I reduced the 9MHz modulation depth by 6dB, and it seems like that gives us several more Mpc. It seems like our sensitivity is really improving when I reduce the modulation depth, although I'm not sure why it has such a significant effect, particularly at high frequencies.  I plot here also 3 of the calibration lines, so you can see that their peaks are lining up pretty well.  If anything, the green and brown traces with the 9MHz at it's lock acquisition value are a bit worse than they look here, since the 1080Hz line should be scaled up by a teeny bit.

Note that the lockloss around 16:45 was me, trying to reduce the 9MHz by another 3dB, but the old script that I use to step by hand further didn't include compensation for the analog CARM gain.  I've fixed the script, so will likely try again next lock.

Rana, Hang

We retook the nonlinear SRCL noise coupling measurement. In our previous study without SRCLFF (LHO:44770), the linear component was too high to expose the nonlinear part. This time we inject SRCL noise with both the MICH and SRCL FF on to reduce the linear component (thanks Sheila et al. for performing the injection; LHO:44772) and, as a result, now we are able to expose the NL couplings.

Please see the first attached plot for the SRCL noise projection. The blue trace is the DARM ASD, the orange is the projection of SRCL based on excess power (which captures both the linear and NL coupling), and the green one is the SRCL projection based on pure linear coherence (i.e., the residual linear coupling after the online FF).

It seems that the SRCL noise is the dominant noise source in the 20-40 Hz band, and is significant all the way up to ~ 300 Hz. Below 30 Hz, the SRCL coupling is still mostly linear from the FF residual, whereas in the 30-80 Hz band the nonlinear coupling dominates.

The ratio between the NL/L PSDs projected onto DARM is shown in the second plot. In the 30-80 Hz band, the NL coupling is typically a few to 10 times higher than the residual linear component in power.

===================================================================================

The data we used for generating the plots started from gps 1224400998 for 1024 sec. The excess power projection also used data from gps 1224399618 for 1024 sec with active SRCL noise injection (both MICH and SRCL FF on).

[Jenne, Gabriele]

We might be limited by intensity noise that we can't sense with the ISS first loop diodes. Now that we got your attention, see the details below. The most interesting plot is the last one.

This morning we checked whether the ISS second loop was open. The output switch of the second loop board (H1:PSL-ISS_SECONDLOOP_OUTPUT_SWITCH_MON) was indeed open, but the input switch on the first loop board was closed (H1:PSL-ISS_SECONDLOOP_CLOSED). So we decided to open it.

We were surprised to see that the sensitivity got significantly worse when the switch was open. See below (H1:PSL-ISS_SECONDLOOP_CLOSED = 1 means switch closed, first loop closed but second loop still open at the output of the second loop board, H1:PSL-ISS_SECONDLOOP_CLOSED = 0 means switch open at the input of the first loop board, but the first loop is still closed)

At the same time we see that the first loop signals (PDA is out-of-loop and PDB is in-loop) also changed. The in-loop signal did not show any difference, while the out-of-loop signal noise floor increased by about a factor of ten. See below (top panel is again DARM, bottom panel is the in-loop signal (dashed) and the out-of-loop signal (solid).

So it looks like the noise in DARM increased at almost all frequencies by about the same factor as the noise seen by the ISS first loop (out-of-loop) PDA sensor.

However, coherence is very poor between DARM and PDA/PDB, even when the noise is high.

It's however clear that the noise increase in DARM is related to toggling the ISS input switch (we tried many times).

Spectrograms of DARM and PDA (out-of-loop signal) in the quiet and noisy periods show that the noise is highly non stationary, so maybe this is enough to explain the lack of coherence.

So the next question is how close is the intensity noise (as measured by the ISS first loop out-of-loop signal) to limiting us in normal condition?

To answer, we computed the ratio of the DARM spectrum over the PDA spectrum while in the noisy state, and use this ratio to project the quiet state PDA spectrum into DARM. We restricted the projection to only those points where the DARM noise got higher by more than a factor of two when in the noisy state. The result, shown below, seems to indicate that we are likely limited by "intensity" noise measured by the ISS first loop sensor, which couples in a non-stationary or non-linear way, so we don't see coherence.

Next steps:

• understand why the noise in the first loop signals got worse with that switch open (maybe a dangling connection?)
• understand if the noise at that input is limiting the ISS first loop performance even when the ISS second loop switch is closed
• check if other PSL signals got worse (frequency noise?)
• do some intensity noise injections, to make sure we see the same with "real" intensity noise

In breif, I measured, fitted and retuned the MICH and SRCL feedforward. The results are good, but the first SRCL FF filter I tries showed the same instability reported in 44740, probably due to the "large" gain below 10 Hz that's neede to have a good fit down to 10 Hz. So I restricted the fit above 30 Hz, and got a new filter, which should have good performance in the 30 to 100 Hz region. I haven't tested it yet, since the IFO unlocked.

