Environmental Noise in Advanced LIGO Detectors
P Nguyen
1
, R M S Schofield
1
, A Effler
2
, C Austin
3
, V Adya
4
,
M Ball
1
, S Banagiri
5
, K Banowetz
6
, C Billman
7
, C D Blair
8
,
2
,
A Buikema
9
,
2
, C Cahillane
6
,
10
, F Clara
10
, P B Covas
11
,
G Dalya
12
, C Daniel
6
, B Dawes
6
, R DeRosa
2
, S E Dwyer
10
,
R Frey
1
, V V Frolov
2
, D Ghirado
6
, E Goetz
3
,
13
,
10
, T Hardwick
3
,
A F Helmling-Cornell
1
, I J Hollows
14
, N Kijbunchoo
15
,
J Kruk
6
, M Laxen
2
, E Maaske
6
, G L Mansell
10
,
9
, R McCarthy
10
,
K Merfeld
1
, A Neunzert
16
, J R Palamos
1
, W Parker
2
,
17
,
B Pearlstone
18
, A Pele
2
, H Radkins
10
, V Roma
1
, R L Savage
10
,
P Schale
1
, D Shoemaker
9
, T Shoemaker
6
, S Soni
3
, D Talukder
1
,
M Tse
9
, G Valdes
3
, M Vidreo
6
, C Vorvick
10
, R Abbott
6
,
C Adams
2
, R X Adhikari
6
, A Ananyeva
6
, S Appert
6
, K Arai
6
,
J S Areeda
19
, Y Asali
20
, S M Aston
2
, A M Baer
21
,
S W Ballmer
22
, D Barker
10
, L Barsotti
9
, J Bartlett
10
,
B K Berger
23
, J Betzwieser
2
, D Bhattacharjee
13
, G Billingsley
6
,
S Biscans
9
,
6
, R M Blair
10
, N Bode
4
,
24
, P Booker
4
,
24
, R Bork
6
,
A Bramley
2
, A F Brooks
6
, D D Brown
25
, K C Cannon
26
,
X Chen
8
, A A Ciobanu
25
, S J Cooper
27
, C M Compton
10
,
K R Corley
20
, S T Countryman
20
, D C Coyne
6
,
L E H Datrier
18
, D Davis
22
, C Di Fronzo
27
, K L Dooley
28
,
29
,
J C Driggers
10
, P Dupej
18
, T Etzel
6
, M Evans
9
, T M Evans
2
,
J Feicht
6
, A Fernandez-Galiana
9
, P Fritschel
9
, P Fulda
7
,
M Fyffe
2
, J A Giaime
3
,
2
, K D Giardina
2
, P Godwin
30
, S Gras
9
,
C Gray
10
, R Gray
18
, A C Green
7
, E K Gustafson
6
,
R Gustafson
16
, J Hanks
10
, J Hanson
2
, R K Hasskew
2
,
M C Heintze
2
, N A Holland
15
, J D Jones
10
, S Kandhasamy
31
,
S Karki
1
, M Kasprzack
6
, K Kawabe
10
, P J King
10
, J S Kissel
10
,
Rahul Kumar
10
, M Landry
10
, B B Lane
9
, B Lantz
23
,
Y K Lecoeuche
10
, J Leviton
16
, J Liu
4
,
24
, M Lormand
2
,
A P Lundgren
32
, R Macas
28
, M MacInnis
9
, D M Macleod
28
,
S Márka
20
, Z Márka
20
, D V Martynov
27
, K Mason
9
,
T J Massinger
9
, F Matichard
6
,
9
, N Mavalvala
9
,
D E McClelland
15
, S McCormick
2
, L McCuller
9
, J McIver
6
,
T McRae
15
, G Mendell
10
, E L Merilh
10
, F Meylahn
4
,
24
,
P M Meyers
33
, T Mistry
14
, R Mittleman
9
, G Moreno
10
,
C M Mow-Lowry
27
, S Mozzon
32
, A Mullavey
2
, T J N Nelson
2
,
arXiv:2101.09935v1 [astro-ph.IM] 25 Jan 2021
Environmental Noise in Advanced LIGO Detectors
2
L K Nuttall
32
, J Oberling
10
, Richard J Oram
2
, C Osthelder
6
,
D J Ottaway
25
, H Overmier
2
, E Payne
34
, R Penhorwood
16
,
C J Perez
10
, M Pirello
10
, K E Ramirez
35
, J W Richardson
6
,
K Riles
16
, N A Robertson
6
,
18
, J G Rollins
6
, C L Romel
10
,
J H Romie
2
, M P Ross
36
, K Ryan
10
, T Sadecki
10
, E J Sanchez
6
,
L E Sanchez
6
, T R Saravanan
31
, D Schaetzl
6
, R Schnabel
37
,
E Schwartz
2
, D Sellers
2
, T Shaffer
10
, D Sigg
10
,
B J J Slagmolen
15
, J R Smith
19
, B Sorazu
18
, A P Spencer
18
,
K A Strain
18
, L Sun
6
, M J Szczepańczyk
7
, M Thomas
2
,
P Thomas
10
, K A Thorne
2
, K Toland
18
, C I Torrie
6
, G Traylor
2
,
A L Urban
3
, G Vajente
6
, D C Vander-Hyde
22
, P J Veitch
25
,
K Venkateswara
36
, G Venugopalan
6
, A D Viets
38
, T Vo
22
,
M Wade
39
, R L Ward
15
, J Warner
10
, B Weaver
10
, R Weiss
9
,
C Whittle
9
, B Willke
24
,
4
, C C Wipf
6
, L Xiao
6
, H Yamamoto
6
,
Hang Yu
9
, Haocun Yu
9
, L Zhang
6
, M E Zucker
9
,
6
, and
J Zweizig
6
1
University of Oregon, Eugene, OR 97403, USA
2
LIGO Livingston Observatory, Livingston, LA 70754, USA
3
Louisiana State University, Baton Rouge, LA 70803, USA
4
Max Planck Institute for Gravitational Physics (Albert Einstein Institute), D-30167
Hannover, Germany
5
University of Minnesota, Minneapolis, MN 55455, USA
6
LIGO, California Institute of Technology, Pasadena, CA 91125, USA
7
University of Florida, Gainesville, FL 32611, USA
8
OzGrav, University of Western Australia, Crawley, Western Australia 6009,
Australia
9
LIGO, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
10
LIGO Hanford Observatory, Richland, WA 99352, USA
11
Universitat de les Illes Balears, IAC3—IEEC, E-07122 Palma de Mallorca, Spain
12
Eötvös Loránd University, H-1053 Budapest, Egyetem tér 1-3, 1053 Hungary
13
Missouri University of Science and Technology, Rolla, MO 65409, USA
14
The University of Sheffield, Sheffield S10 2TN, UK
15
OzGrav, Australian National University, Canberra, Australian Capital Territory
0200, Australia
16
University of Michigan, Ann Arbor, MI 48109, USA
17
Southern University and A&M College, Baton Rouge, LA 70813, USA
18
SUPA, University of Glasgow, Glasgow G12 8QQ, UK
19
California State University Fullerton, Fullerton, CA 92831, USA
20
Columbia University, New York, NY 10027, USA
21
Christopher Newport University, Newport News, VA 23606, USA
22
Syracuse University, Syracuse, NY 13244, USA
23
Stanford University, Stanford, CA 94305, USA
24
Leibniz Universität Hannover, D-30167 Hannover, Germany
25
OzGrav, University of Adelaide, Adelaide, South Australia 5005, Australia
26
RESCEU, University of Tokyo, Tokyo, 113-0033, Japan.
