S2 / S3 Investigation Reports

• Data quality -  Investigations: ( S2 , S3 ) Segment repository (S2 , S3 ) (Vern Sandberg, John Zweizig, Keith Riles)
  Vern reported that he and Ben Johnson have made further investigations on the low-order-byte corruption and have identified a particular interrupt line (IRQ2) in the VME crate as the problem. The clock card actually works without trouble with other interrupt lines; we were just unlucky in the one we chose. So the clock cards can (it's thought) be safely used in the future with other IRQ choices. The crate with the problem was the central crate that ships data to frame broadcasters. For up to 25 microseconds every 1/16 second, the corruption would occur, the variation coming from the response time of the CPU. As a result, it's not yet clear which channels are affected on any transfer and the synchronization is not perfect. As yet, no easy identification has been found for which bytes are affected based on the hardware investigations, but Ben continues looking at data channels for useful patterns.
  
  John intends to try to simulate the effect of the byte corruption on AS_Q and will talk with Vern and Ben to try to make the simulation as faithful as possible, including timing jitter.
  
  Keith reported that the S3 version 1 data quality segments are now released, but the long-delayed S2 version 4 segments are still not yet ready.
  
  
• Calibration    -- Final S2 report   -- Study of fast fluctuations (Gaby Gonzalez, Mike Landry, Brian O'Reilly)
  Brian reported the creation of a nearly final draft (now in the DCC) of the S2 calibration report . Additional changes planned include cross-checks on the sign of the calibration, more information on the alpha and beta derivations, more plots, and a section on the actuation function. Keith wondered if a publication was planned. The answer is yes, but it will be a while most likely.
  
  Gaby reported on recent studies of fast fluctuations in calibration during S2. She reminded everyone that a scale factor error on alpha leads to a frequency dependent error in the calibration function, making the effects of rapid fluctuations worse on signal detection efficiency. A useful tool in the studies is the geometric mean of the average carrier power in the arms and the power in the recycling cavity sidebands, which should govern overall optical gain. A strong correlation is indeed seen between that power measure and the strengths of detected calibration lines. Line strengths in L1 and H1 were weak at the start of S2, but boosted part-way into the run to reduce statistical uncertainties in extracting strengths. In the case of L1, there is evidence that with the increased line strength, statistical errors were low enough (0.7%) to see effects of fluctuating calibrations on minute time scales. The statistical errors in H1 were still large enough (3%) at the end that intrinsic fluctuations were not apparent. A study of data during a short period on L1 with very large calibration line strength confirmed 10-15% peak-to-peak intrinsic fluctuations on time scales of a few seconds (micro-seismic) to 0.5-1 second (stack resonances). It appears that one can track the microseismic fluctuations reasonably well with the power measure (good coherence with calibration line), but the 1-2 Hz disturbances don't always show good coherence, and even when the coherence is high, the transfer function between the two is not flat, as one would have hoped. In addition, the power measure has trouble on long time scales (~hour), presumably because spob_mon is an imperfect measure of true sideband power
  .
• Analysis of S2 inspiral injections (Stephen Fairhurst)
  Steve has carried out detailed comparisons of hardware and software injections, using the same waveforms and checking reconstructed inspiral effective distance against intended injection distance. In general, the software injections agree well (within a spread that increases with effective distance, i.e., with decreasing injection strength, as expected). But the hardware injections are prone to larger spreads and greater reconstructed distance (weaker inferred strength), with the disparity between software and hardware most pronounced for strong signals. A natural explanation is that the software injections are injected and reconstructed with identical assumed calibrations, while the hardware injections are subject to less well known calibration during injection. To test this idea, Steve tried software injections for 2048-second data segments, where he used the alpha/beta coeffiicients for the first minute of the segment throughout the segment, thereby inducing smearing. Qualitatively the smeared software injections do give better agreement with what is seen in hardware, but residual discrepancies remain, particularly in H1 and L1. Steve suspects measurement statistical error is the culprit for H1, while intrinsic calibration variation is more suspect for L1 (Gaby agrees). Studies of chi**2 match between inspiral template and injection reveal a similar pattern, with a residual disagreement in L1 that is plausibly attributed to intrinsic calibration variation. Steve plans next to try injecting chirps with alpha varying during the chirp, to test this hypothesis. Gaby volunteered to check with her code for actual calibration variation during the times of the hardware injections Steve has been studying.
  
