GEOPHYSICS. VOL. 52. NO. II (NOVEMBER 1987); P. 1501-1524.24 FIGS.
`
`Experimental investigation of interference from
`other seismic crews
`
`Walt Lynn*, Mark Doyle*, Ken Larner*, and
`Richard Marschallt
`
`ABSTRACT
`
`In a study of the contamination of reflection seismic
`data by interfering noise from other seismic crews, con(cid:173)
`trolled experiments were performed in the Gulf of
`Mexico and the North Sea. In each experiment, a
`survey ship traversed a line several times collecting both
`data free of and data contaminated by interfering crew
`noise. In the Gulf of Mexico experiment, the "noise"
`ship followed a prescribed course about 11 km from the
`survey ship. In the North Sea experiment, the noise ship
`was positioned at stationary locations 10 and 40 km
`broadside to the survey line. Recorded interference
`noise in both experiments had peak amplitudes well
`above the 0.5 to 1.5 Pa (5 to 15 J..lbar) limit beyond
`which crews typically must agree on time-sharing.
`Despite recorded crew noise that was three to eight
`times higher than levels typically considered acceptable,
`the conventionally processed common-midpoint stack
`of the contaminated Gulf of Mexico data shows only
`slight evidence of the interference noise; in contrast, the
`North Sea stack is severely contaminated by crew noise
`as early as 1 s. However, when each unstacked trace is
`
`scaled by time-varying weights that vary inversely with
`the local power in the trace, the crew noise is no longer
`visible in the contaminated stack of either data set.
`Trace-weight normalization in this process is designed
`to ensure that stacked signal amplitudes are generally
`preserved. A simulated line wherein the actual Gulf of
`Mexico data are contaminated by crew noise five times
`stronger than that recorded in the field [yielding ef(cid:173)
`fective peak noise values of 7.5 to 20 Pa (75 to 200
`J..lbar)] also shows no evidence of crew noise after
`inverse-power weighted stacking.
`When data processing includes conventional stacking,
`we recommend that the specified tolerable amount of
`crew noise be based upon the root-me an-square ampli(cid:173)
`tude of the crew noise computed over an entire record.
`With burst suppression techniques, such as inverse
`power-weighted stacking, we recommend that the speci(cid:173)
`fied level be based upon the duration of the strong(cid:173)
`amplitude burst as well. With both criteria, field specifi(cid:173)
`cations can be chosen that remain conservative while
`tolerating considerably more crew-interference noise
`than in the past. Issues of the influence of crew noise on
`the analysis of prestack data remain for future study.
`
`00 o
`
`INTRODUCTION
`
`Quality control during seismic data acquisition is governed
`oy predefined specifications, or "specs." A spec defines the
`extent to which a given component of the acquisition system is
`allowed to be degraded before it must be repaired or acqui(cid:173)
`sition must be delayed. Examples of specs include the number
`and size of air guns allowed to fail in a source array, the
`minimum number of vibrators in a Vibroseis crew, the maxi-
`
`mum aHowable recorded ambient noise level, and the multi(cid:173)
`pJicity of coverage in each ceH of a three-dimensional (3-0)
`survey. AH specs should be reconsidered periodicaHy to ensure
`that they take into account advances in acquisition and pro(cid:173)
`cessing technology, as well as new demands in interpretation.
`Specs for the tolerable level of ambient noise in the days of
`12-channel or 24-channel acquisition, for example, are not ap(cid:173)
`propriate when we now routinely record with 120-channel and
`240-channel streamers and with 96-channel land cables. Like-
`
`Presented at the 55th Annual International Meeting, Society of Exploration Geophysicists, Washington, D. C. Manuscript received by the Editor
`June 2, 1986; revised manuscript received March 30, 1987.
`*Western Geophysical, P.O. Box 2469, Houston, TX 77252.
`tFormerly Western Geophysical; presently Planning Systems Incorporated, 115 Christian Lane, Slidell, LA 70458.
`(!;:) 1987 Society of Exploration Geophysicists. All rights reserved.
`
`1501
`
`WesternGeco Ex. 1021, pg. 1
`WesternGeco v. PGS
`IPR2015-00311
`
`

`
`1502
`
`Lynn et al.
