RMIT Classification: Trusted

EEET2477 Signal and Systems 2

Lecture 3: Analog to Digital Conversion

Dr Alexandru Fechete (CC) and Dr Dao Thanh Hai (Lecturer)

 RMIT Classification: Trusted

Overview

▪ Nyquist’s Theorem
▪ Ideal Sampling & Aliasing
▪ Practical Sampling of Signals
▪ Amplitude Quantization
▪ Encoding Samples of Signals

2
 RMIT Classification: Trusted

Digital Processing of Sound

3
 RMIT Classification: Trusted

Audio/Speech Processing Scheme

4
 RMIT Classification: Trusted

Analog to Digital Conversion

• To transmit an analog signal or process it using digital technology, the signal has to be converted to a digital signal.
This process is known as the Analog-to-Digital (A/D) conversion.

• Low-pass (anti-aliasing) filtering is the process in which the band-limited analog signal is created.

• Sampling is a process in which a continuous-time signal is sampled by measuring its amplitude at discrete time
instants.

• Quantization is a process of mapping the sampled signal amplitude values into a finite set of amplitude levels.
Sampling and quantization operations are often referred to as the pulse code modulation PCM.

• Binary encoding is the process of assigning binary numbers to the discrete amplitude values of the quantized signal.

5
 RMIT Classification: Trusted

Nyquist’s Theorem

Sampling
• A continuous-time signal can be processed by processing its samples
through a discrete-time system.
• For this purpose; it is important to maintain the signal sampling rate
sufficiently high so that the original signal can be reconstructed from these
samples without error (or with an error within a given tolerance).
• The necessarily quantitative framework for this purpose is provided by the
sampling theorem.
 RMIT Classification: Trusted

Band Limited Signals

• A band-limited signal is a signal m(t) for which the Fourier transform M(ω) of
m(t) is equal to zero for all frequencies above a certain frequency ωM :
m(t )  M () = 0 for |  | M = 2f M
• Where M () = F[m(t )]

𝜹(𝒕)
𝒙(𝒕) 𝒉(𝒕)
𝒚(𝒕)
LTI
h(t) 𝒆−𝜶𝒕u(t)
= system

7
 RMIT Classification: Trusted

Sampling

• The ideal sampling is carried out by reading the signal values at discrete time
interval Ts called the sampling interval.
1 2
fs = Hz s = = 2f s rad / sec
Ts Ts

• Where fs is the number of samples per second and it is called the sampling
frequency.

• What criteria have to be met so that the original signal x(t) could be uniquely
reconstructed from the sampled signal m s(t) = m(nTs)?
8
 RMIT Classification: Trusted

Sampling Theorem
• The sampling operation is performed in accordance with the sampling theorem.

• The uniform sampling theorem states that, the real valued, band limited signal m(t) can be
uniquely reconstructed from the sampled signal m s(t) = m(nTs) IF the sampling frequency is:

f s  2 f max

9
 RMIT Classification: Trusted

Sampling Theorem

• The minimum sampling frequency (or rate) is called the Nyquist rate.

min[ f s ] = 2 f M the Nyquist rate (or frequency),

critical sampling frequency

• The maximum sampling interval is called the Nyquist interval.

1
max[Ts ] = the Nyquist interval
2 fM

10
 RMIT Classification: Trusted

Sampling Theorem

• The minimum number of samples that is required to represent a sinusoid is 2 samples per cycle. If fewer samples are
used the phenomenon known as aliasing occurs where a signal of certain frequency may appear as a lower
frequency after reconstruction.

• The solid line shows a sinusoid which is sampled at less than the Nyquist rate. If the signal were to be reconstructed
from the samples, then the lower frequency signal indicated by the broken line could result . 11
 RMIT Classification: Trusted

Sampling: Ideal Sampled Signal

• Multiplication of m(t) by a unit impulse train δTs (t) yields the ideal sampled signal m s(t):

ms (t ) = m(t ) T (t ) =
s


=  m(nT
n = −
s ) (t − nTs )

• The sampled signal m s(t) is a sequence of impulses located at regular intervals Ts seconds and
having strength equal to the values of m(t) at the sampling instants.

