I have spent some time updating the REFL WFS noise budget. Similar work was done by Evan Hall, (see alogs 63873 and 64448). Updates to IM damping and PR3 damping reduced the WFS noise (see master alog 64573).
See plot 1 for the REFL WFS I pit budget and plot 2 for I yaw
The main contribution to the WFS noise below 10 Hz is PR3 M1 damping from L, P and Y. Between 10 and 30 Hz, there is coupling from the HAM1 TT L4Cs (RX, RY, X and Z), and PR3 V damping. In WFS pit, the strongest coupling in the L4Cs is RY and X, and for WFS yaw it is RX and X, however all four L4C channels have coherence with WFS pit and yaw. The coherence is stronger in WFS A and B, as we have seen before, and we have assumed it is due to the placement of the WFS on the HAM1 table. The PR3 vertical damping only contributes from 10-25 Hz, and it is stronger in WFS pit than WFS yaw and also stronger in WFS A than WFS B.
There were efforts this summer to apply HAM1 L4C feedforward to CHARD pitch (see alog 64183), however they were mostly unsuccessful. If there is noise contributing in the 10-30 Hz band from both the L4Cs and PR3 damping, perhaps that explains why the feedforward was unsuccessful despite the evidence that it should have helped. As a note: Jim left the HAM1 FF from July 27 until Jenne turned it off August 25 while she was working on nonsens. It has not been turned back on, but a comparison of the REFL WFS during that time and after Aug 25 show no appreciable change.
I went back in time and measured the coherence between PR3 vert and the WFS channels. It appears that that HLTS damping updates that Evan and I applied (and Jeff developed) had no effect on this coupling. The only change occured from June 5 to July 20, when I had improved the IM damping loops, but Evan and I had not yet improved the PR3 damping. See coherences from June 5 (before all damping changes), July 20 (after IM changes but before PR3 changes), and Aug 20 (after all changes).
Here are the details of how and when I took my measurements, and how I quantified each noise source on these plots:
- I measured the locked WFS spectra from one of the last locks before the corner vent, Sept 4, GPS time 1346302951
- I converted the spectra from WFS counts/rtHz to W/rtHz with these factors: ct/ct gain of 1 and 1.41 (RF9 and RF45 respectively), transimpedance of 900 and 650 Ohms (RF9 and RF45), whitening gain of 21 and 15 dB (RF9 and RF45), plus conversion gain from RF input to Q output of 22 dB (see DCC T000044), 16 bit +-20V ADC and quantum efficiency of 0.8 A/W.
- I also measured the WFS A and B DC counts at that time. I converted these counts into the power using transimpedance of 1 kOhm, 16 bit +-20V ADC, and quantum efficiency of 0.8 A/W. I calculated the shot noise in W/rtHz with the power (S = sqrt(2*h*f/c * P))
- I measured a "dark" WFS spectra during the vent by taking the IMC offline and measured the WFS with no laser light. I then turned off all whitening and anti-whitening stages on the REFL WFS and remeasured the spectra to allow me to characterize the ADC noise
- REFL WFS 9 and 45 use 2 stages of whitening/antiwhitening. My first dark measurement with whitening can be thought of as dark_meas = N_electronics + N_adc(AW), where AW is the anitwhitening stages. Without whitening/antiwhitening engaged it is dark_meas_nowhit = N_electronics + N_adc. Using these two measurements: N_adc = sqrt( dark_nowhit^2 - dark^2)/sqrt(1 - AW^2) (remember that AW= two stages of antiwhitening applied)
- All four TTL4C channels have a contribution to the WFS, so I applied wiener filtering using the four L4C channels as the witness and each WFS channel as the target. Bandpassing the data around 10-30 Hz before filtering gave the best estimation of the noise contribution
- I applied a coherence projection for PR3 damping contribution. It would probably be best to apply wiener filtering to the L,P and Y channels for the coherence from 0.1-1 Hz but I am having a hard time applying the filtering, and the coherence projection gives a rough idea of the contribution.
- The plots shown only characterize the REFL WFS I channels, as these are the only channels currently used for ASC (CHARD and INP1).
As a last note, in areas where the WFS spectra appears to be shot/electronics-noise limited, there is a discrepancy between my estimation of the shot/electronics noise and the actual floor of the WFS for the RF45 channels. I don't know why this is, and it bothers me!
To make the PR3 contribution more clear, I have attached a WFS pit and yaw sub budget showing the coherence projection from PR3 P, Y and V. Note that V is the main contributor at 10 Hz and above.
Furthermore, I have attached a screenshot showing spectra of PR3 vertical M1 damping inmon and HAM2 ISI PR3 suspoint V. I also included the coherences of the suspoint vertical to each of the REFL WFS, which is coherent to the WFS in the same band as the PR3 M1 damp inmon.
