Krishna, Jim, Ryan,
This morning, we tried some different blend filter and sensor correction configurations on the ISI at ETMX to see if we could improve the low-frequency performance of the platform.
To give some background, the nominal configuration along X axis, at present, is the 45 mHz blends and the LLO ("0.43 Hz only") sensor correction. The very low blend frequency amplifies the motion of Stage 1, likely through tilt reinjection. We wondered if we could do better by going to a higher blend and relying on a broadband low-frequency sensor correction, which uses the tilt-free super-sensor. We used the CPS sensors as the witness sensors, which should be reliable below 0.1 Hz and ST1 T240 as at higher frequencies.
The first attached plot shows the Stage 1 T240 X sensor ASD for the following configurations:
1. Red - 90 mHz blend and Low-Frequency Sensor Correction (LF SC, also known as Rich's filter or Z sensor correction filter).
2. Blue - 90 mHz blend, LF SC and BRS off (no tilt-subtraction).
3. Green - 45 mHz blend, LF SC with tilt-subtraction.
4. Brown - 45 mHz blend, LF SC with no tilt-subtraction.
5. Magenta - Nominal configuration - 45 mHz blend, LLO SC.
The CPS (before sensor correction) signals for the above configurations are shown on the second plot. These should be used as an estimate of the platform motion at low frequencies. The third plot shows an example of the tilt-subtraction. Wind speeds were between 5-10 mph at EX.
As expected (and seen before 15270), the Green/Brown configurations offer best isolation at the microseism. The rms CPS signal for Case 3 is about the same as Case 5. Turning BRS off shows the clear re-injection of tilt in Case 4.
In Case 1/2, there is still a factor of ~5-10 reduction at the microseism and the usual factor of ~200 at 0.43 Hz. The slight surprise was that in Case 1/2 there was a clear amplification of the primary microseism by factor of 4-5 which limits the rms motion of Stage 1. It is not clear why this is being amplified and is worth investigating further.
The tests were concluded by 9:30 AM and the ISI was returned to the nominal configuration.
I've moved around filters (and gains) that are used in the OpLev damping, so that we can make try to increase the bandwidth of our DHARD WFS without driving through the high Q filters that have been used in P2P and Y2Y in the L2 drivealign matrix.
Even though we aren't currently using Yaw OpLev damping on either mass, I did the same for Yaw. Before and after screenshots attached.
The EY 02 Dust monitor has repeatedly gone off this morning, so I disabled it from the medm alarm manager screen. Not sure if it is something anyone has looked into or not.
Fil and Kiwamu,
The following cameras are now reachable from the analog camera screen:
model restarts logged for Mon 02/Feb/2015
2015_02_02 10:28 h1lsc
2015_02_02 10:34 h1broadcast0
2015_02_02 10:34 h1dc0
2015_02_02 10:34 h1fw1
2015_02_02 10:34 h1nds0
2015_02_02 10:34 h1nds1
2015_02_02 10:35 h1nds0
2015_02_02 10:36 h1fw0
2015_02_02 13:41 h1fw0
2015_02_02 20:29 h1fw1
2015_02_02 21:30 h1fw1
LSC code change with associated DAQ restart. several unexpected fw restarts
model restarts logged for Tue 03/Feb/2015
2015_02_03 10:28 h1iopiscex
2015_02_03 10:28 h1iopiscey
2015_02_03 10:28 h1iscex
2015_02_03 10:30 h1iscex
2015_02_03 10:37 h1fw0
2015_02_03 10:38 h1iscex
2015_02_03 10:40 h1iscey
2015_02_03 10:41 h1iscey
2015_02_03 10:46 h1alsex
2015_02_03 10:47 h1alsey
2015_02_03 10:49 h1odcx
2015_02_03 10:49 h1odcy
2015_02_03 10:49 h1pemex
2015_02_03 10:51 h1hpietmx
2015_02_03 10:51 h1hpietmy
