ECE 3331b

Introduction to Signal Processing

Lecture 3

Instructor: Dr. Ilia G. Polushin

Department of Electrical & Computer Engineering

Faculty of Engineering

The University of Western Ontario

Winter 2017
Topic 1: Introduction and Background

Signals and their classification

The Concept of Frequency in Continuous-Time and Discrete-Time Sig-

nals

Analog-to-Digital and Digital-to-Analog Conversion

Analog-to-Digital and Digital-to-Analog Conversion

To process analog signals by digital means, it is first necessary

to convert them into digital form, i.e., a sequence of numbers
(A/D conversion)

In many cases, the processed signal should be converted back

into analog form (D/A conversion).
Analog-to-Digital Conversion

Conversion into digital form (A D):

Sampling
Quantization
Coding
Sampling of Analog Signals
Consider a continuous-time (CT) signal

xa (t), < t < +.

We address periodic or uniform sampling of CT signal xa (t), described by the formula:

x(n) := xa (nT ), n {. . . , 2, 1, 0, 1, 2, . . .}.

where x(n) is the resulting DT signal, and T > 0 is sampling period (sampling interval).
Fs := 1/T is called sampling rate or sampling frequency
Continuous-Time vs. Discrete-Time Sinusoids
Some facts from lecture 2:

Continuous-time sinusoids

xa (t) = A cos (2F t + ) , < t < +

with frequencies < F < + are all distinct.

Discrete-time sinusoids

x(n) = A cos (n + ) , < n < + (DT )

whose frequencies are separated by 2k, where k is an integer, are identical.

Equivalently, discrete-time sinusoids

x(n) = A cos (2f n + ) , < n < + (DT f)

whose frequencies f are separated by an integer number k, are identical.

Discrete-time sinusoids are distinct only within the following fundamental range of frequencies
1 1
or <f < (F R)
2 2

For any discrete-time sinusoid with frequency outside the fundamental range, there exists an identical
sinusoid with frequency from within the fundamental range
 Sampling of Analog Signals
Let xa (t) be a continuous-time sinusoidal signal of the form:

xa (t) = A cos (2F t + ) . (CT )

Sampling (CT) at a rate Fs = 1/T , one gets

x(n) := xa (nT ) = A cos (2F nT + ) = A cos (2f n + ) , (DT )

where
F
f=
Fs
Discrete-time sinusoids with frequencies f separated by an integer number are identical

!
If F1 and F2 are such that
F1 /Fs F2 /Fs = k,
where k is an integer number, then continuous-time sinusoidal signals with frequencies F1
and F2 after sampling at a rate Fs will result in identical discrete-time sinusoids!
This effect (where different continuous-time sinusoidal signals become indistinguishable after
sampling) is called aliasing
Sampling of Analog Signals
Example:
Consider two CT sinusoids:

1 7
x1 (t) = cos 2 t, x2 (t) = cos 2 t.
8 8

Sampling rate Fs = 1 Hz.

Frequencies F1 = 1/8 and F2 = 7/8 satisfy

F1 /Fs F2 /Fs = 1/8 ( 7/8) = 1.

Corresponding DT sinusoids are

1
x1 (n) = cos 2 n = cos n ,
8 4

7 7
x2 (n) = cos 2 n = cos n = cos n .
8 4 4

We have x1 (n) x2 (n), even though x1 (t) = x2 (t).

Sampling of Analog Signals

Example of aliasing:
Two CT sinusoids with frequencies F0 = 1/8 Hz and F1 = 7/8 Hz
sampled at Fs = 1 Hz.

Figure 1.4.5 from Proakis & Manolakis, 2007

 Sampling of Analog Signals

Let xa (t) be a continuous-time sinusoidal signal of the form:

xa (t) = A cos (2F t + ) . (CT )

Sampling (CT) at a rate Fs = 1/T , one gets

x(n) := xa (nT ) = A cos (2F nT + ) = A cos (2f n + ) , (DT )

where
F
f=
Fs
Discrete-time sinusoids are distinct only within the fundamental range of frequencies
1 1
<f < .
2 2

Thus, given sampling rate Fs > 0, the one-to-one correspondence between CT an DT

sinusoids exists only if
1 F 1
< < ,
2 Fs 2

!
or, equivalently,
Fs Fs
<F < .
2 2
Sampling of Analog Signals
Summary:

CT sinusoids
Fs Fs
xa (t) = A cos (2Fk t + ) , Fk = F0 + kFs , F0 , k = 0, 1, 2, . . .
2 2
sampled at a rate Fs all result in the same DT sinusoid

