Unless someone is running a configuration test, this state is not nominal and should be corrected. Usually, we can switch this with out too much stirring up of the platform.
no restarts reported
I spent some time today on the automation of our inital alingment steps.
Following Arnaud's work on the SRM M2 stage (alog 13558), I made a couple of modifications on the L2P filters that he implemented in this past August. It is now engaged, but I did not get a chance to evaluate the performance of it yet.
(Merging the two filters into one)
First, as he mentioned in the alog, a filter in FM2 that he installed was numerically unstable. It seemed that the FM2 filter had been meant for the inverse of the P2P response which grew up at high frequencies as f4. Therefore the filter was accidentally designed to return incredibly high values at the Nyquist frequency. In order to fix this issue, I decided to combine this filter with the other one, i.e. L2P such that the high frequency response becomes flat. Even though this idea of combining two or multiple aggressive filters into a moderate one is generally good, foton did not allow me to do this due to too many numbers of poles and zeros this time. I could have split the filters again into two moderate ones, but instead I decided to drop off some pairs of poles and zeros which are so similar to each other that dropping them did not change the overall response so much. Also, I took out a 2nd order zero at 20-ish Hz in order to let the L2P decoupling filter roll off at high frequencies because I was worried about saturation in DACs especially for the 3f locking. The attached is the transfer function of before and after the modification. Of course, now the filter is not unstable any more.
As seen in the attached, the modified version is accurate until 3 Hz or so and it completely deviates at high frequencies. I am hoping that this is OK because the M2-M3 cross-over is now as low as 4.5 Hz and therefore the discrepancy at high frequencies does not matter. I did not get a chance to see the performance of this filter yet.
no restarts reported
We noticed that in the past there were some long lags between turning the fans on in the Laser Room and turning the two airconditioners on. This led to temperature spikes because the fans rapidly heat the room if the AC is not on. The source of the problem is that the Mitsubishi air conditioners can not be turned on by the same Unitronics control system that controls the fans and must instead be turned on at the thermostats in the Laser Room. This might be hackable, but for now, the proceedure will be to turn on the Laser Room fans at the Unitronics box in the Ante Room instead of at the Unitronix box outside the PSL. The full proceedure is attached to this log and posted at both Unitronics boxes.
Robert, Rick
In a previous alog entry I pointed out that most of the intensity nosie we see in transmission of the IMC is due to input beam jitter converted to RIN due to an IMC misalignment. Today, to prove this, I improved the IMC alignment in three steps:
The third attached plot shows the improvement in RIN in transmission of the IMC. Now it is at a level of 1e-6 at 10 Hz and 2e-7 at 100 Hz. It's almost a factor 10 better everywhere.
It's interesting to note that the RIN is still non stationary, so we should improve further the IMC alignment accuracy. I could not increase more the gains of DOF1 and DOF2, since I got a 1 Hz instability (as expected from the open loop transfer function). However, my intuition is that the IMC mirrors are pretty much not moving, and instead the input beam pointing is moving a lot. So we should servo the input beam to the IMC cavity axis with some Hz of bandwidth. This is what we did at Livingston, where the input beam motion was limited by air currents in the PSL room. I believe a similar approach should be used here. This is much better than high bandwitdh loops on the mirrors, since they should be our best reference.
To study the dependency of RIN on IMC angular fluctuation I used a code developed for a similar task at Livingston. See more details here. In brief, i compute the band-limited RMS of the IM4_TRANS_SUM signal between 50 and 10 Hz, and correlate this with the low frequency content of the IMC alignment error signals. The scatter plot in the 4th attachment shows the correlation between each IMC angular DOF and the BLRMS. There is a clear correlation with DOF1_Y. From the scatter plot we can also see that there is an offset on the error signal.
A more quantitative analysis can be obtained by fitting the BLRMS time series with a linear combination of a constant, the error signals and their squared values. The procedure I used is a slightly modified version of a LSQ fit, and it gives me a ranking of the signals as a function of their importance in improving the fit. The code is attached to this entry. Basically, the first step is to find which one of the channels (constant, error signals or their squares) can be best fit to the BLRMS. finding the minimum residual squared error. Typically a constant is the first winner. Then the procedure is repeated with the residual, looking for the single bets channel to furher reduce the rfit error. At each step I search for the channel that reduces the error the most. The procedure is repeated iteratively.
The 5th plot shows the result of the fit before any improvement on the IMC alignment. Most of the noise fluctuations could be explained by angular motions. The 6th plot shows the ranking of the channels (the bigger the bar, the most important the channel is), confirming that DOF_1_Y is our best candidate.
