E

Self-Noise Models of Seismic Instruments

A. T. Ringler and C. R. Hutt

A. T. Ringler and C. R. Hutt

U. S. Geological Survey, Albuquerque Seismological Laboratory

Online material: Sensor test data to people working in seismic instrumentation and will contrib-
ute to the effort of continually increasing seismic data quality.
INTRODUCTION
METHODS
Based on coherence analysis methods we develop a method for
computing low self-noise models of seismic sensors. We calcu- We developed noise models for 11 different seismic sensors,
late self-noise models for 11 different production seismometers. including all models of sensors currently in use at Global Seismic
This collection contains the majority of sensors currently in use Network (GSN) stations except for Geotech’s KS-36000,
at Global Seismographic Network Stations. By developing these which is currently in limited use. These models were con-
noise models, with a standard estimation method, we are able structed from noise measurements of 25 different noise tests
to make absolute comparisons between different models of seis- (Table 1). All of these tests were conducted at the Albuquerque
mic sensors. This also provides a method of identifying quality Seismological Laboratory (ASL), which is known to have low
variations between two or more of the same model sensor. ambient background noise (McNamara et al. 2004).
Studying Earth’s free oscillations requires a large amount We tested both vault and borehole type sensors. All vault
of seismic data with a high signal-to-noise ratio at long periods sensor tests were conducted in the ASL underground test vault
(Laske 2004). Recent tomographic studies using ambient seismic in order to minimize ambient background noise. To increase
noise (Shapiro et al. 2005) also require the self-noise of seismic high-frequency coherency we collocated sensors on a granite
instruments to be below that of the Earth’s ambient background slab whenever possible. Our borehole tests were done in the
noise, because as they use Earth noise as the seismic signal. It ASL test borehole, which is located near GSN station ANMO
is also important when making temporary sensor deployments (~30 m separation). In all of the tests we used the three-channel
that the instrument’s noise levels are below that of the signals coherence analysis estimate (Sleeman et al. 2006) to measure
being used in the study (Wilson et al. 2002). In order to verify the self-noise of the instruments (Figure 1). The three-channel
that seismic instruments meet the above demands and other user coherence analysis estimate gives good self-noise estimates in a
requirements it is important from a testing standpoint, that one large frequency band (Ringler et al. forthcoming). However, if
be able to measure the self-noise of seismic sensors and develop one is only interested in self-noise at long periods, one may also
baselines for different models of seismic instruments. use the two-sensor coherence analysis (Holcomb 1989).
The different methods used to estimate self-noise of seismic The details of our processing method are as follows: for
sensors have made it difficult to do side-by-side comparisons of each broad-band sensor test we visually inspected for approx-
their performance (Hutt et al. 2009). This lack of a self-noise imately 10 hours of non-stop quiet data on the BH and LH
estimate standard makes it difficult to assess when a sensor’s vertical channels, which are sampled at 20 or 40 samples per
self-noise is above the manufacturers’ specifications, indicating second (sps) and 1 sps, respectively. We then interpolated the
a possible problem with the sensor or noisy site conditions. In data, in the frequency domain, to 40 sps and 8 sps, respectively
sensor development it is important to be able to compare a pro- (Welch 1967). We computed power spectra and cross-power
totype sensor’s self-noise to that of known self-noise levels of spectra between all sensors, in a given test, by way of the modi-
a reference sensor. On top of these complications some sensor fied Welch method. Our Welch method uses a Hanning win-
self-noise models are based on theoretical models, which can dow taper for windows that contain 64,000 points with 87%
over- or under-account for noise contributions to the sensor overlap (Figure 1). We used a very large amount of overlap to
self-noise (Rodgers 1994). remove the need for excessive smoothing. Once the BH and
With the above in mind, we propose a method to stan- LH channel noise levels have been computed, we use the BH
dardize self-noise estimates for seismic sensors and we calculate data channel noise values for frequencies higher than 0.0075
the self-noise of 11 different seismic sensors. Our hope is that Hz and we use the LH data channel for frequencies lower than
this will lead to robust estimates of sensor self-noise for use in 0.0075 Hz. By using two different sample rates, we were able
sensor development, verifying sensor quality, and give overall to extend the frequency band for which the self-noise estimates
characterizations of current sensor performance by model. We remain valid without having to increase the time series of the
hope that the instrument self-noise models will be a useful tool test (Figure 1). We have not smoothed for any discontinuities;

