2. Yunnan Earthquake Agency, Kunming 650224, China
The change of velocity of the underground medium can reflect the change of the physical state of the medium. Since the 1960s, geophysicists discovered that some earthquakes were accompanied by abnormal changes in seismic wave velocity (Savarensky E. F., 1968; Aggarwal Y. P. et al, 1973, 1975; Niu Fengli et al., 2008; Yang Runhai et al., 2011), which is one of the important means to study the evolution process of earthquakes. The airgun's source is less destructive and with higher temporal and spatial resolution, is suitable for 4D seismological studies with timevarying information (Wang Weitao et al., 2017).
The airgun source receives a large amount of energy through the instantaneous release of highpressure gas, and the repeatability is high (Chen Yong et al., 2016). In 2001, during the sealand joint test in the northern part of the South China Sea, the airgun source was first applied to the detection of deep crustal structures (Qiu Xuelin et al., 2003, 2007), which laid the foundation for the development of airgun sources in the field of domestic seismology. In 2006, the Shangguanhu Reservoir experiment in Zunhua, Hebei Province verified the feasibility of airgun source stimulation in terrestrial waters (Chen Yong et al., 2007, 2008). In 2009, the Mapaoquan airgun experiment in Fangshan, Beijing, once again proved that the airgun source can be excited in small land water bodies (Wang Baoshan et al., 2010). Subsequently, the China Earthquake Administration successively carried out airgun continuous excitation experiments with a number of units including the Yunnan Earthquake Agency (Wang Bin et al., 2015; Zhang Yuansheng et al., 2016; Wei Bin et al., 2016; Chen Huifang et al., 2016) and obtained fruitful research results. The airgun source has become an important tool for detecting shallow media structures and spacetime evolution of the continent (Wang Weitao et al., 2017).
The airgun wavelet is divided into a high frequency pressure pulse and a low frequency bubble pulse. The pressure pulse has large amplitude and high frequency, and can obtain high resolution in a small area scale. It is mainly used in seismic exploration fields such as the petroleum industry (Zhou Baohua et al., 1998a, 1998b). The bubble pulse frequency is low and the propagation distance is long. It is mainly applied to the crustal structure (Tian Xiaofeng et al., 2016; She Yuyang et al, 2018) and the media change related to earthquake gestation (Reasenberg P. et al., 1974; Liu Zifeng et al., 2015; Zhang Yuansheng et al. 2017). The scale of seismic signal velocity is usually very small (Wang Weitao et al., 2009; Yang Wei et al., 2010). It has extremely high requirements for the accuracy of the excitation event. Using the airgun signal to calculate the wave velocity variation of the medium is one of the more commonly used methods for studying medium changes. Since dv/v=dt/t(dv and dt are seismic wave velocity changes and travel time changes, respectively, v and t are the original path seismic wave velocity and travel time), the wave velocity change dv is often reflected indirectly by calculating the travel time change dt. The calculation of the wave velocity of the airgun signal is affected by factors such as signaltonoise ratio, calculation accuracy and source interference. After filtering and superimposition, the signaltonoise ratio and calculation accuracy can be improved to some extent, but the influence of the source cannot be eliminated. Therefore, exploring a calculation method to remove the influence of the source is very important for the calculation of the wave velocity variation.
In 2011, the Yunnan Earthquake Agency built China's first airgun signal fixed launch pad in the Dayindian Reservoir of Binchuan County, Dali Bai Autonomous Prefecture, Yunnan Province. So far, many achievements have been made, but many problems remain to be studied. In order to carry out research on the effects of the change of the excitation conditions of the Binchuan airgun signal launcher on the airgun signal and the method of eliminating the source function, the Yunnan Earthquake Agency conducted a dense excitation experiment in February 2017. In this paper, the experimental data is used to analyze the phase and correlation characteristics of airgun signals under different excitation conditions. The travel time variation of airgun signals is calculated by different methods, and comparative analysis is carried out.
1 EXPERIMENTAL OVERVIEW AND DATAThe Dayindian Reservoir is a mediumsized reservoir integrating domestic water and agricultural irrigation. Due to the influence of water consumption and seasonal changes, the water level of the reservoir changes, and the airgun signal and its correlation coefficient will change accordingly (Luan Yi et al., 2016). Airgun signal wave velocity changes are often mixed into source information such as water level changes. The reservoir airgun array consists of four Bolt 1500LL airguns with a single airgun capacity of 2000in^{3}. The square pontoon side of the fixed airgun array is 7m long. The conventional excitation conditions of the reservoir are sinking depth h=10m and working pressure p=15MPa. In 2012, it entered the conventional excitation and continued to this day. In order to carry out research experiments in the early stage of mobile airgun promotion and overcome the shortcomings of low water levels and the inability to be excited, the base of the launching station has drilled a well with diameter of 5m and depth of 20m. The small Bolt airgun with a single capacity of 250in^{3} is used in the well. In order to cooperate with the airgun excitation experimental observation and research, 40 mobile observation stations were reestablished on the basis of the original observation station of the Yunnan Earthquake Agency. The mobile observatory seismometer is equipped with a 2s100Hz CMG40T seismometer manufactured by GURALP from the United Kingdom and a REFTEK130B data collector produced by the US REFTEK company (Wang Baoshan et al., 2012). The commonly used reference station (CKT0) is placed in a swing room with a base distance of about 50m from the airgun array, and the signal is recorded as a source function. The sampling rate of CKT0 is 200Hz and 100Hz, and the sampling rate of other flow observing stations is 100Hz.
