## Minutes of 2008-Jan-04 S5 QPipeline Review Teleconference

### Attendance

Shourov Chatterji, Jonah Kanner, Isabel Leonor, Dave Reitze

Minutes by Isabel Leonor.

### Discussion

* We started approximately where we left off at the last meeting held in Royal Sonesta Hotel in Cambridge, MA, December 12, 2007, i.e. with the flow diagram of the H1H2 pipeline which can be found in the QPipeline technical documentation: http://ldas-jobs.ligo.caltech.edu/~qonline/s5/1.0/documentation/technical/ During the last meeting, Shourov said he would upate this flow diagram to reflect the current pipeline. This has not been done yet; Shourov thinks it might not be trivial to make this reflect the current pipeline, because much of it is contained in the "Q transform" box. Shourov explained what was going on in the Q transform box to the three confused reviewers (and also in response to Dave's question about Eqs. 5.96-5.103, specifically Eq. 5.102, in Chapter 5 of Shourov's thesis): Shourov: The current procedures incorporated into the Q transform algorithm, and the motivations for the changes made, are summarized here: http://ldas-jobs.ligo.caltech.edu/~qonline/s5/1.0/documentation/technical/update.html.20070829 In this new scheme, the coefficients of coherent combinations are calculated after the whitening stage using a high-frequency resolution (implemented at 1/64 Hz). After the coefficients are calculated, the Q-transform of each coherent combination can be calculated, and the corresponding coherent and incoherent energies calculated. It was also noted that in the jargon, E_coherent is also known as E_null. This new scheme also has the advantage of implementing big savings in memeory. In this new scheme, the Q tranform box outputs 4 channels: coherent signal energy, incoherent signal energy, coherent null energy, and incoherent null energy. Isabel: Maybe it's possible to draw a diagram for the Q transform box. Shourov: That's possible. * New topic about how vetoes are implemented in QPipeline. Dave: Worries about correlations in H1H2 pipeline. Where in the pipeline is this dealt with? Shourov: Vetoes are based on null energy, i.e. demand significant null energy to veto signal tiles. Refers to "collocated analysis" section of technical documentation (link given above), specifically the equation which sets the threshold for the null stream significance: Z_ > alpha + beta*Z_0 Dave: How are alpha and beta chosen? Shourov: alpha sets false veto rate, a value of 10 corresponds to about a 1-Hz tile rate. The term beta*Z_0 accounts for calibration uncertainties; beta is square of calibration uncertainty. Values chosen are beta = 0.01-0.05. Jonah: From experience with XPipeline, expects beta to be more like 0.2 or 0.5. Shourov: There are differences in QPipeline and XPipeline, e.g. XPipeline does not use alpha, and XPipeline also looks at many sky positions. Jonah: By making beta lower, then more likely to veto. Surprised that XPipeline does not use lower value. Seems like beta is a strong veto since it has a low value for QPipeline. To illustrate the effectiveness of the null stream veto, Shourov refers to section of technical document titled "Effect of H1H2 null stream veto on playground data", which has plots showing the distribution of H1H2 triggers before and after the veto was applied. Back to discussion of "collocated analysis" section of technical document. Shourov discusses the equations defining time-frequency overlap of tiles. Jonah: How do you know tuning parameters alpha_t and alpha_f? Shourov: The idea was to have minimal tuning for the parameters, so pick something sensible. This can be discussed some more in next week's call. Shourov then goes again to flow diagram of H1H2 pipeline. The dotted lines indicate the post-processing part of the analysis. Here, test for overlaps are done. The plan is to eventually incorporate this into the full matlab pipeline. * New topic on poorly localized bursts and clustering. Dave: QPipeline does not do as well for poorly localized bursts. Should we focus on investigating this case? Shourov: Studies like this have already been done. A summer student worked on implementing clustering extension to QPipeline; results of study showed promise for doing clustering. The simulations included inspirals, white noise bursts, sine-gaussians, ringdowns. Clustering has advantages and disadvantages. One type of clustering studied was density-based clustering, where a metric was used to describe distance between tiles. This kind of clustering worked well for inpirals. Results showed that, for inspirals, the false alarm rate dropped by an order of magnitude after clustering was implemented, given the same efficiency (50%) and the same SNR (25?). But the results also showed decreased performance when applied to sine-gaussians, because the clustering threw out single tiles. Another type of clustering, hierarchical clustering, does not affect sine-gaussian performance. Clustering can be revisited for S5, although it is not expected to give a big increase in sensitivity. For the future, the effects of clustering on different waveforms have to be studied in detail. * It was agreed that the next review telecon will be on January 9 (Wednesday), at 2 PM Eastern.

$Id: minutes_20080104.html,v 1.1 2008/01/31 12:41:56 shourov Exp $