972 Seismological Research Letters Volume 81, Number 6 November/December 2010 doi: 10.1785/gssrl .81.6.972

Downloaded from http://pubs.geoscienceworld.org/ssa/srl/article-pdf/81/6/972/2763093/972.pdf

by Chongqing University user
 TABLE 1
Different noise test configurations used for constructing noise models. All tests were run for multiple days to ensure the
sensors had settled. The right column indicated if any special configurations were used. All tests included recording
vertical component output. Test 9 included data from a reference sensor whose self-noise measurement was not included
at the request of the vendor.
Test Sensor 1 Sensor 2 Sensor 3 Notes
1 STS-2 HG STS-2 HG STS-2 HG 2 with wool caps, 1 in bell jar
2 STS-2 HG STS-2 HG STS-2 HG 3 with wool caps
3 STS-2 HG STS-2 HG STS-1 STS-2s with wool caps
4 STS-2 HG STS-2 HG T-240 STS-2s with wool caps, T-240 in stock pot
5 STS-2 HG STS-2 HG T-240 STS-2s with wool caps, T-240 in plastic shield
6 STS-2 HG STS-2 HG T-240 STS-2s with wool caps, T-240 in plastic shield and styrofoam box
7 STS-2 HG STS-2 HG STS-2 HG 1 with wool cap, 1 in sand
8 KS-1 STS-2 HG STS-1 STS-2 in fiberglass jacket, STS-1 has Metrozet feedback box
9 — — STS-1 STS-1 has Metrozet feedback box
10 KS-2000 STS-2 HG T-240
11 STS-1 STS-1 STS-2 HG
12 T-240 T-240 T-240
13 T-240 T-240 STS-2 HG
14 T-CMPT STS-2 HG STS-2 HG
15 T-120P T-120P T-120P
16 STS-1 STS-1 CMG-3T
17 STS-1 STS-1 STS-2 HG
18 T-120P T-240 T-240
19 KS-54000 KS-54000 CMG-3TB Borehole test
20 CMG-3TB CMG-3TB KS-54000 Borehole test
21 GS-13 GS-13 GS-13
22 KS-54000 CMG-3TB CMG-3TB Borehole test
23 151-120 151-120 151-120
24 151-120 151-120 151-120
25 151-120 151-120 151-120

instead we judiciously picked a frequency to overlap, at which along with multiple tests. Finally, we compared our self-noise
there were minimal changes in the power levels between the models to theoretical self-noise models, when available, and also
BH and LH data. The processing techniques used for our the Peterson new low noise model (Peterson 1993) for reference.
instrument self-noise models are discussed in detail in Ringler We have also constructed a self-noise model for the pri-
et al. (forthcoming) and Evans et al. (2010). mary model of short-period sensor in use at GSN stations,
Having computed self-noise estimates for each test, we Geotech’s GS-13. In this case, we used only EH channel data
then take the minimum self-noise for all of the self-noise tests which is sampled at 200 sps. For this model we still used the
together as a function of frequency for a given model of sensor same taper, window size, and overlap. Since our estimate only
(e.g., we use the three obtained in Figure 1 along with 23 other used 200 sps data, the self-noise estimates are only good at fre-
self-noise estimates for the STS-2 self-noise model). In each case, quencies higher than approximately 0.01 Hz.
we have included the number of self-noise estimates used in the Although the above method appears to give good estimates
instrument’s model figure. This is the low-noise model for the for the self-noise of a particular model of instrument, there are
sensor. We also calculate a median self-noise model for each sen- some limitations to the method (Wielandt 2002). First, all
sor, which is computed by taking the median self-noise from coherence analysis methods for calculating self-noise assume
all of the tests of a given model of sensor as a function of fre- that all the sensors in the analysis have a common input signal
quency. Finally, we apply a 25% logarithmic smoothing (see the from the Earth. This requires that the sensors under test be adja-
Appendix) to both the low-noise and the median self-noise to cent to each other (within about 0.1 wavelength). For measuring
construct the final instrument self-noise models. By applying a self-noise of borehole sensors, this assumption can be violated
large amount of smoothing, our models had minimal local varia- at short periods because of local geology and also the distance
tions in the self-noise. To increase the robustness of our self-noise between the boreholes. Second, small misalignments between
models, we used multiple self-noise estimates from a given test seismometers can result in large self-noise estimates (Holcomb

