Earthquake Reaearch in China  2019, Vol. 33 Issue (1): 9-23     DOI: 10.19743/j.cnki.0891-4176.201901010
New Explorations of Laboratory and On-site Testing of Broadband Seismographs in China
XIE Jianbo1, YANG Dake2, LI Xiaojun3, YUAN Songyong2, TONG Wanglian4, MA Jiemei2, XU Weiwei2, DING Lisha1, YE Shishan1
1. Guangdong Earthquake Agency, Guangzhou 510070, China;
2. Institute of Geophysics, China Earthquake Administration, Beijing 100081, China;
3. Hebei Earthquake Agency, Shijiazhuang 050021, China;
4. Yunnan Earthquake Agency, Kunming 650224, China
Abstract: According to different testing purposes, methods and available environmental conditions, the seismograph testing can be divided into laboratory and on-site testing, respectively. The testing of the seismograph's key parameters and other concerning technical specifications are well described in guide documents (China Earthquake Administration, 2007). This includes seismometer sensitivity, linearity and clip levels based on the shake table test, as well as the seismometer natural period, damping constant based on electrical calibration (Wang Guangfu, 1986; Plešinger A., 1993) and instrumental self-noise collocation estimation (Holcomb L.G., 1989; Sleeman R. et al., 2006). However, with the development of seismic observation technology, many new requirements for the performance evaluation of seismographs have been put forward, and new testing items and methods have emerged.
Key words: Broadband seismographs     Laboratory testing     On-site testing

INTRODUCTION

China is a vast country that has suffered from frequent earthquake disasters. In order to cope with the occurrence and research of earthquake disasters, China has established a large-scale seismic observation network. China's seismic observations have traditionally been distinguished as weak ground motion, strong ground motion, ground strain, geo-electrical and geo-magnetic observations. In China, broadband seismographs are mainly used for weak ground motion observations (Xie Jianbo et al., 2014), which includes the national seismograph network with about 170 stations and 31 regional seismic networks with about 900 provincial-level stations. To ensure the stable and reliable operation of this huge observation network and to ensure the quality of recorded data, the seismograph testing issues are very important. This paper introduces some preliminary explorations on broadband seismograph testing in China.

1 SEISMOMETER SHAKE TABLE TEST

In the seismometer shake table test, some technical details may seriously affect the test results. Here are three typical technical issues that are encountered.

The movement of the shake table top along the guide rail during the test is generally considered to be a linear motion, but in reality, the guide rail of the shake table may incline at a tiny angle. The deviation of the trajectory of the shake table from the straight-line results in a larger value than the expected one when the seismometer sensitivity is calculated according to the set input long-period signal amplitude. However, this error has a certain degree of repeatability in the case where the shake table movement is fixed. The influence factor of the guide rail tilting can be fitted and calculated by multiple repetition tests, and then the sensitivity test in the low-frequency band can be corrected. Assuming that the shake table displacement is always in the interval -xm~+xm, the moving period is T0, and the peak value of the output voltage of the seismometer is Vmax, the first-order approximation of the sensitivity of the seismometer including the inclination of the guide rail is given by Xue Bing as:

 $S = \frac{{{T_0}{V_{{\rm{max}}}}}}{{2\pi {x_m}}}\frac{1}{{\sqrt {1 + \frac{{T_{_0}^4k_0^2}}{{16{\pi ^4}x_m^2}}} }}$ (1)

Where k0 is the influence factor that needs to be fitted to the determination.

In the cross-axis rejection test, the placement error of the seismometer on the horizontal shake table surface, the direction-marking line on the seismometer shell and the orthogonality error of the actual sensing axis are often neglected, which may lead to a result that the cross-axis rejection testing result of the seismometer does not reflect the true performance of the seismometer. In response to this problem, Ma Jiemei et al. (2016)used fixtures and dials to manually set different installation angles of the seismometer, obtained the seismometer output corresponding to the installation angle, and determined the actual sensitive axis of the horizontal movement by a specially designed initial declination calculation program. According to the marking on the dial, they then re-determined the installation posture of the seismometer tested, this method effectively improves the test accuracy of the cross-axis rejection of the seismometer (Ma Jiemei et al., 2016). In fact, after eliminating the installation orientation error of the seismometer, the cross-axis rejection of the three-component seismometer is mainly caused by orthogonality errors. Using the dial's multi-angle installation test and solving the actual sensitive axis direction of the seismometer horizontal movement, a practical seismometer horizontal axis orthogonality testing method is given.

