Ross, Terra, Carl, Tega, Jim, Dave, Matt, Ed, Daniel.
In this alog we report on some of the simulink model changes made in relation to PI work carried out at LHO during the period spanning April to 10th of June.
We currently have two possible error signals for PI damping: the trans-QPDs and the OMC DCPD signals. The trans-QPD signal is more readily available to drive the ETMs, whereas the OMC DCPD signal is more readily available to the ITMs. Since the OMC DCPD signal has a better SNR compared to the trans-QPDs, it can be used to monitor the modes lines before they ring up. The main limitation of the OMC error signal lies with fact that its fidelity may change with better alignment of the X- and Y-arms.
There are currently three different strategies for PI damping in the suspi models, all of which have been implemented in the end station suspi models (h1susetmxpi and h1susetmypi).
The first takes data from the trans-QPDs, processes the data using a band-pass filter and a PLL (iWave) and sends the output to the ESD driver. This is the top-left block in the suspi models called "ETMX_PI_DAMP". It comprises 8 modes blocks each one of which contains an IWave PLL block for dynamic line tracking. The down-conversion block immediately below the trans-QPD mode block provides the means for sending the trans-QPD signals to the corner station (the data transfer for this has not been implemented yet).
We have also adapted the iWave line tracker to PI damping. The simulink model shown below provides the means for the user to enter a threshold value above which the we maximally damp and below which we damp proportionally. There is also a modified IWAVE_AMPCTRL block (in userapps/sys/common/models/IWAVE.mdl) that damps the mode to a manageable level that is again specified by the user so that we can continuously monitor the mode power in the error signal. This block is not currently deployed but should be amongst several variations of PI damping to be tested in the future in order to down-select to the best configuration.
The computation block with 4 outputs is only used in the corner station pi model (h1susitmpi), where we have control of the four ESD quadrants.
The most recent test of this system shows that updating the parameters (SINGAIN, COSGAIN, frequency) of the remote oscillator introduced considerable noise. A short term solution---aimed at demonstrating the viability of the system---is the reduction of the rate at which we update the oscillators. This is the role of the sample-and-hold block in the monitor block below. Whilst the system has been shown to work successfully with this modification, further work is required to realise the full potential of this scheme. For example, some of the amplitude noise in the I and Q values can be eliminated by the constraining them with the tracked amplitude from iWave. This would be particularly useful should the I and Q values suffer the same distortion from the update of the fixed oscillators frequency in the omcpi model (h1omcpi).
To get the signal parameters (I, Q, frequency), we use a modified version of the iWave block that incorporates a fixed phase oscillator. Here, the oscillator serves two main functions: extraction of the I and Q values and providing a dynamic band-pass filter around the mode line of interest.
At the end station, we have companion oscillators whose parameters (SINGAIN, COSGAIN, frequency) we update via EPICs for proof of concept. We note that the first mode uses a slightly different architecture. This was part of the debug process intended to test whether an indirect update of the I and Q values of the oscillators reduces the noise at the output. So far this does not seem to be the case. Although this system currently uses EPICs, this does not have to be the case. The low bandwidth nature of the update means it can be readily added to the existing PI channel for moving the data between the corner and the end stations.
We have also adopted the MIN_MAX_CALC CDS library part for monitoring power build-up of PI modes over a wide range of frequencies and should be deployed shortly. The monitor block resides in userapps/sus/common/PI_MASTER.mdl.
Presently, there is no mechanism for dealing with multiple lines that are separated by less than 1Hz, so we have adapted iWave to function as a user configurable static and dynamic notch filter. These blocks reside in userapps/sys/common/IWAVE.mdl. The idea is to place at least one notch filter in series with iWave in the PI mode block and have it operate in static mode by default. Although preliminary work done on the test stand shows that this does not impact on the performance of the mode block in the absence of any nearby lines and stabilises the behaviour of the iWave PLL in their presence, further work is required to fully stress test this system against wandering lines that cross one another in the frequency domain before deployment.
Evan, Sheila
Earlier today we had several locklosses while trying to engage the soft loops which may have been because SRM was uncontrolled and the POP90 power increased.
We searched for a better error signal for a while tonight, we looked at dither SRM and demodulating DARM (low SNR), demodulating SRCL (good signal but a large offset), POPX, some REFL WFS.
