Congrats, very impressive!

The linked article is a bit light on actual details – could you share the paper/preprint maybe?

Thanks!

https://ieeexplore.ieee.org/abstract/document/11096342

That’s the official paper link. Sorry it’s not open access.

Too bad :( Though I’ve just found a PDF here in the open: https://news.ucsc.edu/wp-content/uploads/2025/08/Pulse_Fi_sh...

Oh great- I forgot we had the open pdf posted somewhere

Awesome job!

A couple questions:

- How was the training/testing data split? Was it split window-wise or participant-wise?

- How does the system perform at elevated heart rates? Seems like it was mostly tested at resting/normal heart rates

1. in this iteration we used a 64% training 16% validation 20% testing split. In our current work we are testing with 10 fold / leave a subject out to get better analysis. 2. the esp dataset had heart rate up to 130 which is relatively high, and the raspberry pi data had people running in place etc... where the heart rate is higher

1. The collected data set has one person in between receiver and transmitter. Will the same model work if there is multiple people between receiver and transmitter?

2. If anyone has a heart rate outside the range of 48-130? can this model detect that heart rate?

1. We are currently working on multi person, this iteration doesn't support it. 2. The range was just what we had in this dataset, if we exposed the algorithm to a broader range, it would work for that too.