The 7th plot shows the measured open loop transfer function of IMC DOF 1 Y before and after my gain increase. The loop had a bandwidth of probably 3 mHz with the original gain of -1. The plot shows the OLTF with a gain of -100, giving a badwidth of 300 mHz. I did not measure the other DOF1 and DOF2 loops, but I could increase their gain in a similar fashion.
After some tweaking, I reduced the gains to -20, since the error signals were showing a large 1 Hz instability, which is consistent with the measure OLTF. However, we are clearly still limited by the residual motion of the beam with respect to the cavity axis.
I'm leaving this configuration running, as shown in the attached screenshot. In case there is any problem, revert back to the original configuration by reducing the gains to -1 and removing the two offsets. No other modification will be needed.
Since I don't like much the idea of running WFS loops with offstes, I tried to zero them by moving the beam on the WFS using the picomotors, but I couldn't get any effect at all, no matter how big step I was moving.
To clarify, all the measured RIN reported here is without engaging the ISS second loop.
model restarts logged for Fri 03/Oct/2014
2014_10_03 02:41 h1fw0
2014_10_03 14:26 h1fw1
unexpected restarts of fw's
Alexa, Sheila, Kiwamu
ALS COMM is robust, we can easily transition from green to IR lock. We measured the noise to be about 20 Hz RMS.
Meanwhile, we had a bit of trouble with ALS DIFF. First the DIFF VCO is railed. To handle this we used an ezca servo to feedback to ETMX. The ezca command we used is: ezcaservo -r H1:ALS-C_DIFF_PLL_CTRL_OUTPUT -g 100 H1:SUS-ETMX_M0_TEST_OFFSET. We also had to set the tune voltage to 5.526V in the Y VCO. This technique would bring the DIFF VCO into range so we could engage the DARM feedback. We were able to close the loop with a lower gain (input matrix set to 0.05). We quickly leanred that the L2P in both ETMX and ETMY are bad (ETMX was worse than ETMY). We need to examine these filters again, and do something similar to what was done with the BS.
I checked the ETMX L2P filters today and found that the L2L gain in the DRIVEALIGN matrix on the L1 (UIM) stage had been set to a wrong value of 10. 
This explains:
I set it back to 1. This should help us finding a neutron star merger.
It seems that the gain had been like this since 20th of last July for some reason. I set it back to 1. Then I checked the performance of the L2P decoupling by injecting a sinusoidal wave at 0.1 Hz (which is kind of the frequency of the signals that we have applied for the ALS DIFF loop) with an amplitude of 104 counts at the input of the DRIVEALIGN matrix. I was able to confirm that the correct L2L gain reduced the angle coupling significantly.
By the way, L1 stage's L2Y decoupling filter had been off by setting the gain to 0. This seems to be the right setting because I did not see a large motion in yaw when driving the longitudinal. Setting the gain to -1 seemed to just introduce a coupling. So I conclude that the L2Y should be off.
The document attached shows the Yaw motion (first page, left) and Pitch motion (second page, left), Friday morning at 1 am PT, for all test masses.
All units were controlled with the blend configuration imported from LLO. The sensor correction was OFF.
Compared to the previous nights measurements, ITMY optical lever has been re-centered. Values look reasobable now. Also the combs in ETMY are gone.
ITMX does look higher than the others, as seen during the previous days by the commissioners in the time series, but it might be due to calibration error (see Jeff's alog).
In general:
- the Yaw rms values seem dominated by micro-seism (or sub-microseism for ETMY)
- the Pitch rms values seem dominated by the 0.5 Hz suspension resonances
Alexa, Sheila, Gabriele
We closed the servo loops for the relf wfs IR y-arm centering. The error signals are fed back to the test masses as follows:
| Pitch Error Sig | Pitch Gain | Yaw Error Sig | Yaw Gain | Filter Bank | |
| ETMY | RELF_B_RF9_I | -0.001 | RELF_B_RF9_I | 0.005 | DHARD |
| ITMY | REFL_A_RF9_I | -0.001 | REFL_A_RF9_I | -0.003 | DSOFT |
The DC centering must be on; these were unchanged. I have also taken a screen shot just in case (the input/output matrices are the same for pitch and yaw).
I wrote a python script that implements the same angular loops, but actuating directly on the alignment offsets. In this way it is not necessary to offload the servo output at the end of the alignment. The loops operates only if the arm transmitted power is above a threshold. They continue to operate as long as the error signals are larger than another threshold value. The check on the error signal is perfomed with a running average, to smooth their noise. All parameters are set at the beginning of the script.
