I reset the dust monitor counts alarm levels to reflect the US Fed 209E cleanroom standards, as defined in T1400024. For Clean-1000 (General room monitoring of LVEA and VEAs) 0.3 minor 2000, major 3000 0.5 minor 750, major 1000 For Clean-100 (Open chambers, cleanrooms with Class-A, etc.) 0.3 minor 250, major 300 0.5 minor 75, major 100 The dust monitor alarm have been set to these levels. I will monitor over the next couple of weeks to see if these levels need to be adjusted.
In general, the 3rd ifo work can go ahead all day Tuesday and Thursday; all other days should be off completely.
Today: Concentrating on initial alignment
Friday: Schnupp length measurement (morning); Locking and alignment (afternoon)
Saturday: Schnupp length measurement
Monday: Locking and alignment
Tuesday: LSC/OMC model split; 3rd ifo work (all day)
Wednesday: Locking and alignment
Thursday: SRC length measurement; Locking and alignment; 3rd ifo work (all day)
Friday: Locking and robustness
This morning at ~0025utc 19 Feb, the End Station HEPI Pump Servos were restarted with the Scan Rate reduced from 0.1 second to 1 second. Also, the database has JeffK's 10mHz PID parameters--see 16782. Also, EndY had the SMOO parameter but that has never been in the database, so any restart requires those to be replaced manually. The smoothing has NOT been reloaded.
See the attached as I going to say that while things are better at both stations, EndY remains noisier than EndX and this is evident in amplitude of the Pressure data and the control signal going to the VFD. The 24 hour plot attached clearly shows the time things were switched 23 hours ago. But whereas Ex Pressure got quieter, EndY's got noisier, likely from the Smoothing going away. Because the pressure signal is noisier at EndY, the drive to the VFD is larger than it is at EndX. I predict that if there is measurable coherence with platform sensors, it will be larger at EndY.
The Pcal calibration line for the LHO END Y is set at 36.7 Hz (40 cts excitation amplitude) and 540.7 Hz (23500 cts excitation amplitude) both producing an SNR of about 20. These frequencies are in line with the agreed upon frequencies as reported in LLO alog #15870.
This morning, Jamie enabled the Guardian SPM on all suspensions. There will be a new yellow bar across the bottom of the GRD screens which may point to settings that have been switched and are discrepant from what the current guardian state says they should be. The system is a work in progress so for now you can ignore this indicator while we learn what comes up in the monitor. To disable the monitoring, set ca_monitor = FALSE in the GRD of choice. Note, while we were working on this the ETMy guardian node completely froze and Jamie had to completely restart it.
While taking acceptance review TFs over the last 2 days, I noticed that the SR2 Pitch TF does not look healthy when the IFO alignment bias is enabled. (Note, acceptance TFs are usually taken with all bias offsets disabled.) Likely the large pitch bias plus the TF excitation is causing an earthquake stop to brush the suspension somewhere. The first attachment is the Pitch TF taken with the full IFO requested bias enabled (P=2789, Y=759), the second attachment is with a lower Pitch bias but the same Yaw bias (P=2589, Y=759). This could be be headache for commissioners. I mapped out that for reduced Pitch bias between ~1500 and ~2600 the TF looks healthy. I did not map out exactly where between ~2600-2789 it goes bad. Indeed, the TF at the aligned state looks to have both Pitch and Yaw motion when viewed on the AS Air camera. The bad TF seems to happen when the beam is somewhere off the top of the camera view. Probably too close for comfort.
On the plus side, I checked the the SRM Pitch TF which is healthy at it's nominal IFO alignment setting. The PR2 has a pretty high Yaw bias enabled (Y = 4311) so should probably be checked. The rest (MCs, PR3, PRM, and SR3) have smaller bias offsets under 2000 so I wouldn't suspect a problem with them.
Kiwamu and I checked that there is no DAQ saturation happening during the SR2 TF. Darn.
model restarts logged for Wed 18/Feb/2015
2015_02_18 14:27 h1fw1
2015_02_18 16:17 h1broadcast0
2015_02_18 20:22 h1fw1
two unexpected restarts. DAQ broadcaster restart to add channels for DMT. Conlog frequently changing channels report attached.
Alexa, Evan, Dan
With the IFO locked on DC readout at 2.8W we measured a DARM OLTF for Kiwamu's calibration. The ESD linearization was OFF. The template can be found under: /ligo/home/alexan.staley/Public/DARM/. I have attached the txt file of the TF.