Seismological Research Letters Volume 81, Number 6 November/December 2010 973

Downloaded from http://pubs.geoscienceworld.org/ssa/srl/article-pdf/81/6/972/2763093/972.pdf

by Chongqing University user
 Self Noise Vertical Components Duration 70334 seconds
−120
Noise of TST1 00
−130 Noise of TST3 00
Noise of TST2 00
−140 PSD of TST1 00
Power (dB) relative to 1(m/s ) /Hz PSD of TST3 00
−150 PSD of TST2 00
2 2

NLNM
−160

−170

−180

−190

−200

−210

−220

−230 −3 −2 −1 0 1
10 10 10 10 10
Frequency (Hz)

▲▲ Figure 1. Power spectra and self-noise for three collocated Streckeisen STS-2 vertical seismometers from one test (Test 1, Table1),
which were used for making the self-noise models of the STS-2 (TST1 was in a bell jar, TST2 and TST3 were covered with wool caps).
We have not done any post-process smoothing to these self-noise estimates. A possible explanation for the elevated noise levels
between 1 Hz and .1 Hz could come from inaccuracies in the sensors’ level bubbles.

1990). This becomes apparent in the 0.1 Hz to 1 Hz frequency For clarity we have included only the self-noise of the digi-
band (Figure 1), where even vertical components see elevated tizer used in the largest number of tests for a given seismometer
noise in the microseism band. A possible cause of this is limi- model. To allow for a comparison of the noise, we have cor-
tations in the manufacturing methods. Small misalignments rected the self-noise of the digitizer by the instrument response
greatly contribute to horizontal self-noise estimates. Third, vari- of the given seismometer.
able quality between sensors of a given model and variations in
the installation methods can produce differences in the self- Guralp CMG-3T/CMG-3TB
noise estimates (Ringler et al. forthcoming). The Guralp CMG-3TB is the secondary borehole sensor used
at GSN stations. It has a flat velocity response from 120 seconds
RESULTS to 50 Hz (Guralp Systems 2006). The self-noise model for this
instrument (Figure 2) was constructed from 15 power spectra
We now discuss the 11 self-noise models constructed in our during three different tests (Table 1). These tests included a
study. When available, we compare our self-noise models with total of four different CMG-3T/CMG-3TB sensors. We did
theoretical self-noise models, as specified by the manufacturer. not differentiate between the two models of instrument as they
We have also calculated the self-noise for the two models of appear to have similar self-noise characteristics. There was no
digitizers used in the study: Quanterra Q680 and Quanterra theoretical noise model for this sensor.
Q330HR (Quanterra 1993, 2007). For all tests using the
Q330HR, the pre-amplifier gain was set to either X1 gain or Geotech GS-13
X20 gain. Our choice for setting the pre-amplifier gain was The Geotech GS-13 is the primary short-period instrument
based on the instrument’s sensitivity. However, the end results used at GSN stations. It has a natural frequency of 0.75 Hz to
were likely not changed by the pre-amplifier gain settings, in 1.1 Hz, which is usually set at 1 Hz (Geotech 1999). We com-
either case, since the digitizer noise was always well below the pared our calculated self-noise model (Figure 3) to the theoreti-
instrument self-noise. We calculated the self-noise of the digi- cal noise model of a GS-13 operating with a Linear Technology
tizers using a shorted input or a terminated input of 120 ohms LT1012 operational amplifier (Rodgers 1994). In total we used
from each side to ground. In all cases, the digitizer self-noise 12 power spectra from one test (Table 1) from three different
was below that of the seismometer for the respective test. GS-13 sensors.

974 Seismological Research Letters Volume 81, Number 6 November/December 2010

Downloaded from http://pubs.geoscienceworld.org/ssa/srl/article-pdf/81/6/972/2763093/972.pdf

by Chongqing University user
 Noise Model for CMG−3T/CMG−3TB
−120
Low Noise
−130 Median Noise
NLNM
−140 Q680 Noise X1 gain

Power (dB) relative to 1(m/s ) /Hz

−150
2 2

−160

−170

−180

−190

−200

−210

−220

−230 −3 −2 −1 0 1
10 10 10 10 10
Frequency (Hz)

▲▲ Figure 2. Self-noise model of the Guralp CMG-3T/CMG-3TB broadband sensor.

Noise Model for GS−13

−120
Low Noise
−130 Median Noise
Theoretical
−140 NLNM
Q330 Noise X20 gain
Power (dB) relative to 1(m/s ) /Hz

−150
2 2

−160

−170

−180

−190

−200

−210

−220

−230 −2 −1 0 1 2
10 10 10 10 10
Frequency (Hz)

▲▲ Figure 3. Self-noise model for the Geotech GS-13 short-period instrument. The theoretical self-noise is for the GS-13 seismometer
connected to a Linear Technology LT1012 operational amplifier.