`
`wise, we should be able to tolerate higher ambient noise levels
`with today's 107 Pa-m (100 bar-m) sources than with yester(cid:173)
`day's 2 x 106 Pa-m (20 bar-m) sources.
`Only through adherence to properly chosen specs can one
`obtain data that, after processing, will yield results not unac(cid:173)
`ceptably compromised by choices made in acquisition. On the
`other hand, specs that are inordinately stringent drive up the
`cost and the time required for a survey. The setting of specs
`that are neither too severe nor too loose is not straightfor(cid:173)
`ward. The task is complicated because the spec requirement
`for any given component of the acquisition system must truly
`depend on the seismic system as a whole, including how the
`data will be processed and how they will be interpreted. His(cid:173)
`torically, specs have been set conservatively, the philosophy
`being that it is better to bear the cost of down-time in acqui(cid:173)
`sition than to suffer any compromising of data quality. Indeed,
`many specs are set by what had worked previously. regardless
`of whether or not previous specs are or were too conservative.
`Here we concentrate on one particular spec used in marine
`acquisition, the tolerable amount of recorded energy from
`other seismic crews. We refer to this noise as "interference" or
`"crew noise." It is not uncommon to have two or more seismic
`boats operating in a given survey area. When the crew-noise
`level exceeds a chosen spec [sometimes stated in terms of the
`largest tolerable peak or root-mean-square (rms) amplitude of
`interfering crew noise; sometimes included in a general "cable
`noise" spec], the solution has been for the crews to time-share.
`In time-sharing, the crews agree to shoot only during prede(cid:173)
`termined time slots and remain idle while awaiting their slot.
`This solution is costly, in terms of both delays in completing
`surveys and costs for idle crew time, particularly in areas such
`as the North Sea and Beaufort Sea, where data can be col(cid:173)
`lected during only limited times of the year. The high cost and
`frequency of time-sharing (as much as 60 percent of sea time
`in some areas during peak activity in the North Sea) have led
`us to question how much of this practice is truly necessary.
`We address this question by examining the processed results
`from two controlled experiments, one from the Gulf of Mexico
`and one from the North Sea. Although our data examples are
`from the marine environment, the conclusions are also appli(cid:173)
`cable to problems of interference in land surveys.
`
`t(cid:173)
`
`oo o
`
`15
`
`20km
`
`Interference Ship-B
`
`10
`
`5
`
`Survey Ship-A
`
`~O 5
`
`10
`
`15
`
`20km
`
`o
`FIG. 1. Relative paths ot a survey ShIP and an interference s.hip
`in a controlled crew-noise experiment in the Gulf of MexIco.
`The small perpendicular tick~ on each tra~kline represent rela(cid:173)
`tive positions of the two ships at 5 km 10tervals along each
`path. The dashe~ line abov~ th~ survey line shows the lo(cid:173)
`cations of the sectIOns shown 10 Figures 7 through 10.
`
`GULF OF MEXICO EXPERIMENT
`
`In this experiment, a 23 km survey line was traversed three
`times: once to acquire data free of interfering crew noise; once
`to acquire data while a second seismic "noise" ship followed a
`prescribed path relative to the survey line (Figure 1); and once
`again to acquire only interference noise while the noise ship
`followed its previous path. We refer to these three lines as the
`"uncontaminated," "contaminated," and "crew-noise-only"
`lines. respectively. 120 channels were recorded. with a receiver
`group interval and shot spacing of 25 m. All traverses of the
`survey line were in the same direction, and the entire experi(cid:173)
`ment was performed within 15 hours, in calm seas. For the
`contaminated and crew-noise-only lines, the interference ship
`was always 9 to 12 km away from the survey ship. (For com(cid:173)
`parison, ships as much as 80 km apart are sometimes forced to
`time-share because interfering noise levels are judged out-of(cid:173)
`spec.) The seismic sources in this experiment had peak-to(cid:173)
`trough strengths (in the frequency range 5--128 Hz) of 4 x 106
`Pa-m (40 bar-m) for the survey ship and 5 x 106 Pa-m (50
`bar-m) for the noise ship. By maintaining different firing inter(cid:173)
`vals for the two sources, we simulated the mis-synchronization
`that would typically occur when two seismic crews are work(cid:173)
`ing in the same area. This mistiming also produced test data
`in which the arrival time of crew-noise bursts was distributed
`with equal likelihood over the length of recorded data.
