1

SBES152
Biomedical Signal
Processing

Lecture 02
Signal Sampling and Quantization

Shereen El-Metwally [email protected]

Credit: Adapted from slides by Dr. Muhammed

Rushdi
2

DSP General Scheme

The ADC unit samples the analog signal, quantizes

the sampled signal, and encodes the quantized
signal level to the digital signal.
3

Sampling Terminology

Sampling interval = T (or Ts)

Sampling frequency =

Or rad/sec
Sampling Terminology

Sample-and-hold analog
voltage for ADC.
4

Sampling Motivation
• Many of the real-world signals are analog.

 Analog signals contain an infinite number of

points.
 They cannot be processed by the digital
signal (DS) processor or computer: infinite
memory + infinite processing power.

• Sampling issues
Sampling in time + quantization in
amplitude
 How to sample in time (sampling period)
Uniform Sampling
Continuous-time
signal xc(t) t

“Windowing”
Δ
As Δ  0
A very thin window
(impulse)

“Sampling”

Discrete -time
signal x[n]
-3 -2 -1 0 1 2 3 4 n
Uniform Sampling
Sampling is, in general, not reversible
1

0.5

-0.5

-1
0 20 40 60 80 100

 Fundamental issue in digital signal processing

 We don’t know what happens between samples.

 If we lose information during sampling, we cannot recover it

 Under certain conditions an analog signal can be sampled

without loss so that it can be reconstructed perfectly
5

Uniform Sampling

• Sampling function
▫ Periodic sequence of impulses of
period Ts

• Sampled signal:

How would the sampled signal look like in the Frequency

Domain???
6

Uniform Sampling
•
7

Uniform Sampling
• Notice that Dk can be obtained by integration
over any period, say [-Ts/2, Ts/2]

Only one term

from the impulse
sequence lies
inside the
integration
interval
Sifting property of delta
function:
∫δ(t) f(t) dt = ∫δ(t) f(0) dt = f(0)
8

Uniform Sampling

Hint: Use the

properties of FT:
1. Linearity

2. Frequenc
y Shifting
1  δ(ω)
ejωot  δ(ω - ωo )

Perio
d
= Ts
The Fourier transform of a periodic train of equidistant delta
functions in the
Uniform Sampling
Samples should be collected at a rate high enough that
the original analog signal can be reconstructed or
recovered later.
T = 0.01 or fs = 100 Hz

fs > 2
fmax

fs < 2 fmax

90 and 10 Hz signals are called “aliases” or indistinguishable of

14

How would the sampled signal

look like in the Frequency
mag
Domain??
Fourier Transform

t f (Hz)
“multiplication” x “convolution”
*

t
T f (Hz)
= fs = 1/T
mag =

-3 -2 -1 0 1 2 3 4 n f (Hz)
fs = 1/T
9

Uniform
Sampling
• Back to our question…
“The sampled signal in the frequency
domain”

Sampling
1
0

Sampling Rate Calculation

• Requirement: No Aliasing
allowed
1
1

Sampling Rate Calculation

Nyquist criterion
1
2

Sampling Rate Calculation:

Example

• The sampled signal can be

represented by:

where,

• Using Ts = 0.4,
1
3

Sampling Rate Calculation:

Example
• Is it periodic?
• What is the period?
New
• Draw using Matlab period
(how?):

Ts = 0.4 sec
1
4

Sampling Rate Calculation:

Example
•

2 >

Nyquist sampling condition is satisfied for T = 0.5, 0.4, and 0.2. But
it is not satisfied for T = 1.
1
5

Sampling Rate Calculation:

Example
T = 0.2 sec
s T = 0.4 sec
s

Ts = 0.5 sec Ts = 1 sec

1
6

Sampling Rate Calculation:

Example

Solution (a):
 The signal is unit pulse.
 Clearly, this signal can be easily sampled by
choosing any value of Ts << 1. ⇒
1
 But, notice that 𝑢(𝑡 )❑
𝑗Ω
Sinc function has no maximum frequency! Not band-
limited! Thus, any chosen value of Ts will cause
aliasing.
1
7

Sampling Rate Calculation:

Example
 Fortunately, the values of the sinc function go
fast to zero, so we can compute an approximate
maximum frequency that covers 99% of the
energy of the signal.
 Using Parsval’s energy relation: The energy of
x1(t) is the area under x12(t)

 This is difficult to calculate…. But can be

approximated.
1
8

Sampling Rate Calculation:

Example
Approximatin
g

a=0;
for k=0:0.001:20
a=a+0.001*sinc(0.5*k)
^2; if a>0.495
k
retur
n
en
d end
2
0

Reconstruction of the Original

Continuous-Time Signal from
Samples
• Assuming band-limited signal and sufficient
sampling, original signal can be recovered by low-
pass filtering.


2
1

Reconstruction of the Original

Continuous-Time Signal from
Samples
• Given,
• Then,
(1)
• Where,

(2)

• Substitute from (2) into (1),

xr(t)

Sinc-interpolation
Ideal Reconstruction Filter

Ideal LPF with cut off frequency of  c= /T

(rad/sec) or fc=1/2T

sin ( 𝜋 𝑡 /𝑇 𝑠 )
h𝑟 (𝑡 )=
𝜋 𝑡 /𝑇 𝑠

29
Reconstructed Signal

sinc function is 1 at t=0

shifted sinc functions at nTs

∞
sin [ 𝜋 ( 𝑡 −𝑛𝑇𝑠 ) /𝑇𝑠 ]
𝑥𝑟 ( 𝑡 )= ∑ 𝑥 [ 𝑛𝑇 𝑠 ]
𝑛=− ∞ 𝜋 (𝑡 − 𝑛𝑇𝑠 ) /𝑇𝑠

The recovered signal is thus an “interpolation” in •

30
terms of time-shifted sinc signals with
2
3

Sampling of Infinite Bandwidth

Signals: Anti-Aliasing Filter
• Anti-aliasing filter: a low-pass filter applied to
the input signal to make sure that the signal to
be sampled has a limited bandwidth.
▫ Applied in all practical analog-to-digital
converters
2
4

Reconstruction of the Original

Continuous-Time Signal from
Samples
• In practice, the exact recovery of the original
signal may not be possible for several reasons:
▫ The continuous-time signal is not exactly band
limited, so there is no max. freq.
▫ The sampling is not done exactly at
uniform times— random variation of the
sampling times may occur.
▫ An ideal low-pass filter cannot be realized.
2
5

The Nyquist-Shannon Sampling

Theorem