RTK GPS: Centimetre Accuracy in Real Time

For everything that needs centimetre-level positioning outdoors — property surveys, infrastructure as-built records, precision planting, autonomous-vehicle navigation, scientific monitoring — RTK GPS is the standard answer. This article covers how RTK works, the variants (classical RTK, network RTK, PPP-RTK), the hardware and infrastructure required, and the use cases that make the investment worth it.

The /learn/differential-gps article covers the classical DGPS technique RTK is built on; this article goes deeper on the carrier-phase upgrade.

The carrier-phase advantage

Standard GPS uses code-based pseudoranges: the receiver measures the time-of-arrival of the satellite's modulated ranging code, multiplies by the speed of light, and gets a distance. The precision of the code-tracking is limited to about 1 metre (the C/A code chip rate is 1.023 MHz, so each “chip” is ~300 m of space; the receiver can interpolate to ~1 m).

RTK uses carrier-phase observations: it measures the precise phase of the GPS carrier wave itself. The carrier is at 1575.42 MHz (L1), with a wavelength of about 19 cm. Modern receivers can measure carrier phase to a few millimetres of the wave's position. That's 100–1000× more precise than code-tracking.

The catch: carrier-phase measurement only tells you the fractional phase at the receiver, not the integer number of whole cycles between the satellite and the receiver. This integer ambiguity must be resolved before centimetre-level positioning is possible.

Integer ambiguity resolution

The RTK algorithm uses several observations (multiple satellites, multiple frequencies, repeated measurements over time) to solve for the integer ambiguities. The standard algorithm is LAMBDA (Least-squares AMBiguity Decorrelation Adjustment), developed in the 1990s and still the workhorse of modern RTK receivers.

Convergence typically takes:

• A few seconds for a dual-frequency receiver under open sky with good satellite geometry.
• 10–30 seconds in slightly more difficult conditions (some multipath, partial satellite view).
• 1–5 minutes in challenging conditions (urban canyon, light canopy).
• Never if the conditions are bad enough — the integer ambiguity can't be resolved at all, and the receiver reports a “float” solution (typically decimetre-level accuracy) instead of a “fixed” solution (centimetre-level).

The user-facing RTK report distinguishes between RTK Fix (integer ambiguities resolved, 1–2 cm accuracy) and RTK Float (integer ambiguities partial, 10–50 cm accuracy). Most surveying work waits for RTK Fix before recording a point.

The base-rover architecture

Classical RTK uses a dedicated base station with a known surveyed position:

1. Base station: a GPS receiver at a surveyed point (typically a permanent geodetic monument or a temporary tripod-mounted setup over a known control point) tracks all visible satellites and broadcasts carrier-phase observations + ephemeris data to the rover.
2. Communications link: between base and rover. Historic options include UHF radio (good for line-of-sight short distances); modern preferred is cellular internet (via NTRIP protocol) or LTE / 5G modem.
3. Rover: the moving GNSS receiver. Tracks the same satellites as the base; applies the base's observations to resolve integer ambiguities and compute its own centimetre-accurate position relative to the base.

The rover's position is computed as a baseline vector from the base, then added to the base's known coordinates. The result is the rover's absolute position in the same reference frame as the base.

The ~30 km baseline limit

RTK accuracy depends on the spatial-correlation assumption: the rover and base see essentially identical atmospheric and orbital errors. This holds well for baselines under about 30 km, beyond which:

• Ionospheric variation between base and rover starts to matter. A solar storm or sub-storm can produce wildly different ionospheric delay over 100 km, breaking the correction.
• Tropospheric variation: weather fronts, height differences (a base in a valley vs a rover on a hilltop).
• Ephemeris decorrelation: the broadcast ephemeris error, while small, doesn't cancel exactly between distant receivers.

At 50 km baseline, RTK accuracy degrades to 5+ cm. At 100 km, 10+ cm. The 30 km threshold is the conventional rule of thumb for sub-2-cm accuracy.

For longer baselines, the alternatives are:

• Network RTK / VRS (covered below)
• PPP-RTK — uses precise satellite products to remove ephemeris and clock errors, leaving only spatial-decorrelation in the atmosphere
• Multiple base stations with weighted combination

Network RTK (VRS)

Network RTK uses a network of CORS stations covering a region. The rover's approximate position is sent to a server; the server interpolates between nearby CORS stations to model the local atmospheric and orbital errors and generates a “virtual reference station” (VRS) correction tailored to the rover's position.

The user experience: the rover gets corrections that look like they come from a base station at the rover's exact location, even though no actual base station is nearby. The network does the spatial interpolation.

Coverage: anywhere within the CORS network (which is dense in most developed countries — North America, Europe, Australia, Japan have nationwide CORS coverage). Accuracy: sub-2 cm with proper network geometry.

Major network RTK providers:

• Trimble VRS Now — commercial, global coverage
• Leica HxGN Smart Network — commercial, Europe + Americas
• Topcon TopNET — commercial
• NOAA NGS — free public service from US CORS network
• EUREF EPN — academic-research network covering Europe
• Various national agencies in Australia, Japan, Korea, etc.

For commercial surveying work in well-covered regions, network RTK has largely replaced classical RTK because the user doesn't need to set up a base station — they just subscribe to the network.

