Minutes of Detector Characterization Teleconference
(October 6, 2000)
Present:
AEI-Potsdam: Chassande-Mottin, Sintes
Caltech: Mours, Shawhan,
Zweizig
LHO: Gustafson, Ito,
Raab, Schofield, Sigg, Whitcomb
LLO: Daw, Marka, Saulson
Loyola: McHugh
Michigan: Riles
MIT: Sylvestre
Oregon: Brau, Frey,
Rakhola, Strom
Penn State: Finn
Syracuse: Penn
Introduction (KR):
-
After summarizing the agenda, KR asked that each person reporting on a
DMT software task state whether that software would be likely to be available
during the upcoming November E2 engineering run. Software contributors
were reminded that delivery to John should not be made at the last minute
if a monitor is to be included in production (details below).
Data Monitor Tool Status (John Zweizig):
-
John has released version 1.2.2 in a distribution kit for downloading
to home institutes. Ed Daw confirmed that he has downloaded and installed
it on a linux machine at LSU. He has written installation
notes for others needing to install on a linux system.
-
The new release contains miscellaneous bug fixes, a major upgrade to the
trend writing class (with changes to calling sequences relevant to previous
users). In particular, multiple channels can be written to one file, and
the writing to file is more automated. Associated changes include handling
of integers and double precision floats by the TSeries and DVector classes.
-
A new root macro (pTrend) is also available to handle trend data displays
(similar to online data viewer). Sergey Klimenko's wavelet class is included
in this release too.
-
Some DMT code has been deliberately excluded from this distribution kit
to prevent inadvertent disturbance of the trigger and monitor managers
running at the sites by users at home institutes. This is a temporary measure;
for the longer term, John will write dummy versions of the relevant server
routines for local debugging use. For example, instead of generating a
real trigger (e.g., writing to meta-database), the dummy routine would
write a file to local disk in the LIGO lightweight format. This should
be available in a week or so.
-
Szabi Marka has written a "monitor monitor", that is, a web
page that automatically keeps track of DMT monitors running on the
sand and stone machines at LHO and delaronde at LLO. The monitor allows
one to see which processes are running, brief descriptions of their function
and their recent output. Szabi urged software writers to keep this framework
in mind when documenting their monitors and asked people to try it out
and send him comments.
-
After some discussion of the E2 run schedule here and later in the meeting,
John requested that all monitoring code to be used in production during
E2 should be given to him for installation no later than October 31.
Status reports on performance characterization priority 1 tasks:
-
Line noise:
-
B. Allen, A. Ottewill (not present)
-
S. Klimenko (not present)
Sergey sent an e-mail later stating that he will be present for the
E2 run and will work on the line source investigation using his code recently
installed in the DMT (see minutes of last teleconference).
-
A. Sintes
Alicia has been looking at the 1994 40-meter data to try reproducing
the results shown by Bernard Whiting at the August LSC meeting (see minutes).
He had found that for some frequencies her line removal algorithm seemed
to create artificial side lobes on either side of the primary line. So
far Alicia has been unable to reproduce that effect, but is not carrying
out the analysis in exactly the same way. She is in contact with Bernard
and hopes to sort this out soon.
-
Characterizing seismic noise:
-
E. Daw (presentation)
Ed has written a general-purpose monitor of band-limited RMS, using
a cascade of elliptical band-pass filters on decimated data. The code is
implemented in the DMT and uses an IIR filter class that Ed created. This
cascade approach is thought to be more accurate than the direct approach
implemented in the LAL library. (Ed tried the LAL filter but had trouble
getting output to appear and decided to start from scratch.)
The filter bands are 0.1-0.3 Hz, 0.3-1.0 Hz, 1.0-3.0 Hz, 3.0-10 Hz,
and 10-30 Hz, with an extra, dedicated band at 4.5-6.0 Hz to monitor a
possible disturbance from an underground pipeline at LLO. The filter class
has been set up to allow easy reconfiguring (with assistance from Matlab).
An example was shown of time series output from the monitor for the various
bands (shifted baselines for clarity).
Release 0.0 is available on the sand, stone and delaronde machines
for testing at /export/home/edaw/bin/blrms in a user-friendly executable.
Ed welcomes input, including criticism, from users. He plans to have the
code running during the E2 run, though perhaps not within the process manager
environment.
Future plans include a filter design tool (to bypass the need for Matlab),
a GUI control (will work with Daniel) and incorporation into the set of
managed DMT processes.
.
-
Inter-channel correlations:
-
B. Allen, A. Ottewill (not present)
-
Bilinear cross-couplings
-
S. Penn (not present, reported by e-mail)
Core C code is complete and conversion to C++ for DMT incorporation
is underway (Steve is learning C++). He hopes to have the code running
by the time of the E2 run for real-time testing. He will be at Hanford
for the entire E2 run.
