back on new Rossa from Xi computing
Update: Sun Nov 3 18:08:48 2019
Update: Fri Nov 15 00:00:26 2019:
We think we got to the bottom of this issue today. The RF signal level going into the demod board is too high. This electronics chain needs some careful gain reallocation.
I was demonstrating to Koji a strange feature I had noticed in the ALS control, whereby when applying a CARM offset to detune the arms, the two arms seemed to respond differently (based on the transmission levels). This kind of CARM-->DARM coupling seemed strange to me. Anyway, I also noticed that the EPICS indicators on the ALS MEDM screen suggested ADC saturations were going on. I had never really looked at the fast time series of the inputs to the phase tracker servos, but these showed saturating behavior on ndscope traces. I went to the LSC rack and measured these on a scope, indeed, they were ~20V pp.
The output of the BeatMouth PDs are going to a ZHL-3A amplifier - we should consider replacing these with lower gain amplifiers, e.g. the Teledyne AP1053. This is relegated to a daytime task.
Other findings tonight:
While working on the PSL table, I somehow put the IMC FSS into a bad state, reminiscent of this behavior. Seems like this is linked to some flaky connection on the PSL table. One candidate is the unstable attachment of the Pomona box between the NPRO PZT and the FSS output - we should install a short BNC cable between these to avoid the lever arm situation we have right now.
The Trillium T240 seismometer needs mass re-centering. Has anyone done this before, and do we have any hardware to do this?
I went to the Trillium interface box in 1X5. In this elog, Koji says it is D1000749-v2. But looking at the connector footprint on the back panel, it is more consistent with the v1 layout. Anyway I didn't open it to check. Main point is that none of the backplane data I/O ports are used. We are digitizing (using the fast CDS system) the front panel BNC outputs for the three axes. So of the various connectors available on the interface box, we are only using the front panel DB25, the front panel BNCs, and the rear panel power.
The cable connecting this interface box to the actual seismometer is a custom one I believe. It has a 19 pin military circular type hermetic connector on one end, and a DB25 on the other. Power is supplied to the seismometer from the interface box via this cable, so in order to run the test, I had to use a DB25 breakout board to act as a feedthrough and peek at the signals while the seismometer and interface boards were connected. I used Jenne's mapping of the DB25--> 19 pin connector (which also seems consistent with the schematic). Findings:
I am holding off on attempting any re-centering, for more experienced people to comment.
I'd like to revive the PRC angular feedforward system. However, it looks like the coherence between the vertex seismometer channels and the PRC angular motion witness sensor (= POP QPD) is much lower than was found in the past, and hence, the stabilization potential by implementing feedforward seems limited, especially for the Pitch DoF.
I found that the old filters don't work at all - turning on the FF just increases the angular motion, I can see both the POP and REFL spots moving around a lot more on the CRT monitors.
I first thought I'd look at the frequency-domain weiner filter subtraction to get a lower bound on how much subtraction is possible. I collected ~25 minutes of data with the PRC locked with the carrier resonant (but no arm cavities). Attachment #1 shows the result of the frequency domain subtraction (the dashed lines in the top subplot are RMS). Signal processing details:
The coherence between target signal (=POP QPD) and the witness channels (=seismometer channels) are much lower now than was found in the past. What could be going on here?
I wanted to restart the c1oaf model. As usual, the first time the model was restarted, it came back online with a 0x2bad error. This isn't even listed in the diagnostics manual as one of the recognized error states (unless there is a typo and they mean 0x2bad when they say 0xbad). The fix that has worked for me is to stop and start the model again, but of course, there is some chance of taking all the vertex FEs down in the process. No permutation of mxstream and daqd process restarts have cleared this error. We need some CDS/RCG support to look into this issue and fix it, it is not reasonable to go through reboots of all the vertex FEs every time we want to make a model change.
I looked into the seismometer situation a bit more today. Here is the story so far - I think more investigation is required:
Attachment #2 has some spectrograms (they are rather large files). They suggest that the increase in noise in the 0.1-0.3 Hz band in the BS seismometer X channel is real - but there isn't a corresponding increase in the other two seismometers, so the problem could still be electronics related.
