Turned ON QDP80 roughing pump for a couple of hours along with the purge air skid to test in preparation for next week's vent. Purge air dew point measured -43degC.

J. Kissel, J. Driggers

In prep for the ISC triggering shut offs, as well as a change in configuration to the dolphin network, all suspension models will need to be restarted today. As such, I've reconciled all suspensions safe.snap SDF files. I attach a full detail of what I've accepted / reverted, but aside from capturing the latest alignment offsets, the only suspensions that had major changes that needed reconciling were the QUADs -- ETMX, ETMY, ITMX, ITMY. With the help of Jenne, we did our best to accept or revert changes on those suspensions that were related to the past ~week's worth of commissioning.

Terry, Daniel, Haocun, Nutsinee

After running into several problems of EOMs not working as they should, we finally installed a New Focus 4002 EOM in the green path with half wave plate at the input and output rotated 45 deg.  The whole set up is to use EOM as an amplitude modulator so we can inject intensity noise into the OPO loop. We have always had a lot of unexplained noise and we're hoping intensity fluctuation+crystal interaction could explain at least some of them.

We hooked up SR785 swept sine excitation to the EOM while OPO is locked. Got about 7% of intensity modulation. TTFSS I-MON goes to CH2 and SHG launch diode readout (monitors what goes to the fiber coupler) goes to CH1.

This data is then multiplied by OPO open loop gain and calibrated into Hz/mW 

(V_ttfss/V_GRPD) * (FWHM Hz/ I-MON PDH Vpk) * (GRPD Transimpedance V/A) * (GRPD Responsivity A/W) 

PD gain and dual DC photodiode amplifier board transfer function are included. 

The diode response "pole" is around 8kHz, which is the turn around point of the plot above.

(A fun fact, the response of the EOM to swept sine signal drops as you go towards DC)

If we were to repeat this transfer function with OPO locked, we got a tail biting effect going on as well. 

Note that the phase here could change depending how much power is going to the temporary EOM we installed. When the rejected (wrong polarization) diode reads about the same as the launch diode the phase above can flip 180 deg. This is likely due to EOM crystal changing the output beam polarization depending how it's heated. With this unreliable response we concluded that the A/EOM set up is no good for power fluctuation control.

Moving on.

Now that we have Hz/mW calibration, we can use the launch diode to figure out how much phase noise do we expect LO to see (calibrated into mW then multiply by Hz/mW).

Note that at 100Hz the frequency noise is almost as bad as Mephisto free running (100Hz/sqrt(Hz) at 100Hz). I pointed out the 1.3kHz line because it also shows up in OPO spectrum, and it is what killed our lock when HAM6 was in air (the peak was much worse in air). This line is to be investigated. 60kHz line may or may not matter. Could it be a laser PZT resonance (?). I will be posting more noise budget (OPO, LO) in a separate log. 

Now convert the plot above into radian and calculate the rms.

It's 1 rad /sqrt(Hz) at 344Hz and doesn't go down to a mrad until 39kHz. The LO loop have to have suppression of 40dB at 100Hz and a UGF of at least at 39kHz to keep sqz phase noise down to mrad level (note that this is all green projection so there's a factor of 2 down, but that still doesn't help). This is why using a PFD and slow path signal to OPO PZT alone didn't work. Currently we feed the fast path signal to TTFSS additive offset which gives LO loop a UGF of ~50kHz.

A BruCo scan for the period listed in 45013 is available here. There are a lot of interesting coherences in the low frequency region (10-30 Hz), some of them already identified and corrected (see 45036). None of them is near one, but they are all not neglegible and very likely enough to explain most of the noise when summed:

• SRCL (figure 1)
• SRC2_P (figure 2)
• CHARD_Y (figure 3)
• MICH (figure 4)

A couple of channels are intriguing, and I don't have a good explanation for them

• ASC-Y_TR_A_NSUM (figure 5)
• POP_A_LF_OUT (figure 6): shows a straight line contribution which seems very close to 1/f^4, and it runs parallel to the measured sensitivity... It would be interesting to check if this coherence is constant over time or non-stationary
• SUS-IM1_M1_VOLTMON_LL_OUT_DQ: (figure 7) between 20 and 70 Hz this channels shows a coherence which puts is very close to the sensitivity. What is this? Some input beam jitter maybe? Interesting, only LL shows coherence, while LR, UR and UL do not show any coherence at all (figure 8 is an example). I wonder if those channels are even working...

