Macro compressive sensing data acquisition

Abstract

The invention relates to a streamer is described. The streamers include one or more streamer segments. The streamer segments may have sensors, channels, and / or an analog array of sensors arranged along their length. At least one of the streamer segments has sensors, channels and / or analog arrays of variable density along the length.

Description

[0001] Divisional Application Statement

[0002] This application is a divisional application of Chinese Patent Application No. 202080008963.X. Technical Field

[0003] The present invention relates to a streamer device, and in particular, a sensor device having sensors, channels or sensor arrays of different densities along a streamer or a section of a streamer.

[0004] The main application of the present invention is seismic data acquisition, but it can also be applied to other fields, such as acoustics and ultrasonics. Background Art

[0005] A towed marine streamer is used for seismic surveys of the seabed to map the characteristics above and below the sea surface. Each streamer includes a cable that is towed behind a vessel below the water surface. The streamer has a series of sensors coupled thereto, which typically measure the pressure difference in the water. Sensors that measure velocity or acceleration (motion sensors) can be used as an alternative, or both types of sensors can be used together to reduce the noise in the output signal, as described below. The sensors are typically evenly distributed along the streamer, and the distance between adjacent sensors can be fixed at about 12.5 m. This configuration results in the density of the sensors not varying along the length of the streamer (e.g., in this example, the density of the sensors will be one sensor per 12.5 m, or 0.08 m -1 , in all directions along the streamer).

[0006] One or more seismic sources are located on the vessel and periodically emit seismic waves that pass through the water surface and reflect off the seabed or the sea floor. These compression waves then return to the ocean surface and are measured by the sensors as a series of compressions and decompressions of adjacent water molecules, thereby providing information about the topography and structure of the ocean surface and the sea floor below the vessel. One streamer can be used, or multiple streamers can be used together and towed side by side behind the vessel to form a sensor blanket (these are called 3D surveys because of the such underwater images that can be made through such configurations). False signals are generated by pressure waves that have passed once near the sensors on the streamer and are reflected back to the streamer by the ocean surface. Typically, false signals do not provide additional useful information, and it is desirable to remove them from the final signal. This can be achieved by using particle motion sensors to supplement the hydrophone sensor readings, which typically results in a better signal-to-noise ratio than the particle motion sensors. The particle motion sensors provide direction measurements and can therefore be used together with the pressure readings to distinguish between waves propagating from the sea surface and the sea floor. The combination of the readings of the two types of sensors allows for the removal of false signals.

[0007] When the towed cable is towed, the received signal is affected by a high level of additional noise due to the natural movement of the sea water and movements other than the ship, as well as the towing of the towed cable itself in the water. Of course, it is desirable to reduce this noise as much as possible and increase the signal-to-noise ratio.

[0008] Traditionally, sensors and output points are evenly distributed at a constant spacing along the length of the towed cable or a section of the towed cable. In this case, the distance between adjacent output points or between adjacent sensors on the towed cable does not change along the towed cable. The selection of the distance between output points and the number of output points required on the towed cable is determined by the signal sampling requirements, and the selection of the distance between sensors and the number of sensors used is determined by the noise sampling and attenuation requirements.

[0009] The Nyquist sampling criterion determines the minimum signal sampling rate to retain a certain amount of information from the signal and stipulates that to sample the signal correctly, an output is required approximately every 3 meters (3.125 m), corresponding to a measurement every 2 ms. However, this does increase the cost and complexity of the equipment and typically provides outputs located every 12.5 meters on the towed cable. This lower sampling rate can lead to aliasing, especially in cases where the higher frequencies and the propagation direction of the compression waves are close to horizontal. Summary of the Invention

[0010] The present invention solves the problems of noise and aliasing in this type of signal, which will be described in detail below. The present invention is applicable to all seismic data acquisition scenarios, including towed streamer, seabed, node, and land seismic data acquisition, but we use the towed streamer scenario as a platform in the following description to explain these ideas. Therefore, when presenting the arguments, towed streamer terminology will be used (where each towed cable may include multiple towed cable sections, such as 100 m towed cable sections), without affecting the generality of the final claims.

[0011] The following terms have a broader meaning to those skilled in the art than the meanings to be presented. These terms are more concise but do not imply or intend to limit the standard meanings.

