What is RTK GPS? Centimeter accuracy, explained

RTK — Real-Time Kinematic — is a way of getting centimeter-accurate position from satellite navigation, by measuring the errors in the system rather than guessing them. A reference receiver at an accurately known position works out how wrong the satellite signals are right now, and your receiver removes that same error from its own measurements. The result is a position roughly a hundred times more accurate than an ordinary GPS fix, available continuously while you move.

That is the whole idea. Everything below is why it works, what it needs, and what it will not do for you.

Why ordinary GPS is only good to 3–5 metres

A satellite fix is a timing measurement. Each satellite broadcasts a signal stamped with the time it left; your receiver measures how long it took to arrive and multiplies by the speed of light to get a distance. Do that for four or more satellites and you can solve for your position.

Light covers about 30 cm in a nanosecond, so every nanosecond of timing error is 30 cm of position error. Four things introduce those errors:

• Satellite clock and orbit error. The satellite’s own idea of the time and of where it is are extremely good and not perfect. Residual error: tens of centimetres to a couple of metres.
• The ionosphere. Charged particles 50–1,000 km up slow the signal by an amount that varies with the time of day, the solar cycle, and your latitude. This is the largest and most variable error — metres.
• The troposphere. Water vapour in the lower atmosphere, adding a smaller but stubborn delay.
• Multipath. The signal bouncing off a shed, a grain bin, or a wet road and arriving twice, so the receiver measures a path longer than the direct one.

Stack those together and an uncorrected consumer receiver lands at 3–5 metres, on a good day, in the open. Your phone is usually worse, for reasons covered in why your phone’s GPS is only accurate to 3–5 metres.

The trick: measure the error, don’t model it

Here is the insight RTK is built on. If two receivers a few tens of kilometres apart look at the same satellite, the signal reaches both through very nearly the same atmosphere, from a satellite with the same clock and orbit error. The errors are almost identical for both.

So: put one receiver at a position you already know precisely. Whatever discrepancy it sees between the position it computes and the position it knows it is at, is error. Send that discrepancy to the second receiver, and it can subtract it from its own measurements.

Nothing has to be modelled. Nobody has to predict the ionosphere. The error is simply measured somewhere it can be measured, and removed everywhere it applies. That is why RTK is dramatically more accurate than techniques that correct from a global model — and also why it has a range limit, because “the same atmosphere” stops being true as the distance grows.

Carrier phase: why centimeters and not metres

Correcting the timing measurement alone would get you to about a metre. That is what DGPS and satellite-based augmentation do, and it is genuinely useful, but it is not RTK.

RTK goes further by measuring the carrier phase — not the data encoded on the signal, but the position within the radio wave itself. The wavelength is about 19 cm, and a receiver can measure where in that wave it is to a percent or so. Millimetres of resolution, in principle.

The catch is that every wave looks like every other wave. The receiver can see it is 7 cm into a cycle, but not which cycle — how many whole wavelengths lie between it and the satellite. That unknown whole number is the integer ambiguity, and resolving it is what RTK spends its first few seconds doing, using multiple satellites, multiple frequencies, and the geometry changing as the satellites move.

When it succeeds you get a fixed solution: centimeters, arriving as a step change. Until then you have a float solution: corrections applied, ambiguities unresolved, accuracy at the decimetre level. The float-to-fixed transition is the single most important thing to watch in any RTK app, because float looks perfectly plausible on screen and is not repeatable at centimeter level.

Base-and-rover, or a network

There are two ways to get the reference side of that equation.

Your own base station. A second receiver on a tripod at a surveyed point, with a radio link to the rover. You own it outright and it works without internet — genuinely valuable in remote work. The costs are real too: a second receiver, a mast, power, a radio pair with range and licence limits, a position that must be established accurately (an error in the base position becomes a constant error in every measurement you take), and someone to notice when it stops.

A network correction service. The reference side is computed across a whole region from permanent stations and delivered to you over the internet. Nothing to survey in, nothing to power, nothing to steal from a field gate — and no baseline limit, because the model is regional rather than a single point. The trade is a data connection and a subscription.

Northing RTK is the second kind, with the correction stream relayed to the receiver over Bluetooth by the phone you already carry — about 1 kbps, no cellular modem in the receiver, no second data plan.

What “centimeter accuracy” actually means

Three words get used interchangeably in this market and should not be.

• Accuracy — how close you are to the true position on Earth. Depends partly on the reference frame the corrections are computed in.
• Precision — how tightly repeated measurements cluster. A receiver can be precisely wrong.
• Repeatability — whether you get the same answer at the same place next week.

For guidance, setting out, and anything you will come back to, repeatability is what you are buying. A pass driven today has to line up with a pass driven in three weeks, and it is repeatability that decides whether it does. That is why the number quoted on this site is “±2.5 cm repeatable, RTK Fixed, open sky” rather than a bare “2.5 cm” — the condition is part of the claim.

Two more honest caveats. Vertical is worse than horizontal, typically by a factor of two to three, for everyone: satellites are above you, never below, so the geometry is weaker. And every accuracy figure assumes the antenna is where you think it is — a receiver on a wobbling mast is measuring the wobble faithfully.

What you need to actually use it

1. A multi-band GNSS receiver. Multi-band is not optional: comparing two frequencies is how the ionospheric delay is measured directly rather than estimated. Single-band chips — the kind in phones — cannot do it.
2. A correction source. Your own base, or a subscription.
3. A link to carry corrections, continuously. Radio, or an internet connection, or a phone acting as the relay.
4. Something to display and use the position — an app, a guidance system, a data collector.

How fast, measured

From our logged dual-receiver drive sessions:

• Fixed within seconds of a cold start — RTK does not converge, it fixes.
• ±2.5 cm repeatable accuracy while fixed, open sky.
• Fixed for 88% of a 40-minute drive that deliberately included tree cover, a bridge and buildings, with 1–2 second recovery after each obstruction.
• 24 seconds back to fixed after a full correction-stream outage, with about a minute of usable accuracy held unaided during it.

What RTK will not fix

RTK removes the errors that are common to you and the reference. It cannot touch anything local to your antenna:

• Blocked sky. No correction service can send you a satellite you cannot see. Canopy, cuttings, and yards remain hard.
• Multipath. Reflections off nearby metal or water are yours alone. Antenna siting matters more than any specification.
• A live correction stream that dies. Accuracy decays as corrections age. Worse, many receivers keep reporting centimeter confidence while it happens — we measured quality flags frozen while true error grew from 7 cm toward 40 cm, which is why every Northing receiver runs an independent watchdog and raises “Stale Corrections” after 30 seconds.
• A bad mount. Centimeter positioning of a bracket that flexes gives you centimeter knowledge of a flexing bracket.

Where the name comes from

Worth thirty seconds, because the name explains the design. Real-Time distinguishes it from post-processing, where you log raw observations and correct them later at a desk — cheaper and often more accurate, but no help to someone steering a machine now. Kinematic distinguishes it from static surveying, where an antenna sits on a tripod for an hour to average its way to a millimetre. RTK is the technique built for the case where the antenna is moving and the answer is needed immediately, and every trade-off in it follows from those two words.

Is RTK what you need?

If the antenna keeps moving and a wrong answer costs you a pass — guidance, planting, spraying, setting out — then yes, and the reason is time-to-accuracy rather than the accuracy number itself. If the antenna can sit still for a few minutes before the number matters, a convergence-based mode may be enough and cheaper; that comparison is laid out in RTK vs PPP.

For what it costs across the market, from five-figure bundles to bare boards, see how much RTK actually costs.

Common questions