`Figure 2 shows five neighboring shot records from the con(cid:173)
`taminated line (taken 9 km from the start of the line). The
`crew noise is the strong coherent energy arriving around 3.5 to
`4.5 s on the left-most record and at successively later times
`from one record to the next. This noise has peak amplitudes in
`the range 1.5 to 4 Pa (15 to 40 J,lbar), wen above the typical
`crew-noise spec of 0.5 to 1.5 Pa (5 to 15 J,lbar). Along this part
`of the line, the shot intervals of the survey ship and the noise
`ship differ by about 700 ms, causing the first arrival of the
`crew noise to be delayed by about that amount from one shot
`to the next. Such mis-synchronization of the survey and noise
`sources gives rise to time misalignment of the crew noise when
`the data are sorted into common-midpoint (CMP) gathers for
`stacking. As shown later, this misalignment is fundamental to
`our ability to suppress the noise.
`line) of
`Figure 3a shows
`the peak amplitudes (solid
`constant-offset (3000 m) traces across the entire 22.5 km (900
`shotpoints) of the crew-noise-only line. The curve has an oscil(cid:173)
`latory appearance because, with the varying arrival time of the
`crew noise, not all traces contain strong crew noise; for some,
`the noise arrived during the short period of time (here 2 s)
`between the maximum recording time and the firing of the
`next shot. The envelope (dashed line) of the peak-amplitude
`values varies roughly from 0.8 to 4 Pa (8 to 40 Jlbar) for these
`long-offset traces. The considerable variation in envelope am(cid:173)
`plitude along the line implies areas of locally strong and areas
`of locally weak crew noise. The corresponding amplitude en(cid:173)
`velope for the uncontaminated survey (not shown here) has a
`similar variability. We conclude that these variations in crew(cid:173)
`noise strength are caused primarily by variations in water(cid:173)
`bottom or sub-water-bottom geology, rather than by array
`effects associated with varying azimuths and distances be(cid:173)
`tween the noise ship and survey streamer.
`The upper two curves in Figure 3b are counterparts of the
`
`WesternGeco Ex. 1021, pg. 2
`WesternGeco v. PGS
`IPR2015-00311
`
`

`
`Interference from Other Seismic Crews
`
`1503
`
`FIG. 2. Five representative neighboring shot records taken 9 km from the start of the crew-noise-contaminated survey
`line in the Gulf of Mexico experiment. The interfering crew noise in these shot gathers is the strong coherent energy
`arriving at around 3.5 to 4.5 s on the left shot record and progressively later on the othe,r records. Peak. amplitudes of
`the crew noise are as large as 4 Pa (40 J1bar), well beyond the typical Gulf of Mexico spec. These data have been gained
`to compensate amplitudes only for geometric-spreading decay.
`
`t"(cid:173)
`oo
`o
`
`.4
`
`5
`
`10
`Distance (km)
`
`15
`
`20
`
`FIG. 3. Amplitude levels on a far-offset (3000 m) trace along the Gulf of Mexico crew-noise-only line. (a) Peak
`amplitude (oscillatory, solid curve). The variation in the envelope (dashed curve) of peak amplitudes is a result of
`changing near-water bottom conditions along the line. (b) rms amplitude. The oscillatory, solid curve is the rms
`amplitude (measured over a duration of 8 s) of the same traces used in (a). The envelope of rms amplitudes (dashed
`line) also shows considerable variation along the line, but remains well below 0.3 Pa (3 J1bar) for most of the line. The
`lowest curve is the rms amplitude of the traces that do not contain the 1 s or so of strong burst-like crew noise.
`
`WesternGeco Ex. 1021, pg. 3
`WesternGeco v. PGS
`IPR2015-00311
`
`

`
`1504
`
`Lynn et al.
