Minutes of Detector Characterization Teleconference
(May 7, 2004)
Present:
Baton Rouge: Giovanni Santostasi
Caltech: Ron Drever, Szabi Marka, Peter Shawhan,
Patrick Sutton, John Zweizig
CSUDH: Ken Ganezer
LHO: Mike Landry, Vern
Sandberg
LLO: Brian O'Reilly
Louisiana Tech: Natalia Zotov
LSU: Gaby Gonzalez
Michigan: Keith Riles
Oregon: Ray Frey, Rauha
Rahkola, Brian Stubbs
Penn State: Mike Ashley
Salish-Kootenai: Tim Olson
UWM: Xavier Siemens
S2 / S3 Investigation Reports
-
Data quality segment repository -- S2 update ) (Keith
Riles)
KR reported release of version 4 S2 data quality segments with
-
Better compilation of times of absent or low calibration line injection
-
"Growly" periods of high 200-400 Hz noise in H1 AS_Q (>= 5 minutes)
-
Incorporation of Allegro science segments, including coincidence with L1
-
Time domain calibration status (pdf)
(ppt ) (Xavier Siemens)
Xavi gave an update on his time-domain calibration work with Bruce
Allen. He has computed S3 alpha and beta coefficients for H1 and L1 on
the fly with varying time scales and compared his numbers with the official
frequency domain estimates. He finds very good agreement for H1, but nearly
30% disagreement for L1 (the product alpha*beta seems to agree well, but
the individual factors do not). There was much discussion on whether this
discrepancy reflected a real problem or just a different convention. (Note
added several days later: the discrepancy does appear to be real, and the
L1 frequency domain numbers in error. Why more than one method of calibration
originally yielded the same results is under investigation.) In addition,
a slightly non-zero imaginary part of alpha seems to indicate a residual
problem in the calibration model. The ~3% imaginary component could be
explained by a 2-degree error in sensing function phase.
On a separate topic, Xavi requested guidance on how frequently to compute
and use alpha values. For H1 it appears that computing alpha once per second
gives the benefit of smoother coefficients without averaging over significant
real variations, while for L1, non-random structure in alpha can be seen
even for alpha's computed once every 1/4 second. KR suggested providing
to the collaboration sets of alphas at 1/16 second and 1 second intervals
and lettting the user choose which to use when computing their own calibration
functions. For the time domain calibration itself, it's not clear what's
best. Gaby suggested experimenting with software injections run through
calibrations at different time scales to see what gives best performance.
In general, it seems reasonable to err on the side of too-frequent computations
rather than not-frequent-enough, particularly for L1. But Xavi's algorithm
uses cubic splines for smooth interpolation, but the splines are forced
to go through each recorded alpha point. So recording alphas so frequently
that statistical measurement error dominates could introduce spurious behavior.
As soon as final S3 calibration functions are released, Xavi is ready to
produce S3 h(t) frames for general use.
-
S2
Calibration systematics - average over chunks (Gaby Gonzalez)
Gaby has examined the assumption used by the inspiral group that a
constant calibration coefficient alpha can be used for each 2048-s "chunk"
of data. Generally, she finds the assumption to be safe for most of the
S2 data, but has compiled a number of exceptions, which will likely contribute
to future DQ flags. Consistency with the assumption is tested via
comparisons of minute-to-minute variations with those expected from random
variation and via comparisons of chunk averages / variances with expectation.
As she has reported before, she again finds strong evidence for true intrinsic
variation in L1 calibration constants on time scales shorter than a minute.
The exceptional segments come in several flavors: 1) large, but apparently
random fluctuations; 2) systematic drifts; and 3) missing calibration lines
at the starts or ends of a segment. Gaby will investigate whether the last
category disappears after use of the version 4 S2 DQ flags.
-
S2
Frequency vs time domain calibration comparisons (Gaby Gonzalez)
As part of the validation of the new time domain calibration, Gaby has
compared the alpha and beta values Xavi computes for S2 with the official
values, looking separately at low-amplitude and high-amplitude calibration
line data. She finds that Xavi's numbers are about 5% lower than the official
numbers, but the width of the ratio of the two is less than 2%. Also, Xavi's
values have a narrower intrinsic width, which is encouraging. Since
the beta parameter did not float during the S2 run, measurements of it
should give exactly one, but histograms of Xavi's computation of it from
the DARM/AS_Q transfer function indicate a bias at the 10**-4 level
and a spread at the 10**-6 level. It is suspected that the infamous low-byte
corruption in DAQ readout is causing the deviation from unity. Gaby
noted that there are eight times when Xavi's beta values differ from one
by much larger amounts, from 10**-3 to 10**-1, which merit investigation.
