Deep dive
The full story: from atomic clocks to the number on your screen
By Bhagya Masalawala · Last reviewed 12 October 2026 · ~9 min read · Tested in Chrome & Safari on Android and iPhone
Trilateration in plain English
Every GPS satellite carries atomic clocks accurate to billionths of a second and continuously broadcasts two things: where it is and exactly when the message left. Your phone records when each message arrives. Light-speed radio waves cover about 300 meters per microsecond, so a timing error of one microsecond is a 300-meter position error — which is why the clocks matter so much and why your phone needs at least four satellites, not three: three solve for latitude, longitude and altitude, and the fourth solves for the bias in your phone's own cheap clock. With five or more satellites — typical outdoors today with GPS plus Galileo or GLONASS in view — the solution tightens further and the accuracy badge on this page drops toward ±5 meters. Your phone repeats this fix about once per second while tracking, producing the trail of timestamped positions from which speed is derived.
Two things degrade trilateration before speed is even computed. The first is blocked sky: tunnels, underground sections, parking garages and deep indoors simply remove satellites from view, and with fewer than four usable signals there is no fix at all — this page correctly shows Searching rather than guessing. The second is multipath: in downtown canyons or near cliffs, signals bounce off glass and rock and arrive late, as if the satellite were farther away. The fix then wanders 20–40 meters even while you stand still. Recognizing multipath is a core skill: if your dot drifts while you are stationary and the badge reads ±25 m or worse, you are seeing reflections, and any instant speed deserves skepticism until you reach open sky.
Two ways to turn fixes into speed
Once positions stream in, speed can be computed two ways, and this site uses both — in order of preference. The first isDoppler speed. Each satellite transmits on a precise carrier frequency (1575.42 MHz for GPS L1). A receiver moving toward a satellite sees that frequency shifted slightly up; moving away, slightly down — the same Doppler effect as a passing siren. Combining shifts from several satellites yields a three-dimensional velocity vector directly, without differencing positions. Doppler speed is remarkably robust: even when your mapped position sits five meters off the road, the velocity can be right within 1 km/h, because errors common to consecutive fixes cancel in the velocity solution. When your browser exposes this as coords.speed (meters per second), this page displays it first, converted to your chosen unit.
The second method is the Haversine fallback for browsers and devices that report position without speed. Take two consecutive latitude/longitude fixes, compute the great-circle distance between them with the Haversine formula —a = sin²(Δφ/2) + cos φ1 · cos φ2 · sin²(Δλ/2), d = 2R·atan2(√a, √(1−a)) with Earth's radius R ≈ 6,371,000 m — and divide by the seconds between them. At 25 km/h you cover about 7 meters per second, comfortably above typical fix noise, so the quotient is meaningful. At a 5 km/h stroll you cover 1.4 meters per second — comparable to the noise itself — so single-second quotients jump. That is why this page smooths the fallback over the last three to four fixes: a short moving average suppresses single-fix spikes while letting genuine acceleration show through within a second or two. Neither method ever invents movement from nothing: below about 0.5 m/s with a solid lock the page reports a calm, explicit Stationary state.
Why speed beats position — with numbers
A concrete example makes the cancellation intuitive. Suppose your true position at second 0 is at the origin and at second 1 you are 7 meters east (about 25 km/h). Ionospheric delay shifts both fixes 4 meters north — a realistic common-mode error. Your mapped positions read (0, 4) and (7, 4): each is 4 meters wrong in absolute terms, yet the difference between them is still exactly 7 meters east, so the speed is exactly right. Real errors are never this clean — some component is always independent between fixes — but the principle holds: correlated errors cancel in velocity while they persist in position. Field consequence: trust a ten-second GPS average at steady pace even when the dot looks restless, and distrust any single-second spike that the average does not confirm.
| Scenario | Position error | Speed error (avg 10 s) |
|---|---|---|
| Park path, open sky | ±4 m | ±1 km/h |
| Suburban road,steady 50 km/h | ±8 m | ±1–2 km/h |
| Downtown canyon, stop-start | ±30 m | ±5–8 km/h spikes |
| Tunnel, 15 s | no fix | held + Searching |
What degrades a fix — and what does not
- Does degrade: tunnels, underground, metal coaches, indoors, dense urban canyons, heavy wet tree cover, phone in pocket or bag, heat-reflective windshields, thick metal cases. Each removes satellites or adds reflections. Expect the badge to climb past ±30 m and the page to show Weak or Searching — believe it.
- Barely matters: clouds, light rain, cold. GPS radio at 1.5 GHz passes through weather with only centimeters of extra delay, corrected by the receiver model. A grey rainy day with open sky still gives ±5–10 m. Do not blame the weather for what buildings cause.
- Matters more than people think: warm-up, motion state, and power saving. The first 20–30 seconds acquire ephemeris data; readings before that are provisional. Standing still, consumer receivers deliberately freeze direction (the compass needs motion); and battery-saver modes that downgrade to Wi-Fi positioning (±500 m) destroy speed entirely. High-accuracy location mode is not optional for this tool.
How this page smooths — and why it shows gaps honestly
Raw consumer fixes are noisy, so every GPS product filters. Ours is deliberately conservative and disclosed: Doppler speed where available, else Haversine distance over time averaged across the last few fixes, with sanity gates — fixes older than a few seconds are marked stale (“last fix Xs ago”), fixes with very poor accuracy are down-weighted, and a live-but-motionless lock reads as Stationary rather than flickering 0.4–1.1 km/h. The alternative used by some apps — heavy filtering that invents a gliding 12 km/h through a tunnel — feels smoother but lies about tunnels. We chose the honest trade: slightly livelier numbers in cities, but never a fabricated cruise through a dead zone. When the signal returns after a tunnel, average and max resume from real fixes; the gap itself is the truth about where GPS cannot see.
Calibrate once: a 10-minute field test anyone can do
- 1. Pick open road as a passenger. Driver drives, you measure — never handle the phone while driving. Choose a straight highway stretch with open sky.
- 2. Warm up, then reset. Tap Start, wait until the badge reads ±15 m or better, then Reset so average and max start clean.
- 3. Hold cruise for 60 seconds. Ask the driver to lock cruise at 80 km/h. Note the GPS ten-second average, not spikes.
- 4. Record the offset. Typical result: dial 80, GPS 76–78. That 3–5% optimism is the manufacturer's legal margin, and now you know your car's personal number.
- 5. Repeat at a second speed. One more run at 50 or 100 km/h confirms the offset is proportional, not a fault. File it mentally: from now on both instruments make sense together.
Cyclists can run the same protocol against a known-distance path: ride one measured kilometer steadily, compare the page's distance tile against the marker, and note the average. Runners: compare a 400-meter track lap. One careful experiment beats a hundred forum opinions, and it teaches the single most valuable GPS skill — reading the average in context instead of reacting to spikes.
Limits and safe use
This technology has firm boundaries. Consumer GPS is a reference, not a certified instrument: it cannot judge a speed limit, stopping distance, or navigation decision where error costs safety, and it is never a substitute for a type-approved vehicle speedometer, marine log, or aviation instrument. It reports speed over ground only — a boat in a 3-knot current or a train on a gradient gets ground speed, not effort or speed through water. It needs sky, power, and honest interpretation: tunnels and underground will always gap, cities will always jitter, and heat plus battery savers will always slow fixes. Mount the phone, start before moving, obey the law, and treat every number as what it is — an excellent, inexpensive, slightly imperfect window onto your motion. Used that way, a free browser page answers “how fast am I going?” about as well as physics at this price allows.