TITLE: 11/06 Day Shift: 16:00-00:00 UTC (08:00-16:00 PST), all times posted in UTC
STATE of H1: Preventive Maintenance
OUTGOING OPERATOR: None
CURRENT ENVIRONMENT:
    Wind: 4mph Gusts, 2mph 5min avg
    Primary useism: 0.02 μm/s
    Secondary useism: 0.31 μm/s
QUICK SUMMARY:

• h1 was at NOMINAL LOW NOISE since 3:48utc & went down at 16:01utc
• Observatory MODE:  Taken to PREVENTATIVE MAINTENANCE
• ISI Config taken to SC OFF NOBRSXY
  • H1 was locked, but during transition PR GAIN & POP_A started to oscillated (& heard verbal alarms for ETMx, ETMy, ITMy).
  • Then we lost locked about a min after transitioning.

We did two rounds of MICH to DARM feedforward tonight by actuating on the ETMs as opposed to the ITMs.

We did the usual measure MICH_OUT to DARM_IN1 with FF off, then measured MICHFF to DARM_IN1, divide them and fit a zpk to them (via Lee's iirrational).  That gave us the result in PDF 1, and the orange lines in PNG 1.  This filter is named "Nov5" in the MICHFF filter module.
The coherence was largely reduced, to be sure, but the performance was mediocre between 20 and 35 Hz due to bad phasing.  

So we iterated on the fit using Jenne's convenient feedforward paper.  We retrieved the current filter in the MICHFF filter module using my filterModuleStateExtractor (call it alpha), and found beta = 1 - (DARM IN / MICH OUT)/[alpha * (DARM IN / MICHFF)], the iterative filter TF.  This gave us the fit in PDF 2, and is the red line in PNG 1.  This filter was implemented as "Nov5b" in the MICHFF filter module.
The coherence was reduced further in the 20 - 35 Hz band.  

This process was useful because the first round of the feedforward reduced the noise sufficiently to give a clean TF of the high-Q notches we see around 18 and 25 Hz.  I figure MICH to SRCL to DARM accounts for most of the remaining coherence, we haven't redone the SRCL to DARM feedforward to the ETMs.
The change from ITM to ETM actuation and MICHFF FM engagement is updated in LOWNOISE_LENGTH_CONTROL.

The templates for the MICH FF measurements are:
/opt/rtcds/userapps/release/lsc/h1/scripts/feedforward/MICH_to_DARM_excitation.xml
/opt/rtcds/userapps/release/lsc/h1/scripts/feedforward/MICHFF_to_DARM_excitation.xml

The results for the DARM IN1/MICH OUT and DARM IN1/MICHFF are actively read by the script in /ligo/home/craig.cahillane/Git/Feedforward/MICH2DARMfeedforward.ipynb.  

Just a quick note in case these show up again:

During one of our locks this evening a number of lines showed up in DARM between 18 and 28 Hz (blue trace, 1225511648) and stuck around for ~half an hour. Their amplitude appeared very stable. We tried to track them down in LSC and ASC signals but didn't find anything convincing.

I turned off the violin mode damping, and the lines (gradually) went away (red trace), when I turned the violin mode damping back on the lines did not return. Not sure if this was because of the violin mode damping or something else. (note: the extra broadband noise in the red trace is due to the feed-forward tuning that was going on in parallel).

There have been a lot of things happening today, most of which is other logs, but I thought it would be useful to make a summary:

• DRMI locking:
  • This morning Hang undid the filter change that he made to the MICH low pass a month or so ago.  Hang made the plot which is the first attachment and shows the lock acquisition times before and after he made his change. The vertical red lines show percentiles of the locking time.  
• Beam dumps:
  • After locking was recovered, Jenne and I added beam dumps for the beams transmitted by the AS AIR and POP AIR beam diverters 45010  More work is needed to dump the popair beam well.
• ASC cut offs:
  • I asked Hang to look at why CHARD Y noise was higher in our noise budget than it was in O2, he changed the low pass so that CHARD Y is now much lower in our noise budget 45023
  • Jenne noticed that SRC2 P had high coherence with DARM, and did noise injections. 45026  Like AHRD Y, this was a significant contribution the the DARM noise budget up to 30 Hz until Hang added roll offs 45032
  • Hang also reduced the gain of SRC2Y and added a cut off there.  This is only engaged in the LOW_NOISE_ASC state because it causes instabilities if used earlier. 
  • We took new noise budget injections for SRC2 P, Y, and CHARD Y after these changes were made. 
• LSC noise:
  • Stefan checked the input matrix element from POP9I to SRCL that is used to cancel the PRCL signal in POP45I.  45034  As we suspected, this coupling seems to be changing from lock to lock. 
  • Stefan also lowered the SRCL gain and put this in the guardian. 45035, then we redid the SRCL injection for the noise budget. 
  • We redid the SRCL and HAM3 noise injections for the noise budget after this work was done. 
  • Craig made a comparison of measurements Jenne and I had from O2 with ones he took recently of the MICH to DARM coupling with and without feedforward, 45029 the main message is that our coupling to DARM is the same but our feedforward performance has been worse. 
  • Gabriele posted an interesting model of how the SRCL FF can be unstable, which is why we need to AC couple it which limits how accurate our feedforward can be.  45030
  • We moved the notches for the beam splitter bound and roll modes into the LSC filter bank, rather than the suspension filter bank.  This was to avoid having them in the feedforward path which makes fitting the feed forward more difficult. We set up the DRMI guardian to acquire with the notches still in the suspensions, so that there high Q filters can be after the triggering rather than before, then after DRMI has acquired we swtich it to the LSC bank.  The real motivation for doing this is that I want to be conservative about changes to DRMI acquisition.  
  • Craig has measured and done fitting to retune the MICH feedforward, sending the feedforward to the DARM actuator rather than the ITMs.  The first round resulted in a similar level of suppression as the existing feedforward, so Craig is now doing a recursive fitting.    Craig was able to reduce the coherence using a recursive fitting 45038
• Georgia found possible down conversion of violin damping 45037

