Things that aren't blocking the project shipping, but are worth doing next. Nothing here is committed — adopt or drop items as priorities shift.
Released work lives in CHANGELOG.md.
Wire ha-airspace end-to-end
The FEED.md contract is in place; actually plumb the Pi → ha-airspace → Home Assistant path with live data and confirm the round-trip.
Flash an ESP32-S3 with ArduPilot/ArduRemoteID. Closes the trust gap the spoofer's encoding quirks (gs 3×, track mod 180°) leave open. ~$10–15 of hardware.
The live feed now carries operator.location_type (takeoff | live_gnss | fixed) so the dashboard can flag a takeoff-point reading instead of showing it as a live operator fix. The history DB (history.py) doesn't store it yet — query_operator only returns the last op_lat/op_lon/operator_id, so /map's operator marker can't make the same distinction for past detections. Needs a schema bump (new op_location_type column, SCHEMA_VERSION → 2 or an additive ALTER TABLE) plus wiring through on_drone_change and the map popup.
ble_feeder.py and wifi_feeder.py each carry their own copy of parse_basic_id / parse_location / parse_system_msg / parse_operator_id / parse_self_id / decode_rid_message. Refactor to a shared odid_decoder.py so the next spec fix is a one-file change.
Promote the high-quality smoke tests we've been writing in heredocs into a real tests/ directory with pytest. Coverage targets: each parse_* parser, the NAN SDA extraction with optional-field variants, tracker callbacks + most-recent-wins, MQTT publisher's latest-wins debounce. Lets v1.x change code without flying blind.
Items we evaluated and decided against, with the reasoning recorded so we don't re-litigate the same call later. Each entry should also note what would change the decision.
Adding a sidecar receiver for DJI's proprietary OcuSync DroneID broadcast — the signal Aeroscope decodes. Reference implementation: RUB-SysSec/DroneSecurity; paper: Schiller et al., Drone Security and the Mysterious Case of DJI's DroneID, NDSS '23.
Why declined:
- RTL-SDR can't do this. DroneID is a 15.36 MHz OFDM signal at 2.4 GHz / 5.7 GHz. RTL-SDR's ~2.4 MHz instantaneous bandwidth and E4000-tuner 2.2 GHz ceiling disqualify it twice over. Would need a HackRF / PlutoSDR / LimeSDR / B205-mini — ~$250 minimum, ~$1500 for the paper's setup.
- Pi Zero W can't decode it. OFDM demod + descrambling + turbo decode at 50 MHz sample rate needs NUC-class CPU, not a single-core ARMv6.
- Range is ~10 m per the paper's own results — practically worse than the BLE/Wi-Fi RID coverage dump3411 already gets.
- DJI-only. Doesn't help with Skydio, Autel, Parrot, Yuneec, ArduPilot-based airframes, etc.
- Shrinking target. Post-Sept 2023 DJI drones broadcast ASTM F3411 RID natively (already decoded here). Pre-RID DJIs are the long tail and the FAA is gradually retrofitting them via broadcast modules.
- Cheaper alternative for the airframe we actually own. A ~$200 Dronetag Beacon clipped to the Phantom 4 Pro satisfies FAA Part 89 and gives dump3411 a native F3411 signal at full BLE/Wi-Fi range — no SDR pipeline.
- Authors' own caveat. Their README explicitly: "not optimized … not meant for productive, reliable localization" — it's an academic artifact.
What would change this call:
- A specific local DJI-detection use case appears (e.g. a recurring drone in the airspace whose owner isn't using a broadcast module).
- A lightweight SDR-based DroneID receiver lands that runs on Pi-class hardware (FPGA-accelerated frontend, etc.).
- A wideband SDR is already in the rack for unrelated reasons, making the marginal cost near-zero.
Until any of those is true, stay focused on the open-standard F3411 path — which is the regulatory direction the airspace is moving in regardless.
Enrich detected drones with manufacturer + model from the FAA Remote ID registry. Reference implementation: jlrjr/faa-rid-lookup — bundles a local SQLite cache built from the FAA's public API (~3,900 exact serials + ~250 ranges) with optional online fallback.
Why declined here:
- dump3411 is a thin producer. It pulls RID off the wire, decodes per ASTM F3411, serves a snapshot per FEED.md. External-registry enrichment is a different concern with its own operational tail (DB freshness, FAA API rate limits, online-fallback policy).
- The dashboard is "is the receiver working?", not the rich drone view. A bare UAS ID is sufficient for verifying detection. The make/model belongs where drone data is actually consumed.
- FEED.md stays clean. Producing
id+id_typekeeps the wire contract on "what came over the air"; baking enrichment fields into the producer would force the schema (and every consumer) to evolve every time a new enrichment source shows up. - Resolution:
ha-airspacehandles it. That project already is the enrichment layer for ADS-B; RID lookup is exact symmetry — same DB-update cadence, same Home Assistant audience.
What would change this call:
- A lightweight static map of CTA-2063-A manufacturer codes (first 4 chars → vendor name from ICAO's MFR registry, ~50 entries) might land in dump3411 later if a "manufacturer hint" in the dashboard turns out to be useful. That's a ~30-line const, no DB, no network, no schema change — not a registry lookup, just a vendor-name annotation. Skipping it for now.