Virgo Logbook

Today we worked on LSC noise budget, Feedback (MICH and SRCL) and Feedforward (especially Alpha) control filters optimization , together with the reduction of the UGF lines position.

Yesterday Michal worked on the Drift control on the B1s RF quadrants during the locks in DC readout. Since they were performing well, today we uncommentend the lines in the ITF_LOCK.py for the control engagement.

At the beginning of the shift we found the ITF locked in LowNoise2.

We already have noticed in the past (https://logbook.virgo-gw.eu/virgo/?r=60456) that for all the DRMI loops, due to the roll-off structures, the computation of the UGF is not done reliably. Thus, all the loops are gaining more wrt the designed UGF. So before starting the activity we performed a quick test online on SRCL, that was reducing the SRCL UGF Setpoint from 5 to 3 Hz (since we know that the band is wrong of about a factor 2), in order to reduce the gain of the loop enough to have more or less the rigth UGF.

GPS of the test:

7.33 UTC UGF set SRCL from 5 to 4
7.42.15 UTC UGF set SRCL from 4 to 3 to bring the real UGF to 6 Hz

After this test we profited of the stability of the lock to work on the Feedforward filters, computed from the injections performed in the latest shifts:

8.07.50 UTC engaged new ALPHA for LN2; this is a good improvement with respect to the previous one, so we left it online.

From here we started to take some CLEAN DATA.

Then we performed a new set of measurements of the LSC loops:

GPS time stamps of the injections:

8.20.35 UTC + 120 sec PRCL noise inj, ampl 5e2 ---> N.B. Due to the roll-off, also PRCL line is too high (in terms of position).
8.27.10 UTC + 120 sec SRCL noise inj, ampl 2e-3
8.34.20 UTC + 120 sec CARM_slow inj, ampl 1e-4
8.42.00 UTC + 120 sec DARM inj DARM_noise ampl 1 (during the injections sidbands of one excited violin mode got excited as well, but it may be just an artifact).
8.53.40 UTC + 120 sec DARM inj DARM_noise_broad ampl 60
8.58.45 UTC + 120 sec MICH noise inj ampl 5e-3

Thanks to these measurements we computed a noise Budget wrt DARM. Analysis of the same set of data to compute the Noise Budget with respect to Hrec will follow. From the DARM one we can see that most of the LSC loops are not that limiting, with CARM now being the worst offender. More on this topic below.

After the measurement we started to work on the new designed control filters for MICH and SRCL (https://logbook.virgo-gw.eu/virgo/?r=60605).

During the same lock we tested the filter without changing the lines position, but only changing accordingly the loops gain through the UGF 'calibrated' setpoints.

GPS of the new filter tests + measurements:

9.06.40 UTC engage new MICH filter (MICH_control10)
9.13.16 UTC MICH UGF SERVO decreased from 7 to 4 Hz
9.21.03 UTC + 120 sec MICH injection ampl 5e-3

The new filter is behaving as expected, by keeping the same accuracy (slightly improved) and is reducing consistently the reintroduction of noise in basically within all the band up to 100 Hz.

Then we measured also DARM in order to check if it was possible to improve the ALPHA filter with the new MICH controller (even if theoretically the feedforward filter design should be independent from the running controller):

9.25.15 UTC + 120 sec DARM noise inj DARM_noise shape, ampl 1

Then we moved to SRCL:

10.02.00 engage SRCL filter LN2 (SRCL_ctrl5b)
10.03.58 UTC we unlocked due to 12 Hz oscillation -> High frequency SRCL oscillation due to the bad UGF computed. We didn't make it on time to adjust the gain.

In the next lock we took our time to work on the reduction filters on the different steps: Alpha for CARM null1f and LN1, plus GAMMA for LN2:

Alpha in CARM_NULL_1F was already good enough, and the measurement we did a few days ago could not compute a better subraction, so we left the old one;
Alpha in LN1 was severely degraded, so the new one computed a few days ago is a strong improvement, and we left it online;
For Gamma we are using a static gain for the time being; we fitted a shaped filter with today's measurement, but no macroscopic improvement could be seen; also, making tests with different static gains were not succesful, and the difference with respect to no feed-forward is marginal; around 15 Hz, the feed-forward seems even detrimental. This topic will be explored more after a new calibration of the NE actuator and arms rebalancing.