[0012] It is important to understand the concepts of input points and output points. The input points correspond to the physical locations of the sensors that sense or measure the raw data, and the output points correspond to the physical locations that provide signal estimates along the towed cable, and the signal estimates are typically provided by combining the data of multiple sensors that form a group by means of hardware or software. The positions of the sensors within the group can be measured relative to the positions of the linked output points, and the positions of the individual output points are used together with the received signal to reproduce the seabed structure by signal processing of the output signals received from the known positions along the towed cable.

[0013] A tow cable, which may be several kilometers long, is typically not a continuous cable but is made by joining together several tow cable segments, for example, each tow cable segment may be 100m, 150m or 200m in length. Combinations of tow cables of different lengths can be used. This allows the length of the tow cable to be varied as required and provides the ability to replace one tow cable segment rather than the complete tow cable in the event of a technical failure.

[0014] As used herein, the term sensor refers to any one or more devices that can be used to measure different properties of interest. Particle motion sensors for measuring seismic signals (such as geophones, accelerometers, and / or rotational sensors) or pressure sensors (such as hydrophones) are the most common and are particularly suitable for use with the methods described below. However, any sensor suitable for measuring particle motion or pressure changes and suitable for mounting on a tow cable can be used.

[0015] Particle motion sensors can be sensors for measuring velocity or acceleration, such as moving coil geophones, MEMS accelerometers, or piezoelectric accelerometers. Hydrophones are sensors for measuring pressure changes, for example, by piezoelectric means.

[0016] An analog array is a group of hardwired sensors that are commonly connected to one electronic channel. These sensors typically can include from 4 to 16 sensors, each sensor being 2m to 15m in length (and may overlap). The sensor distribution can be uniform or non-uniform. The output point is typically located at the center of the array. The center of the array can be the physical center, the center of gravity, or both. For this type of array, at least part of the noise attenuation is achieved by adding together the analog signals from the sensors within the array.

[0017] In the case of using individual sensors rather than sensors forming an analog array, these sensors can each be linked to their own channels, and the channels of multiple sensors can provide input data to a specific output point.

[0018] Channels or electronic channels typically include an amplifier or pre - amplifier and an ADC (Analog - to - Digital Converter). They digitize signals from sensors or from sensor arrays, so there is one channel each time a reading from an analog sensor or analog sensor array is digitized. If only one sensor is connected to a channel, this is called single - sensor recording. If multiple sensors are connected to a channel, this is called sensor - array recording. The number of channels present can correspond to the number of data points transmitted to the ship via a seismic cable. Output points themselves are not equivalent to channels. Output points correspond to the data points recorded to tape or provided as output to a client. These data points can be linked to a channel that itself carries data collected from multiple sensors, or they can be linked to multiple channels that each transmit measurement data collected from a single sensor. Alternatively, multiple channels each linked to more than one sensor can be coupled to a single output point. Thus, each output point need not be associated with one or only one analog sensor array or channel. Other channels carry data (which is typically digitized at this point) from the output point to processing software. This processing software typically receives signals from many output points located at known physical positions along the streamer for the purpose of reproducing the signals received from below the sea and the structure of the sea floor itself.

[0019] For single - sensor recording (one channel for one sensor), the location of the channel or channel location is the same as the location of the sensor. For analog array recording (multiple sensors combined as an array for one channel), the location of the channel is typically at the center or centroid of the array.

[0020] In some cases, noise events are locally coherent. Figure 1 Examples of strong locally coherent noise from towed tests are disclosed. In some data - acquisition scenarios, the data may be affected by noise with a relatively short coherence length. An example is shown in Figure 1 , in Figure 1 which shows accelerometer data obtained in a towed - streamer experiment. The left panel is in the time - offset (space) domain and the right panel is in the frequency - offset domain.

[0021] Especially in cases where locally coherent noise is a problem, signal reproduction can be improved by denser sensor sampling near the output points, as described below. To illustrate this, we again use the example of towed - streamer seismic acquisition. Figure 2 Examples of sensors are disclosed where the sensors are spaced to provide an alias - free output at 125 Hz (higher if the filter includes anti - aliasing). Triangles are used in the figure to indicate the positions of the sensors along a section of the streamer. Figure 2 , 3 , 5, and 7 use arrows to indicate the output - point positions along the streamer.

[0022] This application describes a new method for collecting data, which has a higher signal-to-noise ratio (SNR) than common methods in the presence of local coherent noise. The described system is particularly relevant when the number of sensors is limited and describes how to place sensors to reduce the adverse effects of noise and aliasing while minimizing the number of sensors required on the streamer. Part of how the present invention achieves this is by using a variable density function of sensors, channels, and / or analog sensor arrays along the length of the streamer or a section of the streamer.

