Recall that last night I assembled a preliminary board that I had designed with a VCOCXO on it connected to a 20 turn trimpot. The output goes into an ATMega328p connected to an LCD. Originally, this was intended to be my final clock calibrating solution. I would periodically load firmware in to check the accuracy of the oscillator while feeding the controller with GPS PPS input, and load other firmware to check the accuracy of the clocks.

Despite the minor power setback, I did run it overnight with my little bench supply giving out 3.3 volts. It functioned fine. The board does get warm around the oscillator, thanks to its oven, but I don't think it's going to be a problem. You certainly don't want to attempt to dissipate that heat. If anything, it should help the rest of the analog components to be kept in a relatively stable temperature (whatever impact the absolute temperature has should wind up being cancelled out by the GPS feedback). At steady-state, the oscillator takes around 330 mA or so. It requires much more than that - closer to 800 mA - for the better part of a minute when it's starting cold.

After running overnight, it was running 1 count fast when I visited it this morning. Thats +5 ppb. A little tweak with the screwdriver brought it back down to 0. Even with a 20 turn pot, it's quite touchy if you're trying to manipulate it at the single-digit-ppb level. The spec for stability of the oscillator is only ±20 ppb, however, so there's really no way for me to tell whether there are short term aberrations that are averaging out over my measurement interval (in this case, 10 seconds. For the final 10 MHz product, it'll be 20 seconds), but at least for my use case, none of that is significant.

Once I have a prototype built, I plan on bringing it into the office to run it on a proper frequency counter to see if there are any such short term effects, and if so, their magnitude. My hope is that the stability will not be out of proportion to the longer term accuracy, but we'll have to see.