27
University of Birmingham, Birmingham B15 2TT, UK
28
Cardiff University, Cardiff CF24 3AA, UK
Environmental Noise in Advanced LIGO Detectors
3
29
The University of Mississippi, University, MS 38677, USA
30
The Pennsylvania State University, University Park, PA 16802, USA
31
Inter-University Centre for Astronomy and Astrophysics, Pune 411007, India
32
University of Portsmouth, Portsmouth, PO1 3FX, UK
33
OzGrav, University of Melbourne, Parkville, Victoria 3010, Australia
34
OzGrav, School of Physics & Astronomy, Monash University, Clayton 3800,
Victoria, Australia
35
The University of Texas Rio Grande Valley, Brownsville, TX 78520, USA
36
University of Washington, Seattle, WA 98195, USA
37
Universität Hamburg, D-22761 Hamburg, Germany
38
Concordia University Wisconsin, 2800 N Lake Shore Dr, Mequon, WI 53097, USA
39
Kenyon College, Gambier, OH 43022, USA
Abstract.
The sensitivity of the Advanced LIGO detectors to gravitational waves
can be affected by environmental disturbances external to the detectors themselves.
Since the transition from the former initial LIGO phase, many improvements have
been made to the equipment and techniques used to investigate these environmental
effects. These methods have aided in tracking down and mitigating noise sources
throughout the first three observing runs of the advanced detector era, keeping the
ambient contribution of environmental noise below the background noise levels of the
detectors. In this paper we describe the methods used and how they have led to the
mitigation of noise sources, the role that environmental monitoring has played in the
validation of gravitational wave events, and plans for future observing runs.
1. Introduction
Between 2010 and 2015, the two LIGO detectors at Hanford, WA (LIGO Hanford
Observatory, or LHO) and Livingston, LA (LIGO Livingston Observatory, or LLO)
underwent a period of extensive upgrades to transition from the Initial LIGO stage to
the Advanced LIGO (aLIGO) configuration [1], significantly improving their sensitivity
to gravitational waves [2]. The aLIGO detectors began their first observing run (O1)
on September 12, 2015, and made the first detection of gravitational waves from a
binary black hole (BBH) merger on September 14, 2015 [3], followed by two more BBH
detections before the end of the run on January 16, 2016 [4]. The second observing
run (O2) began on November 30, 2016 after a period of detector upgrades and ended
on August 25, 2017. During O2, in addition to several more BBH detections, LIGO
observed the first binary neutron star (BNS) merger on August 17, 2017 [5]. The
third observing run (O3), which spanned April 1, 2019 to March 27, 2020, came after
another round of major improvements in the performance of the detectors [6] and the
full inclusion of the Virgo detector in the GW network. The first half of the run, ending
on October 1, 2020, LIGO and Virgo observed a total of 39 GW events [7].
Environmental disturbances can significantly impact the data quality of the LIGO
detectors. A gravitational wave (GW) is detected by measuring the differential arm
length (DARM) of an interferometer (and converting it to a GW strain), so coupling
between the external environment and the interferometer readout can reduce a detector’s
Environmental Noise in Advanced LIGO Detectors
4
sensitivity to gravitational waves and potentially produce transient non-astrophysical
signals in the detector. The environment can influence the detector through physical
contact (via vibrations or temperature fluctuations), electromagnetic waves, static
electric and magnetic fields, and possibly high-energy radiation. These effects are
monitored with the physical environmental monitoring (PEM) system of sensors [8].
Studying environmental noise serves two purposes. The first is the validation of GW
events. Environmental disturbances at amplitudes large enough to influence the LIGO
data occur frequently around each detector and can potentially be correlated between
different detectors, i.e. stemming from a common source as opposed to stemming from
chance coincidence. Such correlated noise is not accounted for in the estimation of false-
alarm probabilities, which is done by time-shifting background data from each LIGO
detector to produce long stretches of coincident background. Environmental noise is
particularly important in searches for un-modeled sources of gravitational waves, as these
look for excess power without the use of waveform templates. Thus it is important to
have a quantitative solution for identifying and evaluating the impact of environmental
transients when they occur near candidate events.
The second purpose is to improve the sensitivity of the detector by reducing
contamination from environmental noise. We track down troublesome noise sources
and coupling mechanisms so that we can either remove the noise sources themselves,
isolate them from the detector, or modify the detector to reduce coupling.
Effler et al. (2015) [8] described the methodology for studying environmental
coupling and presented results from the sixth and final science run (S6) prior to the
transition to aLIGO. The methodology has since been improved and expanded, and
sensitivity to environmental effects has changed with the upgrades to the detector.
This paper describes these changes and presents cases where noise sources have been
identified and mitigated between S6 and the end of O3. We also summarize how GW
events are vetted using quantitative results from injections. This paper focuses on noise
investigations at the LIGO detectors; a similar discussion for the Virgo detectors is
provided in Fiori et al. (2020) [9].
There are many techniques for characterizing detector noise beyond those described
here [10, 11, 12, 13, 14]. These include the use of tools for detecting excess power
transients in the strain data [15], categorizing transients using machine learning to
better distinguish them from astrophysical signals [16, 17], searching for correlated
noise between auxiliary sensors and the strain data [18], and many more. Although
these also play a role in achieving the goals above, this paper discusses more direct,
focused techniques for studying, quantifying, and mitigating environmental effects.
This paper is organized as follows. In section 2 we summarize the changes made to
the LIGO detectors and the PEM system since S6. In section 3 we present a method for
quantifying environmental coupling based on data from noise injections. In section 4 we
describe developments in the techniques for performing environmental noise injections.