[0023] Consider two types of sensor layouts, as Figure 2 and Figure 3 shown. The blue triangles represent the relative positions of the sensors and the red arrows represent the output points. In Figure 2 , the required output points are evenly distributed at intervals of 6.25 m. This is designed to provide an aliasing-free output up to 120 Hz, assuming an apparent velocity of 1500 m / s for the slowest seismic signal energy in marine seismic acquisition. This shows a non-uniformly spaced sensor to avoid noise aliasing. If an anti-aliasing filter is used, an aliasing-free output can be achieved up to higher frequencies, but at the cost of some dipping events occurring at higher frequencies.

[0024] The sensor density at a point on the streamer refers to the number of sensors per unit length at that point. If the number of sensors per unit length is fixed, then the sensors can be distributed in various ways. If there are enough available sensors, they can be placed at regular intervals according to the Nyquist criterion, but doing so usually means high cost, high weight, and complexity of the equipment due to the large number of sensors required. If the sensor spacing is fixed, then the density of the sensors does not vary along the streamer. For example, for a sensor spacing of 5 m, at all positions on the streamer, the sensor density will be 0.2 sensors per meter (or 1 / 5 m -1 ).

[0025] An alternative method is to place the sensors pseudo-randomly (at non-uniform intervals) to avoid hard aliasing. A possible pseudo-random sensor distribution is shown in Figure 2 . To provide a clean output at all the required output points, the sensors must be relatively well distributed along the streamer.

[0026] Another method that has been shown to be particularly effective in reducing noise is to concentrate (cluster) the sensors near the output points to better attenuate short coherent length noise. A possible clustered sensor distribution is shown in Figure 3 . This discloses that the sensors are spaced apart to provide a clearer signal at the output points (but only aliasing-free at 60 Hz - higher if the filter includes anti-aliasing). Uniformly spaced output points are shown in the figure, with a spacing distance of 12.5 m. Along Figure 3The sensor density of the shown cable section is now non-uniform. The number of sensors per meter (sensor density) is higher near the output points and lower (down to zero) away from the output points.

[0027] Since sensors are expensive and thus precious, a denser sampling near the output points may require a sparser distribution of the output points (the distance between adjacent output points is on average larger along the cable). Compared to a pseudo-random sampling strategy, by dedicating more sensors to each output point (twice as many in the above example) in the above example, achieving excellent noise attenuation by denser sampling near the output points is at least partially achieved. For example, if the total number of sensors per section is fixed due to cost implications, this means that the number of output points must be fewer. In Figure 3 this is shown for output points spaced at 12.5 m intervals, which is equal to Figure 2 half of the output spacing in the shown example.

[0028] However, if the output points are evenly spaced, then the sparser output points may result in an unacceptable level of signal aliasing. Output points can generally be positioned at uniform or non-uniform intervals. If they are positioned at uniform intervals, the Nyquist criterion sets an upper limit on the length of the interval when non-aliased output data at the required frequency is needed.

[0029] If the number of sensors (e.g., per cable section) is fixed, then this defines the maximum number of sensors that can be allocated to each output cluster. If better noise attenuation is required, then each output point requires a larger number of sensors, and due to the fixed number of sensors, the number of output points needs to be reduced. Using a uniform output spacing, this would mean a longer interval between output points, resulting in aliasing.

[0030] Figure 4 Shows Figure 2 and Figure 3 a comparison of the noise attenuation performance of the two distributions shown (where the output points are evenly distributed along the cable such that the distance between adjacent output points is fixed). The figure shows the power spectral density of the noise versus frequency. We see that compared to different pseudo-random configurations (similar to the configuration shown in Figure 2 ), the clustered sensor configuration (similar to that described in Figure 3 ) achieves a noise attenuation that is approximately 10 dB higher (except at very low frequencies).

[0031] If the output points are evenly spaced, the sparser output points may result in an unacceptable noise / signal aliasing level. For example, a conventional output interval of 12.5 m (the distance between adjacent output points along the cable) will only provide non-aliased output at 60 Hz. Therefore, if the output points are evenly distributed, then Figure 4The noise attenuation shown is only feasible at the cost of allowing aliasing of the output signal.

