independent variable takes on only a discrete set of
values.

4/29/2020
29

Fig. The weekly Dow-Jones stock market index.

Building Block Signals
 Basic building blocks from which many other signals can be
constructed.
1. Unit sample sequence
 often referred to as discrete-time impulse or unit impulse

4/29/2020
sequence;

δ[n-k]=?

0, n≠k, 30
δ[n-k] =
1, n=k.
 Building Block Signals
2. Unit step sequence

4/29/2020
u[n-k]=?

1, n≥k,
u[n-k] =
0, n<k.

Relationship between δ[n] and u[n]

31
 Building Block Signals
3. Complex exponential sequence

x (t ) Ae j0t

4/29/2020
where |A| is the
magnitude of x(t), ω0
is the fundamental
frequency (radians
per second, rad/s);

32
 Building Block Signals
4. Sinusoidal signals
 sine and cosine signals;
 important because virtually every other signals can be thought
of as being composed of many different sine and cosine

4/29/2020
signals;
 to the human ear, a sound that is made of more than one sine
wave will have perceptible harmonics;

x(t) = A cos(ω0t+Φ), for all t,

x(t) = A sin(ω0t+Φ), for all t,

where |A| is the magnitude of x(t), ω0 is the fundamental 33

frequency (rad/s) and ϕ is the phase (rad);
Sine
A

4/29/2020
Cosine

34
Phase of a Sinusoid

4/29/2020
35
Relation between Period and Frequency

4/29/2020
36
Any time-domain signals can be constructed from a
linear combination of different sinusoids.

4/29/2020
37
Beware of phase when you combine sinusoids!!!
Out-of-phase addition (Destructive)

4/29/2020
38
Beware of phase when you combine sinusoids!!!
Constructive addition

4/29/2020
39
Systems
 Systems process input signals to produce output signals.
 Signals contain information that can be used to explain the
underlying physiological mechanisms of a specific event or
system.

4/29/2020
 Mathematical methods can be used to describe signals and to
analyze and synthesize systems.

40
Signal Energy and Power
The total energy of x(t) over the time interval t1 ≤ t ≤ t2
t2 2
 x (t ) dt , where |x(t)| means magnitude of complex number x(t);
t1

4/29/2020
1 t2 2
The time-averaged power is  x (t ) dt
t 2 t1 t1

The total energy of x[n] over the interval n1 ≤ n ≤ n2

n2

 x[n]
2
, where |x[n]| means magnitude of complex number x[n];
n  n1
n2

 x[n]
The time-averaged power is 1 2
n2  n1 1
, 41
n  n1
Linear Systems
 A linear system, in continuous time or discrete time, is a
system that possesses the important property of superposition.
Homogeneity
 Modification in the input signal’s amplitude results in a

4/29/2020
corresponding modification in the output.
T{x(t)} = y(t)
T{ax(t)} = ay(t)
Additivity
 For any two inputs, the sum of the outputs is the same as if
the sum of the two inputs had been processed together.
T{x1(t)} = y1(t), T{x2(t)} = y2(t)
T{x1(t) + x2(t)} = y1(t) + y2(t)

42
Superposition
Continuous time: ax1(t) + bx2(t) = ay1(t) + by2(t)
Discrete time: ax1[n] + bx2[n] = ay1[n] + by2[n]
Practice on Linear Systems
y ( t )  tx ( t )
For x ( t ), y ( t )  tx ( t )
1 1 1
For x ( t ), y ( t )  tx ( t )
2 2 2
For x ( t )  ax ( t )  bx ( t ),

4/29/2020
3 1 2
y ( t )  tx ( t )  t [ ax ( t )  bx ( t )]
3 3 1 2
y ( t )  ay ( t )  by ( t )
3 1 2 2
y (t )  x (t )
The system is linear.
2
For x ( t ), y ( t )  x ( t )
1 1 1
2
For x ( t ), y ( t )  x ( t )
2 2 2
For x ( t )  ax ( t )  bx ( t ),
3 1 2
2 2
y ( t )  x ( t )  [ ax ( t )  bx ( t )] 43
3 3 1 2
y ( t )  ay ( t )  by ( t )
3 1 2
The system is not linear.
Time Invariant Systems
 A system is time-invariant if a time-shift/delay of the input
sequence causes a corresponding shift/delay in the output
sequence.
T{x(t)} = y(t)

4/29/2020
T{x(t-t0)} = y(t-t0)

