alanyu wrote: ↑Thu Mar 28, 2024 8:17 am
It's a firmware issue rather than hardware. Old garmins before X20 are normal/good at gradient while X30/40 are laggy. IMO, Garmin dropped sampling rate a lot to 1 hz to enhance battery life, but gradient needs dozen of points to be calculated, resulting in the lag. Third party graident's on IQstore, which let you choose the filter length, work better than Garmin's own.
And yet those third party fields can only work with the 1 hz data provided by the system. On my 1030+ I'm happy with the gradient, I don't think it's worse than on the 1000 I used before and it's fully in line with what I'd expect from barometric gradient. When climbing on a bike, there's only so many decimeters in elevation change in the timespan people expect their gradient to change before they cry lag, and the tiny pressure change of a tenth of a meter (that's about the resolution commonly available sensor components advertise, but that's in lab conditions), requires a lot of smoothing to filter out the noise picked up by the sensor on the road. Less frequent sampling for better runtime can certainly be part of it, but my impression (based on the performance of my trusty 1030+) is that the lamenting on forums is mostly a mix of
some units with truly defective sensors, multiplied by a large group of users who expect latency comparable to accelerometry based "water level apps" on their smatphone, completely unaware of how that would require careful calibration and would crazy errors from acceleration and braking. My hypothesis is that the biggest change since the era of x20 devices is that people back then just weren't as accustomed to stationary accelerometer based angle measurement as they are now, had more realistic expectations.
That being said, I certainly wouldn't rule out that gradient latency
could benefit from some advanced sensor fusion where barometry gets augmented by a contribution from the 6dof sensors the devices already have (jump detection etc), but we all know why that will not happen: watches aren't mounted at a fixed angle to the ground and Edge aren't a primary development target anymore. And even Edge are not always mounted at a fixed angle to the ground, the bike might happen to be of the mountain type.
One technical detail where the gradient implemention could be worse than necessary, I did suspect my 1000 to suffer from that: for barometric gradient, you need change in altimetry over time, and horizontal progress over time (odometry, aka speed sensor). If the implementation is taken naively from a watch (or just implemented badly), it could be that horizontal progress is taken from GPS even when a speed sensor is present, and possibly even from some raw, unfiltered variation of GPS that would give you a wildly exaggerated speed under bad reception (the extra distance from the coordinates jumping around). If the implementation uses the speed sensor signal when available, but a raw, unfiltered GPS position (as opposed to the prettyfied data on the speed field), then it's also very much a possibility that the grade display degrades (hah!) massively in absence of speed sensor and good GPS reception.
What I do notice, with altimetry on my 1030+, is a wildly off 30s VAM. Grade is fine, as good or bad as is expect, but 30s VAM is wildly off. I run a homebrew cIQ 30s VAM in parallel and the difference is just laughable. I certainly would not bet the farm on my own field being completely without errors (chances are it's 29s VAM or 31s VAM, off-by-one mistakes are easy to make in programming), but the Garmin one is usually off by a huge margin. And my legs certainly
know that I'm not performing at 2000 VAM when the cIQ claims a more realistic 1000, and they also know when I'm not performing far more than 200 VAM when the Garmin field underreads. This 30s VAM wrongness is always the same through a day, but it varies a lot from day to day. Recently it has been a clear offset (as in always 1000 higher, no matter if the cIQ reads 50 or 1000), but if memory serves me right it used to be more like a factor in the past. The underlying bug must be pretty wild, considering how much the symptoms change from day to day (or from activity to activity, or from GPS signal acquire to GPS signal acquire, or from wake-up to wake-up, as I rarely do full reboots). I'd love to hear how others see Garmin 30s VAM compare to a cIQ implemention (haven't jumped through the hoops for publication, sorry, but I'd assume that some of the "custom field build kits" include 30s VAM?)