Koji ISHIDOSHIRO, Masaki ANDO, Kimio TSUBONO and
TAMA collaboration
11th Gravitational Wave Data Analysis Workshop
Systematical veto analysisSystematical veto analysis using all monitor signals using all monitor signals
Abstract
Purpose : fake rejection
Method : systematical veto using all monitor signals
Application: TAMA300 DT9
Results : fake rate was improved 2 orders.
• coincidence analysis (event-by-event veto)• systematical veto setting
ContentsIntroduction
purpose, previous works, our work
Method coincidence analysis (event-by-event veto), systematical veto setting
Data TAMA data, safety of veto
Results signals selection, fake rejection
Summary
ContentsIntroduction
purpose, previous works, our work
Method coincidence analysis (event-by-event veto),
systematical veto setting
Data TAMA data, safety of veto
Results signals selection, fake rejection
Summary
PurposePurpose
Fake rejection using monitor signals
Main signal
Intensity signal
Time series
Norm
alized
am
plitu
de
Monitor signals• L+• l-• l+• Laser intensity• Dark-port power• Bright-port power• Seismic motion• Magnetic field and so on
Monitor signals are recorded with a main output signal of a detector to watch instabilities of the detector.
Previous works
Monitor signals have been investigated for fake rejection and detector characterization.
A. D. Credico (2005), P. Ajith (2006), M. Ando (2005)
They have used only monitor signals having well known correlation with the main signal. They have optimized veto parameters by hands.
Previous monitor signal analysis
Many monitor signals are recorded.
Monitor signals
We should use all monitor signals with optimal parameters. We must optimize many veto parameters. It is difficult to optimize them for instant.
Our workMethod: systematical veto using all monitor signals
parameter optimization signal selection
Monitor signal1
Monitor signal2
Monitor signal3
Monitor signal4
・・・
Signal Selection
Parameter Optimization
•High efficiency•Low accidentalcoincidence rate
X
X
Systematical veto setting
Main signal
Coincidence Analysis
(event-by-event veto)
Coincidence analysis (event-by-event veto)
Systematical veto setting
Application: TAMA300 DT9
ContentsIntroduction
purpose, previous works, our work
Method coincidence analysis (event-by-event veto),
systematical veto setting
Data TAMA data, safety of veto
Results signals selection, fake rejection
Summary
Overview of our methods
Main signalMain signal Monitor signalsMonitor signals
Data conditioningData conditioning-whitening, removal of lines--whitening, removal of lines-
Event extraction
Coincidence analysisCoincidence analysis(often called event-by-event veto)(often called event-by-event veto)
Fake eventsFake eventsGW candidatesGW candidates
Without coincidenceWithout coincidence
vetoveto
With coincidenceWith coincidence
- - Excess-power filter (Δt,Δf, Pth) -W.G. Anderson (2001,1999)
Signal Selection
Parameter optimization
Systematical veto setting
Event extractionExcess-power filter calculates signal power in a given time-frequency window.
When power is larger than a given threshold, we detect burst event.
Main signal time window Δt : 12.8 msec frequency window Δf :800 – 2000 Hz Monitor signals time window Δt frequency window Δf power threshold Pth
optimization
W.G. Anderson (2001,1999)
burst event
Coincidence analysis
Main signal
Monitor signal 1
Burst events
Coincidence
Fake events
Monitor signal 2
GW candidate
Time series
Overview of our methods
Main signalMain signal Monitor signalsMonitor signals
Data conditioningData conditioning-whitening, removal of lines--whitening, removal of lines-
Event extraction
Coincidence analysisCoincidence analysis(often called event-by-event veto)(often called event-by-event veto)
Fake eventsFake eventsGW candidatesGW candidates
Without coincidenceWithout coincidence
vetoveto
With coincidenceWith coincidence
- - Excess-power filter (Δt,Δf, Pth) -W.G. Anderson (2001,1999)
Signal Selection
Parameter optimization
Systematical veto setting
Parameter optimization 1/2
Accidental coincidence rate is estimated by 1-min.time-shifted data.
Monitor signal • Δt, Δf• Pth
Optimization in systematical setting• high veto efficiency• low accidental coincidence rate
20 hours data is used only for systematical veto setting. 20 hours is the least time for us to get statistically-significant.
Veto efficiency =2/3
Accidental coincidencesrate =1/3
Time series [min]
Am
plitu
de
Main signal
Monitor signalTime-shifted monitor signal
1. Pth is fixed in a given Δt, Δf so that accidental coincidence rate is 0.1%.
2. Veto efficiency is calculated using the fixed threshold.
3. These processes are repeated using different Δt, Δf 100 times.
4. The Δt, Δf having the highest efficiency are selected as optimal parameters.
Parameter optimization 2/2
Accidental coincidence rate
Veto efficiency
0.1%
Power threshold 101
1%
10%
100%rate
0.1% is fixed so that total accidental coincidence rate is enough small.
Signal selection
Using the monitor signals having no correlation make accidental coincidence
rate increase without improvement of veto efficiency.The monitor signals must be selected to be used for veto or not.
