Skip to content

Python tool for aircraft Doppler audio geolocation using historical ADS-B flight data (Source-available)

Keep interior CPA minima near search track endpoints

Cinna Daviseb4fc94e858f
diff --git a/src/dopplergeist/search.py b/src/dopplergeist/search.py
index 8742538..a7c82c9 100644
--- a/src/dopplergeist/search.py
+++ b/src/dopplergeist/search.py
@@ -4,6 +4,7 @@ import math
import numpy as np
import pandas as pd
+from scipy.optimize import minimize_scalar
from scipy.signal import find_peaks
from scipy.spatial import cKDTree
@@ -101,6 +102,38 @@ def cpa_table(tracks,points,max_distance,c=343.0,step_s=2.0,chunk_size=512):
distance2=np.sum((pos[sample]-obs[observer_index])**2,axis=1)
previous=np.r_[False,(observer_index[1:]==observer_index[:-1])&(sample[1:]==sample[:-1]+1)]
following=np.r_[(observer_index[1:]==observer_index[:-1])&(sample[1:]==sample[:-1]+1),False]
+ # An endpoint sample can be the sampled minimum even when the
+ # continuous spline turns strictly inside its adjacent interval.
+ # Check the endpoint range-rate signs before optimizing the actual
+ # spline; an endpoint alone is not evidence of a complete pass.
+ for end in (False, True):
+ if not np.any(sample == (len(t)-1 if end else 0)):
+ continue
+ mask = ((sample == len(t)-1) & previous & (distance2 <= np.r_[np.inf, distance2[:-1]])
+ if end else (sample == 0) & following & (distance2 <= np.r_[distance2[1:], np.inf]))
+ edge_indices = np.flatnonzero(mask)
+ if not len(edge_indices):
+ continue
+ oi_edge = observer_index[edge_indices]
+ lo, hi = (t[-2], t[-1]) if end else (t[0], t[1])
+ endpoints = pos[[-2, -1]] if end else pos[[0, 1]]
+ velocities = track.speed_vector(np.array([lo, hi]))
+ radial_start = np.sum((endpoints[0] - obs[oi_edge]) * velocities[0], axis=1)
+ radial_end = np.sum((endpoints[1] - obs[oi_edge]) * velocities[1], axis=1)
+ for j in np.flatnonzero((radial_start < 0) & (radial_end > 0)):
+ observer = obs[oi_edge[j]]
+ fit = minimize_scalar(lambda dt: np.sum((track.position(lo + dt) - observer) ** 2),
+ bounds=(0, hi-lo), method="bounded", options={"xatol": 0.001})
+ when = lo + fit.x
+ if not fit.success or not lo < when < hi or not np.isfinite(fit.fun):
+ raise AcousticSupportError(track.id, "cpa_endpoint_solver_not_converged",
+ emission_intervals_s=[[float(lo), float(hi)]])
+ distance = np.sqrt(fit.fun)
+ if distance <= max_distance:
+ rows.append({"point_id": int(point_ids[chosen[oi_edge[j]]]), "flight_id": track.id,
+ "emission_time": float(when), "receive_time": float(when + distance/c),
+ "distance_m": float(distance),
+ "speed_mps": float(np.linalg.norm(track.speed_vector(when)))})
minima=(previous&following&(distance2<=np.r_[np.inf,distance2[:-1]])&
(distance2<np.r_[distance2[1:],np.inf]))
indices=np.flatnonzero(minima)
diff --git a/tests/test_real_cpa_edge.py b/tests/test_real_cpa_edge.py
index 16c9e8f..8a2c735 100644
--- a/tests/test_real_cpa_edge.py
+++ b/tests/test_real_cpa_edge.py
@@ -4,10 +4,12 @@ import os
from pathlib import Path
import numpy as np
+import pandas as pd
import pytest
from scipy.optimize import minimize_scalar
from dopplergeist.geometry import ecef
+from dopplergeist.search import cpa_table
from dopplergeist.tracks import load_tracks
@@ -42,3 +44,28 @@ def test_real_flight_cpa_inside_final_sample_interval():
if abs(event["emission_time"] - (lo + minimum.x)) < 0.01]
assert len(matches) == 1
assert matches[0]["distance_m"] == pytest.approx(minimum.fun ** 0.5, abs=0.01)
+
+
+def test_real_national_cpa_table_keeps_the_final_interval_pass():
+ source = os.environ.get("DOPPLERGEIST_REAL_REFERENCE_ROOT")
+ if not source:
+ pytest.skip("set DOPPLERGEIST_REAL_REFERENCE_ROOT to the real reference data checkout")
+ root = Path(source)
+ seed = json.loads((root / "results/reference-local/ranked.json").read_text())[0]
+ point = json.loads((root / "results/reference-faint-alltypes-local/ranked.json").read_text())[5153]
+ assert point["point_id"] == 4904
+ clock = seed["recording_start_time_s"]
+ tracks, _ = load_tracks(root / "data/adsb/2026-09-08/reference-region.parquet",
+ time_range=(clock - 120, clock + 3720))
+ track = next(track for track in tracks if track.id == "a52fe1:1")
+ observer = ecef(point["lat"], point["lon"], point["elev_m"])
+ reference = [event for event in track.closest_approaches(observer, max_distance=20000)
+ if 0 < track.time[-1] - event["emission_time"] < 1]
+ assert len(reference) == 1
+ points = pd.DataFrame([{key: point[key] for key in ("point_id", "lat", "lon", "elev_m")}])
+ table = cpa_table([track], points, 20000)
+ near_end = table[(track.time[-1] - table.emission_time > 0)
+ & (track.time[-1] - table.emission_time < 1)]
+ assert len(near_end) == 1
+ assert near_end.iloc[0].emission_time == pytest.approx(reference[0]["emission_time"], abs=0.001)
+ assert near_end.iloc[0].distance_m == pytest.approx(reference[0]["distance_m"], abs=0.01)