During the intensive excitation experiment in February, 2017, the water level of the Dayindian Reservoir was about 19m and that of the well was 15m. The QS05 portable digital seismograph and the TAG33M forcebalanced strong seismometer were installed in the experiment. The portable seismometer sampling rate was 500Hz, and the strong seismic instrument sampling rate was 200Hz. Before the experiment, we numbered all the instruments and performed the consistency test. Except for the abnormality of the Z(vertical) component signal of the instrument of sta37, the consistency of other instruments was good. The E (eastwest) component and the N (southnorth) component of the sta37 instrument correspond well with other instrument signals. The Z component signal is opposite in phase and negative phase to other instruments, so it is necessary to multiply the Zcomponent signal when processing the instrument data of sta37. Phase correction is performed with 1. The reference station in the experiment is the sta06 instrument placed on the bank of the reservoir. Sta06 is about 50m away from the airgun array and about 40m away from the well. In the early stages, portable digital earthquakes were arranged around the airgun source with a radius of 0.5km and 1.7km. The instrument (Fig. 1), in which the GPS was destroyed shortly after the installation of the sta23 instrument, no data was recorded. In the early stages of the experiment, the "control variable method" was used, that is, the reservoir and well excitation were carried out under the working pressures p=9MPa, 12MPa and 15MPa at the depths of sinking h=8m, 10m and 12m, respectively. During the reservoir excitation process, the floating stage was shifted 7m southward and 7m northward at h=10m and p=15MPa respectively. The initial excitation conditions are shown in Table 1. In the later stage, the intensive excitation under normal excitation conditions was mainly carried out, and a line was laid along the northeast direction, and strong earthquake meters and portable seismographs were set up for recording and observation.
In comparison to the single airgun in the well, the airgun array has the characteristics of high energy, low main frequency and long propagation distance. This paper mainly uses the reservoir excitation signal to analyze the correlation and travel time, and uses the preexcitation experimental data recorded by the sta06 instrument (68 guns) to analyze the characteristics of excitation signals in reservoirs and wells under different excitation conditions. Due to the large error of some QS05 portable digital seismograph GPS timing system, this paper uses the data recorded by the original mobile observation network to calculate and analyze the travel time variation.
2 CHARACTERISTICS AND CORRELATION ANALYSIS OF AIRGUN SIGNALS UNDER DIFFERENT EXCITATION CONDITIONS 2.1 Phase Characteristics of Airgun SignalsDue to the mechanical delay of the airgun console and other factors, the recorded excitation log is different from the actual firing time of the airgun. The recorded data and the excitation log are combined with the recorded data and excitation log from the reference station sta06, the Z components of the first reservoir and well excitation signals recorded by the sta06 are respectively taken as the template to obtain an accurate excitation time otime with the crosscorrelation method (Zhang Yunpeng et al., 2017) on the basis of the otime excitation event waveform. In order to analyze the characteristics of the "airgun wavelet" under different excitation conditions, the signals under the same excitation conditions recorded by the sta06 are linearly superimposed, and the "airgun wavelet" corresponding to the excitation condition is represented by the superimposed signal, and under various excitation conditions. The signal is initially aligned and analyzed.
Taking the signal under the normal excitation condition (p=15MPa, h=10m) as an example, the frequency of the signal recorded in the reservoir and well recorded by sta06 is shown in Fig. 2, where: (a) is the reservoir excitation signal and its frequency characteristics, (b) is the excitation signal and its frequency characteristics in the well. The hydrophone installed near the air gun array can clearly record the airgun wavelet (Xia Ji, 2017). In the experiment, the reference station sta06 is located on the bedrock, and the epicenter distance is much larger than the bubble radius. Therefore, the sta06 record signal and near the airgun wavelet recorded by the field hydrophone is more complicated than the phase. The "airgun wavelet" recorded by sta06 includes highfrequency pressure pulses (3070Hz) and lowfrequency (26Hz) bubble pulses, and also includes some signal components between the two frequencies (830Hz). Hu Jiupeng et al. (2017) believe that this part of the signal may be formed by repeated reflections of wavelets caused by irregular water shapes. The excitation signal in the well is limited by the small energy of the single airgun and the insufficient bubble oscillation environment in the well. There is no obvious pressure pulse and bubble pulse phase like the reservoir excitation wavelet, the duration is shorter, the overall frequency is higher, and the frequency is lower focusing on 1015Hz.
