MicroArduCopter, 3" props, Omnibus Nano, Success!

andyp1per · July 23, 2020, 8:34pm

I’ve also added https://github.com/ArduPilot/ardupilot/pull/14891 which could possibly make the FFT a little more accurate.

anbello · July 24, 2020, 9:20am

I did an outdoor flight with my 3 inch copter with omnibus v6 nano and small-copter-4.1 (built 5 days ago) with gyro_rate 1KHz (so nothing special).

I had Potential Thrust Loss errors, never seen before, where do I have to investigate? Battery, ESC, …?

Here the log:
https://drive.google.com/file/d/1KMcnSnuUzRrfgFW86CwTiytZ9y8FSpnh/view?usp=sharing

[Edit]
As battery I use 2 18650 Li-ion cell (Sony / Murata US18650VTC5 2600mAh - 30A), they could go down to 3V each without problems.
Maybe they are a little bit old, too much charge - discharge cycles?

andyp1per · July 24, 2020, 12:04pm

It must be because your desired alt deviates so much from the actual altitude. Did you check that the baro alt is the same?

anbello · July 24, 2020, 1:02pm

BAlt is similar to Alt (obviously with more error).

andyp1per · July 24, 2020, 1:03pm

Did the RCOUT drop at the same time or stay high?

anbello · July 24, 2020, 1:09pm

@andyp1per returning to your last PR on gyro rate I tested your last small-copter-4.1 (as of today) with my other 3 inch copter with revo-mini, same config as post #871

INS_GYRO_RATE = 1
FFT_WINDOW_SIZE = 64

[Edit]
Changed image, it was the wrong one.

anbello · July 24, 2020, 1:15pm

I see strange behavior on RCOUT in the “falling” zone

andyp1per · July 24, 2020, 1:23pm

That looks pretty good to me! The latest update removes the 188Hz pre-filter. You can probably squasuh the remaining peak more by using the double-notch but you may not have enough CPU for that

anbello · July 24, 2020, 1:29pm

That’s what I fear

fill for 20 char min

andyp1per · July 24, 2020, 1:32pm

You might be ok - you only need two peaks and two notches on each - so four total. Remind me what FFT options you are using?

andyp1per · July 24, 2020, 1:33pm

No idea what is going on here

anbello · July 24, 2020, 1:39pm

Here the FFT params:

FFT_ATT_REF,15.00000000
FFT_BW_HOVER,131.06542969
FFT_ENABLE,1
FFT_FREQ_HOVER,341.35119629
FFT_HMNC_FIT,10
FFT_HMNC_PEAK,0
FFT_MAXHZ,480
FFT_MINHZ,180
FFT_SAMPLE_MODE,0
FFT_SNR_REF,25.00000000
FFT_THR_REF,0.12224859
FFT_WINDOW_OLAP,0.50000000
FFT_WINDOW_SIZE,64

anbello · July 24, 2020, 1:39pm

No problem, I will see, Thanks.

andyp1per · July 24, 2020, 1:47pm

And INS_HMNC_OPTS?

You could set

FFT_HMNC_PEAK

To 1 to always track the first peak

anbello · July 24, 2020, 4:13pm

Here INS_HNTCH_*

INS_HNTCH_ATT,40.00000000
INS_HNTCH_BW,150.00000000
INS_HNTCH_ENABLE,1
INS_HNTCH_FREQ,300.00000000
INS_HNTCH_HMNCS,3
INS_HNTCH_MODE,4
INS_HNTCH_OPTS,0
INS_HNTCH_REF,1.00000000

I will try with:

FFT_HMNC_PEAK,1
INS_HNTCH_OPTS,1

anbello · July 24, 2020, 4:48pm

FFT graph with previous config +

FFT_HMNC_PEAK,1
INS_HNTCH_OPTS,1

Accel0

Gyro0

Now I have also a static notch at 120Hz

[Edit]
Top accel
Bottom gyro

anbello · July 24, 2020, 4:51pm

Before latest test I see this on GCS Messages:

