Reports 1-1 of 1 Clear search Modify search
AdV-COM (AdV commissioning (1st part) )
Allocca, Bersanetti, Swinkels - 2:18 Friday 21 April 2017 (37318) Print this report
Alpha subtraction - part I

Tonight we worked on the subtraction of MICH noise in DARM. Since this requires very accurate measurements of the various loops, we spent the first part of the shift on making new filters for shaping the noise so that it closely resembles the quiet error signal. We now use individual filters for the various degrees of freedom, instead of single general one. These filters have been added to the CALnoise Filter Bank and are named "DARM_noise", "MICH_noise" and "PRCL_noise". This is essential for getting measurements that show good coherence over a large frequency range. The same measurements can also be used for the longitudinal noise budget.

After that, we added the necessary modification to the LSC_Acl process. The change is minor, we basically take the correction of the MICH loop, pass it through a filter, multiply it by a gain, and add it to the DARM correction. The filter is defined in the include file LSC_MICH_subtract.cfg, which can be reloaded without unlocking the interferometer.

Finally, we dug up some old Matlab scripts that were used during Virgo+ to calculate the filter. The main changes that had to be done was to update the channel names and use a new way to get the shape of the suppression filter that was running at the moment of the noise injections (in the past, we could get the nominal poles and zeros from the data, now we simply measure it using tfestimate).

We succesfully tested a single iteration of tuning the filter. The suppression was not great yet, since the MICH measurement was taken with the interferometer in a pretty bad state. After this, we did a second set of noise injections, which were used to make the second iteration, but the ITF lock was not anymore stable to allow this test. As in the past, there were some troubles in the fitting with vectfit, which gave negative zeros. Looking at Fig. 1 and 2 (the comparison of the noise projection for DARM and MICH for Alpha suppression off and on respectively), one can see that some improvement is already achieved, as some structure around 100 Hz is suppressed. We can conclude that the noise suppression appears to work as in the past, but more work is needed to reach a good performance..

Hereafter the GPS of the noise injections. They have been performed using the inject_lsc.py in the /virgoDev/Automation/scripts/LSC folder.

'CLEAN' refers to a stretch of data with no noise injections. The amplitude is set for the LSC_noise_colored filter in CALnoise.

The first table refers to the injection performed without any noise subtraction. The second one has been perfomed after the noise subtraction filter was enabled (around 22:28 UTC).

No alpha filter
DOF GPS Duration Amplitude
CLEAN 1176762378 120 --
DARM 1176756933 120 400
MICH 1176758696 120 2

 

First alpha filter
DOF GPS Duration Amplitude
CLEAN 1176762579 120 --
DARM 1176762748 120 400
MICH 1176761980 120 2
PRCL 1176763398 60 10

 

A couple of side notes:

  • there was one hour, roughly from around 11:30 to 00:30 LT, where the Injection system unlocked many times; the on-call was alerted but the intervention was not needed, as the problem went away right after; this is to be better checked by experts tomorrow, but it seemed related to the PMC rampeauto;
  • we unlocked a few times because of a very high line in DARM at around 309 Hz; a good data set for analysis is at GPS = 1176758875 + 120s.
Images attached to this report
Comments to this report:
ruggi - 7:34 Friday 21 April 2017 (37320) Print this report

The 309 Hz oscillation is something related to the marionette reallocation. I supposed to have fixed it, but evidently some change in the marionette filter is needed. Maybe even a notch in the LSC filter could work.

bersanetti, allocca, swinkels - 22:21 Friday 21 April 2017 (37332) Print this report

We implemented a notch filter for the 309 Hz line in the LOW_NOISE_1 state, as a last step with the switch of the coils' relays; in order to do that we added to the DARM loop the 'Arm_notch' filter, but disabling the already existing notch for a 16.85 Hz line, as it induced loop instability, but it was not used anyway. In addition, in order to use the additional filter we had to increase the maximal order of the 'LSC_NONE' filter placeholder from 80 to 100.

Search Help
×

Warning

×