Fig. 3 is the amplitude normalized signal excited by the same working depth h and the different working pressure p recorded by sta06. Combined with the pressure pulse and bubble pulse frequency and phase information of the signals in Fig. 2 and Fig. 3, when the same h, the change of p has little influence on the pressure pulse phase of the reservoir excitation signal, and has a great influence on the bubble pulse phase, and the bubble pulse oscillation period as the increase in p increases significantly. The excitation signal in the well has no obvious bubble pulse, but the first three positive phases with larger amplitude of the signal under the same h are less affected by p, and the latter part of the signal is sensitive to the change of p, showing the same as the reservoir excitation signal.
Fig. 4 shows the airgun signals excited by the same p and different h recorded in sta06. It can be seen from Fig. 4 that the change of h not only affects the bubble pulse of the reservoir excitation signal, but also the slight change of the pressure pulse and its subsequent wave train and amplitude. The second half of the excitation signal in the well is greatly affected by h. The first three positive phases with larger amplitude also change with h.
Fig. 5 shows the signals before and after the displacement of the airgun array recorded by sta06. When the airgun array generates displacement, the signal changes slightly, mainly in the amplitude of the pressure pulse and the phase of the subsequent wave train. The phase of the bubble pulse does not change much. The effect of displacement changes on the phase of the signal has a much smaller effect than the change in p and h.
The signaltonoise ratio of the excitation signal in the same epicenter is higher than that in the well. The correlation of the Zcomponent of the reservoir excitation signal with less time error and the lack of timeshifting is determined by 8 instruments such as sta12.
The crosscorrelation function ρ_{xy} of the two energy finite signals x(t) and y(t) is defined as:
$ \rho_{x y}(\tau)=\int\limits_{\infty}^{\infty} x(t) y^{*}(t\tau) d t $  (1) 
τ is the time delay, and y^{*}(tτ) is the complex conjugate of y(tτ). The selected 68 gun original signals of each station are stacked as the reference signal of the station, and the single gun signal is crosscorrelated with the reference signal to obtain the maximum cross correlation value. At the same time, other stations' signal perform a frequency domain water level deconvolution operation with sta06's signal (Helmberger D. et al, 1971; Zhai Qiushi et al., 2016; Wang Baoshan et al., 2018), get the approximate Green's function, and carry out the corresponding frequency band filtering to improve the signaltonoise ratio after superposition and crosscorrelation operations.
Fig. 6 shows the crosscorrelation results before and after deconvolution of different station signals. The excitation conditions and corresponding serial numbers are shown in Table 1. Among them, the black solid circle is the crosscorrelation value of the original signal under different excitation conditions, and the gray fivepointed star is the crosscorrelation value of the post deconvolution signal. It can be seen from Fig. 6 that the crosscorrelation values of the signals of different excitation conditions of the same station are quite different, and the correlation values of the signals are close under the same excitation conditions, and the change of the excitation conditions can be expressed by the correlation coefficient of the signals. The correlation coefficients of different station signals are not the same. The correlation coefficients of some station signals are very sensitive to the excitation conditions. For example, the signals of sta13, sta28 and sta30 can almost reflect the changes of all excitation conditions. The correlation coefficients of some station signals of different excitation conditions are not much different. For example, the 2231 airgun pressure p is increased from 9MPa to 12MPa, but the correlation coefficients of the sta12, sta15 and sta22 signals are less different. According to the position of the station, it may be related to the media below the station and the station's azimuth. The overall correlation of the signal after deconvolution is improved, and the signal difference caused by the excitation condition is significantly reduced. The correlation coefficient of the signal of different excitation conditions before deconvolution fluctuates greatly, and the correlation coefficient of the signal after deconvolution is flat as a whole.
The correlation coefficient of the deconvolution signal of the sta12 is lower than that of the original signal, but the correlation coefficient of the whole signal after deconvolution is reduced. The correlation coefficient of other stations' deconvolution signals is significantly higher than that of the original signal, and the sta13 deconvolution signal is almost indistinguishable from the original signal. Deconvolution enhances the similarity of airgun signals and reduces the effects of excitation conditions on the signal.
3 USING DECONVOLUTION AND PAIR DIFFERENCE METHOD TO CALCULATE THE CHANGE OF THE MEDIUM TRAVEL TIMEDeconvolution can eliminate the influence of the source to a certain extent and improve the similarity of the airgun signals. Wang Baoshan et al. (2018) proposed that the use of deconvolution signals for the travel time variation can reflect the change of the medium and calculate the change of the travel time of the airgun signals on the basis of the interference method of crosscorrelation delay detection. However, the deconvolution operation process has low stability and reduces the signaltonoise ratio of the signal. Huang Yilei et al. (2017) proposed the pair difference method to calculate the travel time variation without losing the signaltonoise ratio of the signal. In this paper, the deconvolution and pair difference method are used to calculate the travel time variation of the same airgun signal, and the calculation results of the two methods are briefly compared and analyzed.
