Chapter 2

Signal Sampling and Quantization

Edited by Ndegwa 1
 Introduction
• Even most of signals are in continuous-time domain, they should be converted to a number at different discrete
time to be processed by a microprocessor.

• The process of converting these signals into digital form is called analog-to-digital (A/D) conversion.

• The reverse process of reconstructing an analog signal from its samples is known as digital-to-analog (D/A)
conversion.

maps the continuous

Sampler converts the takes the digital signal
amplitude into a discrete
continuous-time signal into a set of amplitudes and produces a sequence
discrete-time sequence of binary code-words

Components of an analog-to-digital converter (ADC)

Edited by Ndegwa 2
Introduction

8 Bit Code

Digital Processor

Edited by Ndegwa 3
 Sampling
• Periodic or uniform sampling, a sequence of samples 𝑥[𝑛] is obtained from a continuous-time signal 𝑥𝑐 [𝑡] by
taking values at equally spaced points in time. T is the fixed time interval between samples, is known as the
the sampling period.

• The reciprocal 𝐹𝑠 is called sampling frequency (cycles per second or Hz) or sampling rate (samples per second).
Sampling rate Sampling period Example sampling period: T = 125 µs.
Sample per second (Hz) 𝐹𝑠 = 1/𝑇 (second)
sampling rate: 𝐹𝑠 =1/125µs = 8,000 samples per second (Hz).

Sample and Hold

Edited by Ndegwa 4
 Sampling Process
input signal
• The sampling of a continuous-time signal 𝑥[𝑛]
is equivalent to multiply the signal with pulse
train signal 𝑆(𝑡). Pulse Train
signal

Sampled
signal

𝑥 𝑡 Input analog signal Sampled signal ∞

𝑥𝑠 𝑡 = 𝑥 𝑡 ∙ 𝑆(𝑡) = ∑ 𝑥𝑎 𝑛𝑇𝑠 𝛿(𝑡 − 𝑛𝑇𝑠)

𝑆(𝑡) Pulse train 𝑛=−∞

Edited by Ndegwa 5
 Sampling Process – frequency domain 1
Fourier Transform

• The signal 𝑥𝑐 [𝑡] and its spectrum 𝑋𝑐 (𝑗Ω)

Inverse Fourier Transform
(1)

• The sequence 𝑥[𝑛] and its periodic spectrum 𝑋 (𝑗𝜔)

(2) Analog Frequency (Hz)

Normalized Frequency
𝑛 (1)(2) (Cycles/ Sample)
• Since 𝑥[𝑛] is related to 𝑥𝑐 𝑡 𝑤𝑖𝑡ℎ 𝑡 = 𝑛𝑇 =
𝐹𝑠
Sampling Frequency (Hz)

• The desired relationship between sampled signal spectrum 𝑋𝑠 𝐹 and the continuous signal spectrum 𝑋𝑐 (𝐹)

From spectral analysis , and after 𝑋𝑠(𝐹): Sampled signal spectrum

1
some mathematical operations 𝑋𝑐 𝐹 : Original signal spectrum
𝑘=−∞ 𝑋 𝐹 ± 𝑘𝐹𝑠 : Replica spectrum

Edited by Ndegwa 6
 Sampling Process – frequency domain 2
∞
1 1 1 1
𝑋𝑠(𝐹) = ∑ 𝑋𝑐 𝐹 − 𝑘𝐹𝑠 𝑋𝑠 𝑓 = ⋯ + 𝑋𝑐 𝐹 + 𝐹𝑠 + 𝑋𝑐 𝐹 + 𝑋𝑐 𝐹 − 𝐹𝑠 + ⋯
𝑇 𝑇 𝑇 𝑇
𝑘=−∞

• Spectrum of 𝑥[𝑛] is obtained by scaling the spectrum of 𝑥𝑐 [𝑡], putting copies of the scaled spectrum 𝑋𝑐 (𝐹),
1
at all integer multiples of the sampling frequency 𝐹𝑠 = 𝑇 .

• The spectrum of 𝑥[𝑛] can be readily sketched if 𝑥𝑐(𝑡)is assumed to be band-limited. 𝑋𝑐 𝐹 = 0 𝑓𝑜𝑟 𝐹 > 𝐹𝐻

• Two conditions obviously are necessary to prevent

overlapping spectral bands:
1. The continuous-time signal must be band-limited.

