S5yr2/VSR1 Coherent Waveburst Review
Checks of segment and livetimes for the cWB-LFS in S5 year 2
ChecksWe want to sign off on review issues related to the livetime and segment lists:
- Livetime: Verify reported livetime in zero lag for each network combination.
- Inclusion: Verify that all segments inside the final segment list were indeed analysed.
- Exclusion: Verify that no segments outside the final segment list were included in the final trigger lists.
ResultsIgor's DQ segments page lists the DQ segments and how they were applied. It also display the results of some nice sanity checks that demonstrate that the segments were applied correctly.
Livetime: Igor's page lists the livetimes after final DQ flags for
the H1H2L1V1, H1H2L1, and H1H2 networks post-CAT3. As a simple check, I counted the
number of injections listed as "processed" by cWB for each of these networks.
The injections were performed in a ``carpet bombing'' approach, covering the
entire calendar time of S5/VSR1 with a mean spacing of 100 sec.
Since the same segments are used for production and simulation jobs, we can
therefore estimate the livetime processed by simply counting the injections
processed and multiplying by 100 sec. Here are the results:
H1H2L1V1 H1H2L1 H1H2 Igor's livetime (days) 57.9 105.7 31.5 "injection livetime" (days) 57.72 105.69 31.37 ratio inj/Igor 0.997 0.9999 0.996
Note: there was confusion over the counting of the WNB injections, because those were done with two different amplitudes in the same MDC set. Omega counted the two amplitudes as separate injections of the same amplitude, while cWB counted them as one injection separate. The difference is evident in compareinj.txt. For this calculation, I doubled the counts of WNBs for cWB, making them similar to the Omega numbers. (Without this factor of 2, there would have been a 17% discrepancy in the livetimes.) For more details, see the discussion on the bursts email list here and up to here.
Inclusion: The simplest check is to verify that the efficiency curves for at least some waveforms go to 100% at large amplitudes (no no significant amount of data is being missed). I've verified this by checking BRST3 efficiency curves for each of the network combinations
analysed (H1H2L1V1, H1H2L1, H1H2V1, H1L1V1, H1H2, H1L1, H2L1).
Exclusion: Igor tested that no triggers in the final set came from
outside the final segment list. This was done for the production triggers (zero-lag and time-shifted) and the largest-amplitude BRST3 MDC triggers
from H1H2L1 times. No errors were found. There is no reason to expect
problems with any of the other networks.
(Note that even if extra livetime was unintentionally included, since we have no event candidates, it would make no difference to the final results anyway.)