`
`3.5
`
`g: 0
`
`-3.5~--~--~2~--~--~4----~---6~--~----
`
`Time(s)
`(a)
`
`3.5
`
`8!. 0
`
`.0
`U
`
`o
`
`.0
`U
`
`-40
`
`- -I'- - -1- -
`
`I
`I
`I
`I
`t
`r I I
`I
`I
`... -1- .. -t-t- t- -
`
`,
`
`-
`
`-
`
`Frequency (Hz)
`(d)
`
`---------------
`
`I
`
`I
`
`I
`
`I
`
`------------
`,
`
`------.---
`
`- 3.5 ...L...-.---'-----l.2---'---.....J
`4' - - - - ' - - - . . . . . .L - - - -L - -
`
`Time(s)
`(b)
`
`5
`
`10
`
`100
`
`Frequency (Hz)
`(e)
`
`o --------------------~~~-~---- ---------------------
`
`I
`
`I
`
`00 o
`
`-
`
`- -
`
`I
`- - - '1 -
`I
`I
`
`-20
`
`.0
`U
`
`-40
`
`- - -1- - -1- - 1 - T - t- -. -'-1- - - - -
`
`,
`I
`I
`
`I
`I
`I
`
`I
`t
`I I I
`
`I . ,
`I I
`
`I
`
`I
`
`I
`
`I
`I
`
`t
`I
`
`t
`I
`I
`
`-3.5~-......L--~2---~--~4~--'---~6~-...L...-.---
`
`- 60 1L...--~--'----'--5::-..L.-...J.!-'-'-1-:-':!0:-----'-------:50:-:----1.':-OO:­
`
`I I I
`I
`
`Time(s)
`(e)
`
`Frequency (Hz)
`
`(f)
`
`FIG. 4. Representative crew-noise traces and their associated amplitude spe~tra. (These data unavoidably contain
`ambient noise as well.) (a) Crew-noise-only trace from the first two-thirds of the Gulf of Mexico line, where the crew
`noise consisted of only 1 s bursts of low-frequency energy. (b) Crew-noise-only trace taken from the last one-third of
`the line, where the crew noise consisted of both low-frequency bursts and 100 nis bursts of high-frequency energy.
`(c) Trace of ambient noise only. (d), (e), and (f) Amplitude spectra for the traces shown in (a), (b), and (c).
`
`WesternGeco Ex. 1021, pg. 4
`WesternGeco v. PGS
`IPR2015-00311
`
`

`
`Interference from Other Seismic Crews
`
`1505
`
`1000
`
`100·
`
`Ii
`~ 10
`Q.l
`"0
`.~
`a.
`E -«
`
`1
`
`.1~ __
`
`2
`
`4
`Time (5)
`
`6
`
`8
`
`FIG. 5. Signal amplitude recorded by a hydrophone group
`1700 m from the source in the uncontaminated line. At each
`time sample, the amplitude is the average trace magnitude
`over 100 successive shot records in the Gulf of Mexico experi(cid:173)
`ment.
`
`t--
`00
`o
`
`s
`
`2
`
`4
`
`6
`
`envelope and amplitude curves shown in Figure 3a, but these
`are rms amplitudes measured over the 8 ms duration of the
`traces. The curves for rms amplitude parallel those for peak
`amplitude simply because the rms amplitudes are dominated
`by the strong early arrivals of crew noise. Significantly, how(cid:173)
`ever, even at the noisiest crew-noise locations (at around 8 to
`10 km) the rms amplitude is just 0.3 Pa (3 J.1bar), within com(cid:173)
`monly used specs for ambient noise.
`Not only the amplitude but also the character of the crew
`noise changes along the line. Figure 4 shows representative,
`constant-offset traces (offset of 3000 m) and their spectra, ex(cid:173)
`tracted from near the beginning and near the end of the crew(cid:173)
`noise-only line. For comparison, Figure 4 also shows a trace
`from a postsurvey, ambient-noise strip. Trace 4a contains a
`low-frequency ( < 25 Hz) crew-noise burst and a relatively con(cid:173)
`stant, low level of noise elsewhere, while trace 4b has a high(cid:173)
`frequency (75-120 Hz), short-duration component as well as
`the low-frequency component. The strong low-frequency noise
`extends over roughly 1 s, and the duration of the strong,
`high-frequency component is limited to about 100 ms. Both
`components are thus of limited duration (i.e., they are "burst(cid:173)
`like") compared to the full 8 s length of the trace.