[0032] As described above, denser sampling around the output points has proven beneficial for attenuating the short coherent noise at that output point. This requires a given minimum number of sensors per output point (to achieve the desired noise attenuation level), but there is also usually a maximum number of sensors achievable per seismic segment (which is typically determined by cost considerations). The result is that there must be fewer output points than would be required to fully sample signals above a certain wavenumber according to classical sampling theory. In particular, in the case of uniformly spaced output points, aliasing will occur for certain signal wavenumbers, preventing correct recovery. Thus, there seems to be a dilemma in that we must either allow more noise than desired or more aliased signals.

[0033] The solution to this is to position the output points at non-uniform intervals to prevent hard aliasing of the signal. The output signals from non-uniform sampling can be regularized (interpolated onto a regular grid) using one of several compressive sensing methods. Otherwise, they can be used directly, for example for seismic imaging, using their known irregular (non-uniform) positions.

[0034] A sensor density function can be defined to describe the density of sensors in a streamer segment centered at a particular location as a function of the location of that streamer segment. The density function is highest where the sensors are closely spaced (high sensor density in the streamer segment where the sensors are closely spaced) and lowest where the sensors are sparsely distributed. In the case where the sensors are clustered in groups, each group will represent a local maximum in the density function. These local maxima do not have to be of the same magnitude, but will be considered local maxima if the density is higher than adjacent locations. The global maximum of the density function for the streamer will represent the highest of these local maxima. Description of the Drawings

[0035] Using a non-uniform distribution of output points along the streamer and sensors of different densities along the streamer, especially in cases where this corresponds to higher density sensors at or near the output points, noise and aliasing can be reduced, and thus a clean signal can be obtained without increasing the number of sensors required. Embodiments of the present invention will now be described by way of example only, with reference to the following drawings, in which:

[0036] Figure 1 Strongly locally coherent noise from a towed test is plotted.

[0037] Figure 2 An example of sensors spaced to provide an alias-free output at 125 Hz (higher if the filter includes anti-aliasing) is shown.

[0038] Figure 3Shows examples of sensors spaced apart to provide a clearer signal at the output point (but only aliasing-free at 60 Hz; higher if the filter includes anti-aliasing).

[0039] Figure 4 Shows test results for sensor placement and filter design for evenly spaced output points.

[0040] Figure 5 Shows examples of non-uniformly spaced output points for avoiding aliasing while allowing good noise attenuation by including sensors clustered near the output point.

[0041] Figure 6 Shows a flowchart of a method for reproducing a signal where the output points are non-uniformly distributed.

[0042] Figure 7 Discloses examples where each local sensor cluster uses more than one output.

[0043] Figure 8 Provides a schematic example of an array non-uniformly distributed along a tow cable.

[0044] Figure 9 Shows the results of a synthetic example using SEAM finite difference data.

[0045] Figure 10 Shows schematic examples of sensors distributed in groups of different sizes along a tow cable. The top row evenly distributes sensor groups along a tow cable segment, and the bottom row distributes the same groups non-uniformly.

[0046] Figure 11 Illustrates a method for reproducing a signal where different groups of sensor clusters are used to represent different frequency ranges in the signal.

[0047] Figure 12 Illustrates a method for reproducing a signal where different groups of sensor clusters are used to represent different frequency ranges in the signal and these groups are non-uniformly distributed along a tow cable segment. Detailed Description

[0048] In some embodiments, the present invention uses non-uniform spacing of input points (sensor positions) and output points to obtain a high-quality (high SNR) estimate of a signal from noise-affected measurements.

[0049] As previously mentioned, a trade-off is needed between allowing more noise than desired or allowing more aliased signals. Non-uniform output sampling can address both signal and noise sampling problems. The aliasing problem can be at least partially overcome by relaxing the requirement that the spacing of output points should be regular and by allowing non-uniformly spaced output points to enable compressive sensing (CS).

[0050] Figure 5 Figure 1 shows a possible configuration of sensors (input points) and output points in an embodiment. The output points are non-uniformly spaced (with an average spacing greater than 12.5 m) to avoid aliasing, while the device allows for good noise attenuation by using sensors clustered near the output points. The sensors are clustered by output and are non-uniformly distributed. Non-uniform distribution means that the distance between adjacent sensors or between adjacent output points is not fixed along the tow cable. For example, for an output point located between two adjacent output points, the distance to the nearest output point in one direction along the tow cable can be less than the distance to the nearest output point in the other direction along the tow cable. If the distribution of the sensors is also chosen to be non-uniform, the same is true for a sensor located between two other sensors on the tow cable. The output points can be spaced along the tow cable section or pseudo-randomly along the tow cable.