Yesterday, Koji and I noticed (from the wall StripTool traces) that the vertex seismometer RMS between 0.1-0.3 Hz in the X-direction increased abruptly around 6pm PDT. This morning, when I came in, I noticed that the level had settled back to the normal level. Trending the BLRMS channels over the last 24 hours, I see that the 0.3-1 Hz band in the Z direction shows some anomalous behaviour almost in the exact same time-band. Hard to believe that any physical noise was so well aligned to the seismometer axes, I'm inclined to think this is indicative of some electronics issues with the Trillium interface unit, which has been known to be flaky in the past.
Attachment #1 - comparison of the POP QPD PIT and YAW output signal spectra with and without them being normalized by the SUM channel. I guess the shape is different between 30-100 Hz because we have subtracted out the correlated singal due to RIN?
This did not have the effect I desired - I was hoping that by normalizing the signals, I wouldn't need to change the gain of the ASC servo as the buildup in the PRC changed, but I found that the settings that worked well for PRMI locked with the carrier resonant (no arm cavities, see Attachment #2, buildup RIN reduced by a factor of ~4) did not work for the PRMI locked with the sideband resonant. Moreover, Koji raised the point that there will be some point in the transition from arms off resonance to on resonance where the dominant field in the PRC will change from being the circulating PRC carrier to the leaking arm carrier. So the response of the actuator (PRM) to correct for the misalignment may change sign.
In conclusion, we decided that the best approach to improve the angular stability of the PRC will be to revive the PRC angualr feedforward, which in turn requires the characterization and repair of the apparently faulty vertex seismometer.
The closest thing I can think of is here.
Is there a loop model of green PDH that agrees with the measurement? I'm wondering if something can be done with a compensation network to up the bandwidth or if the phase lag is more like a non-invertible kind.
During our EX AM/PM setups, I don't think we bumped the PDH gain knob (and I hope that the knob was locked). Possible drift in the PZT response? Good thing Shruti is on the case.
I made a change to the c1ass model to normalize the PIT and YAW POP QPD outputs by the SUM channel. A saturation block is used to prevent divide-by-zero errors, I set the saturation limits to [1,1e5], since the SUM channel is being recorded as counts right now. Model change is shown in the attached screenshots. I compiled and installed the model. Ran the reboot script to reboot all the vertex FEs to avoid the issue of crashing c1lsc.
I tried implementing a basic PRMI ASC using the POP QPD as a sensor. The POP22 buildup RMS is reduced by a factor of a few. This is just a first attempt, I think the loop shape can be made much better, but the stability of the lock is already pretty impressive. For some past work, see here.
The EX PDH setup had what I thought was insufficient phase and gain margins. So I lowered the gain a little - the price paid was that the suppression of laser frequency noise of the end laser was reduced. I actually think an intermediate gain setting (G=7) can give us ~35 degrees of phase margin, ~10dB gain margin, and lower residual unsuppressed AUX laser noise - to be confirmed by measurement later. See here for the last activity I did - how did the gain get increased? I can't find anything in the elog.
Looking at the sensor and oplev trends over the weekend, there was only one event where the optic seems to have been macroscopically misaligned, at ~11:05:00 UTC on Oct 19 (early Saturday morning PDT). I attach a plot of the 2kHz time series data that has the mean value subtracted and a 0.6-1.2 Hz notch filter applied to remove the pendulum motion for better visualization. The y-axis calibration for the top plot assumes 1 ct ~= 1 um. This "glitch" seems to have a timescale of a few seconds, which is consistent with what we see on the CCD monitors when the cavity is locked - the alignment drifts away over a few seconds.
As usual, this tells us nothing conclusive. Anyways, I am re-enabling the watchdog and pushing on with locking activity and hope the suspension cooperates.
The satellite box was reconnected and the suspension was left with watchdog off but OSEM roughly centered. We will watch for glitches over the weekend.
There are no unexpected red-flags in the performance of the DFD electronics. The calibration factors for the digital phase tracker system are 71.291 +/- 0.024 deg/MHz for the X delay line and 70.973 +/- 0.024 deg/MHz for the Y delay line, while the noise floor for the frequency noise discrimination is ~0.5 Hz/rtHz above 1 Hz (dominated by ADC noise).
