Inconsistent PineTime Sampling Frequency

[jones139]

A user has reported observing different sample frequencies for two different PineTime watches (by observing the time taken to collect 125 samples for a nominal 5 second period at 25 Hz).
The user's observation suggested that different phones were giving different sample frequencies rather than different watches.

So I did a test with three different phones (A Samsung A17, A Samsung A06 and Motorolla Moto G13) and 4 different PineTime watches, ranging from one that has been used for a very long time for development to one which is brand new from the June 2026 manufacturing batch of watches. The time taken to collect 125 samples was read off the main activity screen (in advanced mode), for 5 updates in succession and the average calculated. This was repeated with each watch/phone combination with some repeats to give the results below. Note that the error bars are the standard deviation of the data for each watch - the error bars are very small so each watch is producing consistent results.

The data used to produce this graph is shown in this spreadsheet: PineTime Sample Frequency.xlsx

This shows that the time to collect 125 samples is very consistent for a given watch, but different watches give repeatedly different times.
This suggests that the accelerometer sample frequency in the PineTime watches is not particularly accurate - ranging from the expected 25 Hz up to 29 Hz - a variation of about 16%.

Fortunately in most circumstances this will not be significant because of the huge variation in movement frequency for different users (3-8 Hz), but it is possible that if a user makes very slow, large movements that are close to the lower frequency threshold that they might not be detected, and they might need to lower the alarm frequency min setting if they are using the original OSD algorithm.

[jones139]

There are three things we can do about this:

1. Ignore it because the variation is not significant enough given the variation between users is so large.
2. Calculate the actual sample frequency on the phone, and re-sample to 25 Hz by interpolation
3. Investigate what is going on on PineTime more. If we were getting a lower than expected frequency I could imagine it being a software issue because the PineTime operating system is polling hardware rather than relying on interrupts, but this is the other way around - we are getting data more quickly than expected, so is probably an accelerometer hardware issue (or a bug where we are reading the same accelerometer reading multiple times).

[jones139]

I'll ask the InfiniTime developers if they have seen this and thought about it....

[pmithrandir]

Interesting.

For me, it show mostly that pinewatch can be an entry level watch for OSD, but is not as reliable as garmin.
Cardio is less good, now we see the accelerometer is also having inconsistencies...

Until we find out, I would remove the disclaimer on the website advising user to buy that watch and replace it by a proper explanation on what are the pro and cons.

generally speaking, any user who would ask me I would recommend garmin as we can find the forerunner with a week autonomy for 150€ 2nd hand easily, and if needed it can be used for other purpose or resold when leaving OSD, while giving better results meanwhile.

In europe, I always think that people can find 80€ more to spend on their health. It's a bit of money, but not that much.

[jones139]

@pmithrandir, you raise some good points.... but I don't quite agree with all the conclusions :).

Garmin is undoubtedly the 'better' device - the heart rate monitor works when you are moving and the accelerometer is (usually) more accurate (see Forerunner 55 below).

So if you need heart rate monitoring, then Garmin is definitely the way to go, because PineTime is not suitable at the moment (another developer is starting to look into algorithms to filter out movement artefacts from the ppg signal though) - there are a few papers referenced about it that we can look at implementing). I am also wondering about adding in support for generic bluetooth heart rate monitors such as this Coros one.

For movement though, I think this sample frequency variation is more annoying than really significant, because there is such a wide variation of typical seizure movements for different individuals, that this 16% range is not that large. I usually think of seizure movements being in the range 3-8 Hz, which would be 5.5 +/- 2.5 Hz = +/- 45%. Also from a practical perspective, we have been using PineTime for over 18 months now, with a variety of the watches tested, and observe a very high seizure detection reliability, despite the different sample frequencies.

But the main reason I am reluctant to recommend Garmin for new users is the issues we have had with Garmin doing firmware updates and breaking the sort of watches that I recommend to users - the VenuSQ connection reliability problem made it un-usable, and the ForeRunner 55 started sending nonesense accelerometer data - and Garmin have not fixed either of these issues and do not provide a way of down-grading the firmware to a useable one. This leaves me not knowing which Garmin watch to recommend in case it gets broken by firmware update. I have a forerunner 245 that works ok (but battery life is not great). Which Garmin watch do you think we should recommend?

If we can decide what we will say about recommended watches, I'm happy to update the web site to include a better pros and cons discussion though.

Thanks!

Graham.