Finally we decided to try AS45I, where we had a reasonable signal with a zero crossing that gave us good sideband buildups for both pit and yaw. For pit we are using 1*ASA45I +0.76*ASB45I, while for yaw we are using ASA45I+ASB45I. The guardian is engaging these with a gain of -300 in SRC_ASC, which has worked once. It allowed us to turn on the soft loops and corrected the sideband buildups.
Overall violin modes situation improved by a little more than an order of magnitude. Currently the modes with highest magnitudes are 501.092 (IX Mode3, lost lock before I could find the damp setting), 501.606 (IY, damp setting found but didn't have much time to damp), and 507.194 (EX, no damp setting yet).
Modes with newly discovered damp settings:
505.71 & 505.707: EX Mode3: FM1, FM2, FM4, FM5 +20 gain seems to work for both lines. After I noticed the filter stopped damping I turned off Mode 3 gain and turned on Mode2 gain based on Sheila's configuration here. This configuration will damp the modes even further. We lost lock before I could confirm that it won't blow either .71 or .707 up.
505.587: EX Mode1: FM1, FM4 +50 gain
507.159: EX Mode6:FM1, FM4 +gain (forgot to write down how much gain exactly. Probably no more than 100)
507.391: EX Mode8: FM1, FM4 +50 gain
500.05: IX Mode1: FM1, FM4, +50 gain. +100 gain will damp the mode but later will blow it up. +50 gain seems to have worked pretty well.
502.744: IX Mode8: FM1, FM3, FM4, +10 gain (BP, +60deg, 100dB)
502.621: IX Mode7:FM1, FM4 +50 gain (BP, 240dB)
501.606: IY Mode1: FM1, FM2, FM4, +150 gain
All these information has been updated to the new violin mode table but haven't been implemented into the guardian.
Tonight I also confirmed all the filters that VIOLIN_MODE_DAMPING Guardian turns on (I didn't confirm the exact gain, but I did confirm the filter configuration and gain signs). They should be safe to turn on.
[Sheila, Jenne]
While looking at our locklosses today, Sheila noticed that we are using up way more DAC range than is appropriate for the SRM while we are locked using the 3f signals. Looking at a spectrum, it's all high frequency sensor noise.
I lowered the frequency of the cutoff from 300 Hz to 200 Hz. This eats 4 deg of phase at the 40 Hz UGF, so should be plenty fine according to an old OLG measurement Sheila found. We could / should improve this even more, but (a) it's better than it was before, (b) this is only really a problem while we're on our noisy 3f sensor, and (c) we don't want to get too much more crazy without taking a fresh loop measurement, which we don't want to spend commissioning time on right now.
Attached is a set of spectra showing the old SRM output (blue) and the new (red). Don't mind the bounce mode in the red spectrum - damping wasn't on yet during this measurement.
Today, H1 was primarily in the hands of commissioners working on high power. At the end of the shift, H1 was passed over to the SEI team so they could do BRS work (mainly because H1 was starting to break lock quite a bit); this happened as the wind gusts were approaching 40mph.
Guardian was updated (Violin & Bounce mode damping separated). I tried restarting nuc4 to show the updated ISC_LOCK Guardian, but had trouble with bringing it back up (the TV stayed OFF).
Day's Activities
0945 - 1000 hrs local -> To and from Y-mid Opened check-valve bypass valve, opened LLCV bypass valve 1/2 turn -> LN2 at exhaust after 1 minute 10 seconds -> Returned valves to as found Next manual overfill to be Wednesday, June 15th.
Group Status:
Maintenance:
Outreach:
respit from outreach visits. School group on Fri.
1800 hrs. local -> Leaving now
On-Off-On-Off-On.... PT100A back and forth across 5x10-3?
It does not appear to be the interlock. It is still forced on in Beckhoff, and trending it (attached) shows no change while the cold cathode drops in and out.
Krishna, Michael
We took a look at the noise of BRSX with damping forced on and off now that the new damper is installed. For context, the previous damper injected lots of broadband noise which made the BRS unusable during damping. The first plot in the attachment shows that with the new damper we get comparable tilt subtraction whether the damping is on or off. The second plot is a more broadband view of the BRS noise and shows that the damping rings up the torsional mode (~3 Hz) and some other higher frequency modes but doesn't change the broadband noise. The wind picked up right when we forced the damping on so there's an overall increase in low frequency motion.
