Sei sulla pagina 1di 11

GEOPHYSICS, VOL. 71, NO. 2 (MARCH-APRIL 2006); P. E13E23, 20 FIGS., 2 TABLES. 10.1190/1.

2187730

Annual Meeting Selection HFVS : Enhanced data quality through technology integration
TM

Christine E. Krohn1 and Marvin L. Johnson1


INTRODUCTION ABSTRACT
Known problems with vibroseis data include difcultto-pick and inaccurate rst-arrival times, poor well ties, correlation side lobes, harmonic ghosts, and coupling differences. Also, to reduce acquisition costs, various methods are used to record and then separate data from different source locations using vibrators sweeping simultaneously, but these methods suffer from poor data separation and harmonic contamination. A novel combination of heritage Mobil High Fidelity Vibratory Seismic (HFVS) and heritage Exxon vibrator technologies solves these problems with vibroseis data. The method involves vibrator separation combined with vibrator signature deconvolution in such a manner that the output is minimum phase and matches impulsive data. Vibrator signatures are calculated from the vibrator accelerometer measurements. The signatures from multiple vibrators and multiple sweeps are used to design a lter that optimally separates the data from each vibrator and replaces the signatures with a specially designed impulse response. Specic procedures are included to reduce the effects of inversion noise, which can distort the phase. HFVS recording can be used to increase production rates and reduce acquisition costs or to increase spatial sampling and improve data quality. Data recorded from multiple vibrators sweeping simultaneously can be separated by at least 60 dB. After separation, the data from each vibrator can be processed as unique source points. Compared with traditional vibrator arrays, individually processing each source point can result in better data quality by reducing intra-array effects and improving the mitigation of ground-roll noise. With the merger of Exxon and Mobil Corporations, the need arose to integrate the two companies vibrator technologies. Each technology solves different problems with conventional recording, so the integration is not as simple as using both together. Exxon processes for correlated data cannot be applied directly to Mobil inversion ltering processes. The integration of these technologies improves on both. In addition, the integration solves known problems with conventional vibrator processing, such as the match to dynamite data and phase issues with spiking deconvolution, are well documented in the literature (Gibson and Larner, 1984; Baeten and Ziolkowski, 1990). In this paper, we discuss the prior Exxon and Mobil vibrator technologies. Then we focus on a single vibrator to illustrate the inversion process, wavelet shaping, and the desired minimum-phase output. The nal sections discuss vibrator separation and the advantages of processing the data from each vibrator as unique source points.

Heritage Exxon shaped-sweep technology


Heritage Exxon technology (Trantham, 1994) incorporates a nonlinear shaped sweep (Andersen, 1994) with an amplitude spectrum that, after correlation, results in a simple wavelet with side-lobe energy reduced by at least 40 dB. These specially designed sweeps are loaded into the vibrator control system in the eld. Autocorrelations of a linear 696-Hz sweep and a 36-Hz shaped sweep are shown in Figures 1a and 1b; the amplitude spectra are shown in Figure 1e. The autocorrelation of a shaped sweep is a Ricker wavelet (Trantham, 1993a). The eld technique also includes phase rotation of the shaped sweep to suppress harmonics (Rietsch, 1981). Typically, four sweeps with phase rotations of 0 , 90 , 180 , and 270 are utilized, followed by correlation and stack.

Presented at the 73rd Annual International Meeting, Society of Exploration Geophysicists. Manuscript received by the Editor February 1, 2005; revised manuscript received June 17, 2005; published online March 6, 2006. 1 ExxonMobil Upstream Research Company, P.O. Box 2189, Houston, Texas 77252-2189. E-mail: chris.e.krohn@exxonmobil.com; marvin.l. johnson@exxonmobil.com. c 2006 Society of Exploration Geophysicists. All rights reserved. E13