2. The sampling frequency Ω𝑠 must be sufficiently large so that:

Spectrum of continuous-time band-limited signal 𝑥𝑐(𝑡)

Edited by Ndegwa 7
 Sampling Process – frequency domain 3
Case 1: Ω𝑠 > 2Ω𝐻

spectrum of discrete-time signal 𝑥 𝑛 = 𝑥𝑐 𝑛𝑇 with Ω𝑠 > 2Ω𝐻

• The sampling operation leaves the input spectrum 𝑋𝑐(Ω) intact when Ω𝑠 > 2Ω𝐻, therefore, it should be possible
to recover or reconstruct 𝑥𝑐(𝑡) from the sequence 𝑥 𝑛 .

• Sampling at Ω𝑠 > 2Ω𝐻creates a guard band which simplifies the reconstruction process in practical applications.

Edited by Ndegwa 8
 Sampling Process – frequency domain 4

spectrum of x[n], showing aliasing distortion, when s Ω𝑠 < 2Ω𝐻

• If Ω𝑠 < 2Ω𝐻, the scaled copies of 𝑋𝑐(Ω) overlap, so that when they are added together, 𝑋𝑐(Ω) cannot be
recovered from 𝑋(Ω) .

• This effect, in which individual terms overlap is known as aliasing distortion or simply aliasing.

Edited by Ndegwa 9
 Sampling Theorem 1
• Question: Are the samples 𝑥[𝑛] sufficient to describe uniquely the original continuous-time signal and, if so, how
can 𝑥𝑐 [𝑡] be reconstructed from 𝑥[𝑛] ? An infinite number of signals can generate the same set of samples.

• Answer: The response lies in the frequency domain, in the relation between the spectra of 𝑥𝑐 [𝑡] and 𝑥[𝑛] .

different continuous-time signals with the same set of sample values

Edited by Ndegwa 10
 1 1 Sampling Theorem 2
𝑇 = 0.01 sec → 𝐹𝑠 = = = 100 𝐻𝑧
𝑇 0.01
One sample
each 0.01 s

The signal is under-sampled 2 fmax=180 > fs

Edited by Ndegwa 11
 Sampling Theorem 3
• An analog signal can be perfectly recovered (reconstruction filter) as long as the sampling rate is at least twice
as large as the highest-frequency component of the analog signal to be sampled (Shannon sampling theorem).

• Let 𝑥𝑐(𝑡) be a continuous-time band-limited signal with Fourier transform:

Then 𝑥𝑐(𝑡) can be uniquely determined by its samples 𝑥 𝑛 = 𝑥𝑐(𝑛𝑇), where 𝑛 = 0, ±1, ±2, … if the sampling
frequency Ω𝑠 satisfies the condition:

2𝜋
𝐹𝑠 = ≥ 2 𝐹𝑚𝑎𝑥
𝑇𝑠
𝐹𝑠
• Half of the sampling frequency is usually called the Nyquist frequency (Nyquist limit), or folding frequency.
2

Example: To sample a speech signal containing frequencies up to 4 kHz, the minimum sampling rate is chosen
to be at least 8 kHz, or 8,000 samples per second.

Edited by Ndegwa 12
 Example 1
Problem:
Euler's
identity

Solution:
Using Euler’s identity,

Hence, the Fourier series coefficients are:

Edited by Ndegwa 13
 Example 1 – Contd.

a.

b. After the analog signal is sampled at the rate of 8,000Hz, the

sampled signal spectrum and its replicas centered at the
frequencies ±n𝑓𝑠 , each with the scaled amplitude being 2.5/T .
Scaled
amplitude
of 2.5/T

Replicas, no
additional
information.
Edited by Ndegwa 14
 Signal Reconstruction
(Digital-to-Analog Conversion)
• The reconstruction process (recovering the analog signal from its sampled signal) involves two steps.