Conclusion and next steps:
I still don't know what's responsible for the anomalously low noise levels reported by the ALS-X system sometimes. Next test is to check the EX PDH system, since on the evidence of these tests, the problem seems to be imprinted on the light (though I can't imagine how the noise becomes lower?).
We also took this opportunity to re-connect the interlock to the Innolight controller (after it was disconnected for diagnosing the mysterious NPRO self-shutdowns). The diode pump current was dialled down to 0, the interlock wires reconnected, and then the diode current was ramped back up to the nominal 2.1 A. The fan to cool the unit remains mounted in a flaky way as we couldn't locate the frame Chub had made for a more secure mounting solution.
It seems like the pointing of the beam out of the laser head varies somewhat after the startup - I had to adjust the pointing into the PMC a couple of times by ~1 full turn of the Polaris mount screws, but the IMC has been locked (mostly) for the last ~16 hours.
I've checked the state of the laser interlock switch and everything looked normal.
- At the X end, we set up the network analyzer to begin measurement of the AM transfer function by actuation of the laser PZT.
- The lid of the PDH optics setup was removed to make some checks and then replaced.
- From the PDH servo electronics setup the 'GREEN_REFL' and 'TO AUX-X LASER PZT' cables were removed for the measurement and then re-attached after.
- The signal today was too low to make a real measurement of the AM transfer function, but the GPIB scripts and interfacing was tested.
Koji suggested systematic investigation of the ETMX suspension electronics. The tests to be done are:
So the ETMX satellite box is unplugged now, starting 530 pm PDT.
Some ideas that would help increase the locking duty-cycle in the short term.
This is as far as I got last night. The first step is to see how reliable the settings determined last night are, today. I don't understand how changing the output matrix element can have brought about such a significant change in the MICH/PRCL separation in all the RF photodiodes.
There is poor separation of the PRCL and MICH length error signals as sensed in the 3f photodiodes. I don't know why this is so - one possibility is that the MICH-->PRM matrix element in the LSC output matrix needs to be tuned to minimize the MICH -->PRCL coupling.
Over the last few days, I've been trying to make the 3f locking of the PRMI more reliable. Turns out that while I was able to lock the PRMI on 3f error signals, it was just a fluke. So I set about trying to be more systematic. Here are the steps I followed:
Attachment #1 is the result. I don't know what is the reason for such poor separation of the MICH and PRCL error signals in REFL165. The situation seems very different from when I had the DRMI locked in Nov last year.
After this exercise, I tried for some hours to get the 3f PRMI locking going with the arm cavities held off resonance under ALS control, but had no success. The angular motion of the PRC isn't helping, but I feel this shouldn't be a show stopper.
Batteries + power cables replaced, and computers back on UPS from today ~3pm.
The UPS is now incessantly beeping. I cannot handle this constant sound so I shut down all the control room workstations and moved the power strip hosting the 4 CPUs to a wall socket for tonight. Chub and I will replace the UPS batteries tomorrow.
[Liz, Gautam, Chub, Jordan, Koji]
We removed a significant amount of e-waste from the lab. The garbage was moved to the e-waste station in WB SB and are waiting for disposal.
Updated Circuit Diagram and photos: https://dcc.ligo.org/D1500308-v2
- (1) and (6) of the diagram: TFs with various gain slider values for REFL1/REFL2/AO GAIN [ELOG 14948] (gain values and time delay modeling)
- Switching checks, latest photo of the board, Limiter check [ELOG 14953]
- (2): Boost transfer functions [ELOG 14955]
- (3): Slow (aka Length) CM output path [ELOG 14965]
- (4): Pole-Zero filter TF [ELOG 14965]
- (5): TF from TESTA2 to TESTB2 [ELOG 14966]
- (6): AC coupling TF of the AO GAIN stage [ELOG 14967]
- (7): AC coupling TF of the IN2 stage on IMC servo board [ELOG 15044]
Slow path = (1)*(2 if necessary)*(3)*(4 if necessary)
Fast path = (1)*(2 if necessary)*(4 if necessary)*(5)*(6)
gautam 20191122: Adding the measured AC coupling of the IN2 input of the IMC servo board for completeness.