In section 5 we show results of recent studies and provide examples of how environmental
influences have been mitigated. In section 6 we describe the process of vetting GW event
Environmental Noise in Advanced LIGO Detectors
5
Figure 1.
Amplitude spectral densities of the differential arm length displacement
(DARM) at the end of S6 (Feb 27 2010 04:27:47 UTC) and during O3 (Mar 20 2020
00:00:00 UTC).
candidates with examples from real events. We conclude with a discussion of future work
in section 7.
2. aLIGO Upgrades
2.1. Detector Upgrades
When fully commissioned aLIGO is designed to provide an order of magnitude
improvement to sensitivity in its most sensitive band [1], as well as more than an order of
magnitude improvement at lower frequencies due to seismic isolation upgrades. Figure
1 compares the DARM noise spectra of LHO and LLO at the end of S6 to that at the
end of O3. Significant progress has been made in approaching the design sensitivity
of aLIGO, and further improvements are foreseen for the fourth observing run (O4),
expected to begin in 2022. Here we highlight a few of the major upgrades that were
directly relevant to reducing the coupling of ambient environmental noise.
The core interferometer optics (including the test mass mirrors and beam splitter)
are suspended in active multi-stage suspension systems, which in turn are on active
seismic isolation tables [19, 20]. This provided a substantial improvement to sensitivity
below 100 Hz over the initial LIGO configuration. The suspension and isolation tables
also provide useful sensors for the motion of these optics.
Auxiliary sensors used in the control system of the interferometer were moved from
in-air optical tables to in-vacuum, seismically isolated tables. This reduced acoustic
coupling but did not eliminate it. Although the main laser system (PSL, or pre-stabilized
laser) could not be moved into vacuum, an acoustically isolated room was built to house
the laser, and a new optical table with improved isolation and damping of its resonances
Environmental Noise in Advanced LIGO Detectors
6
was installed [21].
To reduce magnetic coupling, magnets and certain magnetic materials are no longer
present on or near the test masses themselves [22]. Instead, the aLIGO test masses are
controlled either by magnets at the upper stages of the suspension system or by an
electrostatic drive. To further reduce the ambient acoustic noise from electronics fans
near the detector, power supplies and most electronics were moved to separate rooms
(called here electronics bays), some tens of meters away from the vacuum system which
houses the interferometer.
2.2. Environmental Monitoring Upgrades
Understanding environmental influences on the detectors requires comprehensive
monitoring of its physical surroundings. This is done through the PEM system
of auxiliary sensors, which consists of accelerometers for high-frequency vibrations
(tens to thousands of Hz), seismometers for low-frequency vibrations (up to tens
of Hz), microphones, magnetometers, voltage monitors that measure the voltage of
electric power supplied to the detector sites, radio-frequency (RF) receivers, a cosmic-
ray detector for high-energy particles, and wind, temperature and humidity sensors.
Detailed information on PEM sensors, including example background spectra and
calibration data, can be found on the PEM website, PEM.LIGO.org [23]. The site
also provides links to long-term summaries of ground tilt, seismic motion, and wind (on
the Environmental Studies pages).
In order to monitor environmental signals that could influence the interferometer,
we use PEM sensors that are demonstrated to be much more sensitive to these signals
than the interferometer is. Sensor locations are chosen with the goal of maximizing
coverage of potential coupling sites. Ideally, if an environmental signal were to reach a
coupling site, nearby sensors should be able to observe the signal at an amplitude equal
to the amplitude at the coupling site. In practice, we place sensors where we expect
the coupling to be strongest, and we may place new sensors during the run to improve
monitoring of important coupling sites.
By focusing on the fundamental interactions that can affect the detector, the PEM
system allows us to monitor potential effects from a large variety of environmental
events. For example, wind can couple through vibrations in the ground and air, so
its effects are monitored by seismometers, accelerometers, and microphones. Lightning
could couple by magnetic fields, power mains disturbances, and electromagnetic waves
at radio frequencies that we demodulate into the detection band, so lightning strikes are
monitored with magnetometers, mains monitors, and RF receivers. The PEM sensors
provide coverage of signals in the detection band of the interferometer (20-2000 Hz),
although we also monitor beyond these frequencies when there are coupling mechanisms
that convert low- or high-frequency signals up or down into the detection band or when
the interferometer performance can be affected by frequencies outside of the detection
band.
Environmental Noise in Advanced LIGO Detectors
7
Figure 2.
The Physical Environmental Monitoring system layout at the LIGO
Livingston detector during O3, as seen on the PEM public website [23]. The path
of the main interferometer laser is shown as a red line; core optics, such as the test
masses, are represented by rectangles inside the vacuum chambers. The most major
changes during aLIGO have been made to the accelerometer locations and the addition
of new magnetometers, e.g. in the electronics bays.
The state of the PEM system at LLO during O3 is shown in Figure 2. A similar
map for LHO is available on the PEM website [23]. Since the transition to aLIGO, many
changes [24] have been implemented to expand the general coverage of the PEM system,
provide additional monitoring near known high-coupling areas, and adapt to the detector
upgrades described in 2.1. Many changes involved the addition of accelerometers or
relocation of existing ones:
•
Vacuum chambers: In iLIGO, most accelerometers were mounted on the seismic
isolation system. These locations became redundant with the introduction of
vibrational sensors as part of the new active isolation systems, so the accelerometers
are now mounted on the chamber walls where they can detect motions that could
modulate laser light scattered off of the chamber walls.
•
Beam tubes: Accelerometers at select sites along the 4-km beam tubes monitor
Environmental Noise in Advanced LIGO Detectors
8
vibrations that affect the modulation of reflected light inside. Coverage now
includes the mid-stations, which is especially important at LHO where significant
coupling has been measured, likely because they contain the smallest aperture
between vertex and end stations.
•
Electronics bays: Floor accelerometers were added to detect vibrational coupling
to the electronics boards (e.g. through resistance variations in poor solder joints)
and to monitor the rooms as seismic sources.
•
Vacuum enclosure areas: Floor accelerometers were added near the vacuum
chambers in order to expand coverage and aid in localizing sources of vibrations
through propagation delays and amplitude differences at locations that do not have
the resonance structure of the vacuum envelope.
•
Pre-stabilized laser table: Coverage of the main laser table was expanded. This
area has continued to be a major source of vibrational coupling.
Many sensors were also upgraded to newer models in order to improve their
performance. Table 1 summarizes the current sensor models and specifications.
