Over the past couple days (in between overhauling malfunctioning PDs), we have been trying to hunt down the excess low-frequency noise. Last week, with Koji's help, we essentially ruled out the possibility of scattered light noise from the transmitted end of the gyro by intentionally reflecting varying amounts of light back into the cavity and observing NO DIFFERENCE in the low-frequency spectrum.
AM from the EOM still seems to be the most likely suspect, and we continue to find ways in which it could couple. Yesterday, we think we traced the crazy AM level modulation that I mentioned in my last post to the iris we are using to isolate the proper beam out of the AOM setup. The double-1st order (desired) and single-1st order (undesired) beams are extremely difficult to isolate from one another, so we had to use a very small aperture that we think was coupling EOM jitter into AM quite strongly. We moved the iris down the beam path closer to the waist(s) and were able to get better isolation with a larger aperture. The drift in AM level now seems absent.
While the AM levels in the CCW and CW beams (as measured on the RF analyzer) are not simultaneously minimized at the same pre-EOM HWP angle, there is no longer a large-angle discrepancy; the difference is <1 degree and the noise in both beams can now be kept fairly low at some compromise angle. I have set up a PDA255 in each path immediately before the cavity to monitor them both simultaneously. By adjusting the HWP before the EOM and fine-tuning the EOM orientation, I was able to get the AM peaks in both beams to be <10 ppm relative to the carrier.
After re-locking the gyro, I saw NO IMPROVEMENT in the low-frequency noise yet again. I replaced the PDA255s and tried looking at the LF noise directly after demodulation. I saw some excess noise above the broadband floor below about 1 Hz, which is roughly where the excess gyro noise begins. Upon deliberately de-tuning the HWP so that the RF peaks were >20x higher, however, I saw NO NOTICEABLE CHANGE in the audio spectrum. Looking at the peaks in the RF spectrum more closely, I discovered that the linewidths are below 1 Hz, suggesting that this may not be the source after all.
I suggest that we vent the chamber and obstruct the cavity so that we can run the same tests with the REFL PDs themselves. Then, we can use the noise we measure in the error signal to easily and accurately estimate the contribution to the NB.