The master conlog process on h1conlog2 stopped running this morning after reporting the following errors:

Dec  2 10:06:32 h1conlog2 conlog: ../ecm_cac.cpp: 515: event_handler: pv name: H1:GRD-HPI_ETMY_USERMSG: args.status is not ECA_NORMAL: ca_message: No reasonable data conversion between client and server types: Exiting.
Dec  2 10:06:54 h1conlog2 conlog: ../conlog.cpp: 331: process_cas: Exception: boost: mutex lock failed in pthread_mutex_lock: Invalid argument Exiting.

'No reasonable data conversion between client and server types' is an error message generated by the EPICS channel access library. My first suspicion was that the data type changed, but this does not appear to be the case. 

It has been restarted.

Changed the matrix values to get pitch motion in the pitch channel (and Yaw) and have the signs respect the right hand rule.  Still no scaling for units or calibration though.  Updated safe.snaps committed to svn.

Fili was modifying the servo board to behave like the end station controllers providing power to the remote pressure sensor amplifier.  The gains of the remote channels are still not correct and we'll correct those down the road sometime--maybe next week.  The signals are noisy again, too bad as we had the ground well done and had low noise before, we'll get it back.

Dan, Dave

Today we rebuilt and restarted the OMC and OMC-SUS models to incorporate Livingston's changes to the ASC for the output mode cleaner.  These changes add outputs from the OMC model that pass ASC control signals to the OMC SUS (L,T,V,P,Y,R).

We've also incorporated a change from Livingston to the OMC master model (userapps/release/omc/common/omc.mdl) which adds a software trigger to the OMC length control signal.  This trigger shuts off the output to the HV PZT (OMC PZT2) when the HAM6 trigger is closed.  The input signal to this software trigger is the output of the PZT1_MON_DC channel; this acts as a readback for the shutter controller state.  The idea is that when the HAM6 shutter controller detects an input condition (light level above threshold), we should disable the length control signal to the OMC.  (Note that this change is somewhat redundant with the input trigger to the OMC length control signal, which enables the OMC LSC when the DCPD sum is above a threshold.)

Awkwardly, the output of PZT1_MON_DC is different between the two sites.  At H1, the PZT1_MON_DC readback out of the PZT driver box is -20V when the trigger is in the nominal/open state, and 0V when the trigger is tripped/closed.  At L1 it's apparently -5V when open and 0V when closed.  Furthermore, at H1 we calibrate the output of PZT1_MON_DC to reflect the voltage that is delivered to the PZT in the chamber: 10V in the open state, 0V in the closed.

The punchline is that I've flipped the polarity of the software trigger to work with the conditions at H1, and I've changed the threshold to be 5V instead of -3.9V.  This means the H1 version of the OMC master model is different from L1 - see the attached figure.  We'd like to understand why the output of the PZT1 DC monitor is different between the sites, but for now we'll need to remember that the master model here at H1 is slightly modified.  I have not committed these changes to the SVN.

Kyle, Gerardo

Used West Crane to "fly" pump cart into Beer Garden area -> Connected to BSC3 annulus -> Pumped annulus volume for a few minutes -> Noticed ion pump was "old style" rebuilt unit from 2002 -> pump has surpassed its lifetime and will be replaced at next opportunity as a matter of course.

I've updated the SSL cert for lhocds.ligo-wa.caltech.edu, and restarted Apache.  I was going to do the same (update the SSL cert) for the internal LDAP server, but it's a bit more involved than I expected so that will wait for another Tuesday.

Checked the Diode Crystal chiller as per OPS checklist and noted that it was full (midway between max and min level).  Log on the front of the chiller shows that Peter King added 250ml on 12/1.

Reset HEPI WD counters for HAM3 and ITMy.

model restarts logged for Mon 01/Dec/2014
2014_12_01 09:17 h1fw0
2014_12_01 10:12 h1fw0
2014_12_01 11:44 h1fw0
2014_12_01 12:46 h1fw0
2014_12_01 13:45 h1fw0
2014_12_01 14:45 h1fw0
2014_12_01 16:52 h1fw0

h1hw0 restarts because of minute trend copy from SSD to HDD. conlog frequently changing channels report attached.

