DUG’s Cadzow filtering algorithm is based on the work of Cadzow (1988) and Trickett (2002). It performs matrix-rank reduction on constant-frequency slices for the purpose of random noise suppression. It is a variant of eigenimage filtering which has been extended to a higher number of dimensions.

Rank-reduction noise suppression techniques are used in many fields and have many names, including principal component analysis (PCA) and truncated singular-value decomposition (SVD). The simple reason that rank-reduction methods are useful is that coherent energy will reside in just the first few principal components (or eigenimages) while random noise will be more evenly spread over all eigenimages.

One of the advantages of the Cadzow approach is that it can be applied simultaneously in any number of spatial dimensions. DUG’s implementation can be used in up to four dimensions (time plus up to three spatial dimensions). Application in multi-dimensional space improves the algorithm’s ability to discriminate between coherent energy and noise.

The algorithm works equally well on flat or dipping events, and is exact for noiseless data having a restricted number of dips.