Stream images looks funny. Powercycled the camera and frame grabber solved the problem.

Conor, Jeff K

Summary:
- Conversion of ISI Suspoint motion into the DoF basis works for the arms, with limited fidelity.
- Corner station DoFs seem not to work, possibly some problem with inclusion of the Beamsplitter.
- IPC communication between ISI and SUS at ETMY is spoilt somehow.

The GS-13s on stage 2 are used as witnesses for residual motion. Their outputs are calibrated to nm, and converted to suspension point motion using Euler matrices (Cart2Eul). The output is in the local suspension coordinates. For example, the channel 'H1:ISI-ITMX_SUSPOINT_ITMX_EUL_L_DQ' is the longitudinal motion of ITMX, which is along the normal vector pointing outward from the HR surface, in the global +X direction.

These suspension point motions are IPC'd to the OAF model, and in the case of the ETMs this involves some multiplexing/AI-filtering for communication down the arm. In OAF they are combined into the IFO degrees of freedom based on  T1500610 .

As a sanity check, I fetched the ISI-model Suspoint channels and matrix'd them into the IFO DoFs in Matlab, for comparison with the OAF DoF channels. For the arms (XARM, YARM, CARM, DARM), things seem to be somewhat okay. I confirmed this by eye, and then took the spectrum of the subtraction (attachment 1, CARM_Spectrum). The residual is consistent with the dynamic range of single-precision floats (about 10^9), at least at high frequencies. A quick software-only transfer function using four poles at 0.1Hz and white-noise injection in a spare filter bank shows a similar total dynamic range (attachment 2, DTT_dynamic_range). The anti-imaging filters, for transmission of the ETM signals along the arm, should allow a considerably smaller residual than we see at low frequencies, and I don't really know what else might cause this.

The corner station DoFs are substantially wrong, visible directly in the time series' (eg attachment 3, MICH_time_series). Since MICH only uses the ITMs (confirmed to be 'okay') and the Beamsplitter, presumably the difference arises there. PRCL and SRCL are similarly incorrect.

Document T1500610 has some minor errors in the Suspoint->IFO-basis definitions, but the current matrix has correct values as far as I could see. It also points to the Suspension model versions of the Suspoint motion, eg 'H1:SUS-ITMX_M0_ISIWIT_L_DQ', but the model uses the ISI versions. Initial testing revealed discrepancies between:
    'H1:ISI-ETMY_SUSPOINT_ETMY_EUL_L_DQ'
    'H1:SUS-ETMY_M0_ISIWIT_L_DQ'
visible in attachment 4 (ETMY_IPC_issue). The other 3 test masses didn't show this gross level of disparity.

Wrote a a new guardian (VIOLIN_DAMPER.py in /opt/rtcds/userapps/release/isc/common/guardian/) that can figure out the best damping phase for the violin mode damper. (This automates a usually very boring task.)

The code relies on each modal damper filter module being set up the same way, with a narrowband filter in FM1, a -60deg filter in FM2 and a +60deg filter in FM3. This setup allows setting the feed-back phase in 60deg increments. E.g. 240deg = negative gain & +60deg filter.

The code loops through the 6 possible phase settings (0deg,180deg,60deg,240deg,120deg,300deg). For each setting it waits for the filter to settle, takes a peak reference value of the mode strength, then waits for a specified time and measures the mode strengh again, calculating the after/before ratio. If the mode grows too fast, the code stops earlier to avaoid a runaway, and procedes to the next phase setting.

After loopig through all settings, it sets the module to the best daming setting.

This seems to work reliably as long as we are not hampered by closely neighbouring modes, and should reduce the repetitive work in modal damping.

As example, I attached the output of the Guardian when it was applied to SUS-ITMY_L2_DAMP_MODE8. Plot 1 shows the violin mode strength over the 15min the code ran, screen shot 2 shows the Guardian message output.