E14

Krohn and Johnson

The second part of the technology is vibrator signature deconvolution (Trantham, 1993b), in which an operator is designed from the sweep and used to convert the data to minimum phase after correlation (Figures 1c and 1d). With the shaped sweep, a correction is made only for phase, minimizing problems with inversion noise; with the linear sweep, the amplitude is shaped as well. In addition, geophone data are rotated by 90 , which applies a phase correction corresponding to a time derivative. As seen in Figures 1c and 1d, the deconvolved autocorrelation wavelets are causal with no energy recorded before zero time. The time-derivative correction is well known; both theory and experiments show that geophone data should be in phase with the time derivative of the groundforce signals (Sallas, 1984; Baeten et al., 1987; Baeten and Ziolkowski, 1990). A reasonable estimate of the ground-force signal, used in vibrator-control electronics, is a mass weighted sum of accelerations measured on the baseplate and reaction mass. The use of shaped sweeps and vibrator signature deconvolution results in clean, approximately minimum-phase wavelets with large rst lobes. The data can be used to pick rstarrival times, which match those of dynamite data. In addition, the phase is optimized for subsequent processing. The use of statistical spiking deconvolution can result in phase errors if the input is not minimum phase. Correlated vibrator data are mixed phase because the correlation process, which is zero phase, is combined with the earth lter, which is minimum phase. The use of amplitude shaping in the eld followed by a phase correction by vibrator signature deconvolution results in approximately minimum-phase data. After statistical deconvolution and nal processing, the data are converted to zero phase for interpretation.

Heritage Mobil HFVS technology


Heritage Mobil technology (see Allen et al., 1998) is referred to as High Fidelity Vibratory Seismic (HFVSTM ), a licensed Exxon Mobil technology. Instead of correlating the data with the pilot signal, which is only an approximation of the signal imparted into the ground, the data are inverted using recordings made of the actual vibrator motion, such as the vibrator weighted-sum signal. These recordings include harmonics and vibrator nonlinearities, which are part of the recorded signal and are included as part of the signature in inversion. HFVS technology also includes the capability to separate optimally data from multiple vibrators operating simultaneously (Sallas et al., 1998). Each vibrator is assumed to radiate a signature into the ground, and each signature is convolved with a different earth reectivity sequence for the vibrator position. The earth sequence can include reectors, multiples, and near-surface effects. A trace d recorded at a geophone is a sum of the convolutions for all vibrators. The data trace di (t) recorded for sweep i as a function of time t is
N

di (t) =
j =1

sij (t) ej (t),

(1)

where sij (t) is sweep i from vibrator j, ej (t) is earth reectivity seen by vibrator j, and denotes the convolution operator. In the frequency domain, this becomes
N

di (f ) =
j =1

sij (f )ej (f ).

(2)

In the rest of this paper, all variables are in the frequency domain. In matrix form for M sweeps and N vibrators, equation 2 is

s21 s31 s41 . . . sM1


or

s11

s12 s22 s32 s42 . . . sM2

. . .

s1N

s2N e1 s3N e2 s4N . . . . . e . N sMN

d1 d2 d3 d4 . . . dM

(3)

Se = d.

(4)

If the number of sweeps is equal to the number of vibrators, this system of simultaneous equations can be solved for e:

e = Fd,
where Figure 1. Autocorrelation of (a) a linear sweep and (b) a shaped sweep as well as a vibrator signature deconvolution of (c) linear and (d) shaped sweeps. The side lobes for the shaped sweep are reduced much more than for the linear sweep. After deconvolution, the autocorrelation wavelets are minimum phase and casual. (e) Deconvolution includes shaping the amplitude spectrum for the linear sweep but not for the shaped sweep.

(5)

F = (S)1 .

(6)

The term F is the lter or operator that, when applied to the data, separates it into individual vibrator records. If there are more sweeps than vibrators, then the solution is determined by least squares. The normal equations are

S Se = S d,

(7)

HFVS: Enhanced Data Quality

E15

where S is the conjugate transpose of S. Solving for the earth responses gives

e = (S S)1 S d.

(8)

A separation-and-inversion lter can be designed from all of the vibrator signatures. The lter is of the form

F = (S S)1 S .

