HAM6 system is running with only the 500 l/s "main" ion pump (valved out the turbo pump), the turbo pump still running.
We'll assess progress tomorrow and determine if we can turn the turbo pump off.
As noted by Evan et al, SRCL coupling to DARM is highly non stationary. I looked into three minutes of data with the SRCL noise injection described in the cited elog entry (GPS time 1113211711 + 180 s).
In brief, the coupling is modulated exactly as the residual motion of ASC-AS_A_RF45_I_YAW_OUT. If we compute the coherence between DARM and SRCL with a gain modulated by ASC-AS_A_RF45_I_YAW_OUT, we get values very close to one. Read the rest of this report if you want to know more.
One strange thing is that it seems that the main contribution to this motion comes from ETMY PIT, even though the AS signal is YAW... See the last plot to see how AS_A_RF45_I_YAW moves in a very similar way to ETMY_L3_OPLEV_PIT...
The first attachment shows a spectrogram of DARM during the SRCL noise injection. The non stationarity behavior of the noise is very evident. The second attachment shows a coherogram, which is basically the same thing as the spectrogram, but showing how coherence changes over time. Again, coherence can go up to values of one, but clearly if you average over the entire time, you lose a lot since the coupling is non stationary. The third attachment shows the transfer-function-gram, which again shows how the transfer function from SRCL to DARM changes over time. You can see from the phase plot that the sign of the coupling flips multiple times. The 4th attachment is an animation of the transfer function between SRCL and DARM over time, which makes even more clear how the coupling changes amplitude and sign.
During the noise injection, I repeated the my standard BLRMS correlation with angular signals. I use the frequency band between 50 and 100 Hz. The correlation with angular signals (using all ASC error signals) is quite good, as shownm in the 5th attachment. I didn't continue further the analysis of which channels are more relevant, since the analysis described in the next section gives a better understanding.
Looking at the transfer function over time (see for example the animation in the 4th attachment) it is clear that the shape remains mostly the same, and only gain and sign changes over time. I therefore computed the average of the TF around 93 Hz, which corresponds to a point where the phase is close to zero (this is just for convenience, in this way the real value of the TF is a good estimate of the gain variation over time). The TF gain varies a lot over time, and as expected changes sign many times. In the 6th attachment I show in the blue trace how this gain varies over time. The red trace is the best fit obtained using my algorithm and all ASC error signals. The green is the residual, which shows hoiw the reconstruction is very good. Using my channel ranking algoritmh I found out that the main contributor is H1:ASC-AS_A_RF45_I_YAW_OUT_DQ is anticipatedm above. The 7th plot shows indeed that the fit is very good even if I use only this signal: so the conclusion is that the SRCL coupling is modulated completely by angular fluctuations visible in the AS port.
The 8th plot shows in blue the coherence between DARM and SRCL, if averaged over the entire 3 minutes of data. Clearly the coherence is kind of low, as explained above. The green trace is instead the coherence between DARM and an 'improved' SRCL signal, built as SRCL * (1 + gain * AS_A_RF45_I_YAW_OUT_DQ), with the gain obtained from the previous fit. The improvement is clear, showing that indeed I was able to recover most of the coupling gain and sign changes.
To get an even better estimation, I did an optimization based directly on the coherence. Basically, I considered the averaged coherence in the band 50-300 Hz, between DARM and a signal given by SRCL(1 + gain * AS_A_RF45_I_YAW_OUT_DQ). The optimal gain is similar to the one obtained with the previous analysis. However, the improved coherence is very close to one, as shown in the 9th plot. This is very efficient way to prove that indeed most of the SRCL coupling is modulated by the above angular signal.
Following up on the TTFSS work in alog 17885, we remeasured the input mode cleaner open loop transfer function.
The ugf is 50 kHz. With 3 dB of additional gain we can push this to 100 kHz and engage the second boost stage. The gain margin will be a little marginal at 1-2 dB, so.
For the measurement the common gain was at 18 dB, the fast gain at –6 dB, the compensation and one boost stage was engaged.
The small feature at 750 kHz is still there but of little relevance. This now closely resembles the iLIGO configuration and behavior.
Again, just because this is how it will be for the forseeable future, I post a more reader-friendly version with notes reflecting the above entry. Hope this helps!