`Apparently the high-frequency component, which appeared
`only in the last 300 shots, is sensitive to the geology near the
`water bottom; but its exact transmission mechanism is un(cid:173)
`known. Where both high-frequency and low-frequency com(cid:173)
`ponents exist, however, we find that the two components
`travel with different group velocities, indicating that the strong
`crew noise propagates as dispersive waves. In routine pro-
`
`s
`
`2
`
`4
`
`6
`
`8
`8
`FIG. 6. CMP gathers from the Gulf of Mexico crew-noise-contaminated line. The crew noise (anomalous amplitudes
`that are most evident after 4 s) is now disorganized relative to the coherent appearance on the shot records (Figure 2).
`
`WesternGeco Ex. 1021, pg. 5
`WesternGeco v. PGS
`IPR2015-00311
`
`

`
`1506
`
`lynn et at
`
`cessing of seismic data, frequency filtering would typIcally ex(cid:173)
`clude the high-frequency noise. Such noise, however, might
`still be observed in the field and influence decisions as to
`whether or not a survey is within spec.
`Consider the zones of low-level noise following the bursts in
`Figures 4a and 4b. Those zones must contain some amount of
`residual, persistent crew noise, whose variance would add to
`the variance of the ambient noise. Since the noise levels in
`those zones are comparable to the level of ambient noise seen
`in Figure 4c, one might conclude that the persistent crew noise
`that follows the burst is weaker than the ambient noise. We
`cannot be certain of the noise level, however, because the
`noise strips and crew noise-contaminated traces could not be
`acquired simultaneously, and ambient-noise levels measured
`on different noise strips varied considerably. (rms values
`varied from 0.05 to 0.1 Pa for this offset, and were as much as
`0.35 Pa for short-offset traces.) Nevertheless, rms amplitudes
`of the portions of traces following the crew-noise bursts
`(bottom curve in Figure 3b) were roughly 0.07 Pa, consistent
`with the measured ambient-noise levels.
`For comparison with the levels of non-burst-like noise seen
`in Figures 3 and 4, Figure 5 shows typical temporal behavior
`of uncontaminated data traces taken from a number of shot
`records. The amplitudes shown in Figure 5 are sample-by(cid:173)
`sample, rms-average magnitudes for the first 100 traces re(cid:173)
`corded by the hydrophone group 1700 m from the source. Just
`as noise amplitudes vary considerably along the line, details of
`the signal behavior can vary substantially as well. We note
`here only general characteristics of the curve in Figure 5: (1)
`early amplitudes exceed 100 Pa, substantially more than the
`1.5 to 4 Pa observed for the crew noise; (2) amplitudes decline
`
`generally monotonically and are still decreasing at a record
`time of 8 s; and (3) the amplitude level at 8 s (.-0.2 Pal is
`higher than the ambient noise level (0.08 Pal observed prior to
`the first arrival.
`The late energy in Figure 5 is either coherent reflections
`from beneath the line of survey and hence can be stacked
`coherently after proper normal-moveout (NMO) correction,
`or it is source-generated noise that mayor may not stack
`coherently. If in an N-fold stack this late energy stacks inco(cid:173)
`herently, the source-generated noise gets the same N -1/2 treat(cid:173)
`ment as does the ambient noise and (as we show later) the
`crew noise; crew noise, if it stacks coherently, is preserved
`intact.
`
`SUPPRESSION OF THE CREW NOISE
`WITH CONVENTIONAL CMP STACKING
`
`As long as the firing intervals of the survey and noise
`sources differ, crew noise arrives at different times on neigh(cid:173)
`boring shot records. Thus, when the data are sorted into CMP
`gathers and corrected for NMO, the crew noise is misaligned
`and will be attenuated with just conventional CMP stacking,
`the most powerful general tool available for suppressing noise.
