Transcript PPT - ACAT`2002

The VIRGO experiment
Data analysis software tools used
during Virgo engineering runs
Review and future needs.
D. Buskulic, ACAT 2002, Moscow
The Virgo experiment

French-Italian Collaboration, 11 laboratories
•Firenze-Urbino
•Frascati
•Napoli
•Perugia
•Pisa
•Roma





•LAPP Annecy
•IPN Lyon
•OCA Nice
•LAL Orsay
•ESPCI Paris
Situated at Cascina, near Pisa in Italy
Arms length : 3 km
Sensitivity frequency domain : 4 Hz - few kHz
Best sensitivity: 3.10-23 Hz-1/2 (at 500 Hz)
Full Virgo commissioning starts beginning 2003
D. Buskulic, ACAT 2002, Moscow
Keeping really quiet
Reach this sensitivity : lower all the noises
• Data sampled at
20 kHz
• Continuous stream
of data
h (Hz-1/2)

f (Hz)
D. Buskulic, ACAT 2002, Moscow
Fighting the noise

Mechanical systems : pendulums and resonating
systems
Most Virgo systems
(mirrors, benches) are
suspended to superattenuators
-> seismic noise attenuated
by at least a factor 1012


Optical systems :
Ultra-stable laser
Electronic systems :
Fighting all the noises !
Example : fighting thermal noise
The more energy is concentrated
in the resonance (high quality
factor Q), the better is the
Low Q sensitivity outside the resonance
High Q
w
D. Buskulic, ACAT 2002, Moscow
Suspended injection bench
Suspended mirror
Laser lab
Detection bench
D. Buskulic, ACAT 2002, Moscow
DAQ System
The Central Interferometer (CITF)
Commissioning of the CITF :
February 2001-July 2002
Try to lock, test systems,
build tools
D. Buskulic, ACAT 2002, Moscow
DAQ System
Engineering runs




Already 4 engineering runs with the central part of
the interferometer (CITF)
3 Days each
1 TB of collected data each time
Learned a lot about the machine




Locking procedure, stability of operation
Collection of data
Data Display/Analysis
Analysis tools :




Data Display
Matlab
VEGA
A little bit of PAW
D. Buskulic, ACAT 2002, Moscow
Data acquisition system and data rates
Fast Digitization
Locking and Alignment Servo-Loops
Photodiodes
Readout
Global
Control
Timing
Crate
Damping and
Control
suspensions
Calibration
Fast Digitization
or Local Servo-Loop
Environment
Monitoring < 1Hz
Frame Builder,
Local Control
Slow Monitoring
Station
GPS
Fast Frame
Builder
Main Frame Builder
Detector Monitoring
Timing System
Slow Frame
Builder
Frame Building
Frame Processing
GPS signals
Timing Information
Data
Archiving
Servo-Loop DOL (Optical Fibers)
DOL (Optical Fibers)
SMS format (Ethernet)
Frame format (Ethernet)





Control signals
Monitoring signals
Interferometer output
Reconstructed signals
Trigger signals
H
Reconstruction
Data
Quality
On-line
Processing
Data
Distribution
stable flux of
4 MB/s (9 MB/s uncompressed)
100 - 125 TB/year
D. Buskulic, ACAT 2002, Moscow
The Frame format




GW data : continuous stream, divided into "Frames"
1 Frame = time slice of all interferometer data :

ADC channels, Monitoring, reconstructed h, etc…

4 MB
Ability to preselect data :

1 Frame is then ~ 100 KB on average
1s
Time
Time
Common format needed for exchange of data
Adopted by all GW experiments in the world, including bar
experiments
D. Buskulic, ACAT 2002, Moscow
Data collection and distribution




DAQ collects Frame pieces and builds output frames
Online processing produces digested data (trend, 50 Hz)
Sent to various data displays and stored on disk
After engineering runs, raw data transferred to Computing centers
(CCIN2P3 Lyon and Bologna) by network
D. Buskulic, ACAT 2002, Moscow
Offline Computing

Size of data set

-> distributed computing
Computing Center
Bologna
CCIN2P3 Lyon
Central
bookkeeping database
VIRGO Experiment
(Computing Resources Cascina)
D. Buskulic, ACAT 2002, Moscow
Data analysis challenges

Pulsar searches
• Rotating neutron stars that have a small
dissymmetry (ellipticity of 10-6 ) generate a
gravitational wave
• Very weak signal (h ~ 10-25 ) buried in noise
• But very long (105 years)
• Need to integrate very long signals (months) to
extract signal from noise



Long quasi periodic signals
Variation of frequency due to earth movement, earth-star relative
movement and position
Signal characteristics dominated by
• Star parameters
• (rotation period, quadrupolar moment, binary system ?)
• Position in the sky
D. Buskulic, ACAT 2002, Moscow
It's a hard analysis : all sky pulsar search