As shown in the attached figure, the loops are working well. Maybe the gains are still a bit low.
To ease the implementation of this alignment techniques into the guardians, I'm not using anymore the threaded trigservo loops as at Livingston, since it was not clear to me how to terminate the threads from inside the guardian. Now everything is local to the main thread. The script is attached. It should be easily configurable to any other configuration and number of degrees of freedom
J. Kissel, J. Driggers There's been some confusion about why the ITM ISIs are performing so differently. This resurrected memories of mistrust in the optical lever calibration, that I'd talked to Thomas about in June (see LHO aLOG 12216). That aLOG leaves it at "we should confirm the calibrations once we have arms again." The idea will be to cross-check the calibration using the alignment sliders, which can be precised calibrated using the green beam and baffle PDs as a 4 [km] optical lever. The details of the process are below, but in summary, the refined calibration of the sliders is IX P = 15.589 [ct/urad] IX Y = 43.835 [ct/urad] IY P = 19.920 [ct/urad] IY Y = 46.700 [ct/urad] which, for IX P is a ~50% correction, but the rest is a ~10-20% correction. We have updated both the OPTICALIGN [ct/urad] gains, and corrected the "cal" [ct/urad] FM0 in the M0 LOCK P and Y filter banks, saved the newly corrected alignment offsets, and confirmed that the refinement of gain and slider value brings the optic to the same place. DETIALS ------------ Currently, the alignment slider calibration gains are P 23.219 [ct/"urad"] Y 51.689 [ct/"urad"] based on dead-reckoned knowledge of the actuation chain (see LHO aLOG 4730). Keita, Alexa, and Shiela recently went through the exercise of finding the green beam down the arms (see LHO aLOG 14201 for ITMX, and LHO aLOG 14161 for ITMY) using baffle PDs and the SUS alignment sliders. From these aLOGs, we know that the ETMY baffle PDs, and ETMX ERM Pattern are at the following ITM alignment slider values: P ["urad"] Y ["urad"] ETMX Top Stop 65.2 ETMX Bot Stop 84.2 ETMX Lef Stop -20.2 ETMX Rig Stop 3.8 ETMY PD1 165 -105.3 ETMY PD4 197 -139 for displacement claimed by the sliders of delta ["urad"] EX P -19 * 2 = -38 EX Y -24 * 2 = -48 EY P -32 * 2 = -64 EY Y 33.7* 2 = 67.4 where the factor of 2 comes from the single bounce, optical lever effect. From an older aLOG when Keita had performed a similar calibration (see attachment to 9087), we know the ETMX baffle PDs are a distance 11.76 [inches] = 0.2987 [m] apart in vertical, and 11.77 [inches] = 0.29895 [m] in horizontal. From D0900949, we know the pattern has an inner-most diameter of 226 [mm] = 0.226 [m]. The length of the arms is 3994.5 [m], confirmed by older measurements during the HIFO days (see LHO aLOG 9635 and 11611). That means the physical displacements are EX P 0.226 [m] / 3994.5 [m] = 56.6 [urad] EX Y 0.226 [m] / 3994.5 [m] = 56.6 [urad] EX P 0.298 [m] / 3994.5 [m] = 74.6 [urad] EX Y 0.298 [m] / 3994.5 [m] = 74.6 [urad] Hence the slider gains should be corrected by IX P 56.6 / 38 = 1.49 [urad / "urad"] IX Y 56.6 / 48 = 1.18 [urad / "urad"] IY P 74.6 / 64 = 1.17 [urad / "urad"] IY Y 74.6 / 67.4 = 1.11 [urad / "urad"] or IX P 0.67138 ["urad"/urad] IX Y 0.84806 ["urad"/urad] IY P 0.85791 ["urad"/urad] IY Y 0.90349 ["urad"/urad] The new slider gains are therefore IX P 23.219 [ct/"urad"] * 0.67139 ["urad"/urad] = 15.589 [ct/urad] IX Y 51.689 [ct/"urad"] * 0.84806 ["urad"/urad] = 43.835 [ct/urad] IY P 23.219 [ct/"urad"] * 0.85791 ["urad"/urad] = 19.920 [ct/urad] IY Y 51.689 [ct/"urad"] * 0.90349 ["urad"/urad] = 46.700 [ct/urad] Which means the former alignment values, in ["urad"], P Y IX Aligned 75.4 -7.6 IX Misaligned 0 -52.75 IY Aligned 182.4 -116.6 IY Misaligned 0 0 now become (["urad"] slider value * [urad/"urad"] = [urad] slider value) P Y IX Aligned 112.35 -8.968 IX Misaligned 0 -62.245 IY Aligned 213.41 -129.43 IY Misaligned 0 0 We have updated both the OPTICALIGN [ct/urad] gains, and corrected the "cal" [ct/urad] FM0 in the M0 LOCK P and Y filter banks, saved the newly corrected alignment offsets, and confirmed that the refinement of gain and slider value brings the optic to the same place.
