Centimeter GPS over a cheap LoRa radio link

Why ordinary GPS is not enough

A standard satellite receiver works out its position from code-phase, or pseudorange, measurements. Positioning company Point One Navigation describes these as accurate but limited to meter-level precision, and gives 3 to 10 m as typical for standard GNSS. That is fine for a car map. It is useless for jobs where a stake has to land in the same spot every time, such as surveying, machine guidance or laying out a field of trees.

Real-time kinematic positioning, or RTK, closes that gap. Instead of relying only on the code, an RTK receiver also tracks the phase of the radio carrier wave itself, which Point One says gives roughly 100 times more measurement precision. Under good conditions it quotes 1 to 3 cm horizontally and vertically.

How a base and a rover cancel the error

RTK needs two receivers. A base station sits still on a known point, and a rover moves around the work site. ArduSimple, which makes low-cost RTK boards, explains the idea simply: two receivers close to each other see roughly the same errors, so the one at a known position can work out the error and tell the other one. The base packages that information as RTCM correction messages, which ArduSimple says are usually sent once per second.

There is a catch in the carrier phase. The GPS L1 carrier has a wavelength of about 19 cm, so the receiver can measure where it is within one wave very precisely but must still work out the whole number of waves to each satellite. Getting that count wrong by one cycle adds about 19 cm of error. While the count is unresolved the receiver reports a float solution at decimeter level; once every count is resolved it reports a fix with full centimeter accuracy.

Corrections only hold near the base. ArduSimple says they are valid up to about 35 km, with some users citing 25 or 50 km, and that beyond that accuracy falls back toward standard levels. The corrections can travel by cable, radio, or over the internet using the NTRIP protocol, which needs a network connection.

A field test with LoRa as the link

A 2026 paper from Universitas Lampung in Indonesia, published in the Journal of ICT Research and Applications, built this kind of system for laying out oil palm plantations, where workers normally use stakes, ropes and tape. Plantations often have no cellular coverage, so the authors skipped NTRIP and sent RTCM corrections from a local base to the rover over LoRa, a license-free, low-power, long-range radio. Both ends used ArduSimple RTK2B boards built around the u-blox ZED-F9P receiver. Planting grids were drawn in GIS software over drone orthophotos and loaded into a phone app that guided the operator to each point.

Against a surveyed benchmark the rover showed a horizontal error of 8.95 cm. In planting trials the mean error was 8.40 cm for a square layout of 15 points and 7.78 cm for a triangular layout of 18 points, both inside the 10 cm tolerance the authors set.

What is hard: keeping the link alive

The weak point was the radio, not the receivers. The team streamed corrections for 60 seconds at each distance from 10 m to 100 m. Every packet arrived at 10 m, but delivery fell steadily to 78.26 percent at 100 m, with packet loss passing 20 percent beyond about 90 m. Errors in the planting trials also tended to grow with distance from the base, most clearly in the triangular layout.

The paper does not report which RTCM messages were sent, the LoRa data rate, transmit power or latency, so the short range cannot be pinned on any one setting. The authors suggest moving the base closer, adding relays or a LoRa mesh, using several bases, or falling back to NTRIP where coverage exists. They also note that their flat orthophoto design ignored slopes, and plan to add elevation models.

The takeaway is a useful one for anyone building cheap precision positioning: the receiver can deliver centimeters, but only for as long as a steady stream of corrections actually reaches it.