• We installed the spare summing node (the prototype which has no serial number) and are using it now, with IN1 enabled on both channels with 0dB gain and polarity ON. 
• We reduced the whitening gain on ASAIR 45 to 6 dB (from 12dB) because the ADC was sometimes saturating causing lock losses when we transitioned DARM to this sensor. 
• We increase the MICH trigger threshold to 40, this has made DRMI lock acquisition with the arms locked on ALS considerably faster. 
• We spent some time on the ALS find resonance, more details latter. 
• We made measurements of the DRMI loop gains at a CARM offset of 30 pm.  We found we had to reduce the PRCL gain by 30% to get the UGF to 70 Hz at this offset, and the SRCL gain by a factor of 2 to get a ugf of 40 Hz.  This is now in the guardian. This change means that we are able to go to a arm build up 100 times higher than the single arm (roughly 20pm CARM stably).  
• We found that when we tried to go to a lower CARM offset, the DHARD loop became unstable and caused us to loose lock.  We are now reducing the gain by a factor of 2, and then manually turning it off.  (For the record, we are running ASC loops to the BS, PRM, and DHARD durring the CARM offset reduction right now.)
• We now see that the recycling gain is lower than it should be, which could be because our alignment is not good.  It might be helpfull to commision some ASC for CHARD and SRM.  We have seen that at around 130 times the single arm transmission, REFL DC is reduced by about 50%, and refl 9 I is coming into the linear range. 
• We have also made various changes to our guardians that should make the locking sequence more robust, like reloading the LSC matrix when it fails, checking if the PSL rotation stage gives us the wrong input power, and actually checking the IR in the CHECK_IR state instead of relying on a person to check it.

ALS IR search

This morning Matt and I did some debugging of the new ALS COMM FIND_FSR, this now works OK as long as the hard coded magic numbers are close to the resonance locations.  Latter this evening, this became untrue, so we added the FINE_TUNE_IR state in again, which was removed last week.  This is a slow search, and doesn't always work.  We could fine tune the parameters to make this better, but we should also find a way of saving the location of the frequency were the last resonance was found, instead of relying on the hard coded offsets.  COMM could also be improved by not discarding the information about any resonances it passes through in searchstate 0. 

The DIFF FIND_IR_BETTER wasn't working, it was actually checking only one of the hard coded offsets (twice).  We fixed this, so now it checks the whole list of hard coded numbers. These numbers also became inaccurate latter in the evening, and when FIND_IR_BETTER tries to search, it always drops the lock.  We reverted to our old FIND_IR instead of trying to debug this. 

These two pieces of code always seem to work for a while after someone has updated the hard coded offsets, but we haven't yet come up with something that is robust to drifts in the frequency. 

We were locked at -11 to -13 sqrt(TRX+TRY) for about 20 minutes (with DHARD off) before we turned DHARD on and lost it at 10:36 UTC December 2nd. 

Since there is no ASC path to TMS, I terminated the initial alignment path for TMS in ASC_MASTER and connected it to the ASC output matrix. (This means that the ASC output matrix, once this change is in, will have 21 rows, which is ugly.)

This is an internal reshuffling of existing signals, and no IPC change is expected (and later no change was found in the new IPC file after building the model).

This will only allow us to control any combination of TMS/ETM/ITM using green WFSs. Any kind of beam position control (e.g. using ITM video) should be done external of the ASC.

I didn't delete the initial alignment thing though it is not used these days.

Only ASC_MASTER.mdl was edited, no change in h1asc.mdl.

h1asc.mdl was compiled successfully at around 20:00 local time with the new ASC_MASTER.

It is not installed today.

08:07 Cris cleaning in the LVEA
08:55 Bob Rhodes heating on site to fix PSL AC unit
09:01 Filiberto and Aaron to end Y to work on PEM. Seisometers and tiltmeters will be disconnected for work.
10:17 Corey to squeezer bay
10:30 Travis and Betsy to LVEA West bay
12:14 Corey done
12:16 Travis and Betsy done
12:19 Filiberto and Aaron done 3IFO work in LVEA
12:53 Jeff B. and Andres done
13:03 Filiberto to mid Y to get chassis
13:58 Sheila and Alexa installing board in electronics racks by H1 PSL
14:14 Sheila and Alexa done
14:35 Corey looking for 3IFO parts at mid Y

Continuing 3IFO QUAD 06 data mining from alog 14151, previously missing data is attached below:

This is a late alog from yesterday when the alog was down. 