After it was decided that the existing design for the TCS HWS table light pipes was insufficient to protect the vacuum envelope we decided to use what everyone else uses, some 10" self sealing flanges and the lexan viewport assemblies. Anyhoo, the commissioners should be happy I won't need to be thumping around on HAM4 any longer.
The rotation stage on ITMy had become unresponsive last week after some cable dressing was done. Restarting Beckhoff had proven ineffective and Fil was unable to find any lose cables. Unplugging and replugging the interlock bypass on the D1300131 box. This possibly had the effect of resetting the relay that blocks communication to the relay stage. Whatever happened the rotation stage is working again.
Andres covered from 10:00 - 11:00. Thank you Andres. 09:15 Betsy to LVEA West bay to work on 3IFO SUS 10:43 Aaron pulling cables between HAM1 and the PSL closet 11:17 Kiwamu to end Y to align optics on the ISCT table 11:56 Corey walking around IMC for ~ 5 minutes 12:00 Chris to end X to check on moth situation 12:04 Aaron pulling cables over HAM1 and HAM2 12:36 Sheila, Kiwamu at end Y 12:40 Aaron done pulling cables over HAM1 and HAM2 12:44 Chris back from end X 14:08 Andres to TMSX lab to look for parts 14:42 Andres back from TMSX lab
Peter K, Gabriele
As reported yesterday, the second loop servo didn't have enough gain, so this morning we modified one switchable stage of the board to include a x10 gain. Unfortunately, that stage is an inverting one, so we can't really switch it on and off with the ISS second loop closed. This stage is controlled by H1:PSL-ISS_SECONDLOOP_ADD_GAIN.
Using everything we have in the board (maximum gain at +40dB, x10 stage, integrator and boost) we could engage the second loop and get the performance shown in the first attached plot. The red and blue traces are the in-loop and out-of-loop signals, calibrated in RIN, when the second loop is open. The green and maroon traces are again the same signals, with the loop engaged with maximum gain. The cyan and magenta traces correspond to the integrator and boost switched on. We're getting a bandwith of about 1 kHz.
We can see that the out-of-loop signal shows an excess of noise with respect to the in-loop signal, up to about 60 Hz. The out-of-loop RIN is about 1e-6 at 10 Hz and 2e-7 at 60-70 Hz. We can conclude that our ISS second loop photodiodes are limited by sensing noise of some kind at this level.
The second plot shows a comparison of the error signals (sum of photodiodes 1-4 and sum of photodiodes 5-8) with one or the other in loop. The performance is very similar in the two cases, so the amouint of excess sensing noise is about the same in both sets of sensors.
The last plot shows that there is some coherence between the out-of-loop signal and the ISS QPDs. This is true at least at low frequency, below 10 Hz. The QPD signals show a structure at ~18 Hz which is very similar to the one visible in the out-of-loop signal, altough there is no coherence. It might be due to a highly non stationary coupling, but we're not sure of that. So it is likely that at least some of the excess sensing noise is due to beam jitter to intensity noise coupling at the ISS array. to imrpove this, we should try to move the beam using the picomotors and looking at the out-of-loop signal noise in the 10-60 Hz region.
The medm screen of the second loop is wrong, in the sense that the buttons are connected to the wrong switches. I created a new screen for the second loop, which for the moment is living in /ligo/home/controls/gabriele/SecondLoop.adl. Once finalized, we should move it to the right place.
After fixing the not taking Damped TF problem and the channel list mismatch problem, we took a full set of transfer functions for 3IFO Quad-09. The results data are posted below and have been added to the SVN repository. There are a couple of anomalies in the data. (1). The last peak of the Damped M0 data in "Y", "R", and "L" are shifted up. (2). The R0 "Y" DOF appears to not be damped where as the other 5 DOFs are. Note: The staging building test stand hardware and software are somewhat down level to the H1 & L1 test stands.
