TL;DR: On boards with two IMUs running at different sample rates, VIBE levels are not comparable between them. Setting both to 1kHz (INS_GYRO_RATE=0) equalises the readings and, on my setup, cut reported vibrations by more than half.
I’m still in the process of figuring out the vibrations issues on my 10inch quadcopter, in the meantime I found some potentially interesting results in regards to vibration levels on Matek H743 slim v4 IMUs.
I’ve noticed, like a few other people already (
- Matek slim h743 v3 vibrations on imu
- Vibrations tab vs AccZ, Accx, 3d printed frame - #50 by dkemxr
- Why do I have two sets of vibe data? - #20 by Adam_Borowski
- 2 IMU Vibe Z values change between the two IMUs Matek 743 - #5 by Michail_Belov
) that vibration levels differ significantly between the two IMUs.
Here we can see for example. that vibration levels drop from 40-50 to 15-20 by looking at the second IMU instead of the first one.
On my board, both IMUs are ICM42688, so this was strange so I investigated further.
On the filter review tool we can see a similar effect on the gyro data.
We can also observe the second IMU sampling at 1kHz while the first IMU samples at 2kHz, which I assume causes a decline in gryo noise above 250Hz (though I’m somewhat confused by that given that I’d only expect frequencies above half the sampling frequency to be “modified” due to lower sampling rate.
It’s also interesting to observe the “stabilization” of the noise graph on the second IMU in the, for example, 150-275 Hz band, where orange curve is much smoother, I have no explanation for that.
I’ve also modified (not very trustworthy modification, mostly provided by LLM, so keep that in mind) filter review tool to provide frequency plots of acceleration data from ACC logs.
The peak noise frequencies appear to match between the accelerometer and gyro(ok, not that interesting, but might save work for others).
For the next test I then modified INS_GYRO_RATE and set it from 1 to 0 to force both IMUs to the same rate.
The VIBE graph now shows great overlap!
Gyro 1 noise data also looks smooth, but also not quite like the gyro 2 looked like in the earlier flight
I’m unable to plot gyro 2 with this modification for some reason.
So I guess there’s a bunch of data here, I still have some questions that I’d appreciate some input on:
- It’s “common knowledge” that vibes above 30 are generally an issue, but it seems from these measurements that vibe levels are highly dependent on the sampling frequency. Should the threshold scale with the sampling frequency?
- Should I lower sampling rate to 1kHz if I care much more about EKF quality/reliability than responsiveness and my vibration levels are borderline, given that this reduces vibe levels in logs?
- Why does gyro noise get lower at lower sample rate even at low frequencies?
- Why is gyro noise graph smoother at lower sampling rate?
Here’s the log with 2kHz/1kHz rate https://drive.google.com/file/d/13_zVxGtrVd5tWsdxOUNGjHTzwG06tJfC/view?usp=sharing
And the log with the 1kHz/1kHz rate