We had an episode of really large beam motion at about 0.6Hz on OMC QPDs as well as OMCR QPDs but not on ASC_AS_C nor AS WFSs. It was very very slow to damp, and at some point it didn't damp further. It was clear that the OMC or OM3 was moving, but the motion didn't show up on none of OMC and OM3 OSEMs. But the OMC SD and RT OSEMs were super noisy.

On inspecting the spectrum of OSEMs, it was clear that the SD and RT OSEMs of OMC had a large broad band noise, the signature of the satellite amp oscillation. However, the difference is that we couldn't see the oscillation in kHz range even using the 16kHz channel (sorry no screen shot). Maybe the oscillation frequency got higher?

Anyway, apparently it has been like that since power outage (attached). Though it's not shown in the spectrum, we confirmed that the broadband noise was there every day after the power outage.

We disabled the OMC damping and the beam motion on OMC QPD immediately got quieter.

We went to the floor and power cycled the satellite amp for RT/SD of OMC (they're on the same box, everything else is on another one), and the noise went away.

Turned on the damping again and it worked fine, no excessive beam motion.

Apparently this happens even though all satellite amplifiers are "fixed". It's worth checking all OSEMs for all optics.

Sheila, Cheryl, Haocun

We re-centered the WFS A and B this afternoon to fix the misalignment (alog27792). Both of them are centered around (0,0) now and we took measurements on the IMC.

The spectrum (measured at test 1) and transfer function (with 20dB gain) of the IMC were taken both before and after re-centering, but there were nearly no differences between them.

For the spectrum, several peaks are:

-55.5 dBm/Hz @ 197.5kHz

-85 dBm/Hz @ 87.5kHz

-83.7 dBm/Hz @ 25kHz

For the Transfer Function:

UGF @150kHz, and we have a bump close to 0dBm at higher frequency.

Jenne, Nutsinee Sheila

We had trouble with locking the arm in IR for inital alingment, because PSL-POWER_SCALE_OFFSET was different from the input power.  We added a statement to the "IDLE" run state of LASER_PWR (the generator of states when not changing the power) that will adjust the normalization if it is more than 0.5Watts wrong. 

            if abs(ezca['PSL-POWER_SCALE_OFFSET'] - ezca['IMC-PWR_IN_OUTMON'])> 0.5:
                ezca['PSL-POWER_SCALE_OFFSET'] = ezca['IMC-PWR_IN_OUTMON']
 

I checked the pickoff mirror, IO_MCR_M14, that I installed on the MCR path on IOT2L (see alog 27725), which picks off the p-pol beam and sends it to a beam dump.  I found that with a locked IMC, the edge of the p-pol beam was getting past the pickoff mirror, so I adjusted the mirror position about 1mm toward the main beam.  By IR card, there's about 3mm between the p-pol beam and the main s-pol beam where the pickoff is installed, so enough clearance that the pickoff mirror should not clip the main beam.

• For our current environment (often windy, low microseism), I think that we should change our nominal BSC to the 250mhz blend, 80mhz sensor correction configuration detailed below. Currently called the Windy state in the SEI_CONF guardian.

• This configuration does a better job of rejecting wind induced tilt and may perform better against microseism than 90mhz blend, .5hz sensor correction that was our previous nominal state.

• This answer may change when the microseism comes up in the winter. We will have to keep this in mind for O2.

Now for the excessive detail.

Before the last couple of months work, we thought that for the end station ISI's, we should use the Quite_90 blends and the Mitt_SC broadband sensor correction. A couple of months ago, Krishna pointed out that the Mitt_SC filter was probably insufficiently rolled off at low frequency, where the BRS actually makes the STS/BRS supersensor worse. There have been a number of iterations on sensor correction filters, but I think we now have a good filter. The first attached images shows a few of the filters we have tried, I won't go through them all, but the dark blue is the Mitt SC filter of old, and the cyan is what we are using currently. The other two are filters that we've tried and discarded, for various reasons,and there are many more in the foton files, but I'm tired of plotting them.

