Hi!
I’m trying to generate a Notch Filter for my Arducopter setup. It’s a moderate sized build with 14" props. However, when I use the webtool to find the appropriate values, I get a graph that looks like this:
As you can see from the pre filter values, it is mostly steady around the -75dB range. The suggestion from the online tool gives me a pair of notch filters, at ~ -62 and -49 dB which look unnecessary, randomly introducing a pair of down-peaks in the post filter data. Based on this, should I use these notch filters?
Additionally, the post filter data shows my Gyro values dropping in a negative exponential graph. Is this normal?
EDIT: After using a Low-Pass Filter of 500 (!) hz, I get a graph that looks like this:
Yes indeed the vehicle is flying really well. Slightly unstable on the descent is all. I haven’t enabled any notch filters at all at the moment, but I do want to add them before starting the autotune.
As @amilcarlucas said, if that log you’re showing us is from flying data, then you don’t need a filter. Everything is well below -50dB so there’s really nothing to be gained from the filter.
This was an altitude test I did this morning where I went up to 250+m and came back down as fast as Arducopter let me. I recall Mission Planner reading something like 2.5ms^-1 for the altitude velocity.
(Had to be a Google Drive link since the file was too large)
I see. This is a perfectly fine configuration for an AutoTune then? Would even a low-pass be worth here? I manage to get a decent roll-off on the 200+hz by just having a 40hz Low Pass.
I’ve never seen a graph so flat. That’s good, but maybe too good. You want as little filtering as possible and looking at that log I wouldn’t use any.
I did notice another issue. There seems to be a really big torque split. Normally that means there is an unwanted twist in the airframe. What is strange here is that the torque split changes, but I’m not seeing yaw inputs to follow. Actually I just looked and for several of these splits the only input is throttle. You should make sure all the motors are straight and level with one another.
It’s odd that you say that. Ordinarily, I’d have my Carrier board mounted via a TPU bed and then mechanically isolated with rubber dampeners; in this case, I’ve just directly mounted my Kore Carrier Board to my airframe. In any case, I’ll leave it alone and try an autotune.
Post flying, I noticed a slight twist in one of the arms, ± 10 degrees of rotation. Perhaps this be the cause? I’ve fixed it completely now and hoping for some decent weather out , so I’ll send that log once I’m able to get out there.
Do you have any recommendations for doing so for Motor mounts clamped to a 16mm tube? I find that it’s really hard to make it perfectly level and then knowing once I get out to the field, they’re all slightly off.
Side note, what Python Program are you using to analyse those logs?
I’ll try that once I get the chance. What does this do?
If you want to log raw IMU data then follow this guide. Raw IMU Logging Otherwise disable Raw IMU in the log bitmask and use fast harmonic notch logging as suggested.
Yup, that would do it. You can look at the tips of the props and from one to the next they should be even. I’ve also take the props off and used the mounting surface of the motors with a long straight edge to get them level. Be detailed, a little twist can create a big split.
Completely forgot to enable Harmonic Notch logging, but this new log is a lot more comprehensive of a flight. I fixed the torque split I think. Much more stable, especially on deliberate swings. I find that the Flowhold mode was terribly unstable however, but I’m not too sure what the cause there was; I think I was just too high for the Optical Flow to work well.
I find that there is now lots more vibrations above -50dB. Is this cause for a notch filter?
I’ll fly again in a couple of hours (After the battery charges) and see what happens when I enable harmonic notch logging.
That log needs a password. But get a log with the logging for the filter, that will be more helpful now. Flowhold isn’t used much any more, most will use loiter and change the position source from GPS to Optical flow. How well the optical flow is working will show up in the log.
You wouldn’t believe it but I’ve literally got my Cube Black (Waiting for a Cube Orange!!) to my Kore Carrier board DIRECTLY mounted to my Tarot Ironman 650 frame. In between are just nylon standoffs, but I’m considered printing a set of TPU ones.
Although the vehicle does fly in the log, the lovely flat line is because the log is missing the data for the flying portion. The easy way to check is to look at the spectogram.
Lots of white means lots of missing data. You need to reduce the logging rate, you might find that going to INS_RAW_LOG_OPT rather than raw sensor in the bitmask is enough. That will remove logging of accels. Other wise batch logging.
I see. The graph looked very wrong from the beginning hence my posting here - too good to be true!!
I’ll see and upload what happens when I enable Fast harmonic notch logging (That’s the one right?)
