Time-Frequency Denoise (TFDN)

Description Suppress anomalous noise from traces when compared with the surrounding windowed traces.
Module(s) Image Gather Processes
Requirements Volume
Related
Works with Gathers


The premise of this noise removal program is to suppress anomalous noise from traces when compared with the surrounding windowed traces. Anomalous noise is identified in narrow frequency bands. Therefore, choosing a sort order where noise is anomalous is important. Often this program will need to be run multiple times in multiple domains. 

It first transforms data to the frequency domain via short-window Fourier transforms. The magnitude of individual frequencies are compared to those of neighbouring traces within the window. Various statistical methods can be used for this comparison (median, median-mean, lower quartile). 

If the amplitudes of the trace under consideration exceed the threshold (the statistic multiplied by some factor), it's replaced with the statistic. 

This process handles NaN values in the input by replacing all such values with zero before it does any other processing. 

Note: The input to this process should always have some type of gain applied (e.g. spherical divergence correction). Always choose a domain where the noise is random and anomalous. When running it on CMP or Shot records, normal moveout can be applied.

  1. In the Control Panel, open the Process tab.
  2. At the tab header, click the Add icon and select New Process.
  3. Scroll down and double-click on Time-Frequency Denoise (TFDN).
  4. Type a name for the process and click OK.

Configuring Time-Frequency Denoise (TFDN)

  1. Volume: Select the input volume.
  2. Start Time: 
    • Constant: Constant TWT in ms
    • Constant and velocity: Constant TWT in ms; a shift in start time is calculated for each trace based on the trace's offset and the specified velocity (in m/s). For both constant modes, a blank constant value is interpreted to mean the start time of the trace.
    • Trace header: For each record, the specified header word on the first trace is used plus an optional bias (both in ms).
    • Trace header and velocity: For each record, the specified header word on the first trace is used plus an optional bias (both in ms). A further shift in start time is calculated for each trace based on the trace's offset and the specified velocity. For both trace header modes, a blank optional bias is interpreted to be a bias of 0 ms. 
  3. End Time of noise removal (ms): The time below which there will be no noise removal. If blank, noise removal continues to the bottom of the trace.
  4. Start time taper length (ms): The length of the taper to apply at the start of the noise removal.
  5. End time taper length (ms): The length of the taper to apply at the end of the noise removal.
  1. Type of thresholding: The statistic used to estimate the typical amplitude.
  2. Number of adjacent traces over which to calculate thresholding: The total number of adjacent traces over which to calculate the typical amplitude statistic.
    • Note: Whether or not to use an absolute threshold. If this is checked, the averaging method does not determine the threshold, but amplitudes exceeding both the average and the absolute threshold are damped towards the average.
  3. Threshold [scalar] at start/end time: The actual threshold (or scalar) used for a given sample is calculated by linear interpolation between these values. The amplitude of a frequency sub-band will be moderated if it's greater than the interpolated threshold, if thresholding is absolute, or the scalar multiplied by the estimated amplitude statistic otherwise.
  4. Window length to be transformed to the frequency domain (ms)      
  5. Time window moveup (%): The vertical distance to move the filter window, as a percent of the window lengths. Larger values will significantly improve runtimes!
  6. Trace sort order: Optionally randomise the order of traces within each record.
  7. Random seed: Optional random seed allowing the results of random trace ordering to be varied.