[0051] If a regular spacing of 6.25 m is chosen, this example contains fewer output points per unit length of the tow cable compared to 16 output points per 100 m tow cable section, but the signal is clearer (i.e., produces output data with a higher SNR). Cleaner output data is more suitable for compressive sensing solutions, including regularizing to a regular grid using state-of-the-art CS methods. Figure 6 Figure 2 shows a flowchart of a method for reconstructing a signal using non-uniformly distributed output points. A compressive sensing algorithm is used to address the fact that the output points are non-uniformly distributed by reconstructing regularly spaced outputs along the tow cable using the non-uniformly spaced outputs at the output points. The number of reconstructed outputs will typically be greater compared to the output points physically spaced along the tow cable.

[0052] Therefore, the above example allows for the use of various noise attenuation methods to produce clearer output traces. As a non-limiting example, assume there are 56 sensors (input points) per tow cable section (which may be 100 m long). If there are 16 uniformly sampled output points as in the above example, each output point will have 3.5 sensors, and these sensors must be distributed fairly well (non-uniformly) along the tow cable to ensure that each of the 16 output points gives a reasonably clean output. Conversely, if fewer output points (but non-uniformly spaced) are of interest, the noise attenuation method can provide clearer results because there will be more sensors available to provide data to each output point (or the device can use fewer sensors). For example, if we find that 8 non-uniformly sampled output points per tow cable section are sufficient, then each output point will have 7 sensors, and if we only need 5 output points, each output point will require 11 sensors.

[0053] A compressive sensing recovery algorithm will be used to reconstruct the signal using the clearer output data. Examples include: IAA-MP and BPDN. Other examples will be obvious to those skilled in the art.

[0054] When implementing embodiments that use non-uniform distributions of sensors and output points, it is advisable to first optimize the non-uniform output point positions. Assuming clean (signal-only) outputs, one can search for the best non-uniform output positions that allow for optimal reconstruction. The next step is to optimize the input (sensor) positions to match the output points.

[0055] Adjust any of our current sensor optimization techniques, and then we will optimize the sensor positions to give us the cleanest output at the selected positions. This may involve clustering the sensor positions around the output points. The sensors within a group associated with a particular output point can be evenly distributed (e.g., a group of 5 sensors can be placed with a spacing of 0.2 meters, and the output point is equidistant between the two outermost sensors, etc.). This will still result in a non-uniform or uneven distribution of sensors along the tow cable because the distance to the adjacent sensor group associated with the next output point along the tow cable will be different from the spacing between the sensors within the group (i.e., it will not be 0.2 meters in that example).

[0056] Of course, in some embodiments, the sensors within a group coupled to a single output point can also be non-uniformly distributed. A non-uniform distribution of sensors or output points may but does not always mean that all the distances between adjacent sensors or adjacent output points on a particular tow cable segment are different from each other.

[0057] However, as long as at least some of the distances between adjacent sensors or adjacent output points on a tow cable segment are different, the distribution of sensors or output points can be considered non-uniform.

[0058] This two-stage optimization can also be coupled to a single stage, but this will require more expensive optimization efforts.

[0059] In some embodiments, it is possible (and advantageous) for each sensor cluster to produce more than one output.

[0060] We can design noise attenuation methods to provide outputs at more than one point for a given sensor cluster. This will be beneficial for reconstruction; for example, by defining a supplementary finite difference signal. An example of such a configuration is shown in Figure 7 . In this example, the average spacing is greater than or equal to 12.5.

[0061] Figure 8 Schematic examples of arrays non-uniformly distributed along a tow cable are disclosed. Each triangle represents a sensor, and a group of sensors connected together represents a sensor array. Note that the sensor arrays themselves are distributed non-uniformly along the tow cable segment. A single channel will convey data from all the sensors within the array to the output points. Although Figure 8The sensor arrays shown are equally spaced along or within each sensor array, but this is for illustrative purposes only. These sensors can also be non-uniformly distributed. The output point locations can correspond to the physical center and / or the center of gravity of each array.