-
Operational state:
-
R. Gustafson, K. Riles (KR reporting)
Since the August LSC meeting, trigger generation for servo instability
detection (based on operational state condition class) has been implemented
and tested, but little progress has been made on integrating the monitor
into the monitor display manager environment. A new version of the monitor
will be given to John for the E2 run, but it may not have full functionality.
If not, a private version of the monitor will be run on the side (on the
sand machine) for further testing/development during E2. In any case, the
tuning of monitor control parameters is best done during the run.
John raised the issue that it's inefficient to run both production
version and private versions of the same monitor. If that becomes an issue
during running, KR suggested the production version be disabled and only
the private version run.
Fred wondered if we have a good feel yet for the cpu demands of all
the monitors to be run during E2. Szabi mentioned that the present load
on sand is 35% of one of the four CPUs. On stone it is about 5%. Depending
on how many monitors are delivered to John by Halloween, we may or may
not have plenty of cpu margin. This issue can be revisited at the organizational
meetings just prior to the run, when hard data is available.
Status reports on performance characterization priority 2 tasks:
-
Bandlimited RMS
-
Time / Frequency plots
-
S. Mohanty (not present - reported later by e-mail)
Since the last meeting, a C++ class has been written to emulate the
Matlab cell array allowing storage of arbitrary-dimension matrices in array
members. This removes an obstacle, and Soumya hopes to have his code (relevant
to time/frequency plots and power spectral transients below) ready in time
for the E2 run.
-
J. Sylvestre
The TID program has been running stand-alone for some time at
both sites. A version based on the DatEnv monitor class has now been written,
and a trigger generator based on pattern recognition in the time-frequency
plane is under development. A GUI control/display program exists now for
TID, but corresponding control/display of the DatEnv version via the monitor
display manager (mdm) awaits stabilization of the mdm environment features.
The TID program includes the features of the event catalog below.
-
Non-Gaussian noise
-
S. Finn, G. Gonzalez (Sam reporting)
This monitor will not be ready for E2, but the prototype version will
be used remotely at PSU during the E2 run.
Status reports on transient analysis priority 1 tasks:
-
Power spectral transients:
-
Servo instability:
R. Gustafson, K. Riles (see above)
-
Event catalog:
-
J. Sylvestre, R. Weiss (see above)
Status reports on transient analysis priority 2 tasks:
-
Flickering optical modes:
-
Transient detection using adaptive denoising methods
-
E. Chassande-Mottin
C code exists for core version of the code. A recent obstacle has been
the lack of Gabor transform code (including inverse transforms) needed
for subband decomposition. Eric hopes to solve this problem soon, by writing
his own transform code, if necessary. He doesn't expect to have the code
ready for the E2 run.
-
Impulse recognition:
-
M. Ito
A monitor based on deviations greater than a certain number of standard
deviations has been written for the DMT environment. Unfortunately, the
present code chews up much cpu time, but it appears that this can be remedied
by reworking the code to eliminate some sorting calls and by adopting IIR
instead of FIR filters. This monitor should be ready for the E2 run.
-
Magnetic field transients
-
R. Frey, R. Rahkola (Ray reporting)
Rauha is taking over the code written by the recently departed Evan
Mauceli and will be at Hanford next week to test it. The original magnetometer
probes showed poor response in the frequency range of interest. They are
being modified, and at least one should be ready in time for E2. The monitoring
code should also be ready to go to use the magnetometer data.
Proposal for Data Summary Set - B. Mours
-
Benoit, Szabi and Virginio have put together a detailed
proposal for a data summary set (DSS) that contains the GW channel
plus a highly distilled summary of monitoring information from other IFO
and environment channels. This reduced data corresponds roughly to what
is called Level 3 data in the LSC analysis white paper. The goal is a data
sample suitable for transmitting over the internet (about 10 KB/s per site)
and for storing months of data on disk.
-
Monitoring information would contain channel summaries (means, rms, band-power,
etc), quality summary (flags on running conditions) and "snapshots" (detailed
information for periods of anomalous IFO behavior). The data would be stored
in large files containing many frames to make efficient use of static data.
-
The quality flags are heirarchical, based on groups of associated channels,
and vary in setting from "fatal" to "suspicious" to "fair" to "gold". The
proposal provides detailed information, including proposed channel groupings
and a summary of information to be stored for different channel types.
-
A prototype DSS builder has been set up and tested online at Hanford. Sample
plots of summary output are shown in the proposal.
-
Benoit and collaborators welcome input on refining the proposal. KR strongly
encouraged further development and suggested incorporating into the quality
flags information derived from online DMT monitors. There was some discussion
of how to test the validity / usefulness of the DSS files. One quick test
is to verify that data periods found suspect by other means (e.g., DMT
transients found) are appropriately flagged as such by the DSS builder.
One expects the process to be iterative, with refinements added to the
DSS as discrepancies are found (and possibly "refinements" added to the
DMT monitors too!).
A.O.B.
-
Next detector characterization teleconference: Friday November 3 at 11:00
a.m. EDT (8:00 PDT).