|
<< Click to Display Table of Contents >> Navigation: Multiples > HR Radon multiple attenuation |
HR Radon Multiple Attenuation applies a high-resolution parabolic Radon transform to attenuate multiple reflections in offset gathers. The method transforms each gather into the Radon (tau-p) domain using an iterative Gauss-Seidel solver, identifies energy that follows parabolic moveout characteristic of multiples, and then subtracts that multiple energy from the original gather to recover the primary reflections.
Unlike conventional Radon demultiple, the high-resolution (HR) approach uses sparse inversion to sharpen the Radon spectrum, making it possible to separate primaries and multiples that are close in moveout. The output gather contains primary energy with multiples suppressed. This module is most effective on common-midpoint (CMP) or common-offset gathers where multiples and primaries have a discernible difference in normal moveout curvature.
The module requires that traces in the input gather be sorted in ascending absolute offset order. At least four traces must be present in each gather for the transform to be applied. An optional AGC pre-conditioning step can be applied before the Radon transform to balance amplitude variations across offsets and improve the quality of the multiple model.
The seismic dataset to be processed. This defines the sequence of gathers that the module iterates over, reading each gather in turn and passing it to the Radon multiple attenuation algorithm.
The individual offset gather on which the Radon transform is applied. Traces must be sorted in ascending order of absolute offset. The gather should contain at least four traces; gathers with fewer traces are skipped. Typical inputs are CMP gathers after NMO correction has been applied, with NMO subsequently removed, or pre-NMO CMP gathers where the moveout difference between primaries and multiples is well defined.
The minimum slowness (curvature) parameter of the parabolic Radon panel, expressed in seconds. This value defines the low end of the moveout curvature range that the transform will model. Negative values of P min capture events with reverse or very shallow curvature (near-flat primaries), while values close to zero represent events with little moveout. The default value is -0.05 s. Set P min to be less than the expected curvature of the fastest primary reflections in the gather. Setting this value too close to zero may cause shallow primary energy to be misclassified as multiples and incorrectly removed.
The maximum slowness (curvature) parameter of the parabolic Radon panel, in seconds. This defines the high end of the moveout curvature range, corresponding to the most strongly curved events (typically shallow multiples or water-bottom multiples with large moveout). The default value is 1.0 s. Set P max to be larger than the expected curvature of the deepest or most curved multiples present in the data. Increasing P max extends the range of multiples that can be modeled, but also increases computation time.
The slowness sampling interval of the parabolic Radon transform, in seconds. This controls the density of the Radon panel: a smaller Delta P creates a finer grid of slowness values between P min and P max, which improves resolution and the ability to separate closely spaced primaries and multiples. The default value is 0.008 s. Reducing Delta P improves separation quality but increases processing time proportionally; increasing it speeds up computation at the cost of resolution. A value of 0.004 to 0.008 s is typical for most survey types.
The normalizing offset used in the parabolic moveout equation, in meters. The Radon transform parameterizes moveout as a parabola relative to this reference offset: a trace at the reference offset has zero moveout contribution from the slowness parameter. In practice, this value is typically set to the maximum offset in the gather, or to the maximum offset that contains usable signal. The default value is 1500 m. Setting this value correctly is important: if it is too small, events at far offsets will be distorted; if it is too large, the transform will under-use the available offset range and resolution will degrade. Only traces with absolute offset not exceeding the Reference Offset are included in the transform calculation.
The lower frequency limit of the band over which the Radon transform is applied, in Hz. Energy below this frequency is not included in the multiple model, which prevents the transform from being distorted by very low-frequency noise or DC bias. The default value is 0 Hz (no low-cut applied). Set this to the lowest usable signal frequency in the data, typically between 3 and 8 Hz for marine data or 5 to 15 Hz for land data.
The upper frequency limit of the band over which the Radon transform is applied, in Hz. Frequencies above this value are excluded from the multiple model, which avoids processing noise-dominated high-frequency content. The default value is 100 Hz. Set this value to match the highest meaningful signal frequency in your data. Using a value well above the signal bandwidth will include noise in the transform and may degrade the quality of multiple suppression.
The length of the time window used during the iterative selection step of the high-resolution Radon algorithm, in seconds. At each iteration the algorithm identifies and picks energy within windows of this size to build up the sparse Radon model. A smaller window gives finer time resolution and better separation of closely spaced events, while a larger window is more stable in the presence of noise. The default value is 0.01 s (10 ms). This value must be at least two samples wide; if a smaller value is entered it is automatically increased to two samples. Typical values range from 8 to 20 ms depending on the dominant frequency of the data.
The number of Gauss-Seidel iterations used to solve for the high-resolution Radon model. More iterations allow the algorithm to progressively refine the sparse solution, producing a sharper Radon spectrum that better separates primaries from multiples. However, each additional iteration increases computation time. The default value is 10 iterations. For most datasets, 5 to 20 iterations are sufficient. Increasing this value beyond 30 typically yields diminishing returns in solution quality. If processing time is critical, start with a low value and increase it until the quality of multiple suppression no longer improves noticeably.
A convergence damping factor that controls the step size of each Gauss-Seidel iteration, dimensionless (range 0 to 1). Smaller values of alfa produce more conservative, stable convergence with less risk of oscillation, while larger values allow the solver to take bigger steps toward the solution and may converge faster but can become unstable. The default value is 0.01. For well-conditioned data with good signal-to-noise ratio, this default is generally suitable. Increase alfa cautiously if convergence is too slow; reduce it if the result shows instability or over-subtraction artifacts.
A regularization weight that controls the sparsity of the Radon solution, dimensionless (range 0 to 1). Higher values of lamda impose stronger sparsity constraints, forcing the algorithm to represent the data with fewer, more focused Radon coefficients. This sharpens the Radon spectrum and can improve the separation between primaries and multiples when their moveout difference is small. Lower values relax the sparsity constraint, which may be appropriate for noisier data. The default value is 0.3. A value in the range 0.1 to 0.5 covers most practical cases. Increasing lamda too aggressively can lead to over-sparsification and loss of weaker primary reflections in the output.
When enabled, an Automatic Gain Control (AGC) is applied to the input gather before computing the Radon transform. AGC equalizes amplitude variations across the gather so that the Radon transform is not dominated by high-amplitude events at short offsets or by amplitude decay with time. This improves the accuracy of the multiple model, particularly on data with strong amplitude-versus-offset (AVO) variations or significant spherical divergence. The default is enabled. The AGC is applied only to the data used in computing the Radon model; the amplitude of the output gather is restored from the original input. Disable this option if the input data has already been amplitude-balanced or if preserving true amplitudes throughout the process is critical.
The half-window length used for the AGC computation, in seconds. The AGC normalizes each sample by the RMS amplitude of the data within a window of this size centred on that sample. A longer window produces a smoother gain function that responds to broad amplitude trends; a shorter window produces a more aggressively equalized trace. The default value is 0.5 s, with a minimum allowed value of 0.1 s. This parameter is active only when Use AGC is enabled. Choose a window length that is longer than the dominant period of the signal but short enough to track genuine amplitude variations with depth.