Concept renderRTK GPS is a receiver mode that uses carrier-phase measurements plus live corrections from a base station, often streamed over NTRIP, to reach centimeter-level relative accuracy. A module such as the ZED-F9P needs a dual-band active antenna on a large ground plane and a data link that stays up. A tracker that only needs meter-level fixes can skip it.
RTK often lands in a tracker spec because someone saw a centimeter figure on a drone or a robot mower. This guide explains how it works, which accuracy figures vendor documents state, and what it costs a small battery device in power, antenna and data link. It ends with a decision table, a test plan and questions for your hardware team.
Figures come from public vendor and agency pages read on November 5, 2026. Two u-blox modules serve as worked examples because both publish full data sheets. This article contains no Duobay price and no test result of ours.
What is RTK GPS and how does it work?
A standard receiver measures its range to each satellite from the code in the signal. RTK (real-time kinematic) positioning also tracks the phase of the carrier wave. The receiver does not know how many whole carrier cycles lie between satellite and antenna. That unknown count is the ambiguity. The receiver resolves it by comparing its measurements with those of a base station at a known position, sent in real time.
The method works across constellations. The ZED-F9P data sheet says all satellites in view, across GPS, GLONASS, Galileo and BeiDou, can feed an RTK solution with correction data, as of November 2026 (ZED-F9P data sheet). That is why engineers also call it RTK GNSS.
Base, rover and baseline
The base is a fixed receiver that sends its observations and position as RTCM messages. The rover is the receiver in your product. The baseline is the distance between them. The rover needs base observations (MSM4 or MSM7) and the base position message (RTCM 1005 or 1006) before it can attempt a fix, as of November 2026 (ZED-F9P integration manual).
Error grows with baseline. The data sheet states RTK accuracy as 0.01 m plus 1 ppm, so by that formula a 20 km baseline adds 2 cm. The manual describes a single reference station within 50 km of the rover, or a network service that builds corrections for the rover location.
Float and fixed solutions
The rover enters RTK float shortly after corrections arrive. Once it resolves the ambiguities, it enters RTK fixed, where the manual says the relative position can be expected to be correct to the centimeter level. The time between the two is the convergence time. If phase lock is lost on too many signals, the rover drops back to float.
By default the rover also stops using corrections older than 60 s and falls back to a standard 3D or DGNSS solution. In NMEA GGA, quality 4 means RTK fixed and 5 means RTK float, as of November 2026 (integration manual). A tracker should send the fix type with every position, so the server never treats a float or fallback point as a centimeter point.
How accurate is RTK GNSS compared with a standard receiver?
Each row carries its test conditions, because a figure without them cannot be compared. CEP is the radius that holds 50% of fixes, so half fall outside it.
| Receiver and mode | Horizontal accuracy (CEP) | Stated conditions | Source |
|---|---|---|---|
| Standard L1 module (MAX-M10S) | 1.5 m | 50%, 24 hours static, signals at -130 dBm. | MAX-M10S data sheet, as of November 2026 |
| RTK module (ZED-F9P), no corrections | 1.5 m | 24 hours static. | ZED-F9P data sheet, as of November 2026 |
| RTK module with SBAS | 1.0 m | 24 hours static. | ZED-F9P data sheet, as of November 2026 |
| RTK module in RTK mode | 0.01 m + 1 ppm | 1 km baseline, patch antennas on good ground planes. The ppm term applies up to 20 km. | ZED-F9P data sheet, as of November 2026 |
| RTK module with PPP-RTK (SPARTN) | Under 0.06 m | IP stream over a low-latency link. Varies by service. | ZED-F9P data sheet, as of November 2026 |
The data sheet gives RTK convergence under 10 s with several constellations and under 30 s with GPS alone, depending on atmosphere, baseline, antenna, multipath and sky view, as of November 2026.
The RTK figure is relative to the base. The integration manual says any error in the base position translates directly into rover error, and that a base on the manual's suggested 60 s survey-in starting point is unlikely to do better than 1 m, even in good satellite visibility, as of November 2026. So a rover can be precise to centimeters against its base and still sit about a meter off on a map. For map accuracy, the base coordinates must come from a surveyed point or from a correction service.
Where do RTK corrections come from?
