Date: Wed, 13 Aug 2003 22:07:20 -0700 (PDT) From: John Zweizig To: Keith Riles Cc: Gabriela Gonzalez , Peter Fritschel , Brian O'Reilly , Rauha Rahkola , Robert Schofield , Peter Shawhan , Stan Whitcomb , Daniel Sigg , Rana Adhikari Subject: RMS and Saturation vetos for H1 Hi Keith, I have created a list in http://www.ligo.caltech.edu/~jzweizig/S2_Data_Quality/H1-veto-report.txt of H1 data segments that I think should be flagged in the data quality segment list. There are three sections: 1) Adc saturations 2) Low frequency band noise 3) Mid-upper frequencyt band noise. The meanings are as follows. I) Adc saturation segments. I ran all the S2 raw data through my ADC backout program. This uses the two phases of AS, REFL and POB to reconstruct the digitized values of AS_I, AS_Q, POB_I, POB_Q, REFL_I and REFL_Q. I then look for the values to repeat themselves ~5 or more time OR for the values to overflow the 16 bit ADC. In general there are "accidental" repetitions every few thousand seconds so I have only included those cases where a value was repeated many times, a repetition was seen on multiple channels or an overflow condition occurred. THe list gives saturation times, segment number and the channel(s) seen to saturate. The overwhelming majority of these times correspond to the last 1-2 seconds of a lock segment, and should probably be interpreted as a lock loss time stamp. II) Low frequency Band noise I have looked through the data quality trends for the 10-100Hz band and selected all minutes where the largest 4-second sample has an RMS^2 > 150. This makes a cleaner discrimination than the RMS averaged over 1 minute. For each such segment, I have listed the start time, the duration, the segment number and a few comments. The use of the maximum 4 seconds is a little dangerous in that it would be more likely to veto a GW burst than would the minute average, but since the cut is enormous compared to the IFO sensitivity in the sensitive band, this shouldn't be a worry. I have noted in the comment field each segment that overlaps those that Gaby identified as being peak-peak outliers. As can be expected there is a big overlap in the segments defined by these two techniques. I am a bit ambivalent about using these segments as a veto. On one hand, the data defined by these segments are very noisy and you can't help but worry that the noise will find a way to leak through to the more interesting bands. On the other hand, there is no evidence for correlated noise in the 100-7kHz bands and no indication that glitches are more common in these segments than out of them. I guess that the decision to veto should be left to the individual analysis groups. III) Mid-Upper frequency noise segments The third table lists segments where the maximum RMS^2 in bands 1-4 is greater than the following: Band Frequency Threshold 1 100-200Hz 0.02 2 200-400Hz 0.0005 3 400-1kHz 0.0005 4 1k - 7kHz 0.002 Each segment is defined by a start GPS, a number of minutes duration (note the non-standard time units), the bands affected, the segment ID and a comment. I have gone through the elog for all of these segments to try to identify the cause of the disturbance. In those cases where there appeared to be something pertenent I tried to summarize it. I have also indicated segments that appear to have large 6-sigma glitch rates as seen by DataQual. For the most part these segments are too long and strong to have been caused by GWs, but a better understanding of the cause of the excess noise in each of these segments would be useful in deciding which to use as vetos. I haven't identified the sources of the excess noise in many of the segments, but they certainly include some or all of the following: * Adcu reboots * Spontaneous broad-band noise (BBN, See E-logs by D. Sigg Feb 15 9:17 UTC and by P. Shawhan Mar 1 3:31 UTC) * Violin mode ring-ups. In two cases the violin modes rung up spontaneously. * Lsc timing Skew (see E-log by P. Shawhan Mar 2 2:52 UTC)? Ciao, John