(9)

The lter is applied then to the data, resulting in separate records for each vibrator position. To prevent the matrix (S S) from being singular, HFVS recording includes phase encoding for each sweep. For example, it is sufcient to have one vibrator sweeping 90 out of phase of the other vibrators for each sweep (Table 1). After separation and inversion, subsequent processing steps include minimum-phase bandpass ltering followed by statistical spiking deconvolution. The spiking deconvolution is used to estimate any differences between the true vibrator signature and the measured vibrator motions. In addition, a model trace can be processed with the data so that phase errors can be corrected.

earth response, and geophone response are minimum-phase functions. A problem with inversion is that all frequencies are not represented in the sweep. An inversion operator is shown in Figure 2a for an 8128-Hz linear sweep. The amplitudes at low and high frequencies outside the sweep band are close to zero; thus, inversion applies a large gain at these frequencies. The application of this inversion operator to vertical seismic prole (VSP) data is shown in Figures 2b and 2c. The inverted data in the time domain (Figure 2b) are obscured by low-frequency noise, and the spectrum (Figure 2c) shows a 30-dB peak at the lowest frequencies. Conventionally, prewhitening is performed by adding noise to the sweep to stabilize the inversion and prevent large gains. Adding 3% noise decreases the gain, but as shown in Figure 2a, an amplitude gain still exists at the lower frequencies, so bandpass ltering is needed. The process of adding noise or prewhitening can distort the phase as can be referred from equation 11 and illustrated with modeling results shown in Figure 3. Equation 11 becomes

1 s(f ) + noise

d(f ) =

[(i2f )s(f )] s(f ) + noise

e(f )g(f ).
(12)

TECHNOLOGY INTEGRATION
The goal of integrating these technologies is to obtain true minimum-phase data with good S/N quality after separation and inversion without the need for a model trace. Our unpublished tests show that while shaped sweeps can be used with HFVS, a different strategy is needed to integrate the technologies fully. Heritage Exxon techniques designed for correlated data cannot be applied directly to inversion.

Inversion
The process of HFVS inversion can be understood from the following equations. We assume the recorded data for a single vibrator in the frequency domain d(f ) are given by

d(f ) = [(i2f )s(f )]e(f )g(f ),

(10)

where s( f ) is the ground-force signal, g( f ) is the geophone response, and e( f ) is the earth response. The factor (i2f ) represents the time derivative. If we apply a lter equal to 1/s(f ), we get

1 s(f )

d(f ) =

[(i2f )s(f )] s(f )

e(f )g(f ).

(11)

Effectively, the lter removes the sweep function, and the remainder should be minimum phase because the derivative, Table 1. Standard HFVS phase rotation for four vibrators and four sweeps. Sweep 1 2 3 4 Vibrator 1 ( ) 0 0 0 90 Vibrator 2 ( ) 0 0 90 0 Vibrator 3( ) 0 90 0 0 Vibrator 4( ) 90 0 0 0 Figure 2. (a) HFVS inversion operator for a linear 8128-Hz linear sweep and (b, c) its application to VSP data. The operator applies a large gain to any noise outside of the sweep band. Both the (b) time domain data and (c) frequency spectrum are dominated by amplied low-frequency noise. The operator gain can be reduced by adding noise to the sweep before inversion (a).

