(Stanford LSC Meeting, July 19-21, 1999)
[The following minutes are meant to be brief
summaries of presentations
and discussions. They are not meant
to be exhaustive and may contain
errors. Speakers or others who
attended are encouraged to provide
corrections. The transparencies of
all speakers will be posted on the
LSC home page shortly.]
MONDAY AFTERNOON SESSION
John Zweizig (Caltech) - Data Monitoring Tool
Status
----------------------------------------------------
The Data Monitoring Tool (DMT) software package
provides an environment
for running offline monitoring algorithms
either in background
(c language) or in foreground (c/c++ language)
using CERN's root
program for graphics and command interpretation.
The background
and foreground programs can be downloaded
from the web:
http://www.ligo.caltech.edu/~jzweizig/cdist.tar.gz
http://www.ligo.caltech.edu/~jzweizig/rdist-0.1.tar.gz
The DMT will run on Sun workstations at the
sites. The Hanford
workstation (E-450) has been delivered and
will be commissioned
shortly.
As part of the brand-new root-version release,
John has defined
a large number of classes, including ways
to access, store,
manipulate and display data. These include
some basic signal
processing tasks, such as windowing, filtering
and fourier
transforms. Example code with graphics output
were shown.
TUESDAY AFTERNOON SESSION
Keith Riles (Michigan) - Update on Performance
Characterization
---------------------------------------------------------------
Performance characterization refers to describing
systematically
the stationary and quasi-stationary behavior
of the interferometers.
A number of volunteers had committed at Gainesville
or in the interim
to completing some high-priority perfchar
tasks by fall 1999. Others
had expressed interest in a number of areas.
KR spent most of this
presentation trying to recruit new volunteers
and to encourage
firm commitments from those who already expressed
interest.
The large task table assembled as part of
the analysis white paper
preparation served as a menu, and many new
commitments were indeed made.
(The current list of tasks/priorities/people/institutions
can be
be viewed at http://www-mhp.physics.lsa.umich.edu/~keithr/lscdc/tasktables.html)
In addition, a number of task definitions
were refined, broadened, or
consolidated.
Fred Raab (LHO) - Transient Working Group
Status
------------------------------------------------
Transient analysis refers to identifying
& characterizing transient
artifacts in the data arising from purely
instrumental or environmental causes.
A web page lists expressions of interest
in a variety of trananal tasks, but
there has been little detectable activity.
A number of urgent transient
signatures and methods were described. There
are a number of data samples
from the 40-meter and from LHO on which algorithms
can be tested now and more
data samples will be on the way soon.
This presentation too used the large task
table as a menu to recruit
volunteers. Again, there were several new
commitments, and some task
definitions were modified.
David Strom (Oregon) - Report from the Data
Set Reduction Working Group
-----------------------------------------------------------------------
This working group is devoted to customized
data sets for detector
characterization and to the more general
problem of LIGO data reduction
for scientific analysis. The group's efforts
are divided into four
broad categories: 1) providing infrastructure
for data distribution from
the sites to LSC institutions, both over
the internet for small data samples
and via data tapes for larger samples; 2)
providing simple working examples of
a variety of tools to look at distributed
data; 3) data compression techniques
(lossless or nearly so); and 4) data reduction
algorithms that exploit
the nature of the data itself (e.g., filtered
decimation, statistical
descriptions). Volunteers to help in this
effort are encouraged to come
forward, especially in the area of reduction
algorithm development &
implementation.
WEDNESDAY MORNING + AFTERNOON SESSIONS
Sam Finn (Penn State) - Simulated Data Sets
-------------------------------------------
This working group is focussed in the near
term on phenomelogical data
set simulation, using parametrization models.
Eventually this effort
will be merged into the more ambitious End-to-End
Model project
(see Yamamoto talk below). The group plans
to provide a first release
of the simulation software in early November,
software that allows
real-time generation of time-domain data
for immediate analysis or
for storage. Ideally, the software will include
certain known instrumental
artifacts (e.g., stationary lines, decaying
violin modes, fast transients)
on top of random Gaussian or non-Gaussian
background noise. Superposition
of astrophysical signals will be supported.
The first (basic) release will
produce data suitable for Matlab analysis.
The second release, scheduled for
December, will produce data in the frame
format and support more of the
ideal features described above. To be most
useful, the simulation should
include the effects of the servo control
loops and whitening filters of
the interferometer. Daniel Sigg & Nergis
Mavalvala offered to provide
the appropriate parametrizations.
Sam Finn - Thermal Noise from Structurally
Damped Systems
---------------------------------------------------------
This was a brief description of an algorithm
used to simulate
thermal noise in damped mechanical systems.
A physical model
of N parallel spring/dashpot connections
to a test mass proves
quite useful in producing compact, causal
digital filters for simulating
the effects of sharp mechanical resonances.