Magnetometer coverage was also expanded, particularly with the addition of
magnetometers on the electronics racks (located in the electronics bays) which were
important noise sources and coupling sites during initial LIGO. Relatively large magnetic
fields are generated by the equipment in the racks and these fields can couple to
components, cables and, connectors in the racks. Additionally, magnetometers in
electronics racks have been useful for identifying sources of narrow spectral peaks even
when the coupling was not through magnetic fields. Cyclical processes producing line
artifacts in the DARM spectrum can be tracked down by detecting currents associated
with those processes. In a sense, we monitor multiple electronic systems at once, using
fluxgate magnetometers in the electronics racks (Bartington-03 series [25]). These are
sensitive enough to detect periodic currents with amplitudes as low as
5
×
10
−
5
A at 1
m from long wires or traces [26].
Additionally, the non-rigid tripods for fluxgate magnetometers were replaced with
rigid ones. Non-rigid tripods lead to increased cross-talk between floor vibrations and the
magnetometer signal, as the magnetometer vibrates relative to the Earth’s magnetic field
at the tripod resonance frequency. This created a peak in the magnetometer spectrum
a factor of three above background, which was eliminated by switching to rigid tripods
[27].
In addition to the fluxgate magnetometers that monitor local magnetic fields,
extremely low frequency induction coil magnetometers (LEMI-120 [28]) were added
to the PEM system in order to monitor magnetic noise from Schumann resonances.
These are global electromagnetic resonances in the cavity formed by the Earth’s surface
and the conductive ionosphere. Lightning strikes around the world excite this resonant
cavity, producing picoTesla-scale magnetic fields that can cause correlated noise in the
LIGO detectors [29, 30, 31]. Two LEMI magnetometers are positioned at each site, far
enough from the detector so that they are not sensitive to the same local magnetic fields
Environmental Noise in Advanced LIGO Detectors
9
Table 1.
Specifications for important PEM sensor types. The operating frequency
range is the range in which the sensor calibration is flat; we often use them over a
broader range. Noise floor numbers are reported in the operating band of each sensor
(seismometer at 1 Hz).
Type
Sensor
Operating freq. Sampling freq. Noise floor
seismometer Guralp
®
CMG-3T [34, 35] 0.1-20 Hz
256 Hz
<1 nm/s
/
√
Hz
accelerometer Wilcoxon
®
731-207 [36]
1-900 Hz
4096 Hz
0.5
μ
m/s
2
/
√
Hz
microphone
Brüel&Kjær
®
4130 [37]
10-900 Hz
16384 Hz
<30
μ
Pa
/
√
Hz
microphone
Brüel&Kjær
®
4188 [38, 39] 8-12500 Hz
16384 Hz
<5
μ
Pa
/
√
Hz
magnetometer Bartington
®
03CES100 [25] 0-900 Hz
8192 Hz
<6 pT/
√
Hz
magnetometer LEMI-120
®
[28]
0.0001-1000 Hz 4096 Hz
<0.1 pT/
√
Hz
radio station AOR
®
AR5000A [40]
24.5 MHz
16384 Hz
—
observed by the fluxgate magnetometers. They are placed at a location between the
corner station and end stations, 100-200 m from the beam tube, one aligned with the
x
-axis and one with the
y
-axis of the interferometer.
An electric field meter was installed in an end test mass chamber at each observatory
[32, 33, 6]. These can detect electric fields generated inside of the chambers as well as
fields from outside the chamber that make it in through glass viewports on the chambers.
3. Coupling Functions
To determine the degree to which the detector is affected by environmental influences
during operation, we inject basic environmental disturbances that produce a response in
DARM. We make acoustic injections with speakers and monitor them with the system
accelerometers and microphones; seismic injections with shakers, monitoring them with
the accelerometers and seismometers; magnetic injections with wire coils monitored with
the magnetometers. The injection methodology is described in more detail in Section
4. To motivate the injection techniques we first discuss the means of quantifying the
coupling.
Suppose there exists only one coupling site, a sensor is placed at the location of the
coupling site, and a noise injection is performed that produces a signal in the sensor and
some response in DARM. A
coupling function
can be computed based on the actuation
measured by the witness sensor and the response measured in DARM [41, 42]. We
compare the amplitude spectral densities (ASDs) of DARM and the witness sensor
during the time of the injection (
injection time
) to their ASDs during a time when both
are at observation-mode noise levels (
background time
). The coupling function at some
frequency
f
is given by
CF(
f
) =
√
√
√
√
[
Y
inj
(
f
)]
2
−
[
Y
bkg
(
f
)]
2
[
X
inj
(
f
)]
2
−
[
X
bkg
(
f
)]
2
(1)
where
X
bkg
(
f
)
and
X
inj
(
f
)
are the ASDs of the witness sensor at background and
Environmental Noise in Advanced LIGO Detectors
10
injection times, respectively, and
Y
bkg
(
f
)
and
Y
inj
(
f
)
are the ASDs of DARM at
background and injection times. We use
coupling factor
to refer to the value of a
coupling function at a single frequency bin.
A sensor’s coupling function can be used to compute the contribution of noise in the
sensor to DARM. For example, when validating GW events, we multiply the coupling
function by the amplitude of any environmental transient observed by the sensor to
predict the corresponding amplitude in DARM. Additionally, multiplying the coupling
function by the sensor’s ambient background level yields the ambient contribution of
noise at the sensor to the DARM spectrum:
Y
(
f
) = CF(
f
)
X
(
f
)
.
Suppose now we expand the scenario such that there are multiple coupling sites,
and a sensor is placed at the location of each site. We can model the response
in DARM to each injection as a linear combination of the sensor signals and their
sensor-specific coupling functions. To solve for the coupling functions, we can perform
multiple injections instead of just one, resulting in a system of
n
equations with
m
unknown coupling functions, where
n
and
m
are the numbers of injections and sensors,
respectively:
Y
i
(
f
) =
m
∑
j
=1
CF
j
(
f
)
X
ij
(
f
)
.
(2)
Here
Y
i
(
f
)
and
X
ij
(
f
)
are the amplitudes of DARM during injection
i
and sensor
j
during injection
i
respectively, and
CF
j
(
f
)
is the coupling function of sensor
j
. One
could solve (2) to determine the coupling functions of all sensors.
We have assumed thus far that the witness sensors are placed at the locations of the
coupling mechanisms, but such perfect placement is not realistically feasible given that
there are an unknown number of coupling sites at unknown locations. A sensor, even if it
is near a coupling site, only measures the injection amplitude at its own location, not at
the coupling location. Therefore, when using real-world sensors, (1) is only an estimate
of the true coupling, and (2) is not an exact model of all the coupling mechanisms.