The rover needs a live stream of corrections. These are the four common paths, as described in the ZED-F9P documents as of November 2026.
| Source | How it reaches the rover | Notes for a device maker |
|---|---|---|
| Your own base station | A fixed receiver sends RTCM messages over your own radio or network link. | The vendor examples put base and rover within a few hundred meters. Map accuracy depends on how well the base position is known. |
| NTRIP service, single base | The rover pulls one reference station stream from a caster over the internet. | u-blox describes a single fixed reference within 50 km of the rover. |
| NTRIP service, network (VRS) | The service builds corrections for the rover location from many base receivers. | Coverage and terms vary by provider. Check every region you ship to. |
| PPP-RTK (SSR, such as SPARTN) | Corrections arrive over the internet, or over an L-band satellite channel with an extra receiver. | Lower accuracy than RTK in the data sheet, with a longer convergence of under 45 s. |
NTRIP stands for Networked Transport of RTCM via Internet Protocol. The BKG Ntrip page describes it as an application-level protocol based on HTTP/1.1, built from three parts: servers that send reference station streams, a caster that passes them unchanged to many clients, and the clients that receive them. Version 1.0 became an RTCM recommended standard in September 2004, as of November 2026 (BKG Ntrip page). On a cellular tracker, the NTRIP client can run in the host firmware and feed RTCM into the receiver's correction input. We have not read the RTCM standards themselves, so we quote no values from them.
In the US, the NOAA CORS Network is managed by NOAA's National Geodetic Survey, and its data are distributed free of charge, as of November 2026 (NOAA CORS). We found no sourced figure for the data volume of a correction stream, so measure bytes per hour on your chosen service before you size the data plan.
What does RTK cost in power, antenna and data link?
The table compares a standard L1 module with an RTK module from the same vendor, as of November 2026 (MAX-M10S data sheet, ZED-F9P data sheet, integration manual).
| Item | Standard (MAX-M10S) | RTK (ZED-F9P) | Design impact |
|---|---|---|---|
| Signal bands | L1 only. | L1 and L2, for example GPS L1C/A and L2C, Galileo E1 and E5b. | Dual-band antenna and RF path. |
| Antenna | Built-in LNA and SAW filter for passive antenna designs. | Active antenna mandatory. External gain 17 to 50 dB at RF_IN. | An extra LNA stage and antenna supply. |
| Patch ground plane | See our antenna placement guide. | Minimum 100 to 150 mm diameter. | Often the limit on enclosure size. |
| Signal level | No RTK line. | Average C/N0 above 40 for optimal RTK. | Measure it in the final housing. |
| Tracking current | 9.5 mA at VCC plus 2.3 mA at V_IO, continuous, 3.0 V supply, default GPS, Galileo and BeiDou. | 85 mA with four constellations, 70 mA GPS only. Peak 130 mA in acquisition. | Both from simulated signals. Measure the whole device. |
| Power save modes | On/off and cyclic tracking. | None listed in the data sheet. Software backup current 1.4 mA. | Plan for continuous running while RTK is in use. |
| Size and weight | 10.1 x 9.7 x 2.5 mm, 0.5 g. | 17.0 x 22.0 x 2.4 mm, 1.8 g. | The antenna sets the size more than the module does. |
| Data link | None needed for a fix. | Live corrections. By default the rover drops corrections older than 60 s. | The modem or another link stays up. |
In continuous tracking, 85 mA against 11.8 mA (9.5 plus 2.3) is roughly seven times the receiver current, before the modem that must stay connected for corrections. Treat it as a starting estimate.
The antenna is often the harder limit. If the enclosure is smaller than a 100 mm disc, a patch cannot get the ground plane the manual asks for inside it. This follows from the dimensions alone. Our GNSS antenna placement guide covers the standard L1 case in a small housing.
RTK also changes the radio plan. Many trackers save energy by letting the modem sleep between reports, and RTK needs a fresh stream plus time to converge after each start. If your budget relies on modem sleep, rerun it with the method in our PSM and eDRX battery life guide.
When does a consumer or asset tracker not need RTK?
