Signal Processing - July 2016 - 63

highest-quality education. Of course, after a few
sobering years of experience, neither extreme actually happened; in this article, we share our collective opinion on this topic.

Introduction

ILINDRI
OTO.COM/M
©ISTOCKPH

As teachers of various signal processing courses, we felt intrigued by the opportunity to bring
our subject to this new teaching paradigm. Of
course, many online resources for signal processing education exist already, including video
courses developed in the 1980s by the Massachusetts Institute of Technology (MIT) Center
for Advanced Engineering Studies [2], [3], various courses available on MIT's OpenCourseWare (OCW), a variety of online interactive
demos, hands-on tutorials and code samples,
and, of course, many collections of lecture
notes and slides. Still, there was no interactive,
full-fledged online course on signal processing available as the MOOC revolution was
unfolding. This presented an interesting challenge and opportunity.
We approached online teaching from
different backgrounds and with different
perspectives, and we all started from our
respective residential classes that range from
mandatory undergraduate-, to master of science- and doctoral-level courses. Specifically, during the last three years, we created
three MOOCs based on the following residential courses:
■ Signal Processing for Communications (COM303), a mandatory undergraduate course taken in the third year at
École Polytechnique Fédérale de Lausanne (EPFL),
Switzerland, became Digital Signal Processing (DSP)
on Coursera and has been offered five times since
spring 2013.
■ Discrete-Time Signals and Systems (ELEC301), a mandatory undergraduate course taken in the second/third year at
Rice University, is the basis for the corresponding edX
course offered twice since fall 2014.
■ Discrete-Time Signal Processing (6.341), a first-year graduate course at MIT that is the basis for 6.341x, made available on edX and offered twice since fall 2014: once in a
limited industrial beta format and once as a fully open
online course.
The goal of this article is to relate the experience gained
in moving "on-campus" material to the MOOC format and
share the lessons learned, the influence this has had on our
on-campus teaching. We hope that our MOOC experiences are
valuable to the signal processing community.

EPFL's Digital Signal Processing on Coursera
Overview and goals
The course Digital Signal Processing, by Paolo Prandoni and
Martin Vetterli, was first offered on the Coursera platform
in February 2013. At the time of this writing, the class has
completed its fifth edition. We authors are with the EPFL,
and the course is based on the residential class COM303
offered by the Communication Systems Department to thirdyear undergraduates. For many SysCom students at EPFL,
COM303 represents the first exposure to a higher-level engineering class after two years focused primarily on introductory subjects. COM303 lists calculus and linear algebra as
prerequisites and recommends familiarity with probability
theory as well. The class is based on the freely available
textbook, Signal Processing for Communications [4], that
we have written.
COM303 is a standard undergraduate-level DSP class
with a slight emphasis on telecommunication systems.
The syllabus starts off in the discrete-time domain and
uses vector spaces and linear algebra as the framework to
introduce signals and signal transforms; as subsequent topics are introduced, the goal is to strike a balance between
solid mathematical foundations and practical applications.
When adapting COM303 to the online medium, we decided
to closely mirror the residential class. We did this for two
reasons: primarily, we wanted to produce a package that,
although aimed at the general public, would retain its focus
on theoretical foundations rather than deliver yet another
hands-on approach to applied DSP, for which countless
tutorials are available on the Internet. Additionally, we
wanted to experiment with the concept of the "flipped
classroom" and be able to minimize standard lecturing to
the advantage of more question-and-answer (Q&A) interaction with on-campus students.

Course organization
Outline
COM303 is composed of 17 lecture days that occur during nine
weeks, and, as shown in Table 1, it is structured around nine
thematic modules; each module is split into a varying number
of small units (the actual videos) with the intent of balancing
the conflicting requirements of a fine-grained subdivision of
the material with the "narrative" needed to provide reasonably
self-contained mini-lectures. Each lecture day provides students with the following:
■ three video units (with associated slides; we should mention that we found it very difficult to produce videos lasting ten minutes or fewer, as per the recommended best
practices, and average video length is 17 minutes)

IEEE SIgnal ProcESSIng MagazInE

|

July 2016

|

63
http://www.ISTOCKPHOTO.COM/MILINDRI
Table of Contents for the Digital Edition of Signal Processing - July 2016