E16

Krohn and Johnson

The process of adding noise makes an amplitude correction to the data without the corresponding minimum-phase correction. If the noise is large, the phase can be distorted. For HFVS data, 3%5% noise may be required, even when followed by multiple bandpass lters. Modeling results for a single-vibrator inversion are shown in Figure 3. The model includes attenuation and a geophone response. The model response (Figure 3a) is minimum phase and casual. The model is convolved with the derivative of a measured ground force signal and inverted. Inversion with the ground force (Figure 3b), as in the original HFVS method, results in a wavelet that is close to minimum phase; however, the wavelet is ringy, even for noise-free model data. Inverting the data by the ground force and doing a 90 phase rotation as in the shaped-sweep method (Figure 3c) does not yield the correct wavelet because the corresponding minimum-phase amplitude correction is not made. Figure 3d shows the data inverted by the derivative of the ground force. Now the pulse matches that of the original. However, the addition of prewhitening noise results in a small precursor. The data are not rigorously minimum phase because the wavelet is not causal. Inversion with a vibrator signature that most closely matches the signal put into the ground, such as the time derivative of the weighted sum, results in cleaner pulses and requires less prewhitening. If the added noise is large, however, phase distortion is observed. Bandpass ltering also is required to reduce noise amplied by the inversion. The problem with added noise in vibrator signature deconvolution is discussed in Bickel (1982) and in Gibson and Larner (1984).

where w(f ) is the impulse response or wavelet designed from the amplitude spectrum corresponding to a characteristic vibrator signature. The amplitude of the impulse response is made to be less than or equal to that of the vibrator signature so that the resulting operator does not exceed 1.0 or 0 dB (Figure 4). The need for prewhitening is signicantly reduced. The separation and inversion process now becomes a separation and deconvolution process, and equation 9 becomes

F = (S S)1 S w.

(14)

An impulse response is designed from the ground force or its derivative. Conventional linear sweeps can be used. The resulting wavelets are simple with large front lobes and can be optimal in terms of bandwidth and pulse length (Trantham, 1993b). Alternatively, the impulse response can be designed to generate broadband data, matching that achieved through spiking deconvolution. Tests show that deconvolution with the derivative of the ground force removes the sweep from the data and replaces it with the desired impulse response. Deconvolution with the ground force without the derivative removes the sweep and replaces it with the derivative of the impulse response, as can be inferred from equation 13.

Wavelet shaping
The integrated, improved solution replaces the inversion with deterministic vibrator signature deconvolution using a specially designed minimum-phase impulse response. Equation 11 becomes

w(f ) d(f ) = [(i2f )s(f )]

[(i2f )s(f )] w(f )e(f )g(f ), [(i2f )s(f )]


(13)

Figure 3. Application of HFVS operators to model data showing (a) the original model, (b) inversion with the ground force as with the original HFVS method, (c) phase rotation of 90 as in the shape-sweep method followed by inversion with the ground force, and (d) inversion with the derivative of the ground force. Panel (d) shows the cleanest result and the best match to the model, but a precursor related to the amount of prewhitening noise is present.

Figure 4. HFVS operator used to remove the vibrator signature and replace it with the impulse response shown in (a) and its application to (Figure 2) VSP data (b and c). The operator amplitudes are less than 0 dB, and no gain is applied to the data outside of the sweep band.

HFVS: Enhanced Data Quality

E17

Minimum-phase output
The goal of shaping is to obtain data that have rst-arrivals which match those of impulsive sources and are minimum phase optimal for subsequent processing. VSP rst-arrivals for the integrated HFVS method match those for an impulsive source as shown in Figure 5; precursors are smaller compared to data correlated conventionally or inverted by the original HFVS method. In Figure 6a, we compare data recorded with a shaped sweep followed by correlation and conversion to minimum phase and data recorded with a linear sweep using the HFVS technique and shaped to the same wavelet. The interleaved seismic traces (Figure 6b) show similar waveforms and comparable S/N quality. The main difference is in the rst arrivals in the gathers (Figures 6c and 6d). The HFVS data do not show the precursor side lobes typically observed with vibrator data, and they are more suitable for automatic rst-break picking. Processing the data to stack also results in data with similar phase and S/N quality (Figure 7). In these tests the same shaped wavelet is used for both techniques, but we have found that we can shape the HFVS data differently to improve resolution. Finally in Figure 8, surface seismic HFVS data processed by these methods and converted to zero phase after surfaceconsistent spiking deconvolution is shown with a zero-phase VSP corridor stack. The VSP corridor stack has been processed with downwave deconvolution and should be true zero phase. The VSP ties the seismic data with minimal need for time shifts or phase rotations.