A noise budget including all of these improvements is attached, there is a trace here for HAM3 coupling as well as SRCL and MICH, but there is a risk that we are double counting now because the HAM3 noise couples to DARM through MICH (and SRCL). The last attachment is a comparison of the DARM spectrum tonight to Friday night in grey.  The improvements today have brought us to the early O2 reference below 25 Hz.  

Peter, Hang

One brute-force way of reducing the LSC MICH noise is to reduce its loop gain, and then we can move the LP filters to lower freqs. To achieve so we need to know the LSC MICH noise budget at low freq (which dominates the rms) and see if it is possible to reduce the input noise to the loop.

In the first plot we show the noise projection to MICH ctrl output based on f-domain MISO coherence analysis. The low-freq rms is largely dominated by the BS local damping noise (mostly from L, with a non-negligible contribution from P and Y). As a comparison, the largest ISI contribution (from BS ISI itself) is almost a factor of 10 below the BOSEM noise.

Since LSC MICH is already controlled with a clean, interferometric signal, it might not be necessary to have high gains on the local damping loops anymore. Thus we might want to reduce the local damping gain and thus reduce the amount of input noise to MICH. This should allow us to further reduce MICH noise to DARM by lowering MICH BW.

We measured the SRCL OLG in nominal now-noise and found it at 20Hz, with plenty of phase margin. We lowered it by 3dB ( H1:LSC-SRCL1_GAIN = -8.8 ), to 15Hz, which seemed to help DARM at 30Hz.

We also looked at MICH - it currently has a 6Hz UGF. Attached is a OLG measurement, but I am not sure I trust the phase at 6Hz. Craig and SHJeila are now working on MICH FF, which currently is worse than O2, and limiting DARM.

We fine-tuned the LSC SRCL input matrix again today by driving PRCL length at 32Hz and 135Hz, and minimizing the lines from the SRCL error signal: 

POP_A_RF9_I ---> SRCL was 0.038 and was set to 0.024

This reduced  the PRCL to SRCL coupling by a factor of 14 to 30 initially, cleaning up the SRCL error signal nicely. However, we noticed that in the next lock the coupling again came up some, to a net reduction of about a factor of 5.
We added this settings to lscparams.py.

I made a rough but representative model of why the SRCL feed-forward causes instabilities. The model works quite well in predicting an instability at about 1 Hz when the SRCL feed-forward is engaged and DC-coupled.

The model is based on:

• only MICH and SRCL longitudinal loops are modeled
• measurement of the optical response of the MICH and SRCL error signals, as measured in 44941. To keep things simple I treated the error signal response as frequency independent and I included only diagonal terms
• the suspension response is taken from the QUAD model plus G1100968 and from 44941
• I assumed the geometrical coupling from MICH to SRCL due to the fact that we are actuating MICH only on BS, and thus creating a significant SRCL motion
• the control filters are taken from the real time model
• the SRCL feed-forward is taken to be zero (no feed-forward), or as a constant (DC-coupled), or a constant times an high pass at 10 Hz (AC-coupled)

With the feed-forward off, the model reproduces well the open loop transfer functions. The key thing to consider here are however the closed-loop responses of the two coupled loops. The plot below shows three traces: in blue (hidden by the green trace) the case without feed-forward, which is known to be stable (as confirmed by the open loop gain); in orange the case with the SRCL feed-forward on, but without the high pass: this trace shows a "wiggle" around 1 Hz; in green the case with feed-forward on and high pass at 10 Hz (this trace hides the blue trace).

