Sergey Klimenko
University of Florida z z z z z z z
What are bursts?
Burst analysis strategy
What is WaveMon?
WaveMon algorithms
WaveMon results for E7 & S1
WaveDSO
Summary
S.Klimenko, October 2002

z z z
Bursts – un-modeled GW sources localized in time
¾ unanticipated sources: rate-?
¾ Supernovae & GRB: rate-several/year
¾ binary systems: mergers and poorly predicted inspirals
¾ N-N: 1.4Mx1.4M: LIGO-I(20Mpc) 1/3ky-1/3y LIGO-II(300Mpc) 1/y-2/day
¾ N-B: 1.4Mx10M: LIGO-I(40Mpc) 1/3ky-1/2y LIGO-II(600Mpc) 1/y-4/day
¾ B-B: 10Mx10M: LIGO-I(100Mpc) 1/300y-1/y LIGO-II(600Mpc) 2/m-10/day
Glitches (background)
¾ non-stationary detector artifacts
¾ detected in auxiliary channels (control, environmental)
Same (or similar) tools are used to detect bursts&glitches
¾ Bursts DSOs: slope, excess power, TFcluster & WaveDSO
¾ Glitch monitors: absGlitch, pslMon, WaveMon
S.Klimenko, October 2002

Compact binary mergers
z
Merger detection rate ~40 higher then inspiral rate
¾ 30Mo<M<200Mo (LIGO-I) 100Mo<M<400Mo (LIGO-II)
S.Klimenko, October 2002

ifo1 data conditioning
DSO
ifo1 auxiliary channels
DMT glitch monitors
Ifo2, … Ifo2, … auxiliary data conditioning channels
DSO
DMT glitch monitors
VETO
VETO coincidence z
GW events
Goal: set limit on the GW burst rate (or discover them) by reducing the false event rate due to random coincidence of glitches and by keeping high efficiency for GW bursts (checked with injected simulated bursts).
S.Klimenko, October 2002

z z interpolating bi-orthogonal wavelet transform binary wavelet tree (linear frequency scale)
S.Klimenko, October 2002

z z z z
Purpose
¾ monitor non-stationary detector noise (glitches)
¾ generate VETO triggers for specified environmental and control channels
Method
¾ Time-frequency analysis in wavelet domain.
Algorithms
1.
wavelet transform: produce wavelet TF plots for master (GW) and slave
(notGW) channels. There could be several slave channels.
2.
percentile transform: select a certain fraction (20%-30%) of TF pixels using the
“percentile selection rule”.
3.
coincidence: produce the coincidence TF plot using “zero time window coincidence rule” for master and slave channels (for the same interferometer)
4.
cluster analysis: find clusters and apply cluster selection cuts using cluster size, asymmetry and likelihood
Output
¾ VETO triggers (estimate TF phase space available for GW bursts)
¾ false alarm triggers (estimate false alarm rate)
¾ dmt-viewer & trend data
S.Klimenko, October 2002

z z given a sample with amplitude A , define its percentile amplitude as a=1/f , where f is fraction of samples with absolute value of amplitude greater then | A| percentile amplitude distribution function: P(a)=1/a 2
¾ the same for all wavelet layers
S.Klimenko, October 2002

S.Klimenko, October 2002
H2:LSC-AS_Q
H2:IOO-MC_F

z z z z threshold on percentile amplitude select a certain fraction of samples (T-F occupancy) don’t care about the data distribution function handle the data dynamic range
H2:LSC-AS_Q
224-240 Hz
S.Klimenko, October 2002
10% 10%

z coincidence TF plot for LSC-AS_Q && IOO-MC_F z z coincidence allows
¾ to set a low threshold at the previous (thresholding) step
¾ to “build” clusters with known false alarm rate.
WaveMon allows different coincidence policies
S.Klimenko, October 2002

cluster – T-F plot area with high occupancy
Cluster halo cluster core positive negative
S.Klimenko, October 2002
Cluster Parameters size – number of pixels in the core volume – total number of pixels density – size/volume amplitude – maximum amplitude energy total energy asymmetry – (#positive - #negative)/size likelihood – log( Π A i
) neighbors – total number of neighbors frequency core minimal frequency [Hz] band frequency band of the core [Hz] time GPS time of the core beginning interval - core duration in time [sec]

z z
S.Klimenko, October 2002

Wavelet coefficients (t) z
Frequency band 224-240 Hz (wavelet layer 14)
¾ red – H2:LSC-AS_Q, black – H2:IOO-MC_F glitches
S.Klimenko, October 2002

z
S1 data
¾ green Æ false alarm rate (random clustering)
¾ red Æ off-time rate (random coincidence of glitches)
¾ blue Æ glitch rate
S.Klimenko, October 2002

z
Total rates for 10 channels:
¾ MICH_CTRL, POB_I, POB_Q,
REFL_I, REFL_Q, LVEA_V1,
MAG1X, PSL, ACCX, MC_F
S.Klimenko, October 2002

z
Glitches and GW bursts compete for the same fraction of the T-F space, defined by the percentile threshold P .
¾ acceptance A – fraction of PV available for GW bursts
A
1
P
1
⋅ V
V i
V i
– cluster volume
V - total T-F volume
H2 H1 z excellent measure of the detector performance in terms of glitches
S.Klimenko, October 2002

Use WaveMon veto to suppress Tfcluster events
S
Combined veto: MICH + POB+CARM+MCF+REFL, gap=0
‹
Veto: H2:LSC-MICH_CTRL, gap=0
VETO efficiency may be limited by Tfcluster false alarm rate.
S.Klimenko, October 2002

E7 playground Data
• black – TFCLUSTER
• blue - WaveMon (only 16-2046Hz frequency band is used)
• red -TFCLUSTR x VETO
L1:LSC-AS_Q x L1:PSL-ISS_ACTM_F
0-3000Hz: 2698 events (88.6%)
200-3000Hz: 1222 events (94.3%)
L1:LSC-AS_Q x ALL
0-3000Hz: 1884 events (92.1%)
200-3000Hz: 727 events (96.6%)
S.Klimenko, October 2002

veto ifo1 percentile selection cluster selection coincidence
(correlation) cluster analysis consistency check cluster selection ifo2 percentile selection veto z z z
Early coincidence allows to set low threshold Æ better sensitivity to GW bursts
Predictable false alarm rate Æ predictable rate of random coincidence after applying glitch veto.
Development is almost completed .
S.Klimenko, October 2002

z z
WaveMon is a tool to monitor the detector performance in terms of glitches.
¾ Can help identify the source of glitches in the GW channel by looking at frequency content and correlation with auxiliary channels
¾ Estimates dead fraction of T-F space due to glitches
¾ Has predictable false alarm rate
¾ Fast: can monitor up to 20 16kHz channels in real time
¾ produces veto triggers
WaveBurst uses similar algorithms as WaveMon
¾ run on S1 data to produce GW candidates
¾ run on S1 data with injected simulated GW bursts to measure the
WaveBurst efficiency.
S.Klimenko, October 2002