A satellite is a clock that announces the time.
Twenty thousand kilometres up, a GNSS satellite does one job: it broadcasts, continuously, what time its atomic clock reads and where the satellite believes it is. Nothing is addressed to you, nothing comes back, and the constellation has no idea you exist. A receiver is a listener — which is why a million gliders and a billion phones can use the same satellites at once.
The signal arrives absurdly weak — below the thermal noise of the receiver itself. It survives because it is spread across a wide band with a code sequence the receiver already knows: correlate the incoming noise against that code and the signal climbs roughly 30 dB out of the mud. The correlation peak is the measurement. Its position in time tells you how long the code took to travel down.
That is the only quantity a GNSS receiver ever measures: elapsed time. Distance, position, altitude and ground speed are all inferences built on top of it — and each one inherits every error in the timing underneath.
The fourth unknown is your own clock.
Radio covers 300 m in a microsecond. To place yourself within three metres you must therefore know the travel time to about 10 ns — and your receiver keeps time with a quartz crystal that costs a few cents and drifts by milliseconds. The satellite end is fine; the panel end is hopeless.
So the receiver does not compute a range at all. It computes a pseudorange: the true distance plus whatever your own clock is wrong by, multiplied by the speed of light. The saving grace is that the error is the same on every satellite at that instant — one bias shared by all measurements, not one per satellite.
That turns a hopeless problem into an ordinary one. Three unknowns for position, one for the clock: four unknowns, so four satellites. That is the whole reason behind the number everyone quotes without explaining. And there is a bonus — because the receiver has to solve for its own clock error, it ends up knowing UTC to within a few nanoseconds. Half the world's telecom and power networks are timed by that side effect.
Find the clock error the receiver cannot see.
Every circle is one satellite saying you are somewhere on this ring. Your clock error inflates or shrinks all three by the same amount, so they stop crossing at a point and open into the triangle navigators call a cocked hat. Slide the error to zero and it collapses. The receiver never sees the cross in the middle — it only sees that the circles disagree, and that disagreement is enough: one extra satellite turns the clock error from an unknown into a measurement.
Four satellites is the floor, not the target.
A modern receiver tracks thirty or more. The extra measurements do three things: they average down the random noise, they let the receiver detect a satellite that is lying to it, and — by far the most important — they improve the shape of the problem.
Shape is everything here. Four satellites bunched into one corner of the sky give you four almost identical equations, and a set of nearly identical equations answers a question badly no matter how precisely each one was measured. Four satellites spread wide give you four independent views, and the same ranging accuracy suddenly buys you a much tighter fix. The number that captures this is DOP — dilution of precision — and it is a pure multiplier: position error = DOP × ranging error.
Split by axis, DOP explains one thing pilots complain about constantly. Horizontal DOP is usually below 1, because satellites surround you in azimuth. Vertical DOP is almost always 1.5 to 2 times worse, and no receiver can fix it: half the sphere is under your feet, so every satellite pulls from above and none from below. GNSS altitude is structurally the weakest number your instrument produces.
Drag the sky. Watch the metres.
Drag any satellite. Herd four of them into one corner and the metres explode while the count on the dial never changes — that is dilution of precision, and it is the reason a receiver reports DOP at all. Then switch constellations on: the sky fills, the shaded spread widens, and the error shrinks without a single satellite getting better. The mask slider is your terrain, your panel and your canopy frame — a valley wall at 25° costs far more than the satellite count suggests, because the low satellites you lose are exactly the ones holding the geometry open. Metres assume a 1.5 m ranging error.
What actually degrades a fix.
Geometry is the multiplier. Below is what it multiplies — the ranging error itself, term by term, for a decent single-frequency receiver on an average day.
The ionosphere dominates, and it is also the one term with an elegant way out. Its delay depends on frequency, so a receiver tracking two bands can combine them into an ionosphere-free measurement and cancel most of it arithmetically. That, not the extra satellites, is why dual-frequency receivers were the biggest single jump in civil GNSS accuracy in twenty years.