So the model explains quite well that adding the current 10Hz high pass remove any perturbation to the closed loop gain at 1 Hz. Morevoer, we can take the ratio [closed loop gain with DC-coupled feed-forward] / [closed loop gain without DC-coupled feed-forward] and fit it. The fit contains indeed an unstable pole at about 1.03 Hz. So the model explains how the interaction of the SRCL feed-forward with the SRCL and MICH loop creates an instability.

Finally, if we reduce the geometrical coupling of BS to SRCL to 10% of the value we get by actuating only on the BS (in other words if we send the MICH control to SRM or SR2 or SR3 and cancel 90% of the coupling), the model predicts that the 1 Hz instability is gone, see the plot below for a zoom of the closed loop transfer function around 1 Hz. Blue trace: stable, no SRCL FF; orange unstable, just with DC couple FF and MICH driving to BS only; green stable, with DC coupled SRCL FF and MICH driving to both BS and SRC, with a precision of 10%.

So the model gives us two ways (at least) to fix this instability, by reducing the MICH / SRCL cross coupling:

• send the MICH control to BS and to SRM / SR2 or SR3 (whichever is practical)
• send the SRCL feed-forward to the ETMs, since this should largely reduce the direct coupling of SRCLFF output to MICH, and thus cut the spurious loop in a different place

Sheila, Craig

I gathered a bunch of times of MICH excitations with feedforward on and off for O2 and O3 to see how well we're doing now as opposed to the past.

We seem to be doing an order of magnitude worse now than we did in O2.

Template exists at /ligo/home/craig.cahillane/Git/IFO/LSC/data/MICH_to_DARM_excitations_O2.xml.

MICH Excitation Times:
O2 MICH FF off: July 1 2016, 9:14:00 UTC
O2 MICH FF on:  June 14, 2017, 00:20:05 UTC
O3 MICH FF off: Nov 4, 2018, 10:29:37 UTC
O3 MICH FF on:  Nov 3, 2018, 22:47:00 UTC

• 15:50 In clean area/LVEA JeffB)
• 16:25 LVEA cleaning (Vanessa)
• 16:52 HAM6 check (Bubba, Jeff J)
• 17:00 Impromptu morning meeting
• 17:07 Schneider elec for Richard
• 18:24 TVo and Daniel to turn off OpLev on SR2
• 18:36 Sheila and Jenne out to ISCT1 and ISCT6
• 18:45 Karen to LVEA to clean
• 19:20 Vanessa to EX to clean
• 19:25 Nutsinee to ISCT6
• 19:28 Karen to EY to clean
• 19:35 Jenne and Sheila out of the LVEA
• 19:44 Nutsinee out of LVEA
• 19:44 TVo and to LVEA
• 20:06 Karen leaving EY
• 22:37 Nutsinee to ISCT6
• 00:17 TVo and Daniel to LVEA ITMY Hartman

OpLev trends for a week - NOTE SR3 optical lever was turned off today for a test, and is still off,  so the change in it's signal is understood.  See TVo for more info.

Laser Status:
Front End Power is 33.26W (should be around 30 W)
70W Output Power is 75.6W
Front End Watch is GREEN
70W Watch is GREEN

PMC:
It has been locked 0 days, 6 hr 29 minutes (should be daysweeks)
Reflected power = 14.87Watts
Transmitted power = 43.86Watts
PowerSum = 58.73Watts.

FSS:
It has been locked for 0 days 3 hr and 17 min (should be dys/weeks)
TPD[V] = 2.939V (min 0.9V)

ISS:
The diffracted power is around 2.2%
Last saturation event was 0 days 3 hours and 17 minutes ag (should be days/weeks)

Possible Issues:

SRC2 is pretty coherent with DARM (attached .png, bottom right corner pink and cyan).  I did some injections, so that Sheila can stick these into the noise budget (.xml attachments).  But, we should roll this QPD loop off faster.

Attached is the DARM spectrum, DARM coherence with some auxiliary channels, and the LSC auxiliary error point spectra for a quiet stretch on Saturday:

1225344730 (Nov 4 2018 05:31:52 UTC).

Note that - compared to Aug 21 2015, PRCL has an excess noise bump up to 60Hz, and an increased high frequency flat noise level (I hesitate to call it shot noise).
That bleeds slightly over to SRCL. Hoverer, the linear noise coupling, at least below 30Hz, is dominated by SRCL.

Gabriele might also post a bruco for this stretch.
Also note that this is from the same lock as Shaila's noise budget (alog 44998)