Figure 5. Application of the different methods to VSP vibrator data compared to data recorded with an impulsive source. Shown are (a) the shaped-sweep method, (b) the original HFVS method, (c) the integrated method, and (d) the results from an impulse. The impulse is generated by applying a pulse to the vibrator actuator. The integrated method has less precursor and is the best match to the impulsive data.

VIBRATOR SEPARATION
When multiple vibrators are used with HFVS, a lter can be designed using equation 14 to provide separate output records for each vibrator. The method requires that the number of sweeps be greater than or equal to the number of vibrators. To ensure matrix solvability, during the sequence each of the vibrators should sweep, at least one time, differently than the other vibrators. A phase rotation of 90 for each of the vibrators during one of the sweeps (Table 1) is sufcient for data separation and to keep the matrix S S in equation 14 from becoming singular. Harmonics are part of the vibrator signature and become part of the signal instead of noise. However, we observe some residual harmonics, which are not represented in the measured ground-force signal and appear as harmonic ghosts in the output. It is helpful to superimpose

Figure 6. (a) Shot records recorded with (left) a two-vibrator array and the integrated HFVS method and (right) a four-vibrator array and the shaped-sweep method. The boxes correspond to zoomed images in (b) and (c). The traces match wiggle by wiggle, seen in interwoven traces (b), in which the rst trace in each pair is the HFVS method and the second is the correlated shaped sweep followed by vibrator signature deconvolution. (d) The rst arrivals for the shaped sweep show the side lobes normally associated with correlated vibrator data, but side lobes are not visible with HFVS data (c). Ground roll is stronger on the HFVS data in this example because of the smaller source array.

E18

Krohn and Johnson

a phase rotation to suppress these residual harmonics For example, for four sweeps, phase rotations of 0 , 90 , 180 , and 270 can be added to subsequent sweeps in Table 1 to generate the phase rotations shown in Table 2. This phase rotation can give up to 20 dB suppression of the residual harmonic ghosts up to the fth order (Rietsch, 1981).

Offset VSP tests


As a rst test, we compare two offset VSPs that were acquired individually and separately using HFVS. All recordings were made with a downhole cemented array of 80 geophones. A vibrator was located at an offset of 300 m and was activated to generate offset VSP data. A second vibrator was located at an offset of 450 m, and a second offset VSP was recorded. Then the two vibrators were operated simultaneously and the records were separated using HFVS. Comparisons of the two VSPs acquired with the two techniques are shown in Figure 9. The data for each offset were subtracted; the amplitude of the subtracted record (the difference) is 40 dB, compared with the 0 dB amplitude of the rst record. This difference can be attributed partly to the difference from repeating the

Table 2. HFVS phase rotation plus rotation to suppress residual harmonics. Sweep 1 2 3 4 Vibrator 1 ( ) 0 90 180 0 Vibrator 2 ( ) 0 90 270 270 Vibrator 3 ( ) 0 180 180 270 Vibrator 4 ( ) 90 90 180 270

Figure 7. Interwoven traces from stacks of HFVS data (rst trace) and shaped-sweep data (second trace). There is an excellent match of waveforms and S/N quality.

Figure 8. HFVS processed and stacked data converted to zero phase and compared to a VSP composite trace. The HFVS data tie the VSP data without a time shift.

Figure 9. Offset VSPs acquired individually and simultaneously. Two vibrators were located at different offsets of (a) 300 and (b) 450 m from the well and operated one at a time. Then they were operated simultaneously, and the records were HFVS separated. The amplitude of the difference in the two methods is less than 40 dB compared to that of the records.

