About Where Am I?
A satellite navigation receiver doesn't measure angles or look at anything. It times signals. Each satellite says where it is and exactly when it sent its message; the receiver notes when the message arrived, and the delay times the speed of light is a distance. Distances from several known points give a position. The catch is the receiver's own clock, which is a cheap quartz crystal, wrong by far more than the nanoseconds that matter, so the receiver solves for its clock error as well as its position. This page shows how, with the real sums.
CIRCLES: a flat toy first. Beacons with perfect clocks stand on a map, and your receiver measures its time to each (1 microsecond of light is 299.8 m, 1 nanosecond about 30 cm). Drag the beacons or the boat. Set how far your receiver's clock is out: every distance comes out too long by the same amount, so the circles miss each other and leave a cocked hat. Slide the correction until the circles meet at one point, or press SOLVE. With two beacons the circles cross somewhere for any correction that lets them reach each other, so the clock error can't be found; a third beacon pins it. In three dimensions there are four unknowns (x, y, z and the clock), so a receiver needs four satellites.
SKY: a sky plot of the satellites in view, the way a receiver shows them (north at the top, the horizon round the edge, the centre straight up), with signal bars. Pick a place, a system and a time; set the elevation mask and what stands round you. ClickTap a satellite to stop or start using it, or let the receiver use them all, the best four or four bunched together. The fix is a real least-squares solution from noisy pseudoranges, started at the centre of the Earth; the card lists each step. The position page shows the fix, fifty more fixes with fresh noise, and the 95% error ellipse worked out from the geometry. When it is one satellite short (three for one system; four with two systems, since each system adds a clock to find), the receiver holds its height, as real receivers do: it takes the height as known (here 0 m, where it truly is; a real set uses its last fix, a map or the sea) and solves for the rest, a 2-D fix. The card says when the height is held. That only works while the height really is known: a wrong held height pulls the position off as well.
ERRORS: where the range errors come from, with typical sizes, and what they do to the fix. Switch each source on or off, put selective availability back on, or use two frequencies or differential corrections, then RUN 100, 1,000 or 10,000 fixes on the sky from the SKY tab. The 95% radius of the results is set against the Expected one from the error budget and the geometry.
MULTIPATH: a reflection reaches the aerial a little after the direct signal and pulls the receiver's timing late (or early, depending on its phase). Pick the reflector (the sea below a masthead aerial, a bulkhead behind it, a warehouse on the quay), its distance and strength, and move the satellite. The charts show the error for every delay (the multipath error envelope) and over a whole pass.
JAMMING: how weak the signals are, and what a jammer does to them. Set a jammer's power, kind and distance: the bars show each satellite's signal-to-noise density falling, and the plan shows inside what range the receiver loses lock or can't start. Also how spoofing works, in plain terms, and the signs a watchkeeper can check.
THE STORY
When Sputnik 1 went up in October 1957, William Guier and George Weiffenbach at the Johns Hopkins Applied Physics Laboratory worked out its orbit from the Doppler shift of its radio signal. Turned round, the idea gives a position to anyone who knows the orbit and measures the shift. That became Transit, the US Navy's satellite system: it was in service for the Navy's missile submarines from 1964 and open to merchant ships from 1967. A Transit fix came from a single satellite pass, from the way the received frequency swept down as the satellite went over, and there could be an hour or more between passes. It closed in 1996.
The US Department of Defense approved the Navstar Global Positioning System in 1973, with Bradford Parkinson leading the programme, and the first Navstar satellite was launched in 1978. After the Soviet Union shot down Korean Air Lines Flight 007 in 1983, President Reagan said GPS would be open to civil use. Full operational capability was declared in 1995. Until 2000 civil users got a deliberately degraded signal, selective availability, which dithered the satellite clocks so that a civil fix was within about 100 m only 95% of the time; it was switched off on the night of 1 to 2 May 2000. Russia's GLONASS (first launched in 1982), Europe's Galileo (initial services from 2016) and China's BeiDou (global since 2020) now fly too, and most receivers use several systems at once.
HOW IT WORKS
The signal. GPS satellites transmit on L1, 1575.42 MHz. The civil C/A code is a pattern of 1,023 chips repeated every millisecond (1.023 million chips a second, so one chip is about 293 m of light), with the navigation message on top at 50 bits a second. The message carries the ephemeris, the satellite's own precise orbit, renewed every two hours or so, and the almanac, rough orbits of the whole constellation that take 12.5 minutes to send in full and stay usable for weeks, so the receiver knows which satellites to look for. Positions come out on WGS 84, the Earth model GPS uses (equatorial radius 6,378,137 m, flattening 1/298.257223563).