[MAV 001:1] PreArm: EKF3 still initialising 
[MAV 001:1] PreArm: Radio failsafe on 
[MAV 001:1] PreArm: Internal errors 0x8000 l:391 ,main_loop_st 
[MAV 001:1] k 
[MAV 001:1] PreArm: EKF3 still initialising 
[MAV 001:1] PreArm: Radio failsafe on 
[MAV 001:1] PreArm: Internal errors 0x8000 l:391 ,main_loop_st 
[MAV 001:1] k 
[MAV 001:1] PreArm: EKF3 still initialising 
[MAV 001:1] PreArm: Radio failsafe on 
[MAV 001:1] PreArm: Internal errors 0x8000 l:391 ,main_loop_st 
[MAV 001:1] k 
[MAV 001:1] PreArm: EKF3 still initialising 
[MAV 001:1] PreArm: Radio failsafe on 
[MAV 001:1] PreArm: Internal errors 0x8000 l:391 ,main_loop_st 
[MAV 001:1] k 
[MAV 001:1] PreArm: EKF3 still initialising 
[MAV 001:1] PreArm: Radio failsafe on 
[MAV 001:1] PreArm: Internal errors 0x8000 l:391 ,main_loop_st 
[MAV 001:1] k 
[MAV 001:1] PreArm: Radio failsafe on 
[MAV 001:1] PreArm: Internal errors 0x8000 l:391 ,main_loop_st 
[MAV 001:1] k 
[MAV 001:1] PreArm: Radio failsafe on 
[MAV 001:1] PreArm: Internal errors 0x8000 l:391 ,main_loop_st 
[MAV 001:1] k 
[MAV 001:1] PreArm: Radio failsafe on 
[MAV 001:1] PreArm: Internal errors 0x8000 l:391 ,main_loop_st 
[MAV 001:1] k 
[MAV 001:1] Radio Failsafe Cleared 
[MAV 001:1] PreArm: Internal errors 0x8000 l:391 ,main_loop_st 
[MAV 001:1] k 
[MAV 001:1] PreArm: Internal errors 0x8000 l:391 ,main_loop_st 
[MAV 001:1] k 
[MAV 001:1] PreArm: EKF3 still initialising 
[MAV 001:1] PreArm: Internal errors 0x8000 l:391 ,main_loop_st 
[MAV 001:1] k 
[MAV 001:1] PreArm: Internal errors 0x8000 l:391 ,main_loop_st 
[MAV 001:1] k 
[MAV 001:1] PreArm: Internal errors 0x8000 l:391 ,main_loop_st 
[MAV 001:1] k

Then I rebooted and done the test OK.

andyp1per · July 24, 2020, 6:27pm

Looks like its worse, probably not enough CPU.

anbello · July 25, 2020, 9:26am

@andyp1per I was thinking about Nyquist frequency, folding, aliases, things I studied many years ago and now I am a little bit confused.

The accel FFT spectrum (pre-filter) on post #887 seems to me sampled at 1KHz and not 2KHz, I say this because is near perfectly symmetrical around 500Hz, 0 to 500Hz the signal, 500Hz to 1000Hz the folding. Am I wrong?

I report here the FFT in question
Accel-Folding

Another thing i would like understand: when we sample, say the accel, at 1KHz all vibration at frequencies higher then 500Hz re-enter in the signal we sample as aliases. Is this right? Could this be cause of problems in stability (leaning)?

From what I remember the only way to eliminate aliases is to filter the signal in the analog domain to not let frequencies above Fs/2 enter the ADC. But in our case both the analog signal (accel) and the ADC are inside the IMU chip.

andyp1per · July 25, 2020, 9:39am

You are correct on all counts!

The MPU6000 will only go up to 1Khz on Accels, but 8Khz on Gyros - my PR allows these to be independent. So when you set the gyro rate to 2kz we downsample gyros to 2kz but accels remain at 1khz. You would need to have a ICM20689 to avoid this.

Aliasing is a problem, and we filter to avoid aliasing effects, but I don’t think it’s the problem.