2015_02_03 10:51 h1pemey
2015_02_03 10:51 h1susetmx
2015_02_03 10:51 h1susetmy
2015_02_03 10:52 h1fw1
2015_02_03 10:52 h1lsc
2015_02_03 10:54 h1asc
2015_02_03 11:10 h1broadcast0
2015_02_03 11:10 h1dc0
2015_02_03 11:10 h1fw0
2015_02_03 11:10 h1fw1
2015_02_03 11:10 h1nds0
2015_02_03 11:10 h1nds1
2015_02_03 12:38 h1broadcast0
2015_02_03 12:38 h1dc0
2015_02_03 12:38 h1fw0
2015_02_03 12:38 h1fw1
2015_02_03 12:38 h1nds0
2015_02_03 12:38 h1nds1
2015_02_03 13:21 h1pemex
2015_02_03 13:28 h1pemex
2015_02_03 13:42 h1pemey
2015_02_03 14:11 h1alsey
2015_02_03 14:15 h1alsex
2015_02_03 14:35 h1alsex
2015_02_03 14:46 h1alsex
2015_02_03 14:46 h1alsey
2015_02_03 16:51 h1alsey
X1PLC1 10:18 2/3 2015
X1PLC2 10:18 2/3 2015
X1PLC3 10:18 2/3 2015
Maintenance Day. ISC end station model restarts with associated DAQ restarts. Jumbo frame upgrade of both frame writers. Bi-weekly restart of h1ecatx1. No unexpected restarts.
SEI: Jim needs DRMI to test configuration changes.
CDS: Investigations on the floor for future PSL shutdown.
3IFO: No disruptive activity.
Commisioners: Last night success for 40sec lock!
Facilities: Bean tube cleaning continues. Next week they will be on the road with cones out.
Safety Meeting: Top 10 Safety Hazards
Safe Safe!!
At 1616utc, I changed the epics PID Proportional and Integral parameters from 35 & 0.45 to 20 & 0.07. These later numbers are what VincentL derived shown in E1100508 in June 2011. Vincent's Plant included no chambers, so it makes sense that these parameters are not ideal. But, in an effort to reduce the coupling of the sensor noise, these lower magnitude parameters should lower our UGF and this coupling. We'll look at coupling to HEPI and ISI signals after a couple hours.
Our next steps will be to check the loop gains before we try turning off the refl offset, and trying to increase the DHARD WFS bandwidth.
We also toggled the noise eater tonight (we thought it might have been oscillating but this probably wasn't the case). We also copied the invP2P and Y2Y filters from the drivealign matrix into the opLev servos so that we can work on the DHARD WFS loops independent of these filters.
Lock loss times: all Feb 4th UTC
4:14:00
5:00:00
7:09:00
I have attached two time series of the lock aquisition (a long and short version), which includes the power trend of DC REFL and the build up in the arms. The third attachement is of the CARM and DARM spectra during our short full lock.
Note: the oscillation at about 1800 sec in the first attachment arises from us trying to transition REFLAIR9 and sqrt(TRX+TRY) too early in the CARM offset reduction.
Once the new IMC power is taken into account, I think the arm buildup on resonance is more like 940 times the single-arm buildup. That means we've achieved an IFO recycling gain of 28 W/W during this lock stretch.
With REFL_A_LF at 21 mW with arms off resonance, and 0.7 mW with arms on resonance (see Alexa's plot above), this means the IFO visibility is greater than 95% if REFL_A_LF is giving an accurate measure of the power reflected back from the PRM.
Back when we digitally normalized the TR sensors (LHO#16211), we did so at 10.95 W of power into the IMC. Correspondingly, the LSC power normalization was set to 10.0 W (i.e., there is a small mismatch).
However, for this lock we had 2.82 W of input power into the IMC, with the LSC power normalization at 3.0 W (another, different mismatch). That means the arm buildup value of 810 that we saw tonight should be corrected to 810×(10.95/2.82)×(3/10) = 944. With a 3% PRM transmissivity, the recycling gain is therefore 28 W/W.
