I got a new RF generator recently and in testing it and checking its internal oscillator, I may have stumbled on an issue with the BG7TBL GPSDO that I've got... I believe it's the most recent version, but I've posted it earlier in this thread (now there's BNC holes through the date on the PCB front panel
)
Anyways, I was using a Fluke PM6690 (12 digits/s 100ps resolution counter with 1s gate time) with a high stability reference (not the highest stability option, but two up from base model TCXO) and my new HP E4431B ESG-D generator, also with the high stability internal reference option. I had the generator going straight into the option C input on the counter and a 1GHz -20dB signal turned on.
With the external reference for both, the timing noise is quite low (both instruments are pretty high spec), amounting to 3x10^-12 peak to peak frequency variation over almost 2000 seconds of recording time:
So using the shared external reference, there is very little variation in either the generation or counting of signals - a sanity check for the quality of the system.
Then with each instrument using their own internal oscillators, you see a predicted fluctuation due to the oven operation, up to 0.7Hz (7x10^-11) peak to peak:
But when you assign one instrument to the external GPSDO reference and one to its own (and this goes for either instrument), while the performance is still slightly better than the internal oscillators... there's a very strange step behavior from the GPSDO timing signal, which is clear and obviously visible. It's also not characteristic of the response of regular drift in a frequency standard - it looks almost like it's the visible steps of the DAC being used to drive the OCXO's adjustment voltage (for reference, the fine adjustment of the HP E4431B's oscillator has a step size of 1/2 to 1/3 of this, somewhere between 100 and 150 mHz at 1GHz generated).
First with the counter locked to the GPSDO:
Then with the generator locked to the GPSDO:
Some really odd looking behavior... and given that the spec was advertised with a least significant digit ballpark of 10^-12.... I don't have real confidence that it's getting there, at least in the current configuration. The GPS was locked to 8+ satellites the whole time (verified through the serial port), and while either instrument with the GPSDO referenced to the other had better performance than just the internal oscillators.... it really wasn't by much.... and if you figure that the runtime for the two instrument oscillators on their own was twice as long, they may actually have had better short term stability than the GPSDO's oscillator. While the GPSDO should still have either one bested in terms of long-term drift, I was a bit disappointed in the short term timing and the strange frequency artifacts i measured in this test.
Probably still worth the $140 or so it cost, but I had higher hopes for it than what this suggests.