pretty obvious algorithm: paint each pulse additively into a time-domain ring buffer, taking into account
the possibility of small time inversions before exporting a prefix of the buffer as completed audio.
could combine all antennas into a single track, or one antenna per track
use signal strength and offset frequency, but ignore noise value
use a windowing function to reduce noise at pulse edges
Allowing users to listen to data files post hoc might help in cases of missing or malfunctioning tags, or reveal problems with the tag finder.