From kats@ligo.mit.edu Mon Mar 31 17:53:05 2008
Date: Mon, 31 Mar 2008 17:50:02 -0400 (EDT)
From: Erik Katsavounidis 
To:  , Glitch Working Group 
Cc: Rainer Weiss 
Subject: low frequency, low Q glitching of the instruments

Hi all,

I have done a more in-depth analysis of the low frequency, low-Q glitches
in the H1 and L1 instruments during the first year of S5. There are some
interesting findings I would like to share with the group.

Following the LSC-Virgo collab meeting at Caltech, Rai Weiss had suggested
to look for the presence of low frequency, low-Q glitches in the
instruments associated with large accelerations of the test masses. Rai
has a scenario of what might be contributing to excessive such glitches-
it is associated with the repositioning of the wire within the cylindrical
groove that holds it in place around the mirrors. To pursue this study I
thought of using the low frequency, low-Q triggers Shourov provided for
the special veto tuning we were discussing the weeks prior to the Caltech
meeting. These are 50-150Hz, Q<10 and SNR>=8 glitches Shourov described in
http://ldas-jobs.ligo-wa.caltech.edu/~qonline/H1/collect/filtered/
http://ldas-jobs.ligo-la.caltech.edu/~qonline/L1/collect/filtered/

In order to get a feeling of the features of these triggers in H1 and L1
cumulative over the first year of S5, you can check out these two links:
http://lancelot.mit.edu/~kats/gdata2/s5/lowf/firstyear/h1/allt/summ-hist.gif
http://lancelot.mit.edu/~kats/gdata2/s5/lowf/firstyear/l1/allt/summ-hist.gif

The lower-right plot "Time within segment" in both of the above links
depicts the time of an event after the start of the segment has been
subtracted, normalized to the duration of the segment. It thus assumes
values from 0 to 1 corresponding to the "start" and "end" of a segment. It
has the convinience of allowing the superposition of any segment,
independent of its duration (but it also has the caveat of not being able
to read much out of its binning). These plots were rather a surprise to
me, as they suggest that there exists *excessive* glitching of the H1 and
L1 instruments immediately following the beginning of a segment.

Excessive glitching at the end of segment is also present, but there is
nothing new to it; we have seen this in similar plots for KW and on that
basis have used a pre-lock loss flags in past runs, including S5. It is
however the first time in recent running that we are seeing this so
pronounced in the beginning of a lock stretch.

I have done a more quantitative description of this higher rate of low
frequency/low Q glitches and comparison with full-band glitches, including
trying to understand why we missed it in our day-by-day analyses during
S5 (...) The summary of findings are in an e-notebook entry in:
http://www.lsc-group.phys.uwm.edu/cgi-bin/bag-enote.pl?nb=burs5veto&action=view&page=30

Day-by-day and global (full year) cumulative statistics and plots of these
low frequency, low Q glitches are in http://lancelot.mit.edu/~kats/gdata2/s5/lowf/
There are tons of plots there that you can browse with your morning coffee.
For comparison, you may browse identical kind of plots on a day-by-day
basis and cumulative over S5 for KleineWelle at
http://lancelot.mit.edu/~kats/s5/kw2/.

I invite people to give an independent look at the data- if what we are
seeing there is right, we should consider adopting a O(100sec)
post-start-of-segment flag in the second year analysis. Moreover, in the
above day-by-day analysis you will see that I have done a coincidence
analysis for the H1 and L1 triggers used in this study (including 100 time
slides). I suggest we look at the 'zero lag' box of this, once we complete
the remaining 5% of the H1L1 livetime coherent-WaveBurst is about to
unveil.

Comments, questions and suggestions are as always welcome.

Regards,
--Erik

E. Katsavounidis
Massachusetts Institute of Technology   |         LIGO Lab, NW22-295
185 Albany Street                       |         tel#: 617 258 9218
Cambridge, MA 02139                     |         fax#: 617 253 7014