Bursts in VIRGO

1
Bursts in VIRGO

• C5 run analysis
• Data statistics
• Burst filters
• Non-stationarity investigation
• Hardware injections

AC Clapson - LAL On behalf of the Virgo
collaboration
2
Interest of VIRGO C5 run
2

• Stable recombined (no PR) optical configuration
• Duration and quality
• Science mode for long stretches
• Hardware injections
• Important transition from
• simulated Gaussian noise.
• Focus on
• Quiet data segment ( 5h).
• Dark fringe signal
• (DC, in-phase, quadrature)

3
Statistical studies tools
3

• Spectrogram
• Rayleigh monitor
• R 1 Gaussian
• R ltlt 1 coherent
• R gtgt 1 non-coherent
• (fast fluctuations)
• Plot 1-R
• Frequency power c2 test
• On log-spectrogram of whitened data, confidence
  level of non-stationarity.
• Event confidence gt 99.9
• Frequency band spectral flatness
• Computed after whitening.
• ? 1 for flat spectrum.
• Plot 1-?

4
Statistical studies overview
4
Frequency (Hz)
c2 test
Rayleighogram
5
Statistical studies frequency view
5

• Approximately Gaussian
• Specific line behaviours
• non-Gaussian
• frequency modulation?
• Most variability
• 0 - 350 Hz
• 3000 - 4000 Hz
• 6000 - 7000 Hz
• Non-equivalent tools.
• Frequency range
• Sensitivity to
• local features.

6
Statistical studies time view
6
Overall limited fluctuations. Small trend in
PSD. No systematic coincidence in peak
location. Information extraction?
Gaussian data reference
7
Burst search methods
7

• Time domain
• Mean Filter (MF)
• Alternative Linear Fit Filter (ALF)
• Correlators
• Gaussian (PC)
• complex Exponential Gaussian (EGC)
• Sine Gaussian tiling based-
• TF domain
• Power Filter (PF)
• Fourier Domain Adaptive Wiener Filter (FDAWF)
• S Transform

(involved methods) (not used here)
NB Not all filters produce SNR consistent
outputs.
8
Burst search methods II
8
Methods involved in C5 investigations
9
Burst search summary
9
Single detection
Double detection
C5 quiet Segment.
Dots for all events Other symbols differentiate
methods.
Using 40 highest energy events for each
method Single detection 47, double 18, triple
11, quadruple 11.
10
Burst search summary II
10

• Many non-coincident triggers.
• Known filter-dependent coupling to waveforms.
• Time varying outputs.
• Partial correlation with statistical overview.
• Focus on different time scales.
• Complementary approaches.
• Quality flag relevance?

11
What do we trig on?
11
In-phase channel
Highest SNR event in segment. Lower energy
events hard to find visually. Veto candidate?
12
Veto investigation with MF
12
Highest SNR glitch in stretch Weak on composite
dark port and demodulated signals,
but clear in photodiode channels
13
Veto investigation II
13
yet invisible in acoustic and magnetometers
channels from central building.
14
Non-stationarity hint
14
Over 5h quiet period, MF trigger density
increases with time
Average SNR evolution
Trigger count evolution

• Computed quantities
• Trigger count
• Averaged SNR
• (over 930s periods)

• Clear increase of trigger density in the 3
  channels.
• (consistent with PSD trend)
• ltSNRgt constant on demodulated signal, increasing
  on DC.

Quiet period 5h
Quiet period 5h
15
and investigations
15

• Gaussian stationary models check
• Compare to simulation data
• Auto-regressive model derived from data PSD.
• Trend not reproduced in Gaussian data.
• Change whitening coefficients
• Training set either at beginning or end of
  segment.
• Trigger count variation but trend maintained.
• Trend not caused by whitening errors.
• Trigger typology
• Observed trend is specific of short windows (lt
  3.5 ms)
• Two local fluctuation periods found for larger
  windows.
• Similar behaviour on all three dark port
  channels.
• Throughout exploitation of methods results.
• Importance of adaptivity time-scale.
• Local fluctuations issue.

16
Hardware injections searches
16

• Injections numerical core collapse,
  Sine-Gaussian, NS-NS
• Burst filters
• MF and PF.
• Noise level issue.
• SNR accuracy?

17
Last word
17

• Relatively short stretch
• Unique observations
• Prototype study
• Involve many complementary tools
• Investigation of deviation
• from stationarity.
• Group activity
• Commissioning Mini-Runs
• LIGO-Virgo joint work

18
Conclusion burst analysis in VIRGO
18

• Large toolbox for
• Data characterization
• Burst search.
• C5 most extensive analysis so far.
• Expectations for C6
• Recycled ITF
• Longer stretches of data.
• Topics to develop
• Multi-channel coincidence
• Integration of methods in synthetic picture.

19
Complements
20
B1 photodiode
Data in WPR_B1_DC
B1
d6
50
99.6
d8
Faraday
50
96
0.4
d2
B1s
Data in WPR_B1p_DC
50
d2
50
d1
50
50
B1p
d1
Data in WPR_B1s_DC
21
Statistical studies encore
Flatness estimator
Lowest frequencies most affected by variability.
22
MF triggers details
23
Burst filter performances
ROC for MF
ROC for PF
SNR 10
SNR 8
SNR 7
SNR 5