`Figure 6 shows five adjacent CMP gathers from the contami(cid:173)
`nated line in our Gulf of Mexico experiment. Because the
`survey source and noise source were not synchronized, the
`noise is sporadic and incoherent along the hyperbolic trajec(cid:173)
`tories that we associate with moveout of signal. As a result,
`conventional N-fold CMP stacking, wherein all traces are
`weighted equally (by liN), will suppress the crew-noise bursts
`relative to signal by a factor M I/2/N, where M is the number
`
`00 o
`
`FIG. 7. Near-trace section of the Gulf of Mexico crew-noise-contaminated line. The crew noise appears as the isolated,
`vertical wave trains that occur sporadically throughout the section.
`
`WesternGeco Ex. 1021, pg. 6
`WesternGeco v. PGS
`IPR2015-00311
`
`

`
`Interference from Other Seismic Crews
`
`1507
`
`FIG. 8. CMP stack of the Gulf of Mexico crew-noise-contaminated line. The crew noise is apparent only deep in the
`section, below 5 s. Compare with Figures 9 and 10. An these data have been processed and displayed so as to preserve
`relative trace amplitudes.
`
`00 o
`
`s
`
`2
`
`4
`
`6
`
`8
`
`FIG. 9. eMP stack of the Gulf of Mexico uncontaminated line. Much of the extraneous coherent energy seen in the
`contaminated line (Figure 8) is also seen here, implying that much of this energy is generated from the survey Source,
`not from the interfering noise source.
`
`WesternGeco Ex. 1021, pg. 7
`WesternGeco v. PGS
`IPR2015-00311
`
`

`
`1508
`
`Lynn et al.
`
`of traces on which the burst noise appears at any given time.
`From Figure 4a, we might call the data between about 1 and
`2 s "burst noise." We also observe that the 1 s burst occupies
`about 10 percent of the 10 s shot interval. We therefore expect
`that at any given time, typically six of the 60 traces within a
`CMP gather of the Gulf of Mexico data will have the strong
`burst noise.
`The power of conventional CMP stacking for suppressing
`crew noise can be demonstrated by comparing a near-trace
`section with a conventional 6O-fold stack of the noise(cid:173)
`contaminated Gulf of Mexico data (Figures 7 and 8). The
`near-trace section is just single-fold, and yet even here the
`crew noise poses a problem only below 2 s. Above 2 s, the
`strength of the survey source, which is significantly closer to
`the streamer than is the noise source, is sufficient to overpower
`any contaminating effects from the crew noise. Below 2 s, the
`crew noise appears as strong-amplitude wave trains on iso(cid:173)
`lated traces.
`The 6O-fold CMP stack in Figure 8 shows none of the noise
`bursts seen in the near-trace section in Figure 7. Close inspec(cid:173)
`tion of Figure 8 does show a variety of events that could be
`coherent noises of one sort or another. However, comparison
`of the con taminated stack with an identically processed 60-
`fold stack of the uncontaminated data (Figure 9) shows many
`of the same events. Such energy, therefore, must have been
`generated by the survey source. Only deep in the data (below
`5 s) does the contaminated stack show evidence of noise that
`is absent from the uncontaminated data.
`The sections shown in Figures 8 and 9, as well as all subse-
`
`quent sections from the Gulf of Mexico experiment, were pro(cid:173)
`cessed identically. The processing included deterministic signa(cid:173)
`ture deconvolution, NMO correction (all with the same veloc(cid:173)
`ity field) and stack, statistical deconvolution after stack, and
`time-varying band-pass filtering. Relative amplitudes between
`all sections have been preserved in the display of each section.
`Just how should the crew noise appear on these processed,
`stacked sections? The answer depends upon the relative posi(cid:173)
`tions and courses of the survey ship and the noise ship, and
`upon the relative synchronization of the two sources. Figure
`10 shows the stacked data from the crew-noise-only line dis(cid:173)
`played with the same geometric-spreading correction and
`other gain treatment as for the contaminated and uncon(cid:173)
`taminated stacks in Figures 8 and 9. The broad arching pat(cid:173)
`tern is characteristic of a noise ship traversing a path across a
`survey line, such as in Figure 1. The fact that the noise ap(cid:173)
`pears from top to bottom throughout the section (although it
`is relatively weak in the shallow portion) stems from the mis(cid:173)
`synchronization of the two sources. Figure 10 thus shows the
`level of crew noise that we should expect in the contaminated
`stack shown in Figure 8. Above 3.5 s, crew-noise contami(cid:173)
`nation is not an issue with these data. Conventional pro(cid:173)
`cessing alone has reduced the crew noise to the extent that
`only at late times does it even become evident.