For one set of parameters (position in the sky, star period…)
do a search in the output signal
Number of different cells in parameter space as high
as 1029 !
To keep with incoming data rate


Brute force method would need 1015 TFlops of processing
power!
With hierarchical methods and Hough transforms, need
1 TFlops
• Still a huge processing power

Able to distribute computing : each node will treat a
frequency band and/or a subset of parameter space
D. Buskulic, ACAT 2002, Moscow
Data analysis challenges


Bursts (supernovae…) -> short signals
Coalescing binary compact objects
(neutron stars or black holes)
Courtesy M. Ruffert, MPA Garching
• Binary neutron stars that coalesce (merge)
after a fall down inspiraling phase produce GW
• inspiraling + merger lasts from a few seconds to a minute
• h ~ 10-23 - 10 -22 still small

Shape of a calculated signal:
chirp with amplitude and frequency
growing in time

Depends mainly on
the masses of the two stars
D. Buskulic, ACAT 2002, Moscow
Binary coalescence search
Theoretical signal
(template)
Experimental (noisy) signal

Optimal filtering
(weighted intercorrelation)

Optimal filtering technique
The search has to be done


For all possible theoretical signals (templates), i.e. for all possible physical parameters
of the system
5.105 templates (waveforms) in the parameter space of interest

-> 300 Gflops needed to do a full search

Even more if take into account more physical parameters


Result :
Theoretical signal
present ?
Alignment of stars spins, ellipticity of orbits, etc…
Easily distributed : a computing node may treat a subset of templates
D. Buskulic, ACAT 2002, Moscow
Data visualization : the Data Display



Home build online display and monitoring tool
Access files or remote frames
Channel browser
D. Buskulic, ACAT 2002, Moscow
Data visualization : the Data Display


Read and display frame files content
Receive and display frames sent over network
Uses
ROOT Libs
(plots, display)
Xforms (GUI)
D. Buskulic, ACAT 2002, Moscow
Data analysis tools

Matlab



Many people in our community used to it
Rich set of signal analysis functions/tools
Difficulty to easily handle the size of data sets available
Night-day seismic noise on site
D. Buskulic, ACAT 2002, Moscow
Data analysis tool : VEGA

VEGA

Offline data handling/analysis environment based on ROOT
• Scripting : CINT
• Data visualization
• Adapted ROOT to handle time-dependent data
(up to a few million points)
• Signal processing
• Interfaced to external libraries (FFTW, SigLib…)
• Home made signal processing lib
D. Buskulic, ACAT 2002, Moscow
Data analysis tool : VEGA
• Data handling
• Meta information in one
place, data in another
• Access frames through a
"channel"
• Can build localy a metadatabase
which is used as an index to
handle a local set of files
http://wwwlapp.in2p3.fr/virgo/vega
D. Buskulic, ACAT 2002, Moscow
Trend data visualization

Trend (downsampled at 1Hz) data displayed on the web in quasirealtime
Uses
• Local metadatabase
• Display by VEGA analysis tool
• Shell scripts
D. Buskulic, ACAT 2002, Moscow
Cooperative analyses and data exchange

The same GW can a priori be seen by all detectors on earth
• Depends on the orientation and amplitude of the wave

Cooperative analysis allows to

extract more information from the signal
• Physical parameters
• Direction of propagation

Need to exchange data

-> same data format… OK, we have the Frame format

Already exchanged some online monitoring data between LIGO
and VIRGO in quasi-real time

Wish to use GRID tools for data exchange
Still problems for GRID middleware compatibility
DataGrid (VIRGO) / GriPhyN-iVDGL (LIGO) interface

D. Buskulic, ACAT 2002, Moscow
Developments around GRID


Use of European DataGrid
Test of a binary coalescence search


Each job treats one subspace of all templates.
Test of a periodic sources search


Hierarchical approach which alternates an FFT step and a Hough
transform step
Each node analyzes a frequency band
 Verified that multiple jobs can be submitted and the output
retrieved with small overhead time
 Computational grids seems suitable to perform data analysis for
coalescing binaries and periodic sources searches

See " A Grid Approach to Geographically Distributed Data Analysis for Virgo", Palomba, Tortone and al.,
GWADW 2002 Workshop
D. Buskulic, ACAT 2002, Moscow
Summary

GW data analysis needs :




Data analysis tools used during VIRGO engineering runs




GW data analysis produces large amounts of data
(in the 100 TB/year range)
Data is continuous -> Frame format
Needs a lot of computing power (TFlops)
Data Display
Matlab
VEGA
Preparing the future :

Needs to exchange data among experiments
• Some exchange already done

Efforts on the way to use GRID
D. Buskulic, ACAT 2002, Moscow