We locked on DC readout at 7:43 UTC on Feb 18, 2015; however I was running a TF the entire time and caused the lockloss. So this isn't worth anyone looking into locklosses or coherences. Tonight we were also able to increase the input power to 7.8W without any trouble while on RF DARM. We did NOT transition to DC readout at this higher power because we noticed that the 13.8 Hz roll mode was rather large; when we tried damping this we ended up blowing the lock. More damping to be done...
J. Kissel, D. Hoak We needed some more damping of the highest vertical and roll modes of the QUADs this evening using DARM as the sensor and the Top Mass (M0) as the actuator. We re-used the same parameters for ETMY vertical at 9.7305 [Hz] (see LHO aLOG 16673). We also tried commissioning some ITMX and ETMX vertical damping toying around with similar filters and gains but were ... inconsistently unsuccessful: I explored the gain and phase parameter space but was either unable to convincingly change the amplitude of the mode, or it broke the lock. We did however pioneer roll damping at 13.81 [Hz] on ETMY. It took us a while to find a sensor (besides DARM) that could help us identify which test mass was rolling -- but L2-stage OSEMs saw ETMY's roll mode loud and clear. Dan designed the filters to have similar affect as the vertical, 9.7 [Hz] filters with a relatively wide band pass surrounding 13.9 [Hz] in FM4, and two trial filters that add or subtract 60 degrees of phase at the 13.9 [Hz] in FM1 and FM2 respectively: FM1 "+60deg" zpk([0],[3.26459+i*18.5144;3.26459-i*18.5144],1,"n")gain(0.0523988) FM2 "-60deg" zpk([0],[1.75001+i*10.3531;1.75001-i*10.3531],1,"n")gain(0.0911865) FM4 "bp13.9" butter("BandPass", 4, 13.3, 14.5)gain(120, "dB") These have been loaded into every QUAD's H1:SUS-?TM?_M0_DARM_DAMP_R bank, but only ETMY's R banks have been commissioned to the point that damping is successful and repeatable (which is the same for V as well). We found that, for ETMY, the +60 deg filter (FM1), (and FM4, obviously), plus a negative gain (we can get as high as -200 or -300), reduces the amplitude of the mode within a minute or two while in the "DARM WFS" ISC_LOCK state, with CARM and TR CARM, DARM is on AS AIR. Also, we could only get any action with the TOP MASS coil drivers in State 1 (LP OFF, or in its highest range mode). ---------- Directions to commission further test masses' loops: - Check that the TOP mass coil driver is in high-range mode -- preferably *before* you're locked, so you can switch it if you're not (I broke the AS AIR, RF DARM lock when transitioning ETMY's M0 drivers from State 2 to State 1). - Try to identify *which* test mass' mode has rung up. Vertical modes appear in M0 OSEMs if rung up particularly bad, and so far I've been able to see roll modes in L2 OSEMs. - Start with only the band pass filter (FM4) on, and gradually increase the gain (with a 5 or 10 [sec] ramp) in the positive direction until you either get close to saturating the DAC, or you start seeing amplitude change on the mode in question. Start with a gain of 1.0, and increase by factors of 2 to 5, depending on how above you are from the normal LF and RT output signals, how close you are to DAC saturation, and how bold you feel today. - If you don't see any change, rotate the phase by 60 degs (i.e. turn on FM1), and increase the gain. Rinse and repeat until you see some action (+120 [deg] = negative gain, and FM2 on, +180 [deg] is negative gain, with no FM1 or FM2 on, etc. etc.) - If you see an amplitude increase, then flip the sign of your gain. - If you see an amplitude decrease, and its slow, try rotating an additional 60 [deg]. - If you see an amplitude decrease, and it's fast, then you've won! Notes: - The definition of "slow" vs. "fast" is that "fast" is that the amplitude noticeably decreases between each average of a 0.1 [Hz] resolution, 70% overlap, 3 exponential overlap, log-scale ASD. "Slow" means you can see it decreasing, but you don't have the patience to wiat for another 30 minutes while it decreases enough that you're happy. - We've had only a few data points, but we've found that ETMY vertical damping signs between using ALS DIFF for DARM, and any red on DARM. This may just have been that particularly bad day though.
Alexa, Keita, Sheila,Elli
This morning we tried to improve the co-alingment of the green and red beams in the Y arm. We ended up adding some offsets to the Y transmon green QPDs, screen shot attached. Elli took some photos using the pcal camera to make estimates of how centered we are on the ETMs.
