Skip to content

Understanding Accuracy

TFD-42 edited this page Aug 13, 2026 · 1 revision

Understanding Accuracy

This tool infers distance from signal strength, then trilaterates. Both steps are approximate — knowing why tells you when to trust the fix.

RSSI ≠ distance

Distance comes from the log-distance path-loss model:

d = d₀ · 10^((PL(d) − PL(d₀)) / (10·n))

d₀ = 1 m reference · n = path-loss exponent · PL = Tx_power − RSSI

It assumes signal decays smoothly with distance. Reality breaks that assumption:

Assumption Reality
Free propagation Reflections off buildings, vehicles, ground
Single path Multipath — copies arrive with different phases
Static environment People, doors, cars shift RSSI 5–15 dB
Isotropic antenna The Flipper's PCB antenna is directional
Known Tx power Actual output often differs from nominal

Multipath is the primary failure mode. A reflection can arrive stronger than the direct path, pushing the estimate in the wrong direction — that's why dense-urban error explodes.

The path-loss exponent n — your biggest knob

Set it in the drawer to match the environment:

n Environment
2.0 Free space / very open field
2.7 Suburban
3.0 Urban (default)
3.5 Dense urban
4.5–5.0 Indoor / heavy obstruction

A wrong n gives systematically biased distances — bad triangulation even with perfect capture geometry. When in doubt, pick the value that matches what's physically around you.

Reading the RMS error

The solver reports RMS residual in metres — how well the distance circles agree:

  • < 30 m — circles agree well, higher confidence.
  • 30–100 m — usable as a search area.
  • > 100 m — noisy environment or bad geometry; re-capture from better spots.

RMS measures internal consistency, not truth. Low RMS with bad geometry can still be wrong — geometry matters as much as the number (see Field Guide).

What the app does to help

  • RSSI smoothing — an exponential moving average damps single-sample multipath spikes before a capture is taken.
  • Outlier rejection — a leave-one-out residual check flags and excludes captures that look like multipath before the final solve.
  • Geometry hint — warns when your captures are too collinear to give an unambiguous fix.

Honest limits

  • ❌ Real-time tracking of a moving transmitter
  • ❌ Sub-10 m precision in any environment
  • ❌ Direction finding (no antenna array)
  • ❌ Reliable indoor / room-level localisation
  • ❌ A replacement for dedicated DF gear

For the measured numbers, see the Field Validation table in the README.

Clone this wiki locally