[0062] Figure 9 Numerical examples are shown. The data represented by the curves in the figures was created using a finite difference code based on the SEAM model. The red box in the top panel shows the portion of the data used for testing. The left panel in the middle row shows the f-k transform of the test data; the right panel shows the aliasing that would occur if the data were sampled regularly at a fixed interval of 12.5. The bottom row shows the test results, comparing the finely sampled reference with the reconstruction performed using the IAA-MP algorithm in a single tx-window.

[0063] If the sampling is non-uniform, the clean signal can be reconstructed (regularized) onto a regular grid. The signal has frequency components up to 125 Hz, so it is significantly aliased if sampled regularly at 12.5 m. Figure 7 The second panel in the bottom row shows an example where 8 output points are non-uniformly distributed over 100 m with an average spacing of 12.5 m. The data is then regularized to a fine grid for comparison with the input. The reconstruction error using the IAA-MP algorithm is less than 5% RMS.

[0064] The density of at least one streamer segment has a density of sensors, channels, and / or sensor arrays that varies by at least a factor of 1.3 over a sliding window of at least 5 m in length. For example, the sensor density within a 6.25 m sliding window may vary along the streamer by more than a factor of 1.5, 2, or 3. The sensor density within a 12.5 m sliding window may vary along the streamer by more than a factor of 1.5, 2, or 3. The sensor density within a 25 m sliding window may vary along the streamer by more than a factor of 1.5, 2, or 3, or the sensor density within a 50 m sliding window may vary along the streamer by more than a factor of 1.5, 2, or 3.

[0065] A sliding window refers to a movable section of a cable or a portion of the cable where the number of sensors can be counted to determine a measure of sensor density. The number of sensors counted, of course, depends on the size of this window. Then, the measure of sensor density at the cable section within the window can be determined by dividing the number of sensors counted by the window size. By sliding the window along a section of the cable (moving the position of the cable section being examined), multiple measures of sensor density representing the position of each cable section can be determined. These positions may or may not overlap. For example, a statement that the sensor density within a 5m sliding window varies by more than 1.5 times indicates that there will be positions on the cable where the window can be positioned to provide a minimum density measure and positions where the window can be positioned to provide a maximum density measure, and that maximum density measure will be more than 1.5 times the minimum density measure.

[0066] In some cases, the density measured at certain positions within the window may be zero. In one example of a cable, the sensors are distributed in groups, each group covering approximately 3m of the length of the cable. These 3m long sensor groups are typically each connected to one output point located at the center of the group (about 1.5m from each outermost sensor), but may also be connected to multiple output points. The output points are spaced approximately 12.5m apart along the cable section.

[0067] The distance between the output points may not be uniform, so the output points are spaced an average of 12.5m apart along the cable section. In this case, the average gap between the sensor groups is about 9.5m. When a 5m window is positioned over this gap, there will be no sensors within the window and the density will be measured as zero. In this case, the ratio of the highest density measured in a sliding window (which can be a 1m, 2.5m, 5m, 6.25m, 12.5m, or 25m sliding window) along the cable to the lowest non - zero density measured in a sliding window of the same size (e.g., the window contains 1 sensor) will exceed 1.5, 2, or 3.

[0068] If a sliding window located at a point on the cable measures zero density, the ratio between the highest density measured in a 5m sliding window and the lowest non - zero density measured in a 1m, 2m, 5m, 6.35m, 12.5m, or 25m sliding window will exceed 1.5, 2, 3, 5, or 10. Any combination of these window sizes and ratios is intended to be covered here.

[0069] For a given length of tow cable, the number of output points required to achieve a given signal quality will depend on the frequency of the signal being measured. For higher frequency signals, the output points will need to be spaced closer together (higher density of output points) in order to achieve a good reproduction of the signal during the processing stage. This is because too low a sampling rate may not be able to distinguish between two signals of different frequencies. For higher frequency signals, a higher sampling rate is generally required. However, in the case of seismic reflections occurring at the tow cable sensors, the noise is higher at lower frequencies (see Figure 4 ). Therefore, for lower frequencies, a lower density of output points is required, but the number of sensors required to feed the signal to each output point needs to be greater to counteract the effect of the higher noise level in the signal.

[0070] To satisfy both the high and low frequencies in the signal while minimizing the number of sensors required for a particular tow cable section, the sensors can be distributed along that tow cable section as shown in Figure 10 . In the top row, the output points (and sensor clusters) are evenly distributed along the tow cable. In the second row, the output points (and sensor clusters) are unevenly distributed. In the example shown, the sensors are located in clusters, each cluster centered around an output point, and the sensors are aggregated around their associated output point as described above.