There is harmonic distortion in the movement of the shake table, and the degree of distortion directly affects testing accuracy. To ensure testing accuracy, the national standard GB/T13823.11-1995 (National Institute of Metrology, China, 1995) stipulates that for test frequencies higher than 0.1Hz, the acceleration waveform distortion of the shake table should be less than 3%, while the International Organization for Standardization of Switzerland requires that the acceleration harmonic distortion of power amplifiers and shake tables in the frequency range from 1Hz to 10kHz does not exceed 2% (International Organization for Standardization of Switzerland, 1999). However, there is no relevant standard requirement for testing in the frequency band lower than 0.1Hz. Today's seismic observation equipment is dominated by broadband seismographs, and there is an urgent need for low-frequency and very low-frequency testing based on shake tables. To this end, Ma Jiemei et al. (2014) combined the theoretical derivation with actual measurements, and concluded that the acceleration waveform distortion has a greater influence on the measured value than the velocity waveform distortion, and once the acceleration waveform distortion index is used as a constraint, it is beneficial to define the reasonable determination range of the shake table test data, which can provide a basis for further research based on the ultra-low frequency seismometer testing technology, and has positive significance for seismometer shake table testing (Ma Jiemei et al., 2014).

2 STUDY ON THE ACCURACY OF SEISMOMETER TRANSFER FUNCTION

The accuracy of the seismometer transfer function could be measured in several respects, e.g. sensitivity is measured by the shake table and the electrical calibration using a step signal is used to measure the natural period and damping constant, while the high-frequency characteristic could be measured based on a pseudo-random binary signal (Berger J. et al., 1979; Wang Guangfu, 1986). The above methods for measuring the seismometer transfer function using electrical calibration signals generally require the application of complex nonlinear fitting iterations (Wang Guangfu, 1986; Plešinger A., 1993) or system identification algorithms (Berger J. et al., 1979). Parameters such as natural period, damping constant, and pole-zero that meet the accuracy requirements can be generally given along with the data approximation residuals during the iterative process and the final stage. The results of the relative electrical method for calibrating the seismometer will vary slightly with the processing method, algorithm and data selection. Therefore, people are constantly researching methods suitable for different needs and improving the accuracy of the results, such as the calibration of ultra and very broadband seismometers (Zhou Yunyao, 2004; Cai Yaxian and Lü Yongqing, 2005); the automatic processing of sinusoidal calibration (Lü Yongqing and Cai Yaxian, 2007); the automatic processing of step calibration (Zhu Xiaoyi et al., 2010); improved pseudo-random binary calibration (Lin Zhan et al., 2015) and etc.

Due to historical reasons, the seismic network of China has only required the accuracy of the seismometer parameters to be within 5% accuracy of the standard value. In 2007, the requirements were raised to 3% in the updated documents (China Earthquake Administration, 2007). When the seismometer parameters deviate from the standard value, the response of the seismometer will be changed. In such cases, the difference in instrument response needs to be given attention. Transfer function simulations can be used to evaluate typical transient responses such as step responses to understand the effect that parameter changes can have on recorded data. The transfer function simulation for parameter variation is much simpler than the nonlinear fitting iteration or system identification algorithm. It can be directly performed in the time domain using a differential equation transformed from the seismometer transfer function by the bilinear transformation. Fig. 1 shows some simulation results calculated from a CTS-1E seismometer whose parameter deviates from standard value (with standard period 120s and damping constant 0.707, in contrast with the measured period 121.7s and measured damping constant 0.746), including comparisons of measured and simulated standard step response with a STS-2 standard response: (a) shows the normalized step calibration source signal and the STS-2 normalized simulation response calculated by the standard transfer function given by IRIS (Incorporated Research Institutions for Seismology), (b) shows the normalized step response of the CTS-1E seismometer with the measured parameter deviation and the simulated standard parameters, and (c) shows the residual of normalized step responses of the CTS-1E seismometer with the measured parameter deviation and the simulated standard parameters referred to a standard STS-2.

