Strange Behaviour in "Acceleration (without g)"

[Physikator]

When recording the accelerations during a car drive (btw: not leads to the "data recording crash" on iOS anymore) I got a battery warning. I hit "ignore" and then restarted data recording. The values were not mostly to zero any more but show a mean around 4m/s². I'm sure I did not accelerate that much for 90 seconds... After some time the mean just jumped back without any interaction. Saved experiment state available on request.
   

[Sebastian Staacks]

I don't think that this is a phyphox issue or that we can do anything about this. The acceleration "without g" is directly provided by the system (both, on Android and iOS) and we only take the data and treat it like raw data. Unfortunately, we do not know in detail how different devices calculate this value, because there is more to it than just subtracting g:

You can imagine the actual accelerometer as a mass on some springs and a clever way of measuring the deflection of that mass. An acceleration leads to a measurable deflection due to inertia, but so does gravity. That's why there is g in the accelerometer's raw data. If the measured acceleration changes, you cannot distinguish if the device has been accelerated (for example countering earth's acceleration) or if it has been rotated, so you need additional sensors to remove g. This is done by fusing the data from the accelerometer with the data from the gyroscope (which is why usually there is no "acceleration without g" on Android phones without a gyroscope - if there still is an acceleration without g, you should be very skeptical about its data).

However, because of the noise from the sensors, some additional tricks and filters are required and that is where we have no idea what Apple (or any other manufacturer) is doing there. An example where you can see this is that (on most iPhones) acceleration without g does not work after rebooting the iPhone will leaving it resting on a table. You can start the measurement in phyphox but you won't get any data until you move the phone a little bit, so these algorithms get some data to calibrate. The system is not sending anything to phyphox until it is ready. We have some Android phones that clearly use the first acceleration reading as a starting point, assuming that the device rests at the beginning of a measurement - leading to really absurd readings if you start the acceleration without g measurement while spinning the phone in a salad spinner.

So, long story short:
I suspect that a (bad) recalibration was triggered when the focus changed to the warning, when the phone was touched or when some other processes in iOS triggered when the battery dropped below the warning threshold. I cannot think of anything we do to the data to explain this.

Of course to be sure, it would be nice to hear if someone has seen this for acceleration with g (where I would not expect this at all).

[Physikator]

Ok, that seems to be right. I do find some change in the absolute value of "acceleration with g" after the "low battery" warning, but only in the order of 0.01m/s². (Dis-)Connecting from/to power causes about 0.1m/s². However, the above behaviour of "acceleration without g" is not easily reproducible.