3.1 Crosscorrelation Delay DetectionFor the seismic signals x(t) and y(t), if the wave forms are highly similar and there is only a delay τ in time, the delay τ can be obtained by calculating the correlation coefficient function of the two signals. The calculation method of the delay correlation coefficient is:
$ {R_{xy}}(\tau) = \frac{{\int\limits_{t  \frac{T}{2}}^{t + \frac{T}{2}} x (t)y(t  \tau){\rm{d}}t}}{{\sqrt {\int\limits_{t  \frac{T}{2}}^{t + \frac{T}{2}} {{x^2}} (t){\rm{d}}t\int\limits_{t  \frac{T}{2}}^{t + \frac{T}{2}} {{y^2}} (t){\rm{d}}t} }} $  (2) 
Where T is the length of the selected time window, and τ and t are the instantaneous times of the signal. The maximum value of the correlation coefficient R_{xy}(τ_{s}) corresponds to the time τ_{s} which is the time delay of the two signals.
3.2 Deconvolution Calculation of Travel Time ChangesThe seismic signal x(t) recorded by the station is usually composed of the source function s(t), the instrument response function i(t), the Green's function g(t), and the noise function n(t):
$ x(t)=s(t) * g(t) * i(t)+n(t) $  (3) 
"*" indicates convolution. In order to get the media time change, we need to remove s(t), i(t) and n(t) to get the Green's function g(t). In the actual processing, the noise function n(t) is difficult to completely eliminate, so it is ignored first. x(t) is removed from the instrument response function i(t) to obtain u(t), and then deconvolutes s(t) to obtain g(t).
The deconvolution process in the time domain is cumbersome. In this paper, the frequency domain level deconvolution method proposed by Helmberger D. et al., (1971) is used to transform the time domain signal by Fourier transform to obtain the seismic signal U(ω), the source function S(ω) and Green's function G(ω) in the frequency domain. According to the value of S(ω), selecting an appropriate level threshold α, increasing S(ω) smaller than the threshold to α, and then performing simple division to obtain G(ω). Then, G(ω) is inversely transformed by Fourier to obtain the approximate Green's function g(t) of the time domain. The calculation method is:
$ G(\omega)=\frac{U(\omega) S^{*}(\omega)}{max \left\{S(\omega) S^{*}(\omega), \alpha max \left\{S(\omega) S^{*}(\omega)\right\}\right\}} $  (4) 
S^{*}(ω) is a complex conjugate of S(ω). Then using the interferometric method based on crosscorrelation delay detection (Wang Baoshan et al., 2008; Liu Zifeng et al., 2015), the signals to be tested are correlated with the template signals to obtain the time delay of each signal and the travel time variation trend of all signals.
3.3 Pair Difference Method to Calculate Travel Time ChangesThe pair difference method is proposed to maximize the signaltonoise ratio of the signals. Huang Yilei et al. (2017) believe that the source function s(t) and the seismic signal x(t) contain the same source change information. The interference method is used to directly calculate the travel time changes of x(t) and s(t), respectively, and then use the travel time variation dt_{x} of x(t) minus the travel time variation dt_{s} of s(t) to remove the influence of the source and obtain the transit time change dt of the medium. The process is as follows:
$ \mathrm{d} t=\mathrm{d} t_{x}\mathrm{d} t_{s} $  (5) 
Due to the GPS timing error of some instruments in the intensive excitation experiment, in order to eliminate the interference of the instrument itself, we use the signal of the Binchuan mobile observing network to calculate the travel time variation. Zhang Yunpeng et al. (2017) analyzed the GPS errors of several reference stations at the transmitting base, all of which are within dozens of μs, far smaller than the dozens of ms of the travel time change. There is no obvious time shift in the data of the selected mobile stations. It is considered that the GPS of the selected stations is accurate. It can be seen from the foregoing that the pressure pulse and its subsequent wave train are less affected by the excitation conditions, and the bandpass filtering of the Zcomponent signal of the 200 Hz channel recording at the CKT0 is performed at 870Hz, with p=15MPa, h=10m, and with no displacement, the signal pressure pulse and its subsequent wave train are excited to obtain the excitation time. At the same time, station 53268 with less interference (the epicenter distance is 6km, as shown in Fig. 1) is selected to calculate the travel time change. In the calculation of the travel time variation of the signals at the station 53268, the timefrequency analysis of the signals recorded at the station 53268 is first performed and filtered according to the frequency range of the airgun signals. The mean and linear drift of the CKT0 signal are removed, and then the CKT0 signals are deconvolved using the processed 53268 signals. The time delay detection of the crosscorrelation is used to calculate the travel time changes of the 53268 signals and CKT0 signals before and after deconvolution. The pair difference travel time variation is obtained by the subtractions of the travel time changes of the CKT0 signals from the travel time changes of the 53268 signals before the deconvolution.
Fig. 7 shows the 53268 signals before and after the deconvolution. The first three signals are the templates by stacking all the signals before the 53268 deconvolution. The last three signals are the templates of all the Green's function after the deconvolution (dec). According to the epicenter distance and the velocity structure of the Binchuan area (Chen Siwen et al., 2016), the red line window is Pwave, and the blue line window is Swave.