Seismological Research Letters Volume 81, Number 6 November/December 2010 975

Downloaded from http://pubs.geoscienceworld.org/ssa/srl/article-pdf/81/6/972/2763093/972.pdf

by Chongqing University user
 Geotech KS-1 sensor. The sensor has a flat velocity response from 0.003 Hz
The Geotech KS-1 is a vault version of the KS-54000 sensor. to 5 Hz (Geotech 2004). We compared our calculated self-
This sensor has a flat velocity response from 0.003 Hz to 5 Hz noise model (Figure 5) to the theoretical noise model of the
(Geotech 2004). As we were only able to acquire one vertical KS-54000. Our self-noise model for the KS-54000 used a total
time series test, we have also included horizontal records in the of 11 power spectra from two different sensors during three
construction of the self-noise (Figure 4). The inclusion of hori- different tests.
zontal data could contribute to an elevated instrument noise-
model. However, the KS-1’s self-noise model is similar to the Nanometrics Trillium 120P
KS-54000’s self-noise model (Figure 5), as we would expect. We The Nanometrics Trillium 120P broadband seismometer is a
have not constructed a median self-noise model because of the vault style seismometer. The sensor has a flat velocity response
limited amount of data. Our model was constructed using three from 120 seconds to 175 Hz (Nanometrics 2006). To calcu-
power spectra from one sensor. We compared our calculated late the self-noise (Figure 7) of the Trillium 120P we used five
self-noise model to the theoretical noise model of the KS-54000. power spectra during two tests (Table 1) from five different
Trillium 120P sensors.
Geotech KS-2000
The Geotech KS-2000 is a compact version of the KS-54000. Nanometrics Trillium 240
The KS-2000 has a flat velocity response from 120 seconds The Nanometrics Trillium 240 broadband seismometer is a
to 50 Hz (Geotech 2006). In our tests only the vault version vault style seismometer. This sensor has a flat velocity response
was tested. As we were only able to acquire two vertical time from 244 seconds to 207 Hz (Nanometrics 2005). Our self-
series tests, we have included horizontal records. Therefore, noise model (Figure 8) for the Trillium 240 was calculated
the median self-noise model is probably elevated slightly. The from 11 power spectra during seven tests (Table 1) from four
self-noise model for this instrument (Figure 6) was constructed different sensors.
from eight power spectra during one test (Table 1) of a single
KS-2000. There was no theoretical noise model for this sensor. Nanometrics Trillium Compact
The Nanometrics Trillium Compact broadband seismom-
Geotech KS-54000 eter is a vault style seismometer. The sensor has a flat velocity
The Geotech KS-54000 is the primary borehole sensor used at response from 120.2 seconds to 108 Hz (Nanometrics 2009).
GSN stations. The KS-54000 is a borehole version of the KS-1 Our self-noise model (Figure 9) for the Trillium compact was
Noise Model for KS−1
−120
Low−Noise
−130 Theoretical
NLNM
−140 Q680 Noise X1 gain
Power (dB) relative to 1(m/s2)2/Hz

−150

−160

−170

−180

−190

−200

−210

−220

−230 −3 −2 −1 0 1
10 10 10 10 10
Frequency (Hz)

▲▲ Figure 4. Self-noise model of the Geotech KS-1 vault seismometer. This model includes horizontal data because we were unable to
get multiple quiet vertical component time series because of the limited number of tests conducted with this model of sensor.

976 Seismological Research Letters Volume 81, Number 6 November/December 2010

Downloaded from http://pubs.geoscienceworld.org/ssa/srl/article-pdf/81/6/972/2763093/972.pdf

by Chongqing University user
 Noise Model for KS−54000
−120
Low Noise
−130 Median Noise
Theoretical
−140 NLNM
Power (dB) relative to 1(m/s2)2/Hz Q680 Noise X1 gain
−150

−160

−170

−180

−190

−200

−210

−220

−230 −3 −2 −1 0 1
10 10 10 10 10
Frequency (Hz)

▲▲ Figure 5. Self-noise models for the KS-54000 compared to the theoretical self-noise for the KS-54000.

