Last week, sts correction gains were measured and calculated on HAM4/5/6 to improve the CPS sensor correction. We measured them today for the remaning chambers.
For the BSC chambers, we locked the platforms to the ground by increasing the blends to 750mHz (cf alog) and turned off the sensor correction. We measured the ratio between the T240 (stage 1) and the STS (ground) signal (should be 1 with the platform moving with the ground). The same measurement was also carried out between STS and L4C for hepi, since IPS sensor correction is used for the Z DOF.
All the measurements using the corner station STS B as input to the sensor correction (ITMX, BS, ITMY) show a factor of two mismatch with the hepi and platform sensors in the X DOF only. There is certainly a calibration error with the STS B seismometer. I checked the calibration filter in the input filter banks and it looks correct (10.17 nm/s /cts). We should investigate more. The other chambers and dofs need a relatively small correction.
The ISI gains (T240/STS for BSCs and GS13/STS for HAMs) are
X | Y | Z | |
HAM2 | 1.036 | 0.995 | 0.877 |
HAM3 | 0.976 | 0.958 | 0.812 |
ETMX (T240 for gnd sensor) | 1.114 | 1.106 | 1.105 |
ETMY | 0.985 | 0.974 | 0.995 |
ITMX | 1.964 | 0.976 | 0.993 |
ITMY | 1.982 | 0.984 | 0.993 |
BS | 1.99 | 0.986 | 0.994 |
The BSC - HEPI gains (L4C/STS) are
X | Y | Z | |
ETMX | 1.110 | 1.107 | 1.091 |
ETMY | 0.998 | 0.991 | 0.969 |
ITMX | 2.167 | 1.063 | 0.972 |
ITMY | 2.011 | 1.006 | 0.963 |
BS | 2.09 | 1.055 | 0.924 |
Attached are examples of the transfer functions for ITMY
To calculate the gains I wrote a script for the bscs in a similar fashion as Sebastien's : it is called BSC_gain_matching_calculation(IFO,Chamber,start_time,duration) located under /ligo/svncommon/SeiSVN/seismic/BSC-ISI/Common/Misc. The results above were obtained using the following parameters :
start_time=tconvert('01/22/2015 15:35') ;
duration = 45*60 ;
The sensor correction matching gains were changed to their new values for the BS ITMX and ITMY in the X dof. The pdf attached are comparing the X platform motion before/after the change, showing improvement (x~2) in the 300mHz 700mHz band.
I confirmed that the factor of two mismatch described in the log above comes from the STS B, looking at the comparison of the three ground seismometers (X dof) of the corner station, see attached screenshot.
It looks like the problem is in the B STS or associated cabling. I took spectra of the raw ADC signal and the B seismometer shows half as much X signal (ADC channel 26 on attached plot) as the other 2 STS's. Arnaud and I went out to check the cabling at the rack, everything is tight there. We are doing measurements with the ITM's right now, so we didn't go out to check the cable on the pod.
I looked though the trends and science channels, and it looks like the auto center on the B STS was pushed at 8:30 am local on Dec 23rd, 2014 and the X channel didn't recover properly. On the first attached trend you can see that before the alleged button push (alleged because there are no logs, I'm just guessing, second trend shows what looks like an auto-centering sequence though) the X channel showed more signal, then after less signal. Y channel looks roughly the same before and after. Attached spectra (3rd png) shows overall spectra is very different, solid red is before, dashed green is after.
Yesterday I dif the ETMs. Today I did all the remainder. All came back under guardian first time.
Have added and committed HPI guardians to the h1 area so as to not by default edit the common file. All other guardian changes from this also committed.
To help with setting up SDF files, I have created lists of channels modified by the H1 guardian nodes for the month of January up to this time. We expect this to be a fairly exhaustive list as it covers the RCG2.9 upgrade on the 13th Jan when all front ends were restarted.
There is a text file per guardian node in the directory /opt/rtcds/lho/h1/data/guardian_channels/guardian_control_channels/2015jan
Attached is a list of the number of channels controlled per guardian node. As expected it is quite small. For example: SUS_ETMX has 1,717 set point channels of which 53 were changed (3%); ISI ETMX (ST1 and ST2) have 1,952 set point channels of which 124 were changed (6%).
Found that the angular control inputs of TMSY were off since Tuesday (restore issue?)
Moved the gain of the TMSY ISC P/Y input modules into a CAL filter. This makes Y the same as X.
The safe.snap was updated.
