Design notes

These pages record why parts of Deepwave are built the way they are: the constraint that forced a decision, the alternatives that were considered and rejected, and the derivation behind a guarantee that the API states in one line. They are developer reference, not user documentation – nothing here is needed to use Deepwave, and none of it describes API you call. If you are looking for how to use a feature, see the user guide and the examples instead.

The audience is anyone maintaining or extending Deepwave, including anyone tempted to change one of these decisions: each page is written so that the cost of the rejected alternative is visible before it is re-attempted.

• PropagatorOutput: DataParallel and pytrees
• Segmented propagation and whole-record resampling
• Normalised cross-correlation: numerical conditioning