Noise Model for KS−2000

−120
Low Noise
−130 Median Noise
NLNM
−140 Q330 Noise X20 gain
Power (dB) relative to 1(m/s2)2/Hz

−150

−160

−170

−180

−190

−200

−210

−220

−230 −3 −2 −1 0 1
10 10 10 10 10
Frequency (Hz)

▲▲ Figure 6. Self-noise model for the Geotech KS-2000 broadband seismometer. This self-noise model includes horizontal component
data because of insufficient vertical data.

Seismological Research Letters Volume 81, Number 6 November/December 2010 977

Downloaded from http://pubs.geoscienceworld.org/ssa/srl/article-pdf/81/6/972/2763093/972.pdf

by Chongqing University user
 Noise Model for T−120P
−120
Low Noise
−130 Median Noise
Theoretical
−140 NLNM
Q330 Noise X20 gain
Power (dB) relative to 1(m/s2)2/Hz

−150

−160

−170

−180

−190

−200

−210

−220

−230 −3 −2 −1 0 1
10 10 10 10 10
Frequency (Hz)

▲▲ Figure 7. Self-noise model for the Nanometrics Trillium 120P broadband seismometer compared to the theoretical noise model.

Noise Model for T−240

−120
Low Noise
−130 Median Noise
Theoretical
−140 NLNM
Q330 Noise X20 gain
Power (dB) relative to 1(m/s2)2/Hz

−150

−160

−170

−180

−190

−200

−210

−220

−230 −3 −2 −1 0 1
10 10 10 10 10
Frequency (Hz)

▲▲ Figure 8. Self-noise model of the Nanometrics Trillium 240 broadband seismometer. We have compared our self-noise model with
theoretical self-noise supplied by Nanometrics for this seismometer.

978 Seismological Research Letters Volume 81, Number 6 November/December 2010

Downloaded from http://pubs.geoscienceworld.org/ssa/srl/article-pdf/81/6/972/2763093/972.pdf

by Chongqing University user
 Noise Model for T−Compact
−120
Low Noise
−130 Median Noise
Theoretical
−140 NLNM
Power (dB) relative to 1(m/s2)2/Hz Q330 Noise X20 gain
−150

−160

−170

−180

−190

−200

−210

−220

−230 −3 −2 −1 0 1
10 10 10 10 10
Frequency (Hz)

▲▲ Figure 9. Self-noise model of the Nanometrics Trillium compact seismometer compared to its theoretical noise model.

calculated from six power spectra during one test (Table 1) sensors. A possible explanation for our instrument self-noise
using one sensor. model being below the theoretical model at high frequencies
could be from improved sensor electronics in newer STS-2
Ref Tek 151-120 HGs.
The Ref Tek 151-120 broadband seismometer is a vault style
seismometer. The sensor has a flat velocity response from 120 DISCUSSION
seconds to 50 Hz (Ref Tek 2010). Our self-noise model (Figure
10) for the Ref Tek 151-120 was calculated from nine power We constructed self-noise models from 11 different models of
spectra during three tests (Table 1) from five different sensors. seismometers, including 10 broadband and one short period
instrument (Figure 13 and Figure 14). Our 11 different self-
Streckeisen STS-1 noise models included all models of sensors that are currently in
The Streckeisen STS-1 broadband seismometer is the primary operation at GSN stations, except for the KS-36000. Although
vault style sensor used at GSN stations. The STS-1 seismom- the number of tests and power spectra for various sensors dif-
eter has a flat velocity response from 360 seconds to 10 Hz fered, in all cases we attempted to use enough different power
(Streckeisen 1987). Our self-noise model (Figure 11) for the spectra to give a good estimate of the self-noise of the particu-
STS-1 was calculated from 16 power spectra during six tests lar model of sensor. Although a “good estimate” is a subjective
(Table 1) from four different sensors. A possible explanation measure, we found that the computed self-noise models did not
for our instrument self-noise model being below the theoretical change dramatically when we increased the number of power
model at high frequencies could be from lower electronics noise spectra used. That is, a self-noise model for a particular sensor
in the new Metrozet E300 replacement feedback boxes, which computed with only a few power spectra (e.g., fewer than five)
were used in several of the STS-1 tests. did not differ much from using a lot of power spectra (e.g., more
than 20).
Streckeisen STS-2 HG One of the important factors in calculating self-noise
The Streckeisen STS-2 high gain (HG) broadband seismom- models is to use power spectra from time periods with low
eter is the main secondary vault style sensor used at GSN sta- noise. This is why all power spectra used, for broadband seis-
tions. The STS-2 HG seismometer has a flat velocity response mometer noise model estimates were low-pass filtered and visu-
from 120 seconds to >50 Hz (Streckeisen 1995). Our self-noise ally inspected to have no large signals (e.g., earthquakes).
model (Figure 12) for the STS-2 HG was calculated from 25 The requirement that the power spectra be computed
power spectra during 13 tests (Table 1) using eight different during quiet time segments comes from the limitations of the

