Satellite positioning is superb with a clear sky and useless without one. In a tunnel the fix vanishes; in a city it reflects off buildings and jumps. Holding an accurate position through those gaps is not a better antenna — it is fusing the satellites with sensors that keep working when the sky does not.
| Input | What it gives | Its limit |
|---|---|---|
| GNSS fix | Absolute position, no drift | Fails in tunnels, reflects in cities |
| Inertial (IMU) | Motion through any gap | Drifts over time if uncorrected |
| Wheel / odometry | Distance travelled, robust | Direction alone is ambiguous |
| Magnetometer | Heading reference | Disturbed by metal and motors |
| Map / road model | Constrains plausible position | Only as good as the map |
| Barometer | Vertical / floor context | Drifts with weather |
The mistake is to treat a satellite fix as ground truth and everything else as backup. In a clear sky the fix is trustworthy; in an urban canyon it can be present, confident and metres wrong, because the signal reached the receiver by bouncing off a building rather than travelling straight down. A fusion system that believes that reflected fix is worse than one that ignored it. So the real work is not adding an inertial sensor as a fallback — it is deciding, moment to moment, how much each source deserves to be believed.
Between fixes, the system dead-reckons: it propagates the last known position forward using how the platform has moved — acceleration, rotation, wheel distance. That coasts cleanly through a tunnel, but inertial estimates drift, because small errors accumulate. Dead reckoning therefore holds an accurate position for the length of a real gap — an underpass, a short indoor stretch — not indefinitely, and the fusion has to know and bound that drift rather than pretend it away.
When the satellites return, the estimate has to snap back to truth without a visible jump, and it has to reject the first few reflected fixes that often arrive at the mouth of a tunnel or between tall buildings. Getting that transition right — correcting smoothly, weighting the returning fixes by their real quality — is what separates a position that feels solid from one that lurches every time conditions change.
Because the problem is physics, not sensitivity. In a tunnel there is no satellite signal to receive however good the antenna, and in an urban canyon the signal reflects off buildings so the fix is present but wrong. No receiver improvement recovers a signal that is absent or corrupted, which is why the answer is fusing GNSS with inertial and motion sensors that keep working when the satellites do not.
Dead reckoning estimates position from motion — how far and in what direction the platform has moved since the last fix — using inertial and vehicle sensors. It coasts through a GNSS gap, but it drifts as small sensor errors accumulate, so it holds an accurate position for the length of a tunnel or an underpass, not indefinitely. The fusion bounds that drift and corrects it the moment a trustworthy fix returns.
By weighting each source by its real reliability and actively detecting multipath. A reflected fix in a city can look entirely valid, so blindly trusting it is worse than ignoring it. The fusion recognises the signatures of an untrustworthy fix — inconsistency with the inertial estimate and the recent track — and down-weights or discards it until a genuinely reliable fix is available.
Written for engineers. Share it with one.
Tell us the specification, the constraint and the deadline. Programmes that cross silicon, radio, embedded and AI are where Faststream is strongest.