I think our fan flow rates at EY are bogus - In order to improve temperature control Bubba and I adjusted flows on SF1 and after no response decided to turn on SF2.

Plot of the idicated flows attached - SF2 (red trace) shows 7 days of somewhat uniform flow even though we just turned it on today. Blue trace does not show expected results of adjusting pitch control as we did today.

If temperature control has improved by tomorrow we will leave the second fan running (two fans operating). Please let us know if there is evidence of increased vibration or noise due to this fan addition.

Changes were made between 4:45  pm and 5:00 pm local time.

I was left instructions to run a2l a few hours into the new lock. It's a bit early but the range has deteriorated too much for my pleasure and the YAW coherence is coming apart pretty bad. I've informed LLO of my action and will also inform them when I'm finished and back in the game.

• Showed up to a locked, observing H1 - Fire Alarm then lockloss (see previous aLog)
• Had some trouble re-aligning arms. PRM had some issues that Jenne and Sheila had to work out - Thank you ladies!
• Spoke to Joe Hanson a couple of times at LLO on the telephone as he didn't realize that his TeamSpeak was a little "hosed"
• I ran a2l once H1 was back to NLN and as I look at it now, it seems to have come apart a little in YAW and perhaps our range is suffering a bit from it.
• Jenne cleared some SDF artifacts of hers and the Intention bit was set to Observing.
• I called LLO to inform them of our intention bit as GWIstat issue is still in process
• Everything seems to be holding steady. No more EY Temp alarms. One TCSY Chiller Flow alarm. 

All times in UTC:

 The attached plot shows that the fier alarm activated at approximately 2 minutes after 00:00UTC.  Lockloss occurred approx 3 minutes after the alarm had been silenced.

It seems that the PSL Mic detected the alarm at roughly 00:02:23 until 00:09:12. The duration of it was approx 7 minutes.

I picked a common, control room StripTool channel, H1:LSC-TR_X_NORM_INMON,  to represent the time of lockloss which seemed to have occurred at approx 00:12:16. Most all of the other mics show a spike feature at the same time as the TR_X plot indicating some correlation. 

03:14UTC

While working on the HVAC Controls Upgrade, Apollo was restarting the air handling unit (AHU-4) which supplies air and heat to the office areas of the OSB. This apparently stirred up some dust in the duct work activating the smoke detector and setting off the building fire alarm. 

This brought about an abrupt end to the weekly meeting and also served as our annual fire alarm drill.

Many parts of our system have not operated correctly for several years and this alarm was testament that this upgrade is paying off in that we will now be able to control our HVAC system the way it was originally intended to perform, possibly even better with the advancement in technology of the replacement parts.
  
KUDOS to everyone on site for doing exactly what they were supposed to do during a fire alarm.    

J. Kissel

I've completed analysis of the new 2017-01-03 reference measurements (LHO aLOG 32942) that were taken after the L2/L3 cross-over upgrade (see LHO aLOG 32933), and including the updates to the AA/AI filtering (see LHO aLOG 32907). I will post more details later.

However, I have not yet pushed these results to DARM loop model parameters or the front end replica, nor have I generated new EPICs records to go with them. Therefore, the time-dependent correction factors, which use the update EPICs records will still be ~15-30% incorrect as they've been since the L2/L3 crossover change (i.e. the entire datset since the restart after the break), and will flag the calibration as bad, which means the ANALYSIS_READY bit will still be flagged as bad the up-coming lock stretches over night.

My hope is to get better information out within 24 hours.

Apologies, and thank you for your continued patience.

~1625 - 1630 hrs. local -> Kyle in and out of LVEA

Shut down rotating shaft vacuum pumps which had been pumping PT180 (BSC8) for the past 8 days.  Also, removed ladder which had been leaning against BSC8.  We should now have enough data to compare and contrast PT180's behavior between when it is exposed/pumped by the YBM to that of when exposed/pumped by a locally mounted turbo.  

Recall that the post-detect era installed Bayard-Alpert gauges PT170, PT180 and PT140 all exhibit a slow upward drift that the iLIGO era Cold Cathode gauges (sampling same vacuum volume) do not.  Understanding/believing our gauges is critical.  

We will decouple the vacuum pumps from PT180 on the next maintenance day.   

State of H1: Observe, and there's a potential for some site noise sources today listed below

Site Activities:

• Bubba raised EY temperature - it seems to have topped out at 68.7F and is currently toggling between 68.6 and 68.7, so might be turning around
• Apollo is on site reworking HVAC in the office area of the OSB - all day, expected to last all week, and maybe into next
• Bubba to EY, see alog 32968 for times the truck was potentially close enough to the CS or EY to inject noise
• Apollo truck happened to rumble loudly enough that I went outside to see what was up, the times I could hear the truck in the Control Room are 23:23:30UTC, 23:45:30UTC

Bubba raised some concerns about the EY HVAC system and after running it by Mike, drove to EY to investigate.