Seismological Research Letters Volume 81, Number 6 November/December 2010 979

Downloaded from http://pubs.geoscienceworld.org/ssa/srl/article-pdf/81/6/972/2763093/972.pdf

by Chongqing University user
 ▲▲ Figure 10. Self-noise model of the Ref Tek 151-120 broadband seismometer compared to its theoretical noise model.

Noise Model for STS−1

−120
Low Noise
−130 Median Noise
Theoretical
−140 NLNM
Q330 Noise X1 gain
Power (dB) relative to 1(m/s2)2/Hz

−150

−160

−170

−180

−190

−200

−210

−220

−230 −3 −2 −1 0 1
10 10 10 10 10
Frequency (Hz)

▲▲ Figure 11. Self-noise model of the Streckeisen STS-1 seismometer as compared to the theoretical self-noise of the STS-1. The
self-noise model being lower than the theoretical model at high frequencies could possibly come from using the new Metrozet E300
feedback boxes, which appear to have lower high-frequency electronic noise.

980 Seismological Research Letters Volume 81, Number 6 November/December 2010

Downloaded from http://pubs.geoscienceworld.org/ssa/srl/article-pdf/81/6/972/2763093/972.pdf

by Chongqing University user
 Noise Model for STS−2 HG
−120
Low Noise
−130 Median Noise
Theoretical
−140 NLNM
Power (dB) relative to 1(m/s2)2/Hz Q330 Noise X1 gain
−150

−160

−170

−180

−190

−200

−210

−220

−230 −3 −2 −1 0 1
10 10 10 10 10
Frequency (Hz)

▲▲ Figure 12. Self-noise model of the Streckeisen STS-2 HG seismometer compared to the theoretical noise model for the STS-2 HG.

coherence analysis method used (Ringler et al. forthcoming). conducted our tests at GSN site locations that may be quieter
In order to get robust measurements it is also important that in the long-period band (e.g. GSN station BFO [Black Forest
the sensors be collocated, since the assumption of the self- Observatory, Schiltach, Germany] or GSN station PFO [Pinon
noise estimate is that all three sensors have a common input Flat, California, U.S.A.]).
signal. Since this assumption is possibly violated with the bore- By calculating self-noise models for both the STS-1 and
hole seismometers tested (which were separated by ~30 m), it STS-2 HG seismometers, we see that there is approximately a
is possible that their self-noise models are slightly elevated at 7-dB difference in noise levels between the sensors at 1 mHz
high frequencies. We did not find any systematic differences in (Figures 11 and 12). This difference in noise level is smaller
instrument self-noise estimates at low frequencies based on the than noted by Sleeman et al. (2006) and Widmer-Schnidrig
location of the instrument’s low-frequency corner. However, (2003). However, it is larger than that noted by Berger et al.
when estimating self-noise it is desirable to use instruments (2004). Possible causes for these differences in noise levels have
with similar expected noise characteristics (Hutt et al. 2009). been attributed to the quality of installation of the sensors and
We also did not find any systematic differences between the the variable quality of the STS-2 seismometer (Sleeman et al.
vertical component of Galperin tri-axial suspension seis- 2006). In our study we tested the STS-2 HG seismometer in
mometers and traditional vertical component seismometers. 13 different tests. These 13 different tests included four differ-
Regardless of suspension type the elevated self-noise estimates ent installation methods and a total of eight different STS-2
in the microseism band do indicate that small misalignment HG seismometers. We believe the diversity of tests and sensors
errors contribute to the self-noise in the frequency band with used helps to exclude the above-mentioned possible causes for
the highest power levels (Holcomb 1990). the differences in the self-noise of the STS-1 and STS-2 HG
We have compared all of our low and median self-noise seismometers at 1 mHz.
models (Figures 13 and 14, respectively) with the GSN noise
model (Berger et al. 2004). At low frequencies, an instru- ACKNOWLEDGMENTS
ment’s self-noise is generally within a few dB of its power spec-
tra (Figure 1). The similar aggregate noise levels between the We would like to thank the vendors of the seismic instru-
GSN noise model and the self-noise model of the STS-1 at low ments used in this study for their encouragement to conduct
frequencies (<0.001 Hz) gives an argument for the robustness this study. We would also like to thank D. L. Anderson and
of our self-noise models. That is, we would not expect to see L. G. Holcomb for help with the noise tests and other useful
large differences in our instrument self-noise models had we discussions. Finally, we would like to thank an anonymous

