re WP 4913
I switched the HEPI pump Servo to run on the true differential pressure as LLO has been doing forever. I did glitch the system through my sillyness. I was all ready to do it without any serious glitch and next time it will be better. Still, the HEPI position loops on the platform did not trip--pretty good support for the hydraulic system. The differential pressure now does look slightly noisier based on the time series. But of course this is now a sum of two noisy signals, maybe time for some smoothing and serious grounding study.
Attached is 45 minutes of second trends for the Pump station channels and some cartesian basis HEPI L4C signals. I don't see any ill affects of this transition so far.
So the alarm manager is okay as I changed the database alarms. The medm though is also common so if you open the EndY HEPI Pump Servo medm, it will show pressure not okay. When it is deemed okay to continue down this path, I'll update the medm when I install the same database changes at EndX.
no restarts reported
Alexa, Nic, Evan, Dan, Sheila
Tonight we went back to trying to get ALS DIFF working well. We aren't sure what the problem is.
First Nic checked that the gain of the 2 ESDs were the same. The old X gain in L3 LOCK L was 0.2, the new one is 1.12 The old Y gain was 0.7 the new one is 0.384. We also checked the crossover by using the green control signal.
Attached are some plots, based on the suspension model that Jeff used to design the DIFF loop, and filters downloaded from foton. This is using none of the boosts for DIFF, which is how we have been running the last few nights without being able to use the ESD. The two plots on the left are the cross overs for our loop, and the open loop. We can use these plots to compare what we have now to what Jeff originally designed, (G1400146-v2) We also copied the livingston filters, see alog (12590), their crossover is the third plot while their open loop (we added a scaling factor to get the right ugf) is the last plot on the right.
Btoh ESDs are driving, but we seem unable to use them in the loop.
Nic, Alexa, Sheila, Evan, Dan
We wanted to measure the crossover between L1 and L3 of ETMY. We took a transfer function from LOCK_L1/L3 to ALS-Y_REFL_SERVO_CTRL_OUT with green locked to the yarm. We found an old template made by Sheila and Stefan for the L1 stage. We repeated the measurement for a few data points (green trace), and it seemed the same as the old measurement (blue) so we aborted the measurement and went with the assumption that it was still a valid measurement. We then measured for the L3 stage (red trace). Clearly the crossover between L1 and L3 is about 2 Hz as expected. Nic plans on dividing these two transfer functions tomorrow morning...
Fo reference the xml file is located here: /ligo/home/sheila.dwyer/ALS/HIFOXY/Y_UIM_2.xml
Here is the ratio of the ETMY UIM and ESD actuator gains (red and blue traces in original post).
It looks like our crossover is about 2Hz and the phase margin is about 45 degrees.
After spending far too long scratching my head about calibrating the OMC DCPD signals, I've measured the RIN of the OMC transmitted beam, and it's not pretty.
To make the measurement I locked the OMC on the side of a carrier TEM00 fringe, as described here. I measured the open-loop transfer function; this is the first plot attached (UGF is 100Hz, which is about the same as when we're dither-locked on the peak of the fringe with much higher gain in the LSC servo). To calibrate OMC-DCPD_NORM into true RIN I divided by 1/(1+G) in DTT so that the loop suppression is accounted for. To calculate the RIN seen by other PDs, I divide by the PD sum at DC.
The results, in the second plot, show that the beam incident on HAM6 has some issues. At high-ish frequencies, 100Hz and up, the OMC Trans intensity noise is due to noise out of the mode cleaner. It's coherent with IM4 Trans and the ISS second loop PDs. Probably this noise can be mitigated using the ISS second loop and also reducing the IMC angular fluctuations that were described by Gabriele.