12
 RMIT Classification: Trusted

Sampling: Ideal Sampled Signal

• It can be shown that the Fourier Transform of the ideal sampled signal m s(t) is:


1
M s ( ) =
Ts
 M ( − n
n = −
s )

• Where M () = F[m(t )]

• The Fourier transform of the ideal sampled signal consists of an infinite number of copies of the
message signal spectrum M(ω) (original base-band spectrum) placed at the integer multiples of
the sampling frequency ωs.

• M s(ω) is a periodic function of 𝝎 with period ωs.

13
 RMIT Classification: Trusted

Sampling: Ideal Sampled Signal

14
 RMIT Classification: Trusted

Ideal Reconstruction: Frequency Domain

• The spectrum of a band-limited signal m(t)
sampled at the minimum rate : fs = 2fM
samples per sec can be expressed as:

M ( ) = H rec ( ) M s ( )

• If the sampling frequency ω s ≥ 2ω M then,

the base band signal m(t) can be fully
reconstructed by applying a low-pass filter
with the cutoff frequency fc = ωM = ωs/2 to
the sampled signal m s(t).

15
 RMIT Classification: Trusted

Ideal Reconstruction: Frequency Domain

16
 RMIT Classification: Trusted

Ideal Sampling

• To avoid the spectral overleap we must have:

( s −  M ) −  M  0
 s − 2 M  0
 s  2 M there is no overlap between the spectral
components of the sampled signal

s  2M there is an overlap between the spectral components of

the sampled signal. This is called aliasing or undersampling . 17
 RMIT Classification: Trusted

Aliasing

• For example, a component of frequency f = fs/2 + fx that falls into the lost folded yellow tail shows up
or is aliased by a component of lower frequency falias= fs/2 - fx in the reconstructed signal.

• All the spectral components at frequencies > f s /2 reappear as spectral components at frequencies < f s /2.
This spectral tail inversion process is known as spectral folding or aliasing.
• Due to the aliasing process all frequencies above f s /2 up to f max are lost.

18
 RMIT Classification: Trusted

Example 1
An audio signal was passed through a low-pass filter with the cut-off
frequency of 22 kHz.

1. What is the signal bandwidth?

2. What is the minimum sampling frequency given by the Nyquist
criterion that allows full reconstruction of the analog version of this
signal?

19
 RMIT Classification: Trusted

Example 1
An audio signal was passed through a low-pass filter with the cut-off
frequency of 22 kHz.

1. What is the signal bandwidth? (22kHz)

2. What is the minimum sampling frequency given by the Nyquist
criterion that allows full reconstruction of the analog version of this
signal? (min fs = 2 x 22kHz = 44kHz)

20
 RMIT Classification: Trusted

Example 2

What is the bandwidth and the minimum sampling frequency for the
following signal?

x(t) = cos(8000πt+70) – 5sin(2000πt+20)

21
 RMIT Classification: Trusted

Example 2

What is the bandwidth and the minimum sampling frequency for the
following signal?

x(t) = cos(8000πt+70) – 5sin(2000πt+20)

ω1 = 2π x 4000 therefore f1 = 4kHz

ω2 = 2π x 1000 therefore f2 = 1kHz

The signal bandwidth is 4kHz, therefore the min fs = 8kHz

22
 RMIT Classification: Trusted

Example 3

Consider a band limited signal with 5kHz bandwidth sampled at

8KHz.
fs = 8kHz and fmax= 5kHz

• What is the minimum sampling frequency given by the Nyquist

criterion?
• Is aliasing occurring in this case?

IF YES then:

• What is the frequency range of the folded spectral tail?

• What is the alias frequency for f = 4.5kHz in the sampled signal?
23
 RMIT Classification: Trusted

Example 3
Consider a band limited signal with 5kHz bandwidth sampled at 8KHz.
fs = 8kHz and fmax = 5kHz

• What is the minimum sampling frequency given by the Nyquist criterion?

fNq = 2 x fmax = 2 x 5kHz = 10kHz

• Is aliasing occurring in this case?

YES

IF YES then:

• What is the frequency range of the folded spectral tail?