In this paper, the travel time variation of the Pwave and Swave before and after the deconvolution of the signals at the 53268 is calculated. At the same time, according to the timefrequency characteristics of "airgun wavelet" in Fig. 2, the CKT0 signals are divided into pressure pulse and bubble pulse, and their travel time changes are calculated separately, as shown in Fig. 8 as the experimental excitation condition change and the travel time variation result. The Pwave and the Swave dt trend of the signals before the deconvolution are consistent. The working pressure p decreases, dt decreases (dv increases), and vice versa. dt is positively correlated with p change. The sinking depth h becomes shallower, dt increases (dv decreases), and vice versa. dt is negatively correlated with h change.When the displacement s changes, dt fluctuates, but the change trend is not obvious enough for p and h. It may be because p and h after s change are consistent with p and h of the template signal, and the signal changes are small. When the excitation moment is obtained by the correlation, the travel time difference caused by s is greatly reduced. The change trend of the bubble pulse dt is consistent with the trend of the dt change of Pwave and Swave before the deconvolution at the 53268. The change trend of the pressure pulse Ecomponent dt is consistent with the bubble pulse change, but the value is significantly smaller than the bubble pulse dt. Compared with the Ecomponent, the N, Zcomponents of the pressure pulse have stable travel time and there is no obvious change as in the Ecomponent of the pressure pulse.
The results of the pair difference method show that the travel time difference between the body wave and the bubble pulse is greater than the travel time difference between the body wave and the pressure pulse. Except for the large travel time variation fluctuation between the Ecomponent of the Pwave and the bubble pulse, the other pair difference method results all show the change in the excitation conditions. The pair difference method does not eliminate the influence of the excitation condition. The results of the deconvolution travel time variation show that the Ecomponent dt is related to the change of excitation condition. The Ncomponent of dt is affected by the excitation condition except for some periods (about 60 shots), but most other periods are independent of the change of the excitation condition. The Zcomponent of dt is irrelevant to the excitation conditions. The pair difference method cannot eliminate the influence of the excitation condition, and the deconvolution cannot completely eliminate the excitation condition either. But the deconvolution effect is obviously better than the pair difference method, and the Zcomponent dt cannot see the influence of the excitation condition.
4 DISCUSSION AND CONCLUSION 4.1 DiscussionIn this paper, the characteristics of the "airgun wavelet" waveform under different excitation conditions are analyzed by using the Binchuan airgun source intensive excitation experimental data. The results show that the effect of the depth of sinking on the phase of the pressure pulse and its subsequent wave train is greater than the influence of the working pressure on it. Bubble pulses are sensitive to changes in sinking depth and working pressure. The change of "airgun wavelet" is related to the difference of the bubble pressure, bubble radius and water body pressure excited in different sinking depths and under different working pressures (Xia Ji, 2017). At the same time, according to the signal characteristics of different sinking depths, the phase change may also be related to the difference in the boundary reflection distance of the reservoir bottom and the water surface. When the airgun array is displaced, the overall signal change is small. After the amplitude is normalized, the phase and amplitude of the signals moving southward are basically the same with those of the predisplacement signal. The signals moving north, compared with the predisplacement signals, have subtle changes in the amplitudes of the pressure pulse and its subsequent wave train, but the other amplitudes and phases are basically the same. The depths of different positions in the bottom of the Dayindian Reservoir are different (Wang Bin et al., 2015, 2016). Therefore, the signal difference before and after the displacement may be related to the difference between the distance between the airgun arrays and the water bottoms in different positions and the distance between of the airgun arrays and the CKT0 station.
The influence of the deconvolution operation on the correlation of airgun signals is analyzed. The results show that deconvolution can reduce the signal difference caused by the excitation conditions and further improve the correlation of airgun signals. In this paper, wave filtering is done before and after the deconvolution process. In order to investigate whether the correlation of the signal after the deconvolution is caused by filtering, we select the signals of the sta12, sta13 and sta15 in the above analysis, filter after the spectrum respectively. We calculate the correlation of the signals before and after the filtering and after the deconvolution as shown in Fig. 9. The results show that the correlation between the sta13 and sta15 signals after the filtering is improved, but much lower than the correlation coefficient of the signals after the deconvolution. Except for the 55th gun signal the correlation of the sta12 signals after the filtering is improved, but the other signals are not much improved compared with the original signals, and most of the signal correlation coefficients are smaller than the correlation of the deconvolved signal. Therefore, deconvolution can improve the similarity of the signal, and this improvement is not completely caused by the noise reduction process such as filtering before deconvolution.