 Fig. 1 Difference in step response of CTS-1E parameter deviation
3 COMPARATIVE STUDY OF TIME DOMAIN AND FREQUENCY DOMAIN DIFFERENCES IN SEISMIC RECORDS

The accuracy of seismometer parameters can be determined from shake table testing and electrical calibration. The transfer function simulation method can be used to evaluate the differences of the transient or steady-state response of the instrument caused by parameter deviation. Besides, we also care about the difference in real observation records of the different instruments in daily life. Xie Jianbo et al. introduced two kinds of analysis studies on the amplitude absolute difference and correlation comparison of full-band signals in the time domain (Xie Jianbo et al., 2014).

In order to understand the detailed influence of the parameter difference, the analysis of the measured data could be done in several separated frequency bands by using these two time-domain methods. Fig. 2 shows the waveform comparison of the M8.1 earthquake in Nepal on April 25, 2015 recorded by the system "EDAS-24L+CTS-1EF" and the reference system "Q330HRS+STS2.5" at Jiujiang station for the east-west component, in which a filter with frequency band from 0.00833Hz to 0.02Hz is applied, and the transfer function correction is performed within the filter band. The results show that a larger difference in the coda wave at about 1100-1500s than the other parts. Fig. 3 shows a correlation analysis result of the up-down component in the frequency band from 0.0167Hz to 10Hz. The results show that there could be a time difference of about 1.63 times for the sampling interval between the on-site record and the reference instrument record at Jiujiang station.

 Fig. 2 Absolute amplitude comparison in time domain

 Fig. 3 Time domain correlation comparison

Differences can also be compared for the recorded data spectrum. Fig. 4 is a comparison of frequency domain differences in the same frequency band as in Fig. 2. The results show that the long-period components recorded by the on-site instruments are higher than those by the reference instruments at Jiujiang station.

 Fig. 4 Comparison of frequency domain differences
4 COLLOCATION TESTS ON SEISMOMETER SENSITIVITY

Seismometer sensitivity needs to be calibrated by a shake table. Pavlis and Vernon proposed a new way to evaluate the sensitivity of seismometers using the empirical transfer function estimation in the frequency domain (Pavlis G. L. and Vernon F.L., 1994), in which they argued that the seismometer to be tested and the precisely calibrated reference seismometer should be installed as close as possible and in parallel (so as to assume they recorded the same ground motion). They used a very complicated quantile statistical algorithm to weight the smoothed data spectrum to obtain the empirical transfer function estimates. Havskov J. and Alguacil G. used another method of cross-correlation between the records of two co-located seismometers based on the general principle of signal processing to smoothly suppress uncorrelated spectral noise to obtain the empirical transfer function estimates (Havskov J. and Alguacil G., 2007). Lin Zhan et al. calculated the sensitivity of the instrument under test by calculating the power of two co-located records in a narrow band to avoid the problem of large dispersion of signal power spectral density at a single discrete frequency, and evaluated the accuracy of the sensitivity result by calculating the signal-to-noise ratio of the measured data using the square coherence function (Lin Zhan et al., 2013).

The collocation test of seismometer sensitivity is simpler and easier than the shake table test, and the cost is also comparatively lower. On the other hand, comparing to the electrical calibration test, it avoids the problem of the calibration coil parameters that may have uncertainty. It was demonstrated that the sensitivity can be evaluated with sufficiently high precision under conditions that the co-located sensors were placed close enough and parallel with each other in the installation, and the data is spectra-smoothed (Pavlis G.L.and Vernon F.L., 1994; Havskov J.and Alguacil G., 2007) or filtered in a narrow-band (Lin Zhan et al., 2013) to achieve a sufficiently high signal-to-noise ratio.

5 COLLOCATION TESTS ON SEISMOMETER SELF-NOISE

Self-noise is an important technical indicator of a seismometer and plays an important role in recording data quality. To objectively and accurately assess seismometer self-noise, Holcomb and Sleeman et al. have successively proposed self-noise testing methods using two and three co-located seismometers which are mounted as close and parallel as possible to record data simultaneously. Rather than extracting correlated signal information from the co-located recorded data in seismometer sensitivity testing, the seismometer self-noise testing aims at extracting uncorrelated noise information from the recorded data (Holcomb L.G., 1989; Sleeman R. et al., 2006).

Many factors could have influence on the seismometer self-noise test results. The most important ones are the orientation consistency of the co-located seismometers (Holcomb L.G., 1990), the temperature of the test environment and airflow pressure disturbance (Ringler A.T. et al., 2011). Sleeman et al. proposed to pay attention to the effects of digitizer self-noise when measuring seismometer self-noise (Sleeman R. et al., 2006). Li and Teng analyzed the influence from the digitizer self-noises using simulation method. They found that when the digitizers with higher noise levels were employed, the obtained seismometer self-noise result would show a correspondingly higher level (Li Caihua and Teng Yuntian, 2014).

