As you are suggesting, testing an additional Kakute mini board will be very helpful. If the board has uneven voltage supply to the sensors and lower rated power draw, an additional cap at the 3.3 voltage regulator prior to the gyro power supply could possibly be tested. Requires some tiny soldering. I have not seen this with the normal sized board, ie, the one with the ICM20689. I have only tested this FC on a larger sized copter.
I think I could do that… A small leaded cap over the top of the 3.3v reg should work. Looks like the IMU is on the top of the board and has the 3.3v regulator next to it. So, plenty of room for the cap and all. That’s easier than replacing the board, so maybe I’ll start w/ that.
Have you guys actually looked for clipping in your logs? My latest flight with fast sampling on for example, had no obvious issues, no ekf warnings, etc, so I wouldn’t have known about the 90,000 clips if I wasn’t looking at the graphs.
I was going to ask you about this. With that sensor “fast sampling” means a 1kHz HW filter on the accels and a 250Hz filter on the gyros. The gyros are further filtered in software but the accels are not for the EKF. Instead the EKF is supposed to “learn” the noise through it’s state variables. So although the logs look bad, I wonder whether they really are since there is no obvious vibration on the gyros at all - which makes me think it’s all artificial. That said your learned AZBias makes it look like the EKF is struggling so artificial or not I suspect it’s a problem.
Without fast sampling the accels have a HW filter at 220Hz - I just wonder if putting some kind of SW filter at this frequency would help. Everyone will tell you that’s a big no-no for an EKF, but I’m more of a try it and see kind of guy 
Right, ok. So a ‘clip’ event is when the magnitude of the acceleration is outside the sensor’s measurement range? (as opposed to the frequency of the vibration being outside of the sensor’s measurement range?)
1 Like
_clip_limit = 15.5f * GRAVITY_MSS
I installed a cap on the 3.3v reg next to the IMU, and initial results are very good… (Just holding in my hand full throttle).
I’m charging up for a flight now.
It did still clip hundreds of times, but before I was getting thousands…
Flight will tell us more… Be back in about 20 min…
2nd results looking good… But a few seconds after I took off the FC rebooted while I was in the air… Everything looked ok, so I flew around again w/out any problems, but just over my yard… SO, checking over everything again, recharging, and maybe I’ll be brave enough to do my usual flight around the area so I can compare it to previous flights.
Vibes are all around 5-10. I’m getting hundreds of clips… it was up to about 900, but before the cap, I bet it would have been 10x that.
It’s only marginally better… I’d say not enough to be sure the cap is actually helping.
Still, like my last few flights w/ all my vibration mitigation, I didn’t have any problems. Vibes peak at about 25 instead of 30. Clipping about 30,000 times instead of 90,000. But, looking at all my logs, I think that’s w/in the range I was getting before.
New FC just arrived. I’m thinking it won’t make a difference… But maybe I’ll still try it.
Can you try on 3s before you put it in a frame?
Already finished…
This one is the same… Again, no problems w/ the flight, but about 30,000 clips.
Vibes peak to about 30, average is x=4 y=9 z=2.
No clipping w/ fast sampling off.
I have a 3s battery, but not one this copter can lift… Not sure how much data I could get just on the bench. But now I’ve got the originial FC sitting here, so can do testing w/ it.
As far as this copter goes, I think it’s fine w/ fast sampling off. Maybe it’s fine w/ it on and I can ignore the clipping, but not sure. And otherwise, I’ll probably rebuild it some day and try soft mounting it on gel pads. Not for a while, though… I’ve rebuilt this thing way too many times the past few days :).
I’m happy to do more testing and supply logs, if you think the clipping might be a problem in the code you want to figure out.
Whow, that’s no good.
Thanks for sharing the information you have gathered from the tests. You know your soldering, wicked1!
I’ve briefly looked at the backend drivers. It seems as noise increases at the higher filter bandwiths, but I’m not sure if I am interpreting these tables correctly.
The reasoning is the benchmarking you have done against the mpu6000. Gyro noise is similar for the ICM, but accel noise should be significantly lower from looking at the datasheet. I understand from your tests that the EKF handles the accel data well at the lower filter setting. I.e., there should be no reason for the increased noise, even artifacts, with the ICM (with all other variables otherwise the same).
“I think some of this is because for sensors that only support sampling below 32k we can essentially pick the unfiltered option, but for sensors where the unfiltered option is 32k we pick 8k with a lot of filtering rather than 8k with little.”
