Minutes of Detector Characterization Teleconference
(January 5, 2001)
Present:
Caltech: Barish, Mours,
Shawhan, Whitcomb, Yamamoto, Zweizig
Carleton: Christensen
Florida: Klimenko, Rakhmanov
Glasgow: Drever
LHO: Gustafson, Ito,
Raab, Savage, Sigg
LLO: Daw, Marka, Weiss
Loyola: McHugh
Michigan: Riles
MIT: Adhikari, Shoemaker
Oregon: Rakhola
UTB: Romano
Preliminary E2 Investigation Reports:
-
Correlations (N. Christensen reporting)
(See preliminary
report web page for details) This study uses DMT software written last
year by Adrian Ottewill for quantifying correlations among data channels.
Many clear correlations are observed between the GW channel and various
other channels over different frequency ranges. Dozens of channels were
examined. Plots are shown of correlation strength (0-1) vs frequency. Strong
correlations are observed with a number of inteferometer control signals.
Nelson solicited advice on what areas to focus on. It was suggested that
purely environmental channels be checked for correlation, since cause and
effect can be cleanly separated there. Complicating this group's task is
the inadvertent mixing of longitudinal control signals during the run by
a bug in the controller software (see below).
-
Data Compression (B. Mours reporting)
(See preliminary
report web page for details) Different data compression techniques
have been tried on the E2 reduced data set (RDS) channels and statistics
compiled. Although the nominal factor of two lossless compression was found
for frame-level compression of integers, the compression factor for floating
points numbers (which dominate the RDS) was only about 1.3, and the compression
algorithm slow. The team also looked at two lossy compression techniques
for floats: 1) straight float to int conversion with bit-number control
and 2) wavelet compression. The first method is fast, robust, and introduces
white noise at a level several orders of magnitude below (present) detector
noise for a factor of 2 compression (32->16 bits) and added noise at the
few percent level. The second method has the advantage of giving stronger
compression (up to factor of 11 for parameters tried) and tunable shape
for the added noise. Its disadvantages are slower speed and "touchiness"
w.r.t. the character of the data.
In addition, a data summary set (DSS) was produced with statistical information
on many channels and selected "raw" data channels.
Daniel raised some technical issues. He wondered if the data compression
could keep up with real-time data taking and whether the benefits of compression
were outweighed by the cpu costs of decompression before analysis. Benoit
believed the standard method could comfortably keep up with a 20 MB/s data
rate. Sergey said the wavelet method could handle a few MB/s on a single
processor and suggested multiple processors would be easy to implement.
Daniel objected to using a computing farm merely for compression/decompression.
He also reminded everyone that AIT tape drives can attain comparable
lossless compression at the hardware level "for free", using roughly equivalent
techniques.
Peter stressed that any candidate compression algorithm must eventually
be subjected to a test that includes downstream astrophysical source searching.
-
Information on error signal couplings (R. Adhikari and S. Whitcomb
reporting)
(See note for details) Because of
an inadvertent coupling of the interferometer length servos (software bug),
it is now quite difficult to extract clean measures of error and control
signals in the bandwidths of the servos. Although some transfer function
measurements were carried out at the time, not enough were measured to
permit a complete unfolding of the effects. Stan suggested that above the
unity gain frequency of the highest-bandwidth servo, one should be able
to extract fairly clean error signals by "inverting" the known software
bug. After the meeting Rana provided a matlab
template for carrying out such an inverstion.
Data Monitor Tool Status (John Zweizig):
-
The E2 data replay occupied most of John's time in the last month. The
replay was done in two batches, for 2-arm and 1-arm data and ran with seven
monitors at about three times the real data taking rate. He was not sure
what limited that rate, but suspected I/O overhead, despite running off
the 1 MB/s RDS instead of the original 3 MB/s raw data.
-
During the E2 run 500k 1-second frames were recorded with a few thousand
lost due to disk change glitches and the like. There were 211k DMT triggers
generated for the meta-database. the 0.4 Hz trigger rate was handled comfortably
by the meta-database. About 100 of the triggers were lost initially because
of a technical "race condition" which is understood and fixable. A larger
fraction of triggers were initially lost from the 1-arm replay because
of another technical problem which is also fixable.
-
The stability of the sand and stone machines at Hanford seems much improved,
and the frame broadcaster was recently replaced, eliminating a choking
problem in the data pusher.
-
In response to a KR question, John confirmed that trend data can be
read with existing DMT infrastructure. Peter reminded everyone that LIGOtools
is available for off-site data access, including to trend data.
Status reports on performance characterization priority 1 tasks:
-
Line noise:
-
B. Allen, A. Ottewill (not present)
-
S. Klimenko
No news since December teleconference.
-
A. Sintes (not present)
-
Characterizing seismic noise:
-
E. Daw
The band-limited RMS monitor is now fully installed and running
under the DMT process manager at both sites (KR note: Hurray!).
A problem reported last year with recurring crashes has been fixed with
John's help. The monitor is general purpose but will be made easier to
use. It now monitors seismic noise in several bands, as described at previous
telecons. At Livingston it has been observed that noise in the 1-3 Hz band
is correlated with losses of Mode Cleaner locks. Fred remarked that the
data being generated at Hanford by the monitor is very nice to have, but
that the calibration scale factors seem off. Ed will look into it.
-
Inter-channel correlations:
-
B. Allen, A. Ottewill (not present)
-
Bilinear cross-couplings
-
Operational state:
-
D. Chin, R. Gustafson, K. Riles (KR reporting)
Dave has rewritten the OSC package virtually from scratch to make it
more flexible and efficient. He is now trying to make it work again and
will be at Hanford for three weeks, starting January 15 where he hopes
to finish shaking down the new code. New versions of the servo instability
monitor and lock loss monitor, which now use the old OSC code, will
also be commissioned at Hanford. He will also work on integrating these
monitors into Daniel's Monitor Display Manager (MDM) program to allow GUI display
of results (and eventually parameter control). Tables and graphs of E2
locked stretches longer than 1 minute have been posted on the Lock Loss
Investigation web
page, extracted after the recent data replay (see above).
Status reports on performance characterization priority 2 tasks:
-
Bandlimited RMS
-
Time / Frequency plots
-
S. Mohanty
Not much progress since last month, due to travel. Soumya expects to
have his code (relevant to power spectral transients) running by the end
of the month.
-
J. Sylvestre (not present)
-
Non-Gaussian noise
-
S. Finn, G. Gonzalez (not present)
Status reports on transient analysis priority 1 tasks:
-
Power spectral transients:
-
Servo instability:
-
D. Chin, R. Gustafson, K. Riles (see above)
-
Event catalog:
-
J. Sylvestre, R. Weiss (not present)
Status reports on transient analysis priority 2 tasks:
-
Flickering optical modes:
-
Transient detection using adaptive denoising methods
-
E. Chassande-Mottin (not present)
-
Impulse recognition:
-
M. Ito
Masahiro is examining the output of the monitor that was run during
the E2 replay, sorting through false alarms to see how to better tune the
monitor parameters. A scan of seismic data found four distant earthquakes
that appear to have knocked the interferometer out of lock.
-
Magnetic field transients
-
R. Frey, R. Rahkola (Rauha reporting)
As reported at the last telecon, a new monitor based on John's PSL
monitor is being written. Data from E2 is being studied to determine reasonable
thresholds to set for magnetometer channel transients.
A.O.B.
-
Next detector characterization teleconference: Friday February 9 at 11:15
a.m. EDT (8:15 PDT).