Input referred offsets on the IN1/IN2 were tested with different gain settings. The two inputs were plugged by the 50 ohm terminators. The output was monitored at OUT1 (SLOW Length Output). The fast path is AC coupled and has no sensitivity to the offset.
There is the EPICS monitor point for OUT1. With the multimeter it was confirmed that the EPICS monitor (C1:LSC-CM_REFL1_GAIN) has the right value except for the opposite sign because the output stage of OUT1 is inverting. The previous stages have no sign inversion. Therefore, the numbers below does not compensate the sign inversion.
Attachment 1 shows the output offset observed at C1:LSC-CM_REFL1_GAIN. There is some gain variation, but it is around the constant offset of ~26mV. This suggested that the most of the offset is not from the gain stages but from the later stages (like the boost stages). Note that the boost stages were turned off during the measurements.
Attachment 2 shows the input refered offset naively calculated from the above output offset. In dependent from which path was used, the offset with low gain was hugely enhanced.
Since the input referred offset without subtracting the static offset seemed useless, a constant offset of -26mV was subtracted from the calculation (Attachment 2). This shows that the input refered offset can go up to ~+/-20mV when the gain is up to -16dB. Above that, the offset is mV level.
I don't think this level of offset by whichever OP27 or AD829 becomes an issue when the input error signal is the order of a volt.
This suggests that it is more important to properly set the internal offset cancellation as well as to keep the gain setting to be high.
The output stage (and AO GAIN stage) of the MC board was modelled. The transfer function was measured with the injection from EXC B. The denominator was TESTB2, and the numerator was SERVO OUT.
This stage is AC coupled by 2x 1st order HPFs. Firstly, this transfer function was measured with AO GAIN set to be 0dB. (Attachment 1)
This TF was used to characterize the cutoffs of the HPF stages, represented as the following ZPK:
Then the AO GAIN was already measured as seen in [ELOG 14948]. The AO gain TF was then modeled by LISO with the above HPF as the preset. This allows us to characterize the time delay of the AO GAIN part.
For the CM board modeling purpose, the transfer function from TESTA2 to TESTB2 was needed. (Attachment 1)
The ZPK model of this part is
CM Board Slow out (digital length control) path transfer function / pole-zero filter pair (79Hz/1.6kHz) transfer function
The excitation was given from EXC A. The denominator was TESTA2, and the numerator was OUT1.
Attachment 1 shows the measured transfer function with and without PZ filter off and on. The PZ filter provides ~26dB attenuation at high frequency. The output stage has a single order 100kHz LPF and it is visible in the transfer function.
The transfer function without the PZ filter was modelled by LISO as the following PZK representation. There looked a small step in the TF which caused the additional PZ pair (66~67Hz) but has very minor effect in the mag and phase.
The transfer function of the PZ filter was separately analyzed. The TF with the switch ON was normalized by the one with the switch OFF. Thus it revealed the pure effect of the switch. The PZK model of the stage was estimated to be
[Jon, Yehonathan, Gautam, Aaron, Shruti, Koji]
We get together on Wednesday afternoon for cleaning the lab. Particularly, we collected e-wastes: VME crates, VME modules, old slow control cables, and other old/broken electronics. They are piled up in the office area and the cage outside rioght now (Attachments 1/2). We asked Liz to come to pick them up (under the coordination with either Gautam or Koji). Eventually this will free up two office desks.
Also, we made the acromag components organized in plastic boxes. (Attachment 3)
Hardware issues that need addressing:
This problem has re-surfaced. Is this indicative of some problem with the on-board VGA? Even with 0dB of whitening gain, I see PDH horns that are 10,000 ADC counts in amplitude, whereas the nominal whitening gain for this channel is +18dB. I'll look at it in the daytime, not planning to use REFL55 for any locking tonight.
After making sure the beams were hitting the 3f photodiodes on the "AP" table, I was able to lock the PRMI with the sidebands resonant inside the RC using 3f error signals. This would be the config we run in when trying to lock some more complicated configuration, such as the PRFPMI (i.e. start with the arms controlled by ALS, held off resonance). Tonight, I will try this (even though obviously I am not ready for the CARM transition step). The 3f lock is pretty robust, I was able to stay locked for minutes at a time and re-acquisition was also pretty quick. See Attachment #1. Not sure how significant it is, but I set the offsets to the 3f paths by averaging the REFL33_I and REFL33_Q signals when the PRMI was locked with the 1f error signals.