F0 + kFs
x(n) := xa (nT ) = A cos 2 n+
Fs

= A cos (2nF0 /Fs + + 2kn) = A cos (2nf0 + ) ,

where f0 = F0 /Fs .
CT frequencies Fk = F0 + kFs , k = 1, 2, . . . are indistinguishable from F0 after sampling with
rate Fs . Thus, they are aliases of F0 .
The one-to-one correspondence between sinusoidal signals before and after sampling exists if
and only if
Fs Fs
<F < ,
2 2
where F is frequency of the continuous-time sinusoidal signal and Fs is sampling rate.
Sampling of Analog Signals

Relationship between CT frequency F and DT frequency f for a given sampling rate Fs :

The frequency Fs /2 is called folding frequency.

Sampling of Analog Signals

Example 1.4.2 (Proakis & Manolakis, 2007): Consider the analog signal

xa (t) = 3 cos(100t).

Problem (a): Determine the minimum sampling rate required to avoid aliasing.
Solution. The frequency of the continuous-time signal xa (t) is F = 50 Hz. To avoid aliasing, the
sampling rate must be at least twice higher than F . Therefore, the minimum Fs = 100 Hz.

Problem (b): Suppose the signal is sampled at Fs = 200 Hz. What is the DT signal obtained
after sampling?
Solution: n
x(n) = 3 cos 100 = 3 cos n .
200 2
Sampling of Analog Signals
Example 1.4.2 (Proakis & Manolakis, 2007): Consider the analog signal

xa (t) = 3 cos(100t).

Problem (c): Suppose the signal is sampled at Fs = 75 Hz. What is the DT signal obtained after
sampling?
Solution:

n 4 2 2
x(n) = 3 cos 100 = 3 cos n = 3 cos n = 3 cos n .
75 3 3 3

Problem (d): What is the frequency 0 < F < Fs /2 of a CT sinusoid that yields samples identical
to those obtained in part (c)?
Solution: Continuous-time sinusoidal signal x(t) = 3 cos(2F t) after sampling with Fs = 75 Hz
must satisfy

n 2F 2 2F 2
x(n) = 3 cos 2F = 3 cos n = 3 cos n ) = ) F = 25 Hz.
75 75 3 75 3
Sampling of Analog Signals

Question: Given an analog signal, how to select the sampling rate Fs ?

Some information regarding frequency content of the signal is neces-

sary!

In particular, we need to ensure that the signal does not have significant
frequency content above certain frequency Fmax (i.e., sinusoids with fre-
quencies F > Fmax are negligibly small in amplitude/power)

Example: Speech signals generally do not contain significant content

above 3000 Hz
Sampling of Analog Signals

CT signal:
N
X
xa (t) := Ai cos (2Fi t + i )
i=1

We need to ensure that

max Fi Fmax ()
i=1,...,N

In particular, (*) can be achieved by filtering of CT signal prior to sam-

pling.

Sampling rate Fs should be selected such that

Fs > 2Fmax

Why? To avoid aliasing. More specifically, to guarantee that

1 Fi 1
fi := for all i = 1, . . . , N.
2 Fs 2
Sampling of Analog Signals
If Fs > 2Fmax , the continuous-time signal can be reconstructed exactly.
Sampling Theorem: Suppose the continuous-time signal can be represented as a sum of
sinusoids, as follows
N
X
xa (t) := Ai cos (2Fi t + i) , max Fi Fmax := B.
i=1,...,N
i=1

Suppose the CT signal is sampled at a rate Fs > 2B. Then xa (t) can be exactly recovered
from its sample values x(n) := xa (nT ), T = 1/Fs .
Fn := 2Fmax = 2B is called Nyquist rate.

One possible reconstruction algorithm:

+1
X sin 2Bt
xa (t) = xa (nT )g(t nT ), where g(t) = .
n= 1
2Bt

The above reconstruction formula involves computing infinite series for each t; it is mainly
of theoretical interest. Practical reconstruction formulas exist (see, for example, Chapter 6,
Proakis & Manolakis, 2007)
Sampling of Analog Signals

Problem 1.8 (from Proakis & Manolakis, 2007):

An analog electrocardiogram (ECG) contains useful frequencies up to 100 Hz.
Question (a) What is the Nyquist rate for this signal?
Answer. Fs = 2Fmax = 200 Hz.
Question (b) Suppose that we sample this signal at a rate of 250 samples/s.
What is the highest frequency that can be represented uniquely at this sam-
pling rate?
Answer. Fmax = Ff old = Fs /2 = 125 Hz
Analog-to-Digital Conversion

Conversion into digital form (A!D):

Sampling
Quantization
Coding
Quantization
Quantization is the process of conversion of a discrete-time continuous
amplitude signal into a digital (discrete-time discrete amplitude) signal.