Nevertheless, as explained above, we distribute sensors to maximize coverage of coupling
sites and find that this has been sufficient for producing reliable coupling functions for
all sensors, as discussed further in Section 3.1.
One hurdle remains in attempting to solve (2). In practice, typically
n < m
due to
logistical constraints on the number of injections one could perform during a realistic
time window, which makes the system of equations underdetermined. The problem
can be simplified by instead approximating
CF
j
(
f
)
for each sensor independently of
other sensors. Given a sensor
j
, we can re-purpose (1) (replacing
X
with
X
ij
and
Y
with
Y
i
) to compute a single-injection “coupling function”
CF
ij
(
f
)
for each injection,
then combine those to produce an approximation to
CF
j
(
f
)
. The closer an injection
is to a sensor, the more accurate the computed
CF
ij
(
f
)
would be to
CF
j
(
f
)
, since the
DARM response would be dominated by coupling near sensor
j
. Since it is impractical
to produce an injection at each sensor, the approach we have adopted for combining the
CF
ij
(
f
)
is to construct a
composite coupling function
whose value at each frequency bin
Environmental Noise in Advanced LIGO Detectors
11
is the coupling factor corresponding to the nearest injection, determined by the highest
sensor amplitude (using the assumption that injection amplitudes are equivalent). That
is, for a frequency
f
k
and a set of injections
I
, we measure the sensor amplitudes
{
X
ij
(
f
k
)
|
i
∈ I}
, compute the single-injection coupling functions
{CF
ij
(
f
k
)
|
i
∈ I}
,
and compute the composite coupling function as
̃
CF
j
(
f
k
) :=
CF
lj
(
f
k
) where
l
=
argmax
i
∈I
(
X
ij
(
f
k
))
.
(3)
If the distribution of injection locations provides sufficient coverage of sensor
locations, then
̃
CF
j
(
f
)
≈
CF
j
(
f
)
. We discuss shortcomings of this assumption in Section
3.1.
Computing the single-injection coupling functions
CF
ij
(
f
)
(example shown in
Figure 3) requires a significant difference between the injection and background signals
in the sensor and in DARM. To distinguish between measurements and upper limits,
thresholds are chosen for the sensor and DARM in the form of a ratio between the
injection ASD and background ASD. For each frequency bin, if an injection produces a
large enough signal to exceed both the sensor threshold and DARM threshold, then a
coupling factor can be measured via Eq. 1. If the injection exceeds the sensor threshold
but not the DARM threshold, then we instead compute an upper limit by omitting the
DARM background term. These thresholds are typically chosen to be a factor of two
in DARM and a factor of a few in the sensor, based on the typical level of fluctuations
observed in the spectra.
The composite coupling function computed via (3) is used for comparing coupling
between different sensor locations and producing estimates of DARM amplitudes, e.g.
as part of event validation (see Section 6). Therefore we refer to a sensor’s composite
coupling function simply as its coupling function from here on. Figure 4 provides an
example of an estimated ambient for an accelerometer on the HAM6 vacuum chamber
(which houses the interferometer output optics). The PEM website provides coupling
functions for all accelerometers, microphones, and magnetometers produced from the
most recent campaign of injections [23].
3.1. Uncertainties and Limitations
We characterize coupling using the coupling function defined in (1) instead of a transfer
function because we do not assume perfect coherence. Low coherence can arise either due
to non-linearity in the coupling or due to the spacing between the sensor and coupling
site.
To measure coupling, we inject signals large enough to produce a response in
DARM, but the maximum amplitude of injections is limited by the sensitive range
of the environmental sensors (saturation produces an overestimate of coupling). This
effectively limits how far below the DARM background we can probe for coupling or
establish upper limits.
Environmental Noise in Advanced LIGO Detectors
12
Figure 3.
Vibrational coupling excited by a broadband (60-200 Hz) acoustic injection
near the output arm of the interferometer. The left plot shows the displacement of
an accelerometer in the PSL room during background time (black) and injection time
(orange). The middle plot shows the interferometer readout during background time
(black) and injection time (orange). Estimated ambient levels for the accelerometer
are also shown as dark blue dots, with upper limits shown as light blue crosses;
they are produced from the single-injection coupling function in the right plot. A
vibrational single-injection coupling function represents meters of differential test mass
displacement per meter of sensor displacement, hence the units of m/m.
Figure 4.
Ambient noise level for the LHO HAM6 Y-axis accelerometer estimated
from a composite coupling function, using acoustic and seismic injections near the
output arm. For simplicity only five injections were used to produce this example,
however in practice the number of injections performed near a sensor can be many
times higher.
Equation (1) relies on two assumptions about the coupling mechanism. First, the
coupling is assumed to linear, e.g. doubling the amplitude of the injection would double
the amplitude of the response in DARM. We check this by repeating injections with
different amplitudes. Second, the coupling function ignores any up- or down-conversion
of the signal between the sensor and DARM. This non-linear coupling can be very
significant for scattering noise and bilinear coupling but is not accounted for in the
estimates of linear coupling. One way we detect non-linear coupling is by sweeping
Environmental Noise in Advanced LIGO Detectors
13
single frequency injections over time and searching for off-frequency response in DARM
spectrograms. Frequency changes from non-linear coupling can be an issue in broadband
injections where up- or down-converted noise in DARM appears in the injection band,
resulting in artificially higher estimates of linear coupling. We split broadband injections
into smaller frequency bands to avoid this effect when necessary. One approach for
quantifying non-linear coupling is presented in Washimi et al. (2020) [43].
As mentioned above, the use of (2) relies on the assumption that the environment
is monitored at the coupling site. The density of sensors is not great enough for this
to be strictly true, especially if the source of the environmental signal is closer to the
coupling site than the sensor is. The finite spacing of sensors leads to imperfect coupling
functions but, for environmental signals that are generated at a distance greater than
the typical sensor spacing of a few meters (the external signals that are the focus of
PEM), the uncertainty can be estimated based on the differences between injections
made at different locations. We choose a sensor near a known coupling site and find the
variance between single-injection coupling functions measured for that sensor. Figure 5
shows single-injection coupling functions for an accelerometer measured from shaker
injections produced from three locations. Since the injection locations are close enough
to the accelerometer, we can assume that the variance is entirely due to the distance
between the sensor and the coupling site. Variations in injection location result in
a multiplicative scaling of the single-injection coupling functions, so we quantify the
variance by computing the geometric standard deviation of coupling factors in each bin.
