Representation, Isolation & Detection

The same time-series problem — different physical domains
Acoustic Time-Series Representation
Array geometry · beamforming · DTFT / windowed spectrum
  • Hydrophone geometry and sound speed bound observability
  • Multipath and thermoclines reshape energy over time
  • Windowing, filtering, and DTFT define what survives
  • These choices are lossy and irreversible
  • Stabilization must occur before any inference
The Same Problem I Solved in Text
Text treated as time-series · semantic windowing
  • Text processed as a time-ordered signal
  • Sliding windows instead of whole documents
  • Post-processing on windows: embeddings, tags, clustering
  • Meaning emerges through persistence, not single passes
  • This code predates this sonar work
Ambiguity Is the Baseline
Noise floor · impulses · regime transitions
  • Most acoustic windows contain no actionable signal
  • Different regimes require different detectors
  • False positives are expected
  • Early confidence hides structure
Detection as Time-Series Navigation
Window overlap · FrameSessions · PilotWave persistence
  • Detection treated as hypothesis generation
  • Multiple isolators operate on the same windows
  • Context persists across time via FrameSessions
  • PilotWave reinforces weak but coherent structure
  • Confirmation is delayed and explicitly released
Tools reused and adapted: sliding windows · post-window processing · embeddings · tagging · clustering · DTFT / windowed FFT · beamforming · band-pass filtering · ImpulseIsolator (TKEO) · IsolationForestProcessor · DcmtIsolator · FrameSessions · PilotWave