Multi-Constellation GNSS
Why GPS + Galileo + BeiDou is not simply more satellites
Why it matters
Combining constellations roughly triples or quadruples the satellites in view, which is often the difference between a usable fix and none at all in cities, under trees, or at high latitudes. But it only works if the inter-system biases are handled correctly; done naively it degrades accuracy. It is also the foundation of modern integrity (ARAIM) and precise positioning.
Prerequisites
The temptation, and the trap
By now you can take a handful of pseudoranges and turn them into a fix with single-point positioning. And at the end of acquisition we noted that a modern receiver locks onto dozens of signals at once — not just GPS, but Galileo, BeiDou, and GLONASS, all overhead, all transmitting. The temptation is obvious and overwhelming: throw them all into the same least-squares solve. Thirty satellites instead of eight! The fix should be unstoppable.
Try it naively and the fix gets worse. A receiver that mixes GPS and Galileo as if they were one system can end up metres off — more satellites making things worse, which is exactly the kind of result that means a concept is hiding underneath. So before we reap the (very real) reward, we have to understand the trap.
The idea, before any mathematics
Go back to what a pseudorange actually contains. It is the geometric range plus the receiver's clock offset — the same offset on every satellite, which is precisely why we can solve for it as one extra unknown. That logic has a silent assumption buried in it: that "system time" means the same thing for every satellite. For a single constellation it does. Across constellations, it does not.
GPS keeps GPS Time. Galileo keeps Galileo System Time. BeiDou keeps BeiDou Time. Each is a superb timescale in its own right, but they are steered independently, and they sit tens of nanoseconds — metres of range — apart from one another. (GLONASS adds its own twist, running on a UTC-based scale and historically using a different signal scheme entirely.) So when your receiver measures the offset between its own clock and "system time," the answer is different depending on which system's satellite it is looking at. A GPS pseudorange and a Galileo pseudorange, taken at the very same instant by the very same receiver, carry different clock offsets. Lump them into a single clock unknown and the solver is forced to reconcile two numbers that genuinely differ — and it pays for the contradiction by shoving the position.
The fix is as clean as the problem: give each system its own receiver-clock unknown. Each one quietly absorbs that system's timescale offset plus the receiver's hardware delay for that signal — together the inter-system bias (ISB). GPS satellites share one clock unknown, Galileo satellites share another, and the position is common to all of them, tying the whole solve together.
Every constellation tells time slightly differently. You do not force them to agree — you let each keep its own clock in your solution, and share only the one thing they truly agree on: where you are.
That single idea sets the whole accounting. You no longer have one clock unknown; you have one per system. There is a reference-frame echo of the same story — GPS broadcasts in WGS84, Galileo in GTRF, BeiDou in CGCS2000, GLONASS in PZ-90, all realisations of the global ITRF that agree to a few centimetres — small enough to ignore for a metre-level fix, but real, and part of the same lesson that constellations are not interchangeable.
Seeing it
Picture your sky. With GPS alone, maybe eight satellites are up, and if you stand in a street between tall buildings, half of them are hidden — you are scraping by on four, with lopsided geometry. Now switch on Galileo and BeiDou: the open strip of sky above the street fills with a dozen more satellites from the other systems. Your geometry, measured by DOP, tightens dramatically, and you sail through the canyon that would have stranded a GPS-only receiver. The cost was small — a couple of extra satellites to pay for the extra clocks — and the payoff was the difference between a fix and no fix.
Try it yourself
Toggle constellations on and off, and raise the urban-canyon mask that blocks a slice of sky.
- Satellites used
- 3 of 6 visible
- Constellations
- 1
- Unknowns (2 + clocks)
- 3
- Redundancy
- 0
- HDOP
- 2.58
- Position error
- 0 cm
Turn on just GPS and widen the urban canyon: as buildings swallow the sky, satellites drop out, the redundancy hits zero, and the fix dies. Now switch on Galileo and BeiDou — the open sky refills with their satellites, HDOP falls, and the fix comes back stronger than GPS ever had it alone. That is why every modern receiver is multi-constellation. But notice the cost in the read-out: each constellation adds an unknown, because each runs its own system time, so you need one extra satellite per system before any of them helps. Finally, untick clock per system: without a per-system clock to absorb the inter-system bias, combining constellations drives the position wrong — more satellites making things worse, the trap that makes this a concept and not just a bigger number.
