|
<< Click to Display Table of Contents >> Navigation: Geometry > FK Interpolation |
FK Interpolation reconstructs missing or irregularly sampled traces within a seismic gather by working in the frequency-wavenumber (F-K) domain. The module transforms the input gather into the F-K domain and uses a high-resolution, anti-leakage eigenvalue decomposition (Berkout method) to estimate the true wavefield and then insert new traces between the existing ones. This approach handles spatially aliased data more robustly than simple trace-by-trace interpolation because it leverages the coherency of the entire gather simultaneously.
Use this module to regularize under-sampled pre-stack gathers before NMO, migration, or AVO analysis, or to fill in dead traces caused by receiver dropout. The frequency band and the number of eigenvalues control the trade-off between reconstruction fidelity and noise amplification. Note: this module is deprecated. It is retained for backwards compatibility with existing workflows. For new projects, consider using the 5D interpolation modules available in g-Platform.
Connect the seismic data source here — typically a SEG-Y file handle or an internal data item containing the pre-stack gathers to be interpolated. This item provides both the trace data and the geometry (trace headers) that the module reads to set up the F-K grid.
The current gather to be processed, delivered one gather at a time from the connected data source. Each gather (for example, a CMP or common-offset gather) is transformed to the F-K domain independently. The traces within the gather should be sorted in a consistent spatial order (e.g., by offset or receiver position) so that the F-K decomposition correctly represents the lateral wavefield sampling.
The lower boundary of the frequency band (in Hz) over which the F-K interpolation is performed. Default: 0 Hz. Frequency components below this value are excluded from the reconstruction. Set this to a value above the dominant low-frequency noise (for example, ground-roll or swell noise) to prevent the interpolator from fitting coherent noise instead of signal. For typical reflection seismic data a value of 5–10 Hz is common.
The upper boundary of the frequency band (in Hz) over which the F-K interpolation is performed. Default: 125 Hz. Set this to the highest signal frequency present in the data — typically the Nyquist frequency or just below it. Frequencies above this value are not reconstructed and are zeroed out in the output. Reducing this value when the useful signal bandwidth is limited can improve stability and suppress high-frequency noise in the interpolated traces.
An integer multiplier that controls how many new traces are inserted between each pair of existing traces. Default: 1 (no interpolation — the gather passes through unchanged). A value of 2 doubles the number of traces (one new trace inserted between each pair), a value of 3 triples it, and so on. The minimum value is 1. Use higher values when the original trace spacing is too coarse to satisfy the spatial anti-aliasing condition for the dominant dips in the data. Be aware that very large factors require more computation time and may amplify noise if the original fold is insufficient to constrain the eigenvalue decomposition.
The maximum number of eigenvalues to retain during the F-K domain matrix decomposition (Berkout anti-leakage method). Default: 1000, minimum: 1. In practice the effective rank of the data matrix is determined by the number of coherent events in the gather. A large value (the default) allows the algorithm to model a complex wavefield with many dipping events. Reducing this value acts as a rank-truncation filter: it retains only the strongest coherent components and suppresses weaker signals and noise. If the data are simple (few events, good signal-to-noise ratio), setting this to a smaller number — for example 10 to 50 — may produce cleaner interpolation and reduce computation time.
A dimensionless noise regularization factor (prewhitening level) applied during the eigenvalue decomposition to stabilize the inversion. Default: 0.01. This parameter sets the assumed noise floor relative to the signal energy. A higher value (for example 0.05–0.1) adds more damping to the solution, which suppresses noise amplification but may smooth out weak reflectors. A lower value (for example 0.001) trusts the data more and produces sharper interpolation, but risks amplifying noise in the reconstructed traces. For most field datasets a value between 0.01 and 0.05 gives a good balance. This parameter is equivalent to a Tikhonov regularization weight in the least-squares sense.