New tidal code was loaded for both arms which delays the tidal servo by 5 seconds after the arm has locked.
Also fixed the long standing problem that some of the values in the PLL/PHD autolockers were not restored automatically after a PLC restart.
J. Kissel While the BSC-ISIs were in an odd configuration anyway for obtaining sensor correction gain matching data (see LHO aLOG 16200), I took the opportunity to clean up, clean out, and fill in the ST1 and ST2 BLND_NXT filter banks with the correct, high performance, Ryan DeRosa, blend filters. Recall that the CUR (for "current") and NXT (for "next") filter banks should have the same exact filters in all filter modules, such that they can be used for the automatically switching between blend filters on the fly (see T1200126). All NXT banks for ETMX, ETMY, ITMX, ITMY, and BS now have identical filters as what's in place in the CUR banks (HAMs are already up-to-snuff). Further, I've cleared out all "Unknown", gain-of-one, filters which seemed to have been sporadically peppered in the higher numbered FMs. Finally, when transitioning back from the "Start" filters to Ryan's filters (by hand), I've left the RZ filter banks with all FMs OFF, because we don't run the RZ isolation loops. For the record, I performed the copy and paste by-hand in foton, saved filters via foton, and reloaded all coefficients simultaneously using the GDS_TP screen. Further, I've committed all affected filter files (and any other I found different from the SVN -- the HAM's .txt files because of Jim's recent improvement of the isolation loops [aLOG pending], and HAM4 and 6's .fir files, because Jim installed them Dec 2nd [aLOG "pending"]). Also -- I did not have enough time to *test* whether the newly implemented filters allow for smooth switching between the low-noise filter set and any others, but my suspicion is that it *still* won't work, given that the blend switching software cannot handle filter sets that occupy two FMs, which is the case for the RX and RY filters on ST1.
J. Kissel, A. Pele In order to gather measurement data for GND to ST1 sensor correction, I've switched all chamber's ISIs except for HAMs 4-6 to a high-frequency blend. For the HAMs this means the "750mHz" blends, and for the BSCs the "Start" filters (see attached) and turned both ISI X&Y and HEPI Z sensor correction OFF. Yesterday, worried that there still might be residual garbage from initial install in these filter banks, we've confirmed that all chambers involved have the same filters as attached. The chambers have been set up in this fashion since 1105975887 Jan 22 2015 15:31:11 UTC (We blends switched by done by 15:05 UTC and 15:22 UTC in the HAMs and BSCs, respectively, but I forgot to turn OFF the sensor correction until Arnaud reminded).
BSCs and HAMs have been restored as of: 1105980152 Jan 22 2015 16:42:16 UTC XARM Restored 1105981034 Jan 22 2015 16:56:58 UTC YARM / BS Restored 1105981219 Jan 22 2015 17:00:03 UTC HAM23 restored
model restarts logged for Wed 21/Jan/2015
2015_01_21 01:47 h1fw0
2015_01_21 13:36 h1fw1
2015_01_21 15:05 h1fw1
2015_01_21 15:29 h1fw1
2015_01_21 15:43 h1fw0
2015_01_21 17:26 h1fw1
2015_01_21 17:40 h1fw0
2015_01_21 18:43 h1fw0
2015_01_21 19:08 h1fw0
2015_01_21 19:40 h1fw0
2015_01_21 20:42 h1fw0
2015_01_21 23:03 h1fw1
2015_01_21 23:29 h1fw1
2015_01_21 23:45 h1fw0
All unexpected restarts. Inexplicably both frame writers have become unstable, starting about 24 hours after Tuesday's solaris reboots. I may try another solaris reboot today.
I've attached a trend of the h1fw0 and h1fw1 restarts for this month (h1dc0 restart is included to show intentional DAQ restarts). The instability starts 1/13 when we upgraded to RCG2.9.
Conlog frequently changing channels list attached.
Alexa, Evan, Sheila, Koji, Rana
We have been locking the arms tonight, with DRMI on 3F and trying to transition CARM to sqrt(TRX+TRY).
So far no success, although we found several setings which were wrong.
We tried moving the CARM offset (using the COMM VCO) to -200 Hz (green), and adding a small amount of gain, 1/10 of our nominal gain. Rana saw that this adds a lot of high frequency noise to CARM.
Sheila, Kiwamu, Rana, Evan, Daniel, Alexa
We had hope to explore the phase-space of DRMI and improve the lock acquistion time; however, we did not get very far.
