|
<< Click to Display Table of Contents >> Navigation: Multiples > Radon multiple attenuation |
Radon multiple attenuation suppresses water-bottom and interbed multiples in NMO-corrected CMP gathers using a high-resolution parabolic Radon (tau-p) transform. After applying stack-velocity NMO correction, primary reflections are flattened and cluster near zero residual moveout (p = 0), while multiple reflections retain residual curvature and map to larger p-values in the tau-p domain. The module exploits this moveout difference: it transforms each gather into the tau-p domain using a least-squares high-resolution method, separates primaries (small |p|, up to P mid) from multiples (large |p|, above P mid), models the multiple wavefield, and subtracts it from the input to recover the primary-only gather.
The transform is applied in time-varying windows, each with its own reference offset, allowing the moveout separation to adapt to the depth-dependent velocity field. Because the method is data-driven, it works without a water-depth model. Input gathers must be NMO-corrected, CMP-offset sorted, and have all static corrections applied before using this module.
The primary data connection carrying the seismic dataset. This item transports the SEG-Y file handle, trace headers, and geometry objects that are passed through to the output unchanged alongside the demultipled gather.
The NMO-corrected, CMP-offset sorted seismic gather to be demultipled. The gather must have all static corrections applied and must contain at least four live traces. Traces must be sorted in ascending offset order.
Minimum curvature parameter (slowness) of the Radon transform, in seconds. Default: -0.5 s. This value defines the most negative parabolic curvature modelled, corresponding to the largest over-correction of primaries at far offsets after NMO. Start with a value around -0.3 s and allow the primaries to show a slight over-correction. Making P min more negative increases computation time.
The curvature boundary separating primaries from multiples, in seconds. Default: 0 s. Radon components with curvature between P min and P mid are treated as primaries and are preserved in the output; components with curvature between P mid and P max are treated as multiples and subtracted. Set this value to match the residual moveout of primaries — typically near 0 for well-corrected gathers. Moving P mid slightly positive (e.g., 0.02–0.05 s) preserves slightly under-corrected primaries.
Maximum curvature modelled, in seconds. Default: 1 s. This represents the largest under-correction expected for multiples at far offsets. Choose this value based on the maximum multiple residual moveout observed in the data; begin with approximately 1 s and adjust. Computation time increases with the square of the number of p-values, so keep P max as small as the data allows.
Curvature sampling interval (step between parabolas) at the reference offset, in seconds. Default: 0.008 s. This parameter controls the resolution and computation time of the Radon decomposition. A value of 8–32 ms is usually sufficient; smaller values improve resolution but increase computation time. Delta P should be approximately equal to the dominant signal period (1/dominant frequency) of the data.
Width of the linear taper zone applied around the P mid boundary in the Radon domain, in seconds. Default: 0.05 s. A taper is applied symmetrically around P mid so that the transition from primary to multiple energy is smooth rather than abrupt. Increase this value if leakage artefacts appear at the primary-multiple boundary; decrease it for sharper separation when the moveout difference is large.
Length of the time-domain taper applied at the boundaries between adjacent time-varying windows, in seconds. Default: 0.200 s. The module processes the gather in time segments defined by the TimeOffset table; this taper blends the contributions from adjacent windows to avoid discontinuities at segment boundaries. Use longer tapers when segment boundaries are close together.
Container group defining a time taper applied to the subtracted multiple model. This allows the multiple attenuation effect to be ramped on and off over a user-defined time interval, blending smoothly between the demultipled gather (inside the window) and the original input (outside the window). This is useful when multiples affect only a specific time zone in the record.
Start time of the attenuation window ramp-on, in seconds. Default: 0 s. Below this time the original gather is output without multiple subtraction.
End time of the attenuation window ramp-on, in seconds. Default: 0.1 s. Between T1 and T2 the multiple subtraction is linearly ramped from zero to full strength.
Start time of the attenuation window ramp-off, in seconds. Default: 9.9 s. Between T2 and T3 the full multiple subtraction is applied. Set T3 and T4 to match the bottom of the zone containing multiples.
End time of the attenuation window ramp-off, in seconds. Default: 10 s. Between T3 and T4 the multiple subtraction is linearly ramped from full strength to zero. Below T4 the original gather is restored.
A table of time-offset pairs that defines the reference offset used for the parabolic transform in each time segment. Add one row per time zone; each row specifies a Time (segment boundary, in seconds) and a Reference offset (m) for that segment. The reference offset should be close to the maximum source-receiver offset of the input gather at that time. Using multiple segments with depth-appropriate reference offsets improves separation quality for surveys with large offset ranges. Default: one segment at Time = 0 with Reference offset = 1550 m.
Lower frequency limit of the Radon transform pass-band, in Hz. Default: 0 Hz. The high-resolution least-squares Radon computation is performed only within the frequency band defined by Frequency min and Frequency max. Set these values to match the usable signal bandwidth of your data to avoid processing noise-dominated frequencies.
Upper frequency limit of the Radon transform pass-band, in Hz. Default: 100 Hz. Set this to the highest usable signal frequency in the data. Frequencies above this value pass through unchanged.
Stabilisation factor for the least-squares inversion used in the Radon transform, as a percentage of the maximum spectral power. Default: 0.1 %. A small pre-whitening factor (0.1–1 %) stabilises the inversion in the presence of noise without significantly reducing resolution. Increase this value if the output contains numerical artefacts or instabilities, particularly when the gather has few live traces or high noise levels.
When enabled, applies an automatic gain control (AGC) equalisation to the gather before the Radon transform, then reverses the gain after the multiple subtraction. Default: enabled. AGC balances the amplitude across traces and times so that deep, weak reflections receive the same weight as shallow, strong ones during the transform. Disable this option only when amplitude preservation is critical and the data already has balanced amplitudes.
Length of the AGC operator window, in seconds. Default: 0.5 s. A longer window produces a smoother, gentler gain variation; a shorter window is more responsive to local amplitude changes. Only used when Use AGC is enabled.
When enabled, preserves the mute pattern already applied to the input gather: any sample that is zeroed in the input remains zeroed in the output, preventing the Radon reconstruction from filling in muted regions. Default: enabled. Enable this option when the gather has been offset-muted before multiple attenuation, to avoid reintroducing energy in the muted zone.
When enabled, the module automatically connects to adjacent modules in the processing flow.
Controls how corrupted (NaN) samples are handled. Fix replaces bad values before processing. Notify reports the problem and stops. Continue proceeds without correction.
Number of CPU threads used for parallel Radon computation. Set to the number of available physical CPU cores for best performance.
When enabled, this module is bypassed and the input data passes through to the output unmodified. Use this option to temporarily disable multiple attenuation without removing the module from the flow.
The output data connection carrying the processed dataset with the same SEG-Y file handle, trace headers, and geometry objects as the input.
The demultipled CMP gather with primary reflections preserved and multiple energy attenuated. The gather retains the same geometry and sample rate as the input. Connect this to the next processing step, typically NMO inverse correction followed by stacking.