Based on crosscorrelation delay detection and interferometry, we discuss whether the difference between the pair difference method and the deconvolution method can eliminate the interference of the excitation condition in the calculation of travel time change. The results show that the difference between the pair difference method and the deconvolution cannot completely eliminate the excitation condition influence, but the effect of excitation conditions after deconvolution is significantly reduced. We perform multivariate regression analysis of all travel time changes dt and excitation conditions as shown in Fig. 10. In the figure, "Ppr", "Pbu" and "Spr" respectively represent Pwave and pressure pulse, Pwave and bubble pulse and the difference dt between the Swave and the pressure pulse. In the paper, the difference between the Pwave dt and the bubble pulse dt of the 53268 signal component E and the excitation condition change trend (Fig. 10(a)) is large and the multivariate regression results have the lowest correlation. The correlation coefficient between other pair difference method result and the regression results of the excitation condition is high, and the regression curve has the same shape. The overall displacement has the least influence on dt. We average the results of the multivariate fitting of the pair difference method other than the result of "Pbu"(FIg. 10(a)), and obtain the multiple linear regression equations of the travel time variation dt and the sinking depth h, the working pressure p and the displacement s:
$ \mathrm{d} t=(2.23.1 h+2.5 p+0.24 s) \times 10^{3} $  (6) 
The fitting results show that dt is inversely proportional to h, and is proportional to p and s, and the influence of h is greater than the influence of p, and the influence of s is the smallest, which is consistent with the phase information of the "airgun wavelet" recorded by the sta06. The relationship between the travel time variation and the excitation conditions obtained in this paper is different from the conclusions obtained by Zhou Qingyun and Chen Junlei (2018), which mainly reflects in the influence of h on the travel time. By comparison, we found that Zhou Qingyun and Chen Junlei (2018) performed bandpass filtering on the CKT0 signal at the frequency of the bubble pulse before intercepting the data, that is, using the bubble pulse to obtain the excitation time. Therefore, it can be inferred that the selection of the template signal during the data interception also has a certain influence on the change of travel time. Besides the higher correlation between the dt of the Ecomponent and the excitation condition Fig. 10(d), Fig. 10(h), the deconvolution calculation results also show that the dt of the Ncomponent is greatly affected by s, which may be related to the northsouth movement of the airgun array, and others have a low correlation with the excitation conditions. The Zcomponent dt is completely unrelated to the excitation conditions, but the overall fluctuation is large, which may be related to the fact that the excitation conditions change too radically and the airgun signal penetration depth of 53268 is shallow and the atmospheric pressure is relatively large. In reality, h and p are relatively fixed or slow, and the calculation results of the predecessors indicate that such large fluctuations will not occur in the short term (Liu Zifeng et al., 2015). The 4756 gun signal has a relatively stable change in travel time. The main reason is that these airgun signals have the deepest sinking depth and the highest working pressure, so the airgun array has the highest energy (Xia Ji, 2017) and the highest signaltonoise ratio. The results of multivariate regression further indicate that the pair difference method cannot remove the travel time change.
The deconvolution cannot be completely removed, but the deconvolution can improve the correlation of the signals and greatly reduce the excitation condition interference, especially in the vertical components.
4.2 ConclusionIn this paper, the phase and correlation characteristics of airgun signals under different excitation conditions in Binchuan are analyzed. The effects of the pair difference between the excitation and the deconvolution method are compared. The following conclusions are drawn:
(1) The pressure pulse is less affected by the excitation condition and can be applied to obtain the excitation moment of the event.
(2) The influence of the sinking depth on the phase and travel time of the airgun signals is slightly greater than the change of the working air pressure. The influence of the displacement change can be eliminated by the cross correlation or deconvolution in the vertical signals, but it is difficult to completely eliminate in the horizontal signals. The airgun array should have the consistent excitation condition as much as possible during the excitation.
(3) The change of excitation condition can be expressed by the correlation coefficient of the signals. Deconvolution significantly reduces the signal difference caused by the excitation condition and improves the correlation of the signals.
(4) When the pair difference method is used to calculate the travel time change, the influence of the excitation condition in the travel time variation cannot be eliminated. Although the deconvolution method cannot completely remove the influence of the excitation condition, the travel time variation after the deconvolution is significantly reduced by the excitation condition, especially in the vertical components.
The calculation of travel time variation has important reference value for the spacetime evolution of medium. In actual conditions, the excitation conditions are relatively fixed or slow to change. When the airgun is excited, the consistency of the excitation conditions of the source should be kept to the maximum to ensure the calculation accuracy of the timevarying changes. Based on the good feasibility of deconvolution to eliminate the influence of excitation conditions, the travel time variation of the vertical component Green's function can be used as the reference object of medium change.
ACKNOWLEDGEMENTThe authors would like to thank Professor Wang Baoshan from the University of Science and Technology of China for the crosscorrelation delay detection program and the data interception program provided by the researcher Wang Weitao from the Institute of Geophysics of the China Earthquake Administration. Thanks to the data provided by the "Active Source Innovation Team" of the Yunnan Earthquake Agency, and to the reviewer for the revised comments.
Aggarwal Y.P., Sykes L.R., Armbruster J., Sbar M.L. Premonitory changes in seismic velocities and prediction of earthquakes[J]. Nature, 1973, 241(5385): 101104. DOI:10.1038/241101a0 
Aggarwal Y.P., Sykes L.R., Simpson D.W., Richards P.G. Spatial and temporal variations in t_{S}/t_{P} and in P wave residuals at Blue Mountain Lake, New York:application to earthquake prediction[J]. Journal of Geophysical Research, 1975, 80(5): 718732. DOI:10.1029/JB080i005p00718 
Chen Huifang, Lin Binhua, Jin Xing, Wu Lihua, Cai Huiteng. An experimental study on optimization of largevolume airgun source excitation conditions in a reservoir[J]. Earthquake Research in China, 2016, 30(3): 355363. 
