Related: PUM P2P and Y2Y inversion filters:
https://alog.ligo-wa.caltech.edu/aLOG/index.php?callRep=10610
I injected into PUM drivealign P2P, through P2P inversion filter, and measured the TF from injection to the OL P (red thick) and Y (red thin dashed). Did the same thing to Y2Y to OL Y (blue thick) and P (blue thin dashed).
In a frequency band where I managed to get a high coherence (0.2Hz to 4Hz) P2P and Y2Y look pretty good. Also P2Y is good. However, Y2P (blue thin dashed) looks problematic between 1 and 2Hz (and maybe 0.4-0.6Hz).
I'll make a Y2P filter that eliminate this.
Apollo (Mark, Scotty), Greg Grabeel Taking advantage of the window in between the work on HAM 4 and HAM 5, Apollo installed the HWS viewports. Because of the hole pattern on the flange they are not able to be pointing directly down in regards to the scribe on the face, but they are fairly close. Hopefully close enough that the slotting on the light pipe will make up any difference.
Last night, we finally obtained successful transfer function data on both the main and reaction chains of the ETMy. As expected, the main chain looked clean (we'd taken sneak peeks prior). However, the reaction chain looked bad. So, we went out and found a tough-to-see earthquake stop touching the reaction chain top mass. This cleaned up the TFs. However, alleviating the stop made the reaction chain pitch change. Realigning the chain made the lower stage sensors fall out of range of course. After fixing all of these, we discovered the TF's had again gone to h*ll. I scrutinized the EQ stops, all flags, and all other potential spots for mechanical interference. Finding nothing in error, I gave up for lunch. TFs of the main chain are still clean so the interference must not be between the chains but to the structure, BOSEMs, or EQ stops. Will go hunting again after lunch.
We'll need to abort the spool replacement effort scheduled for this afternoon since we will continue to need the IAS equipment set up there until this QUAD satisfies both alignment and noise parameters. As usual, it is difficult to get a QUAD to satisfy both at the same time. I expect another round of TFs to be taken tonight since convincing ones self that the TFs look good is particularly difficult during low-coherence daytime measurements.
I committed an update to ^/trunk/Common/MatlabTools/SingleModel_Production that optionally corrects (for the single model only at this point) a long-standing known error with the B matrix in the Matlab state space for all models (single, double, triple, quad), whereby the coupling from structure pitch to optic longitudinal was zero.
The trouble is that the SS matrix elements have been derived for infinitely flexible wire. For realistic, stiff wire, the wire flexure correction would ideally be implemented with code like the following
where dblade, dpitch, dyaw1 and dyaw2 are the vertical and horizontal offsets to the wire attachment points, flex0 is the vertical component of the flexure length, and si0 and c0 are sine and cosine of the angle of the wire to the vertical. This amounts to insetting the ends of the wire by one flexure length at each end. However for historical reasons there is no dblade or analogous quantity defined in any of the Matlab models for the very top wire attachment, so that line has had to be left out. The A matrix turns out completely OK because it only depends on the shortening of the wire - the pendulum has effectively been hitched up by flex0 because dblade couldn't be increased, but it retains all the same pendulum resonances. However it has the effect of zeroing out the structure pitch to optic longitudinal coupling in the B matrix, because even with dblade=0, there's supposed to be a lever arm of flex0.
The impetus to fix this is that Fabrice has been doing a series of experiments with a seismometer mounted as an pendulum with the same structure as for HAUX/HTTS (two blades, two wires), and is seeing a coupling from pitch of the structure. Therefore it is of interest to have a debugged model with the proper pitch-to-longitudinal coupling for comparison purposes.
To implement the fix, I created a new Mathematica model with an explicit dblade parameter, ^/trunk/Common/MathematicaModels/TwoWireSimpleBladesED (ED=extra "d"), updated the Matlab-export code to match, and exported a new set of Matlab matrix elements symbexport1bladesEDfull.m. I then hacked ssmake1MBf.m to use these elements when pend.dblade is defined. The usual use case will be pend.dblade=0, but other values also work. The attached plot is generated by edplot.m and shows the P to L transfer function for the default case of the TwoWireSimpleBladesED model which has dblade=0.001 and flex0=0.0009687. The plot generated from the Matlab model is in blue, and is exactly overlain by comparison data exported from the equivalent Mathematica in red.
