GPSDO: PLL or MCU controlled?

[thinkfat]

I'm wondering about the merit of using a Kalman filter for this purpose. Usually, you need a dynamic model of your system that generates an estimate of its next state which the Kalman filter then combines with the next measurement to find the "true value". This estimate in your case seems to be the "DPPS" - what is that exactly, the divided-down OCXO clock?