`The weak amplitude of the crew noise in the contaminated
`stack is consistent with the measured amplitudes of the noise
`bursts (4 Pal and the source-generated amplitudes seen in
`Figure 5. If we assume that at any given time the burst noise
`occurs on six of the 60 traces in an NMO-corrected CMP
`
`1··-5
`
`2
`
`00 o
`
`5
`
`2
`
`4
`
`8·
`
`FIG. 10. CMP stack of the crew-noise-only line displayed with the same gain as the contaminated and uncontaminated
`stacks in Figures 8 and 9. Crew noise is clearly not an issue above 3.5 s.
`
`WesternGeco Ex. 1021, pg. 8
`WesternGeco v. PGS
`IPR2015-00311
`
`

`
`Interference from Other Seismic Craws
`
`1509
`
`gather and that the burst noise on those six traces stacks
`incoherently, the crew-noise level after stacking is about
`(6)1/2(4)/60 = 0.16 Pa. For comparison, from Figure 5 the am(cid:173)
`plitude of coherent signal would be about 0.3 Pa at 5 sand
`about 0.2 Pa at 8 s. These implied signal and noise levels are
`consistent with the levels of signal and crew noise seen in
`Figure 8.
`
`SUPPRESSION OF CREW-NOISE BURSTS
`WITH WEIGHTED CMP STACKING
`
`The character and organization of crew noise in shot re(cid:173)
`cords and CM P gathers suggest two general approaches for
`attacking the noise beyond what can be achieved in conven(cid:173)
`tional CMP stacking. On shot records, the noise is coherent
`from trace to trace and could be attacked with a dip filter,
`provided the moveout of the noise across a record differs suf(cid:173)
`ficiently from that of the signal. Moveout of crew noise, how(cid:173)
`ever, depends upon the azimuth and distance between the
`interfering source and the survey streamer. When the inter(cid:173)
`fering source is ahead of the streamer, moveouts of crew noise
`and signal will be in the same direction and can also be com(cid:173)
`parable; hence dip filtering to suppress the noise could attack
`the signal as well. Such is the case for the crew noise seen on
`the shot records in Figure 2. Similarly, in areas of complex
`structure where the signal can have a wide range of moveout
`on shot records, an attack on crew noise based on its moveout
`would discriminate against portions of the signal. Moreover,
`whereas ship A may be situated such that the crew noise from
`ship B can be discriminated on the basis of moveout, it is
`possible that ship A's energy cannot be discriminated from the
`signal recorded by ship B. Fortunately, dip fiHering is unnec(cid:173)
`essary.
`For noises, such as crew noise, that are burst-like and occur
`on only a portion of the traces across each CMP gather, the
`10cal1y anomalous amplitudes of the noise itself can be ex(cid:173)
`ploited to supplement the noise-suppression power of the
`CMP stack. Examples of approaches to crew-noise suppres(cid:173)
`sion that discriminate on the basis of noise amplitude are
`median stacking and trimmed-mean stacking (Farmer and
`Haldorsen, 1985), comparison of contaminated data traces
`with uncontaminated reference traces (Akbulut et aI., 1984),
`and noise blocking, either along traces or across different
`gatherings of traces, such as CMP gathers and common-offset
`gathers. Median stacking and noise blocking can be effective,
`but they generally require considerable computation; more(cid:173)
`over, noise blocking requires selection of a specified threshold
`parameter for defining the amplitudes that are considered
`anomalous.
`An efficient alternative to these approaches is to apply data(cid:173)
`determined, independent time-varying weights to the traces
`within each CMP gather. Two particularly simple types of
`weighting schemes are inverse-amplitude and inverse-power
`weighting, wherein rms amplitude and power are measured
`over short, sliding time windows (Embree, 1968; Gimlin and
`Smith, 1980). To preserve general signal amplitudes in any
`such scheme, the sum of the weights for all the traces in a
`CMP gather is made equal to unity. Inverse-amplitude
`weighting is analogous to performing a fast automatic gain
`control (AGC) on the data prior to stacking. Stacking with
`
`00
`o
`
`inverse-power weighting, sometimes referred to as diversity
`stacking, is particularly potent in suppressing anomalously
`strong noise of limited duration. This trace-weighting scheme
`requires only that signal amplitude be roughly uniform from
`trace to trace.
