Last night we saw a familiar bit of excess noise return -- a double-hump feature between 500Hz and 1400Hz, see first plot. This noise had been intermittently observed up until March 3 (see Gabriele's discussion here) -- after this we were successful in closing the corner station ASC loops in low noise, and we thought we had driven it away. But last night it was back.
The noise decays over a timescale of tens of minutes, which makes us think of a slow alignment loop that drags the IFO towards an operating point where the noise that is causing the bump doesn't couple to DARM. Also the intermittent appearance of the noise makes us think that the coupling depends on the initial alignment, or at least something that changes from lock to lock.
The only alignment loops that were changing on the timescale of the noise last night were the YAW loops in the IMC. The second plot attached shows a bandpassed trend of DARM_IN1 (top) and a trend of IMC-DOF_3_YAW, for the two hours as the noise disappears.
The third and fourth plots are a two-hour trends of other ASC signals (same timescale as the second plot). The corner station ASC (third plot) and the ETM loops (fourth plot) arrive at their preferred output values fairly quickly, much faster than the decay rate of the noise.
Summary:
It's the BS butterfly thing at 2449Hz, down converted by the OMC dither at 3300Hz to make 851Hz line, and then the OMC LSC feedback shakes the OMC length at that frequency finally upconverting low frequency OMC length signal into the OMC transmission at around 851Hz.
Details:
The twin peaks are centered at 851Hz (top). Blue is when it was with a hump, red is when it died down. You can also see that the line at 2449Hz was a factor of 30 or so larger when there was a hump.
According to Dan, 2449Hz is the BS butterfly mode, and 851 Hz is the difference between the BS butterfly and 3300Hz OMC length dither.
On the bottom plot, the OMC PZT2 monitor shows a huge 851Hz line, and this line was also about a factor of 30 larger when there was a hump. This is the PZT used for OMC length control.
On the second plot, I shifted the top plot such that the frequency axis starts at 851Hz, and on the bottom plot I put the OMC length error signal. This shows that the features match up, and the width of the half-hump roughly agrees with that of the OMC length.
These things all mean that we're unnecessarilly shaking the OMC length at 851Hz with too large an amplitude, and that the noise will probably go away if we notch this line somewhere in the OMC length control path. We can also more aggressively low-pass the OMC length feedback.
There is now a 6th-order butterworth bandstop filter in the input to the OMC LSC demodulation for 2.4k-2.5k. There are also additional butterworth rolloff filters 1k in the demodulation products (OMC-LSC_X_SIN and COS) in case any of the other high frequency bumps in the spectrum are due to the OMC length drive.