Signal Generation

[wbenoit26]

• Parameter distributions are specified using bilby. The PriorDict we use matches the distributions used by the R&P group for O3 injections.
• Waveform polarizations are generated using bilby's WaveformGenerator
• Interferometer responses to polarizations are computed in ml4gw. We test to ensure that our results match results from bilby.
• The same set of parameter distributions is used for both training and testing waveforms
• We generate waveforms that are 8 seconds long. Our smallest mass combination, 5-5 solar masses, takes ~5.5 seconds to merge once the system reaches a frequency of 32 Hz, which is the highpass frequency we use during whitening (see Training Dataset and Augmentations #473 and Testing Dataset #474)
• 100,000 polarizations are generated for training, though we effectively show our network many times this number, see Training Dataset and Augmentations #473. See /home/ethan.marx/aframe/methods-paper-data/train/signals.h5 for an example training signal file.
• For testing, we generate enough waveforms to fill all the timeslides of our testing background with the specified spacing, see Testing Dataset #474. This amounts to 1,347,512 waveforms. For an example of a testing waveform file, see /home/ethan.marx/aframe/methods-paper-data/test/waveforms.h5