Triangulation

RSSI multilateration from multi-node detection logs. Two paths:

• Live — the Map tab runs the correlator every refresh over the last 5 min of detections from every source (server + every agent). Fixes show as dashed crosshair markers; popup lists contributing nodes, per-observation RSSI, and calibration coverage. Nothing to run — it's always on.
• Post-hoc (sdr.py tri <db>...) — replays saved sessions end-to-end, configurable path-loss parameters, can push results to ATAK.

How It Works

1. Correlation — matches detections across nodes by device ID (keyfob, TPMS, BLE), channel (PMR, POCSAG), or frequency (GSM, LTE, LoRa) within a configurable time window
2. RSSI → Distance — log-distance path loss model: distance = 10^((RSSI_ref - RSSI) / (10 * n))
3. Multilateration — 2 nodes: weighted midpoint (ambiguous). 3+ nodes: least-squares optimization

Node Spacing Guidelines

Scenario	Min spacing	Notes
Keyfob / TPMS (parking lot)	30-50 m	Short range, steep power falloff
BLE (building)	10-20 m	Very short range
PMR walkie-talkie (outdoor)	100-300 m	Stronger signal, flatter curve
GSM / LTE uplink	200-500 m	High power, need wider baseline

Nodes should be spaced at least as far apart as the expected distance to the emitter. 3 nodes in a triangle is much better than 3 in a line.

Path Loss Parameters

Configurable via --path-loss-exp and --rssi-ref:

Signal type	Exponent (n)	RSSI at 1m (dB)	Environment
BLE	2.5	-40	Indoor/short range
WiFi	2.7	-30	Indoor/outdoor
Keyfob / ISM	2.7	-30	Outdoor
TPMS	2.5	-35	Outdoor, near ground
PMR446	2.2	-20	Outdoor LOS
GSM / LTE	3.0	-20	Urban
LoRa	2.3	-30	Outdoor long range

These are the model defaults; per-node hardware bias is handled separately by calibration (below). --use-snr remains available as a workaround when calibration isn't set up yet.

Calibration

power_db from an RTL-SDR is dB relative to ADC full-scale, not absolute dBm. Two dongles reading the same signal can disagree by 10–20 dB due to gain variation and temperature drift — bigger than the path-loss model error. Calibration removes this per-node bias.

Calibration is automatic — if output/calibration.db exists, sdr.py tri loads per-(node, band) offsets and subtracts them from every observation before multilateration. Triangulation prints how many observations got a calibrated reading vs. fell back to raw power.

Building a calibration

Match captured detections against emitters whose position and TX power are known:

# One-time: tell the calibration DB where this node physically sits.
sudo venv/bin/python3 src/sdr.py calibrate set-position --node-id N01 \
    --lat 42.5098 --lon 1.5361 --alt 1050

# Ingest captured sessions and solve per-band offsets.
sudo venv/bin/python3 src/sdr.py calibrate ingest --node-id N01 \
    --emitters configs/calibration_emitters.json \
    output/wifi_*.db output/fm_*.db output/adsb_*.db

# Inspect solved offsets.
sudo venv/bin/python3 src/sdr.py calibrate show

Reference emitters go in configs/calibration_emitters.json (see .example):

• WiFi APs — BSSID, lat/lon, EIRP in dBm (typical home AP ~20 dBm)
• FM broadcast stations — frequency, lat/lon, EIRP (public databases)
• Cell towers — CGI, frequency, lat/lon, EIRP (forward-compatible; inert until scanners decode serving cell)

Passive sources that don't need a registry:

• ADS-B — aircraft self-report position + altitude; TX power inferred from ICAO category table. Dormant until scanners/adsb.py captures per-message RSSI (tracked in roadmap).
• AIS — vessel class derived from message type; class A ~12.5 W, class B ~2 W. Same RSSI gap as ADS-B today.

Run sdr.py calibrate ingest periodically (or calibrate watch) as new sessions accumulate; offsets refit from the last 7 days of samples by default. Calibration is node-local in output/calibration.db — it stays with the hardware across sessions.

Skip calibration with sdr.py tri --no-calibration for before/after comparisons.

Known Limitations

• RSSI is noisy in practice (multipath, obstructions, antenna orientation). Expect room-level accuracy at best, even after calibration.
• 2-node solutions are inherently ambiguous. 3+ nodes strongly recommended.
• Calibration handles per-node hardware bias. It doesn't solve path-loss model mismatch (log-distance still assumes a simple exponent). For sub-10 m accuracy, TDOA is needed — see the roadmap.
• ADS-B and AIS targets self-report position — triangulation is not needed for them, but they are still useful as passive calibration references.