We began with DRMI+arms off resonance under the nominal gaurdian configuration. We acquired the lock 5 times, and found the following acquisition times:
Lock time (Jan 22, 2015 UTC) | Acquisition Duration | |
1 | 00:43:02 | 8 min |
2 | 00:49:40 | 3 min |
3 | 00:52:53 | 3 min |
4 | 00:57:28 | 4.5 min |
5 | 01:19:30 | 7 min |
At this point we checked the demod phases for the 1f signal, and decided to adjust the REFL45 phase from 137 deg to 142 deg. We also changed the MICH upper trigger to 5.0 from 10.0. and touched up the alignment (this probably ruins the validaty of our test). With this configuration, we repeated the above and found:
Lock time (Jan 22, 2015 UTC) | Acquisition Duration | |
1 | 02:00:00 | 1.5 min |
2 | 02:00:57 | 0.5 min |
3 | 02:04:00 | 2 min |
4 | 02:05:30 | 10 sec |
5 | 02:44:29 | 10 min* |
* flipped CARM offset at one point.
Overall this test was pretty inconclusive, but I thought I would post the lock aquisition times. This evening, we have been running with the MICH upper trigger threshold at 5.0 and it takes about 15 min or so to lock again ...
A few filter changes for the MICH / BS tonight:
My guess from the noise that we hear in the speakers is that the BS coil driver DAC channels has been lightly saturating whenever we reduce the CARM offset. Its pretty close to the rails before we start and the signals are only noisier when we're on the side of the CARM fringe.
The first attachment compares the ELP40 with the RLP65.
The second attachment compares the M2 LOCK signal with 1f lock (PURPLE), 3f lock (BROWN), and 3f lock with the new filter (BLUE). The new filter gives us a 3-4x reduction in the RMS. I think this should give us a more smooth 3f lock and a smoother CARM reduction.
J. Kissel, A. Pele While Seb suggests that moving the HAM3 blends to the "250mHz" configuration is a good configuration to "solve" the 0.6 [Hz] feature (see LHO aLOG 16100), we argue that it would be better to live with a 0.6 [Hz], digitally-sharp, peak than to increase the X motion by the factor of several. OR we should find a different set of blend filters for RY that *don't* increase the low frequency motion. I attach 2 supporting documents: (1) 16196_20150121183131_H1-SEI_QUIET_3C14C7_SPECTRUM-1105747216-86400.png The performance comparison between all HAM platforms from the Jan 20th summary pages. On this day, HAMs 2, 4, 5,and 6 are in the "nominal" LHO configuration, all using "01_28" blends on all degrees of freedom, with sensor correction ON for all DOFs, GS13s in high-gain mode, with newly improved isolation loops. HAM3 however, was in Seb's suggested configuration -- the same as the other HAMs *except* for the RY blend filter, which has been changed from "01_28" to "250mHz." (2) 2015-01-21_H1vL1_BlendFilter_Comparison.pdf A comparison between the H1 "01_28," H1 "250 mHz", and L1 "400" RY blend filters, as well as the H1 "01_28" and L1 "250" X blends. This shows that (a) In RY, (i) the H1 "01_28" and L1 "400" are virtually identical, as expected. To be precise, for some reason, the L1 GS13 filters are 8% higher in gain, but otherwise identical. (ii) The H1 "250mHz" blend is actually *lower* in blend frequency than the H1 "01_28" filters. This means the low-frequency roll-off of the GS13 high pass is much slower for the "250mHz" than for the "01_28". (b) In X, (i) there is a good bit of difference in the 0.2 - 4 [Hz] region. We suspect this is because the H1 "01_28" filters were copied over from LLO in Oct 2013, before Ryan had made further tweaks to these blend filters to improve performance around the SUS resonances. The performance is most different at 0.75 [Hz], where the displacement sensor low-pass is lower by a factor of 10. (ii) it's concerning that the GS13 high-pass -- though 19% different in overall gain -- does *not* differ between the two site's versions of the filter. This must be what Ryan means when he suggests his filters are "almost" complementary. While all points are interesting with respect to the sociology and history of the SEI commissioning, point (2)(a)(ii) is the key for HAM3. Because the RY GS13 filter rolls off slower and lower for the "250mHz" blends, the differential vertical GS13 noise is reinjected into the RY loop. This excess residual RY motion couples to X via tilt-horizontal coupling, which degrades the low-frequency noise performance in the same frequency region. Bad. We'd discovered this nasty cross-coupling path as far back as The Stone Ages. We should find/design an RY filter for HAM3 that *increases* the blend frequency from the H1 "01_28" filters, if we intend to run for a while in such an odd configuration. We may try copying over a few other L1 blend filters. This is all still lower priority compared to the *rest* of the to-do list, but Arnaud and I are worried that the excess low-frequency motion in HAM3 alone might be causing problems with cavity motion.