Measurement templates are attached, since I don't have permissions to copy them in /opt/rtcds/userapps/release/lsc/h1

The plot below shows the DARM spectrum without any feedforward (red), with MICH only (blue) and woth MICH+SRCL (black).

The configuration that gives the good MICH filter and the (tentative) good SRCL filter ois shown in the last attached screenshot.

The h1seih23 and h1seih45 computers had their dackill glitches again.  This is what killed the several hour lock at 22W that was left overnight.  I requested the HAM 2,3,4,5 HEPI and ISI guardians to "ready", then ssh'ed to the computers and ran the startWorld script, as in alog 44672 and alog 44695.

Also, just in case some settings get missed, I took the SEI_CONF guardian to WINDY_NOBRSXY and then back to WINDY_NOBRSX, to force any sensor correction settings to be put back.  The SEI_CONF medm screen made it look like everything was in its expected state, and all of the lights (except end X due to BRSX issues) were green, but since this was a problem last week, hopefully this ensures all sensor correction is properly turned on.

Rana, Craig

Today we calibrated the CARM servo board spectrum into Hz.

First, we drove a strong line in the MC2 suspensions at 71.1 Hz, and measured the transfer function from REFL_SERVO_ERR_OUT_DQ (we assume counts from the common mode servo board error signal) to IMC_F_OUT_DQ (calibrated into kHz).
We got 3.1e-7 kHz/cts, or 0.5 Hz/V, at 71.1 Hz, assuming an ADC of 2^16 cts / 40 V. 
Assuming a CARM pole of 0.5 Hz, this gives a CARM optical gain at DC = 284 V/Hz.

We've also measured the CARM OLG, giving a UGF of 10 kHz.  
By extrapolating the CARM OLG with a 1/f^2 line, we can approximately calibrate CARM into Hz of frequency noise incident on the interferometer:


where 
 is the CARM OLG, 
 is the common mode boost (z=4kHz,p=40Hz), 
 is the REFL9 sensing chain,
 is the CARM plant, and 
 is the common mode board test1 point.

The calibrated spectrum is pretty flat, quickly reaching an RMS of 2 Hz at high frequencies.  This RMS frequency noise seems too high to me, and the spectrum doesn't have the shape I expected.

Sheila, Georgia, Craig, Hang, Rana

We have a calibrated DARM spectrum that we can compare to the noise during O2, and it looks like our noise from 40-100 Hz is very similar to the noise floor after the Montana earthquake last July (ignore the large injection just above 70 Hz).  The first attached screenshot shows the noise before and after the earthquake and now for a comparison. The second screenshot shows the coherence of DARM with ASC and LSC signals, none of which are high above 35 Hz.  

Done tonight:

• We removed the changes to the ITMX ESD filtering from the guardian (that doesn't matter for feedforward). 
• We changed the guardian to engage the new MICH FF which Gabriele made earlier (44767), and the SRCL FF that Craig designed last night (44744).  We haven't tried Gabriele's SRCL FF yet.  
• We also removed the ISS second loop from the guardian because it was causing locklosses. 
• Started noise budget injections: IMC PZT P+Y, CHARD P+Y, DHARD P+Y, MICH P +Y (very small compared to O2), LSC MICH +SRCL
• Started a SRCL excitation for Hang at midnight (7 UTC) exactly, we turned it off at 7:19 UTC.
• At 7:20 UTC we are leaving the interferometer undisturbed.  

Craig, Rana

We checked the RF levels at the REFL9 demod board RF MON (-23 dB coupled output) with 50 Ohm terminated oscope.

So to convert this trace back into the input of the demod board we multiply by ~14.

From the scope trace, we can see that the fastest slew is ~140 mV / 10 ns (10 V/us). This is not a problem for the FET mixers on the board.

The slew rate limit of the RFPD output amp is 450 V/us, so we are from hitting the hard slew rate limit. The question that remains is how close can we be before there is some non-negligible downconversion.

Thinking back to the iLIGO days, we used to have trouble with the MAX4107 (450 V/us slew rate) when it was putting out ~150 mV-pk at 2*25 MHz (~30 V/us).

So I think if the LMH6642 distorts at the same fraction of slew rate as the 4107, we are close to seeing this excess noise. To be sure, we'll need to do a more careful and systematic test with some spare RFPD and demod electronics. Tricky.

