The latest boards arrived today. They are version 1.4, and they're much better than the 1.2.x boards.

I put a scope on the 3.3 volt power bus at various points, and it appears that all of the isolation and filtering (not to mention the LDO replacing the buck converter) has had a real and positive impact. The analog and oscillator supply rails now exhibit less than 10 mV p-p of noise and ripple, according to my measurements. There's more than that in the digital section, and the inductor between the two is successfully keeping most of the noise confined. There is still some 10 MHz switching noise coming through, but I am not sure it's worth taking further steps to try and kill it.

I also took a moment to test the DAC granularity of the new design. It's right around 200 ppt per DAC unit, which is twice as much granularity as before. That's pretty much exactly as expected given that the buffer amp is compressing the DAC excursion range by almost exactly 51% (100k/51k). The DAC slope is inverse now, so the code had to change to accommodate that.

Preliminary tests taken with a frequency counter with a 10 ms gate time and a sample rate of approximately 0.25 seconds show an Allan deviation of less than 5e-10 at tau < 1s. The deviation plots tend to stay under 1e-9 out to 10^1, where the GPS module's stability dominates. At the moment, my antenna placement is not ideal, so the 10^2-10^3 numbers sometimes rise up into the 2-5e-9 range. I may wind up moving the whole experiment to a better GPS location and getting real deviation plots with known ideal GPS reception.

For those who wish, the firmware has a PLL mode you can turn on if you like. The difference between the two is that the FLL (the default behavior) only looks at the most recent error to discipline the oscillator. What happened in the distant path no longer matters. With a PLL, instead, the total, all-time error rate contributes to the discipline, so you can be sure that the total number of cycles will be correct, so long as you look at a large enough value of tau (sampling period).