Ideal analog vs simulated digital
comparisons of transfer functions and power
spectra were shown
Hiro Yamamoto (Caltech) - End to End Model
Update
-------------------------------------------------
The End to End Model is an ambitious project
to simulate the interferometers
from first principles, including the effects
of optical losses, servo control
electronics, misalignments, ambient noise,
etc. Ultimately an entire
interferometer will be simulated in detail
in both the frequency and
time domain, but work is proceeding according
to the subsystems coming
online first. Simulation of the pre-stabilized
laser system and of
the input optics is nearly complete. Work
is underway on the suspensions
& seismic modelling. Simulation of complex
optics configurations with
servo controls is available now in multi
mode acquisition code.
Plans for the future include implementing
more complex subsystems,
streamlining of code to allow more complex
optics and fields,
model validation using real data, enhancing
the user interface,
adoption of the ansi standard template library,
implementing thermal
lensing, speeding up of the simulation code,
and documentation.
Robert Schofield - Ambient & Diagnostic
Magnetic Fields Measured
----------------------------------------------------------------
(Oregon)
inside of a BSC Vacuum Chamber at Hanford
--------
-----------------------------------------
Knowledge of magnetic fields in the vicinity
of the test masses
is important because of the magnets attached
to the masses for
actuation. Measurements have been carried
out to determine both
the ambient fields in one of the Hanford
vacuum chambers and the
transfer function from fields generated immediately
outside the chamber
to the inside. Two magnetometers were mounted
on a fiberglass rod suspended
in the chamber, and measurements were made
along two different lines
in the chamber. Field magnitudes and gradients
were determined for
frequencies up to 800 Hz. Effects of the
clean room fans were
clearly visible in the data.
From the ambient field and field gradient
measurements, a model
(D. Coyne) was used to predict resulting
test mass displacement
noise. First indications are that noise is
generally below the
LIGO 1 requirement, at least for the chamber
measured. The measured
transfer functions are in rough agreement
with a simple model
of eddy current damping in the chamber walls.
Daniel Sigg (LHO) - On-Site Seismic Correlation
-----------------------------------------------
Seismometer measurements were taken at two
different sites at Hanford
(corner station and one mid-station) over
30 512-second intervals. Power
spectra, cross-power spectra and coherence
were calculated using a Hanning
window. Measurements were taken along 3 dimensions
at each site. All
individual power spectra show the micro-seismic
peak near 0.2 Hz while
noise at much lower frequencies is attributed
to instrument drift.
The noise bottoms out near 2 Hz and is still
rising at 10 Hz (limit
of measurements shown). Coherence between
orthogonal axes of measurement
is generally weak, even for the same seismometer,
but coherence between
parallel axes of the two seismometers is
quite strong in the vicinity
of the micro-seismic peak, reaching to 80-90%.
The phase angle of
the cross spectrum at that frequency (about
60 degrees) is roughly
consistent with the expected speed of seismic
waves.
Gabriela Gonzalez (Penn State) - Data Analysis:
A Test for Stationarity
-----------------------------------------------------------------------
Techniques for testing and quantifying stationarity
have been tried out
on 40-meter data. If noise is Gaussian and
stationary, the distribution
of successive spectrogram bins has a Rayleigh
distribution with a
mean value equal to the standard deviation.
If one divides the
standard deviation by the mean (positive
definite), one ideally expects
a flat distribution in frequency at the value
one. The 40-meter data
display a somewhat ideal distribution below
about 80 Hz and above
about 400 Hz. In the intermediate range,
though, where one expects
thermal noise to be important, the distribution
shows much larger
standard deviations than mean (except narrow
bands where violin modes
dominate). Most of this behavior turns out
to be due to a small number
of rapid transients. Removing the affected
time intervals yields a
distribution only slightly higher than ideal
and very nearly ideal
at the violin mode frequencies.
Soumya Mohanty - Non-Parametric Method for
Detecting
----------------------------------------------------
(Penn State) Non-Stationarity
------------ -----------------
The goal is to create a test for non-stationarity
that does not depend
on knowing a priori the right noise distribution.
Ideally the test should
detect true transients with good efficiency
but should also have a low
false-alarm rate. It is meant for generic
bursts where one does not
know the burst waveform, precluding optimized
filters. Put simply, the
test used here compares power spectral density
estimates computed at
two different times (with normalizations
determined by the measured
variances) for various bins in frequency.
From these comparisons, a
"carpet plot" (differential form of "ordinary"
carpet plot)
is generated as time passes. A true rapid
transient is characterized by a
double band in time of large values over
some frequency range. The
test has been evaluated on various background
noise distributions
and power spectrum shapes, and the thresholds
set to give false alarm rates
of 1/hour. With those settings, one can obtain
80% burst detection efficiency
for various burst types (narrowband Gaussian
at different central frequencies
and white Gaussian) with maximum excursions
in amplitude of order a
few standard deviations. The algorithm will
be converted to run in the
Data Monitor Tool environment this fall.
Soma Mukherjee - Simultaneous Dynamical Tracking
& Removal of Multiple
----------------------------------------------------------------------
(Penn State) Violin
Modes
------------ ------------
The goal is to track closely the noise in
narrow violin mode resonances
using a Kalman filter technique to allow
accurate subtraction of
the known lines from the data. The Kalman
filter uses a dynamical model
of a viscously damped simple harmonic oscillator
driven by white noise.