[0071] Clusters of 5 sensors (green / black triangles) are placed along the tow cable every 12.5 meters. Smaller clusters of 3 sensors (pink / lighter triangles) are placed every 6.25 meters. Some (in this case 3) of the sensors in each larger cluster also form one of the smaller clusters of sensors. Some or all of the sensors in the 5-sensor clusters are thus used to provide outputs for both lower frequency signals and higher frequency signals. Both frequencies meet the requirements in terms of noise reduction and sampling rate, but fewer sensors are used than in the case of clusters of 5 sensors placed every 6.25 meters along the tow cable.

[0072] Of course, the exact distances given are only examples, and the distances between the clusters and the number of sensors in each cluster can be optimized for a particular case or when a particular type of signal is to be detected. Clusters of 3 to 12, preferably 4 to 10, more preferably 4 to 6, and most preferably 5 sensors can be set along the tow cable every 10 to 25 meters, preferably every 10 to 20 meters, more preferably every 11 to 15 meters, and most preferably every 12.5 meters, and smaller clusters of 1 to 5, preferably 2 to 4, and most preferably 3 sensors can be set at shorter intervals (e.g., every 2 to 15 meters, preferably every 4 to 8 meters, and most preferably every 6.25 meters).

[0073] The distances between the clusters can be averaged so that the distribution of the output points along the tow cable section and the distribution of the sensors along the tow cable section are uneven to reduce the effects of noise and aliasing as described above (as in Figure 10as shown in the lower example distribution). If a non-uniform distribution of output points is used, then fewer points may be required along the tow cable and the average distance between the larger and smaller groups may be able to be increased. A particular pattern of groups can be repeated multiple times along each tow cable segment, and multiple such tow cable segments can be connected to form a longer tow cable.

[0074] To account for higher frequencies, additional individual sensors can be included between the groups such that there is an output point every 3.125 meters, or, if a non-uniform distribution of groups and / or output points is used, such that there is an average of one output point every 3.125 meters. These sensors will be located between a group of 5 sensors and a group of 3 sensors. If a non-uniform group distribution is used, each pair of adjacent large and small sensor groups will have an individual sensor midway or on average midway between them.

[0075] The following table summarizes the estimated requirements for each frequency range. The table is particularly relevant in the case of using particle motion sensors, but may also be relevant in other cases.

[0076] Although it is preferred that the distribution of sensors and output points is non-uniform along the tow cable segment, this is not to tailor the frequency to the needs of using different groups. For example, the sensors can be positioned at fixed intervals along the length of the tow cable, and the output points for different frequency measurements are simply linked to specific groups of these sensors, thus meeting the requirements for the number of sensors and the sampling frequency.

[0077]

[0078] Figure 11 A flowchart of a method for reproducing a signal is shown, which includes two sets of sensor groups to illustrate low and high frequencies in the signal. In Figure 11 the method shown, the groups are evenly distributed along the tow cable (which generally means that the output points are also evenly distributed as these are typically located at the center of each group of sensors). The first group of sensors is specifically designed to attenuate noise in the first frequency band, while the second group of sensors is specifically designed to attenuate noise in a higher second frequency band. These sensors are spaced along the tow cable such that the second group is located between the first groups. Some sensors will belong to both groups simultaneously. The noise attenuation output in the first frequency band is obtained from the sensors in the first group and spatially interpolated to a smaller spacing to correlate with the spacing of the output points of the second group. The noise attenuation output in the second frequency band can be collected for the output points at the smaller spacing for both groups (or only the sensors that form part of the second group). These outputs for the two frequency bands are combined, resulting in an output for the two frequency bands at the smaller spacing of the second group. When obtaining the signal in the first frequency band, only the signal from the larger first group can be used (as this can avoid using noise signals from the second group with fewer sensors). In some cases, the signals from both groups can be used simultaneously.

[0079] Figure 12 is a flow chart of a similar method for signal reproduction. Figure 12 The method shown involves non-uniformly spaced sensor clusters along the streamer. In this case, the output points linked to the first cluster and the second cluster are non-uniformly spaced. This method is similar to Figure 11 method, but it must be used when processing signals from two groups of sensors Figure 10 Method to create a set of noise-attenuated regularly spaced outputs from non-uniformly distributed output points.