The night before last, OMC injections suggested that our scattering problem did not involve a back-reflection from the OMC. While looking at the data, I noticed that the injections seemed to excite a 0.05 Hz oscillation (servo?) that did produce scattering up to 100 Hz. The time series in the figure show the 85-90 Hz band of DARM against vertical and horizontal OMC OSEMs. The longitudinal injection is between about 1400 and 1800s and is visible on the T3 OSEM but not on the vertical OSEM. The 0.05 Hz oscillation starts near the end of the injection. The level of motion that produced shelves to 100 Hz was not that much greater than normal motion. This suggests that we should try exciting the OMC in other degrees of freedom, in case the scattering is in another direction besides the main beam direction. I think vertical injections would be the place to start because vertical seems to correlate the best; perhaps light is scattering back from the shiny table top just below the upside down OMC.
Robert, Koji, Sheila
We checked the coefficient between the RIN and the PZT of the laser. We swept from 1 kHz down to 10 Hz, but we only got good coherence below 20Hz. The measured value for H1:PSL-ISS_PD[AB]_OUT (Volts) / H1:IMC-F_OUT (kHz) at 10 Hz was –40 dB. Dividing by 10 V (ISS PD DC level) and multiplying by the measurement frequency of 0.01 kHz we get
RIN/rad = 10–5.
The ISS inner and outer loops were off during the measurement.
The HAM6 Vent and Table payload changes caused a change in the Free Hanging Position of the Optical Table. This resulted in tripping of the ISI during Isolation, 17816, and subsequently, a resetting of the CPS Target Locations.
With a couple nights of DC Readout Locking, I've gotten the go ahead from DanH, Koji, Sheila, and Kiwamu. I've now updated the safe.snap and the HAM6 SDF is now dull again. Attached is the change to the Targets we've been running with for almost a week now.
MICH feedforward seems to be doing its job, although there is room for improvement by implementing a frequency-dependent subtraction.
SRCL coupling into DARM seems to be very nonstationary. Consequently, the feedforward is not working.
We injected band-limited white noise (elliptic bandpass, 10 Hz to 1 kHz, 6 ct amplitude) first into MICH, then into SRCL, to test the feedforward that was implemented a few weeks ago.
For MICH, frequency-independent subtraction is fair to middling (red) compared to no subtraction (blue); at best we get 20 dB of subtraction around 150 Hz. Note that the TFs in this plot use the whitened DARM channel. The whitening is undone for the spectra in the fourth pad.
For SRCL, the 1/f2 feedforward via ITMY L2 gives no subtraction at all. The attachment shows the TF of SRCL control → DARM with the feedforward off and with broadband noise injected into the SRCL error point. Unlike MICH, appearance of this excitation in DARM is highly nonstationary, fluctuating by a factor of 2 or so in a frequency-dependent way. Additionally, the coherence is poor above 20 Hz, despite the excitation elevating the DARM noise by more than an order of magnitude from 20 to 100 Hz.
The shape of the excess noise is more or less the shape of the 100 Hz elliptic cutoff that we put into SRCL a few weeks ago. Is it possible that the SRCL control noise explains the nonstationary, 100 Hz "scattering" shelf that we've seen in the DARM spectrum this past week?
Using the measurements described above, here is a projection of MICH and SRCL control into DARM. It seems that these two noise sources, along with DAC→ESD noise, can explain most of the DARM noise from 10 to 70 Hz. There is still some excess from 80 to 200 Hz, and an overall excess in the high-frequency noise floor.
For MICH, I used the coherent transfer function we measured earlier. For SRCL, I estimated the TF magnitude by dividing the ASDs of DARM and SRCL (after subtracting off their quiescent values). The dtt files are in evan.hall/Public/2015/04/FullIFO/Noise
as MichNoise.xml
and SrclNoise.xml
.
Some times (all UTC):
After these measurements, I also tuned the PRCL→SRCL subtraction in the LSC input matrix from 0.005 to -0.04 (using in-vac POP). This reduced the appearance of a 122 Hz PRCL excitation in the SRCL error signal by 20 dB.
For completeness, here is the same budget as above, with intensity and frequency noises included.
We suspect that the sharp shelf at 100 Hz in the frequency noise projection might be coupling via SRCL, rather than directly to DARM. So between the frequency and SRCL projections, there may be some double-counting of noise in DARM.
Frequency, intensity, and DCPD dark noise are not enough to explain the excess noise between 200 Hz and 4 kHz. It seems they can somewhat explain the uptick in noise above 4 kHz.
Slightly updated/corrected version attached.