Around the same time, I realized that using the Quite_90 blend may not roll off the ISI T240's quickly fast enough at low frequency to avoid injecting platform tilt and that we should use a higher blend, at least while the microseism is low. Currently we're using the Quite_250 blends for a high blend, but there may be room for optimization there.

The next plot is the estimated supprestion of the "old" BRS configuration (the Mitt_SC filter, Quite_90 blend), the "new" SC/blend configuration (Warn_SC_v3, Quite_250 blend), and a couple relevant CPS blend filters (which are a good first approximation to the low frequency performance, without sensor correction). JeffK has a log (594) in the seismic log that explains this estimation of the expected sensor correction performance, for those interested.

We know from experience the 45mhz blends are hopeless in 10+ mph winds but suppress the microseism well, and the Quite_90 are okay in winds to 20 mph, but are no good in high (~>.5 micron BLRMS) microseism. I've tried using the Quite_250 blends alone with low microseism, but I think the motion around the first TM modes was too high (the spots on the AS port were moving around a lot at something like .5hz). The new configuration has allowed us to lock in 30-40mph winds (and lower), but microseism is very low. Based on the the simple model used in the second plot, it should do better than the Quite_90 blends alone in high microseism. It even looks like it should do better that the 45mhz blends at the secondary microseism (.150 hz), but the gain peaking is right on top of the primary microseism (~30-80mhz), which we don't see often.

The gain peaking of the new configuration looks like it would be worse than either the 45mhz blends or the Quite_90s, but this model doesn't include platform tilt, which is filtered out by using a higher blend on the ISI and a tilt-subtracted STS ( at the ETM's) or a low tilt STS ( ITMY)  for sensor correction. I think Krishna showed that this "new" configuration is in practice better, in his alog 27735.

LLCV bypass valve 1/2 turn open, and the exhaust bypass valve fully open.

Flow was noted after 52 seconds, closed LLCV valve, and 3 minutes later the exhaust bypass valve was closed.

After Sheila's log about the BS causing locklosses, I wanted to check a few things in the guardian code and I found what looks like conflict in the code. In /opt/rtcds/userapps/release/isi/h1/guardian/  ISI_BS_ST2.py, line 4

ISOLATION_CONSTANTS['CART_BIAS_DOF_LISTS'] = ([ ] , [ ])

the empty brackets at the end indicate the ISI is not set to restore any Cartesian locations.  This is supposed to be the code that sets what dofs are restored when the ISI re-isolates, and is particular to the beamsplitter.

However, in  /opt/rtcds/userapps/release/isi/common/guardian/isiguardianlib/isolation/ CONST.py lines 102 (for BSC ST1) and 122 (for BSC ST2) both read

            CART_BIAS_DOF_LISTS = ([], ['RZ']),

CONST.py is common code, and I would interpret this to mean that all BSCs are restoring a stored RZ location. This isn't a problem for the other BSCs, because they never change state, but if we turn on the ST2 loops for the BS, this code could force the ISI to rotate some after the loops come on. The attached trend shows the last ten days of the BS ST2 RZ setpoint and locationmon, and they track each other, so I dont think the BS is returning to some old RZ location, but someone who understands this code should explain it to me.

Raised CP5 Dewar pressure (3/4 turn at pressure regulator + 1/4 turn yesterday). Will take days to read a pressure response. Current pressure is 17.5 psi in tank and 15 psi at exhaust. 

Also filled CP5 to 100% with 1/2 turn open on LLCV bypass valve. 

Evan, Stefan

We didn't get past the ASC engaging tonight.

The one thing that clearly improved stability was setting the MC2 cross-over up by 4dB - before that even the slightest disturbances kicked the x-over into an oscillation.

After that fix, we at least didn't have any more immediate losses. 