As usual, there's a lot of angular motion of the POP spot on the CCD monitor, but the lock seems to be able to ride it out.
Lock-settings (I modified the .snap file accordingly):
REFL33_I --> PRCL, loop gain = -0.019, Trigger on POP22, ON @ 20cts, OFF@0.5cts.
REFL33_Q --> MICH, loop gain = +1.4, Trigger on POP22, ON @ 20cts, OFF@0.5 cts.
It would be good if you and Shruti can look at how to change the parameters in Zero so as to do a fit to the measured data. Usually, in scipy.optimize we give it a function with some changeable params, so maybe there's a way to pass params to a zero object in that way. I think Ian and Anchal are doing something similar to their FSS Pockel's cell simulator.
I installed nds2 again, this time successfully with
conda install -c conda-forge python-nds2-client
I installed nds2 on donatello with yum, but still can't import nds2.
Looking at the old latch.st code, looks like this is just a heartbeat signal to indicate the code is alive. I'll implement this. Aesthetically, it'd be also nice to have the hex representation of the "*_SET" channels visible on the MEDM screen.
Latch logic works. But latch alive signal is missing.
The boost filters of the CM servo board were tested. Their ZPK models were made.
The transfer functions of the boost filters were measured with the SG output of a SR785 connected to IN1. The IN1 gain was set to be 0dB. The transfer function was taken between the IN1 input and the TEST1A output.
With no boost and normal boost, the input signal amplitude was fixed to 20mVpk. For the other boosts, however, I could expect large gain variation through a single sweep. Therefore automatic SG amplitude tracking was used. The target was to have the output to be 1V with maximum amplitude of 100mV.
Attachment 1 shows the measured transfer functions.
The pole and zero frequencies of the boosts were estimated using LISO. Here the TFs were normalized by the TF of 'no boost' to cancel the delay of the other stages including that of the monitor channel.
ZPK model of Normal Boost:
ZPK model of Super Boost (State1):
ZPK model of Super Boost (State2):
ZPK model of Super Boost (State3):
In preparation for some locking work tonight, I did the following at the POP in air table with the PRMI locked on carrier:
== Test Status ==
[done] Whitening gain switching test
[done] AA enable/disable switching
[0th order] LO Det Mon channel check
[none] PD I/F board check
[done] QPD I/F board check
[done] CM Board
[none] ALS I/F board
The photos of the latest board can be found as Attachments 3/4
With some input signals, the functionarities of the CM servo switches were tested.
After the tests the LSC cables were reconnected (Attachment 6)
I think this offset setting thing is not so good. People do this every few years, but putting offsets in servos means that you cannot maintain a stable alignment when there are changes in the laser power, PMC trans, etc. The better thing is to do the centering of the WFS spots with the unlcoked beam after the control offsets have been offloaded to the suspensions.
I simulated this circuit with zero, but haven't gotten the results to match the measurements above.
There is ~ 7% variation in the power seen by the MC2 trans QPD, depending on the WFS offsets applied to the MC2 PIT/YAW loops. Some more interpretation is required however, before attributing this to spot-position-dependent loss variation inside the IMC cavity.
Attachment #1: This shows a scatter plot of the MC2 transmission and IMC REFL average values after the WFS loops have converged to the set offset positions. The size of the points are proportional to the normalized variance of the quantity. The purpose of this plot is to show that there is significant variation of the transmission, much more than the variance of an individual datapoint during the course of the averaging (again, the size of the circles is only meant to be indicative, the actual variance in counts is much smaller and wouldn't be visible on this plot scale). For a critically coupled cavity, I would have expected that the TRANS/REFL to be perfectly anti-correlated, but in fact, they are, if anything, correleated. So maybe the WFS loops aren't exactly converging to optimize the inoput pointing for a given offset?
