How Does a GPS Speedometer Work? The Science Behind Accuracy

1. The 30-second answer

Navigation satellites broadcast their location and the exact time, kept by atomic clocks. Your phone times those signals, works out how far away each satellite is, and intersects the distances to fix its own position — a trick called trilateration. Repeating the fix every second gives a trail of positions, and speed falls out of that trail: either measured directly from the Doppler shift of the radio waves, or computed as distance travelled divided by time. Outdoors with a settled fix, that chain lands within about 1–3 km/h of your true speed. Everything below is that sentence, unpacked — and the unpacking is where the interesting errors live.

2. Four clocks and some geometry

Each satellite in the GPS constellation (America's system, joined today by Europe's Galileo, Russia's GLONASS and China's BeiDou — your phone typically hears several at once) carries multiple atomic clocks stable to a few billionths of a second per day. It continuously transmits two essentials: where I am (its orbital ephemeris) and when I sent this (a timestamp). Radio waves travel at the speed of light, roughly 300 metres per microsecond, so timing is everything: a single microsecond of error is 300 metres of position error.

Your phone compares arrival times across satellites. Three satellites narrow you to two possible points in space (one is usually absurd — deep underground or far in orbit — so it is discarded), giving latitude, longitude and altitude. But there is a catch: your phone's clock is a cheap quartz crystal, nowhere near atomic grade. So a fourth satellite is needed to solve for your clock's own bias as a fourth unknown. That is why receivers quote a minimum of four satellites, and why fixes keep improving as five, eight or twelve come into view — extra measurements let the solver average down the noise, and the ± meters badge on our speedometer visibly drops toward ±5 m on open ground.

The system repeats this roughly once per second while you track. Speed, distance, average and max are all downstream of that one-per-second heartbeat. If the heartbeat stops — tunnel, underground car park, deep indoors — there is nothing honest to display, which is why this site shows a Searching state instead of coasting on imagination. A gap in the record is information, not failure.

3. Two ways to turn positions into speed

Once fixes stream in, there are two fundamentally different ways to derive speed, and our speedometer uses them in a strict order of preference. Understanding both explains nearly every confusing readout you will ever see.

Method A — Doppler velocity (preferred)

GPS satellites transmit on precise carrier frequencies — 1575.42 MHz for the classic L1 signal. A receiver moving toward a satellite observes that frequency shifted slightly upward; moving away, slightly downward. It is the same physics as an ambulance siren changing pitch as it passes, applied to radio. By combining the shifts from several satellites, the chip computes a full three-dimensional velocity vector directly, without subtracting one position from another. Browsers expose this as coords.speedin metres per second, and whenever it is available our page displays it first, converted into your chosen unit. Doppler velocity is the reason a phone can report a rock-steady 89 km/h while its own map dot wobbles across two lanes.

Method B — breadcrumbs (distance over time)

Some browsers and older hardware report positions but no speed field. The fallback is beautifully simple: take two consecutive latitude/longitude fixes, compute the great-circle distance between them with the Haversine formula, and divide by the seconds elapsed. At 25 km/h you cover about 7 metres per second — comfortably above typical fix noise, so the quotient is meaningful. At a gentle 5 km/h stroll you cover 1.4 metres per second, which is smallerthan the noise itself, so raw second-by-second quotients jump wildly. That is why the fallback is smoothed over the last few fixes: a short moving average suppresses single-fix spikes while letting genuine acceleration show through within a second or two. It also explains why pedestrian readings look bouncier than highway readings on every GPS product ever made — the physics gets harder at walking pace, not the software lazier.

4. The paradox: why speed is steadier than position

Newcomers find this hard to believe, so here is a concrete example. Imagine you cycle due east at 25 km/h — 7 metres in one second. A layer of ionospheric disturbance delays every signal slightly, shifting both of your consecutive fixes 4 metres north. Your mapped dots read wrong by 4 metres each. But the difference between them is still exactly 7 metres east — so your speed is exactly right. Errors shared between neighbouring fixes cancel in velocity while persisting in position.

Real-world errors are never this tidy — part of every error is independent between fixes — but the principle survives contact with reality: correlated error cancels, and much GPS error is correlated over seconds. The field skill this buys you is the single most valuable one in this article: trust a ten-second average at a steady pace even when the dot looks restless, and distrust any one-second spike the average does not confirm. That jump to 68 km/h while cruising at 50 is almost certainly a reflected signal, not a gust of hidden horsepower.

5. The error budget: what actually eats accuracy

Think of accuracy as a budget that the environment spends. Here are the line items, roughly ordered by damage:

And what barely matters: clouds, light rain and cold. GPS radio at 1.5 GHz sails through weather with centimetres of extra delay that receivers already model. A grey rainy day under open sky still delivers ±5–10 m. When someone blames the clouds for a jumping readout downtown, the buildings did it.