DMT Software Status & Plans:
-
Status of DMT infrastructure:
-
John Zweizig: Version updates, etc
John reported that new documentation on the DMTGen simulation package has
been added to the DMT project
page. John has been using the simulation package to study the low-byte
corruption issue by injecting random clusters of hexadecimal FF into AS_Q.
To gauge the frequency of the problem, he has looked at excitation channel
readback where he applies notch filters to remove the three calibration
lines and looks for residual activity. Indeed he sees occasional glitches,
presumably from the corruption. More information can be found here.
-
Patrick Sutton: Calibration infrastructure
Patrick has been tied up with other work for the last couple of weeks,
but is now getting back to finishing up the infrastructure for the DMT
calibration class. New urgency comes from the project to reproduce the
WaveBurst analysis pipeline in the DMT as part of the burst analysis review,
for which Sergey Klimenko needs calibrated spectra.
-
Status of DMT monitors under substantial development / revamping / debugging:
-
Tim Bodiya / Keith Riles: SpecMon
KR reported that SpecMon's residual instabilities seem to have been
fixed. It has run for many days at a time without crashing. Tim is working
on cleaning up the graphical display and on implementing on-the-fly test-band
settings.
-
Ken Ganezer: seis_blrms
Ken reported on recent attempts to reproduce offline the glitching seen
online during short frame broadcaster data dropouts. Although he hasn't
yet reproduced the effect, he can create something similar at monitor startup.
Peter Shawhan suggested the effect looked like filter ringing. KR suggested
looking at the driver code to see how it deals with non-contiguous data.
(After the meeting KR and Ken exchanged further e-mails and looked at the
code; there does appear to be a likely culprit in the driver code.)
-
Sergey Klimenko: BurstMon
No update.
-
Tim Olson: SpectrumArchiver
Tim has finished a rewrite of the code that allows config file specification
of channels to archive and will ask John to restart the monitor at the
sites. The code will be placed in the DMT CVS archive in a day
or so. John mentioned that he will be at LHO in June and may wait until
then to do any monitor restarts. Tim is now resuming work on the GUI development
for easy retrieval of archived spectra.
-
Steve Penn: BicoMon and BicoViewer
No update.
-
Giovanni Santostasi: PulsarMon
Giovanni showed some sample plots (based on his own simulations assuming
S3 sensitivities) of figure of merit plots to be produced by the new FOM
monitor. He showed effective strain sensitivities for one year of running,
along with energy-conservation upper limits on strain from all ~1500 known
pulsars with measured spindown rates. He also showed the implied integration
times needed to reach those energy conservation limits with S3 sensitivity.
The Crab limit could be reached after about 150 days (~7 days at design
sensitivity). All other known pulsars require considerably longer. Another
plot showed limits on ellipticity one could reach for the known pulsars.
For one year at S3 H1 sensitivity, the best limit one could obtain is 4*10**-7.
KR suggested that for the actual monitor it would probably be desirable
to restrict attention to known, and perhaps only isolated, pulsars,
with signal frequency above 50 Hz.
Remaining DMT reports were cut off by expiration of the telecon
reservation.
-
Brian Stubbs: SuspensionMon
-
Julien Sylvestre: TFCMON (?)
-
John Whelan: StochMon
John reported by e-mail that he has a new undergraduate with some C++
experience who will start full-time work on StochMon next week.
-
Natalia Zotov: PTMon
-
John Zweizig: LSCMon et al
John reported by e-mail that he is modifying PSLMon (infrastructure
class for several monitors, including DataQual) to allow explicit disabling
of parts of bands in band rms plots, to eliminate lines from 60 Hz harmonics,
calibration injections, violin modes, etc.
A.O.B
-
Next detchar telecon: Friday May 21 at 1:30 p.m. EDT (focus on S2/S3 investigations)