Seismological Research Letters Volume 81, Number 6 November/December 2010 981

Downloaded from http://pubs.geoscienceworld.org/ssa/srl/article-pdf/81/6/972/2763093/972.pdf

by Chongqing University user
 Low Noise Models for All Sensors

Power (dB) rel to 1(m/s2)2/Hz

−120
CMG−3TB
−140 GS−13
KS−1
−160 KS−2000
KS−54000
−180 NLNM
GSN NM
−200
−3 −2 −1 0 1 2
10 10 10 10 10 10
Frequency (Hz)
Power (dB) rel to 1(m/s2)2/Hz

−120
T−120
−140 T−240
T−Compact
−160 151−120
STS−1
−180 STS−2
NLNM
−200 GSN NM

−3 −2 −1 0 1 2
10 10 10 10 10 10
Frequency (Hz)

▲▲ Figure 13. Low self-noise models for all seismometers in the study.

Median Noise Models for All Sensors

Power (dB) rel to 1(m/s2)2/Hz

−120
CMG−3TB
−140 GS−13
KS−2000
−160 KS−54000
NLNM
−180 GSN NM

−200
−3 −2 −1 0 1 2
10 10 10 10 10 10
Frequency (Hz)
Power (dB) rel to 1(m/s2)2/Hz

−120
T−120
−140 T−240
T−Compact
−160 151−120
STS−1
−180 STS−2
NLNM
−200 GSN NM

−3 −2 −1 0 1 2
10 10 10 10 10 10
Frequency (Hz)

▲▲ Figure 14. Median self-noise models for all sensors in the study. We have not included a median self-noise model for Geotech’s KS-1
because of insufficient data.

982 Seismological Research Letters Volume 81, Number 6 November/December 2010

Downloaded from http://pubs.geoscienceworld.org/ssa/srl/article-pdf/81/6/972/2763093/972.pdf