FIrst we tried to recover the alignment that we had last week durring the 8 Watt lock. Keita noticed that the Y arm QPDs had moved significantly since that time. We tried alingning the both TMSs using first the BOSEMs, then the baffle PD pointing. With the WFS running to align the red to the X arm, we tried to move the test masses to restore the old positions on the QPDs. This resulted in a very bad green alingment, we could have tried to recover this using QPD offsets and transmon pointing, but this didn't make a lot of sense so we didn't do it. Instead we just tried to make sure the two beams were both aligned in the Y arm when the input pointing was alinged to the X arm, and MICH was well aligned. We improved the situation compared to last night, but the co-alignment is still not great.
As mentioned in the previous alog (alog 16780), I suspected that the linearization in the ESD may not be behaving as expected.
We have been assuming that the ESD linearization does not change the actuator gain from the non-linearized case, but this turned out to be wrong according to a measurement (and analytical calculation).
It seems that either fixing this behavior in the actual system or correctly incorporating the linerization in the suspension model would reduce the discrepancy between the model and measured ESD response to a level of several 10 %.
On the other hand, I got another mystery where I am unable to explain the amount of change between the linearized and nonlinearized cases. The work continues.
(A single frequency measurement)
I did a quick measurement in this morning, comparing the linearized and non-linearized cases on ETMY (which was the only available ETM in this morning with the green light locked). I drove ETMY ESD at 2.261 Hz with an amplitude of 300000 cnts at L3_LOCK_L_EXC. The amplitude is set close to saturation at the DAC in order to get highest signal-to-noise ratio. The bias voltage was set to 9.5 V. There were no filters between the excitation point and the DAC. The L2P and L2Y were disabled. The oplev damping was active at the L2 stage only in pitch. The force coefficient was set to -180000 cnts in order to be identical to ETMX.
I used the green PDH control signal to monitor the displacment on the test mass. Also, since I used the green PDH signal instead of the IR locking signal this time, I was not able to get great SNR at 13 Hz which is the ferquency I have been using.
(Results)
The attachment shown below is the result.
As I switched the linerization on, the peak height at the excitation frequency increased by a factor of roughly 1.65. In addition, the 2nd harmonic peak decreased as the linearization is supposed to eliminate the nonlinear terms. The residual in the 2nd harmonic could be due to some charge on the test mass.
If this is true in ETMX as well, this will reduce the discrepancy between the suspension model and measurement down to 25% (which was previously reported to be discrepancy of a factor of 2.07 in alog 16780).
(Analytical calculation does not match the measurement)
On the other hand, according to my math (see the second attachment), the linerization was expected to increases the response only by a factor of 1.45 instead of 1.65. Although the difference between the two is only 13%, this still makes me think that something is not right as this calculation is relatively straightforward. Just for reference, I also attach the liearization simulink model that we use on ETMs.
If my math is correct, in order for the ESD to have the same actuator response gain, the absolute value of the force coefficient has to be the same as the bias counts at the DAC. Since we use bias voltage of 9.5 V (or 125418.4 cnts), the force coefficient should be -124518.4 cnts as well, but it was set to -1800000 cnts in reality. Since the response gain is proportional to the force coefficient, this should give us an extra increment of 1800000/124518.4 = 1.45 in the gain. I have no idea why the measurement differs from the expected.
I will repeat the measurement at some point soon when I get a chance.
The charge on the test mass is probably affecting your calibration. It would be useful to measure the charge on all 4 of the test masses using the optical levers and excitation of the quadrants independently. Look at Stuart Aston's and/or Borja's log entries at LLO and LHO. Also, the charge will be variable as long as the ion pumps are open to the vacuum system.
PR3 oplev died at about Feb 17 2015 20:00:00 UTC, that's about noon local time.
The commissioners had to stop using PR3 oplev for some ASC purpose yesterday evening.
I thought I had powered-on all Op Lever Whitening Chassis after my work on Tuesday. I went out this morning and powered unit back on.
H1BROADCAST0.ini has 3 additional channels in r9856. h1broadcast0 was restarted, its channel count increased from 686 to 689.