Multipath is the term with no ceiling. A signal that bounces off a hangar wall, a wet wing or a metallised canopy arrives late, and late means far. Open sky is nearly immune; a glider parked between two hangars can sit thirty metres from where it is. It is also the reason a fix that looks perfect on the ground can be the worst fix of your day.
Four constellations, and why the fifth satellite is not free.
GNSS is the family name. GPS (United States), GLONASS (Russia), Galileo (Europe) and BeiDou (China) each fly a constellation of twenty-four to thirty-plus satellites, and any modern chipset tracks all four at once. In open sky that is thirty-plus satellites in view instead of eight — redundancy that turns a marginal urban or mountain fix into an ordinary one.
There is a catch worth knowing. Each system runs its own time scale, so mixing two constellations adds one more unknown — the offset between them. Two systems need five satellites, three need six. The receiver can shortcut this with broadcast offsets, but the principle stands: the first satellite of a new constellation pays for itself only once a second one arrives.
Coverage differs too. GPS orbits are inclined 55°, so at Alpine latitudes there is a permanent thin patch in the northern sky that no amount of waiting fills. GLONASS flies at 64.8° and covers it — one concrete reason a multi-constellation receiver behaves better in a north-facing valley.
Augmentation: metres, then centimetres.
SBAS — EGNOS in Europe, WAAS in the States — is the one that matters in a cockpit. Ground stations measure the errors, geostationary satellites broadcast the corrections on the same L1 frequency, and any receiver that listens gets roughly 1 – 2 m instead of 5 – 10, plus an integrity message telling it when not to trust a satellite. It is free, it needs no extra hardware, and it is what puts vertical guidance on a GNSS approach — EGNOS is why LPV minima exist in Europe.
RTK goes further by changing the measurement. Instead of timing the code, it counts cycles of the carrier wave itself — a 19 cm ruler instead of a 300 m one — and resolves how many whole cycles are in flight by differencing against a base station within tens of kilometres. The result is centimetres, at the price of a data link and a base. Surveyors and autonomous tractors live here; gliders do not. PPP reaches similar accuracy from a global correction stream with no local base, but takes minutes to converge — fine for a research aircraft, useless for a winch launch.
In the cockpit.
Altitude is not what you think it is. GNSS height is measured from a mathematical ellipsoid, not from sea level. Over central Europe the geoid — the surface that sea level actually follows — sits roughly 45 m above that ellipsoid, so raw GNSS altitude reads high by about that much until the receiver applies a geoid model. Add the structurally poor VDOP and the conclusion is the one every regulator reached: fly airspace on a calibrated barometric altimeter. IGC files record both heights side by side for exactly this reason.
Ground speed is better than position. A good receiver does not derive velocity by subtracting successive positions — it measures the Doppler shift of the carrier, which is an independent observation accurate to a few centimetres per second. That is why the position on a moving map jitters while the ground speed sits rock steady, and it is what makes an in-flight wind computation possible at all.
Rate and latency are performance figures. A 1 Hz fix arrives up to a second stale; add 200 ms of pipeline latency and at 90 knots the glider is typically 30 m — worst case over 50 m — past the point on the screen. And a circling glider changes heading meaningfully inside one second. Fast, low-latency fixes are what let a system compute wind from a single circle rather than averaging over three.
The antenna is still the whole story. GNSS signals are right-hand circularly polarised and arrive from the whole upper hemisphere, so the antenna needs sky — not a carbon deck above it, not a metallised canopy, not the shadow of an engine pylon. The failure mode is identical to the FLARM one, and so is the fix: give it a view.
And the cold start: a receiver needs each satellite's orbit data before it can use it, which arrives at a leisurely rate in the navigation message. That is why a glider out of the hangar after a winter takes a minute to fix and the next flight takes seconds — the data is still cached. Antenna placement, in detail.
What to take to the aircraft.
Trust the horizontal position, treat the GNSS altitude as a cross-check and never as a clearance, and remember that both come from the same weak signal reaching one antenna. Multi-constellation and SBAS have made the fix itself very good; almost everything that still goes wrong goes wrong between the sky and the connector.
The next thing your computer does with all of this is turn a sequence of fixes into a wind vector — which is the wind article.