[Georgia, Patrick]

We ran the usual charge measurements today - the quadrant-by-quadrant effective bias voltage measurements on both ETMX and ETMY, and the additional measurements of combinations of the four parameters on ETMX (measured by driving combinations of the signal and bias electrodes).

The ETMY effective bias trend remains consistent with ~50 V offsets on the UL quadrant (coupling to pitch), the UR quadrant (coupling to yaw), and the LL quadrant (both pitch and yaw). The first attachment shows this over the last few months.

The ETMX effective bias is trending back towards zero after the bias was switched on September 11. The second screen shot shows the quadrant-by-quadrant measurement for ETMX.

This trend is also seen in the four-parameter measurement which makes sense. The third attachment shows the alpha and gamma parameters in the first column, these are associated with the ESD actuation strength and should not change over time. The second column is the beta+beta2 parameters, as measured by driving the bias and measuring the response on the op lev, and by driving the signal electrodes and measuring the response on the op lev. This measurement should be immune to space-charge polarisation which we assume is symmetric. The bias-driven beta+beta2 seems to have taken a jump recently (though I don't want to jump to conclusions based on two data points), while the longitudinal signal drive did not. This is strange. The third column shows beta-beta2 and the Veff, which are sensitive to space charge polarisation and consistent with the quadrant-by-quadrant measurements.

Note when running these measurements - it can be hard to tell if the HV is on or off. Even with the HV off a small signal is seen in the ESDAMON channels whin driving the quadrants. These channels are calibrated into volts and should see a large excitation from the charge measurements (100s of volts, exact numbers depend on the measurement). Up until today the ETMY L3 ESDAMON and LVESDAMON filters were turned off, leaving those channels in counts, not volts. I have switched those filters on now.

J. Kissel, S. Dwyer, E. Goetz, L. Sun

Lilli Shiela and I were able to successfully update the calibration of DELTAL_EXTERNAL today. The residual systematic error is no worse that 15 % and 7 degrees (likely due to a slightly wrong coupled cavity pole, and slightly wrong actuator strengths, and slightly wrong sensing vs actuation relative delay). Recall that this calibration is informed by quick, limited frequency band, measurements, so I'm happy that the results are even this good.

The big gotcha that took us so long: the MEDM screens (recently upgraded by Jeff) do not correctly reflect what's happening in the front-end in each actuator replica's ESD path. Namely, with the replica ESD linearization bypassed, the replica BIAS bank has no influence on the ESD path (this was not Jeff's impression of how it should be, thus he didn't update the MEDM screen that way). Thus, when we had installed the correct replica bias -- and it just happened to be negative -- it did not affect the replica signal path. Thus -- unless we change the model in the future to be more reflective of real life -- we need to apply the bias voltage *sign* and only the sign somewhere in the replica control path. Combined with several other sign flips between the O2 configuration of interferometer control and the current interferometer control (a different arm, a different output matrix, a different bias sign), this was very difficult to find (which is why we needed Sheila's new set of eyes to help figure out the problem).

I've accepted all of the settings in the CAL-CS model as they stand now, so we should have this good(ish) calibration permanently.

I've confirmed the updated calibration and it's systematic error with a PCAL to DARM sweep and an DARM Open Loop Gain, which can be found here:
/ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/O3/H1/Measurements/FullIFOSensingTFs/
    2018-10-23_H1DARM_OLGTF_23to800Hz_10min.xml
    2018-10-23_H1_PCAL2DARM_TF_23to800Hz_2min.xml
they've been exported for future processing by Evan and Lilli. These measurements and their exports have been committed to the above location in the repo.

The attached 2 screenshots show two things:
 - The PCAL to DELTAL_EXTERNAL transfer function (left two panels of first attachment), which quantifies the residual systematic error in the calibration
 - A calibrated (to the accuracy quantified above) amplitude spectral density of DELTAL EXTERNAL.

There are plenty of qualifiers on the noise of this ASD -- it's in the middle of tonight's round of commissioning. The biggest know problem is that they're actively working on tuning the LSC feed-forward, so there is no MICH, PRCL or SRCL feed-forward on during this ASD. Please focus on the fact that the PCAL lines (in green X's) at 35 and 330 Hz are matching the LIVE DELTAL_EXTERNAL (in red).

I determined that something was amis after having replaced BSC10's AIP pump body this morning and, so, swapped the controller out as well.  Something still isn't working - I don't believe the pump current data even though the absolute value is plausible.  I'll continue to investigate at the next opportunity.  Until then, I am leaving the AIP energized but have shut down the pump cart and local turbo.   