Xu Weiwei et al. introduced the simultaneous synchronization test of multiple seismometers and improved test efficiency and reliability of seismometer self-noise test results (Xu Weiwei et al., 2017). They argued that the consistency test of a large number of seismometers of the scientific seismic array of China is of great significance. They also showed the effects of digitizer self-noise on the seismometer self-noise testing using real data examples, e.g., Reftek 130 and Reftek 130S digitizer tests show a significant difference when different gain modes are adopted in the tests. It is typically 20dB higher using the normal gain mode versus using the high gain mode. When testing the CMG-3T seismometer self-noise under the normal gain mode configuration of a Reftek 130S digitizer, the test result only reflects the digitizer contribution at the frequency band above 0.3Hz.

In the process of seismometer self-noise test, even if the data and algorithm used are exactly the same, the different results would be obtained if the calculation parameters are different. To this end, Ringler et al recommended using uniform parameters for data processing (Ringler A.T. et al., 2011). Li Xiaojun et al. conducted research on the possible effects of different parameters in detail (Li Xiaojun et al., 2015). They adopted the three-channel method proposed by Sleeman et al. to calculate the seismometer self-noise (Sleeman R. et al., 2006), and used Welch's modified periodogram to calculate different combinations of parameters during data processing (Welch P.D., 1967). The distributions of self-noise levels at different frequencies under different parameter combinations are analyzed. It is found that the distribution of the extreme values of the self-noise at different frequencies under different parameter combinations is not fixed, but the overall trend has a certain regularity, that is, the self-noise computation result decreases with the increase of K-line, and reaches a minimum when the window length is set to 98% and the overlap rate is set to 98%. The overall maximum difference is greater than 20dB, corresponding to more than one order of self-noise level difference. Therefore, it is necessary and reasonable to adopt the unified reasonable parameters for spectrum analysis and calculation, and only by this method, the different estimation results could be comparable. Their study also analyzes the causes of distortion due to unreasonable parameter combinations and gives a preferred interval of parameter combinations by the analysis of the bias and the variance of the spectrum.

High-quality seismometers must have both low self-noise and long-term stable performance. In order to be able to objectively and accurately evaluate these two properties of a seismometer, Tong Wanglian proposed establishing a complete and standardized seismometer self-noise test procedure, which was named by Tong Wanglian as the "Standard Evaluation". The key of his suggestion lies in the unification of test time, data processing and parameter settings. Upon power up, a seismometer usually needs some time to settle down for stable operation, this procedure for a broadband seismometer typically takes 5-7 days. Generally, continuous data recording of at least 10 days (240 hours) is required after the seismometer has run stabilized. The data is calculated in segments of 1 hour, and calculations are performed successively without the need of picking out any events or abnormal records. Uniform calculation software, consistent Welch spectrum estimation algorithms and parameters are used for noise processing. Tong suggests to establish the seismometer quality rating for the "Standard Evaluation" from the self-noise estimation value at 0.01Hz. Additional indicators at 800s and 8Hz frequencies are also suggested to characterize the ultra-long period and high-frequency performance respectively for a more complete and reasonable evaluation. The evaluation criteria includes self-noise total average, self-noise dispersion, four quality grading thresholds, grading compliance average, grading compliance rates, and two-sensor simultaneous compliance rates. Table 1 shows the "Standard Evaluation" results of three broadband seismometers tested at the Jiuliancheng seismic station. The results show that all the three seismometers reached the optimal grade A for the 120s broadband seismometer scoring standard for self-noise performance. However, after they are further analyzed for the total score, the self-noise total average value and the compliance rate, it showed that the STS2.5 seismometer's result was significantly better than the two BBVS-120 seismometers; in addition, between the two BBVS-120 seismometers, the one numbered G10497 was slightly better than the G10496; but at 8Hz, the self-noise levels of the two BBVS-120 seismometers are slightly lower than that of the STS2.5 seismometer.

