Sampling and Reconstruction

Introduction
This section introduces the relationship between continuous signals/systems and their representation as
discrete-time signals/systems that allows the direct use of digital computing technology. Sampling forms
the bridge between these two representations of the world.

Sampling
Analog Discrete-Time
System System
Reconstruction

A Discrete-Time system has as its input and output sequences of values. Here it is assumed that these
sequences represent samples of analog signals that are equally spaced in time (Ts seconds apart). Extensions
to applications involving functions of other variables (e.g. distance, force, pressure. velocity, pixels, …) will
be apparent.

x(n*Ts) y(n*Ts)
Discrete-Time
System
xn yn

In order to develop the analysis tools for such a system, we must form an equivalent analog system.
x(t) x(t, Ts)
y(t)
Sample Re-Sample Reconstruct
(Ts sec.) Analog (Ts sec.)
System
Where:
x n = x (n * Ts ) = x (t ) t = n*T
s

and
y n = y (n * Ts ) = y (t ) t =n*T
s

Analyzing this equivalent system using standard analog tools will establish a set of discrete-time tools
including the Discrete Fourier Transform (DFT) and the z-Transform.
Another reason for examining the sampling/reconstruction process is that they form the actual hardware
interface between analog and discrete-time systems. We will then be prepared to perform Digital Filtering
of analog signals.

J. N. Denenberg Copyright DTS September 27, 2001

Sampling Tutorial Page 2 of 6

Sampling
Pulse Amplitude Modulation
PAM is where the analog signal after sampling takes on the form of a sequence of pulses Ts seconds apart.
Each pulse carries information about the analog signal’s amplitude at the time the pulse was generated.
There are two variants of PAM. The first, currently the dominant approach, called Natural Sampling is
where the shape of each pulse is affected by the changing input during the pulse. The second is called
Uniform Sampling (mainly used in the older laboratory grade samplers) where all of the pulses have the
same shape, but different amplitudes. We will analyze both.
Natural Sampling in the Time Domain
Assume that we have a generalized, time-limited pulse centered at t = 0 as shown below.

p(t)

-2Ts -Ts 0 Ts 2Ts

The Fourier Transform of this pulse is to denoted P(jω). A periodic version of this pulse where the original
pulse now repeats every Ts seconds is then:

J. N. Denenberg Copyright DTS September 27, 2001

Sampling Tutorial Page 3 of 6

pT(t)

1.2
A
1
0.8
0.6
0.4
0.2
0
-30s
-2T -25 -20 -15
-Ts -10 -5 0 5 10 15
Ts 20 25 30s
2T

[ ]
Note: lim pTs (t ) = p (t )
Ts → ∞

pTs (t ) , the pulse train, is periodic with period Ts and we can express it as a sum of individual pulses.
∞
pTs (t ) = ∑ p(t − n * T ) s
n = −∞

“Natural Sampling” is then done by multiplying x(t), the input signal, by pTs (t )

x(t) ∞
x (t , Ts ) = ∑ x (t )* p(t − n * T )
*
s
n = −∞

pTs (t )

J. N. Denenberg Copyright DTS September 27, 2001

Sampling Tutorial Page 4 of 6

Uniform Sampling in the Time Domain

The second type of sampling, Uniform Sampling, results in a train of pulses in which each pulse has the
same shape. Each pulse has a “strength” which is set by the value of the input at the sampling “instant”.
This method also goes by the name “Flat Top Sampling” when the transmitted pulses are rectangular. One
way to implement uniform sampling is to precede a natural sampler with a “Sample and Hold” circuit.

x(t) xsh(t) ∞
x (t , Ts ) = ¦ x (n * T )* p(t − n * T )
S&H
*
s s
n = −∞

pTs (t )

The Sample and Hold circuit has the following effect on an analog signal.

2.5

1.5
xsh(t)
x(t)
1

0.5

0
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
1
2
3
4
5
6
7
8
9

This is a non-linear distortion of the signal, but now the signal is constant during each pulse of the pulse
train and multiplying leaves all of the resulting pulses having the same shape. A delay of ½ Ts must be
introduced in the pulse train to center each pulse in a constant region of the modified signal. This has the
effect of delaying the signal itself by Ts/2.

J. N. Denenberg Copyright DTS September 27, 2001

Sampling Tutorial Page 5 of 6

Frequency Domain Analysis of Sampled Signals

Introduction
Both types of sampling are practical. One, Natural Sampling, is simpler in the frequency domain while the
other, Uniform Sampling, is simpler in the time domain. In the limiting case, as the pulse width approaches
zero while holding the pulse area constant, both systems yield the same result. This limiting case is called
Ideal (or Impulse) Sampling where each pulse is now the “Impulse” or “Dirac Delta” function, δ (t − n * Ts ) ,
and is the form of sampling described in many textbooks.
Natural Sampling in the Frequency Domain
Let P(jω) be the Fourier transform of p(t), the single sampling pulse of width ∆t < Ts. The Fourier series for
the periodic pTs (t ) is:
∞
2π
pTs (t ) = ∑ P *ε n
jnω s t
ωs = , the sampling rate in radians per second.
n = −∞ Ts
Where:
Ts
1 1 1  2π 
Pn = ∫2 p (t ) * ε − jnω s t dt = P ( jnω s ) = P  jn 
Ts  Ts 
T
Ts − s
2
Ts
Which results in:
∞
 2π   2π 
{
ℑ pTs (t ) = } 1
Ts
∑ P jn T  * δ  ω −
Ts
 = PTs (ω )
n = −∞  s   

1.2

1
ω)| -------
|P(jω

0.8

0.6

0.4

0.2

0
1 2 3 4 5 6 7 8 9 10 11 12 13
-0.2

-0.4

Where the height of each “bar” denotes the strength of the delta function at that frequency.

J. N. Denenberg Copyright DTS September 27, 2001

Sampling Tutorial Page 6 of 6

Since the Fourier Transform of a product of two time functions is the convolution of their individual Fourier
Transforms. The Fourier Transform of the sampled version of the input, x(t), is:

1 ∞
 2π    2π 
X Ts ( jω ) = ∑ P  jn
 T   * X  j  ω −  = PTs ( jω )
Ts n = −∞  s    Ts 

1.2
ω)|
|X(jω

1
ω)| -------
|P(jω
0.8
ω-ω
|X(jω ωο)|

0.6

0.4
ω-2*ω
|X(jω ωο)|

0.2

0
1
5
9
13
17
21
25
29
33
37
41
45
49
53
57
61
-0.2

-0.4

Notes:
1. Each repetition of X(jω), the Fourier Transform of x(t), is identical. It only varies in magnitude
according to the magnitude of P(jω) at the center frequency of that repetition.
2. If ωs (or fs) increases (i.e. Ts decreases), the repetitions remain equally spaced in the frequency
domain, but move further apart.
Aliasing
TBD
Uniform Sampling in the Frequency Domain
TBD

J. N. Denenberg Copyright DTS September 27, 2001