8:51 Kiwamu to IOT2L
9:05 Bubba to EX
9:06 Karen to EY
9:08 Corey to Squeezer Bay working on 3IFO ISC
9:15 Cris to MX
9:37 Corey to EY looking at ALS optics on table
9:38 Andres working on 3IFO storage cleanup in LVEA
9:44 Kiwamu done in LVEA
9:53 Gerardo X beam tube ion pump work
10:11 Kiwamu and Elli to ISCT6
10:11 Corey back from EY
10:24 Karen done at EY
10:27 Bubba done at EX
10:43 Corey back to Squeezer Bay
10:47 Andres out of LVEA
11:02 Jeff and Andres to LVEA for more cleanup
11:10 Rick and Sudarshan to LVEA PSL area
11:20 Jodi to LVEA taking pics
11:22 Jeff and Andres out of LVEA
11:32 Jodi out of LVEA
12:02 Gerardo out of LVEA
12:22 Corey out of Squeezer Bay/LVEA
14:02 Corey back to Squeezer Bay for 20 min.
(Kyle R, Gerardo M)
The new ion pump became railed sometime after decoupling the pump cart yesterday, and remained railed overnight.
This morning after pumping down the annulus system with the pump cart, we determined that the main valve was leaking.
So, to minimize exposure of the annulus system to atmosphere, we decided to add another valve in series with the leaky valve.
After pumping down on the system for about 25 minutes, the Ion pump started operating as it should, and is currently at 1.5 amps.
or perhaps 1.5 mA
Just like ETMX earlier, restarted the HEPI guardian at ETMY to restore the same position for all dofs. No problems again. Attached is similar plot showing how different the alignment is after the restart. Again, the HEPI is exactly the same, cause we made it so. And the ISI, even though it restores no dofs, holds to its previous position to within 1 urad ( worst is RX=Pitch at ETMY,) all other dofs much less.
As a part of the DRMI investigation, we checked the laser power of CO2X and found that the power had dropped by a factor of two or so a week ago. However it turned out that this was due to a calibration filter (FM2 of TCS-ITMX_CO2_LSRPWR_MTR) which had turned off at that time for some reason. We switched it back on and also edited safe.snap accordingly. The attached is a 100days trend of relevant channels. As shown, the input of the monitor channel stayed at a constant value approximately in the past month or so while the output suddenly changed a week ago, indicating that the calibration filter was taken out.
Math puzzle: What's wrong with this UGF Servo?
its installed for use in the OMC and LSC at the moment, and could be used in the ASC if we find we want to hold the UGFs constant during TCS tuning.
We want to measure . The algebra tells us that
,
.
When the excitation is sufficiently large that the noise is negligible, we get .
and
are complex quantities, and the real and imaginary parts are the I and Q outputs of the demods, or
,
.
In the version of the UGF servo that Rana posted, the phases would be chosen such that the imaginary (Q) outputs are zero, thus
.
This is OK as long as G only ever changes in magnitude, otherwise .
To get the correct measurement in the case of a changing phase, one must do the following:
This is implemented in the attached simulink model.
(First of all I don't know the answer yet)
I believe G only changes the gain most of the case.
The problem was
a/e = G/(1-G) and b/e = 1/(1-G) change their phase as you change the gain of G.
We usually don't care the phase of G, but only care the magnitude of G as the phase of G is fixed.
Therefore what we need is to take the ratio of the magnitude of a and b
|a| = |G|/|1-G|, |b| = 1/|1-G|
|a|/|b| = |G|, where |a| = sqrt(aI2+aQ2) and |b| = sqrt(bI2+bQ2)
(Ed: I'm suggesting to take SQRT(I^2+Q^2) to eliminate the frequently-omitted-effort of adjusting the demod phase correctly.)
Meh, I say it's overkill. As Nic mentions, this works just fine if the UGF phase is not changing, so long as you set the demod phase correctly. As Koji mentions, the UGF phase should not be expected to change that much. It is already a second-order effect, so what is there now should be fine, unless we really want to accommodate wild loop phase fluctuations at the UGF. Is there a reason that's ever something we want?