Lucas Silva Lopes (Student ID #: 0887016)

Chinh Nguyen Ngoc Tran (Student ID #: 50649247)

BME 60A
Project 3
Heart Rate Monitor

Dear Dr. Botnivick, Eric, and Jue,

In this project, we tried to build an optical heart rate monitor that reads the heart rate
as accurately as possible. In order to do so, many steps were required. We used
paperboard, electrical tape, LEDs and a light sensor along with the Arduino board to
create the device. We then used Labview to write the code. The hardest part was to
figure out an absolute way to distinguish between a good signal and a bad signal, so
we could show only reliable values of heart rate. Many steps were described in the
report. We hope you like our project.

Sincerely!










Approach Description:

We divide our project into four parts: Data acquisition, signal processing, interface
and data logging.

In the first part of our Project, which was data acquisition, we used paperboard and
electrical tape to make a support for the LEDs in order to keep them stable on top of
the sensor and pointing to it. Also, we found out that using just one LED was not
sufficient; especially in a dim room with only light coming from the outside, more
LEDs were required for the experiment to be effective. Thus, we used three LEDs
instead of just one. By keeping the LEDs still, above and pointing to the sensor while
one little finger lie on it, we were able to acquire good data several times in at least
two very different conditions (day and night). Therefore, we made this our standard
device.

In the second part of our project, we had to write a code to smooth the signal,
detrend it, and find its period. We used methods described in class, such as: moving
average filtering, median rank filtering, linear fitting and autocorrelation. The period
was found after the autocorrelation function graph.

In the third part, we had to create an interface for the device user. We had to decide
which values to show and which values not to show. More precisely, we had to find a
way to decide, regardless the ambient conditions: is the signal pattern acquired
mostly due to the pulsatile nature of the circulatory system or not? After trying
different ways of measuring this attribute, we found out that the difference between
the center peak of a normalized autocorrelation curve and the peak to its right is
really helpful to decide whether a signal is acceptably periodic or not.

There are some other attributes we could have used to decide if a signal is good or
not, such as the area under the processed signal after taking its modulus; however,
even though these attributes work well visually, when you try to normalize them, they
are no longer eligible to differentiate between a good signal and a bad signal. Its
important to normalize because periodic functions are characterized by different
attribute values depending on the ambient light conditions. We also tried other ways
besides normalization to make these values the same regardless the ambient
conditions, but they all turned out unsuccessfully.

In the last part, we added data logging into our project, so data can be opened in a
spreadsheet program for viewing or creating graphs.




Code Description:

In the first part of the code, we just acquire data continuously from the sensor. We
use an acquisition speed of 200 Hz and acquire 600 samples, giving a total of 3

seconds of data acquisition.

In the second part, we process the acquired signal. First we use a filter express vi
and select smoothing as filtering type. We select a 5 half-width moving average, in
other words a moving average of 11 points. We then convert the filtered signal, which
is in dynamic data type (includes additional attributes such as date and time the data
was acquired) to waveform data type. Later we ignore the 11 first elements of the
curve because the moving average algorithm creates an artificial rise in the
beginning of the curve. We use a median filter vi, with left and right ranks equal to
five. We calculate a linear fitting and subtract it to detrend the data. At last, we
calculate the autocorrelation function of the signal.

In the third part, we first normalize the autocorrelation function, we get rid of the
negative values by adding 1 to everything, and make the maximum value 1 again by
dividing the function by 2. Now the distance between the center peak and the peak to
its right is a number between 0 and 1, which is more intuitive, similar to percentage.
Then we find the peaks. The difference between their indexes is the period of the
signal and the difference between their values is the number between 0 and 1. Now
we calculate the heart rate by changing the units of the period from indexes to
seconds, taking its inverse (which is the frequency in hertz) and multiplying it by 60,
which gives the heart rate in beats per minute. We then convert it to a string
constant, so when the distance between the peaks is less than 0.3 (value found by
observation of various periodic and non periodic signals) we show the heart rate, and
when it is greater than 0.3, we show Try again.

In the last part, we allow the user to define a path to the data. It can be appended to
a preexisting file or saved into a new one.