On Friday evening, I acquired the primary error signal into the ADC with the cavity obstructed so that it would not resonate. I tuned the HWP before the EOM so that the error signal was centered about zero. I wanted to see if the RFAM was truly oscillating in amplitude or just monotonically increasing to some level after I had minimized it. Here is a short (5-hour) time series beginning right around then. The following 48 hours looked essentially the same, with a slight variation in the oscillation period and peak-to-peak level.
Today, I tried to get a quantitative handle on the RFAM level, but I was sidetracked by the weird behavior. I once again tried to minimize the circular light getting through the input optics by iteratively adjusting the initial QWP and the HWP immediately after it and minimizing the P light getting through the PBS. I was able to obtain a maximum contrast of 55.5 uW to 159 mW over the full range of the HWP for the optimal QWP setting, so that only ~0.03% of the light remained circularly polarized.
Then, I adjusted the HWP just down the line (immediately before the EOM) to center the far-from-resonance error signal about zero. I observed the same drift as before. I then put my hand on the metal case of the EOM and the drift increased rapidly, indicating that the time dependence of the RFAM is likely due to thermal coupling.
I verified that the beam was not clipping on either end of the EOM (there was some scatter from the edges of the beam, but the spot was centered on the aperture). I tried changing the alignment slightly to see if the drift lessened, but this failed.
I then tried swapping out the HWP for another---no dice. Then, I swapped the EOM for another broadband ThorLabs modulator of the same model. This seemed to work for a few minutes, as the drift appeared less pronounced, but it worsened quickly. It also responded the same way to the "grab the damn thing" test.
An interesting observation: the HWP orientation that minimizes the RFAM when looking at the signal from the PD on the RF analyzer IS NOT the same one that centers the error signal DC level at zero (in fact, the cavity will not lock at this setting). Conversely, when the error signal is centered at zero by setting the HWP, the PD signal shows strong RFAM (~40 dBm above noise level). This is baffling.
As my frustration was maturing, I caught the following sight (several of them, actually, but this was the only one I froze in time) on the scope:
Green is TRANS_DC, yellow REFL DC, blue ERR_INMON (note the offset), and purple the actuation signal to the PZT. The cavity has been locked for quite some time, but the slow loop is not engaged. As the length of the cavity slowly drifts, the feedback signal hops discretely from one level to another. This particular one is about a 3.5-volt step, corresponding to about 14 MHz. I am no expert, but this looks to me like some sort of mode hopping. There is not (and has never been) a dedicated faraday isolator at the experiment input, so I think it is quite possible that back-reflections are getting back into the laser head. From what I understand, mode hopping is uncommon in a short NPRO, and I'm not sure how this would affect the output polarization of the laser, but it seems suspicious. By the time I gave up, the circular light contrast had decreased by a factor of more than 4 (i.e. the minimum achievable power with the same QWP orientation was > 200 uW).