by Chongqing University user
 reviewer, L. Astiz, L. G. Holcomb, and R. A. Uhrhammer for Sleeman, R, A. van Wettum, and J. Trampert (2006). Three-channel cor-
very detailed and helpful reviews of this manuscript that ulti- relation analysis: A new technique to measure instrumental noise of
digitizers and seismic sensors, Bulletin of the Seismological Society of
mately improved the presentation. America, 96 (1), 258-271.
Streckeisen, G. (1987). Very-broad-band Feedback Seismometers, STS-1
REFERENCES V/VBB and STS-1 H/VBB Manual, 46 pps.
Streckeisen, G. (1995). Portable Very-broad-band Tri-axial Seismometer
Berger, J., P. Davis, and G. Ekström (2004). Ambient Earth noise: A STS-2 Low-Power Manual, 50 pps.
survey of the global seismographic network, Journal of Geophysical Welch, P. D. (1967). The use of fast Fourier transform for the estima-
Research, 109, B11307, doi:10.1029/2004JB003408. tion of power spectra: A method based on time averaging over
Evans, J. R., F. Followill, C. R. Hutt, R. P. Kromer, J. M. Steim, R. L. short, modified Periodograms. IEEE Transactions on Audio and
Nigbor, A. T. Ringler, and E. Wielandt (2010). Method for cal- Electroacoustics AU-15, 70–73.
culating self-noise spectra and operating ranges for seismographic Widmer-Schnidrig, R. (2003). What can superconducting gravi-
inertial sensors and recorders, Seismological Research Letters, 81, meters contribute to normal-mode seismology?, Bulletin of the
(4), 640-646. Seismological Society of America, 93 (3), 1,370-1,380.
Geotech Instruments (1999). Operation and Maintenance Manual Wielandt, E. (2002). Seismic sensors and their calibration. In
Portable Short-period Seismometer, Model GS-13, Stock Number New Manual of Seismological Observatory Practices, 1–46.
990-55400-9800, 50 pps. GeoForschungsZentrum Potsdam. is http://www.gfz-potsdam.de/
Geotech Systems (2004). Borehole Seismometer System Model 54000- portal/
0103 Operation and Maintenance Manual, Part Number 990- Wilson, D., J. Leon, R. Aster, J. Ni, J. Schlue, S. Grand, S. Semken, S.
54000-9803. Baldridge, and W. Gao (2002). Broadband seismic background
Geotech Systems (2006). Broadband Seismometer Models KS-2000 and noise at temporary seismic station observed on a regional scale in
KS-2000M operation manual, 30 pps. the southwestern United States, Bulletin of the Seismological Society
Guralp Systems (2006). CMG-3TB Operator’s Manual, Part MAN- of America, 92 (8), 3,335-3,342.
BHO-0001, Issue C, 78 pps.
Holcomb, L. G. (1989). A Direct Method for Calculating Instrument U.S. Geological Survey
Noise Levels in Side-by-Side Seismometer Evaluations, USGS Open-
File Report 89-214, 34 pps.
Albuquerque Seismological Laboratory
Holcomb, L. G. (1990). A Numerical Study of Some Potential Sources of P.O. Box 82010
Error in Side-by-Side Seismometer Evaluation, USGS Open-File Albuquerque, New Mexico 87198-2010 U.S.A.
Report 90-406, 41 pps. [email protected]
Hutt, C. R., R. L. Nigbor, and J. R. Evans, eds. (2009). Proceedings of the (A. R.)
Guidelines for Seismometer Testing Workshop, Albuquerque, New
Mexico, 9–10 May 2005 (“GST2”), USGS Open-File Report 2009-
1055, 48 pps. APPENDIX
Laske, G. (2004). The needs for/of low-frequency seismology. IRIS
broadband instrumentation workshop: http://www.iris.edu/sta- In order to remove large variations in our noise estimates and
tions/seisWorkshop04/abstracts/laske-abstract.pdf. construct sensor self-noise models, we employed a moving aver-
McNamara, D. E. and R. P. Buland (2004). Ambient noise levels in the
continental United States, Bulletin of the Seismological Society of
age where the number of points averaged is based on the num-
America, 94 (4), 1,517-1,527. ber of points that have not yet been averaged in the power spec-
Nanometrics (2005). Trillium 240 seismometer user’s guide, Part tra. A similar smoothing method has been used for two-sensor
15672R3, 42 pps. self-noise estimates (Holcomb 1989). The assumption used in
Nanometrics (2006). Trillium 120P seismometer user’s guide, Part the smoothing method is that since there are fewer power spec-
15149R3, 44 pps.
Nanometrics (2009). Trillium Compact seismometer user’s guide, Part
tra values at low frequencies, one does not want to bias the low-
16889R1, 67 pps. frequency end of the power spectra by averaging these points
Peterson, J. (1993). Observations and Modeling of Seismic Background with high-frequency power spectra values.
Noise, USGS Open-File Report 93-322, 94 pps. Suppose the length of our power spectrum, to which we
Quanterra (1993). Q680/SXS-G Operation documentation for software want to apply our smoothing, is of length m, and we want to
release 34/02-RL1231s, 144pps.
Quanterra (2007). Q330 Operations guide: Q330HR/Q330 operation
smooth by 25% of the remaining length. Let x(n) denote the
overview of support tools baler operation, Document 304808, 318 nth value in our power spectra and let r = 0.25(m – n). We
pps. further assume that x(n) is ordered so that x(1) is the highest
Ref Tek (2010). 151 Broadband Seismometer Operations, 53 pps. frequency value and our power spectra is ordered in terms of
Rodgers, P. W. (1994). Self-noise for 34 common electromagnetic seis- decreasing frequency. Then the averaged value y(n) for x(n),
mometer/preamplifier pairs, Bulletin of the Seismological Society of
America, 84 (1), 222-228.
when n > r is given by taking the mean of the values x(n – r),
Ringler, A. T., J. R. Evans, C. R. Hutt, L. D. Sandoval (forthcoming). A x(n – r + 1), …, x(n + r – 1), x(n + r). In order to deal with the
comparison of instrument self-noise analysis techniques. Bulletin of high-frequency points, where there are not enough points of
the Seismological Society of America. higher frequency that are still lower than the Nyquist number
Shapiro, N.M., M. Campillo, L. Stehly, and M.H. Ritzwoller (2005). to be averaged, one can apply a forward difference or simply
High resolution surface wave tomography from ambient seismic
noise, Science, 307 (5,715), 1,615-1,618.
remove those points.

Seismological Research Letters Volume 81, Number 6 November/December 2010 983

Downloaded from http://pubs.geoscienceworld.org/ssa/srl/article-pdf/81/6/972/2763093/972.pdf

by Chongqing University user