Alexa, Dan, Elli, Evan, Koji, Sheila
We also locked with both end station beam diverters closed last night, we had no problems locking this way, but also saw no improvement in DARM. Now we are leaving the beam diverters at both end stations and OMC refl closed all the time. AS air and POP we have been closing sometimes in full lock, but we don't see a notiicable change in DARM. The AS air beam diverter has krytox, so we should feel free to close it whenever we want a low noise lock, but the POP beam diverter has not had krytox so we may want to avoid closing it without a specific motivation. We have not yet tried closing the refl beam diverter.
Sheila, Gabriele, Evan, Koji, Dan, Alexa, Elli
Yesterday Koji, Evan and Sheila locked at 15W, and after an hour and a half the lock became unstable and they reduced they power due to suspected PI. A spectrogram ('darmspectrogram3_1.png ') of last night's lock shows a 844Hz line grow in the DARM spectrum (see attached spectrogram from 2015-04-15 09:30:00 UTC). This corresponds to a PI at frequency 15540Hz, which is the first PI LLO saw (LLO alog 15934). The line appears about an hour after the 15W lock began and the power was reduced after 1hr25min. Once the power was reduced to 10W the line got smaller.
This evening we locked at 15W at 2015-04-15 23:40:00 UTC and we saw the same line grow at 843.4Hz, corresponding to a PI at 15540.6Hz (see spectrogram 'spectrogram16April_843HzPI.png '). This line grew untill we lost the lock at 1hr55 mins later ~ 2015-04-14 01:35:00 UTC (not due to the PI). We measured the DARM spectrum around 15.5kHz using the SR785. Attached is a plot '16AprilDARMSpectrum.jpg ' of the growing 15.54kHz line.
Grabriele saw that the 843Hz line was coherent between between the X-arm ASC QPDs and Darm, so we turned on the ETMx ring heater (see alog 17899). The ring heater is requesting 1W total, or 0.5W each on the upper and lower segments. We locked for a second time tonight at 15W at 2015-04-16 05:18:10 UTC. With the ETMX ring heater running, we didn't see any growth in the 15.54kHz line, as measured by the SR785. See 2nd plot '16AprilDARMSpectrum_withETMXRH.jpg '.
Measurement notes:
-Spectrogram generated at LigoDV https://ldvw.ligo.caltech.edu/ldvw using channel H1:CAL-DELTAL_EXTERNAL_DQ.
-Spectrum meaurement taken with SR785 connected to OMC DCPD readout. There is a script in /ligo/home/eleanor.king/netgib/SR785/SPSR785omcdcpds.yml that is used to take the spectrum (use the command < ./SRmeasure SPSR785omcdcpds.yml > to take a measurement).
-Dan points out there are also OMC DCPD 64kHz testpoints which are channels H1:IOP-LSC0_MADC0_TP_CH12 and H1:IOP-LSC0_MADC0_TP_CH13, which correspond to DCPD_A_INMON and DCPD_B_INMON.
How exciting! We have found that 0.4W per segment is best at present. At 0.55W there is another mode that rings up in a very short time, it appears around 1360, or 15004 in the IOP channels. You get a lot earlier warning looking at the IOP channels as mentioned, The IOP ASC_TR_ channel test points at are also nice as they have a lot less mess. Fitting this data as was done by Mathew Evans with in the PI observation paper results in a mechanical Q of 6.9M, the fit is not very good though with essentially 2 data points. Specifically the equation for the fit is τm = 2Qm / (ωm(Rm − 1)) τm - time constant of ring up, Qm - mechanical mode Q, ωm - mechanical mode frequency, Rm mechanical mode parametric gain = const * Power where 'const' is dependent on frequency overlap condition, spatial overlap, Q factors and some other stuff. So it assumes generally the only thing that is varied is the Power.
The numbers are relative to the power delivered to HAM6. We still need single bounce OMC lock data in order to infer the OMC throughtput.
Obviously the systematic error was larger than the statistical error noted...
Sheila, Gabriele, Elli
ETMx ring heater turned on at 15:04:16 1:56:00 UTC requesting 0.5 W in upper and lower segments.
The ring heater is on because it looks like we have been watching a parametric instability build up at 15540.5 Hz over the last hour or so of lock. There is coherence between the X-arm ASC QPDs and Darm at this frequency, which is why we are trying using the ETMX ring heater. There is a script in /ligo/home/eleanor.king/netgib/SR785/SPSR785omcdcpds.yml that we have been using to track the amplitude of this mode (use the command < ./SRmeasure SPSR785omcdcpds.yml > to take a measurement).