Averaged across all bins, the geometric standard deviation between injection locations is
1.4, i.e. coupling functions measured from vibrational injections, as well as vibrational
noise projections to DARM, vary by a factor of 1.4. A similar study combining
geometric standard deviations for various magnetometers at both observatories shows
that magnetic coupling measurements vary by a factor of 1.7 [44].
In the case of acoustic injections, the uncertainty in a coupling function can be
exacerbated when nodes and anti-nodes in the acoustic signal coincide with the location
of a sensor. This results in peaks and troughs in the sensor spectrum at frequencies that
have a node or anti-node at the sensor location, respectively. These artifacts can impact
any sensor, but are more noticeable in microphone spectra than accelerometer spectra,
possibly because the stiffness of the vacuum enclosure results in effectively averaging
over a larger area; in microphones, the peak-to-trough ratio is typically a factor of a
few. The peaks and troughs are present in the sensor but not in DARM, because the
sensor monitors a single point whereas the coupling to DARM is spread across a large
enough area for the effects of nodes and anti-nodes to average out. Consequently, this
effect imprints troughs and peaks onto the coupling function. The artifacts can be
smoothed out of the spectra by computing a moving average over
X
inj
(
f
)
. The peak-
to-peak distances are typically a few Hz, so we smooth the spectra enough to remove
features up to a few Hz across. This is acceptable in broadband acoustic injections
which are designed to not produce any other spectral features at this scale.
Although the injections used to measure coupling functions are designed to best
Environmental Noise in Advanced LIGO Detectors
14
Figure 5.
Single-injection coupling functions for the HAM5 Y-axis accelerometer
from shaking injections made from three different locations (on top of HAM5, on top
of HAM6, and on the HAM5 chamber door) show the typical spread in coupling that
results from varying the injection location. Multiple injections at different frequency
bands are shown for each source location. On average the coupling measured from
different locations varies by a 1.4.
replicate environmental noise, there are still differences and it is useful to test the
coupling functions with different environmental events by comparing noise seen in
DARM during such events to noise levels predicted by PEM sensors and their coupling
functions. Thunderstorms are known to produce short-duration transients in DARM
at tens of Hz. At LLO, coupling functions for several accelerometers at the Y end
station, where vibrational coupling was the highest, were capable of estimating the
amplitude of multiple transients in DARM to within a factor two during a particularly
loud thunderstorm [45]. Helicopter flyovers can produce narrow-band features in DARM
up to tens of seconds long. Coupling functions of various sensors at both interferometers
predicted the amplitudes of lines produced by multiple helicopter flyovers during O3
to within a factor of two in most cases [46]. Vibrational noise from rain and the
building HVAC, which produce much longer-duration noise in DARM, have also been
well estimated by coupling functions at LHO [47, 48].
4. Injection Methods
The basic methodology of environmental noise injections is described in [8]. Here
we summarize the methods and describe improvements made to the hardware and
techniques since then.
Injection locations are chosen to best mimic disturbances from outside the detector.
To do so we choose them to be as far from the detector and environmental sensors as
possible, but we are usually limited by the size of the detector sites themselves (some
injections can be made from outside). We perform injections from as many locations as
Environmental Noise in Advanced LIGO Detectors
15
Table 2.
Specifications for injection equipment.
Equipment
Injection type
Custom enclosure with two 14-in. speakers
Acoustic
Various smaller speakers
Acoustic
APS 113 Electro-Seis
®
Long Stroke Shaker [49]
Vibrational
Piezosystem
®
[50] shaker with custom reaction mass
Vibrational
Brüel & Kjær
®
[51] EM shaker with custom reaction mass Vibrational
1 m diameter copper coil (100 turns)
Magnetic
3 x 3 m and 5 x 5 m coils (80-100 turns)
Magnetic
Figure 6.
Injection equipment photos. From left to right: wall-mounted magnetic
field injection coil; 14-in. speakers; APS 113 shaker connected to the door of a
vacuum chamber by a rigid fiberglass rod; modified Piezosystem shaker clamped to
an electronics rack; modified B&K shaker clamped to a beam tube support.
time allows in order to maximize coverage of potential coupling sites. Increased time
allocation towards environmental studies in recent years has allowed for a significant
increase in the number of injection locations.
Table 2 summarizes the current equipment used and Figure 6 shows photos of
some of the equipment. Seismic injections at low frequency (up to tens of Hz) during
initial LIGO were performed with small electromagnetic and piezoelectric shakers and
a weighted cart. A large shaker [49] has been used since the beginning of noise studies
for O3.
Two new injection techniques have been developed for localizing vibration coupling
sites connected to the vacuum enclosure, such as locations on the vacuum enclosure that
reflect scattered light. The techniques rely on the slow propagation speeds (hundreds
of meters per second) of vibrations on the steel vacuum enclosure walls or, for acoustic
injections, in air.
The first technique is narrow-band, and involves vibrating the vacuum enclosure
at two slightly different frequencies, each injected from a shaker or a speaker at a
different location (e.g. a shaker at one location injects a sine wave at frequency
f
and a
shaker at the other location injections at frequency
f
+0
.
01
Hz). The two injections are
Environmental Noise in Advanced LIGO Detectors
16
Figure 7.
Example spectrograms showing a vibrational beat injection using two
shakers to localize the coupling site responsible for a 48 Hz peak in the DARM
spectrum. The shakers were injecting at 48 and 48.01 Hz. The Y-axes of the
spectrograms are centered along at 48 Hz and show the combined signal in each sensor
modulating at the beat frequency (0.01 Hz). This set of spectrograms suggests that
the accelerometers on the input test mass (ITM) chambers and the Y-axis HAM2
accelerometer are likely not close to the true coupling location, since the beat envelopes
are the furthest offset from the beat envelope in the DARM response. Multiple other
injections were made (not shown here) with varying shaker locations in order to rule
out other sensors until the most likely candidate remaining was the HAM3 Y-axis
accelerometer. Black glass was used to block scattered light at this location and the
peak was eliminated for the second half of the O3 observation run.
adjusted in amplitude to produce strong beats in DARM. Because the injection locations
are different, the relative phase of the two injected signals varies with location on the
vacuum enclosure. As a result, the phase of the beat envelope varies with position,
and different sites experience maximum chamber wall motion at different times. The
sites with accelerometer signals that have the same beat envelope phase as DARM are
candidates for the scattering sites on the vacuum enclosure walls (Figure 7). Other
sensors that are not near the coupling site may also match the phase by chance, but
these false positives can be rejected by varying the locations of the shakers.