The Kalman filter uses a minimum mean square
estimation to derive a best
estimate of the underlying oscillation strength
(and associated
"mode temperature"). The algorithm has been
tested on 40-meter data
from 1994 to track simultaneously about 25
violin modes where a
subtraction of derived mode signals reduced
the total variance in
the 565-610 Hz range enormously. This algorithm
too will be converted
to run in the DMT environment this fall.
Bernard Whiting (Florida/ANU) - Progress in
Noise Characterization
------------------------------------------------------------------
The Florida and Australia groups have investigated
using spectral
uniformity as a means of characterizing stationarity
of noise.
Histograms of the magnitude, and the real
and imaginary parts of
FFT coefficients at different frequencies
in 40-meter data show
clear deviations from Gaussian behavior for
frequency bins affected
by power line noise (60 Hz & harmonics).
It was demonstrated that
mere numerical inaccuracy (e.g., digitization
noise) or mismatch of
frequency line and frequency binning can
give misleading effects in
power spectra, and frequency aliasing is
a prevalent hazard.
In general, it was shown that one must be
careful in interpreting
spectral features and be aware of artifacts.
Sergei Klimenko (Florida) - Input Optics Simulation
Status
----------------------------------------------------------
The goal is to build an End-to-End module
for simulating the
input optics that includes the mode cleaner
and Faraday isolator
and that interfaces to the pre-stablilized
laser and core optics
simulations. The simululation includes control
servos and effects of noise.
Work is far enough along to verify that explicit
time domain modelling
gives results consistent with frequency-domain
modelling.
The optics part of the IO simulation is complete.
Work is underway
on the mode cleaner length control servo,
integration with the
PSL system and implementing the mode cleaner
wave front sensing servo.
Completion of the input optics simulation
is expected in early September.
Neven Simicevic (Louisiana Tech) - New Seismic
Measurements at Livingston
-------------------------------------------------------------------------
Seismometers more sensitive than the nominal
LIGO sensors have been tested
at Livingston, allowing the limits of sensitivity
at low frequencies
to be decreased by an order of magnitude.
Four sensors were placed on
site, one at the corner station, one at one
end station and two at or
near the other end station. Time series were
taken over several days.
Aside from giving a more precise measure
of ambient noise at the site,
the measurements revealed daily disturbances
from trains passing to the
south, disturbances that can be distinguished
clearly from background for
durations of about 15 minutes.
Discussion on data tape technology
----------------------------------
Various data tape technologies were discussed,
using a web page table
of options (created by Stuart Anderson) as
background. Given concerns
about cost and reliability of other technologies,
it was decided that
the Sony AIT-2 8-mm choice was most desirable.
One can obtain single
tape drives for about $3k. The technology
allows data transfer rates
up to 6 MB/s, and a single tape holds 50
GB. One significant drawback,
however, is that each tape costs about $100.
Discussion on data tape mounting at sites
-----------------------------------------
Various off-site LSC groups will want data
from the IFO sites from
time to time to carry out detector characterization
tasks. Although
the infrastructure is being put in place
to produce such tapes,
recording data on them and shipping them
could be a significant
manpower drain at Hanford and Livingston.
The Lab is reluctant to
commit to that task and has suggested that
LSC members provide such
manpower as part of participating in the
project.
Some comments made during discussion:
* It's too expensive for individual
groups to send a person to a site
merely to record data on a tape.
* It's quite reasonable to expect the
LSC to help detector operations
in this way as part of helping
"run shifts" at the sites. In the long
term one expects/hopes several
LSC members to be on site at any given
time.
* Data tape recording requests should
be made via informal proposals
to the Lab, perhaps by the detector
characterization w.g. collectively.
* Data is not stored long at the sites.
It will not be possible to request
full data samples very long
after the fact.
* Without someone on site to verify
channels are connected properly and
sensors are operating, data
taken at a random time during commissioning
is likely to be worthless.
* It will be desirable to arrange in
advance for short data taking periods
in the coming months from which
many LSC groups can benefit, rather than
forcing each group to make its
own arrangements.
Conclusion:
No real decision was made, but for the
short term (next 6 months or so),
there will probably be two modes for
getting data on tape:
* A group sends someone to the site
not only to take care of the
data recording, but also the
verification of the data's integrity.
* In coordination with commissioning,
we collectively arrange for occasional
data recording periods and the
shipping of tapes to interested groups.
In either mode, a participating group must
provide its own raw tapes
for recording. It should be kept in mind
that small data transfers are also
possible over the internet.
Discussion on analysis white paper
----------------------------------
Some suggestions for improvement:
* Add a data flow diagram and explanatory
introductory text.
* Prune the sections on astrophysical
sources.
* Clarify the role of non-Lab LSC groups
who participate in
commissioning / operations.
A number of groups would like to
help out at the sites but do
not want to violate NSF rules on
double-funding installation/commissioning.
These suggestions have been forwarded to
Rai Weiss and the
other editors of the white paper.