At low frequencies, between about 0.2 and 3Hz, there is huge intensity noise on the input beam to HAM6 beam that is not seen just after the IMC (neither IMC_TRANS or the ISS second loop PDs). The noise is seen in HAM6 by the OMC QPDs (in green - limited by electronics noise above 50Hz?) and by ASC-AS_C (dashed black). At high frequencies the intensity noise in HAM6 is coherent with the noise out of the IMC, but the loud stuff around 1Hz is likely due to the clipping that Keita measured yesterday.
The third plot shows an attempt at characterizing the length noise of the OMC; the loop-corrected RIN was converted into meters using the derivative of the usual Fabry-Perot transmission formula at half-resonance (the conversion factor is 3.7e8 RIN/meter). This is a first step towards building an OMC noise budget, following the work by Zach at LLO. At high frequency (around a kHz), the noise begins to drop below the limit of 3x10^-16 meters/rt[Hz] prescribed by Valera's estimate in G1100903, but there's a forest of lines around 1kHz that's not unlike what was observed at L1. At lower frequencies, the noise seen at the DCPDs is dominated by the input intensity fluctuations, and it's not possible to measure the length noise that's intrinsic to the cavity. The photocurrent during these measurements was about 3mA for each DCPD (shot noise RIN of 10^-8/rt[Hz]).
Most of the noise between 50 and 800 Hz is largely non stationary, as visible in the spectrogram (figure 1). It is also clear that the non-stationarity is very similar to the one we see directly in transmission of the IMC
This non stationarity is, as expected, closely related to angular motions of the IMC. In particular, the noise fluctuates in the same way as the IMC transmitted power (figure 2). My guess is that the DC alignment of the IMC is again not very good.
Evan, Alexa
Even after the ETMY L2P work performed yesterday (alog 14832), last night we found the L 2 angle for ETMY was still bad. So today we did some more investigation, again. I don't think we made any big head-way, but I will post what we did and the results anyways...
Setup: We excited ETMY_L1_LOCK_L at various frequencies with an amplitude of 5e5 cts and 0 deg phase. The LOCK filters that are nominally installed were left on. Using the pitch lock-in, we demdoluated the pitch oplev signal at the respective frequencies and noted the I and Q signals. We then turned off this excitation and excited ETMY_L1_DRIVEALIGN_Y2P_EXC. This Y2P filter bank was empty, and we changed the gain from 0 to 1. The phase of this excitation was also set to 0 deg. We adjusted the gain such that the magnitude of the I,Q signals was close to that induced by the LOCK_L excitation. Then, we adjusted the phase of the Y2P excitation until we reduced both I, Q signals to zero. In other words, we drove in length with the LOCK_L, and examined the motion in the pitch oplev. We reduced this motion with an excitation in the Y2P drive align matrix. We had hoped this information would help us create a patch and improve our L2P in L1.
Results:
0.8 Hz
LOCK_L EXC = 5e5cts only | Y2P Gain = 16, 0 deg only | |
I | -0.412 | 0.174 |
Q | -0.082 | 0.387 |
|z| | 0.42 | 0.43 |
ang(z) | 11 deg | 66 deg |
With both excitations, and Y2P phase set to 55 deg, then I = -0.026, Q = -0.006
0.6 Hz
LOCK_L EXC = 5e5cts only | Y2P Gain = 16.5, 0 deg only | |
I | 0.54 | -0.585 |
Q | 0.44 | 0.46 |
|z| | 0.7 | 0.74 |
ang(z) | 39 deg | 38 deg |
With both excitations, and Y2P phase set to 1 deg, then I = -0.01, Q = -0.01
0.5 Hz
LOCK_L EXC = 5e5cts only | Y2P Gain = 10, 0 deg only | |
I | 1.02 | -0.85 |
Q | -0.33 | 0.73 |
|z| | 0.42 | 1.12 |
ang(z) | -18 deg | -40 deg |
With both excitations, and Y2P phase set to -22 deg, then I = -0.06, Q = -0.1
0.4 Hz
LOCK_L EXC = 5e5cts only | Y2P Gain = 3.8, 0 deg only | |
I | 0.045 | -0.092 |
Q | 0.083 | -0.019 |
|z| | 0.095 | 0.094 |
ang(z) | 3 deg | -2 deg |
With both excitations, and Y2P phase set to -5 deg, then I = -0.003, Q = -0.095
We then decided to take a transfer function from L1_LOCK_L_EXC to the pitch oplev, from 0.1 to 1 Hz, which was our region of interest. We forgot to save the DTT template and closed it, but the results were consistent with the measurement above within 10 dB. So we created a notch at 0.5 Hz since it seemed like we were over compenstating this resonance. However, our oplev spectrum did not improve at all. So we moved onto the damping which seemed to help a good amount at these frequencies. One thing we still need to check is confirm which of the stages has the worst L 2 angle.