Table 1 Seismometer test evaluation score sheet at Jiuliancheng seismic station (from Tong Wanglian)
6 COLLOCATION TESTS OF SEISMOMETER ALIGNMENT AND TRIAXIAL ORTHOGONALITY

The azimuth accuracy of the seismometer installed at the station is important for many seismological studies, and the seismometer triaxial orthogonality is another aspect related to the azimuth determination. Therefore, many research works have been done to promote the accuracy of instrument installation orientation (Holcomb L.G., 2002; Chiu H.C. and Huang H.C., 2003; Lü Yongqing et al., 2007; Ekström G. and Busby R.W., 2008; Niu Fenglin and Li Juan, 2011; Ringler A.T. et al., 2013; Xie Jianbo, 2014).

At present, a shake table test can only be used to test the seismometer orthogonality between two horizontal components, and it is impossible to test the orthogonality between sensors of the vertical and the two horizontal components. Tasič I. and Runovc F., (2013) and Xie Jianbo et al.(2018) separately proposed their own methods for testing the spatial orientation and triaxial orthogonality of seismometers by collocation method. However, the results of the collocation test are relative and have certain limitation (Xie Jianbo et al., 2018a).

Tasič I. and Runovc F. used the Euler rotation method to solve the relative azimuth between the target seismometer and the reference seismometer, and defined the axial deviation of the measured seismometer based on the reference seismometer axis. They deduced a set of equations for solving the Euler rotation angles and gave a set of Euler angle solutions (Tasič I. and Runovc F., 2013). Xie Jianbo et al. found that each equation deduced by Tasič I. and Runovc F. can have two possible solutions, but only one was used originally. By using the solution set given by Tasič and Runovc, the calculated azimuth relative to the reference seismometer is ambiguous in direction. Xie Jianbo et al. supplemented another set of possible solutions to the equations of Tasič I. and Runovc F. They suggested that the physical condition constrant, that the seismometer could not be mounted upside down, should be applied in the angle solving procedure. Therefore, an azimuth solution with the correct direction sign relative to the reference seismometer can be obtained. The supplemented solutions are as follows (Xie Jianbo et al., 2018a),

 $\psi = {\rm{atan}}2[({a_{21}}{a_{32}} - {a_{22}}{a_{31}}), ({a_{12}}{a_{31}} - {a_{11}}{a_{32}})]$ (2)
 $\theta = {\rm{atan}}2[{a_{31}}, - ({a_{11}}{s_\psi } + {a_{21}}{c_\psi })]$ (3)
 $\phi = {\rm{atan}}2[({a_{11}}{s_\psi }{c_\theta } + {a_{21}}{c_\psi }{c_\theta } - {a_{31}}{s_\theta }), ({a_{21}}{s_\psi } - {a_{11}}{c_\psi })]$ (4)

The definition of Tasič I. and Runovc F. is used in the formula.

The solution of Tasič I. and Runovc F. intuitively gives the orienting deviation of the sensing axes between the co-located instruments, but the relative space state between them is not clear. Xie Jianbo et al. defined a set of azimuth and coordinate tilt angles in a three-dimensional Cartesian coordinate system, which can clarify the spatial relative state between the co-located instruments and the orthogonal deviation of the axes can also be shown. Let the O-NEU represent the orthogonal reference geometric triaxial coordinate system, and O-YXZ represent the three-axis coordinate system of the testing seismometer. Y′, X′, and Z′ represent the projection of Y, X, and Z on the O-NE plane, the three pairs of azimuths and the coordinate tilt angles are shown in Fig. 5. Xie Jianbo et al. gave the relationship between the six angles and the matrix elements representing the transformation from O-NEU to O-YXZ coordinate system, which can be obtained by least square fitting of the collocation test data.

 Fig. 5 Azimuth and coordinate tilt angle definition

After using the complete solution with additional physical constraints, the results of Tasič and Runovc are consistent with the results of Xie Jianbo et al.(2018a). Xie Jianbo et al. also studied the influence of timing errors in recorded data and found that the angle βud can reach a minimum value when the timings of the co-located recorded data is consistent in time through back and forth sample shifting (Xie Jianbo et al, 2018b). Fig. 6 shows the timing error estimation results of the numerical experiment data without timing errors (N-βud) and the real collocation test data with possible timing errors (J-βud), which corresponds to the results of N-θ and J-θ in the context of Xie Jianbo et al. (2018a).

