LSC-Virgo Burst Analysis Working Group
Review Committee Meeting Monday 6 February 2006 09:00 PST / 12:00 EST
Minutes: Monday 6 February 2006 09:00 PST / 12:00 EST
- Continue the review of the GRB GWB search
We continued to go through Isabel's technical document
- In response to last week's requests, Isabel has posted data conditioning sanity checks
- Consider spectral leakage. Certainly some of these plots show
low frequency structure, e.g., Plot 9.
However, it is not so clear if there is spectral leakage of low frequency
content in, e.g., Plot 2.
- The fact that 4-second DFT and 1-second DFT look similar is
encouraging but not yet conclusive. It would be nice to have smoother
spectra to compare to or perhaps calibrated spectra so they can be
compared to official noise curves (to see if low frequency structure
- Also, will need similar plots for H1, H2, L1, for representative
S2, S3, and S4 data (nature of ASQ changes from run to run and IFO to
- ACTION Jolien will figure out exactly what should be plotted.
- Consider time series continuity. Most of these plots,
e.g., Plot 4 seem quite continuous.
However, a couple, e.g., Plot 8 and
Plot 9 show what appears to be a
discontinuity on the transition boundaries.
- ACTION Need to examine these transitions more. Perhaps there
is significant fluctuations in the spectra near these transitions
that is causing the whitening of one segment to be significantly
different from other segments.
- Idea: when making plots like these, it may be useful to overlay
two time series that are processed similarly but on slightly shifted
data so that the transitions occur at different places. Then
discontinuities would be more distinct.
- Several issues for future analyses were raised. These include:
- Overwhitening: would naturally suppress lines and focus analysis
to the most sensitive bands while rejecting noisy bands.
- Why cross-correlation? Another possibility would be to take the
minimum of the two IFO's autocorrelation. Which is better?
Jolien has posted his GWDAW-2000 presentation which argues that
the minimum-method might be better for sufficiently non-Gaussian
[ PDF ].
- More time-domain filtering. Why is the band-passing done in
the frequency domain?
- Is there a better way to do the whitening? E.g., power spectrum
averaging (either time average or frequency bin average) to get a more
stable spectral estimation.
- Calibration: are the most recent version of S4 (V04) calibration used?
- No: at the moment V03 calibration is being used.
- How much difference? Only seems to be a couple degrees in phase.
- Amplitude error (15% in Livingston) is unimportant for the search
algorithm but it is important for determining sensitivity.
- Options: Either re-run analysis using V04 calibration for
S4 or at least redo the simulations to re-evaluate sensitivity. (I.e.,
inject with V04 calibration, search with V03 calibration.)
- Calibration: is DARMERR or ASQ being used? A: ASQ. Should move to
DARMERR in future.
- Calibration: are there any sign errors between H1 and H2 (note: search
is insensitive to sign flips between H and L)? Almost certainly not.
ACTION Need to double check with Mike Landry that the calibrations
that were used for S2, S3, and S4 had correct relative sign between H1 and
- Why was the frequency band 40 Hz to 2000 Hz used? A: 40 Hz is
pretty arbitrary. 2000 Hz is the same as with the untriggered search.
- Why does the "whitened" spectrum look bad below 70 Hz? Not clear.
(Not sure what data set the plot is from.) ACTION Isabel will
investigate a bit.
$Id: minutes-2006-02-06.html,v 1.2 2006/02/10 20:44:49 jolien Exp $