For some reason all of the gains on the ETMx upper and lower segments were set to 1, so I have changed them back to what they were in February (which is also what the ITMs are currently set to).
H1:TCS-ETMX_RH_SET UPPER&LOWER DRIVECURRENT_GAIN =12.5
H1:TCS-ETMX_RH_UPPER&LOWER VOLTAGE_GAIN =3.5
H1:TCS-ETMX_RH_UPPER&LOWER PCB_GAIN=3.5
H1:TCS-ETMX_RH_UPPER&LOWER CURRENT_GAIN=-0.08
I've also changed the resistances for all of the RH segments (H1:TCS-ETMX_RH_UPPERRESISTANCE etc.) to the resistance values Aidan meausred (see alog 16655).
I added one additional ADC to the x1lsc0 and x1oaf0 chassis. This completes the 3IFO IO-Chassis inventory.
Following up on the IMC alignment improvement described earlier, I took a look at the residual fluctuations in the IMC transmitted RIN. Indeed, the RIN is much better on average when the correct IMC alignment offsets are used, but there are still quite large fluctuations. These are very well correlated to the residual angular motion, as visible in the IMC ASC error signals.
To quantify this phenomenon, I computed the band limited RMS of the ISS signal in the 100-400 Hz band and computed my usual linear regression analysis using all the IMC WFS and QPD signals. The first attachment shows that all the residual RIN fluctuations are predictable based on angular motion. The most relevant signal is WFS_B_I_YAW (see second attachment). Looking at the spectrum of the signal, it is quite clear that the low frequencies are dominated by a wide and smooth shoulder, likely due to input beam jitter created by air currents in the PSL room.
If this hypothesis is correct, we could implement an additional control loop, that servos the input beam to WFS_B. Since the DC of the input beam is already servoed to MC2_TRANS and since the DC of WFS_B is already used in DOF1 and DOF2, this additional loop must be AC coupled.
[Alastair]
I'm turning the Y-arm CO2 laser on to test a new script that sets the rotation stage position. There is a beam dump in place on the table, so the output will not reach the CP.
Laser turned off at 17:07 local time.
Testing of script:
Power request Measured power
300mW 0.3009W
400mW 0.3996W
200mW 0.1991W
600mW 0.5996W
I've taken Amplitude Spectral Density measurements on all of the suspension coil drivers that utilize the quad monitor bd. Below are screenshots of the data from the ones that I think may require some investigation. This data will be commited to SVN.
In list form, these are the suspects: ETMX L2 UR -- (State 1 too large) BS M2 UL, UR -- (State 1 & 2 too large on UL, No change in noise between states for UR) ETMY L1 UL, LL, LR (No change in noise between states) PR3 M3 LR -- (No change in noise between states) PRM M3 LR -- (No change in noise between states) SRM M3 UL, LL, LR -- (No change in noise between states) SR3 M2 UL, UR, LR -- (UL close to ADC noise, UR, LR no change in noise between states) SR3 M3 UR, LR -- (No change in noise between states) Ed didn't mention, but these measurements are not calibrated, i.e. they're in units of ADC counts of monitor board voltage (hence the comparison between ADC noise in counts). To calibrate into current across the coils, one needs to invert (divide by) the response of the noise monitor circuit (D070480), zpk([0,0,0,0],[5,5,5,5,4.8e3,4.8e3],196) which takes you to, Vout, the voltage across the differential legs of the output op-amps, before the output impedance network of each driver, then you need to divide Vout by the output impedance network and the coil, to get the current across the coil, Ic, Ic = Vout / (2*coilDriver.Zout + osemCoil.Zc). Note that these are suspects because we've simply played the "this performance doesn't look like the others" game. As the list of suspects is smaller than "all of the noise monitors," we can now focus our attention here so it's less overwhelming to get answers to questions like - is it right? - why doesn't it look like the others? - is it the driver response itself or the monitor? etc.
This morning and early afternoon I worked on the IMC alignment. My goal was to reduce the coupling of input beam jitter to intensity noise in transmission of the IMC. In brief, I dithered the input beam with the PZT at 80 Hz in pitch (amplitude of 3 cts.) and at 110 Hz in yaw (amplitude of 2 cts.) and looked at the ISS second loop power. I wrote a python script to demodulate the ISS second loop signal at tghose two frequencies, so that I could produce two error signals. They turned out to be very sensitive to the IMC alignment.