Since we were at 2.8 W of input power to the mode cleaner rather than the "nominal" 10.9 W that we've been using previously, and since the LSC power normalization is only dialed in to the nearest integer watt, there is a small amount of systematic shift in the arm buildups during this lock.
Congratulations!!!
The first noise spectrum! Nice going.
Awesome! Nice going - Congratulations!
Excellent news! Congratulations.
J. Kissel More interesting data on the HEPI Pump Servo. Looking to ID the response of the HEPI Differential Pressure "Plant," i.e. from the control drive output to differential (supply - return) pressure change, I turned OFF the servo and put small steps in the control output of varying sizes (see pg 1 of 2015-02-03_H1HPI_PumpServo_SysID.pdf). First linearly increasing step amplitudes to determine the DC gain (see pgs 2 and 3). Then, to determine the frequency response, I put in many steps of the same size to fit the impulse response (see pg 4). The results: As it stands, the HEPI pump servo plant can be well-approximated by a single pole at 24 +/- 1 [mHz] and a DC gain of 0.0546 [PSI/ct]. Note, E1100508 suggests that the plant is a single pole a 7 [mHz], with a DC gain of 0.087 [PSI/ct]. Looks like not. With this information, and the current PID settings of (35,0.45,0) respectively, I can predict the current open loop gain (see pg 5). I model a UGF of 39 [mHz], with a suppression of 0.37 @ 10 [mHz], 0.2 @ 1 [mHz], below which the 1/f integrator kicks in. Note, E1100508 suggests using PID params of (20, 0.07, 0), which would result in even lower a UGF and less suppression. With such a simple plant, it's really tough to find PID parameters that aren't stable, so I also show a toy set of parameters with much higher values, (P,I,D) = (300,30,0), where the integrator dominates. With these parameters, we get a UGF of 390 [mHz], with suppression of 0.24 @ 100 [mHz], 0.035 @ 10 [mHz], and so on. Why not increase the UGF to infinity? Because there're worries out there that above a few [Hz], poorly-matched impedance, transmission-line-like effects take over the frequency response, which may cause sharp, phase-full features in the plant that are not characterized by a simple pole. Why does this not-at-all match what's measured when comparing Closed vs. Open. vs. No Sensor ASD spectrum? (see 2015-01-27_H1HPI_DifferentialPressure_Open_v_ClosedLoop_Comp.pdf, and 2015-02-02_H1HPI_PumpServo_ADCNoiseCharacterization_DiffASD.pdf, which is not new data, but copied to this aLOG for convenience) Open vs. Closed loop ASDs show that there is suppression of 0.1 @ 10 [mHz], not the modeled 0.37 @ 10 [mHz]. Further, the measured spectra shows some gain peaking of a factor of 2, where the model predicts no region where the suppression is above unity. It really does seem like the ASD behavior of the signal is totally disconnected from the time series behavior of these sensors. I really hope this whole saga isn't some false alarm from EPICs vs DTT nonsense... ----- Raw Data lives here: /ligo/svncommon/SeiSVN/seismic/HEPI/H1/Common/2015-02-03_H1HPI_PumpServo_StepResponse.mat Data Processing Script lives here: /ligo/svncommon/SeiSVN/seismic/HEPI/H1/Common/H1HPI_PumpServo_StepResponse_20150203.m
As it happened the new als models contain another experimental feature: An align/misalign button for the input beam steering. This feature was added at the end of the integration module. The aim is to have separate values for the misaligned and aligned state. The alignment values are the output bias of the integrators. This allows for instantaneous and glitch-free off-loading as well as a single button misalignment. The misaligned state can either be a fixed value or a value which is added to the aligned value. An additional calibration gain has been added to the output to make sure the alignment values are in physical units. All state changes are done with a smooth ramping process.
All integrators have been updated to include a ramp time.