PPP-RTK and convergence-free alternatives

A relatively new development: combining PPP (Precise Point Positioning) with RTK-style ambiguity resolution. Uses globally-published precise satellite products (clocks and orbits) without needing a local reference station, while still achieving the integer-ambiguity fixed solution.

Accuracy: 1–10 cm globally, with convergence time of seconds to minutes depending on the implementation.

Major PPP-RTK services:

• Galileo HAS (High Accuracy Service) — free, broadcast directly via Galileo satellites, sub-20-cm accuracy
• Trimble RTX — commercial, sub-5-cm accuracy
• NovAtel TerraStar — commercial

PPP-RTK is the modern frontier: it bypasses the base-station infrastructure entirely while approaching RTK-level accuracy. For applications where setting up a base or subscribing to a network is impractical (offshore platforms, remote agricultural work, single-vehicle deployments), PPP-RTK is increasingly the preferred solution.

Use cases

Where RTK is the right answer:

• Land surveying — property boundaries, cadastral surveys, geodetic control. Sub-2-cm accuracy is the regulatory standard for boundary work in most jurisdictions.
• Construction stake-out — setting out building corners, road alignments, utility positions. RTK rovers replace optical surveying equipment.
• Precision agriculture — automatic-steered tractors plant, spray, and harvest with rows aligned to within centimetres. Saves 5–10 % on fertiliser and seed costs across large fields.
• Autonomous vehicle HD mapping — building high-definition maps used by self-driving cars requires lane-level positioning of every road feature.
• Infrastructure as-built records — recording the precise position of installed pipes, cables, foundations for future maintenance.
• Scientific monitoring — tracking glacier flow, tectonic-plate motion, subsidence at millimetres per year. Repeat RTK observations over months capture the drift.

For consumer applications (smartphone navigation, fitness tracking, recreational hiking), RTK is overkill — standard multi-GNSS positioning at ~5 m is sufficient.

Hardware and cost

RTK hardware varies by tier:

• Survey-grade RTK rover: $5,000–$20,000 USD for the receiver plus controller. Brand examples: Trimble R12i, Leica GS18, Topcon Hiper VR. Includes dual-frequency multi- GNSS receiver, RTK firmware, NTRIP client, rugged enclosure.
• Mid-tier RTK (precision agriculture, mid-range survey): $1,500–$5,000 USD. Brand examples: u-blox ZED-F9P with a smartphone, Emlid Reach RS+.
• Hobbyist / open-source RTK: $500–$1,500 USD. Boards like u-blox ZED-F9P, ArduSimple, Drotek, with open-source RTK software like RTKLIB.
• Smartphone RTK: a few premium smartphones (Samsung Galaxy S, Google Pixel) now include dual-frequency GNSS chips that, combined with a paid RTK correction service (subscription ~$30/month), can achieve decimetre-to- centimetre accuracy in apps like SwiftNav's Skylark. The boundary between “survey-grade” and “smartphone” is blurring.

The biggest barriers to RTK adoption for general use are historically hardware cost and the need for a base station. Network RTK + cheap dual-frequency receivers + smartphone integration are gradually removing both.

Common misconceptions

“RTK works anywhere with cell signal.” Network RTK does — anywhere covered by the CORS network. Classical RTK needs the rover within ~30 km of the base station, regardless of cell signal. Some remote areas (large parts of the western US, parts of Africa, much of Siberia) aren't well-covered by network RTK; PPP-RTK fills the gap.

“RTK is instant.” No — even under good conditions, the receiver needs several seconds to converge from “float” to “fixed” solution. In challenging conditions, the fix may never be achieved. RTK quality is about how often you get a fix and how fast it converges, not just the accuracy when fixed.

“You can use any GNSS receiver for RTK.” RTK requires dual-frequency carrier-phase tracking, which is a specific hardware capability. Single-frequency receivers (most smartphones until ~2018) can't do RTK at all. Modern multi-band receivers (u-blox ZED-F9P, smartphone chips since 2022) can.

“Network RTK is always free.” Free public network RTK is available in some regions (US via NOAA NGS, parts of Europe via national agencies). Most commercial network RTK services are paid subscriptions ($20–$100/month per rover). For commercial surveying work, the subscription cost is typically a small fraction of the labour cost.

“RTK is just for surveyors.” Precision agriculture and autonomous vehicles are the biggest growth markets for RTK; surveying is the historical use case. Self-driving cars increasingly use RTK + inertial sensors for lane-level positioning where the HD map demands it.

“PPP-RTK replaces all other forms.” Not quite — PPP-RTK is converging on RTK accuracy but currently has slower convergence (especially in challenging conditions) and slightly worse precision. For survey-grade work in well-covered network RTK areas, classical or network RTK remains the standard. PPP-RTK is the right answer when no local network is available.

“RTK only works outdoors.” Largely true — RTK requires line-of-sight to GNSS satellites, so indoor positioning falls back to other technologies (UWB, beacons, inertial). But some “outdoor-adjacent” situations work surprisingly well: a vehicle moving slowly through light tree canopy, a tractor under a partial barn roof, a surveyor working in an urban courtyard. The receiver tracks whichever satellites it can see; partial sky coverage often still yields a fix, just with worse geometry. Pure-indoor RTK is impossible without ground-based pseudolite infrastructure (rare except in specialised airport / mining applications).