EE325 Signals & Systems II Lecture §7.

Reading

From Signals & Systems, 2nd edition by Oppenheim and Willsky:

Chapter 7: Sampling

• §7.0 - Introduction
• §7.1 - Representation of a Continuous-Time Signal by Its Samples: The Sampling Theorem
– §7.1.1 - Impulse-Train Sampling

7.0 - Introduction

Under certain circumstances, a continuous-time signal can be uniquely represented by and completely recon-
structed by its sampled discrete-time representation. This chapter covers the conditions under which this
holds (the sampling theorem), effects when it doesn’t hold (aliasing), and how discrete-time signals can be
used to process continuous-time signals.

7.1 - Representation of a Continuous-Time Signal by Its Samples:

The Sampling Theorem

In general, a continuous-time signal cannot be uniquely represented by its samples as a discrete-time signal.
There are an infinite number of continuous-time signals that share the same values at integer multiples of
T , i.e.,
x1 (kT ) = x2 (kT ) = x3 (kT ) . . . (1)

Figure: Continuous-time signals with identical sample values (Fig. 7.1)

If the continuous-time signal is bandlimited, however, if the sampling rate is high enough, the sampled signal
will be a unique representation of the continuous-time signal.

©2022 Laura E. Boucheron 1 Spring 2022

EE325 Signals & Systems II Lecture §7.1

Example: The “in-between points”

Imagine that you are given a thermometer and asked to record the temperature outside between 9:00am and
10:00am in ten minute intervals. You plot the temperature readings you took and get the following plot.

This is a discrete-time signal since we only have samples at regular points in time. We know that a discrete-
time signal is undefined for any points in time that don’t correspond to a sample point. If someone were
to ask you to use this plot to tell them the temperature at 9:15am, your answer, at least mathematically,
should be “I don’t know.”
However, I am now going to ask you to take your best guess about what the temperature was at 9:15am.
There is no right or wrong answer here, but if you wanted to try to be as accurate as possible in your guess
about the temperature at 9:15am, what would you say? Write your answer here for reference.

How confident are you in that answer? What if the spacing between the samples increased? What if the
spacing between the samples decreased?

How consistent are the results across the entire class? Why is there such consistency in how people answer
this question?

©2022 Laura E. Boucheron 2 Spring 2022

EE325 Signals & Systems II Lecture §7.1

7.1.1 - Impulse-Train Sampling

We need a means of representing a continuous-time signal in terms of its value at evenly spaced intervals.
We can use a periodic impulse train to sample the continuous-time signal, a process known as impulse-train
sampling.
xp (t) = x(t)p(t) (2)
X∞
p(t) = δ(t − nT ) (3)
n=−∞

Due to the sampling property of the impulse function, we know that x(t)δ(t − t0 ) = x(t0 )δ(t − t0 ). We can
thus express the sampled continuous-time signal xp (t) as
∞
X
xp (t) = x(nT )δ(t − nT ) (4)
n=−∞

Figure: Impulse-train sampling in the time-domain (Fig. 7.2)

You are now ready to take Quizzes 7.1.1 Sampling in the Time Domain available in Module 1 on
canvas.

• Competency Quiz 7.1.1: Sampling in the Time Domain will test you on material related to
sampling as implemented and interpreted in the time domain. This may include choosing the correct
plot of a sampled signal given a sampling period and/or frequency, specifying the sample values for a
given function both for t = nT and t 6= nT , and a refresher on conversion between time and frequency
variables.
• Mastery Quiz 7.1.1: Sampling in the Time Domain will involve similar questions, but with
additional complications or thought required.
• The optional Challenge Quiz 7.1.1: Sampling in the Time Domain will be a single question
that will push your thought process on time-domain interpretation of sampling.