Attachment #2: Maps of the transmission/reflection as a function of the (YAW, PIT) offset applied. The radial coordinate does not yet mean anything physical - I have to figure out the calibration from offset counts to spot position motion on the optic in mm, to get an idea for how much we scanned the surface of the optic relative to the beam size. The gray circles indicate the datapoints, while the colormaps are scipy-based interpolation.
Attachment #3: After talking with Koji, I explicitly show the correlation structure between the IMC REFL DCMON and MC2 TRANS. The shaded ellipses indicate the 1, 2 and 3-sigma bounds for the 2D dataset going radially outwards. The correlation coefficient for this dataset is 0.46, which implies moderate positive correlation. 🤔
The following was implemented in a python scipt:
I am now setting the offsets to the WFS QPD loop to the place where there was maximum transmission, to see if this is repeatable. In fact it was. Looking at the QPD segment outputs, I noticed that the MC2 transmission spot was rather off-center on the photodiode. So I went to the MC2 in-air optical table and centered the beam till the output on the 4 segments were more balanced, see Attachment #4. Then I re-set the MC2 QPD offsets and re-enabled the WFS servos. The transmission is now a little lower at ~14,500 counts (but still higher than the ~14200 counts we had before), presumably because we have more of the brightest part of the beam falling on the gap between quadrants. For a more reliable measurement, we should use a single-element photodiode for the MC2 transmission.
The logic chips 74ALS573 were replaced. And now the gain sliders are working properly.
== Test Status ==
Last week we found that the logic chip for the REFL1 gain switching was not transmitting the input logic. I went to Downs and obtained the chips. After some inspection some other latch chips were suspicious. Therefore U46, U47, and U48 (#1, #3, and #4 from the top) were replaced. After the replacement, the gain measurements were repeated. This time the test for the AO gain was also performed. Now all three slideres show the gain as expected except for the consistent -0.2dB deficit.
Note that the transfer functions for the REFL gains were measured with the input at IN1 or IN2 and the output at TESTA1. The TFs for the AO gain was measured with the excitation at EXC B, the input at TESTB2 and the output at the SERVO output. The gain and phase variantions for the AO gain at low frequency is the effect of AC coupling existing between the excitation and the servo output.
[Update on Oct 14, 2019]
The measured transfer functions show the phase delay determined by the opamps involved. The phase delay well below the pole frequencies can be represented well by a simple time delay (a phase delay linear to the frequency). Attachment 7 shows the time delay estimated by LISO for each gain setting of each gain stage. REFL2 has particularly large phase delay because of the use of OP27s. The delay is even larger when the gain is high presunmably because of the limited GBW.
Now with the CM board tested with the signal injected, it turned out that the latch logic was flipped. As the default state locked the digital levels, the buttons other than the mbbo channels were inactive.
By giving 0 to C1:LSC-CM_LATCH_ENABLE, the modification of the digital state is enabled. And with the value of 1, the digital bits on the board is locked.
In order to reflect this, latch.py was modified and now the controls are all activated.
See trend. This is NOT symptomatic of some frozen slow machine - if I disable the WFS servo inputs, the lock holds just fine.
Turns out that the beam was almost completely missing the WFS2 QPD. WTF ðŸ˜¤. I re-aligned the beam using the steering mirror immediately before the WFS2 QPD, and re-set the dark offsets for good measure. Now the IMC remains stably locked.
Please - after you work on the interferometer, return it to the state it was in. Locking is hard enough without me having to hunt down randomly misaligned/blocked beams or unplugged cables.
I took this opportunity to do some WFS offset updates.
Mon Oct 7 14:51:53 2019. I closed the PSL shutter to measure the WFS head responsivity.
I made a thru calibration as in this elog, treating laser, reference PD, and WFS RF output as a three-port device. The DC current supplied to the laser is 20.0 mA in all cases. The Agilent spectrum analyzer supplies a -10 dBm excitation to Jenne laser's AM port, and A/B is measured with 20dB attenuation on each input port. Results are in /users/aaron/WFS/data/191007/. The calibration had 100 averages, all other measurements 32 averages; other parameters found in the yml file, same folder as the data.
I normalized the result by the difference between the dark and bright DC levels of each segment.
Mon Oct 7 17:29:58 2019 opened PSL shutter.