As might be hoped, the value at f=0 is -(dblade+flex0)= -0.0019687. The sign is negative because +pitch is right-handed around +y=left, i.e., nose down, so as pitch increases the effective flexure point moves backwards. See the attached diagram.
For now I'll manually post these but soon these will be posted by a robot.
model restarts logged for Tue 11/Mar/2014
2014_03_11 09:57 h1ioppemmy
2014_03_11 10:05 h1ioppemmy
2014_03_11 10:06 h1ioppemmy
2014_03_11 10:14 h1ioppemmy
2014_03_11 10:15 h1ioppemmy
2014_03_11 10:16 h1ioppemmy
2014_03_11 10:46 h1asc
2014_03_11 11:12 h1lsc
2014_03_11 13:25 h1ioppemmy
2014_03_11 13:26 h1ioppemmy
2014_03_11 13:27 h1ioppemmy
2014_03_11 13:27 h1pemmy
2014_03_11 21:37 h1lsc
2014_03_11 22:02 h1lsc
2014_03_11 22:06 h1lsc
2014_03_11 11:40 DAQ Restart
They weren't empty but they were wrong. The IPS raw signals only became useful Monday after we finished the final alignments after the Actuator connection. So if you care (likely not) about the ETM HEPI position before now (0955pdt) you'll have to look at the local coordinates. Now you can look at the cartesian values for positions. Remember, I zero'd (<50 counts, 655cts.0.001" or 38.8nm/ct) Monday. The raw IPS are all running under 300cts now (HEPI is still unlocked) but the ACB weight decreased slightly yesterday and I'm not surprised to see a little drift as well. The cartesian are all running under 10um or urad and most much less. The nominal position for control will be zero.
Yesterday we tacked both doors on HAM4 with four bolts. We did not complete a chamber closeout checklist: we are not going to pump down on this volume until after we revisit alignment of the SR2 there, i.e. we cannot pump down until going back in and completing the closeout.
I started to try to commision the slow feedback to ETMX again. (We previously had this working, but had never charachterized it, and it was causing pitch problems. I would like to have a feedback to both HEPI and the top mass, and Arnaud has been working on L2P decoupling.)
It seems like the L2P filter right now is unstable, the optic is pitched around alot everytime I try to engage it. Currently I have a X arm gain of -0.8, the notch in MO_LOCK_L, and no boost. The feedback is stable, but the UGF is low. The guardian is set up to set these settings, and running tonight.
From the transfer function measurements I've made (none of which have verry good coherence) it seems like we should be able to increase the gain, the problem is that we start to excite pitch too much and loose lock.
If opwsws4 has not crashed by morning, would someone from the red team hit save on my dtt session, please? Thank you.
Your data is saved and ETMX is misaligned now.
Although Adam warned me about making sure we fixed all the connections in the LSC model, I miscommunicated with Dave and we installed it with out fixing the connections. Daniel and I now fixed the connections, restarted the model, and commited it to svn.
The Y2Y filters for ETMX ITMX and ITMY were installed today in the top mass drivealign matrix. The ITMX filter has been copied from ITMY and the design method is the same as in alog 10610.
The open loop (top mass yaw drive to test mass yaw through the invY2Y filter) (should) give a single pendulum transfer function with a resonant Q of 3 @ 0.6Hz, for the 3 quads.
The design scripts are living in :
/ligo/svncommon/SusSVN/sus/trunk/QUAD/H1/ITMY/SAGM0/Scripts/itmy_Y2Y_inversion.m
/ligo/svncommon/SusSVN/sus/trunk/QUAD/H1/ETMX/SAGM0/Scripts/etmx_Y2Y_inversion.m
And the measured data is from the dtt templates :
/ligo/svncommon/SusSVN/sus/trunk/QUAD/H1/ITMY/SAGM0/Data/2014-03-03_H1SUSITMY_M0_YtoPY_WhiteNoise_0p1to10Hz.xml
/ligo/svncommon/SusSVN/sus/trunk/QUAD/H1/ETMX/SAGM0/Data/2014-03-10_H1SUSETMX_M0_YtoPY_WhiteNoise_0p1to10Hz.xml
Alexa, Sheila
We have been unable to lock ALS COMM for more than about 5 minutes tonight, I'm not sure if this is because of high seismic, or for some other reason. (The microseism is around (or just below) the 90th percentile, and we have convoys of trucks) We have made a few measurements of spectra, not with the WFS engaged. The attached screen shot shows two spectra, the red trace was taken with a gain of 8dB on in 1 of the common mode board, the green trace with 19 dB. We wanted to reproduce that the high gain loop really has lower noise, but keep dropping lock.