©2022 Laura E. Boucheron 3 Spring 2022

EE325 Signals & Systems II Lecture §7.1

We can also express the effects of impulse-train sampling in the frequency domain. It is in the frequency
domain that the sampling theorem can most easily be visualized.
From the multiplication property of the Fourier transform, we know that
Z ∞
1
Xp (jω) = X(jθ)P (j(ω − θ))dθ (5)
2π −∞

What is the Fourier transform P (jω) of the impulse train p(t)?

We thus have Z ∞ ∞
1 2π X
Xp (jω) = X(jθ) δ(ω − kωs − θ)dθ (6)
2π −∞ T
k=−∞

∞
1 X
= X(j(ω − kωs )) (7)
T
k=−∞

At this point, we see that Xp (jω) is a superposition of shifted replicas of X(jω), scaled by 1/T .
You are now ready to take Quizzes 7.1.1 Sampling in the Frequency Domain available in Module 1
on canvas.

• Competency Quiz 7.1.1: Sampling in the Frequency Domain will test you on material related
to sampling as implemented and interpreted in the frequency domain. This may include choosing the
correct plot of a sampled signal spectrum given a sampling period and/or frequency, specifying the
band edges or other key frequencies in a given spectrum, specification of location of spectral replicas,
and specification of amplitude scales.
• Mastery Quiz 7.1.1: Sampling in the Frequency Domain will involve similar questions, but
with additional complications or thought required.

©2022 Laura E. Boucheron 4 Spring 2022

EE325 Signals & Systems II Lecture §7.1

On the previous page, we saw that Xp (jω) is a superposition of shifted replicas of X(jω), scaled by 1/T .
From this representation, we can also readily obtain the requirements on the bandlimits of X(jω) to assure
perfect reconstruction.

Figure: Impulse-train sampling in the frequency domain (Figs. 7.3, 7.4)

Based on the sketches of impulse train sampling in the frequency domain, state the constraints on ωM and
ωs to assure the ability to exactly reconstruct X(jω) from Xp (jω):

©2022 Laura E. Boucheron 5 Spring 2022

EE325 Signals & Systems II Lecture §7.1

Nyquist Sampling Theorem:

If x(t) is a bandlimited signal with X(jω) = 0 for |ω| > ωM , then x(t) is uniquely determined by samples
x(nT ), n = 0, ±1, ±2, . . . if ωs > 2ωM where ωs = 2π/T . Signal x(t) can be reconstructed by a lowpass
filter with gain T and cutoff frequency between ωM and ωs − ωM .
The frequency 2ωM is known as the Nyquist rate. The frequency ωM is known as the Nyquist frequency.
You are now ready to take Quizzes 7.1.1 Nyquist Sampling Theorem available in Module 1 on canvas.

• Competency Quiz 7.1.1: Nyquist Sampling Theorem will test you on material related to ban-
dlimits and the Nyquist sampling theorem. This may include specification of the bandlimit given a
signal spectrum and/or equation, use of a bandlimit to specify a sampling frequency according to the
Nyquist sampling theorem, or use of the sampling frequency to infer the bandlimit of a signal.
• Mastery Quiz 7.1.1: Nyquist Sampling theorem will involve similar questions, but with addi-
tional complications or thought required.
• The optional Challenge Quiz 7.1.1: Nyquist Sampling Theorem will be a single question that
will push your thought process on bandlimits and the Nyquist sampling theorem.

Reading for Lab #2

The following material will be covered as part of Lab #2.

From Signals & Systems, 2nd edition by Oppenheim and Willsky:
Chapter 7: Sampling

• §7.1.2 - Sampling with a Zero-Order Hold

You may choose to read this material now since it follows on from Section 7.1.1 material or to wait until
Lab #2 is assigned.

Reading for Lecture 7.2

The following material will be covered as part of the next lecture in class.
From Signals & Systems, 2nd edition by Oppenheim and Willsky:
Chapter 7: Sampling

• §7.2 - Reconstruction of a Signal from Its Samples Using Interpolation

©2022 Laura E. Boucheron 6 Spring 2022