IEEE Signal Processing - May 2018 - 115

is generic, rather than considering real-
world heart rates, and the exercise is of a
theoretical nature.
To complement [12, Problem 2.4]
with a hands-on practical example, we
instructed the students to test the concepts
of biased and unbiased estimators by
deriving their own heart rates from their
ECG recordings, as outlined in "Excerpt
#1 from the Coursework Assignments"
and Figure 7.
Another assignment involves extract-
ing the breathing rates from the recorded
ECGs, as respiration modulates ECG
through respiratory sinus arrhythmia, as
outlined in "Excerpt #2 from the Course-
work Assignments" and illustrated in
Figure 7(b). As mentioned previously,
the first trial is a "baseline" experiment
whereby students are allowed to breathe
in an unrestricted fashion. The second
and third trials, however, require the stu-
dents to breathe at specific rates following
metronome-like on-screen instructions.
This gives the opportunity to extract
the respiration rate from the ECG enve-
lope using signal processing algorithms
applied to the individual ECGs.

Student feedback
Over the academic years 2014-2015,
2015-2016, and 2016-2017, we ana-
lyzed student feedback from the Imperial
College student online evaluation system
(SOLE). Table 1 shows the number of
students enrolled in our EE3-08 ASP
course and the percentage of students
who gave their feedback through SOLE.
Figure 10 shows that the percentage of
students who found the course to be
intellectually stimulating (by checking
the "Mostly Yes" and "Yes" boxes in the
questionnaire) was above 80%, peaking
at 85% in 2016-2017. It was our impres-
sion from 2015-2016 that, by providing
MATLAB scripts for the preprocess-
ing of raw ECG data, this somewhat
decreased the incentive of some more
adventurous students to apply their own
preprocessing algorithms. This tradeoff
between giving too many versus too few
instructions is an ongoing challenge
faced by educators. Figure 11 shows the
statistics of student satisfaction with the
module over the corresponding 2015-
2017 period, with a similar trend as in

excerpt #1 from the Coursework Assignments
Using your own RR-interval (RRI) signal r [n] from Trial 1 (unconstrained breathing), obtain the heart rate h [n] from
h [n] = 60 .
r [n ]

(S1)

To obtain a smoother estimate of the heart rate, use the following method to
average every ten samples of the heart rate:
10

ht [1] = 1 / ah [i], ht [2] = 1 / ah [i], f,
10 i = 1
10 i = 11

(S2)

where a is a scalar.
a) Plot the probability density estimate (PDE) of the original heart rates h [n]
and the averaged heart rates ht [n] for a = 1 and a = 0.6.
b) Comment on the shape of the PDE of the averaged heart rates compared
to the original heart rates. How does the constant a affect the PDE?

excerpt #2 from the Coursework Assignments
Breathing at regular rates will make the presence of respiratory sinus arrythmia in the cardiac (ECG) data more pronounced.
a) Apply the standard periodogram as well as the averaged periodogram
with different window lengths (e.g., 50 s and 150 s) to obtain the power
spectral density (PSD) of the RR-interval (RRI) data. Plot the PSD of the RRI
data obtained from the three trials separately.
b) Comment on the differences between the PSD estimates of the RRI data from
the three trials. Can you identify the peaks in the spectrum corresponding to
frequencies of respiration (and their harmonics) for the three experiments?

Figure 10 and with more than 80% of
students satisfied with the module.

Conclusions
We have explored ways to enrich
some classic estimation theory topics
by "signal processing for wearable
health" aspects. The aim was to enable
students to explore interdisciplinary
signal processing challenges that are
often encountered in their future jobs.
This has also provided an opportunity
to bring biopresence into the curricu-
lum through the challenge of analyz-
ing students' own physiological data
and to enhance their engagement
through a de facto part-ownership of
their course.
Overall, the students appreciated
the experience of performing biosig-
nal analysis on their own data and felt
IEEE Signal Processing Magazine

May 2018

Table 1. The number of students enrolled in
the EE3-08 course and the percentage of
those who gave their feedback online.
Year

2015

2016

2017

Number of students

Percentage who gave
feedback

39%

54%

60%

that such a biopresence in the curricu-
lum solidified the theoretical concepts
taught in the class and in an enjoyable
and engaging way. This approach fur-
ther encourages the students to
■ explore additional concepts not cov-
ered, or only partially so, during
the lectures
■ enhance their creativity both in
learning and applying the concepts
with the additional "opportunity
115

Table of Contents for the Digital Edition of IEEE Signal Processing - May 2018