baro sensor location?

[AgentSmooth]

Hi, does anyone know where the baro sensor is located on different watch models? (especially interested in Ionic at the moment)

Baro data seems wildly inconsistent in high speed situations (we are talking up to half an order of magnitude error in speed calculations based on baro data), and I'm thinking this is caused by bad shielding against oncoming wind. I would like to try improvising a shield, if practical, in order to improve consistency (since I'm already massaging data and would definitely like to keep this practice to a minimum).

As an alternative, can anyone think of a good data smoothing strategy for baro readings that wouldn't be too taxing on watch resources? I'm currently doing simple batched-average and it's not cutting it.

Thanks for any tips and pointers.

[Gondwana]

It's possible to calculate a running average without summing all of the relevant values every iteration.

Might help: https://dev.fitbit.com/build/reference/device-api/signal/

[AgentSmooth]

I keep rolling averages for other things, but for baro data in particular there's no point since it's bound to change, and significantly, due to the nature of the activity. So what I'm doing is collecting a batch then averaging _within_ the batch in hopes that it would reduce the significance of an individual reading error. However, what I'm observing from live data is that, regardless of batch averaging, baro data comes in with randomly spaced "glitches" (I'm speculating, but IMHO this is caused by sideways gusts of wind, since I was able to reproduce this behavior on a minor scale by blowing along the face of the watch) that _sometimes_ coincide with increased delays between batches (based on timestamp data). So either the watch can't cope with the stream of data speed-wise and starts choking (document.* update delays?), or baro shielding is a factor (I've seen other equipment misbehave in ~150mph wind, though proper shielding reduces variance significantly). For the time being, I'm resorting to also averaging between adjacent batches, but this is not ideal since it reduces precision.