The botom panel shows the coherence with the ISCT1 accelerometer that Robert moved onto the periscope today.
J. Kissel, A. Pele, A. Staley All HSTS and HSTS at both sites have incorrect M2 and M3 coil/magnet COILOUTF gain signs. While once again banging our heads against the IMC crossover measurements against Alexa's discrepant ALS COMM model, we had one more epiphany regarding the SUS -- RESPECT THE PHASE. Her lowest frequency cross-over, the M1-M2 crossover, which should have a phase of -180 [deg], was measured to have a phase of 0 [deg], but otherwise the magnitude matches exquisitely. Then I remembered her mentioning a month ago that LLO had to account for an unknown minus sign on there lower stages of MC2 to get models to match measurements. Having just recently found an AOSEM vs. BOSEM COILOUTF gain sign bugs on the QUADs during the coil balancing, it occurred to me to check that the M2 and M3 stages of the MC2 matched the PUM stage of a QUAD. They did not, and are indeed incorrect. BIFF! Worried that it was a systematic flaw copied and pasted everywhere (and because green team need mode cleaner, but not PRM), I checked PRM. Also incorrect. To confirm that that sign orientation *is* in correct, we drove a DC offset on the lower stages of PRM, and indeed, a +L requested drive moves the suspension in -L (as measured by the OSEM sensors). I asked Arnaud to quickly plot the phase of the acceptance measurements that he had for one of these suspensions. ZLONK! +180 [deg] measured phase at DC, where we should expect (from any suspension on the planet) 0 [deg] phase at DC. Sad about this, I then modified the acceptance measurement scripts to plot the phase of a random smattering all of the M2 and M3 acceptance measurements that have been taken of all HSTS at both sites. All have a phase of +180 [deg] -- see attached. POWIE! The reason for the flaw in magnet compensating gains? AOSEMs and BOSEMs have coils wound in exactly opposite direction. So a positive current sent to a BOSEM coil will result in magnetic fields opposite in direction than the AOSEM coil. We realized this halfway through defining all of our suspensions' lower stage magnet signs, as is reflected by the poor quality of information in our sign convention documentation, T1200015, specifically the Cheat Sheet, where it shows a table of COILOUTF gains for "Any 4 OSEM Stage" as UL - LL + UR + LR - which is the current sign distribution on the M2 and M3 stages for HSTSs. However, this table was created / determined assuming a BOSEM winding. You'll notice that the difference between -v2 and -v3 is "adding a AOSEM vs. BOSEM push/pull table," and the (unfinished!) statement that "AOSEM and BOSEM coils are wound exactly opposite in direct[[ion]]." CLANK-EST! Naturally, because the HLTS EPICs values began as a copy-and-paste of the HSTS, any HLTS has this same flaw of its lower stages, but we haven't used them enough to notice. Conclusions: - All HSTS and HSTS at both sites have incorrect M2 and M3 coil/magnet COILOUTF gain signs. That's MC1, MC2, MC3, PRM, PR2, PR3, SR3, SR2, and SRM. There signs should be: UL + LL - UR - LR + - I blame myself. - I clearly need to create a version 4 of T1200015. - Of course this affects all IFO loops that use the HSTS, so I have not and will not change the signs unless I have whomever designed the ISC loops sitting next to me.
T1200015 has been updated to version 4 with explicit calls to the differences between an AOSEM and BOSEM, 4-OSEM stage.
- Hanford Fire Maintenance fixing hydrant 1 by the woodshop
- Jim and Cyrus at MY working on PEM card.
- Dave B. two DAQ reboots today.
- Jeff B. and Andres R. at HAM4
- Robert S. installing accelerometers on ISCT1L
- Apollo installing both doors on HAM4 with 4 bolts
- CDS implementing LSC and ASC model changes
- SUS doing transfer functions at EY
- Patrick T. doing a CONLOG update
- Karen at EY to clean
Until the guardian installation, I've been asked to write a quick procedure on how to turn back the BSC-ISIs on after a trip.
Right now, the best configuration for BS, ITMX, ITMY and ETMX is the same: lvl3 controller on ST1 and ST2, Tcrappy blend filters on ST1 and ST2. No sensor correction.