My first attempt was to move the beam on MC2_TRANS QPD in order to minimize the jitter to RIN coupling. Practicallt, I moved the QPD offsets to zero the error signal produced as explained above. The script I wrote implemented a slow servo to do this automatically. As shown in the first plot, this worked fine: this is a comparison of the RIN before and after the adjustment of the beam position. RIN is a factor 10 lower than before almost everywhere below 200 Hz.
Unfortunately, on a long timescale the offset servo is diverging, as shown in the second loop. The reason seems to be some interaction with the IMC ASC loops: moving the beam on MC2_TRANS adds offsets on the WFS, and then the IMC ASC loops respond slowly. The result is somehow drifting away in DC. So this is not a good soluition.
Duiring my previous attemps I found out that the IMC alignment was responding incredibly slowly to my action. Therefore I estimated the loop bandwidths by measuring their step response time constants:
Pitch | Yaw | |
---|---|---|
DOF1 | 40 s | 80 s |
DOF2 | 5 min | 4 min |
DOF3 | 30 min | 30 min |
Clearly they were too slow, so I increased all gains to have a bandwidth of about 50 mHz for all of them. To doi this while the loops were closed, I changed the input matrices, as shown in the third attachment. I checked that after this modification all loops have indeed a step response of the order of 20 seconds. There is however a very larg cross coupling of all loops, confirming the result explained in the previous section.
Although this is a less clean solution, I tried to minimze the intensity noise by acting on the DOF_1 and DOF_2 offsets, or in other words of the WFS_A/B offsets. It turned out that the best choice is to act on DOF_1_Y and DOF_2_P, since this is the combination that effectively zero the error signals without affecting significantly the IMC transmitted power. I adapted my script to servo those two offsets to move the error signals to zero. The result is shown in the fourth attachment.
You can use the attached script, provided that you first switch on the two following dither lines:
H1:IMC-PZT_PIT_EXC ampl. 3 frequency 80 Hz
H1:IMC-PZT_YAW_EXC ampl. 3 frequency 110 Hz
The coupling is still fluctuating quite a lot, expecially for the yaw degree of freedom.
I left some offsets in the two degrees of freedom, as found by the script servo: DOF_2_P = 135.9, DOF_1_Y = 47.5
Isn't the bandwidth f_BW = 1/(2pi tau)? Meaning, more like 10mHz with a 20sec response time.
Continuing on from yesterday's TTFSS work. All notches were removed, and the transfer function was measured (see NONOTCHS.tif). The peak at ~760 kHz is of no big concern. However the peaks at around 1.77 MHz are. These were notched out with C50 (1-65 pF) and series with L2 (220uH). C50 was adjusted to suppress the two peaks. The transfer function was re-measured (see OLTF.tif). The unity gain frequency is around 450 kHz. Currently the common gain is set to 27 dB and the fast gain to 5 dB. Should the loop break out into oscillation, reduce the fast gain to 0 dB, wait until the oscillation stops and then increase it back to 5 dB. Rick, Peter
I looked at the FITS estimate for the optocoupler. The demonstrated FITs for the part is < 5555. Which if I understand things equates to a mean time to failure (MTTF) of just over 20 years. In a nutshell, I do not know why the part failed. Let alone two of them.
Because this looks to be the "final" configuration for a while, I again post an annotated version of the open loop gain transfer function for the PSL's (table top) FSS. In summary, - The unity gain frequency is 449 [kHz], with a phase margin of 58 [deg]. - The feature at 1.8 [MHz] (believed to be a resonance in the EOM) has been notched out, but with a pretty skinny / high-Q notch. - This [creates a new / pushes up the] feature a little higher in frequency, but the gain margin appears to be OK for now. - The 770 [kHz] feature is one of those "lucky" resonances where the zero is before the pole in frequency, so we win phase instead of losing it. So, there's no need to compensate for it, there's plenty of phase margin, and we aren't going to do anything about it. - In the future, we'll install a lower Q / wider notch to completely suppress this feature entirely without creating any new features, but at this time that's been deemed of lower priority.
Koji, Evan, Sheila
Tonight we saw the shelf from 50 up to 100 Hz, however we are back to a range of around 35 Mpc. We saw this last week and hoped that it was related to the HAM6 cleanroom. We have done several things:
The first plot was measured with the injection of 20000cnt@0.2Hz to H1:SUS-OMC_M1_TEST_L_EXC.
The second plot was measured with the injection of 10000cnt@0.2Hz to H1:SUS-OM1_M1_TEST_L_EXC.