HFVS: Enhanced Data Quality

E19

experiment. Even after removing the strong rst arrivals and imaging the reectors (Figure 10), there is little difference between separate or simultaneous acquisition of the vibrator data. In a second test at a different site, we compare the residual left on each record after separation from the other vibrators operating at the same time. This test was performed prior to recording a 3D VSP with four vibrators and an 80-level receiver tool. During the survey, HFVS was used to record simultaneously two vibrators centered at two different shotpoints, reducing acquisition costs. To evaluate the separation effectiveness, we located the four vibrators at offsets of 90, 760, 1100, and 1500 m from the well and operated them simultaneously. Separated HFVS data using eight sweeps are shown in Figure 11. The high-amplitude tube waves and rst arrivals visible on the near-offset record (Figure 11a) are not visible in the other records. P-wave and S-wave reections are evident on the far-offset data, but the tube waves corresponding to the near-offset data cannot be located. The noise before the rst arrivals comes from a nearby gas plant. The process has separated these events, showing no contamination of the records by the other vibrators operating simultaneously. We used fk analysis to search for linear tube waves on the far-offset data in a time window corresponding to the tube-wave arrivals on

the near-offset data. They could not be observed and were below the background noise. Since the signal-to-ambient-noise ratio was approximately 60 dB, the data separation was at least 60 dB in this example.

Surface seismic tests


Separation of surface seismic data is shown in Figure 12. Again, after separation the cross-contamination between vibrator records is not evident and is below background noise. These data were recorded at the start of a 3D survey. The test involved locating four vibrators at the intersections of a receiver line and different source lines. The vibrators were approximately 130 m apart. When the data from a single sweep were correlated, the records showed the superposition of data from the four vibrators (Figure 12a). After separation, the strong rst arrivals and ground roll from the different sources could not be located on the other records as in the separated data for the second vibrator (Figure 12b). The separation exceeds 60 dB (Figure 12c). One to two traces corresponding to geophones at the same stations as the vibrators are overdriven and leave some noise on the separated traces.

Separation quality control for acquisition


It is important to monitor data quality during acquisition for all land operations. For HFVS acquisitional, standard vibrator attributes and the condition number from the matrix inversion in equation 14 can be used to monitor possible problems in the separation process. The process is robust. In production 3D surveys, we typically nd only a few records that are noisy or show problems in separation. If a vibrator is not performing properly and the ground-force signal becomes corrupted, then full separation may not be achieved, although the data still may be useable. In Figure 13, the rst arrivals from the nearoffset records are visible on the far-offset records. The vibrator signature (Figure 14) indicates that one vibrator has developed a serious problem: The amplitudes in the middle of the

Figure 10. Images made from the data shown in Figure 9 after removing the rst-arrivals and imaging the reections using VSPcommon-depth-point transform for data acquired (a) separately and (b) simultaneously. The rst image in each panel corresponds to a source offset of 300 m, and the second corresponds to a source offset of 450 m. There is little difference between the images.

Figure 11. Data recorded using HFVS and four vibrators located (a) 90, (b) 760, (c) 1100, and (d) 1500 m away from the well and operated simultaneously. Strong direct arrivals and tube waves visible from the near-offset vibrator record (a) cannot be located on the separated far-offset record (d).

E20

Krohn and Johnson

sweep are half of the normal amplitudes, and the waveform shows a doublet. Even so, the data are useable; reections are readily visible in the far-offset data. With HFVS, it is important to track the geometry through the separation process because the number of records after separation can differ from those acquired. New SEGD le format standards, which allow for multiple shotpoints per record, and recording systems in which vibrator accelerometer and geophone records are recorded together will facilitate HFVS acquisition and separation.

PROCESSING VIBRATOR DATA AS UNIQUE SOURCE POINTS


During conventional acquisition, data from vibrators operating simultaneously form a source array. There are advantages, however, in separating these data so that unique records are obtained for each vibrator. With HFVS, there is no increase in source effort and acquisition time. The cost differential should be small and is associated with recording vibrator motions and storing data uncorrelated and unstacked.