From fs/2 to fmax. This is from 4kHz to 5kHz

• What is the alias frequency for f = 4.5kHz in the sampled signal?

fx= f – fs/2 = 4.5kHz - 4kHz = 0.5kHz therefore
falias = fs/2 – fx = 4kHz – 0.5kHz = 3.5kHz
24
 RMIT Classification: Trusted

Practical Sampling
Practical Sampling: Natural Sampling
• Although instantaneous sampling is a convenient method, a more practical way sampling a band-
limited analog signal m(t) is performed by high-speed switching circuits.

• The sampled signal xns(t) can be written as : xns (t ) = m(t ) x p (t )

• Where xp(t) is the periodic train of rectangular pulses with period Ts, and each rectangular pulse
in xp(t) has width d and unit amplitude.

• The sampling here is called natural sampling, since the top of each pulse in xns(t) retains the
shape of its corresponding analog segment during the pulse interval.
25
 RMIT Classification: Trusted

Practical Sampling : Natural Sampling

26
 RMIT Classification: Trusted

Practical Sampling: Natural Sampling

27
 RMIT Classification: Trusted

The Aperture Effect

• The flat-top sampling is equivalent to passing an ideal sampled signal through a filter
having the frequency response H(ω) = P(ω).
• The high frequency roll-off characteristic of P(ω) acts as a low-pass filter and attenuates
the upper portion of the message spectrum.
• The loss of the high frequency content is known as the aperture effect.
• The larger the sampling pulse duration d, the larger the effect. Thus, the aperture
effect can be reduced by making d smaller.

• In practice the aperture effect can be neglected as long as:

28
 RMIT Classification: Trusted
Amplitude Quantization and Binary
Encoding
Quantization: Uniform (Linear) Amplitude Quantization
• In the uniform quantization process, the sampled analog signal is mapped into a finite set of equidistant amplitude
levels.

• We assume that the amplitude of the sampled signal xs(t) is confined to the dynamic range (-m p, m p). This range is
divided into N zones, each of step size Δ called the quantization step or the resolution of the quantizer, and it is
given by:

29
 RMIT Classification: Trusted
Quantization: Uniform (Linear)
Amplitude Quantization
• A sample amplitude value is approximated by the midpoint of the interval in which it lies.

• There is a discrepancy between the quantized signal xsq(t) and the sampled signal
xs(t). Thus, the quantization process introduces an error. The larger the quantization
step size Δ, the larger quantization error is introduced.
30
 RMIT Classification: Trusted

Uniform Quantization: Noise (Error)

• The difference between the input and the output signals of the quantizer becomes the
quantizing error, or quantizing noise. For a random input signal, the quantizing error qe
varies randomly within the interval:
− 
 qe 
2 2
• Assuming that the error is equally likely
to lie anywhere in the range:
− 
 , 
 2 2
• The mean-square quantizing error per
quantization interval Δ, is given by:
𝑚𝑝 2
2
𝑞𝑒 =
3𝑁2

• The larger is the quantization step Δ, the larger the error.

• The larger is the number of quantization levels N, the smaller is the error.
31
 RMIT Classification: Trusted

Quantization: Signal-to-Noise Ratio

• Thus, for an n bit quantizer the quantization signal-to-noise ratio (SNR) is
defined as :
mean squared signal 3 2 n
SNR = = 2
mean squared noise 2

• Where:

n = log 2 N

• or expressed in dB:
3 
SNRdB = 10 log10  2 2 n  = 6.02n + 1.76 dB
2 
32
 RMIT Classification: Trusted

Quantization: Signal-to-Noise Ratio

• Thus, for an n bit quantizer the quantization signal-to-noise ratio (SNR) is

defined as : mean squared signal 3 2 n
SNR = = 2
mean squared noise 2

• Where: n = log 2 N

2 𝑚𝑝 2 2𝑚𝑝 2
𝑞𝑒 = mean squared signal = (Assuming a sine wave input)
3𝑁2 2 2

• or expressed in dB:
3 
SNRdB = 10 log10  2 2 n  = 6.02n + 1.76 dB
2  33
 RMIT Classification: Trusted

Quantization: Uniform (Linear) Quantization

• The input-output characteristics of a uniform quantisation are:

• The graph shows the output value that the quantizer assigns to a given value
of the input signal. Note : that the quantization step Δ is the same for all values of the input signal.
34
 RMIT Classification: Trusted

Quantization: Non-Linear Quantization

• For many signals the uniform quantizing is not efficient.
• For example in speech, the basic problem with linear quantisation is that the constant
quantization step size will usually map quite precisely amplitudes of large peaks (high
SNR) in speech signal but for low-amplitude speech components it will have a large
quantization error (small SNR).