Another thing... I talked to Arnuad at LLO and all their L2P, L2Y filters in the ETMs are simple low pass filters with the DC gain set to minimize the coupling. The LOCK filters then have nothces at 9, 10 Hz to handle those resonances, but no other resonance are compenstated for. Their M0 damping filters are "aggressive" and help reduce the angular motion.
Alexa, Evan
The ETMY oplev damping was off. We have turned both picth and yaw damping back on.
Additionally, I have tweaked the filters and the gain slightly in order to reduce gain peaking:
According to conlog, the loop gains appear to have been randomly adjusted several times over the last week.
We hope that this might improve the robustness of the ALS locking. If not, we might try a similar recommissioning of the ETMX oplev damping (since this are also off).
Nic, Evan
Last night, we were playing around a bit with the loop shape of the ETMY pitch oplev damping. We returned to the original shape, but found we could not reengage the loop without causing the optic to oscillate. It is currently locked with a factor of 10 less gain (0.3 → 0.03), which is stable but is probably not suppressing the optic motion at all.
I and Daniel were talking about the reason why there are bounce and roll coupling, and the bounce should be the local gravity axis versus the arm angle.
I looked at old DCC document (T980044) and found that Y arm is almost parallel to the vertex local Y axis while X arm is almost parallel to the X end local X axis.
Below is a table of angle deviation from pi/2 between the relevant arm and the local vertical defined by the gravity. Deviation=0 means that the arm is orthogonal to the local gravity. FYI, the difference in the local gravity angle over 4km is about 2*pi*4km/40000km = 628 urad.
EX (X end vertical VS global X) | EY (Y end vertical VS global Y) | IX (Vertex vertical VS global X) | IY (Vertex vertical VS global Y) | |
LHO | 8 urad | 639 urad | -619 urad | 12 urad |
LLO | 315 urad | 19 urad | -312 urad | -611 urad |
The bounce coupling is directl proportional to these numbers.
LHO EY is 80 times worse than LHO EX. At LLO EX is worse than EY, but LLO EX should be a factor of 2 better than LHO EY.
By Keita.
Roll mode:
If the roll motion happens in the plane that bisects the angle formed by the HR surface and AR surface, there's some coupling to length because of the wedge angle.
If that's the mechanism, the coupling depends on the centering on the mass.
By Keita
Jeff pointed out that some incoherent sensor noise (I think that's the phrase he used) in Krishna's post from earlier today. He suggested this may be due to RY coupling to X. We might be able to fix this by pushing up the blend on RY or use a blend the rolls off the seismometers at low frequency. RX and RY already use Ryan's 250 mhz blends, so at first glance we only have 1 filter with a higher blend. I've attached 4 plots, the first three of which show CPS, L4C and T240 components respectively of our available blends. The fourth plot compares the T01_28 (dashed) components and the current 250mhz blend (solid). It looks like the T01_28 maybe (?) fits the bill for more low frequency roll off of the seismometers.