For those interested in looking closer at QUAD model parameters, attached are plots comparing all of the QUAD Main Chains when suspended with wires and also when suspended with fibers. Note, if QUAD data is missing for one of these configurations it's because there was no clean data available. Between the 2 plotted configurations, all 12 (H1, L1, and 3IFO) QUADs are represented. Note, I tried to chose data sets that had the same or similar environmental conditions, but it was difficult due to the fact that some QUADs were reworked on test stands and some were reworked in chamber. In all cases they were mounted on Solid Stack Test Stands or Locked ISIs and in-air.
Data is committed to the svn and can be found at:
/ligo/svncommon/SusSVN/sus/trunk/QUAD/Common/Data/
There does not seem to be a pattern in the data of the 2nd pitch mode peak which are clustered by a specific type of suspension (ETM vs ITM, or wire segment hang vs wire loop hang).
And now with some cursors and in a second format for Brett.
As suggested, I looked at the stiffnesses of the Top Mass blades to see if there is a correlation with the second pitch mode frequency shifts. I don't see it. In order of the peaks on the P to P plot, starting with the lowest frequency to the highest the blade sets used in each QUAD are:
H1ETMx - SET 9 (~1.28Hz)
L1ETMy - SET 13
L1 ETMx - ?
L1ITMx - SET 14/15
L1ITMy - SET 12
Q8 ETM - SET 8
Q9 ETM - SET 2
Q6 ITM - SET 10 (~1.531 Hz)
The blade sets go in order of stiffness from highest to lowest, so SET 2 is stiffer than SET 15. SET 14/15 is a mixed SET with blades still of adjascent stiffness.
I took the two wireloop quads that have the highest and lowest 2nd pitch mode frequencies and made a fit to them. These measurements and their respective made-to-fit models are shown in the attached plot. QUAD06 (H1 QUADTST) is the highest, X1 ETMX is the lowest.
I previously did a fit for QUAD06, see log 14235. The fit for X1 ETMX was made simply by taking the QUAD06 fit and subtracting 3 mm from dn, which works quite well.
Since the outliers are 3 mm apart on dn, the other quads seem to have an even spread between those, and no correlation with spring stffness is evident, then a possible explanation is that our tolerance on positioning the top mass blade tip height is +-1.5 mm.
Attached is a prediction of what +-1.5 mm on dn would look like for the fiber quads.
The black is a model of H1ETMY (which has been the default fiber model for some time) where dn=1.78 mm; blue is the same model but with dn=0 mm; red is again the same model but with dn = 3 mm. Some data is included as well. The H1ETMY measurement is in orange, which matches well because of the previous fitting of H1ETMY. In purple is H1ETMX. I think H1ETMX corresponds to the wireloop quad X1ETMX, which was the low outlier on dn for the wireloop configuration. In that configuration a dn of 0 mm worked quite well to the fit model to the data. Here the same 0 mm dn makes almost as good of a fit. There is not data matching the dn=3mm. +3 mm was found to work well for the high dn outlier wireloop QUAD06, which is not yet a fiber quad.
So it seems that for the existing fiber quads, +-1 mm on dn explains the spread well. However, the most recent 3rd IFO quads, still with wireloops, are the stiffest yet in pitch, so they would be expected to bring this to +-1.5 mm and line up with the dn=3mm red curve.
Posting some notes from recent email converstions looking into the large apparent shifts in dn (top mass blade tip height) and d2 in the all metal build (PUM wire loop prism).
Attachement PUMCOMDetails.pdf is from Eddie Sanchez and is a drawing showing that the position of the PUM wire loop break off in the all metal build is basically the same as where it should be in the final fiber build. However, the model fitting suggests the actual break off is about 1.8 mm lower. So Betsy took some photos of this prism on a suspended metal quad. See image files 1445.jpg to 1447.jpg. Since the prism is round, it could be the wire does not have a clean break off. The pictures seem to indicate the wire has a significant length of a line contact. The 1.8 mm shift could be within this line contact.
The last image, 1449.jpg, shows a picture of the top mass blade spring tip in a suspended top mass. The spring looks pretty well centered, not consistent at all with +3 mm of apparent shift in dn for this quad. Quoting some numbers from Betsy:
"The top surface of the blade, as close to the tip as possible, is supposed to be at 9.6mm down from the top of the bridge notch. The notch is 14.6mm wide, the blade is 5mm wide, therefore the bottom of the blade should line up with the bottom notch. No gauge blocks needed. From the picture, this looks very close to lining up."
switched to hi gain ~0730pdt Tuesday.