[0080] So far, we have only discussed operations in the common source setting. We can also consider the possibility of including non-uniformly sampled source points into the high-dimensional reconstruction problem. It is well known that signal representations in high dimensions are sparser than those in low dimensions, thus improving the quality of reconstruction using CS techniques. This in turn can allow for a sparser sampling of the output points in the first place.

[0081] The non-uniform sampling source points mentioned here can be considered as the coaxial direction (the general direction of the streamer) and the cross direction. Therefore, the coaxial (in-line) direction refers to the direction along the longitudinal direction or parallel to the longest side of the streamer. The cross-line direction refers to the direction perpendicular to the coaxial direction, or the direction generally parallel to the shortest side of the streamer. Sensors and / or output points may be non-uniformly distributed along the coaxial direction of the streamer and along the cross direction of the streamer, or may have a varying density. In addition, streamers with uneven spacing in the cross direction can also be considered, which increases the sparsity of the signal in the high-dimensional signal reconstruction method using the compressed sensing method. The same non-uniform distribution of output points and sensors can be applied to the cross direction and or instead of the coaxial direction. In order to extend the sensor coverage in the cross direction and along each streamer, several streamers need to be towed side by side away from the rear of the ship, or the streamer section needs to have a certain thickness to allow the sensor to be installed on a specific streamer in a 2D rather than a 1D configuration.

[0082] In one embodiment, the sensor density is highest around the output point, and the sensor density decreases between the output points. In another embodiment, there is no clustering around the output point. In these examples, the distribution of sensors is uneven.

[0083] Fewer output points (tracks) will enable cheaper compressed sensing algorithms. Although these methods are usually computationally expensive; their use will be more affordable given the small number of output tracks that need to be processed compared to the number of input tracks.

[0084] We use the term "cluster" in a general sense, as there can be sensors between these positions. This approach can find applications in streamer, seabed, node, borehole, and land seismic data acquisition scenarios.

[0085] Typically, a streamer will consist of multiple streamer segments. In the case of only one streamer segment, the streamer segment can be referred to as a streamer.

[0086] The arrays (if used) can all be of the same type or vary depending on the position on the streamer.

[0087] The non-uniform distribution of such sensors as already discussed can be applied to the channels and analog arrays of the sensors. The channels, sensors, and analog arrays of the sensors can be collectively referred to as the measurement body. For the present invention, any of the described measurement bodies can be used.

Claims

1. An ocean seismic streamer, comprising at least one streamer segment, said at least one streamer segment including a plurality of sensors and a plurality of output points, data from a set of sensors being fed to said output points, wherein, the density of said sensors is highest around said output points and lower between said output points; said output points are positioned at non-uniform intervals.

2. The ocean seismic streamer according to claim 1, wherein said sensors are concentrated in positions closer to said output points to attenuate short coherent length noise in the seismic data.

3. The ocean seismic streamer according to any one of claims 1 and 2, wherein the distribution of said sensors at said output points is regular.

4. The ocean seismic streamer according to any one of claims 1 and 2, wherein the distribution of sensors at said output points is irregular.

5. The ocean seismic streamer according to claim 4, wherein said output points are distributed along said streamer segment at an average spacing higher than the spacing required by the Nyquist sampling criterion.

6. The ocean seismic streamer according to claim 4, wherein compressive sensing is used to reconstruct a set of outputs regularly spaced along said streamer from the signals at said output points, wherein the number of said regularly spaced outputs is greater than the number of said output points.

7. The ocean seismic streamer according to any one of claims 1 to 6, wherein the sensors employed are pressure sensors, particle motion sensors or a combination of pressure sensors and particle motion sensors.

8. A method for optimizing the distribution of a plurality of sensors and a plurality of output points along a streamer, data from a set of sensors being fed to said output points, said method comprising: determining the positions of said output points along said streamer; and using the determined positions of said output points, configuring a plurality of sensor groups to feed data to each output point and arranging said plurality of sensor groups along said streamer, wherein the positions of said sensors are determined such that the density of said sensors is highest around the output points and lowest between the output points, wherein, the step of determining the positions of said output points along said streamer comprises: arranging said output points at non-uniform intervals along said streamer.

9. The method according to claim 8, wherein said method comprises using a compressive sensing algorithm to reconstruct a set of outputs regularly spaced along said streamer from the signals at said output points, wherein the number of said regularly spaced outputs is greater than the number of said output points.

10. The method according to any one of claims 8 and 9, comprising attenuating short coherent length noise in the data by concentrating said sensors in positions closer to said output points.

Images