9:15 Fil Aaron to EX working on cabling
10:00 Alastair to LVEA TCS-Y table
10:15 Hugh, Gerardo to LVEA
10:30 Alexa transitioning EY
11:30 JeffB to LVEA, moving SUS pallets for 3IFO
11:00 Kyle to EY cleaning up tools, done 12:00
13:00 JeffB, Bubba to LVEA looking at craning operations around High Bay, out 13:15
14:00 Gerardo to LVEA
14:30 Hugh, Suresh to HAM2 HAM3 for oplev work
K. Venkateswara
I had installed temperature sensors on BRS and GND_T240 yesterday as described in 14825. The first plot shows the trend over a day along with the PEM_VEA temperature sensor. The count to Kelvin conversion was expected to be 1.56e-3 K/count. This seems roughly consistent with the temperature of the T240. The BRS temperature sensor shows a lower magintude and a phase offset due to it's extra thermal shielding and larger mass.
The attached pdf shows the ASD of the temperature sensors and the coherence between them and their respective instruments. The temperature sensors are mostly limited by ADC noise. An op-amp based pre-amp of 50-100 gain would be useful. BRS_RY_Out shows a little bit of coherence near few mHz while T240 X, Y and Z show no coherence with the temperature sensor.
You should add the ADC noise to this plot
I've added a comment about it in 14909. I'm not sure how to display the ADC noise in DTT. In any case, it shouldn't be limited by ADC noise any more.
J. Warner, K. Venkateswara
We have turned on sensor correction on Z and X at ETMX since noon today.
Z on ETMX had been switched to TBetter blend by an unknown person during the weekend. Jim switched it back to the 90 mHz blend(Ryan's LLO blend). We then also turned on the X sensor correction which is currently using the tilt-subtracted super sensor.
The first two pdfs show the before and after plots for the two configurations. Before was taken ~3 am this morning and after was at ~12 noon. Note that the improvements in Yaw/Pitch at low frequencies were largely due to going to the 90 mHz blend and then using Z sensor correction.
The third plot shows the improvement in X and Z transfer functions from the ground to Stage 1 before and after the configuration change.
It seems to be helping over a large frequency range above 10 mHz and not hurting anywhere, so we will leave it on for overnight monitoring. If it affects commissioning in any way, it is easy to turn off through the ISI screens.
edit: Z sensor correction is to HEPI_IPS and X is to Stage1_CPS
J. Kissel, J. Warner In the "after" configuration, given that the coherent linear transfer function ("SCimprovement.pdf") only shows a factor of a few amplification at 50 [mHz], yet the ASD shows a factor of 20-30 amplification, I argue that this is even more evidence for re-injection of residual tilt noise of ST1. I haven't done the usual X_blend * RY * g/w^2, but I've suggested that Jim try to play around with the RY blend filter (i.e. just use one of the other blends we already have in the bank) and see if the change in performance of RY improves or worsens the X performance. #weneedamodel P.S. The optic looks like it was misaligned for the "after" measurement, or there was a laser glitch during the measurement, because the optical lever noise floor seems supremely high, and I'm confident does not reflect the displacement of the optic. The "before" looks OK.
As a follow up to my alog showing the H1 science frame usage per subsystem, I've found a way to show this as a readable pie chart
Gerardo & Hugh
We recentered these oplevs to make better use of their tracking. And especially HAM2 as it seemed to be way off on one quadrant. Luckily we didn't lose the beams.
First attached is a three day trend showing yesterdays fun with HAM2 and the recentering this morning.
Next is the last eighty minutes, maybe given the wandering, the X-Y plots on the medm should have their window zoomed out some. This seems to be an issue mostly on HAM3 Yaw.
The third plot shows the five hour detail around yesterday morning on HAM2 with the HEPI Cartesian RZ (yaw) RY (pitch) and HP, the horizontal pringle mode. This further shows even though the HEPI RZ has servo'd back to its reference, the OpLev Yaw would disagree. Likewise, even though the HEPI pitch shifted way off for most of the time and then came almost all the way back, the OpLev Pitch doesn't follow that in any reasonable way. Even though these OpLevs have no caibration as I understand, this says alot about system distortion. I've included a feducial from the large shift in the HP calculation but it doesn't appear to correlate to anything in the OpLev response.
