SKYLAB provides technical support for GNSS modules, and 'drone drift' is the issue we encountered.
When it comes up, the usual description is that while hovering the aircraft slowly slides to one side. The pilot's first instinct is to tune the flight controller — recalibrate the IMU, check the propellers and motors, and go through the control parameters again. None of that is on the module side.

We usually start by asking them to hold the aircraft still on the ground and record nothing but the module's coordinate output for ten minutes.
If the airframe has not moved but the coordinates keep creeping in one direction, the position solution is drifting, and the control loop has nothing to do with it. In the field, an unstable control loop and coordinate drift are both called "drift", but they lead to different checks.
By the relationship between the airframe and its coordinates, "drift" falls into three types. Below we discuss only coordinate drift.
If the aircraft really is moving and the coordinates move with it, the aircraft is not under control — that is an attitude and control-loop matter. If the aircraft is still and the coordinates move, that is position-solution drift: in a static test it shows up as a slow one-way offset in the coordinates, and in flight as a uniform shift of the whole track. The third type is a combination of the two, which a static test cannot show.
How far the coordinates drift can be measured. Leave the module stationary for several hours and compute CEP50, CEP95 and maximum drift; then repeat with the same board at a site with different sky visibility and compare the two sets of numbers. With the airframe fixed, what you measure is the drift of the coordinates themselves.
Before the static measurement, note which output you are using and what update rate is configured. Take the SKM7506T, our module for drone applications, as an example: it outputs both NMEA 0183 and NAV-PVT adapted for drone flight controllers. The NMEA stream is what engineers read in the logs — GGA fix quality, the satellites used in the solution from GSA, and C/N0 from GSV are all there. NAV-PVT carries the position, velocity and time solution. To tell whether the coordinates are drifting, compare readings over time within one stream only; do not match NMEA coordinates against NAV-PVT coordinates.
C/N0 alone cannot tell you whether a satellite's signal is corrupted
Near trees, glass curtain walls or power lines, reflected and direct signals add together: out of phase, C/N0 shows a clear dip; in phase, it can actually be pushed higher. So using a single absolute value to decide whether a satellite's signal is corrupted misses the in-phase cases. Read it comparatively instead: in the same environment and on the same module model, first compare satellites against each other, then look at the pattern of one satellite's C/N0 over time, and finally match the azimuth of the suspect satellite against the azimuth of the obstruction on site. Only record it as a multipath suspect if they line up.
A higher satellite count does not mean a more accurate position.
Position error is roughly DOP × ranging error: DOP is the amplification factor from satellite geometry, while ranging error comes from multipath, obstruction and receiver noise. Satellite count only tells you how many are available; whether the geometry is good is a question of DOP. If the count is high but HDOP is large, check the actual list of satellites used in the solution in GSA first. GSV reports satellites in view and GSA reports the satellites actually used in the solution, so the two counts are not expected to match. First see how many of the visible satellites made it into GSA, then read that together with DOP.
When you see "missing satellites", do not assume a signal problem. Each NMEA 0183 GSV sentence carries at most four satellites; with multiple constellations the output is split across several sentences, so dropping one sentence loses several satellites. A capture window that is too short also makes it look like satellites are missing. Confirm that every sentence was captured before concluding that satellites were lost.
Losing RTK corrections does not mean an immediate drop to meter-level accuracy.
When the correction link breaks, the solution degrades in three stages: RTK fixed → RTK float → single point. The float state in between is easily mistaken for flight-controller drift — the position moves slowly without jumping, but accuracy is already declining. Tuning control parameters will not change the position solution. The GGA fix quality field reports these states directly: 4 = RTK fixed, 5 = RTK float, 1 = GNSS fix (single point). Tracking how it moves between these values tells you which stage you are in.
How the flight controller flies the aircraft is not ours to decide. What we deliver on the module side is the coordinates, plus NMEA logs that record the conditions under which those coordinates were computed.
The module we recommend for drones is the SKM7506T. It integrates a quadrifilar helix antenna and the RTK engine into one module, removing one RF feed line from the airframe. It receives L1 + L5 dual-frequency signals from GPS, BeiDou (including BDS-3), GLONASS, Galileo, QZSS, IRNSS and SBAS, with 200 tracking channels. Outputs include both NMEA 0183 and NAV-PVT adapted for flight controllers; RTCM3 correction data is fed in through RX1, and the position update rate can be set to 1, 2, 5, 10 or 20 Hz.
With an RTK fixed solution, horizontal accuracy is 1.0 cm + 1 ppm and vertical 1.5 cm + 1 ppm; without corrections, single-point positioning has a horizontal accuracy of 1.0 m CEP and a vertical accuracy of 2.0 m CEP. The two modes are roughly two orders of magnitude apart in accuracy. Neither accuracy figure applies to the float state, so when you run the statistics, segment the data by GGA fix quality first and compute each segment separately.
Next time a drone seems to be drifting, do not start with the flight controller: check first whether the coordinates still move with the airframe fixed, then read the GGA fix quality on the same output stream.
Next:No Next Page
Why Does a Drone "Drift"? The Answer May Not Lie in the Flight Controller
2026-11-18Giving Livestock a "Smart ID Card": This Bluetooth Ear Tag with a 2.5-Year Battery Life Makes Herd Location & Management Easier
2026-11-11Replacing the GNSS Module and Antenna Did Not Improve Accuracy: A Seven-Step Troubleshooting Record
2026-11-04
Copyrights© Shenzhen Skylab Co.,LTD All Rights Reserved.