Found coherence between the HAM1 TTL4Cs and the PR3 M1 vert damp in 10-30 Hz region. Perhaps this is the same seismic motion coupling into the WFS in two ways?
Coherences, oh my!
Following up on several comments, suggestions, and questions, I have made a series of plots:
- Coherences between the PR3 M1 damp inmon and HAM2 ISI GS13 cartesian projections, LTV and RPY
- Coherences between the REFL WFS and the HAM2 ISI GS13 cartesian projections, PIT and YAW
- Coherences between the HAM1 TT L4Cs and the HAM2 ISI GS13 cartesian projections, RX/RY and X/Z
- ISI HAM2 GS13 and PR3 suspoint spectra, GS13 and PR3. Spectra are in uncalibrated (or whatever the channel is calibrated to) units/rtHz. Interesting to me is the GS13 Z and the PR3 V seem to be moving ~factor of 10 more than the other dofs
- PR3 M1 DAMP IN1 spectra, here, once again in uncalibrated units/rtHz. The Vertical dof is not flat from 10-20 Hz like the other dofs are
What does this mean? At first it appeared to me that we had two separate problems: table motion of HAM1 coupling to the WFS, and PR3 motion coupling to the WFS. Now I am starting to wonder if this is related noise. For example, we feedforward the HAM1 table motion, but it has no effect on the WFS spectra because PR3 is still experiencing the same motion and coupling it into the WFS. I'm not sure if that is correct, but that is one way I am thinking about it. I also think it is strange that the PR3/HAM2 motion seems to be strongly vertical/z direction, while the HAM1 L4Cs are showing the motion in all cartesian dofs. Is it possible the projection into cartesian of the four L4Cs is not quite right, so there is a bit of Z motion in all four coordinates? All 6 GS13 cartesian coordinates are coherent with the PR3 M1 vertical as well.
I feel fairly certain this is not sensor noise from PR3, as Evan and I retuned the damping loops to ones that Jeff designed to avoid sensor noise reinjection. As I said above, the coherence from WFS to PR3 vertical inmon stayed consistent even after the damping loop changes. I already posted those plots, but I am reposting them with labels to make the point more clear. I chose June 5, before we made any damping loop updates, July 20, after the IM damping loops were changed, and Aug 20, after the PR3 damping loops were changed. The coherence exists on June 5 in almost all sensors, and is present on July 20. It appears unchanged on August 20. Therefore, I don't think this is sensor noise reinjection from 10-30 Hz, as any sensor noise would have been greatly suppressed by the PR3 damping loop updates. If anything, updating the IM damping loops increased this coherence between PR3 and the REFL WFS, possibly because the IM damping loop noise was the dominant noise source in that frequency band in the REFL WFS at that time.
I finally found the error that has plagued my mind for some time regarding my noise budget for the REFL WFS. If you look at my budgets in the original alog, you will see that there is a gap between the measured RF45 WFS spectra, and the known noises, especially the dark noise trace, which I believe should be our limiting noise source in RF45 just as it is in RF9.
I forgot to ensure that I have the correct in-lock gains on the RF45 chain. I was using values that may have been correct at some time, but due to IFO changes, they are no longer correct. Namely, when unlocked, there is a 6 dB gain factor applied to the RF45 path, and then when locked there is an additional 6 dB of digital gain to compensate for the modulation depth reduction of 6 dB when we power up. That means my conversion factor for the ADC and dark noises should have been divided by 6 dB gain and my full lock spectra should have been divided by 12 dB gain. (My error was that I had divided 3 dB of gain across the board for RF45).
With that correction, the budget no longer has that annoying gap, and means we can stop coming up with silly reasons for the discrepancy. Whew.
We could consider adding L-4Cs to stage 0 of HAM2 (and HAM3). Jim Warner has been able to do very good feedforward with these sensors and improve the performance in the 10-30 Hz region. I've attached an overlay plot of the Z motion for HAM2 vs. HAM5 for March 28, 2023. Original plots are from LDAS page for HAM2 and HAM5. The plot shows that HAM5, which has feedforward L-4Cs attached to stage 0 inside the vacuum system has ~5-10x better performance at the frequencies (10-20 Hz) where the PR3 damping and the HAM-ISI Z motion have high coherence. Since the two plots use the same color scheme, I've added a label to show which of the green lines belongs to HAM2 vs. HAM5. HAM5 is the quiet one.
Spectra from last night comparing HAM3, HAM5 and HAM7 up to 75 hz. First plot is the Z gs13s. If we are improving the sensors on HAM2-3, we could also thing about doing the cps upgrade, HAM7 FC1 suspoint is better than HAM3&5 over .5-20hz, mostly due to the lower noise CPS. It's worse above 20 hz because it doesn't have the passive isolation from HEPI.
