Here is the current gyro noise spectrum, as computed from both the AOM and PLL control signals:

It is not any better than before, but there is one important thing going on here: the noise in the PLL signal is lower than that in the AOM signal between ~1-100 Hz. The noise in this band is dominated by oscillator phase noise (from the AOM driver or from the PLL LO, depending on the signal). This marks the first time that we can verifiably show that some noise in the locking loops is suppressed in the transmission readout. This has been predicted by our noise model for quite some time, but we haven't really been able to demonstrate it yet.

In fact, it's been there ever since we rebuilt the gyro, but I had forgotten a factor of two in the AOM calibration due to the double-pass. If you look at some older plots, you can see that the noise at high frequencies differed between the two signals by just enough to make the mid-frequency noise look the same (while even making the AOM noise look BETTER than the PLL noise at low frequencies).

I'm pretty confident that what we see now is correct. The noise above ~200 Hz is well described by the "spillover" noise, so the only mystery that remains is the excess low-frequency noise. It appears to be exactly the same in both loops, so this isn't something that is being suppressed by the CW loop gain. This has been verified by watching the LF noise as we turn the CW boosts on and off.

The main suspect at the moment is the RFAM drift we see as a modulation of the primary-loop error signal offset, so this will be our next focus. I will soon have a quantitative analysis of the AM/PM levels we are getting from the EOM, and drifts with time therein.