|
<< Click to Display Table of Contents >> Navigation: Noise attenuation > Wave separation (HRWS) |
The Wave separation (HRWS) module separates coherent seismic wavefields using the High-Resolution Wave Separation (HRWS) algorithm based on the Berk antileakage Fourier transform. Operating on multi-gather data volumes, the algorithm decomposes each local time-space window into its frequency-wavenumber components using an iterative antileakage FFT that suppresses spectral leakage from irregular spatial sampling. The wavefield of interest is then extracted in the frequency-wavenumber domain and transformed back to the time-space domain. This approach is particularly effective for separating upgoing and downgoing wavefields in VSP data, for ground-roll attenuation, and for up/down wavefield separation in ocean-bottom seismic data.
The module supports distributed multi-node execution for processing large 3D volumes. It reads input data from a SEG-Y file and writes the separated wavefield output to a new SEG-Y file. An optional velocity model can be supplied to enable slope-constrained or NMO-domain separation.
The sorted index item that controls the order in which input data is processed. This is the standard sequence control input.
A handle to the SEG-Y file or data store containing the input seismic data to be wave-separated. The module reads data in spatial blocks of the size defined by the Horizontal sliding window parameter.
The sorted header index that defines the gather ordering for the input data. This is used to assemble the spatial windows for the HRWS algorithm.
An optional RMS velocity model used when NMO-domain separation is enabled (Use NMO option). When connected, the input data can be NMO-corrected into the zero-offset domain before applying the Berk transform, enabling better separation of reflection energy from coherent noise with conflicting moveout. Leave unconnected if not using NMO mode.
The path and filename for the output SEG-Y file that will receive the separated wavefield. The module writes the extracted wavefield (or the enhanced data if Output is enhanced data is enabled) to this file.
When enabled (default: true), the module writes output traces directly to the output file as each spatial block is processed, without buffering the entire volume in memory. Disable this option only if the processing order requires re-sorting the output before writing, which may be necessary in some distributed processing configurations.
The container grouping all HRWS algorithm parameters.
The number of traces included in each spatial processing window. Default: 40 traces. A larger window captures more spatial information and improves the frequency-wavenumber resolution of the Berk transform, but increases memory usage and computation time. The window slides along the spatial dimension with 50% overlap to avoid edge artefacts. Typical values range from 20 to 80 traces depending on the lateral coherence of the target wavefields.
The length of the time processing window, in seconds. Default: 0.050 s (50 ms). Each time window is transformed independently. A shorter window provides better temporal resolution for non-stationary data (e.g. data with rapidly changing noise character), while a longer window improves frequency resolution and stability. Typical values are 50–200 ms.
The minimum frequency (in Hz) included in the separation. Default: 0 Hz. The Berk transform and wave separation are applied only within the frequency band defined by Min frequency and Max frequency. Set Min frequency to exclude very low frequencies where the separation becomes unstable or where no signal of interest is expected.
The maximum frequency (in Hz) included in the separation. Default: 800 Hz. Set this to the Nyquist frequency or to the highest signal frequency in the data. Limiting the bandwidth to the signal band reduces computation time without affecting the result.
The number of dominant eigen values (plane-wave components) to preserve in the Berk antileakage decomposition. Default: 1. This controls the rank of the signal model — a value of 1 separates a single dominant dip event (e.g. the primary downgoing wave), while higher values allow separation of multiple overlapping events. Increase this value if the target wavefield contains several distinct dip components that cannot be separated with a single component, but note that too high a value may allow noise to leak into the output.
Controls how the frequency-wavenumber decomposition is applied. All frequency applies a single Berk transform across the full frequency band within each window. Each frequency applies the decomposition independently at each frequency slice, which can improve separation quality when different frequencies have substantially different moveout characteristics, but increases computation time.
When enabled, the output contains the enhanced (signal) component of the wavefield — the portion retained after separation. When disabled (default), the output contains the separated (removed) wavefield, which is the coherent noise or the secondary wavefield that was subtracted from the input. Enable this option when you want to output the cleaned signal rather than the extracted noise.