`Consider, for example, stacks of the schematic six-fold CMP
`gather shown in Figure 11. Trace 4 contains an isolated burst
`that is substantially stronger than the aligned signal. With
`uniform weighting, this strong burst-like event will dominate
`the stack trace even though stacking reduces its strength rela(cid:173)
`tive to that of signal by a factor of six. Weighting each of the
`traces by the inverse of its rms amplitude or power (computed
`over, say, a 200 ms window) will further reduce the relative
`contribution of the noisy trace when the data are stacked.
`Consequently, the stack will show signal with the burst noise
`virtually eliminated.
`Such weighted stacking techniques have been used suc(cid:173)
`cessfully in land acquisition and processing for years, and yet
`they are relatively unused in the marine environment. Inverse(cid:173)
`power weighting (diversity stacking) of consecutive Vibroseis
`records, for example, is commonly used to reduce cultural
`noise such as that caused by motor vehicles or animals. Such
`noises in land acquisition are considerably stronger (relative
`to signal) than any crew noise encountered in marine sur(cid:173)
`veying. Although diversity stacking in Vibroseis acquisition
`typically applies to vertical stacking, as opposed to the CMP
`stacking discussed here, the rationale (i.e., attacking anoma(cid:173)
`lous, high-amplitude noise) is the same for both applications.
`Let us compare the actions of uniform (i.e., conventional),
`inverse-amplitude, and inverse-power weighting on the con(cid:173)
`taminated Gulf of Mexico data. We focus on a zone deep in
`the data where the crew-noise contamination is greatest.
`Figure 12a shows a closeup view taken from the lower right
`corner (the noisiest portion) of the contaminated section in
`Figure 8. These data show coherent energy (crew noise) that
`dips from the lower left to the upper right corners. Figure 12b
`shows the stacked data that result from the application of
`inverse-amplitude trace weighting. Here the rms amplitude
`used in the weighting was computed over a 200 ms sliding
`time window, and, as required to help preserve the amplitude
`
`t In
`
`E
`o
`o
`
`~ l
`
`W1 W2 W3 W4 Ws Ws
`LW1 = 1
`
`FIG. 11. Schematic six-fold CMP gather with one trace con(cid:173)
`taminated by burst-like noise. By choosing the trace weights
`Wj to be inversely proportional to the amplitude or power of
`each trace (computed over a short time window, for instance
`150 or 200 ms), the burst-noise contamination of the stack
`trace can be dramatically reduced.
`
`WesternGeco Ex. 1021, pg. 9
`WesternGeco v. PGS
`IPR2015-00311
`
`

`
`1510
`
`Lynn et al.
`
`of the signal, the trace weights sum to unity. The coherent
`crew noise that was so evident in the uniform weighted stack
`is reduced substantially in the inverse-amplitude weighted
`stack, but it is still present, especially in the lower left of the
`section. However, inverse-power weighting (with weights also
`computed over 200 ms windows, Figure 12c), has removed all
`remnants of the crew noise.
`Figures 12a, 12b, and 12c differ only in the amount of crew
`noise that has survived the stack. The signal is virtually identi(cid:173)
`cal in every case. For reference, the same portion of the sec(cid:173)
`tion from the uniformly weighted stack of the uncontaminated
`data is shown in Figure 12d. The uncontaminated section
`differs only slightly from the inverse-power weighted stack of
`Figure 12c. The differences, which are due in part to different
`realizations of ambient background noise in the two data sets,
`are remarkably small considering that the paths of the boat
`and streamer could not have been identical for the two line
`traverses. The slight differences between the signals, which are
`also evident between the conventional stacks of the contami(cid:173)
`nated and uncontaminated lines (Figures 12a and 12d), sup-
`
`port the contention that the differences are not due to inverse(cid:173)
`power weighting. Moreover, such differences in signal exist in
`zones clearly uncontaminated by int