Didn't want to clog up the benchmark thread with this but it's an interesting problem with FIT files.
Smart-recording is evil, experimented using my old vivoactive vs fenix on my long run and they are a quarter mile off despite no GPS loss.
The route looks accurate on both without GPS drift.
The problem must be the coastline paradox, where if you measure more points, you get more distance https://en.wikipedia.org/wiki/Coastline_paradox
Almost 8000 records on the fenix, only 1300 on the vivoactive, it adds up.
What's interesting is that Stryd also appears to not use 1-second recording but some kind of hybrid, it's FIT has almost 4300 points but that's still half of the Fenix, it's less than a tenth of a mile difference.
Has anyone had success interpolating points, distance and elevation, when smart-recording is on?
There are curve formulas for point to point because people tend to "arc" and not go in straight lines when there is velocity so you have to look at points before and after.
I am wondering if anyone has more datasets with 1-sec vs smart-recording, I bet I could find some on Rainmaker's site if I hunt for them.
Smart-recording is silly with today's cpu and storage on watches, I could see a full day ultra maybe needing it but eventually efficiency will allow that to have 1-sec too, storage is definitely not a problem
I can get several elevation databases but distance is another matter.
By the way does anyone know if modern GPS watches use the doppler-effect available in GPS in addition to the lat/lon ability? It's a fascinating principle I was not aware of, it can measure speed based on the slew of the signal (like how a siren changes tone when moving past you)
This library supposedly has the most accurate math for distance between two close GPS points, better than Haversine or Vincenty
https://geographiclib.sourceforge.io/
There's no PHP version but a JS version exists so was thinking of porting it.
https://geographiclib.sourceforge.io/scripts/geographiclib.js