The last paragraph about the last plot is confused by the reversed Pitch and Yaw of this OpLev, see 14965.
Evan, Sheila, Nic, Lisa To answer Peter's questions , we have relocked the DRMI (without arms) on 1f and 3f (10W input power, 27 mW on the BBPD photo-detector). The DRMI was very stable in both cases, here are the good times: Nov 5, 6:00 - 6:15 UTC DRMI locked on 1f, WFS on Nov 5, 6:30 - 6:45 UTC DRMI locked on 3f, WFS on Evan is about to post plots with error signal spectra. It is probably a good idea to make some plots with ground motion/ISI/optical lever signals to "capture" these good times.
These are the corresponding RFPD spectra. The analogous LLO measurement is LLO#15430. LLO's measurement of RF135 seems pretty surprising, since the 3f demodulated spectra appear to be almost entirely noise-dominated above 20 Hz.
What's this 2449.25Hz line that grows over time?
In the attached left, the error signal for PRCL and SRCL grew larger and larger toward the end of the lock.
In the attached right, red, blue and green correspond to the end, middle and the beginning of the lock. The difference in noise floor might be that they were switching from 1f and 3f or vice versa.
Regardless, a line at 2449.25Hz grew larger over time and it was dominating the error signal RMS.
Is this something intentional? Oscillation somewhere? Rogue line?
I am not sure what the 2449.25 Hz line is; there is not a line there that I know of...
In terms of the difference in the noise floor, I can confirm that this comes from the 1f to 3f transition. Keita's green trace was taken from 5:50 PM UTC, blue trace from just after 6:20 PM UTC, and red trace was taken at 6:50 PM UTC. I have attached a dataviewer snap shot showing the LSC PD input matrix of the MICH (2_23), PRCL (3_20), SRCL (4_22) 3f signals around that time. The MICH and PRCL signal transistion at around 6:17 PM UTC, which explains why the blue and red traces in those plots go up. Meanwhile, SRCL transitions a bit later at 6:26 PM UTC, which explains why only the red trace goes up in the noise floor.
According to Dennis the lowest elastic mode of the beamsplitter is expected to be at 2458 Hz (right circular cylindrical without wedge angle, bevels, wire standoff-prisms).
See the attached for some manipulated data.
There are now pressure signals coming from the End Station VEAs just before the BSCs fluid distribution manifolds. So the pressure before the actuators and the pressure just after the actuators (before and after the distribution manifolds.) See the first attachment--you can see that there is some 6psi pressure drop from the last transducer on the Pump Station Manifold to the Transducer just before the supply distribution manifold at the chamber. This is a distance of some 80 or 90 feet of 1" tube.
The idea is that the Actuators are meant to operate at a consistent pressure drop and having the sensors in the area of operation and where we have tighter temperature regulation would be a better thing. While these epics channels can be conditioned (smoothed, averaged,...) the second attachment shows how much noisier these raw signals would be to produce the differential pressure signal for the servo. I've subtracted the Return Pressure from the Supply to get the Differential; the vertical scales are the same for the two EX signals,and the same for the two EY signals. All plots are in PSI.
Why are they so much noisier? Let me see, maybe the 80 or 90 feet of cable? We do have an at chamber active signal amplifier, don't know the DCC off hand but later.
Anyway, I may be a little reluctant to switch HEPI to these signals. I don't think any one has complained about HEPI because we are running on the direct supply pressure rather than the differential.
As RichM pointed out to me, in the second plot, the lower traces of Output Pressure are in loop and the True Differential plotted above are out of loop. I would expect these situations to reverse whence the servo switches to the latter. Additioinally, the first plot which has the max & mins, show the overall noise on the three channels are really the same at the stations; there is certainly more noise on the EY sensors by something like a factor of 2. But you can see how much quieter the mean is on the servo'd channel.
So, I don't think there is sensor noise issue here to worry about.