Virgo Logbook

Detector Characterisation (Glitches)

direnzo, dalcanton - 0:55 Thursday 18 January 2024 (63006)

Welcome the "elbow glitches" - Some observations on the waveform of the annoying 25-minute glitches

Following the analysis initiated in logbook entries #61944 and #62161, I report here the plots of some of the glitches included in the dataset in logbook entry #62965 for the mini-Engineering Run.

Figure 1 is an animated GIF with the first 20 glitches in the miniER dataset for the raw channel V1:Hrec_hoft_20000Hz. No filters have been applied. They almost all show some steep turnings at the times marked by the dashed line, which gives them an "elbows-like" waveform. Not sure if it matters but this turning seems to be "always" upward... Notice also that the time of the dashed line has been triggered by the excess energy in the whiten data, and it is slightly delayed from the omicron trigger time reported in the dataset. I assume the latter to be a characteristic time of the tile corresponding to the glitch.

Figure 2: the same behaviour is observable in Hrec_raw, for all the glitches.

Figure 3: unfortunately, the glitch presence is a bit less recognizable in DARM. (the fast oscillations seem to correspond to the frequencies of the violin modes ~450 Hz) The one shown in the figure corresponds to that in the previous figure for Hrec. Notice one peculiar thing, shown in the next figure...

Figure 4: whitening the data, for better visualizing the glitches, we can notice that the glitches in Hrec are delayed 1145 ms from those in DARM. Additionally, the phase seems to be the opposite: upward turnings in Hrec convert to downward ones. These seem to be general properties of all the 25-minute glitches examined. I don't know how expected or informative is this observation. Maybe reconstruction experts can comment further...

Images attached to this report

Comments to this report:

direnzo - 9:13 Thursday 18 January 2024 (63008)

ERRATA CORRIGE: concerning the delay between the glitches in DARM and Hrec, the delay is 114.5 ms. I've missed the dot in the above entry.

mwas - 11:02 Thursday 18 January 2024 (63010)

How are the data whitenened, and how does this differen between h(t) and DARM? I exepect the ~100ms delay is due to difference in the whitening, and that the whitening is done in a causal way, which means that it introduces a delay.

Figure 1 and 2 show a time-frequency decompostion of h(t) and DARM made using the old matlab implementation of Omega pipeline and there is no time delay of ~100ms between the two. From what I remember Omega uses zero phase delay whitening filters (equivalent ot applying the same filter twice once in the time forward direction and once in the time backward direction).

/users/mwas/detchar/hoftGlitch_20230118/arches.m

Images attached to this comment

63010_1705576037_Screenshot from 2024-01-18 09-58-36.png

63010_1705576037_Screenshot from 2024-01-18 09-58-17.png

mwas - 13:51 Thursday 18 January 2024 (63013)

The systematic kink in the h(t) time series is interesting. If one assumes that it is due to a current discharge in the mirror coils, we can try to use it to estimate what is the current involved.

Looking at the figure my imperssion is that the change of the kink is of the order 0.5e-18 over 40ms, which using the 3km arm length to convert strain into DARM correspond to a change of speed of 4e-14 m/s.

The mirrors have a weight of 42kg, and the coil/magnet pairs have a response of 0.0034 N/A (VIR-0296B-16). So the change in momentum correspond to an impulse of 1.6e-12 N s and a charge going through the coils of 5e-10 C.

If this charge correspond to a discharge of a capacitor with 10V at its edges, the 5e-10C correspond to a capacitor with 50pF capacitance. The question is if we can imagine somewhere a ~50pF capacitor, being charged over 25 minutes with a ~3e-14 A current, and discharged in a few ms once it reaches 10V. (Or something with similar order of magnitude).

direnzo - 15:00 Thursday 18 January 2024 (63011)

I confirm the impression of Michal on a potential non-zero phase shift in the transformed data. I've used the gwpy standard whitening filter for both Hrec and DARM in Figure 4 of the above entry. This filter is based on an FFT-convolution of a FIR filter obtained from the inverse of the estimated signal ASD, used as a transfer function. I have contacted Duncan MacLeod for a better understanding of these functions.

To confirm that the glitches in DARM and Hrec are in fact synchronous, I've used a Butterworth forward and backward ("filtfilt"), bandpass filter between 20 and 300 Hz to focus on the frequencies of interest for these glitches. As visible in the attached figure (where DARM has been multiplied by 10^-13 to have a similar scale as Hrec), the glitch in the two channels occurs in practice at the same time.

Images attached to this comment