Selection by the veto efficiency <0.5 % Do not use for veto 0.5 - 2 % Use for veto > 2% Use for veto with re-optimization lower threshold: accidental ~ 0.5%
We would like to use the monitor signals having strong correlation with the main signal more effectively.
These monitor signals are re-optimized so that the power threshold become lower.
Example of Signal selection 1/2
Monitor signal do not have significant correlation.rate
100%
10%
1%
0.1%
Power
simulated data
We do not use this signal for veto.
Veto efficiency
Accidental coincidence rate
Example of Signal selection 2/2
Monitor signal have strong correlation.rate
100%
10%
1%
0.1%
Power
simulated data0.01%
We use this signal for veto with lower threshold.
Veto efficiency
Accidental coincidence rate
ContentsIntroduction
purpose, previous works, our work
Method coincidence analysis (event-by-event veto),
systematical veto setting
Data TAMA data, safety of veto
Results signals selection, fake rejection
Summary
TAMA300 dataData: 200 hours in TAMA DT9 (Dec. 2003 – Jan. 2004)(20 hours data is used for only parameter optimization )Monitor signals: 64 channels (HDAQ 3ch, MDAQ 61ch)
HDAQ: 20kHz, 16bit
MDAQ: 316.5Hz, 16bit
Safety of vetoHuge GWs may make burst events on monitor signals.We confirmed the safety of veto by hardware injection test during DT8 and after DT9.
Sine Gaussian waves were injected into L- feedback signal.
We compared veto efficiencyand accidental coincidence rate.
Threshold
rate
Accidentalcoincidence
1 sigmaSignificant differences did not exit for all monitor signals.
Even huge GWs did not makeburst events on monitor signals.
Veto efficiency
ContentsIntroduction
purpose, previous works, our work
Method coincidence analysis (event-by-event veto),
systematical veto setting
Data TAMA data, safety of veto
Results signals selection, fake rejection
Summary
Selected signals
Laser
Bright-port PowPD
L+
Trans. Pow
Dark-port Pow
Intensity
• SEIS Z • Magnetic field
l-, l+
These 10 monitor signals were selected.
Intensity and l- signals were re-optimized to have lower threshold.
Selected monitor signals• Laser intensity• l-• L+• l+• Dark-port power• Bright-port power• Seismic motion• Magnetic field
Fake rate
Without veto
With veto
hrss threshold
Fake r
ate
[H
z]
Fake rate was improved 2 orders @ hrss = 10-18 .Maximum amplitude of fakes was improved by 1/4 .
Power threshold
hrss threshold
Software injection test
1/100
1/4
Accidental coincidence rate3.2%
Dead time0.2%
ContentsIntroduction
purpose, previous works, our work
Method coincidence analysis (event-by-event veto),
systematical veto setting
Data TAMA data, safety of veto
Results signals selection, fake rejection
Summary
SummarySystematical veto method using all monitor signal
coincidence analysis (event-by-event veto) systematical veto setting
Analysis with TAMA DT9 data• 200 hours data (10% are used for only parameter optimize)
• 10 monitor signals were selected.• Fake rate was improved 2 orders @hrss=10-18 with 3.2% accidental coincidence rate (or 0.2% dead time).
Future works• We would like to apply this method to online study for TAMA300 and CLIO
• Understandings of fakes origin were obtained. (such as unexpected correlation)
Convert from power to hrss
Signal power: dimensionless signal-to-noise ratio→ physical value: GWs RSS (root-sum-square) amplitude
Software injection test: sine-Gaussian signals f=850Hz, 1304Hz, Q=8.9
Log(hrss)
Power
Injection events
Fitting lineFitting line
Data conditioningRaw data: non-stationary, frequency dependence, line noises
Data conditioning filter
Normalization by averaged power before 10 min
Removal lines
Selection frequency bands to be analyzed
Fourier domain
Frequency
power
Time series
Before data condition
Afterdata condition
Equivalent period
Time series
Amplitude
375Hz signal
20kHz signal
Different sampling signal: out of synchronization
High speed DAQ 20kHz, Middle speed DAQ 375Hz
Common signal: dark-port power
Minimum T
Correlated signals
Main signal
Intensity signal
Time series
Norm
alized
am
plitu
de
PD
Laser
Intensity
Correlated signals
Main signal
Magnetic fields
Time series
Norm
alized
am
plitu
de
Correlated signals
Time series
Main signal
Norm
alized
am
plitu
de
Vertical seismic motion
Coincidence analysis
Threshold
Time series data
Poweroutput
burst duration time Main signal
Threshold
Monitor signal
Coincidence analysis
Threshold
Time series data
Poweroutput
burst duration time Main signal
Threshold
Monitor signal
DefinitionVeto efficiency: the rate of burst events rejected
Accidental coincidence rate: the probability of burst events rejected accidentally
Accidental coincidence rate was estimated by four different time-shifted data.
Significant differences did not exit.
We select 1 min. time shift for easy.
Time
Total: 200 hours ⇒ 180 hours are used to set an upper limit. 20 hours are used only to set veto parameters.
⇒ 200 hours are used to search evens.