When enabled, the wavefield separation is constrained to a specific dip range defined by the From slope and To slope parameters. This prevents energy outside the expected dip range from being assigned to the extracted wavefield and improves separation when the signal and noise have overlapping frequency content but different apparent dips.
The minimum apparent dip (slope) of the target wavefield, used when Use constrains is enabled. Energy with slopes below this value will not be attributed to the extracted wavefield.
The maximum apparent dip (slope) of the target wavefield, used when Use constrains is enabled. Energy with slopes above this value will not be attributed to the extracted wavefield.
When enabled, traces whose energy falls outside the defined slope constraints are excluded from the output rather than being passed through. Use this option to blank traces where no valid wavefield separation can be performed.
An optional parameter group that controls datum-level time shifting applied before the Berk transform. Shifting the data to a common flat datum before separation can improve performance on data with significant topography, where the apparent dip of flat events would otherwise vary across the processing window.
When enabled, the module applies a time shift to flatten data to the specified datum before the separation, and removes the shift from the output. Enable this on land data with pronounced topography.
The datum elevation (in metres) to which data is shifted when Use shift to datum is enabled. Set this to the flat processing datum used in the rest of the processing flow.
The near-surface velocity (in m/s) used to compute the static time shift when shifting data to the datum. Set this to the replacement velocity or weathering velocity for the survey area.
An optional parameter group controlling NMO correction applied before the Berk separation. Applying NMO correction flattens reflection events across the gather so that they appear as horizontal events in the time-offset plane, making them easier to separate from dipping noise events.
When enabled, the input data is NMO-corrected using the connected velocity model before applying the Berk transform, and NMO is removed from the output. This improves separation quality when noise events have moveout that conflicts with the reflections. Requires the Vrms model input to be connected.
Selects whether the separation is performed in the CMP (common midpoint) domain or the CSP (common shot point) domain. CMP domain is the standard choice for reflected wave separation. CSP domain may be preferred for ground-roll or guided-wave attenuation.
Specifies whether the input velocity model is referenced to the processing datum (true) or to the topographic surface (false). Set to match how the velocity model was built.
When enabled, applies a monotonicity constraint in the reverse direction to the velocity-based time shift computation. This prevents physically unrealistic negative velocity gradients from producing unstable time shifts.
When enabled, applies a monotonicity constraint in the forward direction to the velocity-based time shift computation, ensuring the velocity field increases monotonically with depth where required by the physical model.
Controls the degree of NMO stretch muting applied before the separation. High NMO stretch at near offsets at shallow times can distort the amplitude spectrum and degrade the Berk transform. Set this to the same NMO stretch threshold used in the NMO correction module.
Controls the amount of spatial smoothing applied to the velocity model along the offset axis before it is used for the NMO-based datum shift. Larger values produce a smoother velocity field and reduce spatial discontinuities in the output.
An optional de-spiking post-processing step applied to the separated output to remove high-amplitude spike artefacts that can arise from the Berk transform at locations with very irregular spatial sampling or poor wavefield separation.
When enabled, the de-spiking filter is applied to the output after wave separation. This is recommended when the input data has irregular spatial sampling or when the output shows isolated high-amplitude artefacts.
The number of traces in the spatial window used by the de-spiking filter to estimate the local signal amplitude for spike detection.
The length of the time window (in seconds) used by the de-spiking filter for local amplitude estimation.
The step size of the sliding time window (in seconds) used by the de-spiking filter. A smaller step gives more overlap between consecutive windows and produces a smoother result.
The amplitude threshold used to classify a sample as a spike. Samples whose amplitude exceeds this multiple of the local RMS amplitude are replaced by the local median. Lower values (e.g. 2–3) are more aggressive and remove more artefacts; higher values are more conservative.
An optional parameter group that limits processing to a subset of bins in the input volume.
The index of the first bin to process. Default: -1 (process from the first available bin). Set to a positive value to start processing partway through the dataset, for example when reprocessing a sub-area or testing parameters on a subset of the data.
The index of the last bin to process. Default: -1 (process to the last available bin). Set together with First bin to define a processing sub-range within the full dataset.