Sheila, Elli, Dave O, Alexa, Evan, Daniel

This morning when we arrived all the control room computers were off, the
PSL was down, and the alog was down.  Over the phone Dave walked us
through restarting cdsfs0.  The PSL watchdog had apparently tripped, although
Elli and Dave O trended the power out of the diodes (H1:PSL-NPRO_D1PWR and
H1:PSL-NPRO_D2PWR) and they didn't seem to be dropping before the watchdog
tripped.

We also had to adjust the Y arm fiber polarization.

We installed the common mode summing junction which Filiberto repaired 2
weeks ago. Channel B seems to be busted, the output is railed with a small input. Evan and I pulled the
board and put it in Fil's office, we asked Richard and Fil for a spare so
we can try locking tomorrow.

We also switched the order of the filters in ETMX L1 so that both boosts
are before the control filter.  We are hopping that this will help relieve
the intermittent problem that DIFF loses lock while engaging the boosts.

We added a new MEDM screen, which has an overview of the ISC guardians used for full locking attempts. 
There is a small and large version, they are both linked from the guardian overview screen.  

Since our calibration of the CARM offset in physical units is not very accurate, I post here a couple of plots with powers vs TR_CARM_OFFSET (sqrt(TX + TY) = -TR_CARM_OFFSET). 

The highest power build up we reached so far had TR_CARM_OFFSET = -15 (see first plot, arm power = 115 x single arm, as measured by TR_Y_QPD_B_SUM,  entry 15323 ); in this state the BS correction signal became very large and we unlocked in ~ 1 min. Even if we didn't stay long on the previous state (TR_CARM_OFFSET = -12, arm power = 74 x single arm), we didn't notice any instability there; so I think it is fair to say that TR_CARM_OFFSET = -12 is the highest stable state we reached.

 On Saturday , Stefan left the IFO locked with TR_CARM_OFFSET = -8 (arm power = 32 s single arm, Guardian state TR_QPD_TRANSITION). It stayed locked for about 1.5 h! (second plot). The current last step of the Guardian (which has been failing on Saturday) goes directly from -8 to -15 (corresponding to the state CARM_20PM). 

We will stop around -12 next time. 

J. Kissel

In summary:
(1) Turn on X, Y, and Z sensor correction for HAMs 2 and 5, using the standard Hua Filter scheme (see T1200285), with tuned gains.
(2) Use LLO's M1 OSEM Damping filters and gains.
(3) Turn off optical lever damping so we don't have worry about maintaining optical levers to as great care.

-------

This continues (and hopefully resolves) the study of why PR3 and SR3 (both HLTSs) are angularly noisy, began by Kiwamu (see LHO aLOG 15048), and continued in a prior aLOG by me (see LHO aLOG 15154). I had started today thinking that I would do the usual full modeling suite, and this time include the optical lever damping. But after a little bit of exploring, I found that the L1 HLTS, H1 PR3, and H1 SR3 were using in various, completely different damping schemes, the performance of the optical levers are radically different, so a noise projection would be difficult, and the L1 vs. H1 HAM ISIs perform significantly different. So, since a "representative" model seemed impossible, as did the thought of making an individual model for all four suspensions and comparing, I've just spent the time gathering proof of what we need to do to make them much better. Once we get all of the above three steps completed *then* I'll make a full model suite of the performance.