This would be a good example to test this, with a small copter, (if the board is ok and if this is how the IMU behaves with the current config).
As noted, I have seen little problems with larger copters running the normal sized Kakute F7.
So overall flight performance is similar to the nano?
Yes I agree. @wicked1 I’d be happy to do a build for you that allows you to configure an anti-aliasing filter in software if you want to try. I’m really not convinced these are real vibrations.
Ceitean,
Yes the performance seems the same as the nano.
Andy,
I’d be happy to experiment.
Wouldn’t extra filtering eventually stop the benefits of the faster sample rate? I don’t know the math, so maybe that’s not an issue for the sample rates and filter frequencies we’re using…
Here’s another log… The one I posted at the beginning of this thread was particularly bad. Now I’ve got my better balanced props and secured all my antenna, etc… None of that was necessary w/ the nano… but anyway, it’s much better than it was in my original log.
Interesting, now it doesn’t look like vibes - just high offsets. There is a clear correlation between throttle and offset which makes me still think it’s power related:
Unless you are accelerating at a constant 5m/s/s this has got to be artificial
Andy, more data related to the kakute mini. Continuing from your notch filter thread. While it’s related to vibrations and filters, it seems a bit off topic for your thread.
I did a basic 1.5 minute hover inside, and the logs look pretty good. Definitely don’t have the vibration issues I do outside in a real flight. Something interesting on the accels…
I’m posting this mostly just for the sake of putting info out there. But if you have any ideas about why my accel pre-filter looks the way it does or how to mitigate it, I’d be happy for any advice.
There are very large peaks that I’m not sure are related to motor noise… Looks like harmonics, so good for the harmonic filter… But not sure if I should target them or the motor noise… (Maybe I need two harmonic filters :)…)
Post filter top, pre filter bottom, vibes right.
Also, I can see my static notch at about 90hz in the pre-filtered FFT on the gyro. Not sure what’s going on w/ that. If I disable that notch, the peak there is bigger.
I’ve been looking at the ICM20689 docs - I think there are a few configuration options related to noise that we should experiment with.
Last night I logged disarmed and tapped the frame, and the FFT of tapping it has all the same noise peaks. So this frame really does have some resonance. The 2nd harmonic of the frame resonance is right at the motor noise frequency. There are probably some interactions there…
So, I guess what we are seeing is real noise. It’s noise the mpu6000 couldn’t see, but this ICM20689 can. (or maybe the kakute board has resonance around the same frequencies, so is more susceptible to the noise in this copter) Whatever the reason, I think this copter is ringing like a bell. I can hear it when I tap the frame.
I’ve got a few copters on the same frame, and I can see the same noise peaks on all of them… With the omnibus/mpu6000 the peaks are tiny… They’re pretty substantial w/ the kakute.
I don’t know why it seems to only affect the accel’s… The gyros look similar between the omnibus and kakute, but the accels are a mess in my kakute logs.
Anyway, I guess there’s not a problem w/ anything… I’ve just noticed some differences between the IMU’s. While I’ve been searching for info over the past several days I’ve noticed people in the betaflight forums discuss it frequently. I’ve seen some manufacturers release FC’s with an option for either IMU and in their sales literature says something like “if your copter is too noisy for the icm20689, you can select the mpu6000 option”.
If you do want to try some things, though, let me know. I’m happy to do the testing. This particular copter might be a good test platform if you want to make it work with noisier vehicles…
Cool.
I was going to ask about your frame - I am using an Armattan Gecko 3" frame which is very stiff carbon fibre and I don’t think has much resonance. If you have a softer frame I can believe the problem is worse.
Also I am using these: https://www.quadcopters.co.uk/gemfan-flash-3052-propellers - which are much better from a vibration perspective.
Can you post a picture of your build?
Yeah, it’s a bit of a mess since I’ve been rebuilding it so frequently the past several days.
The frame is ultra light, so is a little flexible. I guess this is the compromise with a light frame…
I’ve been working with two different copters since I have two boards now, but this is the one I’ve been using for 99% of our discussions… (The two are basically the same… Different ESC’s and cameras, but everything else the same. Vibrations look identical between them)
Looks cool. Can you wire the runcam into a UART? Then you could try my runcam driver
Yes… Plenty of IO on this kakute mini