LSC-Virgo Burst Analysis Working Group

Navigation

Burst Group Home
Wiki
ExtTrig Home
CBC Group Home
LSC, LIGO, Virgo

Documents

'How-to' docs
Agendas/minutes
[2008, 2007, earlier]
Talks/Posters [pre-wiki]
Papers
Paper plans
White papers
ViewCVS

Investigations

Analysis projects
Old notebook [General, S2, S3, S4, S5]
Virgo workarea
External collabs

Review

Main review page
Telecons

S5 Qpipeline High-Frequency Review Telecon 7th of August 2008 11:00 Eastern

Agenda and minutes of the S5 Qpipeline High-Frequency Review Telecon for Thursday the 7th of August at 11:00 Eastern time.

Agenda and Contact Info

Agenda

Action items
  1. CorrPower tests
  2. update on astro waveforms
  3. Comments on report on Post.Proc. scripts (more checks/scripts)
  4. Table of review items
  5. AOB

Contact Info

  InterCall telecon service:

    USA (toll-free): 1-866-616-1738
    UK (toll-free): 0800 073 8914
    Italy (toll-free): 800-906-494
    Germany (toll-free): 0800-1014-907

    participant code: 251 288 9495#

Minutes

Minutes by Jonah
August 7, 2008

Michele walks us through his link:

http://www.ligo.mit.edu/~zanolin/high_freq_search/

Michele:
1) Waveforms by Biatti - you will see that my 
intervention was fairly mild.

2) You will see that I've explained conversion factor
for distances

3)  I will explain how to include errors
Generation error is about 10%
Error from only using one polarization - should
include systematic for this

Considering the script:

http://www.ligo.mit.edu/~zanolin/high_freq_search/astro_wf_treatment.m

Changed begginning paragraph to explain factor from geometric units
to physical units.  "The factor" is the geometric units length scale 
divided by 10 kiloparsecs expressed in kilometers.

This script can be run in matlab with input files
('D1_Hcross.dat');  
('D1_Hplus.dat'); 
('D4_Hcross.dat');  
('D4_Hplus.dat');
which live in the same directory - this produces all plots for today's call.


Consider http://www.ligo.mit.edu/~zanolin/high_freq_search/6.jpg
Blue is Biottie's + waveform
Red is windowed waveform with mean subtracted
Down sampling and low-pass filter gives green one - this is 
injected waveform.

See 2.jpg - Freq Domain
Blue is Original spectrum
Dip in red curve is at Nyquist freq.  
Dip in green is 8000 Hz - Nyquist
The physical part of green curve is between 1 and 6 thousand hertz.
One of the reviewers should play with the script, can blow up relevant
region

Title of each plot is hrss that arrives at earth in band of interest.  
Notice that + pol and x pol are different by 2 orders of magnitude.  

Plots 9 through 12 are results of filter applied to random noise.  
Blue is initial noise - green is final re-sampled one.  

Siong:  For clarity, these spectrums are square of FT

Michele:  Yes, you can see details in script.  I do 
a FFT, then take the absolute magnitude (this is amp, not
the square)  You also need a normalization factor for
sampling rate.  

Discuss the plots

Michele:  We will likely put these plots in a webpage.

Corr Power Test

Brennan:
http://emvogil-3.mit.edu/~bhughey/high_freq_search/corrver/
Average spectrums DO look white.  You can see notch filters, but that
is probably OK, since there IS a notch.  

Siong:  This looks good - I am satisfied about this.

Siong:  I assumed that H1H2 cuts are applied through CorrPower
- the amplitude cuts and the H0 cuts
In S2, S3, S4, when we used CorrPower, we usually make a cut
on the amplitude - h1 hrss w/in factor of 2 of h2 - and another
cut on r0 that h1h2 should not be phase shifted

Brennan:  No, we don't make Ro cuts, because of phase shift
in H1 H2 at high frequencies.  This is do to phase uncertainty
at high frequency.  The amplitude cut I never actually looked at - 
I can check that.  

Jonah:  We should look at how close the thresholds are to 
Gaussian noise thresholds, to evaluate how much the data 
could potentially be cleaned up by glitch rejection (such as 
amplitude cuts).  This could be part of sensitivity estimation
calculation.

Siong:  It might be better to do it with Q-pipeline triggers - 
more obvious relationship between threshold and Gaussian noise
FAR.

Jonah:  Yes, this is wise.

Siong:  Let's go on to Post-proc. scripts.

Matt and Brennan walk through Matt's issues and Brennan's 
responses.  For completeness, Brennan's e-mail response is
included at the end of thses minutes.  

http://emvogil-3.mit.edu/~bhughey/high_freq_search/morechecks.html

Siong:  It might be wise to move them to a different directory
and freeze the code.

Brennan:  Yes, this is wise.

Siong:  This looks good to me.  

Siong:  Let's look at rev. table.  I've added items
for sense. estimate and H1H2 only analysis.

Brennan:  Should have update on H1H2 analysis update next
week.  

Siong:  In past, we used wave-burst, corr-power, but 
there was no evidence of how to make sensitivity esitmate 
of corr-power.  We should maybe have a discussion with Laura.


12:17 Meeting adjourned

Attendance

Brennan Hughley, Jonnah Kanner, Matt Pitkin, Michele Zanolin, Ik Siong Heng

MEETING ADJOURNED

Brennan's e-mail about Post Processing Scripts

Brennan's e-mail response to concerns in Post-Processing scripts:

Checking scripts for efficency curves and output checking - It seems I
previously overwrote the results of this test with a later check, I've
rerun the original so we should now obtain the same
output. "Countallevents.script" checks the number of events going into
CorrPower and "countallinputs.script" checks the number of events going
out of CorrPower for each amplitude and waveform.  These should nominally
be the same.  However, there is a little glitch in the way the CorrPower
jobs are setup: there are 72 jobs distributed to the nodes and, if there
are less than 72 events, CorrPower decides to run all events in each of
the 72 jobs.  So if the number of inputs is greater than 72, the number of
input events and events actually run should match.  If the number of input
events is less than 72, the number of output events is 72Xthe number of
input events.  Generally, this will be at low amplitudes and these events
will be killed by the CorrPower Gamma cut, but if they survive they are
obvious enough to be dealt with by hand, since the numbers will be a very
obvious factor of 72 higher than they should be, and always divisible by
72.

getgoodevents.script:  Sorry, I must have changed this script when
checking the alternate set of cuts resulting from switching lengths of
coincidence window.  At any rate, none of my surviving events have
gamma<6.2, so this was obviously done as stated at the time.

Efficiency curve plotting scripts:  Yeah, I think the uncertainty formula
isn't actually appropriate, but this script will actually be superceded by
Laura's script for the paper, so I don't think this is a huge concern.

Also, executables combineseglists.out and rejectseglists.out are in
/archive/home/bhughey/noise_S5firstyearfinal/postprocess