Before today's work the models running on h1iscex, h1iscey were:
core | model | ex dcuid | ey dcuid |
1 | IOP | 83 | 93 |
2 | PEM | 84 | 94 |
3 | ISC | 85 | 95 |
4 | ODC | 86 | 96 |
5 | CAL | 124 | 125 |
After the split the models now are:
core | model | ex dcuid | ey dcuid |
1 | IOP | 83 | 93 |
2 | PEM | 84 | 94 |
3 | ALS | 85 | 95 |
4 | ODC | 86 | 96 |
5 | ISC | 126 | 127 |
teal = model not changed
New PEM = OLD PEM + OLD CAL
NEW ALS = ALS component of old ISC (inherrits old ISC core and dcuids)
NEW ISC = old ISC minus ALS (uses old CAL core and brand new dcuids)
Spliting the End Station ISC Models Into Two
Daniel, Jim, Dave:
WP5034. The problem: end station isc models are running long, so their RFM IPC to the corner station do not make it in time. The solution: split the ISC model up into ALS and ISC, with the latter running faster and able to send RFM IPC to the corner in the allocated time (60uS).
Because the h1iscex,ey computers did not have any spare cores (iop, pem, isc, odc, cal) for now we added CAL to PEM to free up a core. Hopefully a better long term solution is to combine isc and odc.
Daniel had done a lot of early legwork getting the models into shape. He hand edited the H1.ipc file to change the source model name for channels moving from ISC to the new ALS models. Daniel also handled the filter module files.
We identified all models which receive IPC channels (including Dolphin) from the new ISC and ALS models. These were also recompiled and restarted.
First thing this morning we moved the new H1.ipc file into place and did a round of make
and make-install on the related models. The testpoint.par file needed editing during the "make install" process due to the model-dcuid change for the isc models.
The rtsystab file was modified for the new model layout in the end station isc front end.
Then all the related models were stopped:
h1lsc, h1asc, h1susetmx, h1susetmy, h1hpietmx, h1hpietmy, h1pemex, h1iscex, h1odcx, h1calex, h1pemey, h1iscey, h1odcy, h1caley
The h1iopiscex and h1iopiscey were restarted. We saw an awgtpman autostart problem with these models.
The end station models were restarted (h1pemex, h1iscex, h1odcx, h1alsex, h1pemey, h1iscey, h1odcy, h1alsey). We had to play around with the safe.snap files for the ISC and ALS models to get them to startup.
The receiving models were restarted (h1lsc, h1asc, h1susetmx, h1susetmy, h1hpietmx, h1hpietmy)
The IPC from the end station ISC models to the LSC are now without any errors. The ALS model still has some IPC to the LSC and this retains its error rate (this model was not substantially sped up).
The DAQ master file was reconfigured for the new model layout, and H1EDCU_DAQ.ini was resynced.
I checked the DAQ data for the PEM and found it was corrupted. This was tracked to the change of the ADC part name in the model when PEM and CAL were combined, and we forgot to remap the bus selector parts attached to the ADC. This was fixed, there is a two hour gap in PEM data.
CDS OVERVIEW MEDM
Dave:
The H1CDS_STATE_WORD_CUSTOM.adl overview MEDM screen was updated:
Another reboot of h1alsey was needed to fix a wrong ADC channel assignment. If memory serves right, there is a busted ADC channel in EX but then this "fix" was propagated to EY when the new ey model was templated from ex.
Current max CPU times:
The IPC errors from the red transmitted power and the red QPDs have disappeared. We still have the channels H1:ALS-X/Y_ARM with IPC errors. These are sent from the als models. They are no longer needed and can be eliminated.
IPC errors from the corner to the ETM SUS have not changed and are around 10-20 Hz. This is not surprising, since the corner models were not updated.
Noticed that the ODC models in EX/EY run at 32kHz. Since the end stations only run at 16kHz, this is unnecessary.
All safe.snap and filter files are up to date. medm screens updated where needed. All changes in svn. Looks like this change is a success.
Still to do:
Here is the link to the busted iscex ADC channel which prompted the software fix.
These trends were taken while there was someone in the enclosure. New ones will be taken on 02/04.