Quantization is usually performed by rounding of the signals amplitude

to the nearest quantization level:

xq = Q [x(n)]
Quantization
Sampling

Illustration of sampling & quantization processes:

Analog signal xa (t) = 0.9t , sampling period Ts = 1 s, = 0.1.
Quantization
The distance > 0 between two successive quantization levels is called quan-
tization step or quantizer resolution

Quantizer resolution:
xmax xmin
= ,
L 1
where L is the number of quantization levels, and xmax , xmin are the maximum
and the minimum values of x(n).
Quantization
The error introduced by quantization is called quantization error or quan-
tization noise:
eq (n) = xq (n) x(n).

For quantization by rounding, the quantization error satisfies

eq (n) .
2 2

Quantization error decreases as quantization step decreases.

Sampling vs. Quantization

Sampling Quantization

Sampling vs. Quantization:

Sampling of a continuous time-signal does not lead to a loss of information

if Fs > 2Fmax .

Quantization is generally an irreversible process and always results in a loss

of information.
Analog-to-Digital Conversion

Conversion into digital form (A!D):

Sampling
Quantization
Coding
Coding

Coding is the process of assigning of a unique binary number to each quantization level

With a word of length b > 0 on can create 2b binary numbers. Thus, the number b of bits
required for coder must satisfy
b log2 L,
where L is the number of quantization levels.
Coding

Problem 1.10 (from Proakis & Manolakis, 2007):

A digital communication link carries binary-coded words representing samples of an
input signal
xa (t) = 3 cos 600 t + 2 cos 1800 t.
The link is operated at 10,000 bits/s and each input sample is quantized into
1024 dierent voltage levels.
Question (a) What is the (maximum possible) sampling frequency?
Answer. Number of bits per sample=log2 1024 = 10. Fs = 10, 000/10 = 1000 Hz.
Question (b) What is the Nyquist rate for the signal xa (t)?
Answer. Fmax = 900 Hz. FN = 2Fmax = 1800 Hz.
Question (c) What is the resulting discrete-time signal x(n)?
Answer. x(n) = 3 cos (2 (0.3)n) + 2 cos (2 (0.1)n) .
Question (d) What is the (minimum) resolution of the quantizer ?
Answer. = (xmax xmin ) /(L 1) = (2 5)/(1024 1) = 10/1023
Digital-to-Analog Conversion

Digital-to-Analog conversion is usually performed by the following three components:

Digital Sample Low-pass Analog

D/A
input signal & hold smoothing output signal
Converter
filter

D/A Converter accepts a binary word and produce an output proportional to the value
of the binary word

Sample-and-Hold (S/H) Interpolator holds the signal constant between samples

(other types of interpolation are possible)

Response of an S/H interpolator

to a DT sinusoidal signal
Figure 6.3.8 from
Proakis & Manolakis, 2007

Low-Pass Filter smooths the output of S/H to remove discontinuities.

Content:
Introduction: Signals, Systems and Signal Processing, Classification of Signals, The Concept
of Frequency in Continuous-Time and Discrete-Time Signals, Analog-to-Digital and Digital-to-
Analog Conversion.

Discrete-Time Signals and Systems: Discrete-Time Signals, Discrete-Time Systems, Anal-

ysis of Discrete-Time Linear Time-Invariant (LTI) Systems, Discrete-Time Systems Described
by Difference Equations, Implementation of Discrete-Time Systems

The z-Transform and its Application to the Analysis of LTI Systems: The z-Transform,
Properties of the z-Transform, Rational z-Transforms, Inversion of the z-Transform, Analysis of
LTI Systems in the z-Domain, The One-sided z-Transform

Frequency Analysis of Signals: Frequency Analysis of Continuous-Time Signals Frequency

Analysis of Discrete-Time Signals Properties of the Fourier Transform for Discrete-Time Signals

Frequency-Domain Analysis of LTI Systems: Frequency-Domain Characteristics of LTI

Systems, Frequency Response of LTI Systems

The Discrete Fourier Transform: Frequency-Domain Sampling: The Discrete Fourier Trans-
form (DFT), Properties of the DFT

The Fast Fourier Transform