Once in LowNoise2 we worked online on the position of the UGF lines. We decided to change DRMI UGF lines to the following positions:

MICH 13.3 Hz;
SRCL 11.1 Hz;

We didn't perform any test on PRCL, but plausibly we will move the line from 64.4 to 31.1 Hz.

GPS time stamps of the test:

12.27 UTC changed online MICH line to from 21.7 to 13.3 Hz (with the standard filter)
12.28.45 disabled Pydiag servo

After the change, the servo started to compute the right UGF (9Hz).

12.38.44 engage new MICH filter (MICH_control10)

From here clean data.

12.46.20 UTC changed online SRCL line from 26.6 to 11.1 Hz

we waited the servo to converge to the right UGF, and increased a bit the SRCL amplitude

12.59.00 UTC engaged new SRCL filter (SRCL_ctrl5b)
13.00.07 Unlock while the UGF servo was decreasing the gain of the loop. 3.75 Hz oscillation, probably low gain oscillation?

Observations:

In the next days, data of today will be analyzed to well understand the effect of the new rolloff filters on Hrec.
The filters engagement worked well, MICH especially is reducing a lot of noise in the whole band of interest;
SRCL filter stability and robustness need to be further studied. So other tests will be performed in the next usefull shifts;
In addition, different optimization of the SRCL_ctrl5b performed by Henrich should be tested (see pictures attached). These optimized filters are designed in order to keep the same accuracy (LF) and to reduce further more the reintroduction of noise in the HF zone. Phase margin is more or less 25 deg.
Today indeed we only performed online test with the change of position of the UGF lines. Before doing the8/9 modification, we need to update accordingly the python script of the DRMI diagonalization of the pydiag process (since today we switched it off during the test), as it relies on the hardcoded values for the UGF lines frequencies

Automation:

we reduced a couple of timers from 20s to 10 in ACQUIRE_LOW_NOISE_2, as now we are confident that the transition is correct and they should not be needed anymore; old ones are left commented in ITF_LOCK.py;
in view of the change of the DRMI UGF lines frequencies, those are already reverted to the standard values during the lock acquisition;
the engagement of the new MICH_control10 filter (and the new target UGF) has been inserted in the automation, but for now we left it commented; before engaging it by default in LN2, we need to ultimate the UGF lines modification and the integration with the PyDiag process.

ATTACHED PICTURES:

Fig. 1: comparison of Alpha subtraction in LN2, before (purple) and now (blue); we kept the " blue filter";
Fig. 2: LSC Noise Budget of DARM: as stated above, CARM is the most limiting one, with MICH/PRCL/SRCL relatively below the DARM noise;
In Fig. 3 is reported the comparison of MICH spectra and correction with the different filters (purple: nominal, orange: MICH_control10 with wrong gain, blue: MICH_control10 with gain adjusted);
In Fig. 4 is reported the comparison of SRCL spectra and correction between the nominal controller (purple) and the new one, SRCL_ctrl5b (blue) with the wrong gain;
Fig. 5: comparison of Alpha subtraction in CARM_NULL_1F, before (purple) and now (blue); we kept the "purple" filter;
Fig. 6: comparison of Alpha subtraction in LN1, before (purple) and now (blue); we kept the " blue filter";
Fig. 7: comparison of Gamma subtraction in LN2: in purple the old static gain; in blue the new shaped filter; in red no feed-forward at all;
Fig. 8/9: suppression and fit for Alpha in LN2;
Fig. 10/11: suppression and fit for Alpha in LN1;
Fig. 12: comparison of MICH spectra and correction with the different filters and the UGF line moved from 21.7 to 13.3 Hz (purple: nominal, blue: MICH_control10);
Fig. 13: comparison of SRCL spectra and correction with the different filters and the UGF line moved from 26.6 to 11.1 Hz (purple: nominal, blue: SRCL_ctrl5b);
Fig.s 14/15/16: Optimized control filters for SRCL, to be tested yet.

At the end of the shift we left the ITF to the next CAL shift.