Intra-array effects
CMP gathers of HFVS data after separation can show a substantial intra-array effect, especially in areas with large surface relief. Data from a sand dune and sabkha Figure 12. Surface seismic gathers recorded by four vibrators located 130 m apart area with 60-m dunes are shown in Figalong one receiver line. (a) When the data from a single sweep are correlated, arures 1517. Prestack-migrated images (Figrivals from all four vibrators are evident in the record. When the data are separated ure 15c) for HFVS are substantially better using HFVS, the rst arrivals and ground roll from neighboring vibrators are not with improved reector continuity and represent in the record. Only the record for the second vibrator is shown in (b). (c) duced noise compared with those for conThe HFVS data plotted on a decibel scale. ventional correlated data (Figure 15a). Part of this improvement arises from the correction of intra-array effects (Figure 15b). Data are shown for the negative offsets of a CMP gather in Figure 16. There are several skips because the vibrators could not reach the top of the steepest dunes. The four traces close together are generated by separating the records for the four vibrators at each source station. In conventional recording, the vibrators form an array, and there would be a single trace. The vibrators are located bumper to bumper, and the pads are approximately 13 m apart. As can be seen in Figure 16a, there are misalignments between events across each set of four traces because of elevation changes and differential moveout between vibrator positions. In this example at an offset of 1.6 km, there is 12 ms of misalignment across the 39-m array. Of this misalignment, 4 ms is caused by statics corrections and elevation changes and 8 ms is caused by difFigure 13. Separated data from two vibrators centered at an offset of 90 m (a and b) and two vibrators centered at an offset ferential moveout. With conventional recording, these traces of 1500 m (c and d). In comparison to Figure 11, some of the would be summed by the array as shown in Figure 17a. With rst arrivals from the near-offset data are present on separated HFVS, corrections can be made for intra-array statics (Figdata for the far offsets. Ground-force data are shown in Figure 16b) and NMO moveout (Figure 16c), resulting in better ure 14.

HFVS: Enhanced Data Quality

E21

alignment and more coherent reections after the array sum (Figure 17b).

Migration with greater spatial sampling


Further improvement for these data can be obtained by binning the HFVS data at 6.25 m instead of the natural 25-m CMP bins. Kirchhoff prestack time migration was performed with the smaller CMP bins (Figure 15c) and output with 25-m bins. This can be compared to Figure 15b, where corrections

Figure 14. Vibrator signature data for the record in Figure 13, indicating a corrupted ground-force signal from vibrator four for (a) the waveforms and (b) the spectra.

Figure 15. Migrated images from a sand dune and sabkha environment, comparing (a) conventional vibrator data with (b and c) HFVS data. (b) Correction is made for intra-array effects, an array sum is performed, and the data are prestackmigrated with 25-m CMP bins. (c) The ne trace spacing is carried through migration. Correction is made for the intraarray effects, and the data are migrated with 6.25-m bins and output at 25 m. Correction of intra-array effects and migration with smaller bins both contribute to the imaging improvement over conventional data, especially noticeable between 1.8 and 2.0 s.

Figure 16. Negative offsets for a CMP gather showing substantial intra-array effects. The four neighboring traces are from data separated from four vibrators located 13 m apart. With conventional recording, these would be summed. Shown are the (a) raw data, (b) after static corrections, and (c) after NMO. The data are better aligned for the array sum after corrections are applied.

E22

Krohn and Johnson

are made for the intra-array effects, array summed, and then migrated with 25-m bins. Gaps in the source coverage at the top of the dunes result in poor noise attenuation by the stack after migration. Migrating with the smaller bin size improves noise suppression in this example. Alternatively, the data can be output with the smaller bin size.

roll noise is not aliased (Figure 18b) and can be suppressed. A phase-match lter is used to remove ground roll, and then the four records for each station are summed. Reections are now visible under the ground roll cone (Figure 19b). Unmigrated

Ground-roll suppression
HFVS recording with multiple vibrators at a small source interval can be used to better sample and suppress strong ground-roll noise. In the area shown in Figures 1820, conventional recording is performed with four vibrators per station. Raw shot gathers after HFVS separation of the four vibrators are shown in Figure 18. Ground roll is aliased by the receiver spacing for both the individual gathers (Figure 18a) and the array sum (Figure 19a). If a supergather is constructed by sorting the separated data on station and offset, then the ground-

Figure 18. Shot gathers for separated HFVS data with four vibrators per station. (a) The ground roll is aliased in the four shot gathers. (b) If a supergather is constructed by sorting into station and offset order, the ground roll is not aliased and can be removed.

