Match
Match is used to fine-tune a noise model before it's subtracted from the seismic data. One example of an algorithm that produces a noise model is SRME. We seek to better match the noise model to the noise in the data, in order to obtain a more-thorough noise removal.
The danger of Match is that it doesn't know what's noise and what's primary energy. It uses a minimum-energy criterion in its objective function. Therefore, Match needs to be used conservatively in order to minimise primary energy being introduced into the noise model.
Match allows for the simultaneous subtraction of one or more noise models from the seismic data. A single output filtered noise model is created from the input noise models. Note: when multiple input noise models are used the runtime will be significantly increased compared to using a single input noise model.
- In the Control Panel, open the Process tab.
- At the tab header, click the Add icon and select New Process.
- Scroll down and double-click on Match.
- Type a name for the process and click OK.
Configuring Match
-
Volume:
- Reference: This is the data to which the noise volumes are matched.
- Noise: The noise models that will be matched to the reference volume.
More than one noise volume can be entered. The matching of the noise volumes is simultaneous.
For a workflow process, the workflow input can be used for either the reference volume or the first noise volume. This is specified with the checkbox that appears while editing a workflow.
- Pattern Matching: Uses principal component analysis (PCA) to learn key patterns, each of which are independently matched to produce the output. Just like the conventional match, the process operates in overlapping windows of data defined by spatial and temporal window sizes. Furthermore, each window is divided into overlapping patches based on the patch percentage to form an over-complete dictionary matrix for the determination of key patterns. Pattern matching only accepts one noise model.
- Spatial window length (# of traces): More traces in the design window result in a more-stable system.
- Temporal window length (ms): A longer design window results in a more-stable and less-aggressive subtraction.
- Filter length (ms): A longer filter will result in a more aggressive subtraction and hence potentially the removal of (more) primary energy.
- Zero time in the filter (ms): If the data has been zero-phased, zero time needs to be in the middle of the filter. Causal data is better served with a one-sided or near-one-sided filter, as a symmetric filter is unnecessary and can result in noise being introduced above the predicted multiples.
- Horizontal filter size (traces): Must be an odd number. A wider filter will result in a more aggressive subtraction and hence potentially the removal of (more) primary energy. For a trace-by-trace operation, enter 1.
- Maximum filter coefficient: If the maximum filter coefficient is zero, it has no effect. If it's greater than zero, then filter coefficients are limited to this value. If it's less than zero, and a filter coefficient is larger than the absolute value of this limit, then that filter coefficient is set to zero.
- Whitening (%): The smaller the percentage, the more aggressive the subtraction.
- Output type: Choose between the matched noise model, or the primary volume with the matched noise model already subtracted.
- Output filter volume: Outputs a second volume, suffixed with '_filter', containing the filter coefficients, suitable for convolving over the noise volume. This is calculated for the first noise volume only using a single temporal window that covers the full trace.
- Allow missing traces: If unchecked, it is an error for a trace to exist in either the noise or primary volume but not the other.
- Match amplitude/lag/phase: If unchecked, the system will attempt to find a simpler filter that matches only the selected attributes. When constructing this simpler filter, parameters such as 'filter length', 'zero time in filter', 'maximum filter coefficient' and 'whitening' will have no effect.