• CS: 13:52:30-13:54:13, driving from main parking lot to beam tube road just beyond the CS building (camera)
• EY: 22:03, Bubba called to say they had arrived at EY
• EY: 22:34:20, Bubba calls, leaving the end station
• CS: 22:42:27, truck cleared LN2 tank, and parked at VPW
• CS: 22:54:44, truck leaves VPW
• CS: 22:55:28, truck clears South East corner of LVEA section of the OSB

I have produced filters for offline calibration of Hanford data from the beginning of O2 A until the end of 2016. The filters can be found in the calibration SVN at this location:
/ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/O2/GDSFilters/H1DCS_1163173888.npz

For information on the calibration model, see
https://alog.ligo-wa.caltech.edu/aLOG/index.php?callRep=31693
https://alog.ligo-wa.caltech.edu/aLOG/index.php?callRep=32329
https://alog.ligo-wa.caltech.edu/aLOG/index.php?callRep=32907

For suggested command line options to use when calibrating this data, see:
https://wiki.ligo.org/Calibration/GDSCalibrationConfigurationsO2

The filters were produced using this Matlab script in SVN revision 4050:
ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/O2/H1/Scripts/TDfilters/H1_run_td_filters_1163173888.m

The parameters files used (all in revision 4050) were:
ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/O2/Common/params/IFOindepParams.conf
ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/ER10/H1/params/H1params.conf
ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/ER10/H1/params/2016-11-12/H1params_2016-11-12.conf
ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/O2/H1/params/H1_TDparams_1163173888.conf
ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/ER10/H1/Scripts/CAL_EPICS/D20161122_H1_CAL_EPICS_VALUES.m

Several plots are attached. The first four (png files) are spectrum comparisons between CALCS, GDS, and DCS. Kappas were applied in both GDS and DCS plots with a coherence uncertainty threshold of 0.4%. Time domain vs. frequency domain comparison plots of the filters are also attached. Lastly, brief time series of the kappas and coherences are attached, for comparison with CALCS.

Starting CP3 fill. LLCV enabled. LLCV set to manual control. LLCV set to 50% open. Fill completed in 28 seconds. LLCV set back to 15.0% open.
Starting CP4 fill. LLCV enabled. LLCV set to manual control. LLCV set to 70% open. Fill completed in 79 seconds. TC A did not register fill. LLCV set back to 35.0% open.

State of H1: in Observe

Site Activities:

• snow removal, but only small equipment used around the OSB
• one vendor to OSB, only office area
• Hanford Fire here for maintenance - checked in with Richard before startin

Current Weather Conditions:

• temperature: -6C, 20F
• winds: 16.1-32.2kph, 10-20mph
• useism: low

I reset the dark offsets for all LSC and ASC PDs.  The ASC diodes already had scripts in ..../userapps/asc/common/scripts/dark_offset/, so I copied the style of those into an LSC script that now lives in ..../userapps/isc/common/scripts/dark_offset/. 

I also created a bash script that will call each of those scripts in succession - it's natively in the isc folder, but linked in the asc folder:  all_offsets_LSCandASC.

Summary:
Calibration measurements utilizing Pcal to DARM transfer functions can be impacted by non-unity gain when making corrections for the modeled frequency response of the AA filtering. At LHO, the ER10/O2 model had an analog AA transfer function with non-unity gain based on the LTI measurements from ER8 (at LLO, the LTI model for ER10/O2 was normalized to 1). The analog AA model at LHO has a gain of ~0.99. Below we detail the impact on sensing function and actuation coefficients. In summary, the sensing function gain is ~1% larger than originally modeled, and the actuation coefficients are ~1% smaller. This would imply that the inspiral range is ~1% higher than currently predicted.

This new understanding means that the analysis code needs to be re-run for the optical response parameters and actuation coefficients. The front-end calibration will need to be updated, and, finally, the GDS pipeline needs new filters generated and installed.

Details:
The calibration of the Pcal channels (ex: H1:CAL-PCALX_RX_PD_DQ) determines the watts reflecting from the ETM per count of the channel at DC. Whatever gain of the analog AA, this is already accounted for in this calibration procedure. This gain is implicitly accounted for when the value of the calibration is installed in the front-end filter module.

A Pcal to DARM transfer function was previously understood as follows (note that PD calib., susnorm, m/N coeff are taken care of in the front end and (1+G)*(1/f^2) are taken care of in analyzing the measurements):

DARM   (IFO opt. resp.) (OMC DCPD TF) (AA(a) freq. resp.) (AA(a) gain) (AA(d) TF)
---- = ----------------------------------------------------------------------------------------------
PCAL   (PD calib.) (AA(a) freq. resp.) (AA(a) gain) (AA(d) TF) (susnorm) (m/N coeff.) (1 + G) (1/f^2)

where AA(a) is the analog AA, AA(d) is the digital AA, and "freq. resp." means the normalized transfer function, and G is the open loop gain. In the above (incorrect) understanding, the AA(a) and AA(d) terms cancel. The reason this is incorrect is that the AA(a) gain has an inverse in the PD calibration factor. So the real (correct) Pcal to DARM transfer function is:

DARM   (IFO opt. resp.) (OMC DCPD TF) (AA(a) freq. resp.) (AA(a) gain) (AA(d) TF)
---- = --------------------------------------------------------------------------------- .
PCAL   (PD calib.) (AA(a) freq. resp.) (AA(d) TF) (susnorm) (m/N coeff.) (1 + G) (1/f^2)

Thus, to isolate the IFO optical response, we need to divide out the modeled AA(a) gain. Since the gain is ~0.99, then the gain of the optical response should go up by ~1%.