• To make the SNR distribution more uniform across signal amplitudes, logarithmic PCM
systems are used with quantization step constant in the logarithmic scale.

• This gives a small quantization step at low amplitudes and large at high
Continuous-Tim Signal
amplitudes.
• This approach gives the same quantization SNR for all amplitude levels.

35
 RMIT Classification: Trusted

Quantization: Non-Linear Quantization

36
 RMIT Classification: Trusted

Encoding
• An encoder translates the quantized samples into code numbers which are then
converted into binary sequences.

37
 RMIT Classification: Trusted

Bandwidth Requirements for PCM

Suppose that N quantizing levels are used, satisfying:
N = 2n and n = log 2 N
For this case each sample is represented by n binary pulses (n-digit codeword).
If the sampling rate is 𝑓 𝑠 ≥ 2𝑓 𝑀 , then 𝑛𝑓 𝑠 binary pulses must be transmitted per
second.
And that should be equal to 2 times the maximum frequency of the PCM signal:
Thus nf s = 2 f PCM
n n
Or
f PCM = f s  2 f M = nf M
2 2
The minimum required bandwidth for PCM is proportional to the message signal
bandwidth fM and the number of bits per symbol n.
38
 RMIT Classification: Trusted

Bandwidth Requirements for PCM

Therefore we need to transmit out of the binary encoder at 𝒏𝒇𝒔 bits per seconds
Transmission is via a binary signal 𝑚 𝑏 (𝑛) and if we assume serial transmission then
the minimum bandwidth 𝑚 𝑏 (𝑛) is defined by 𝑓 PCM :
n n
f PCM = f s  2 f M = nf M
2 2
39
 RMIT Classification: Trusted

Questions
• What is the minimum sampling rate (Nyquist rate) for a speech signal with the 4kHz bandwidth?

• What is the minimum sampling frequency in Hz (Nyquist rate) for the following signal:
f(t)=sin(2000πt)+cos(4000πt) ?
• An A/D speech conversion system has an anti-aliasing low pass filter with the cutoff frequency of
8kHz. What is the minimum sampling frequency (Nyquist’s rate) for this system?

• An A/D speech conversion system uses the sampling frequency of 44kHz. What is the highest cut-off
frequency of an appropriate anti-aliasing filter that can be used by this system?

• Which practical sampling system has a smaller aperture effect; one with the 5 milliseconds sampling
pulse or one with the 10 milliseconds sampling pulse?

• An analog speech signal with 4kHz bandwidth was sampled at 8kHz rate and amplitude-quantized
using PCM with 256 amplitude levels. What is the bandwidth of the amplitude-quantized signal?

40
 RMIT Classification: Trusted

Questions
• What is the minimum sampling rate (Nyquist rate) for a speech signal with the 4kHz bandwidth?
8kHz

• What is the minimum sampling frequency in Hz (Nyquist rate) for the following signal: f(t)=sin(2000 πt)+cos(4000πt)
4kHz

• An A/D speech conversion system has an anti-aliasing low pass filter with the cutoff frequency of 8kHz. What is the
minimum sampling frequency (Nyquist’s rate) for this system?
16kHz

• An A/D speech conversion system uses the sampling frequency of 44kHz. What is the highest cut-off frequency of
an appropriate anti-aliasing filter that can be used by this system?
22kHz

• Which practical sampling system has a smaller aperture effect; one with the 5 milliseconds sampling pulse or one
with the 10 milliseconds sampling pulse?
5 milliseconds

• An analog speech signal with 4kHz bandwidth was sampled at 8kHz rate and amplitude -quantized using PCM with
256 amplitude levels. What is the bandwidth of the amplitude-quantized signal?
32kHz
41
 RMIT Classification: Trusted

Acknowledgment

• I would like Thank and to Acknowledge the support of : A/P Margaret Lech,
the course coordinator of EEET2113 Signal and Systems 2 course from RMIT
University Melbourne.

42
 RMIT Classification: Trusted

Copyright © 2022 RMIT University Vietnam