The pseudorange. The receiver lines up its own copy of the code with the incoming one; the shift it needs, times the speed of light, is the pseudorange: the true range plus c times the receiver clock bias, plus errors. Every pseudorange carries the same clock bias, so it is one extra unknown, shared by all of them (a second system such as Galileo keeps its own time, so mixing systems adds one more unknown).
The fix is a least-squares solution by Gauss-Newton iteration: guess a position and clock (here the centre of the Earth and zero), work out the ranges that guess would give, and solve the linear sums for the correction that best explains the difference; repeat until the correction is tiny. The first step jumps thousands of kilometres, the last few millimetres, five or six steps in all.
DOP (dilution of precision) comes from the geometry alone. Each row of the matrix H is the unit vector from receiver to satellite (east, north, up, sign reversed) with a 1 for the clock. The diagonal of (HTH)-1 gives HDOP (east and north), VDOP (up), PDOP (all three), TDOP (time) and GDOP (everything). Position error is roughly the range error times the DOP, so satellites bunched in one part of the sky give a big DOP and a poor fix even with good ranges. Height is usually worse than horizontal position, because every satellite is above you. At high latitudes the 55° orbits never pass overhead, which leaves a hole in the northern (or southern) sky.
The errors. The ionosphere slows the code (and is worst near the horizon, where the path through it is longest, about three times the overhead delay); a broadcast model takes out roughly half. Because the delay depends on frequency, a receiver with two frequencies can remove it almost entirely, at the price of about three times the noise and multipath (for L1 and L2 the factor is 2.98). The troposphere is smaller and well modelled. Satellite clock and orbit errors are now well under a metre. Multipath and receiver noise depend on the site and the receiver. Differential corrections from a reference station nearby take out most of what is common to both. All the figures on the ERRORS tab are typical, single-frequency, illustrative ones: real ones vary with the receiver, the time of day and the solar cycle.
Multipath on the code: the receiver's delay lock loop compares an early and a late copy of the code, a fixed spacing apart, and steers until they match. A reflection adds a second, later, weaker correlation peak, so the match point shifts. With a 1-chip spacing and an in-phase reflection of strength a and short delay δ the error is aδ/(1+a); with a narrow 0.1-chip spacing it can't exceed about a times half the spacing; reflections more than a chip and a half late (about 440 m) do nothing. As the satellite moves the reflection's phase turns, so the error swings back and forth.
Jamming. The minimum L1 C/A signal at the Earth's surface is −128.5 dBm (−158.5 dBW), about 18 dB below the thermal noise in its own 2 MHz band; the receiver digs it out by correlating with the code. Typical clear-sky signal-to-noise density is 40 to 50 dB-Hz. Interference adds to the noise as J/(Q·Rc), with Rc the chip rate and Q a spectral factor (the textbook values are 1 for a narrowband jammer, 1.5 for one matched to the signal and 2 for wideband noise). The page assumes a receiver keeps lock down to 28 dB-Hz and needs 34 dB-Hz to start, typical round figures.
WHERE IT SIMPLIFIES
The constellations are tidy Walker patterns: 24 GPS slots in six planes at 55° (two turns per sidereal day, about 11 h 58 min, 20,200 km up) and 24 Galileo slots in three planes at 56° (about 14 h 5 min, 23,200 km up). The real GPS has about 30 satellites in unevenly spaced slots, so a real sky on the same day looks different; the page's time of day is its own. Orbits are circles. The sums leave out the signal's flight time (about 0.07 s, during which the Earth turns), relativity and the satellite clock terms, on both sides, so they cancel. Each fix draws fresh errors; in a real receiver the bigger errors wander slowly over minutes, so the fix drifts rather than jumps. The least squares are unweighted. The flat toy uses straight-line distances on a plane. The correlator is ideal (infinite bandwidth). The jamming sums are free space, line of sight within the radio horizon and nothing beyond it, with no reflection off the sea; the aerial gain towards the horizon is taken as −5 dBi. Spoofing is described, not simulated, and the page gives no instructions for jamming or spoofing.
Not for navigation.
FURTHER WATCHING AND READING