Where the metres of error in plain GPS come from
Your phone has GPS too and it also shows you a dot on a map. The difference between it and a survey receiver is not that one of them "catches" more satellites, but what it does with the signal. A receiver works out its distance to each satellite from the time the signal takes to travel from orbit to the antenna. On the way, that signal passes through the ionosphere and the troposphere, which delay it differently at different times of day; the satellite clocks have small errors; and the orbits the receiver thinks it knows are not perfect either.
Add all of that up and you get an error in the order of metres. Fine for navigation. Not fine for a property boundary or a column axis.
The base and the rover see the same error
The idea behind RTK, Real Time Kinematic, is almost embarrassingly simple. You put one receiver on a point with known coordinates. It also measures wrongly, but since it knows where it is, it knows by how much it is wrong at every moment. It sends that difference to the second receiver, the rover, which is on the pole in your hands. The rover is a kilometre or ten from the base, so the signals reaching it pass through nearly the same atmosphere and carry nearly the same errors. Subtract them and what is left is a position relative to the base, good to a few centimetres.
That is why receiver specifications always say something like "millimetres plus ppm". The millimetres are the constant part. The parts per million grow with the distance to the base, because the further away you are, the more the atmosphere above you differs from the one above the base. Twenty kilometres out, you are no longer measuring the same sky. For urban and construction sites the base is usually a few kilometres away and that part is negligible; it starts to matter on long alignments and when one base serves a whole district.
The carrier phase
There is a second reason for the centimetres. A navigation receiver uses the code in the signal, which has a long "wavelength" and is therefore coarse. A survey receiver uses the phase of the carrier wave itself, which is about twenty centimetres long. It can measure exactly where within that wave it sits, but it does not know how many whole waves lie between it and the satellite. This is the ambiguity, the integer unknown.
Resolving it is the mathematics you pay for in the receiver. While it is searching, the rover shows "float" and the accuracy is at decimetre level. When it finds the answer it shows "fixed" and you are at centimetres. Recent receivers do this quickly; the A66 MAX, for instance, states initialisation under 5 seconds. But in poor conditions the solution can be lost and you drop back to float, and then the only right thing to do is wait.
Your own base or a network
Corrections can come from two places. The first is your own base, set up in the morning on a known point, with a radio link to the rover. It works everywhere there is no mobile coverage and does not depend on anyone's service. The downside is that you carry two receivers, and that radio range is something a mountain shortens without warning. If you run your own base, put it in the open and high up, away from metal fences and roofs, and leave it on the same point for the whole job. Moving the base at midday without a check is the classic way to end up with two halves of a site that do not match.
The second is a network of permanent stations, with corrections arriving over the internet via NTRIP. Then you take out only the rover, switch on mobile data and work. Networks compute a virtual base close to you, so the ppm part of the error stays small. The weak spot is the connection. In central Sofia that is not a problem; in a gully between two ridges it is. That is why we like a receiver that has both a modem and a radio. The FOIF A66 MAX has built-in 4G and a 1W radio, and the FOIF A90 has internal and external radio plus WiFi. The same crew can work on the network in the morning and on its own base in the afternoon.
What breaks RTK in the field
First, obstructions. A building, a rock face, dense forest overhead. Signals reflect off walls, the receiver gets the same wave by two routes, and that is called multipath. Sometimes you see it as a fixed solution with wrong centimetres, which is more dangerous than float because it looks right. When in doubt, measure the point twice a few minutes apart, or come back later when the satellites have moved.
Second, pole tilt. If you measure with a tilted pole and no IMU, the error goes straight into the coordinate. With IMU compensation, which both of our receivers have up to 60°, it is calculated out, but the IMU also wants to be initialised properly. Third, the operator. A wrong antenna height is the most common mistake we see, and it is entirely ours. Fourth, the time of day. Satellites are not spread evenly across the sky and there are periods when their geometry over a given spot is poor. The software shows this as PDOP; when that number climbs, accuracy drops, even though the screen still says fixed.
In short: RTK gives centimetres relative to the base, under clear sky and with the right antenna height entered. Each of those three conditions that is missing takes something off the accuracy, and it is you rather than the receiver who has to know it. The rest is in the GNSS receivers category.