Start with GPS only and widen the urban canyon. Satellites vanish behind the buildings, the redundancy falls to zero, and the fix dies. Now switch on Galileo and BeiDou: the open sky refills, HDOP drops, and the fix returns — healthier than GPS alone ever was. Watch the read-out as you add each system: the unknowns tick up by one every time, because each constellation brings its own clock, so the first satellite of a new system merely pays for its own unknown and only the second begins to help. Finally, untick clock per system. With a single shared clock the inter-system biases have nowhere to go, and the position error jumps from centimetres to metres — the trap, made visible.
Now the mathematics
Single-constellation positioning solved for position and one clock. With systems present, the state grows by one clock per system:
where is the receiver-clock offset (in metres) relative to system 's time scale. The measurement model for a satellite belonging to system is
so its Jacobian row is the usual line-of-sight direction followed by a in only that system's clock column and zeros in the others:
The unknown count is therefore in this 2D course ( in the real 3D world). The redundancy is
for satellites. This is the precise statement of "not free": two GPS and two Galileo satellites give against unknowns — a bare solution with zero check. The fifth satellite, from either system, is the first that buys any redundancy. But from there the extra satellites are pure geometry, and the position cofactors of — the HDOP — fall fast as the new satellites fill directions the first constellation left empty.
From the idea to working code
@gnerd/gnss-core models it directly. solveMultiGnss2D takes satellites tagged with their
constellation and estimates position plus one clock per distinct system, reporting the counts,
redundancy, and combined HDOP. A singleClock flag forces the wrong, shared-clock model so the bias is
demonstrable rather than merely asserted:
import { solveMultiGnss2D } from "@gnerd/gnss-core";
// GPS + Galileo satellites, each system carrying its own clock offset in the pseudoranges.
const observations = [
{ system: "gps", position: gps1, pseudorange_m: r1 },
// ... more GPS and Galileo satellites ...
];
const correct = solveMultiGnss2D(observations); // one clock per system
const naive = solveMultiGnss2D(observations, { singleClock: true }); // one shared clock
// correct.numClockUnknowns -> 2, correct.position ~ truth, correct.rmsResidual_m ~ 0
// naive.position is biased by the inter-system offset it cannot absorb; its residuals stay large
As with every module in the course, the measurement formation is explicit, the clock model is a stated choice rather than a hidden assumption, and the geometry diagnostic (HDOP) falls out of the same normal matrix the solve already builds.
Check your understanding
- Why can a GPS pseudorange and a Galileo pseudorange, measured at the same instant by the same receiver, not share one clock unknown? (The two systems run independent time scales, so the receiver's offset from "system time" differs between them — the inter-system bias.)
- You add BeiDou to a working GPS + Galileo fix but see only a tiny improvement from BeiDou's first satellite. Why? (That satellite first has to pay for BeiDou's new clock unknown; only the second BeiDou satellite adds redundancy and geometry.)
- A receiver combines four constellations yet reports a worse position than GPS alone. What is the most likely cause? (It is not estimating per-system clocks, so uncorrected inter-system biases are corrupting the fix.)
Where this goes next
Multi-constellation is the quiet revolution that made GNSS work where it used to fail — dense cities, deep tree cover, high latitudes — simply by having more sky to lean on. It is also the launchpad for the next tier of techniques: ARAIM, the advanced integrity that leans on cross-constellation redundancy to certify aviation; multi-frequency processing that uses signals from each system to kill the ionosphere; and precise positioning that must now track a forest of per-system and per-frequency biases to reach centimetres. Every one of them starts from the humble bookkeeping you just did: share the position, respect each clock, and count the cost honestly before you spend the reward.
Common misconceptions
- Combining constellations just means more satellites in the same solve. Each system has its own time scale, so its pseudoranges carry a different clock offset; they cannot share one clock unknown.
- More satellites always means a better fix. Each added constellation also adds an unknown (its clock), so a fix needs one extra satellite per system, and ignoring the bias makes accuracy worse.
- All GNSS use the same coordinate frame. GPS uses WGS84, Galileo GTRF, BeiDou CGCS2000, GLONASS PZ-90; all are ITRF realisations but differ slightly, which matters for precise work.
Key takeaways
- Each constellation runs its own system time scale and reference-frame realisation.
- A multi-constellation fix estimates one receiver-clock offset per system (the inter-system bias).
- Unknowns are 2 position + one clock per system, so you need one extra satellite per constellation.
- The payoff is a fuller sky - lower DOP, higher availability, more redundancy for integrity.
- Ignoring the inter-system bias biases the position; it is not optional.