Minutes of 2007-Nov-29 S5 QPipeline Review Teleconference
Shourov Chatterji, Jonah Kanner, Isabel Leonor, Dave Reitze
Minutes by Jonah Kanner.
Dave: We were supposed to have familiarize ourselves with basics of search. Shourov should pick up where he left off. Shourov: Did you read through chapters 3 and 5? All: We read much of it. Jonah: How does clustering work in Q-Pipeline? Shourov: Clustering is not a part of the standard q-pipeline. S-G are minimum t-f uncertainty basis. For a non-localized signal, we hope to represent it with the least number of tiles possible. Minimum t-f tiles lessens need for clustering. Clustering algortihm is under development - not part of S5 search. We found that, for inspirals, can improve search if you do a good job of clustering. Paper to come out soon. SG detection efficiency can be damaged by clustering algorithms. On the other hand, current Q-pipeline does very well for non-localized signal. The current problem is an OR problem - any one tile can trigger a detection. For insprial, ~50% of energy can be seen in one tile. Dave: What wave-forms do you use for testing Shourov: Standard mdc's - we use a variety Dave: What is freq. band of search? Shourov: 48 Hz - 2 kHz ??? Dave: Is it true that noise is constant over freq. band? Shourov: No, but that "assumption" is never used in search. Dave: What is the meaning of the statement: Q-Pipeline does not need to worry about noise when choosing basis? Shourov: Basis is not astrophysical. Search "looks" for signals that are S-G AFTER whitening. Jonah: Why does coherent-pipeline threshold "early" Shourov: We don't do it that way anymore. In current algorithm, weighting is done in frequency domain, but summing is done after Q-transform. Shourov: I'll go through thesis, chapter 5. Dave: Why a bisquare window over a Hann window? Shourov: The bi-square window was simpler analyitically when finding the normalization factor. Otherwise, they are very similar. - on pg. 122, we see important restrictions to avoid aliasing. Q > ~3. Code issues errors if constraints are violated. For higher values of Q, get narrower tiles, that can go closer to Nyquist frequency. Isabel: Does mis-match metric still apply for bi-square window? Shourov: Becuase windows are similar, the metric is negligibly affected. Simulations and Matlab experiments show this to be so. -Normalization is chosen so mag (h) is recovered by Q transform coefficient -Further, the normalization leads to expected coefficient = power spectral density for noise -characteristic strain amplitude = hrss - Can predict algorithm performance from first principles - Statistics: If you meet some requirements, you have enough statistics that noise tiles follow nice distributions. You are able to add together tiles from different detectors, and know they share a distribution. - Standard normalized energy for Q-pipeline is defined. - Normalized energy is simply related to significance and SNR Dave: I need to digest chapter 5 better - can do by next meeting. Can we meet at GWDAW? Maybe after dinner or go out to dinner. All: Yes. Plan to meet at GWDAW. End