The Block-Normal Event Trigger Generator John W. C. M c Nabb, for - - PowerPoint PPT Presentation

LIGO

The Block-Normal Event Trigger Generator

John W. C. Mc

¯Nabb, for

The Penn State University Relativity Group

Mike Ashley, Lee Samuel Finn, John Mc

¯Nabb, Eric Rotthoff, Amber Stuver,

Tiffany Summerscales, Matt Tibbits, Keith Thorne, Tamara Valinoto, Kristina Zaleski

mcnabb@gravity.psu.edu

Penn State Center for Gravitational Wave Physics LIGO Scientific Collaboration

McNabb GWDAW-8: Milwaukee,Wisconsin 1/15

LIGO

Outline

Block-Normal Overview From Time-Series to Triggers Change Points Blocks and Events Clusters to Triggers Tuning Block-Normal Thresholds Sensitivity to Block-Normal parameters Reconstruction of Simulations Conclusions

McNabb GWDAW-8: Milwaukee,Wisconsin 2/15

LIGO

The Problem

Finding unmodeled bursts in time-series. “Unmodeled” means one must look for general features of bursts. Allows for many different methods. A solution: Block-Normal generates triggers for more computationally expensive multi-interferometer analysis

McNabb GWDAW-8: Milwaukee,Wisconsin 3/15

LIGO

Block-Normal Overview

Baseband data into frequency bands Break data into blocks: Character of data changes between blocks. Character of data within a block roughly constant. Cut on “unusual” blocks : events Collect adjacent events into a cluster Look for coincident clusters to generate triggers

McNabb GWDAW-8: Milwaukee,Wisconsin 4/15

LIGO

Block-Normal is a spectral analysis

Data conditioning steps: basebanding, line removal, whitening Analysis carried out on separate bands Coincidence insists on events with same spectral character Choice of bands: Avoid violin modes which are too non-stationary to track note 900-930 Hz band for comparison with bar results Frequency(Hz) 128-192 192-320 384-512 512-640 704-1024 1065-1365 1408-1708 1758-2048 900-930

McNabb GWDAW-8: Milwaukee,Wisconsin 5/15

LIGO

Change Points

Characterize data using parameters of normal distributions: mean and variance. Find change points

the probability of data to either side being drawn from one distribution.

the probability of the data being drawn from two different distributions.

If change point found: (

). iterate over the two newly formed blocks. Second pass examining consecutive blocks: in the limited interval is change point still significant? (

)

McNabb GWDAW-8: Milwaukee,Wisconsin 6/15

LIGO

Blocks

blocks characterized by: start and end times mean(

• )

variance(

) events defined by: comparing block’s character to that of long (>50s) epoch (

• ✂☎✄

)

OR

• ✠
• ✂

• ✝

and

• ✝

are called event thresholds

McNabb GWDAW-8: Milwaukee,Wisconsin 7/15

LIGO

Triggers

Cluster adjacent events into a single cluster. Calculate calibrated energy of cluster.

L1 H1 H2 time Coincidences: L1 H1 H2 L1 H1 H2 frequency band

Do coincidence between detectors based on: frequency band time of “loudest” block within cluster. consistency of calibrated energy. consistency of duration. Clusters that pass get marked as triggers.

McNabb GWDAW-8: Milwaukee,Wisconsin 8/15

LIGO

Intermission

Block-Normal is a time-domain search for bursts that Breaks data into roughly stationary normal blocks. Identifies unusual blocks as events Triggers on coincident clusters of events in multiple interferometers. Characterizes triggers by Frequency, Energy, Duration, and Peak-time in each IFO. Parameters of the search are the frequency bands, the change point thresholds (

,

) and the event thresholds (

,

• ✝

)

McNabb GWDAW-8: Milwaukee,Wisconsin 9/15

LIGO

Change Point Thresholds

Change-Point false rate as function of acceptance threshold for different data segment lengths, variances

McNabb GWDAW-8: Milwaukee,Wisconsin 10/15

LIGO

Sensitivity/Effi ciency

Tune for best in-band signal sensitivity Model signal:

• ✟✂✁

• ✂

band 1:

Hz ,

ms Four-parameter efficiency model

• ✡

• ✁

Fit accommodates non-zero background

McNabb GWDAW-8: Milwaukee,Wisconsin 11/15

LIGO

Sensitivity/Effi ciency

0.2 0.4 0.6 0.8 1

E1/2 = (∫h2 dt)1/2

ε

H1 H2 L1

McNabb GWDAW-8: Milwaukee,Wisconsin 12/15

LIGO

Reconstruction of Simulations:

Timing: Event time: time at reconstructed signal peak Precision: Better than 8 samples for 50% of detected signals [(25,75) quantile range] Better than 40 samples for 90% of detected signals [(5,95) quantile range]

2 4 6 8 10 E1/2

h (h sec1/2)

Event time precision (samples) Lock 61 band 1 IFO H1 CG 576 Hz τ=100 ms ρA ρR [9 8] µT νT [2 2] Central 50 quantile 10 20 30 40 50 Eh

1/2 (h sec1/2)

Event time precision (samples) Lock 61 band 1 IFO H1 CG 576 Hz τ=100 ms ρA ρR [9 8] µT νT [2 2] Central 90 quantile

McNabb GWDAW-8: Milwaukee,Wisconsin 13/15

LIGO

Reconstruction of Simulations:

“Energy” Squared strain in event Reconstruction is very precise Reconstruction is accurate at level of the calibration systematics

Ei (h2 s) Er (h2 s)

McNabb GWDAW-8: Milwaukee,Wisconsin 14/15

LIGO

Conclusions

Block-Normal parameters: Change-Point thresholds control sensitivity to changes in mean, variance; specified by desired change-point false rate Event thresholds control sensitivity to differences between block, background character; specified relative to background mean and variance. Sensitivity: Comparable to existing burst search methods Reconstruction: Energy accurate at level of calibration systematics precise to greater than calibration systematics Time Resolution:

• ✁

samples

McNabb GWDAW-8: Milwaukee,Wisconsin 15/15