|
Now actually works on turbo (was previously emitting "temp low" instead of "temp high" while in direct-drive, probably due to an overflow).
Made stepdown work based on an average of the last few temperature predictions instead of just the most recent one. (reduced noise sensitivity)
Made each temperature sample based on 8 measurements instead of 4, to reduce noise.
Made standby mode re-init thermal measurement arrays, to avoid weird behavior next time light is used.
Reduced fixed-point precision to avoid overflows.
Reduced prediction strength to encourage stepping down faster while hot.
(unfortunately also steps down later, I think, if it wasn't already hot)
Not totally happy with new algorithm, but it's the least crappy of a whole bunch of things I tried.
(for example, a PID approach with correction based mostly on I... didn't work very well)
(taking an average of the last few predictions is very similar though, and works)
(but the result is still kind of meh)
Saving this so I'll have a functional base next time I try to improve it.
|