The second injection technique, which is broad band, involves propagation delays
in impulse injections. Impulse injections are performed by striking the vacuum
enclosure directly with enough force to produce a transient in DARM and in nearby
accelerometers. The vibrational impulse propagates through the structure of the vacuum
Environmental Noise in Advanced LIGO Detectors
17
enclosure, arriving at different accelerometers and coupling sites at different times. We
can distinguish these arrival times because the propagation velocity is much slower than
in solid material, and is only roughly 300 m/s in our case. Using time series plots, the
arrival time of the impulse in DARM is compared to the arrival time of the impulse in
multiple accelerometers (Figure 8, left). The accelerometers that have the same arrival
time as DARM are more likely to be near a coupling site than those that observe the
impulse much earlier or later than DARM does. Again, varying the location of the
injection eliminates sensors that match the DARM time-of-arrival by chance but are
actually far from the coupling site. An additional consistency check is that the coupling
of accelerometers near the coupling site will vary less between different impulse locations
than that of accelerometers far from the coupling site. Finally, if the accelerometer is
at the coupling site, the impulse in DARM will have a resonance structure that is
similar to the resonance structure of the accelerometer signal, which can be judged from
spectrograms (Figure 8, right).
These two techniques aided in the localization of a coupling site that was producing
a 48 Hz peak in DARM throughout the first half of O3 [52]. The peak was present before
the start of the run, and shaker and acoustic injections suggested the source was likely
at the corner station. Impulse injections pointed to the highest coupling being near the
vertex and input arm. Using the double-shaker beat injection method, with frequencies
of 48 and 48.01 Hz, it was found that the timing of the beat envelope in DARM best
matched that of the HAM3 door, even after varying the shaker locations and using
a temporary accelerometer to test other nearby locations. This led to the discovery
that the 48 Hz peak was a result of scattered light at the HAM3 viewports, which was
promptly eliminated by blocking it with black glass, removing the 48 Hz peak from the
DARM spectrum for the remainder of the observing run.
Improvements have also been made to the magnetic field injection equipment. In
order to generate fields strong enough to couple into DARM using the 1 m magnetic
field coils made during initial LIGO [8], we must focus the power of the coil into narrow
bands and combs instead of injecting broadband signals. This was sufficient in initial
LIGO when strong magnetic coupling occurred primarily through permanent magnets.
However, due to the removal of permanent magnets from the test masses, coupling
from those sources has decreased and cables and connectors have become the dominant
coupling sites above about 80 Hz, introducing more structure to the coupling functions
and requiring stronger injections.
To achieve high-amplitude broadband magnetic injections, seven wall-mounted
coils, each one a 3 m x 3 m or 5 m x 5 m square of 80-100 turns, are being installed
at each site; three at the corner station and two at each end station. These coils are
fixed in place and can be operated remotely, allowing for weekly injections to monitor
variations in magnetic coupling caused by changes to electronics. Figure 9 compares the
old and new magnetic injections. Some coils were installed and operated at the sites
during O3; the project will be completed by the start of O4.
Environmental Noise in Advanced LIGO Detectors
18
Figure 8.
Left: Example time series of a single impulse injection signal in DARM
and various output optics accelerometers. Multiple sensors observe an impulse time-
of-arrival matching that of DARM, but repeating the injection from various other
locations rules out sensors that do not match DARM consistently across multiple
injections. In this case the septum (separating the HAM5 and HAM6 chambers)
accelerometer signal matched the DARM signal most consistently (other injections not
shown for brevity). Right: Spectrograms of the same impulse injection for DARM and
the three sensors with the closest matching time-of-arrival to DARM. The similarity
between the frequency structure of the septum accelerometer and that of DARM
further supports the septum as a dominant coupling site in the output arm.
5. Mitigation of Environmental Effects in aLIGO
We use the methods discussed thus far to track down noise sources whose estimated
ambient level in DARM is more than a tenth of the DARM background. Mitigation
can be accomplished in three ways: by removing or modifying the source itself, by
isolating the source or otherwise addressing propagation of the signal to the detector,
or by reducing the coupling itself through some modification to the detector. Here we
provide several examples of environmental effects that were mitigated based on results
from noise investigations.
Environmental Noise in Advanced LIGO Detectors
19
Figure 9.
DARM response to old (left, comb) and new (right, broadband) magnetic
injections. Black and orange lines show the DARM ASD before and during the
injections, respectively. The comb injection curve is a composite of multiple comb
injections, with fundamental frequencies of 7.1 Hz, 14.2 Hz, and 49.7 Hz, made at
different times. The broadband injection spectrum is a composite of a 10-100 Hz
injection and a 100-1000 Hz injection made at different times, hence the break at 100
Hz.
5.1. Seismic and acoustic influences
Figure 10 shows the ambient contribution of vibrational noise during O3, produced by
combining the highest coupling factors among accelerometers and microphones measured
from an injection campaign at the beginning of O3. At the end of O3, the vibration
noise background at both observatories was dominated by input beam jitter above 100
Hz (discussed in Section 5.1.1). At LHO, the dominant coupling region below 100 Hz
was the output arm. At LLO, the dominant coupling regions were the Y-end in the
40-60 Hz band and the output arm in the 60-100 Hz band.
5.1.1. Input beam jitter.
Alignment fluctuations of the beam (beam jitter) entering the
interferometer cause variation in the coupling of the fundamental mode into the arm
cavities, producing amplitude noise. In addition, the varying beam position relative
to defects in the test masses causes a variation in the balance of light between the
two arms, further contributing to the noise [6]. The dominant source of alignment
fluctuations was turbulence in the laser cooling system, causing vibration of mirrors
and other optics on the table and in the laser, producing peaks in DARM at mechanical
resonances of the optics and their mounts. A second mechanism may be variation in
beam diameter associated with the turbulent cooling. In addition to vibrations from
the cooling system, transient vibrations, such as those made by large vehicles or heavy
footsteps in the control room, produced transients in DARM by temporarily increasing
alignment fluctuations.
Mitigation has included removing turbulence-producing connectors, sharp turns in
the coolant lines, and abrupt diameter changes within the cooling system; and reducing
the flow of coolant [53]. In addition, injections were used to identify the optic mounts
Environmental Noise in Advanced LIGO Detectors
20
Figure 10.
