Arduplane reporting noise on simulated NED data?

I’m currently trying to run a full SITL simulation stack in order to test some UAS autopilot capabilities.

Everything has worked spectacularly with the exception of the GPS and to a much greater extent, NED noise that appears to be coming out of ardupilot’s streams on ports 5760 (MavProxy) and 5763 (Mission Planner).

Here is a plot showing how noisy the throughput NED and GPS signals are:
Where NED is shown in blue, GPS is shown in green, and my attempt at filtering is shown in red. There are two of each on the same flight path (but different simulations) to show how non-random this noise is. As you can see there are definite patterns to the noise (It’s always aligned to the manhattan axes, indicating that there’s only ever noise in one of the three: north, east, or down) and it always happens in the same places. This makes me believe that this is way too weird to be an interpretation bug on my side (mis-flipping a received mavlink packet would result in noise in all three axes, not just one at a time).

Is anybody familiar enough with the Ardupilot/Mavlink/SITL stack to point me in the direction to go to in order to solve this noise? Is it programmed into Ardupilot intentionally and something I can shut off? Or does anybody else have any ideas as to how and why the noise is limited to the same places/times and always behaves in the manner indicated above?

1 Like

I’m not familiar with and the data you show. Can you tell me which MAVLink messages you are seeing noise in? Several messages have NED axes. Is it accelerometer? Or position data? Or something else?
We have quite a few SIM_* parameters for noise levels, if you can tell me which noise it is I can suggest what to change