Week 9 Journal

This week, our main focus was to debug the ICA algorithm to get a more accurate heart rate reading. While looking through the algorithm we have been following, we realized that the algorithm was assuming that the columns of our input were individual signals. However, the input we have been feeding it so far has the rows as the sources. Because of this, we traced through the ICA algorithm, making sure the sizes of variables matched throughout the computation. Then we fed in a transposed version of our input.

While looking at MicroPython libraries, I came across a library called ulab, which had a lot of numpy like functions for arrays - from finding statistics of arrays to computing the eigenvalues to finding the FFT and IFFT. Jennifer was able to download it to her laptop by updating the OpenMV's firmware, so we will be replacing the helper function we made for the ICA algorithm. Hopefully, using functions from the library will allow us to focus on debugging the ICA algorithm instead of a million helper functions.

In the process of debugging the ICA algorithm, I realized that the mean function in numpy has a parameter for axis, which determines if the mean of a 2D array irrespective of the direction or column wise or row wise. Therefore, couple of the mean functions we used in the algorithm were incorrect. So, I coded two more mean functions - one that took the mean of each row and one that took the mean of each column. A similar problem occurred with the center function, so I recoded so that the mean of each row was subtracted from each row's elements.

Next week, we plan on going through the rest of the ICA algorithm and replacing helper functions like the DFT function and the mean functions we wrote with functions from the ulab library. Hopefully, this will correct any computational errors we had and lead to a more accurate ICA algorithm, which will yield a more accurate heart rate reading.