【Craig, Evan】
We took a look at the noise is AS A RF45Q in and out of lock. This sensor is used for dHard and we wanted to see if we could gain some insight into the cross-coupling issues between DARM and the angular loops (some previous insight here and previous decoupling work here). We mostly looked at the sum channel (essentially an rf DARM readout) rather than pitch/yaw. There aren't any stunning conclusions here, but if the sub-10 Hz noise is better understood it might provide some clues about how to make improvements.
We found some nonstationarity in the dark noise (wandering lines). With light on the diode in full lock, in the region 10 Hz to 100 Hz it seems like there is some broadband excess that is not a straightforward optical signal (from DARM or dHard).
We wonder if it would make sense to try more whitening on this sensor, as currently there is only one stage engaged (AS B RF45, which is not used for anything, has three stages engaged and hence better dark noise performance).
Dark noise
The first attachment shows the dark noise versus the in-lock noise of this sensor. During the maintenance period we unlocked the IMC and proceeded to measure the dark noise, and we were surprised to find a slowly wandering line in the vicinity of 4–7 Hz along with its harmonics (measurements in green and magenta). We were again surprised when we looked at the individual Q segments and found different responses to these lines (second attachment). For example, Q1 seems entirely insensitive to the fundamental of this line, but is the most sensitive out of the four quadrants to the harmonics. This second attachment also shows the dark noise of the AS B RF45Q quadrants, but since AS B RF45 has more whitening than AS A RF45, the two diodes do not readily lend themselves to comparison.
We also looked at some past times during maintenance days when it seemed the diodes had no light on them (although the IMC was unlocked); one such time is also plotted in the second attachment. Here there are no wandering lines, but the noise at and below 1 Hz is worse (the whitening settings have not changed, according to time machine). Maybe there was some signal since the IMC was still locked.
In-lock noise
The first attachment again shows the noise in AS A RF45Q sum during the full lock. The spectrum at and below 10 Hz is presumably optical signal (DARM plus other effects). The noise above 100 Hz is white, consistent with shot noise. Between 10 Hz and 100 Hz the high-frequency noise appears to aquire some shape that is not white. This region is not coherent with DARM except near various lines (e.g., the dither lines).
The rough calibration was reckoned as follows. Based on previous estimates of the AS port sideband content and the rough number of 250 mW of power leaving the SRM (via the calibration of AS C into milliwatts and the 800 ppm transmission of OM1), it seems that the 45 MHz beatnote signal should be of order 2 mW. Together with the dc value of the AS A RF45 sum channel in counts (about 10000), this can be used to infer that the total measured noise at high-frequency is about 10−7 mW/Hz1/2.
Comparison with DCPD sum
We also made a comparison of AS A RF45Q against the DCPD sum (third attachment). The rf sensor is scaled to match the dc sensor in the region 5–10 Hz, where the two are highly coherent. Above 10 Hz, the rf sensor's noise dominates over the dc sensor. The interpretation of the noises below 5 Hz is less straightforward. One thing that is apparent is that there are regions of high coherence (e.g., around 0.8 Hz) even though the two sensors show spectra with vastly different amplitudes. The sensors could be seeing the same noise, but with different coupling factors. Alternatively, a sensing noise peculiar to the dc readout (e.g., from the OMC) could be being impressed onto the loop and then witnessed by the rf sensor.
Plots for a longer stretch (6 hours, starting at 2019-09-21 09:00:00) of median-averaged data are also attached, with smaller binwidth. Here the large variability in the low-frequency coherence is more apparent.
A next step would be to run bruco on the DCPD sum and the RF45Q sum to see what the major contributors are at these low frequencies. I tried running bruco on Caltech ldas-pcdev12 (for only 1000 seconds), but it just hangs.
I am also attaching a comparison between AS A and AS B RF45 sums, along with some ADC count comparisions. AS B RF45Q maintains coherence with DARM up to higher frequency and has an overall lower noise floor than AS A RF45Q. AS B has not received the same careful rephasing as AS A (there is some DARM sensitivity in the I quadrature), but nonetheless is phased so that DARM mostly appears in Q.
It is also interesting to note that the low-frequency regions of high coherence with the DCPDs (0.8 Hz, 1.27 Hz, etc.) show up in both quadratures for both AS A and AS B even though (at least in the case of AS A) the phasing minimizes the appearance of DARM in I above 10 Hz.
The ADC count comparisons seem to indicate that turning on two extra whitening stages on AS A will increase the rms from 200 ct to something like 500 ct.
We turned on whitening stages two and three for AS_A_RF45. We did this in lock by switching the DHARD sensor from AS_A_RF45 to AS_B_RF45 with a gain of -1.25 in both the PIT and YAW sensing matrix. Preliminary results suggest no DARM noise improvement, but lowered dark noise in the AS_A sensor above 10 Hz. Noise was lowered by around a factor of 2. Accepted this change into the SDF, and are back to observing.
There was actually a bit of improvement in DARM between 15 and 20 Hz, though it's hard to see without sufficient averaging.
I took an hour of "before" data starting at 2019-09-27 17:30:00 and an hour of "after" data starting at 2019-09-27 20:25:00 and median-averaged it. The plots are attached (I omitted the DARM / dHard yaw coherence to keep the plot from getting overcrowded -- the overall coherence is lower than the pitch DOF). The improvement is subtle, but it's there, particularly at 19 Hz. The changes in coherence and in noise level are roughly consistent with the noise at 19 Hz being dominated by angular fluctuation.
Nothing is particularly surprising here given the previous noise budgeting injections.