• First: the samples 𝑥 𝑛 (digital signal )are converted

into a sequence of ideal impulses 𝑥𝑠 𝑡 , in which
each impulse has its amplitude proportional to digital
output 𝑥 𝑛 , and two consecutive impulses are
separated by a sampling period of T.
∞

𝑥𝑠 𝑡 = ∑ 𝑥 𝑛 𝛿(𝑡 − 𝑛𝑇𝑠)
𝑛=−∞

• Second: The analog reconstruction filter (ideal

low-pass filter) is applied to the ideally recovered
signal 𝑥𝑠 𝑡 to obtain the recovered analog signal.
The impulse response of the reconstruction filter
is 𝜋𝑡
sin( )
𝑇𝑠
ℎ𝑟 𝑡 = 𝜋𝑡
𝑇𝑠 Edited by Ndegwa 15
 Signal Reconstruction – Contd.
• Before applying the reconstruction filter,
a zero-order hold is used to interpolate
between the samples in xs(t).
Ideal low-pass
filter

• Reconstruction filter

An ideal low-pass reconstruction filter is

required to recover the analog signal
Spectrum ( an impractical case). Practical low-pass
filter

A practical low-pass reconstruction

(anti-image) filter can be designed to reject
all the images and achieve the original signal
spectrum. Edited by Ndegwa 16
 Signal Reconstruction – Contd.

the condition of the Shannon sampling theorem

is violated. We can see the spectral overlapping
between the original baseband spectrum and the Aliasing
spectrum of the replica (add aliasing noise)

 Perfect reconstruction is not possible, even if we use ideal low pass filter.

 if an analog signal with a frequency f is under-sampled, the aliasing frequency

component 𝑓𝑎𝑙𝑖𝑎𝑠 in the baseband is simply given by:

Edited by Ndegwa 17
 Example 2
Problem:

Solution:

Using the Euler’s identity:

Edited by Ndegwa 18
a.

b.
The Shannon sampling theory condition is satisfied

Edited by Ndegwa 19
 Example 3
Problem:

Solution:
a. b.

Edited by Ndegwa 20
 8 Quantization
• During the ADC process, amplitudes of the analog signal to be converted have infinite
precision.

• Quantization : The quantizer converts the continuous amplitude signal discrete amplitude
signal.

• Encoding: After quantization, each quantization level is assigned a unique binary code.

A block diagram for a DSP system

Edited by Ndegwa 21
 8 Quantization – Contd.
• A unipolar quantizer deals with analog signals ranging from 0 volt to a positive reference voltage

∆ : Step size of quantizer (ADC resolution)

𝑥𝑚𝑎𝑥 : Max value of analog signal ∆=
𝐿
𝑥𝑚𝑖𝑛 : Min value of analog signal

𝐿: Number of quantization level

𝐿 = 2𝑚
𝑚 : Number of bits in ADC

𝑖 : Index corresponding to binary code 𝑖 = 𝑟𝑜𝑢𝑛𝑑

∆

𝒙𝒒: Quantization level 𝑥𝑞 = 𝑥𝑚𝑖𝑛 + 𝑖 ∙ ∆ 𝑖 = 0,1, … , 𝐿 − 1

∆ ∆
𝑒𝑞: Quantization error 𝑒𝑞 = 𝑥𝑞 − 𝑥 𝑤𝑖𝑡ℎ − ≤ 𝑒𝑞 ≤
2 2

Edited by Ndegwa 22
 8 Quantization – Contd.
Example:3-bit ADC channel accepts analog input ranging from 0 to 5 volts,

• bipolar quantizer deals with analog signals ranging from a negative reference to a positive
reference.

Edited by Ndegwa 23
 9 Periodicity
• In discrete-time case, a periodic sequence is a sequence for which
where the period N is necessarily an integer.

Discrete sinusoidal: 𝑥 𝑛 = 𝐴 𝑐𝑜𝑠(𝜔𝑛 + 𝜃)

Pulsation: 𝜔 (rad/sample)
Phase shift: 𝜃
𝜔 periodic sequence with N = 7 samples
Frequency: 𝑓 = cycle/sample
2𝜋
1
Period: 𝑁 =
𝑓

𝜋 𝜋/ 1
Example 1:
𝜔 4 1
𝜔= → 𝑓= = =
4 2𝜋 2𝜋 8 → 𝑁 = 𝑓 = 8
Period of N = 8

Edited by Ndegwa 24
 9 Periodicity – Contd.
Example 2: Not periodic with N = 8

3𝜋 𝜔 3 𝜋/8 3 1 16
𝜔= → 𝑓 = 2𝜋 = 2𝜋 = → 𝑁 = = 3×
8 16 𝑓 3
But periodic with N = 16 (Must be integer)

Edited by Ndegwa 25