 Practice on time-invariant systems

y ( t )  sin[ x ( t )] y ( t )  x ( 2t )
By shifting the input by t0, By shifting the input by t0,
y1( t )  sin[ x ( t  t 0 )] y1( t )  x ( 2t  t 0 )
By shifting the outputby t0, By shifting the output by t0,
y ( t  t 0 )  x ( 2 ( t  t 0 ))
y ( t  t 0 )  sin[ x ( t  t 0 )]
y1( t )  y ( t  t 0 ) 44
y1( t )  y ( t  t 0 )
The system is time varying.
The system is time invariant.
LTI Systems & Convolution
 Linearity and time invariance (LTI) play a fundamental role
in signal and system analysis.

4/29/2020
1. A very important way to analyze an LTI system is to study
the output signal when a unit impulse signal is used as an
input.
Impulse response

2. The output signal can be used to infer properties about the

system’s structure and its parameters θ.
45
Discrete-Time LTI systems: The Convolution Sum

 Convolution is an operator that takes an input signal x[n] and

returns an output signal y[n] based on the knowledge about

4/29/2020
the system’s impulse response h[n].

46
 Convolution Illustration (cont.)

4/29/2020
47

Ref: wikipedia
Convolution (cont.)
 Drawback of convolution in time domain
 Convolution is NOT a simple multiplication. Each
output instance y is the SUM of x multiplied with h

4/29/2020
with all time offsets.
 A very simple and time-effective way of performing the
convolution process is using the Fourier Transform.
 The Fourier transform of the convolution is the product of
the two Fourier transforms! − The Convolution Theorem

48
Assignment II

1. Determine if the following systems are linear and time-

invariant.

4/29/2020
(a) y ( t )  x (t 2 )(b) y (t )  x (t  2)

49
Lecture Contents
Part I:
 What is Digital Signal Processing (DSP)?
 Signal basics

4/29/2020
 Systems and their properties
 Sound creation and its intensity
 Sound intensity in dB scale
Part II:
 Analog to digital conversion
 Sampling
 Quantization
 Encoding 50

 Application areas of DSP

 Analog-to-Digital Conversion (ADC)
 The conversion is a two-stage process, which is sometimes
referred to as Pulse Code Modulation (PCM):
1. Sampling (  discrete-time signal)
 obtaining values of signal at discrete instant of time; the more

4/29/2020
samples you take, the more accurate the representation is;
2. Quantization
 replaces amplitude at sampling instant with an approximation
from a finite set of discrete values (levels);
 bit depth >>> no. of bits used to encode a discrete value;
 n-bit bit-depth-> 2n levels
Encoding ( digital signal)
 converts each discrete value into a discrete binary number with
a specified bit-depth; 51
 the higher the bit-depth, the larger the file size but the better the
signal resolution;
x[n] = {6.5, 3.3, 2.0, 2.5, 4.4, 5.0, 3.7, 2.2, 2.0}

4/29/2020
3-bit quantization bit-depth

52

52
 Sampling
 Sampling converts an analog signal into a series of impulses,
each representing the amplitude of the signal at a given
instance in time.
 A discrete-time signal x[n], can be typically obtained by

4/29/2020
periodic sampling of a continuous-time signal xc(t).

x[n] = xc(nT ) where T is sampling period and T=1/fs

where fs is sampling frequency.

The more samples you take, the more accurate the

representation is;
53
 4/29/2020
54
The Effect of Fs on Signal Reconstruction
Example 1  In moving pictures, which consist of a sequence of
individual frames, each of which represents an instantaneous view
(i.e., a sample in time) of a continuously changing scene. When these
samples are viewed in sequence at a sufficiently fast rate, we perceive

4/29/2020
an accurate representation of the original continuously moving scene.
Example 2  Bouncing Ball
If a ball is photographed (sampled) at one instant during a single
bounce, you cannot tell anything about the path of the ball except that
it is off the floor. You can’t tell whether it is going up or down or the
distance of its bounce. If you take more photos at equally-spaced
instants during one bounce, the more accurately you can determine the
path of the ball as it bounces.
55
The Sampling Theorem

4/29/2020
Q. A continuous-time signal xc(t) is sampled at fsample = 8 kHz to
obtain a discrete-time signal x[n]. If the xc(t) is uniquely
recoverable from x[n], what is the highest frequency of xc(t)?
Ans. 4 kHz 56
 Aliasing & Anti-aliasing Filter
 Aliasing occurs when the sampling frequency is not at least twice the
signal frequency.
 If an analog signal contains frequencies above the Nyquist frequency,
these frequencies overlap into the spectrum of the sampled signal and