Chen Siwen, Wang Baoshan, Tian Xiaofeng, Wang Fuyun, Liu Baofeng, Li Lu. Crustal structure from YunxianNinglang wideangle seismic reflection and refraction profile in northwestern Yunnan, China[J]. Seismology and Geology, 2016, 38(1): 91106 (in Chinese with English abstract). 
Chen Yong, Zhang Xiankang, Qiu Xuelin, Ge Hongkui, Liu Baojin, Wang Baoshan. A new way to generate seismic waves for continental crustal exploration[J]. Chinese Science Bulletin, 2007, 52(16): 22642268. DOI:10.1007/s1143400702474 
Chen Yong, Liu Lanbo, Ge Hongkui, Liu Baojun, Qiu Xuelin. Using an airgun array in a land reservoir as the seismic source for seismotectonic studies in northern China:experiments and preliminary results[J]. Geophysical Prospecting, 2008, 56(4): 601612. DOI:10.1111/j.13652478.2007.00679.x 
Chen Yong, Wang Baoshan. Preface to the special issue on advances in seismic exploration and monitoring with active sources in China[J]. Earthquake Research in China, 2016, 30(3): 281283. 
Helmberger D., Wiggins R.A. Upper mantle structure of midwestern United States[J]. Journal of Geophysical Research, 1971, 76(14): 32293245. DOI:10.1029/JB076i014p03229 
Hu Jiupeng, Wang Baoshan, Chen Yong. The influence of water shape on land airgun triggered signals[J]. Journal of Seismological Research, 2017, 40(4): 543549. 
Huang Yilei, Yun Naidan, Zhou Shiyong. A study on "pair difference" method to improve precision of signal travel time drift under low signalnoise ratio[J]. Journal of Seismological Research, 2017, 40(4): 595604 (in Chinese with English abstract). 
Liu Zifeng, Su Youjin, Wang Baoshan, Wang Bin, Yang Jun, Li Xiaobin. Study on analysis of travel time variation of seismic wave of active source in Binchuan[J]. Journal of Seismological Research, 2015, 38(4): 591597. 
Luan Yi, Yang Hongfeng, Wang Baoshan. Large volume airgun waveform data processing (1):Binchuan, Yunnan[J]. Earthquake Research in China, 2016, 32(2): 305318 (in Chinese with English abstract). 
Niu Fengli, Silver P.G., Daley T.M., Chen Xin, Majer E.L. Preseismic velocity changes observed from active source monitoring at the Parkfield SAFOD drill site[J]. Nature, 2008, 454(7201): 204208. DOI:10.1038/nature07111 
Qiu Xuelin, Zhao Minghui, Ye Chunming, Wang Tiankai, Wang Ping, Zhang Yixiang, Xia Kanyuan, Lee C.S. Ocean bottom seismometer and onshoreoffshore seismic experiment in northeastern south China Sea[J]. Geotectonica et Metallogenia, 2003, 27(4): 295300 (in Chinese with English abstract). 
Qiu Xuelin, Chen Yong, Zhu Rixiang, Xu Huilong, Shi Xiaobin, Ye Chunming, Zhao Minghui, Xia Shaohong. The application of large volume airgun sources to the onshoreoffshore seismic surveys:implication of the experimental results in northern South China Sea[J]. Chinese Science Bulletin, 2007, 52(4): 553560. DOI:10.1007/s1143400700511 
Reasenberg P., Aki K. A precise, continuous measurement of seismic velocity for monitoring in situ stress[J]. Journal of Geophysical Research, 1974, 79(2): 399406. DOI:10.1029/JB079i002p00399 
Savarensky E.F. On the prediction of earthquakes[J]. Tectonophysics, 1968, 6(1): 1727. DOI:10.1016/00401951(68)900231 
She Yuyang, Yao Huajian, Zhai Qiushi, Wang Fuyun, Tian Xiaofeng. Shallow crustal structure of the middlelower Yangtze River region in eastern China from surfacewave tomography of a large volume airgunshot experiment[J]. Seismological Research Letters, 2018, 89(3): 10031013. DOI:10.1785/0220170232 
Tian Xiaofeng, Wang Fuyun, Liu Baofeng, Yang Zhuoxin, Zheng Chenglong, Gao Zhanyong. Upper crustal velocity structure along the Yangtze River from Ma'anshan to Anqing[J]. Earthquake Research in China, 2016, 30(3): 430438. 
Wang Baoshan, Zhu Ping, Chen Yong, Niu Fenglin, Wang Bin. Continuous subsurface velocity measurement with coda wave interferometry[J]. Journal of Geophysical Research, 2008, 113(B12): B12313. DOI:10.1029/2007JB005023 
Wang Baoshan, Yang Wei, Yuan Songyong, Guo Shijun, Ge Hongkui, Xu Ping, Chen Yong. An experimental study on the excitation of large volume airguns in a small volume body of water[J]. Journal of Geophysics and Engineering, 2010, 7(4): 388394. DOI:10.1088/17422132/7/4/005 
Wang Baoshan, Ge Hongkui, Yang Wei, Wang Weitao, Wang Bin, Wu Guohua, Su Youjin. Transmitting seismic station monitors fault zone at depth[J]. EOS, Transactions American Geophysical Union, 2012, 93(5): 4950. 