However, the current ASC engaging logic doesn't work. It also doesn't seem to make much sense. We have convergence checkers, but they are currently used before all loops are closed. Thus we wait for one loop to drag us off into some direction, only for the next loop to come on and load up the previous loop again.

Peter, Sheila, Evan

We have been having trouble keeping the IMC locked when powering up (without the interferometer).

We found that the UGF of the loop was something like 110 kHz. During O1, we ran with a UGF that was more like 50 kHz. Recall also that the IMC loop at one point had some questionable resonance features above 100 kHz, so it is probably in our interest to keep the UGF on the low side (though we should confirm this with a high-frequency OLTF measurement).

We turned down the loop gain by 4 dB, giving a UGF that is more like 60 kHz. We were able to power up from 2 W to 23 W without lockloss.

Attached is an OLTF measurement after turning down the gain. As usual, the last point is garbage.

C. Cahillane

I have revamped the uncertainty budget to include covariances between all stages of actuation and all time-dependent parameters.
I computed each parameter's covariances in real and imaginary coordinates to provide a consistent basis.  I then compiled an 6 x 6 Actuation Covariance Matrix C_A, a 2 x 2 Sensing Covariance Matrix C_S, and an 8 x 8 Kappa Covariance Matrix C_K.  Then I compile them into a giant covariance matrix C:

     _             _
    |  C_A  0   0   |
C = |   0  C_S  0   |
    |_  0   0  C_K _|     

Then, I multiply by some conspicuous Jacobian vectors J(f) to get the final 2 x 2 uncertainty matrix σ_R^2(f):

σ_R^2 = J * C * J'

where J looks like:

        _                            _
       |  d Re(R)    d Im(R)          |
       | ---------  ---------   ....  |
       | d Re(p_i)  d Re(p_i)         |
J(f) = |                              |
       |  d Re(R)    d Im(R)          |
       | ---------  ---------   ....  |
       |_d Im(p_i)  d Im(p_i)        _|

(I was able to use complex differentiation and Cauchy-Riemann here to make the derivatives easier.  Recall that R = 1/C + D*A.  Now I can compute dR/dA = D and dR/dC = -1/C^2 to form J(f), thanks to 200 year old mathematics)

Finally, to make the uncertainties readable by humans, I divide σ_R^2(f) by |R(f)|^2, rotate σ_R^2(f) by angle(R(f)) via a rotation matrix, and read off the square roots of the diagonal of the rotated σ_R^2(f) to get the magnitude and phase uncertainties plotted below.

I have plotted the uncertainty at GPSTime = 1135136350, the time of the Boxing Day Event.

The plot shows an overall increase in magnitude uncertainty of about 1% at low frequency.
Phase uncertainty increased by about 0.5 degrees at low frequency.

The effects are more dramatic at Livingston.  Check out LLO aLOG 26542.  

Jenne, Peter, Jeff, Terra

We took charge measurements on the ITMs by driving 20.1 Hz into H1:SUS-ITMX/Y_L3_DRIVEALIGN_L2L_EXC with 100k cts and looked at the coupling to DARM. We stepped up and down the ESD bias voltage from zero and found the bias that gave zero coupling to get charge measurements, where V_charge= (20/2¹⁸) * bias * 40. ITMX zeroed with bias = 3k, ITMY zeroed with bias = 2.5k. See bias stepping ITMX and ITMY spectra attached. 

ITMX charge: 9.15 V,   ITMY charge: 7.6 V

ITMX: 

ITMY:

Approximating alpha: The first term in the expression for the force produced by the ESDs is the attractive force between the ESD fringe fields and the test mass: F = alpha(V_bias - V_signal)^2, where alpha is a constant of proportionality. With the known bias voltage V_b, signal drive voltage V_s, drive frequency f, and now the peak amplitude x_pp , we used the largest bias offset (100k) to approximate alpha for both ITMs. Work is attached. 

ITMX: alpha = 1.78 x 10^-11 N/V²,   ITMY: alpha = 1.85 x 10^-11 N/V²