Terry Daniel

Added a Thorlabs FLH1064-8, a FLH532-4 and a FLH532-4 in front of the DCPDs which monitor the IR laser light, the SHG green light and the green light launched into the fiber, respectively. This blocks the ambient light and makes us insensitive to the table lights.

We tried to do some DARM noise budgeting based on its coherence with the auxiliary degree of freedoms. The data we use for this analysis are starting from GPS 1224301218 and for a duration of 1024 sec. The auxiliary loops we consider so far includes LSC MICH/SRCL/PRCL/MCL, ASC CH/CS/DH/DS P/Y, as well as the ones we used for jitter noise subtraction during O2 (PSL-DIAG_BULLSEYE_PIT/YAW & IMC-WFS_A/B_DC_PIT/YAW). We only focused on the freq range from 10-100 Hz as this should be the band where the auxiliary DOFs' coupling being most significant.

In the first attached plot, the blue trace is the raw DARM output, the orange trace is the one after we removed all the linear coherence with the aux loops we considered (computed based on the f-domain MISO coherence; the subtraction is also done in the f-domain, i.e., the subtraction is assumed to be ideal), and the green trace is an O2 H1 reference before the Montana earthquake. There exists some calibration uncertainties and we simply scaled the current and O2 DARM to match at 100 Hz. Nonetheless, it suggests that we have some excess noise below 60 Hz compared to the O2 ref, and those noise cannot be solved by tuning LSC FF nor A2L.

In the second plot we show the noise projections from the LSC auxiliary loops, in the third plot the projection based on the ASC loops, and the last plot the projection based on the jitter channels.

For the LSC, coupling right now is dominated by SRCL (and its slope is much sharper than 1/f^2 as I would naively expect due to DARM detuning...).

For the ASC, we have large DS_P coupling because we are trying to suppress the 0.5 Hz oscillation. However, as we have SRC ASC controlled it seems unnecessary to have such a high BW DS_P loop. The noise above 20 Hz is then dominated by CH_Y and DH_Y. For DH_Y we have some margin and should be able to improve the LP relatively easily. For CH_Y, however, as we kept losing phase for that loop at ~ 3 Hz the optimization might be challenging... Need to think more carefully how to handle it.

For the jitter noise, we do not see significant couplings to DARM.

===============================================================

In principle the all the linear coherence can be subtracted out and should not contaminate DARM. Ideally the subtraction should happen online, but in the worst case we can still remove them from offline regressions (as we did for this study). Thus those noise should be the easy ones to tackle.

The subtraction residuals, if not explained by the coherence with aux dofs we haven't considered so far, are probably from nonlinear couplings and would be harder to remove. That would be the hard part of the commissioning work. From the current study, we do see excess noises in the <60 Hz region that cannot be explained by the linear couplings between the major LSC/ASC loops and DARM, and thus we might need to start thinking about the origins and solutions to those noises.

Sheila, Hang, Craig, Georgia, Danny, Thomas

• Hang lowered the gain in MICH P to avoid the possibility that it was unstable at 1.1Hz 44739.  We later realized that the locklosses we've been having where MICH P rings up at 1.1 Hz were actually caused by the SRCLFF, not by ASC
• We left the SRC ASC on as we increased the power, and this seems fine so it is now in the guardian.  
• We also added the ISS second loop back into the guardian. 
• In the low noise length control state there were some gain changes that didn't make sense when we looked at the OLG measurements, so we have commented them out.  This leaves us with a SRCL ugf of 35 Hz, PRCL at 60 Hz ,and MICH at 9.5 Hz. 
• We engaged the MICH FF and it still seems to be doing a good job of reducing the coherence.  
• We found that the SRCL FF we have is unstable, we made the measurements to redo the fitting.  We saved templates in userapps/lsc/h1/templates  Craig is working on fitting and has a filter that is ready to try
• We ran Gabriele's A2L script, it worked and reduced the coherence with ASC. 
• We lost lock for an unknown (fast) reason at 3:07 UTC, 4:15 UTC, 5:36 UTC (may have been a small EQ) and 6:09 UTC
• ETMY mode1 gets rung up in about 1 out of every 5 locklosses.  I am not sure why this mode in particular should be rung up, it is not at the edge of any of our notches.  The filter used for damping is narrower, so it might ring longer on locklosses than the wider filters do. 

Related to WP 7881, modified and recompiled the PSL ISS model, to change the name of a filter bank from PSL-ISS_SECONDLOOP_REFERENCE_DIFFRACTION_CAL to PSL-ISS_SECONDLOOP_REFERENCE_DFR_CAL

Modified accordingly the MEDM screen in /opt/rtcds/userapps/trunk/psl/common/medm/ISS/PSL_CUST_SECONDLOOP_SLOWOFFSET.adl