Ambient vibrational noise at LHO (top) and LLO (bottom), shown in
dark blue (measurements) and light blue (upper limits). The values are produced
by selecting the highest-amplitude composite coupling function at each bin across all
sensors at each observatory. The black and orange lines show the DARM background
and the aLIGO design sensitivity, respectively.
that produced the largest peaks in DARM, and mass-and-viton dampers were added to
the mounts. This resulted in motion reductions by factors of a few [54]. Finally, the
resonances of optic mounts on the periscope that raises the beam from the laser table
level to the interferometer level, were tuned by adding small masses to the optic mounts,
so that their resonances would not overlap in frequency with periscope resonances that
would increase the motion of the mounts [55].
5.1.2. Vibrational coupling at resonances of the vibration isolation system.
One of the
most troubling environmental couplings early in aLIGO was vibration coupling at 100
Hz and above through the seismic isolation system. Not only were ambient vibration
levels producing noise within a factor of two of the DARM background around 1000
Hz, but the coupling was highly non-linear (see non-linearity discussion in Uncertainties
and Limitations), and it was the only vibrational coupling observed that could produce
noise in the detection frequency band around 100 Hz from a source near 1000 Hz.
One could imagine a rising frequency signal (chirp) from the startup of a motor with
squealing bearings, for example, that would have been able to produce a chirp in DARM
Environmental Noise in Advanced LIGO Detectors
21
in the 100 Hz band. This problem required special vetting of the first GW detections
since, normally, the vetting procedure only assumes linear coupling (discussed further
in Section 6).
The coupling was due to little or no isolation in certain frequency bands associated
with mechanical resonances of the isolation system. The active system vibrationally
isolating the in-vacuum optical tables works mainly below 20 Hz. For higher frequencies,
there are one (HAM chambers) or two (BSC chambers) passive layers associated with
the suspension of the optical tables. But, at the many resonances (violin modes) of the
multiple wire-like flexures that suspend the tables, there was little isolation, allowing
vibration at these frequencies to couple to DARM. This lack of isolation produced
linear vibration coupling at multiple optical tables and, at the dark port table, non-
linear coupling due to an intermodulation of vibration and a strong length dither used
in controlling the length of the output mode cleaner (OMC). The coupling was mitigated
by attaching Viton
TM
[56] to the suspension flexures at tables with coupling to DARM.
To further reduce non-linear coupling, the amplitude of the OMC length dither was
reduced as far as possible [57].
5.1.3. Coupling of wind through ground tilting in the 0.1 Hz band.
Vibrations from
wind affect the interferometer directly in the 10-100 Hz band. At lower frequencies,
particularly in the band around 0.1 Hz, pressure fluctuations associated with wind can
affect performance and duty cycle by tilting the ground. Performance can be affected
by direct tilt of optical table supports or by tilt of ground motion sensors used in the
active isolation system, producing inaccurate signals from sensors that do not distinguish
between tilt and acceleration. Even far from the buildings, we found that the ground
tilts in wind (about 1e-8 radians/sqrt(Hz) at 0.1 Hz in wind reaching 15 m/s at LHO),
to a degree that is consistent with spatially varying wind speeds and Bernoulli forces.
But the tilting in the buildings was a factor of several times larger, and found to be
greatest near the building walls. The pressure fluctuations on the walls are thought to
tilt the wall supports which, in turn, tilt the ground at their base. The coherence length
of floor tilt measured at Hanford was a couple of meters, indicating that the cement
slab does not tilt as a unit. Instead, the tilting is local and mainly within meters of
the base of columns that support the walls, consistent with an elastic dimpling of the
ground around the support [58, 59].
The localized nature of the dominant tilt has led to the simple mitigation technique
of moving ground sensors as far from the walls as possible. While certain sensors could be
moved, the large vacuum chambers near the wall could not, and for future installations,
we have recommended that the chambers be placed at least 10 m from the base of wall
supports.
In order to further mitigate the effects of wind-induced tilt, tilt meters with
improved sensitivity were designed and deployed [60, 61]. The first versions were
produced to correct the artifacts that tilts produce in seismometers, but a table-top
tilt meter is also being developed in order to mitigate the effects of the tilt of the optical
Environmental Noise in Advanced LIGO Detectors
22
tables in the chambers.
Wind fences have been used to reduce wind in agricultural and recreational settings,
and modeling suggested that wind fences may be useful for reducing the effects of wind
pressure on the building walls. For this reason a wind fence was constructed at Hanford,
and is currently being evaluated [62]. One remaining question is how effective a wind
fence is in the troubling frequency band around 0.1 Hz where the length scale is 100m
for 10 m/s wind.
5.1.4. Vibration modulation of scattered light paths.
A major source of detector noise
and reduced sensitivity to GWs is the scattering of light from the beam spot on a test
mass or other optic to surfaces that are moving relative to the optic, like vacuum chamber
walls. A very small fraction of the light reaching the moving surface is reflected to the
originating or another beam spot, where it scatters back into the main interferometer
beam. As the distance to the moving surface changes, the phase of the returning light
changes relative to the main beam, producing fluctuations in the amplitude of the beam,
that, at 1 part in 10
20
can be on the scale of those produced by gravitational waves.
In addition to this sensitivity to recombined scattered light, the scattering noise is
problematic because of non-linear coupling when the path length modulation becomes
comparable to the wavelength of the light, producing noise at harmonics of modulation
frequencies [63].
The subtlety of scattered light noise is illustrated by the mechanism that was behind
a mysterious glitch in DARM that turned out to be produced by ravens [64, 12]. The
Rube Goldberg–like mechanism began in the desert sun at LHO, where ravens pecked
at ice accumulations on a cryopump vent tube just outside of an end station building.
The vibrations from pecking were transmitted through the vent tubes to the cryopump
inside the building. The cryopump was attached to the beam tube, and the vibrations
were transmitted through the beam tube to a calibration structure located inside of
the vacuum, which vibrated slightly with each peck. The structure was angled so as
not to retro-reflect light scattered from the test mass, about 10 m away. However,
polishing grooves on the surface reflected a small fraction of the light back to the
test mass, where a small fraction recombined with the main beam. The interference
between the light in the main beam and the tiny amount of light reflected from the
grooves varied with the motion of the calibration structure produced by each peck.
The varying interference caused fluctuations in the light of the main beam, similar to
the fluctuations produced by gravitational waves. After the discovery of this coupling
mechanism, the calibration structure was baffled to reduce the light scattered back into
the interferometer, eliminating the raven glitches and other similar vibrational signals.
Scattered light baffles can themselves be problematic - for example, vibrations in
the 5-30 Hz band, such as from nearby truck traffic, produced transient noise and limited
detector sensitivity in early aLIGO. Investigations using vibration injections and laser
vibrometry showed that the coupling was due to light reflecting from imperfect light
baffles. The ground motion was amplified by the resonances of the baffles (quality factors