Start from the job the position does. These rows are engineering reasoning from the figures above, and none describes a specific product.
| Use case | Accuracy the job needs | Likely answer |
|---|---|---|
| Find a lost pet, bag or bike | Meter level. | Skip RTK. A standard module lists 1.5 m CEP. |
| Geofence alerts with a radius of tens of meters | Meter level. | Skip RTK. |
| Reports every few hours, modem asleep in between | Meter level, lowest energy. | Skip RTK. It needs a live stream and time to converge after each start. |
| Indoors, in containers or inside vehicles | Any level. | RTK will not help. It needs strong open-sky signals. |
| Which yard row or parking bay holds a trailer | Meter to sub-meter. | Test SBAS or PPP-RTK on your site before a full RTK design. |
| Mower boundary, machine guidance, lane-level position | Centimeter level. | RTK is a candidate. Plan antenna, power and link from the start. |
Write the requirement as a number with a percentile and conditions, such as a CEP in open sky. A vague word like "precise" in a PRD tends to pull RTK, with its antenna, power and service cost, into a product that does not need it.
What should you test before you commit to RTK GNSS?
Vendor figures come from controlled setups. We have not run these tests for any product. Set each pass line before the first run, and run the same route with RTK off as a control.
| Test | How to run it | What to record |
|---|---|---|
| Fixed-solution availability | Run your real route with the final enclosure and log the fix type every epoch. | Share of time in fixed, float and fallback, by location. |
| Convergence time | Time from correction start to first fixed solution, after a cold start and after a short signal loss. | Time to fixed per run, with satellite count and C/N0. |
| Correction outage | Cut the correction link on purpose. | Time until fallback, and whether the payload marks those points as non-RTK. |
| Base position check | Compare the rover result with a point whose coordinates you know independently. | The offset, which shows any base position error. |
| Energy per report | Measure the battery rail from wake-up to last byte, correction download included. | Charge per report with RTK on and with RTK off. |
What should you ask your hardware team about RTK?
Get a written answer to each before the antenna and module are chosen.
- What accuracy does the product need, as a number with a percentile, and who wrote it down?
- Is that accuracy absolute (on a map) or relative to a base, dock or other device?
- How often, and for how long, does the device need a centimeter-level position?
- Where do corrections come from in each country we ship to, and who holds the account?
- Does the enclosure have room for a dual-band active antenna and its ground plane?
- What does the device report when it loses the fixed solution?
- Have we tested SBAS or PPP-RTK against the same requirement?
For how we scope tracker hardware, see LTE and GNSS tracker development.
Is RTK the same as differential GPS?
Both use corrections from reference stations. RTK also uses carrier-phase measurements and resolves their cycle ambiguities, which is how it reaches centimeter-level relative accuracy. The ZED-F9P data sheet lists RTK under differential GNSS using RTCM 10403.3 messages, as of November 2026 (ZED-F9P data sheet).
Can a small tracker use RTK?
The antenna is the main limit. The ZED-F9P integration manual asks for an active dual-band antenna and, for a patch, a ground plane of at least 100 to 150 mm diameter, as of November 2026 (integration manual). If the enclosure cannot hold that, plan an external antenna or ask the antenna vendor what its part needs.
Do I need my own base station?
How could Duobay help with an RTK decision?
A Gate 0 Feasibility study can turn your accuracy need into a written requirement and set the antenna, power and data-link budget and test plan for your product. We have no RTK test results to publish here, and we do not publish prices. Tell us about your device.
Sources. All accessed November 5, 2026. Check the current revision before you design.
- u-blox, ZED-F9P-04B data sheet, UBX-21044850 R05: https://content.u-blox.com/sites/default/files/ZED-F9P-04B_DataSheet_UBX-21044850.pdf
- u-blox, ZED-F9P integration manual, UBX-18010802 R16: https://content.u-blox.com/sites/default/files/ZED-F9P_IntegrationManual_UBX-18010802.pdf
- u-blox, MAX-M10S data sheet, UBX-20035208 R08: https://content.u-blox.com/sites/default/files/MAX-M10S_DataSheet_UBX-20035208.pdf
- BKG, Ntrip overview: https://igs.bkg.bund.de/ntrip/about
- NOAA National Geodetic Survey, CORS Network: https://geodesy.noaa.gov/CORS/
- GPS.gov, Augmentation systems: https://www.gps.gov/augmentation-systems
If you are deciding whether your tracker needs RTK, tell us about your device and the accuracy it must reach.
Planning a tracker? Tell us what it needs to do. A Duobay engineer replies within 1 business day. Send a project brief →