Python tool for aircraft Doppler audio geolocation using historical ADS-B flight data (Source-available)
Python acoustic-geolocationads-baircraft-trackingairplaneaudio-forensicsdoppler-effectdoppler-shiftgeolocationosintsignal-processing
"""Clock pruning checked against archived real flight and acoustic observations."""
import json
from pathlib import Path
import pandas as pd
from dopplergeist.matching import association_candidates
from dopplergeist.tiled_search import _association_clock_window
def real_associations():
path = Path(__file__).parent / "data/reference-cpa-real.json"
recorded = json.loads(path.read_text())
table = pd.DataFrame(recorded["cpa_rows"])
candidates, counts = association_candidates(table, recorded["events"])
return table, recorded["events"], candidates, counts
def scheduled_intervals(candidates, window, sweep_s):
"""Join the same observed association identities after the clock restriction."""
retained = {}
for candidate in candidates:
lower, upper = candidate["recording_start_bounds_s"]
interval = (max(lower - sweep_s, window[0]), min(upper + sweep_s, window[1]))
if interval[0] > interval[1]:
continue
identity = (tuple(candidate["event_indices"]),
tuple((row["flight_id"], row["emission_time"]) for row in candidate["passes"]))
assert identity not in retained
retained[identity] = interval
return retained
def bounded_associations(table, events, window, sweep_s):
center, tolerance = _association_clock_window(window, sweep_s)
return association_candidates(table, events, time_origin=center, time_tolerance_s=tolerance)
def test_real_recording_window_avoids_unrelated_clock_work_without_losing_trials():
table, events, original, original_counts = real_associations()
# Historical source-informed development clock; not independent field truth.
window, sweep_s = (1788898640.0, 1788898660.0), 5.0
bounded, bounded_counts = bounded_associations(table, events, window, sweep_s)
expected = scheduled_intervals(original, window, sweep_s)
assert len(original) == 10549 and len(expected) == 26
assert scheduled_intervals(bounded, window, sweep_s) == expected
assert bounded_counts["work_units"] < original_counts["work_units"] / 10
def test_real_association_clock_endpoints_remain_inclusive():
table, events, original, _ = real_associations()
# Use interval endpoints from measured passages, including subsecond Unix
# times. No acoustic sample, trajectory, event or flight identity is changed.
for candidate in (original[0], original[5274], original[-1]):
lower, upper = candidate["recording_start_bounds_s"]
for sweep_s in (0.0, 5.0):
for instant in (lower - sweep_s, upper + sweep_s):
window = (instant, instant)
expected = scheduled_intervals(original, window, sweep_s)
assert expected
bounded, _ = bounded_associations(table, events, window, sweep_s)
assert scheduled_intervals(bounded, window, sweep_s) == expected