IEEE Signal Processing - May 2018 - 114

the random modulation/demodulation
background into a biosignal analysis task.

summary of the experimental instruc-
tions and a sample ECG recording that
corresponds to the different trials out-
lined previously. Observe the sinusoidal
modulation effects of the respiration on
the ECG envelope in trials 2 and 3 in Fig-
ure 7(b); a perfect vehicle to incorporate

Preprocessing
The raw ECG recordings from the
iAmp typically exhibit baseline drifts
and other artifacts that often interfere

Amplitude (mV)

12
11
10
9
8
7

0

10

20

30

40
Time (s)
(a)

50

60

70

Amplitude (mV)

11
R
10
9

RR Interval

R

R
T

T

T
P

P

P

8

6.5

7
Time (s)
(b)

7.5

8

Figure 8. A student's own real-world arm ECG recording of more than 75 s. (a) A typical raw ECG
recording from iAmp that exhibits baseline drifts. (b) The magnified ECG plot with P-, R-, and T-waves.

ECG

4

× 104

2
0
-2

0

10

20

Difference
Between R Peaks (s)

Filtered ECG

30
40
Time (s)

50

Estimated R Peaks

60

70

Ampthresh

(a)
0.95
0.9
0.85
0.8
0.75

The assignments
0

10

20

30
40
Time (s)
(b)

50

60

70

Figure 9. The detection of R-peaks in the ECG and the generation of HRV. (a) The band-passed
filtered ECG and the suggested amplitude threshold (broken blue line). (b) The RRI obtained from the
R-peak detection algorithm.
114

with the analysis conducted by the stu-
dents. Figure 8(a) shows an example of
an ECG recording that contains low-
frequency drifts. As an optional exer-
cise, the students are asked to design a
digital filter to remove unwanted drifts
to yield a clean signal that resembles a
"clinical" ECG reading, such as the
ECG trace in Figure 8(b).
To simplify the preprocessing steps,
we also provided MATLAB scripts
to identify the heartbeats (R-peaks) in
the ECG and to generate the RRI time
series (in a way similar to that in [11]), as
the main focus of the assignments was
on processing the respiratory data
and variations of the heart rate in the
frequency domain and not on artifact
removal. In the preprocessing step, the
MATLAB script first performs band-
pass filtering of the raw ECG recording
with a fourth-order Butterworth fil-
ter with cutoff frequencies at 5 and
20 Hz. This removes high-frequency
noise and low-frequency fluctuations,
while also suppressing some features of
the P-, S-, and T-waves to assist the
RR-interval detection.
In the next step, R-peaks are detect-
ed in the signal based on an amplitude
threshold (the broken blue line in
Figure 9), which can be adjusted by
the student. The so-identified peaks and
the obtained RRI time series are simi-
lar to the example shown in Figure 9;
the students were able to fine-tune
the detection parameters to suit their
own recorded data. Next, the provided
MATLAB script identified suspected
anomalies in the ECG traces (e.g., the
ectopic peaks) and students were given
the option to discard them. Finally, the
RRI time series was resampled at a sam-
pling frequency of 4 Hz, to produce the
interpolated time series suitable for fur-
ther analysis.

IEEE Signal Processing Magazine

|

May 2018

|

To illustrate the usefulness of the adopt-
ed data-driven approach as a comple-
ment to the standard theory-based
teaching, we revisited the concept of
biased estimators from the perspective of
probability density functions of weight-
ed average heart rates [12, Problem 2.4].
In the problem, the notion of heart rates
Table of Contents for the Digital Edition of IEEE Signal Processing - May 2018