6. Your phone matters more than you think

Not all receivers are equal, and the differences are physical. Flagship phones from roughly 2018 onward increasingly ship dual-frequency GNSS — tracking both the classic L1 signal and the newer L5/E5a signals — which lets the chip directly measure and subtract most ionospheric delay instead of modelling it. In open sky that roughly halves the error; in cities the extra signal diversity helps reject reflections. A second factor is constellation count: a phone hearing GPS plus Galileo plus GLONASS has far more satellites to choose from, so losing two behind a building hurts less.

Then there is software. Your browser's Geolocation API is a middleman: it decides update rates, may quantise or delay fixes, and on some platforms withholds coords.speedentirely — which is exactly why our page carries the Haversine fallback rather than showing a dash. Desktop browsers without any GNSS hardware may silently substitute Wi-Fi positioning (±hundreds of metres), which cannot produce meaningful speed at all; the accuracy badge exposes that substitution the moment it happens. The takeaway: two phones side by side can honestly disagree by a few km/h in a street canyon, and both can be behaving correctly for their hardware.

7. GPS against the dashboard: the optimism rule

Drivers comparing GPS to the dial usually discover the dial reads higher — and assume the phone is wrong. More often it is the other way round. Regulations in the EU, US, India and elsewhere permit speedometers to read high by a margin (never low), and manufacturers use it: a dial showing 100 km/h typically corresponds to 94–98 km/h of true speed. It is a deliberate safety margin, not a fault, and changing tyre size shifts it further.

So at a steady cruise on open highway with a green accuracy badge, the browser readout is frequentlycloser to physical truth than the needle. But “closer to truth” is not “certified”: the dashboard is type-approved, always available, and stable in tunnels; GPS needs sky and comes with no certificate. Use them as a pair. If both agree within a few km/h at cruise, both are healthy. If they disagree by more than ~10% repeatedly in good signal, suspect tyres, a failing wheel sensor — or a phone baking in direct sun with throttled updates — and investigate rather than picking a favourite. And whatever you do, the dial remains the only legal instrument: GPS is a respected second opinion, never a defence in a speeding dispute.

8. Myths worth retiring

• “More bars means more accuracy.” Signal strength bars describe your mobile network, not satellites. Watch the ± meters badge instead — it is the receiver telling you its own uncertainty.
• “GPS needs the internet.” Positioning is pure satellite radio. Data only loads the page; afterwards you can track in airplane mode with location on. (Assisted-GPS downloads ephemeris faster with data, but the fix itself is offline physics.)
• “A paid app is more accurate.” Apps share the same OS location feed and the same physics. What money buys is features — logs, maps, exports — not a better Doppler shift. Our free page reads the identical sensor.
• “Speed apps drain GPS; the sensor wears out.” The receiver is passive — it only listens. Battery drain comes from keeping the chip, CPU and screen awake, not from any transmission. Nothing wears out by measuring.

9. Calibrate your phone in 10 minutes

Theory is nice; one experiment teaches more. As a passenger (never handle the phone while driving), pick a straight highway stretch with open sky. Open this site's speedometer, tap Start, and wait until the badge reads ±15 m or better — then Reset so average and max start clean. Ask the driver to hold cruise at 80 km/h for a full minute and note the ten-second GPS average, ignoring spikes. Typical result: dial 80, GPS 76–78. Repeat once at a second speed to confirm the offset is proportional. Congratulations: you now know your car's personal optimism number and your phone's honesty in good conditions — two facts no forum thread could give you.

Cyclists: ride one measured kilometre steadily and compare the page's distance tile with the marker. Runners: compare a 400-metre track lap. In each case the protocol is identical — warm up, reset, hold steady, read the average — because the error physics is identical and only the scenery changes.

10. Habits for honest numbers

1. Seek open sky first. A footpath beats a balcony; a park beats a downtown canyon. Buildings are error source number one.
2. Mount it, don't swing it. A swinging hand adds metres of phantom motion per second. Dashboard cradle, handlebar mount, or flat and still at chest height.
3. Wait out the warm-up. Give it 20–30 seconds after Start; reset once the badge turns green so the session stats start clean.
4. Read averages, not spikes. Ten seconds of steady average is the truth; a one-second peak is usually a reflection.
5. Keep high-accuracy location on. Precise Location granted, battery-saver exemptions for your browser on long sessions, phone shaded from direct-sun overheating.

FAQ

Questions this article answers

Try the science yourself

The fastest way to feel everything above is a two-minute outdoor test: start the live tool, watch the badge settle, and compare the average against a steady pace.