I got the second prototype boards today. They swap out the ATMega328 for an ATTiny4313, which worked perfectly. Unfortunately, it does require me to abandon the Arduino IDE, firstly because there isn't enough flash for the whole thing and the Arduino libraries, and secondly because my ATTiny mods for my (ancient) version of Arduino don't support the 4313.

It wasn't so bad, though. This project doesn't use Liquid Crystal, and translating the serial I/O stuff into direct register calls turned out to be quite simple.

For the next set of boards, I've decided to put a buck converter on the board rather than continuing to use the LT1117 3.3v LDO. I've made good use of the MC34063 on my Hydra and OpenEVSE II projects, and believe I can come up with something that will make less ripple than the LDO and wall wart combination are doing presently. At least, I hope so.

Hopefully, that will be the final design.

Meanwhile, I was able to hook the first prototype up to an Agilent frequency counter at work. It said that the board was running 7 Hz slow, but there's no telling how accurate its own internal timebase is. What it did say, however, was that it was staying within 0.01 Hz of what it was reading, which is 1 ppb. So that's a good sign. I'm going to try and read the manual to figure out some of the statistics that it can gather. I think it can give me a standard deviation value, which would be almost as valuable as an Allen deviation plot.

Meanwhile, according to the debug serial output, prototype number 2 is presently showing a drift of less than 600 ppt over the last 500 seconds. My own frequency counter shows 10.00001 MHz, but the next step will be to attempt to use the output of this board as that counter's reference.

EDIT: I also found an Agilent N5182A signal generator. That one had a calibration sticker on it. I fed the same frequency counter's reference input from the GPSDO and had it measure the frequency of the N5182A, which it said was 0.0465 Hz high. That means that the two were 4.65 PPB (or so) apart.