After taking the measurements with the damping forced on (~1 hr), we found that the BRS had started drifting rapidly which decreased the ability to subtract tilt. This is due to the heat generated by the stepper motor that drives the turntable. In normal operating conditions, the damper shouldn't ever need to run for this long and thus shouldn't create such a large of a drift.
The take away from this is that the BRSX sensor correction need not be turned off if damping does trigger. The tilt subtraction might not be very effective shortly after damping but it should still be useable.
Michael, Krishna, Jim, Hugh
Here's a brief look at the performance of the BRSY sensor correction during Wednesday's ~20-40 mph winds. We were using 250 mHz blends and Warn_SC_2 sensor correction filters for Y axis (the beam axis). The interferometer was able to lock well as noted in 27658.
The first plot in the attached pdf shows an ASD of the ground seismometer Y, the BRS rX, the tilt-subtracted seismometer Y and the ST1 seismometer Y signals. The tilt-subtraction worked well and kept ST1-Y motion small even as isolation was provided above 0.1 Hz from the sensor correction filter. Note that ST1 still sees the ground tilt and the seismometer output is contaminated by tilt, hence the effect of the isolation is not clear in the ST1 Y channel but we expect it to be there. Ideally one would like to measure the tilt going into ST1 from the ground and cancel that (rX sensor correction), but as Robert showed in his alog a few weeks ago, ground tilt varies significantly on the LVEA floor and one would need multiple angle sensors to map it well or one (good) angle sensor on Stage 1.
The second plot shows the coherence between the ground seismometer Y and BRSY rX and that between the tilt-subtracted Y and BRSY rX. The high coherence in the first is good, but the high coherence in the second indicates that the tilt-subtraction is not optimal yet and improvements of ~2 could be had at low frequencies.
The third plot shows has the same lines from the first, along with the buried seismometer Y for comparison. One can see that the tilt-subtraction is more effective and is also more useful above 0.1 Hz.
The fourth plot has two additional lines - ST1 CPS and it's RMS. The RMS motion was kept below 1 micron but had an odd bump at 8 mHz. Jim and I tried to add a notch to the sensor correction at 8 mHz but were unsuccessful in suppressing it well.
This morning I modified the high pass filters for the tilt-subtracted seismometer channels as shown in the second attachment. The new (red) filters suppress signals below ~20 mHz with some gain-peaking at 40 mHz (which is acceptable) and decent phase above ~60 mHz. These seem to work well and keep the CPS signals small. They are installed at both ETMX and ETMY. The third attachment shows the sensor correction with this new filter . The wind-speeds were only ~10 mph but the CPS signal is smaller and doesn't show the ~8 mHz peak.
Will comment this entry when leaving
1610 hrs. local -> Leaving site now.
We are still losing lock more than 50% of the time once we start to increase the power into the interferometer. I slowed down the rotation stage to 10% of its normal speed. In this way I was able to make it to 10 W once. On the second try, the interferometer lost lock right after a gain step in the summing node board. The behavior of the CARM error signal does not seem any different from past successful power increases, so it is not clear to me what could be wrong with the summing node board.
In order to make it easier to engage the soft loops, I made the 20 dB boosts come on only when the soft loop error signals are less than 0.01 ct.
I also turned down all the hard loop gains by 6 dB and re-engaged the bounce mode notches in the suspension PUM drives, because the bounce modes were becoming unmanageable.
Today I redistributed gain in the CARM and IMC boards so that we can use the CARM gain slider to power up, rather than the SNB gain slider. I powered up to 10 W three times and didn't have any locklosses
I reoptimized the green QPD offsets for the Y arm in full lock.
Bounce damping setting was changed for IY and EX so they damp at all. Don't know why the old settings don't work.
For EX I just flipped the sign.
For IY, I rotated the phase by 210 degrees in effect and increased the gain significantly. Gain increase might be unnecessary but I leave it like that because it seems to work faster.