Additional channels are:
+[H1:LSC-PD_DOF_MTRX_1_1]
+[H1:LSC-TR_X_QPD_B_SUM_OUTPUT]
+[H1:LSC-TR_Y_QPD_B_SUM_OUTPUT]
LVEA: Laser Hazard Observation Bit: Commissioning 08:21 Betsy – Running TFs on ETMX and SR2 08:23 Corey – 3IFO work in H2 Squeezer bay 08:25 Filiberto – Pulling cables at HAM6 08:49 Mitch – 3IFO work in West Bay 08:55 Cris – Cleaning at Mid-X 08:55 Karen – Cleaning at End-Y 09:03 Mitch – Out of LVEA 09:15 Travis – Going into the LVEA looking for tooling 09:22 Doug – Going to optics cabinet near H1-PSL enclosure 09:35 Travis – Out of the LVEA 09:39 Corey – Going to End-Y 09:41 Apollo – Starting up A/C units for DCS 09:55 Richard – Swap IR Spool (X arm) camera zoom lens 10:20 Jodi & Travis – Going to Mid-X and Mid-Y to look for parts 10:24 Elli – Going to X-Arm Spool 10:52 Dan – Giving a tour of LVEA for students 10:53 Jodi & Travis – Back from the Mid stations 10:55 King Soft on site to take water samples 11:10 Richard – Out of LVEA 13:07 Filiberto & Ed – Going to End-Y to remove 3IFO power supplies 14:25 Corey – Going into the LVEA 14:31 Corey – Out of the LVEA 15:00 Guardian training in OSB large conference room
Bottom line--If we collect fewer channels, that is, if the database has fewer Analog inputs to process, the database process time changes. This may seem obvious, but come on, we are collecting 15 pressure sensors, doing two or three calcs and the PID. Oh yeah, there is the 1 sec heartbeat, so sure, this processor is really stressed!
The first two attachments are matlab histograms of the DT field of the the epics PID. This is the time between the PID process iterations used in the PID calculation. In alog 16619 JeffK reports a DT value of 0.55sec; this is with 15 sensors collected and the epics running at 10 hz---the processor is only getting to the PID calculation every .55 sec! When I reduce the number of channels collected to 7, DT drops to 0.312 secs (first attachment.) In the second attachment is the histogram when the epics is set to run at a 1 second SCAN rate. Here, the PID record is processing at ~1second but is multimodal with about a 1mhz variation.
The third attachment is the Power Spectra of the Differntial Pressure running the Pump Servo and its coherence with the HEPI L4Cs at the BS. In the spectra you can see the zero related to the update period of the PID: Blue--15 sensors 10hz epics ==> 0.55sec update= 1.88hz; Red--7 sensors 10hz epics ==>0.312sec=3.2hz. And clearly Red: 15 chanels at 1hz ==> 1 sec update=1 hz.
The lower panel of the third plot shows the coherence with the Pressure to the BS HEPI L4Cs, it shows just H3 which had the strongest coherence in our normal configuration early Tuesday morning. The Blue Trace is that normal configuration of 15 sensors collected at a 10hz epics scan; the Green trace is when the sensors collected dropped to 7 as does our coherence with the reduced gain peaking. When we shift the process to 1hz scan shown in the Red trace (without adjusting the PID parameters!) the gain peaking increases as does our coherence again. As Jeff has done in alog 16782, we need to recalculate the PID parameters if we wish to reduce the epics scan rate.
A couple of comments / clarifications: (1) I attach the extra bit of information -- we now have three different data points for these silly sods of CPUs: Station nSensors CPU Clock Cycle EX 6 0.288 Corner 7 0.312 Corner 15 0.552 A linear fit of these numbers reveals that we pick up 29 [ms] per sensor. Gross. (2) Hugh's histograms of the clock-cycle were produced from ~5000 data points, querying the PID's subfield DT as he says. What's interesting is comparing the histograms from the three corner station data points, 7 Sensors 10 [Hz] mostly-uni-modal, with a few slips -- 0.21% of the queries 15 Sensors 10 [Hz] uni-modal 15 Sensors 1 [Hz] tri-modal where we're taken care to span the same clock cycle range, and to have the same bin-width in all histograms. Also note that when refiring to the time between modes in the 15 sensors, 1 [Hz] data, they're 1 [ms] (millisecond) apart, not 1 [mHz]. Interesting? Yes. Important? Probably not. If we're sampling at 1 [Hz], then a clock uncertainty at 1000 [Hz] should make very little difference to us. It's more important that we can freely add and subtract sensors without having to worry whether the sampling rate will change, and/or be different than we request. So, we'll stick with a 1 [Hz] requested sampling rate, and use the design from 16782 and confirm goodness.