Concepts and wording

Datamodel

The following overview summarizes the underlying data processing model of this framework:

• Noise cancellation methods are called filters to have a shorter name

• The goal of the filters is to make an as good as possible prediction of the disturbance channel based on the target and witness signals.

• For that, the ideal filter is calculated from a conditioning dataset (comparable to a training dataset in machine learning) and then applied to a evaluation dataset (comparable to a test dataset) to analyze its performance. The conditioning and evaluation datasets are independent of each other, i.e. they contain different realizations of the signals.

• A dataset defines a set of signals that can be used to test filtering techniques

  • Each dataset contains one target channel, one or multiple witness channels and optionally one GW signals channel

  • The GW signal or short signal channel contains the useful signal that should be recovered through cancellation

  • In a real-world application, only the witness and target channels would be available. The exact realization of the GW signal and disturbance channels would be unknown

  • To evaluate the performance on simulated data, the GW signal channel can be added. The disturbance channel is implicitly defined as the difference between target and GW signal

• The residual signal is the most relevant indicator of the cancellation performance. It indicates the difference between the prediction and the targeted disturbance

Concepts

• Hashing is used to prevent unnecessary rerunning of calculations

  • All objects that are part of an evaluation run configuration run provide a comparable hash that is a unique, comparable indicator of their configuration and version

  • These hashes are e.g. used in filenames to determine if a result for a given calcuation is already available

  • The python __hash__ interface is randomized for every session of the python interpreter for security reasons. It is therefore not usable here.