Old | NEW | |
EX |
Gain= -0.03 (negative), FM1 (+60dg), FM4 (bp) |
Gain=+0.03, FM1 (+60dg), FM4 (bp) |
IY |
Gain= -0.03 (negative), FM1 (+60dg), FM3 (BP), FM6 (+30dg) negative sign + 60 deg + 30deg = -90 deg total |
Gain= -0.15 (negative), FM2 (-60dg), FM3 (BP) negative sign -60 deg = +120 deg total |
ISC_Lock was changed but not loaded.
Travis and I added a filter with negative gain to IY bounce (and EX) damping, so the gains for all bounce and roll damping filters will be positive from now on. The guardian is updated for this.
We found that in the state DARM offset, we got good damping for ETMX with a positive gain and -60 degrees of phase. This worked better than the settings that Keita described in the DARM offset state, but we also used Keita's settings sucsesfully in DHARD WFS.
TJ and I saw that with the Keita settings at DARM_OFFSET we were ringing up the bounce mode for ETMX. So, now in the Resonance state the bounce damping gets set to the Sheila settings. Unclear why this one needs a 120 deg phase change but no others do. Perhaps related to why the settings needed changing in the first place? When the bounce damping is first turned on in DHARD_WFS, it goes back to the Keita settings.
We have been using a bandpass filter for ETMX bounce mode damping that is designed for the ETMY bounce frequency. For the other optics we use a broad bandpass, so I've now set the gaurdian to do that for ETMX as well. The new settings are:
FM3 (broad BP), and FM8 (gain -1). These are in the guardian for now, to be used in all states, so I've commented out the filter change in RESONANCE.
When we have the notches in the ASC off, we can damp the bounce modes as seen in the DARM spectrum. However, even when they appear to be reduced in the DARM spectrum they seem to be increasing in the MICH ASC signals.
I copied quad bounce and roll notches to the BS ASC loops as well.
I looked at a 1 mHz spectrum of the bounce modes from last nights lock, and the frequencies are consistent with what we reported one year ago:
ETMY | 9.731 Hz |
ETMX | 9.776 Hz |
ITMY | 9.831 Hz |
ITMX | not rung up last night but should be 9.847 Hz |
The confusion about the ITMY mode frequency (corrected by Evan in the comment) was unfortunately propagated to our monitoring filters. I've corrected the monitor bandpass for ITMY and made all of the bandpasses 10 mHz wide (before they were all 30 mHz wide, meaning that the monitors could not distinguish between the ITMs).
ETMX Mode3: FM1, FM2, FM4, +10 gain (BP, -60deg, 100dB) seems to have worked quite well. No need to turn on a damping for 505.707 (which should be checked again with 0.0001 BW to see if it's really exists). 505.707 should be watched while damping 505.710 just in case.
The new violin mode table is being uplated slowly. The modes with Damp Setting column filled out are the one that's been sucessfully damped by the filter settings indicated in the table.
Jenne, Sheila
With the long lock tonight we got to do some more violin damping. I haven't updated the wiki or guardian.
At first we thought the largest mode was the 505.71 mode that Nutsinee logged about, and tried to damp it without sucsess using the settings she described. After getting a 0.002 Hz spectrum we could see that it was really 505.706Hz. (on the wiki there is a note asking if this mode really exists, looks like it does) I tried to damp this with ETMX MODE 2, which has a narrower filter, and in the last 5 minutes of the lock it looked like a phase of -60 and gain of 30 was working, with the 505.71 notch engaged, but we unlocked before we could be really sure it was working.
We damped the mode at 506.922 (ETMX mode5) using a phase of 0 degrees and a gain of 30. FM1 and FM4. (This is different from what is on the old wiki) the new wiki has this mode not damped.
We also damped a 500.062Hz mode (ITMX mode 1, 0 degrees gain +30, also different from the wiki) With higher gain and the same filter settings we rang the mode up.
The top panel shows the fundamental violin mode forest at the begining of the lock in blue, and the end in red. (There is a tiny improvement in 505.707Hz) The bottom panel shows the 0.002 Hz spectrum with cursors at 505.71 and 505.707Hz.
A negative gain for the 505.707 mode worked for me (although it's not clear whether it damped 505.707 or 505.710).
The setting I had seems to work on both 505.707 and 505.71.