Wang Baoshan, Tian Xiaofeng, Zhang Yunpeng, Li Yulan, Yang Wei, Zhang Bo, Wang Weitao, Yang Jun, Li Xiaobin. Seismic signature of an untuned largevolume airgun array fired in a water reservoir[J]. Seismological Research Letters, 2018, 89(3): 983991. DOI:10.1785/0220180007 
Wang Bin, Wu Guohua, Su Youjin, Wang Baoshan, Ge Hongkui, Jin Mingpei, Yang Jun, Zhang Qiming, Li Tao, Liu Jiwu. Site selection and construction process of Binchuan earthquake signal transmitting seismic station and its preliminary observation result[J]. Journal of Seismological Research, 2015, 38(1): 16 (in Chinese with English abstract). 
Wang Bin, Li Xiaobin, Liu Zifeng, Yang Jun, Yang Runhai, Wang Baoshan, Yang Wei. The source and observation system of Binchuan earthquake signal transmitting seismic station and its preliminary observation results[J]. Earthquake Research in China, 2016, 30(3): 316325. 
Wang Weitao, Wang Baoshan, Ge Hongkui, Chen Yong, Yuan Songyong, Yang Wei, Li Yijin. Using active source to monitor velocity variation in shallow sediment caused by the Wenchuan earthquake[J]. Earthquake Research in China, 2009, 25(3): 223233 (in Chinese with English abstract). 
Wang Weitao, Wang Baoshan, Jiang Shengmiao, Hu Jiupeng, Zhang Yuansheng. A perspective review of seismological investigation on the crust at regional scale using the active airgun source[J]. Journal of Seismological Research, 2017, 40(4): 514524, 677 (in Chinese with English abstract). 
Wei Bin, Su Jinbo, Wang Haitao, Zheng Liming, Wang Qiong, Zhang Wenlai, Yuan Shun, Wei Yunyun, Chen Hao. Site selection and construction of Hutubi airgun source signal transmitting seismic station and its characteristic of source[J]. Earthquake Research in China, 2016, 32(2): 222230 (in Chinese with English abstract). 
Xia Ji. Study on the Characteristics of Large Volume Airgun Source[D]. Harbin: Institute of Engineering Mechanics, China Earthquake Administration, 2017 (in Chinese with English abstract).

Yang Runhai, Wang Bin, Zheng Dingchang, Pang Weidong, Min Zhaoxu, Qin Jiazheng, Wu Guohua, Xu Yaji. Variation of wave velocity and strong earthquake preparation in Yunnan region[J]. Journal of Seismological Research, 2011, 34(4): 420427 (in Chinese with English abstract). 
Yang Wei, Ge Hongkui, Wang Baoshan, Yuan Songyong, Song Lili, Jia Yuhua, Li Yijin. Velocity changes observed by the precisely controlled active source for the Mianzhu M_{S}5.6 earthquake[J]. Chinese Journal of Geophysics, 2010, 53(5): 11491157 (in Chinese with English abstract). 
Zhai Qiushi, Wang Baoshan, Yao Huajian. Study on the deconvolution method and processing flow of airgun source data[J]. Earthquake Research in China, 2016, 30(3): 394404. 
Zhang Yuansheng, Guo Xiao, Qin Manzhong, Liu Xuzhou, Wei Congxin, Shen Xuzhang, Yan Wenhua, Zou Rui, Yin Liang, Wang Yahong, Sun Dianfeng, Feng Hongwu, Zhang Lifeng, Guo Yingxia. The construction of active source repeated monitoring in the Qilian Mountain, Gansu Province[J]. Earthquake Research in China, 2016, 30(3): 333341. 
Zhang Yuansheng, Wang Baoshan, Chen Yong, Wang Lanmin, Zou Rui, Qin Manzhong, Guo Xiao, Shen Xuzhang, Wei Congxin, Liu Xuzhou, Wang Yahong, Sun Dianfeng, Guo Yingxia, Yin Liang. Traveltime variations before and after two major earthquakes derived from activesource seismic data[J]. Chinese Journal of Geophysics, 2017, 60(10): 38153822 (in Chinese with English abstract). 
Zhang Yunpeng, Li Xiaobin, Wang Weitao, Wang Baoshan, Ye Beng, Yang Jun, Wang Bin. Data service system and data quality assessment for mobile observation of transmitting seismic stations in Binchuan, Yunnan[J]. Journal of Seismological Research, 2017, 40(4): 525533 (in Chinese with English abstract). 
Zhou Baohua, Liu Weibei. The development and application of airgun source[J]. Equipment for Geophysical Prospecting, 1998a, 8(1): 16, 50 (in Chinese with English abstract). 
Zhou Baohua, Liu Weibei. The development and application of airgun source (2)[J]. Equipment for Geophysical Prospecting, 1998b, 8(2): 15, 9 (in Chinese with English abstract). 
Zhou Qingyun, Chen Junlei. Influence of different triggering conditions of airgun source on travel time changes[J]. Journal of Seismological Research, 2018, 41(2): 264272 (in Chinese with English abstract). 