Figure 17. CMP gather from Figure 16 after the array sum. (a) The array sum is performed with the uncorrected data from Figure 16a followed by NMO and static correction. (b) after An array sum is made of the corrected data from Figure 16c. Both positive and negative offsets are displaced without the near-offset traces. The gather with intra-array corrections made before the array sum shows better reector continuity.

Figure 19. Array sums for the data in Figure 18, showing (a) the array sum followed by ground-roll suppression as in conventional recording and (b) ground-roll suppression using the supergather followed by the array sum as can be done with HFVS. More ground roll can be removed, with the HFVS data showing reectors under the air blast.

HFVS: Enhanced Data Quality

E23

Optimal separation is achieved by HFVS through the design of a single separation lter based on all sweeps at a location. Separation is greater than 60 dB. This method gives the survey designer increased exibility. A survey can be designed traditionally with three or four vibrators per source point, or it can be designed to have one vibrator per shotpoint with a decreased shot interval. Data quality can be improved by correcting for intra-array effects and improving noise suppression. Alternatively, in good data areas single vibrators can be located at each source point and the data collected simultaneously, reducing costs. This paper demonstrates the advantages of single-source recording without arrays. There also should be advantages in reducing the size of receiver arrays. With larger channel-count recording systems and with HFVS, the potential exists for further data-quality improvements with better sampling of both signal and noise.

REFERENCES
Figure 20. Unmigrated stacks for an area with large groundroll noise for (a) shaped-sweep data and (b) HFVS data. Ground roll can be removed better with the HFVS sampling (Figure 19), and the data also can be shaped to a greater bandwidth. stacks show reduced noise and wider bandwidth from the use of HFVS (Figure 20).
Allen, K. P., M. L. Johnson, and J. S. Ma, 1998, High delity seismic (HFVS) method for acquiring seismic data: 68th Annual International Meeting, SEG, Expanded Abstracts, 140146. Andersen, K. D., 1994, Shaped-sweep technology: U. S. Patent 5 347 494. Baeten, G., and A. Ziolkowski, 1990, The vibroseis source: Elsevier Science Publishing Co. Baeten, G. J. M., J. T. Fokkema, and A. M. Ziolkowski, 1987, Seismic vibrator modeling: Geophysical Prospecting, 36, 2265. Bickel, S. H., 1982, The effects of noise on minimum-phase vibroseis deconvolution: Geophysics, 47, 11741184. Gibson, B., and K. Larner, 1984, Predictive deconvolution and the zero-phase source: Geophysics, 49, 379397. Rietsch, E., 1981, Reduction of harmonic distortion in vibratory source records: Geophysical Prospecting, 29, 178188. Sallas, J. J., 1984, Seismic vibrator control and the downgoing P-wave: Geophysics, 49, 731740. Sallas, J., D. Corrigan, and K. P. Allen, 1998, High delity vibratory source seismic method with source separation: U. S. Patent 5 721 710. Trantham, E. C., 1993a, Seismic vibrator signature deconvolution: U. S. Patent 5 400 299. , 1993b, Minimum uncertainty lters for pulses: Geophysics, 58, 853862. , 1994, Controlled-phase acquisition and processing: 64th Annual International Meeting, SEG, Expanded Abstracts, 890 894.

CONCLUSIONS
Tests show that integrating heritage Exxon shaped-sweep and heritage Mobil HFVS technology improves data quality. The data after separation are minimum phase and match the data recorded with impulsive sources. Traveltimes can be picked accurately for checkshot times, tomography, and static corrections. In addition, the phase is optimized for subsequent processing, and after conversion to zero phase it matches zerophase VSP well data The S/N quality improves, and the need for further bandpass ltering is reduced.

Potrebbero piacerti anche