Signal Processing - July 2016 - 81

instructors, TAs, and other kind peers." Regarding the forum, a student commented, "The instructors did a fantastic job in interacting
with the students. I cannot recall one single question that was not
properly addressed. They are very kind and responsive." To us, this
indicated that better learning outcomes in 6.341x were strongly
facilitated by a high level of staff involvement on the forum.
In terms of the organization of the courseware, students
reacted especially positively to the organization of the course
topics, verifying the staff's intuition that weaving video segments with short exercises would provide a natural mechanism
for students to check their understanding before moving too
far along. The interactive numerical content in the courseware
was well received and generally viewed as a unique feature
of 6.341x. Some students had been hoping for an opportunity
to write and test their own signal processing code, which fell
outside of the scope of the interactive problems. Many of those
students desiring additional code practice were typically sufficiently motivated to do so on their own and post the results
of their efforts on the discussion forum, significantly enriching
the discussion among the community of learners.
A number of individuals also commented that, overall,
6.341x had a significant impact on them both personally and
professionally. In particular, students indicated the immediate applicability of 6.341x to their professional work in fields
ranging from software-defined radio to the design of particle
accelerators. A recurrent theme in student feedback was also
that 6.341x helped them to decide to change their field to signal
processing, which the staff was delighted to hear. Overall, these
comments indicate to us that 6.341x is a challenging course, but
for those who complete it, it is also a very rewarding experience.

Next steps
Going forward, there are several potential modes and roles for
6.341x content. In the context of its incorporation into the MIT
residential course, there is no question that it has contributed significantly to the educational experience, and it will continue to be
incorporated and developed further in the context of the MIT residential graduate course. We welcome the opportunity for it to find
a similar role in many residential courses at this level elsewhere.
In the form that 6.341x ran as a MOOC in spring 2015, the
content was released on a fixed schedule, and learners were
expected to commit to that schedule. Although that mode of
delivering difficult course content of this depth and sophistication is typical in a university environment, rigid pacing is
not necessarily well matched to participants outside of the
university environment or to those with other significant time
constraints and deadlines. In future deployments of 6.341x, we
envision a self-paced mode on a more local platform so that
pacing of the content can accommodate the needs of particular
groups of participants. As one model for use in an industry
environment, a company could perhaps subscribe to the content and platform for use internally. The oversight, pacing, and
staffing of the course could then be managed internally to
match the needs and schedule constraints of participants.
Our current plan in the near term is to release the total
6.341x course content in a form that is freely available to

learners on the edX platform, for the purpose of self-study.
In this mode, autograding the exercises and problem sets will
be activated, but no discussion forum or support staff will be
available. Specifically, it will be accessible in a manner similar
to that of course content on MIT OCW, although with a more
interactive component. When it becomes available, the material will be accessible at http://www.rle.mit.edu/dspg/6.341x.

Conclusions
The experience of the three DSP courses presented in this article clearly suggests that online platforms and content offer rich
opportunities for teaching signal processing. How to best affect
this is not yet clear, and "best practices" can be very dependent on
the demographics of the learners and the objectives and personal
style and preferences of the course developers and instructors in
adapting residential course content to an online environment.

Commonalities
Perhaps the most important commonality across the three
courses is the focus on solid theoretical foundations. In that
and a number of other respects, they have a different purpose
and target audience than many other online courses, for which
content is primarily oriented toward a high-level overview of a
topic area. This difference is clearly a key factor in the drop-off
level of active involvement from registration (i.e., many registrants are motivated mainly by curiosity) to course completion. This is inevitable for online courses that attempt, to the
extent possible, to provide participants with the same depth
and sophistication as a residential course. It is also important
to recognize that a MOOC is no substitute for a well-taught
residential course that incorporates significant interactive face
time with a knowledgeable and motivated staff.
Because the three courses are based on residential classes at
different levels, the backgrounds and expectations of the participants somewhat differ. However, in a broad sense, a serious
background in signals and systems at some level was common,
and quite often, a more advanced background including industrial project experience with a partial motivation to refresh that
background was helpful.

Differences
Two of the courses were offered on edX and one on Coursera.
The differences between platforms are certainly many but not
profound enough to significantly affect the way the material was
structured and presented. The three courses were, in fact, more
distinct in their handling of numerical exercises and examples.
EPFL ultimately gravitated toward Python (via IPython Notebooks), Rice experimented with a tight integration between
MATLAB and edX, and MIT developed specific extensions
to the edX platform to provide in-browser numerical exercises
independent of any specific package or programming language.
Each of these approaches has potential advantages and
drawbacks. MATLAB offers perhaps the most complete
signal processing sandbox and a very user-friendly learning
curve, but its scripting language does not please those students with a more rigorous background in computer science.

IEEE Signal Processing Magazine

July 2016

http://www.rle.mit.edu/dspg/6.341x

Table of Contents for the Digital Edition of Signal Processing - July 2016