The phase detector appears to be doing fairly well now.

The performance out to tau 400 seconds or so is on a par with what it was before. But from there, the ADEV graph takes a downward turn, as is to be expected with GPS disciplined oscillators generally. By contrast, the older design continued straight to the right out to around 2000 seconds, and then started its downward trek. In general, the long term performance at the moment appears to have at least a half an order of magnitude of "left shift" - that is, the performance is the same at perhaps a quarter of the tau. Where this really shows is the unwrapped phase differences (removing the linear residual) between old and new. The old code allowed excursions of 2e-7 on a fairly regular basis. The new variant is almost entirely inside of 3e-8 (30 ns) in the testing I've done so far. Given that the GPS module's nominal stability is 10 ns, I'd call that a win.

And that sort of makes sense. The current firmware samples at a rate of 5 seconds as opposed to the old code's 100 seconds. I tried at 1 second, but that seemed to me to be prone to overshoot. Still, I think a fair bit more tuning might make the result a little bit better.

Also still to do is to revamp the LED support. At the moment, the LEDs don't have a particularly calibrated meaning to them like they did before. The "blinking" state means a GPS unlock, and the 0 state means the PLL is unlocked, but that leaves states 1-3 to be defined. I'm considering making them the absolute value of the phase delta (with some appropriate scale factor), since it does appear that the rate of change of phase does seem to correlate rather strongly to the stability of the output.