Here's the details explaining how I can to these conclusions. They're supported by the first and only attachment.
(3) Turn off optical lever damping.
    Pages 1 and 2 of the attached show the wide variety of performance on the optical levers. Page 3 and 4 show the H1 the levers are in loop, but only with a bandwidth from 0.4 to 1 [Hz] (see design for SR3 in LHO aLOG 14719). There seems to be some effort toward rolling off the noise, but it seems quite unrelated to the actually noise performance of the levers at high frequency.
    LLO *had* used optical lever damping sporadically on L1 PR3, but they're currently not using it and haven't since Oct 17 2014. Given that the damping is so much strong and the input motion is so much smaller, this makes sense that its not needed. Further -- even with L1 SR3 aligned, the location to which, presumably, the optical lever has been centered -- the performance of the optical lever spectra is not limited by residual ground motion of the optic. So it's most certainly unusable for control. 
    Including optical levers in the local damping scheme complicates the remaining dynamics of the suspension (and perhaps more to the point, the subsequent modeling of it), and getting used to relying on them means they'll be left on and most certainly reinject noise above their bandwidth unless the loops are custom tailored to the ever-evolving optical lever noise. So if we can achieve the same level of local damping with the top mass OSEMs, and improve the performance of the ISIs, let's do it. 

(1) Turn on X, Y, and Z sensor correction for HAMs 2 and 5, using the standard Hua Filter scheme (see T1200285), with tuned gains.
    Pages 5 through 7 compare the performance of the HAM2 and HAM5 ISIs, highlighting the degrees of freedom which contribute to L, P and Y at the optic. 
    Remember, the L at the suspension point is the dominant contributor to L *and* P at the test mass, at all frequencies (see pgs 5 and 33 of the second attachment to LLO aLOG 7907). In turn, X and RY (for PR3) and Y and RX (for SR3) are the dominant contributors to L at the suspension point. Y at the optic is all Y at the susp. point, which is all RZ of the table.
    Though I'm not sure what the H1 HAMs have worse performance than the L1 HAMs between 2 and 8 [Hz] (that should be investigated further), I certainly know that the *drastic* difference between 0.3 [Hz] and 1 [Hz] is because LLO has sensor correction for all DOFs turned on. Poking around at LLO, I've found that the sensor correction is nothing particularly fancy -- it's just the standard Hua filter scheme, with a single, gain only Match filter at the output to tune better match STS gain to the displacement sensors (those gains are a correction of ~10-20%, matched to a ridiculously high precision [where its unclear if the precision is needed]). HAM2 uses the HAM2 STS (STS A), and HAM5 uses the HAM5 STS (STS C), as expected.
    At the first L and P resonances of the HLTS, there's a possibility for the following improvement if we get to LLO's level of isolation:
Frequency [Hz]   Table DOF      Performance Ratio
    0.64          HAM2 X       6.23 / 0.11 = 56.6
                  HAM5 Y       5.29 / 0.08 = 66.1 
    0.74          HAM2 X       1.87 / 0.03 = 62.3
                  HAM5 Y       1.42 / 0.02 = 71.0
and as we know by now, its these lowest resonance frequencies that dominate the RMS motion of the optic.
    All this being said, except for between 0.2 [Hz] and 0.6 [Hz], LLO is kicking the snot out of the "requirements." Nice job! I'm very confident that we can do just as well here at LHO. The trick will be to get the HAM2 and HAM3 sensor correction up at the same time, so that we don't introduce and relative low frequency noise in the recycling cavities.

    P.S. There're some pretty nasty sharp features and associated harmonics in the L1 HAM5 ISI's RX and RZ spectrum ... we should get that fixed -- they're obviously electronic, particularly ugly, and might affect pulsar searches.

(2) Use LLO's M1 OSEM Damping filters and gains.
    Pages 8 and 9 highlight the DRASTIC difference in damping loop filters. I hesitate to call the H1 HLTS filters a "design," because I know they were copied from the QUADs (hence the 0.43 [Hz] and 0.56 [Hz] resonant gains in the L and P filters, respectively). There's no reason at all we shouldn't just switch to the LLO design immediately -- these aren't under global control so we need not worry about changing any global control transfer functions, and though I haven't modeled it (yet) the increase in gain at just about all frequencies, especially what's focused at the *actual* first L and P modes of the HLTS. With the switch, we would get a factor of 16.2 increase in gain at 0.75 [Hz] in the L loop (which presumable will hit the same mode in P as well), and a factor of 44 increase in gain at the first, 1 [Hz], Y resonance.

With steps (1) and (2) complete, that means we can expect to improve the Y and P motion at the optic, at the main resonances by as much as three orders of magnitude.