The above equations are laid out in a graphical subway map schematic in G1501518-v14.

The actuation coefficients will also be impacted by this, although the coefficients will be multiplied by the AA(a) gain so that the overall DARM OLG remains unchanged.

I have pushed the changes to the DARM model code and scripts that account for this. Specifically:
computeSensing.m (r4025)
create_partial_td_filters.m (r4026)
create_full_td_filters.m (r4027)
fitDataToC_20161116.m (r4028).

Re-running analysis of optical response parameters requires re-running the fitDataToC_20161116.m script first (with printing data to file), then running the fitCTF_mcmc.m script. Re-running analysis of actuator coefficients requires re-running actuatorCoefficients_Npct.m (with printing data to file), then re-running fitActCoefs_Npct.m.

Hopefully this takes care of everything. Unfortunately, I cannot verify these changes with Matlab because I have no way of running Matlab offsite (need a network license). :(

Summary: shutting down the main HVAC system increased the range by 2-3 Mpc. The shutdown produced a large change in DARM, 8-10 Hz, and ~ 3% change between 48-130 Hz. The feature in the 8-10 Hz band may be due to turbulence in the Y-end chilled water flow and would probably be reduced by lowering the frequency of the VFD at Y-end. It is not yet known if this will reduce the noise in the 48-130 Hz band.

Thursday, Dec. 22, before we shut down, I cycled the main HVAC multiple times in order to see if it was costing us range. In the blue “off” periods in the figures, the chiller pad chillers and water pumps were off, the turbines were off at EX, EY, and CS, and the OSB fan was off. Figure 1 shows that the range improved by 2 to 3 Mpc during the “off” periods (we were averaging about 70 Mpc).

Figure 2 shows that the DARM spectrum improved in the 6-18Hz region, by a large factor between 8 and 10 Hz, and by roughly 3% in the 48-130 Hz regions. It is not clear whether the noise produced by the HVAC in the 48-130 Hz band is produced by vibration at lower frequency or by direct coupling. Low coherence with vibration sensors (< 1% in most of the 48-130 Hz band), and the observation that there is little change in peaks in that band that are known to be driven by vibration, suggest that the increased noise in this band is produced by vibration at lower frequencies and not by linear coupling. However, a preliminary look at PEM injection suggest that it is not impossible that the noise in the 48-130 Hz band is produced by direct vibration coupling. We will investigate this with further analysis of data from PEM injections.

The biggest difference between “on” and “off” was in the 8 -10 Hz band (Figure 2). Figure 3 shows that Y-motion of ETMY ST2 is quite coherent with DARM, while X-motion at ETMX is not coherent with DARM. The CS is also not as coherent at these frequencies. The chilled water flow at EY has previously been shown to produce vibration in this band from turbulence (https://alog.ligo-wa.caltech.edu/aLOG/index.php?callRep=11466), so reducing the setting of the variable frequency drive below the current 50Hz level (I recommend 35-45Hz in the linked log) might reduce the 8-10 Hz region of DARM. Further analysis of PEM injections and additional injections at 9 Hz at EY would help us determine if the 8-10 Hz drive at Y-end is also responsible for the noise in the 48-130 Hz band.

Shutdown times:

UTC Dec. 22

21:24:00 shutdown starts, 21:26:00 complete

21:34:00 startup starts,  21:36:40 complete

21:44:00 shutdown starts, 21:45:40 complete

21:54:00 startup starts, 21:55:40 complete

22:04:00 shutdown starts, 22:05:40 complete

22:14:00 startup starts, 22:15:40 complete

Evan G., Aaron V.
I have checked a new filters file into the calibration SVN for offline (C01) calibration. The filters were made using revision #3987 using the script 
/ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/O2/H1/Scripts/TDfilters/H1_run_td_filters_1165799036.m
The file can be found here:
/ligo/svncommon/CalSVN/aligocalibration/trunk/Runs/O2/GDSFilters/H1DCS_1165799036.npz

In order to achieve agreement with GDS calibration, it was necessary to add a delay of 2 16kHz clock cycles to the actuation filters. Evan G. is reviewing the DARM model codes to determine where this timing discrepancy originated. 

The attached plots include:
1) h(t) spectrum from CALCS and DCS
2) ASD ratio (DCS / CALCS)
3 - 5) plots of filters and errors