  
• Timing (Szabi Marka, Daniel Sigg) - No update.
  
  
• Glitches / Burst/Inspiral search group vetoes (Laura Cadonati, Erik Katsavounidis)
  Laura gave an update on glitch studies in the burst and inspiral groups for both the S2 and S3 runs. In brief,
  S2:
  
  • Verified full S2 data consistent with playground w.r.t. glitchmon triggers
  • No veto being used in high-threshold untriggered burst analysis (although L1:AS_DC works well for WaveBurst)
  • Working on veto strategy of low-threshold analysis - four WB events could well be consistent with common H1/H2 acoustic source.
  • Some strong inspiral triggers in H1 correlated with glitches in AS_I, but not safe or efficient to use AS_I as veto.
  • POB_I looks promising for vetoing triggers based on post-newtonian expansion BH-BH inspirals.
  • But the more generic BCV templates (no chi**2 cut is practical) seem more prone to strong clustering of large glitches
  S3:
  
  • Focussing early work on online triggers from glitchmon to get an idea of promising channels to pursue offline
  • Starting to look at inspiral triggers in playground. Noise looks better than in S3 w.r.t. a BNS search.
  • Some loud triggers correlated with seismometers
  • PTMon getting exercised
  Keith asked about the 4 WB events. Likely to be acoustic or merely cannot be ruled out as acoustic? Robert: not clear, but would be leery claiming  anything after seeing the WaveMon AS_Q SNR vs accelerometer SNR. The 4 events lie near a diagonal band consistent with acoustic disturbance.
  

• Interchannel correlations   (Nelson Christensen)
  Nelson gave a run-down on recent S3 correlation studies between H1:AS_Q and microphones, accelerometers, and seismometers.A variety of coherence peaks are seen (0.01 Hz resolution), particularly between AS_Q and microphones. In some cases, no apparent peak is visible in AS_Q's spectrum. Nelson's undergraduate is putting together ascii and html tables of identified peaks in coherence to serve as possible frequency range vetoes for the pulsar group. They are starting with S3 H1 but will do all three IFO's for S2 and S3. Keith suggested going after S2 IFO's once S3 H1 is done.
  
  
• Violin modes (Sergey Klimenko, Jason Castiglione) - No update
  
  
• Bilinear couplings (Steve Penn) - No update
  
  
• Correlated inter-site environmental transients & local environmental disturbances  (Robert Schofield)
  Robert cited the recent work by Vern and Ben as a healthy reminder that chronic problems can prove solvable. It is hoped that fixing the clock card problem will lead to a reduction of the long-standing 16 Hz peaks in LHO/LLO coherences. After overcoming some technical problems, Robert is now running his S3 PEM correlations code over the dedicated PEM reduced data set.
  
  Robert has been studying the effects of dust on AS_Q. He has verified that only dust at the dark port table seems to matter. It will be important to monitor it carefully in future runs. A new enclosure with purging seems to reduce glitch rates to neglible levels after dust is allowed to settle. Unfortunately, it is not straightforward to identify dust glitches directly from their effect on AS_Q, since the response is very similar to that for a step function disturbance. An earlier hope that one could use wave front sensing photodiode signals looking at the same beam did not pan out. Similarly, AS_DC/AC do not seem effective. For now, Robert recommends a 10-20 hour epoch veto after anyone has entered the dark port enclosure.
  
  The 60-Hz disturbance from pulsed heating coils that Robert has reported on in the past is being attacked on several fronts. One approach being tried is continuous heating with cold water modulation. This method, though, still contributes a steady 60 Hz signal that contributes about 30% of the ambient 60 Hz field in the LVEA. Another approach is redesigning the heating coil shape away from a loop toward something with less inductance, and another is variac control of the coils.
  
  Dewar lock-dropping glitches continue to be a headache. An idea to insulate the dewars seems to be proving too expensive. Robert has verified that dewar glitches seems to be occurring at Livingston too.
  

A.O.B.

• John announced that the compiler on the saiph DMT machine at Caltech was recently changed, and DMT code is not yet working again because the LDAS FrameCPP component is incompatible with the compiler. No observatory DMT machines are affected
 
• The next detchar telecon will be in two weeks: Friday May 7 at 1:30 EDT. The main subject will be DMT software status.