4/29/2020
interference occurs. This unwanted condition is known as aliasing.
 An alias is a signal which has a frequency that is less than the highest
frequency in the analog signal being sampled and thus falls within
the spectrum or frequency band of the input analog signal causing
distortion.
 Low-pass filtering is necessary to remove all frequency components
of the analog signal that exceed the Nyquist frequency.
 Aliasing can also be avoided by sufficiently increasing the sampling
frequency.
 E.g. If a sound signal is not sampled at a high enough rate, the effect 57
of aliasing will become noticeable with background noise and
distortion.
Assignment III

1. A continuous-time signal xc(t) is sampled at fsample = 400

Hz to obtain a discrete-time signal x[n]. If the xc(t) is not

4/29/2020
uniquely recoverable from x[n], what is the possible highest
frequency of xc(t)?

2. Consider a continuous-time signal xc(t) with the highest

frequency of 8 kHz. In order to perfectly reconstruct xc(t)
from its samples, determine the minimum sampling
frequency that should be used.

58
 Quantization
 The process of mapping a large set of input values (real numbers) to
an approximation from a finite set of discrete values (levels);
 n-bit bit-depth-> 2^n levels;
 bit depth = no. of bits used to encode a discrete value;

4/29/2020
 the higher the bit-depth, the more accurate the representation;
 Rounding and truncation are typical quantization processes.
 Forms the core of all lossy compression algorithms.

Fig. Quantization (rounding).

59
 Quantization (cont.)
sample-and-
analog hold signal
signal
3 Fig. Bit-depth = 2.
2

4/29/2020
1
0

Fig. Bit-depth = 4.

60
Fig. Bit-depth = 4 (± values).
Importance of Bit-Depth Two-bit quantization
Just an example

4/29/2020
Reconstructed

61
Four-bit quantization

4/29/2020
The higher the bit-depth,
the more accurate the
representation;
62
 Quantization Noise/Error
 Quantization is an irreversible process (i.e., the same output
value is shared by multiple input values but it is impossible in
general to recover the exact input value.)
 Quantization noise/error is the difference between the actual

4/29/2020
value and the quantized value.

63
 Bit-Depth vs. Quantization Noise
 The higher the bit-depth, the less the quantization noise and the
better the signal resolution, but the larger the size of the
encoded signal will be. (the less compression ratio)

4/29/2020
64
 Recap: What is DSP?
 Represent signals by a sequence of number
 Sampling or analog-to-digital conversions
 Perform processing on these numbers with a digital processor
 Digital signal processing (DSP)

4/29/2020
 Reconstruct analog signal from processed numbers
 Reconstruction or digital-to-analog conversion

digital digital
analog signal signal analog
A/D DSP D/A
signal signal

65
 4/29/2020
66
Application Areas
 Audio Segmentation
 Segmenting of audio stream into speech, music or silence
 Audio Classification
 Classifying an audio stream into different sound classes
 Audio Indexing

4/29/2020
 To facilitate effective management and use of enormous
amounts of audio data
 Audio Watermarking
 To prevent unauthorized copying and distribution
 Audio Coding
 To obtain compact digital representations of high fidelity
audio signal
 Music Summarization
 To generate a concise and informative content that best
summarizes an original content;
 Music indexing, content-based music retrieval and web- 67
based music distribution;
 Emotion classification
Application Areas
 Electrocardiogram (ECG)
 Electrical activity of heart
 Detection of ECG waveform components
 Segmentation

4/29/2020
 Heart beat type classification
 Phonocardiogram (PCG)
 Heart sounds vibrations
 Segmentation
 Detection and classification
 Electromyography (EMG)
 Human muscle movement
 Muscle activity detection
 Electroencephalogram (EEG)
 Summation of the electrical activity caused by the random
firing of neurons in brain 68
References
[1] Alan V. Oppenheim, Alan S. Willsky, and S. Hamid Nawab,
Signals & Systems, Second Edition, Prentice Hall, 1996.
[2] Thomas L. Floyd, Chapter 12: Signal Conversion and
Processing, Digital Fundamentals, Global Edition, Pearson, 2015.

4/29/2020
69
 4/29/2020
Thank YOU!!!

70