The measurements in this plot were made using the 13dB LO mixer. We increased the gain on the SR560 going into the daq to 1000 and AC coupled it to get above the daq noise floor.
Measurement 1 was made when the phase of the clocks was not quite exactly pi/2. The issue here is that we can be sensitive to amplitude fluctuations as well as phase. The output voltage from the mixer was at -200mV. We then swapped the bnc cable lengths to get as close as possible to a pi/2 phase difference. In the end we got to 20mV (the calibration here is 1v/pi rads and at 0v the signals have pi/2 phase).
Rana noticed that according to the clock's spec they are pretty sensitive to temperature fluctuations. The third measurement in these plots was with some 'thermal insulation' on the clocks. We didn't leave them long enough to reach any sort of equilibrium temperature, but it should have blocked out the external fluctuations a little. Koji and Jenne then stole the thermal insulation from our experiment - thus putting personal comfort above scientific discovery (though Koji at least did elog what time they did it).
I changed the mixer in the phase measurement for a 7dB local oscillator one at around 7pm. Data from before this time should still be calibrated using the 1V / pi rads value. At this point I also had to change the gain on the SR560 to 100 to stop it railing.
When I changed the mixer this evening, I also tried unplugging the 1pps signal that is locking the clocks. There was no noticeable drift, and I think we could easily make a measurement by leaving them synced and then unplugging them just before we measure.
Rana also suggested that we try making making a measurement with a phase difference of 0 rad, to make an amplitude noise measurement.