- Reset the watchdogs
- Check that HEPI position loops are still on
- Open the ST1 CART BIAS screen and click on Reset CPS offset then Store target offsets
- Open the commands screen and click on !Isolate <CHAMBER> lvl3
- A black xterm window will show up. While this pop-up window is there, don't touch anything.
- The isolation loops are now on. Next step is to switch to the good blend filters. Wait for the T240 signals to go down! The overview T240 indicator has to be ALL GREEN before doing anything.
- When you think the platform has settled down, open ST1 BLEND screen -> SWITCH ALL and select Tcrappy.
- If this process makes the ISI trips, that means you haven't waited long enough. Restart the all process. A trick that could work is to start switching the vertical blend filters (Z, RX and RY) first, then the horizontal ones (X, Y, RZ).
- Repeat the same process with ST2 BLEND.
Remember that we are dealing with seismic signal and very low frequency instruments. Be patient!
Andres & Jeff Using a 3 axis translation stage on an optics post (see photos) we glued a new magnet/dumbbell assembly to the UR side of the SR3 optic, to replace the one that was knocked off on Monday. The process is straight forward and was completed in a couple of hours. Procedure: 1) Lock the Intermediate Mass in its level hanging position; then lock the Optic in its level hanging position. 2) Adjust the placement and X, Y, Z location of the 3 axis translation setup so the magnet flag holder is aligned on top of the ring of glue left from the original magnet/dumbbell assembly. 3) Lock the Z and Y axis stages of the 3 axis translation setup in place so they cannot shift. 4) Retract the X axis stage as far as it can go to pull back the magnet flag holder away from the Optic face. 5) Using Acetone and Alpha swabs carefully clean the old glue from the face of the Optic. It helps to hold an Acetone soaked swab against the glue spot for several minutes to soften the old EP-30. Then with a second swab gently rub the glue spot. If the glue does not start to come off hold the Acetone swab against the glue spot for a few more minutes. 6) Repeat the wetting and swabbing process until all the old EP-30 has been removed. 7) Mount a new magnet/dumbbell assembly into the flag holder and apply a dab of EP-30 to the dumbbell. 8) Slowly turn in the X axis stage until the dumbbell makes contact with the face of the Optic. 9) Lock the X axis stage so it cannot shift while the glue dries.
Jonathan Hanks, Jim Batch, David Barker Reconfigured the badger.ligo-wa.caltech.edu web server to no longer accept http access using a common login. Replaced with https LIGO.ORG authenticated service, which displays the Jenkins service running on x1boot. Existing users may notice cached links to http pages need to recache as https
Assuming the frequency calibration of the network analyzer is accurate, we can compare the measured PRC length with the measured mode cleaner length. This was measured in alog 9679.
Parameter | Value | Unit |
---|---|---|
FSRPRC | 2.600075 | MHz |
LPRC | 57.6508 | m |
FSRMC | 9.099173 | MHz |
LMC | 16.473612 | m |
FSRMC / 3.5 - FSRPRC | -306 | Hz |
(1 - FSRMC / 3.5 FSRPRC) LPRC | 6.8 | mm |
Compared with the modeclaner, the power recycling cavity is about 7 mm too short. The other way around, the modecleaner is about 2 mm too long.
We hooked up the network analyzer to the timing comparator/frequency counter and set it to 40 MHz sharp at 0 dBm. The readback value was dead on, occasionally we would read 1 Hz higher. Conclusion: the frequency of the sweep is no more than 1 Hz off, even at 100 MHz.
I've redone the fits using both the magnitude and the phase. The fitting function is now the usual Fabry–Pérot reflectance function, with a complex magnitude to allow for global amplitude rescaling and global phase offset.
Nominal FSR | Frequency (Hz) |
−39.5 | −102 701 040 ± 200 |
−39.5 | −102 700 900 ± 220 |
−26.5 | −68 900 600 ± 180 |
−12.5 | −32 500 080 ± 100 |
12.5 | 32 501 720 ± 110 |
26.5 | 68 903 940 ± 200 |
39.5 | 102 704 700 ± 200 |
The linear fit now gives an FSR of (2 600 073 ± 9) Hz. This is consistent with the previous fit, and anyway the total error is still dominated by some systematic, as seen by the fact that the residuals are excessively large.
Taking a systematic 400 Hz uncertainty on the residual for the 12.5 FSR measurement gives a systematic uncertainty of 32 Hz on the PRC FSR. Propagating foward gives (57.6508 ± 0.0007) m.