 Fig. 6 βud angle results of timing error estimation
7 RESEARCH ON SEISMIC DIGITIZER TIMING ACCURACY AND EMPIRICAL TRANSFER FUNCTIONS

In comparison to a seismometer, a seismic digitizer is a more ideal linear system. Correspondingly, there are many mature industry standards for performance evaluations and technical parameter testings of a seismic digitizer. Due to system determinacy, the amplitude and phase frequency response characteristics of a digitizer are determinate. With the deepening of research work, scientists have gradually discovered that some technique details of the seismic digitizer can affect the seismic data research or the evaluation of other performances of the seismograph. For example, the digitizer self-noise level could affect the self-noise estimation of the seismometers; the possible timing offset between different digitizers could affect the data processing result severely. SNL's test report (Kromer R.P., 2006a, 2006b) and methodological report (Merchant B.J. and Hart D.M., 2011) are relatively comprehensive in this area.

The test method for the timing accuracy of a seismic digitizer in the current national measurement specification for technical parameter and performance of a seismograph (China Earthquake Administration, 2017) is too simple to estimate the timing accuracy on the order of a millisecond or better. Referred to SNL's test report and methodical technical report (Kromer R.P., 2006a, 2006b; Merchant B.J. and Hart D.M., 2011), Xie Jianbo et al. used the record of timing pulses generated by a time server to analyze the timing accuracy of a seismic digitizer (Xie Jianbo et al., 2018b). Since the timing pulse train used for testing is a time-determined known signal, Xie Jianbo et al. further processed the spectra of the recorded data and the input signal to estimate the empirical transfer function of the seismic digitizer. They tested several types of seismic digitizers, and found that there could be about 1 sample point timing offset between digitizers widely used in the seismic network of China, which should be unacceptable for some studies with high time accuracy requirements.

Fig. 7 shows the test results of a single rising edge of two main types of digitizers in the seismic network of China using the timing pulse method (Xie Jianbo et al., 2018b). Since the test results are relative and different instrument manufacturers employed different time tag references for time corrections, it is not easy to say which one is more accurate. However, the difference in the time tag reference is reflected in the time consistency of the recorded data. Taking the two digitizers in Fig. 7 as an example, there is a timing offset about 6.6 milliseconds between them. The differences of the measured timing consistency of Xie Jianbo et al. (2018b), the time difference result in correlation analysis of Fig. 3, and the timing error test results of Fig. 6 all show that for the several types of digitizers used in the seismic network of China, there could be a time consistency difference of about 1 sample point between each other.

 Fig. 7 Single rising edge test results of timing pulse train for two types of digitizers

The timing difference caused by the different time tag reference selection is a system deviation, but it can still be corrected. On the other hand, the long-term stability of timing service is more important. Therefore, Xie Jianbo et al. also conducted long-term stability tests under the state with and without time synchronization separately (Xie Jianbo et al., 2018b). Their test results under time synchronization show that the timing accuracy stability and the clock error log record of the digitizer are completely consistent. In addition, when the timing synchronization signal is lost, the timing accuracy stability has a certain relationship with the environmental temperature variation, but there is no clear rule found.

The measured results of the empirical transfer functions of digitizers tested by Xie Jianbo et al. are consistent with the theoretical responses computed using the FIR filter coefficients and corresponding delay times provided by the equipment manufacturers (Xie Jianbo et al., 2018b). Moreover, the measured results also reflect some characteristics of the analog circuit before the AD conversion. For example, the measured digital output response amplitude is slightly lower than the theoretical response calculated from the filter coefficient when the frequency is higher than 10Hz. Xie Jianbo et al. also found that the delay time correction parameters of the actual output data of a digitizer are not only related to the FIR filter coefficients designed by the equipment manufacturers, but also related to the time tag reference points selected. The delay time correction parameters affect the actual transfer function of a digitizer, which are very important for a work that needs the correction of transfer function of the digitizer, and the accurate delay time parameters are required in this case.

8 CONCLUSIONS

We try to introduce the new works and achievements in the field of seismograph testing as objectively as possible. But the information that we consulted was not comprehensive enough and some excellent and important works were not involved, especially the environmental adaptability related tests and performance tests related to new instrument development.

ACKNOWLEDGEMENT

Some works described in this paper are the results of our team's project sponsored by the Department of Earthquake Monitoring and Prediction, China Earthquake Administration. We would like to thank the Department of Earthquake Monitoring and Prediction, China Earthquake Administration for supporting the project and all individuals who have given support to the team.

REFERENCES