next up previous contents
Next: SUSHAPE - Wiener shaping Up: 1D Filtering Operations Previous: SUCONV, SUXCOR - convolution,   Contents

SUPEF - Wiener predictive error filtering

The prediction error filtering method, also known as Wiener filtering, is the principle process of traditional Wiener-Levinson deconvolution. The reason that this subsection is not headed ``deconvolution'' is because there are two additional issues that have to be addressed before prediction error filtering can be used for effective deconvolution. These issues are preprocessing of the data, and postprocess filtering. Indeed, much feedback has come back claiming that supef doesn't work properly, when in fact, it simply being used improperly. (This is our fault, for not supplying sufficient documentation.)

Using supef, itself, generally requires that the value of maxlag be set. This may be determined by first establishing the size of the wavelet being spiked, through application suacor.

As the preprocessing step, you will probably need to use sugain to remove any decay in amplitudes with time that may result from geometric spreading.

As a postprocessing step, you need to remove any increase in frequency content which has occurred due to the whitening effect of the prediction error filter. The demos in the demos/Deconvolution directory demonstrate the functioning of the program. However, these examples are a bit unrealistic for field data. For example, in the demos, the data are spiked, and then reverberations are removed via a cascaded of supef calls with maxlag=.04 and minlag=.05 maxlag=.16, respectively. However on field data, we would probably only use a single pass of the filter to remove reverberations, or to spike arrivals.

The prediction error filter has a spectral whitening effect, which is likely to put high frequencies in the data that were not there originally. These must be filtered out. It is a good idea to assume that there is a loss of frequency information, and design the parameters for the postprocessing filtering, using sufilter to slightly reduce the frequency content from its original values.

Also, an added feature of the Release 32 version of supef is the mixing parameter, which permits the user to apply a weighted moving average to the autocorrelations that are computed as part of the prediction error filter computation. This can provide additional stability to the operation.


next up previous contents
Next: SUSHAPE - Wiener shaping Up: 1D Filtering Operations Previous: SUCONV, SUXCOR - convolution,   Contents
John Stockwell 2007-04-10