Notch filter and tuning of a copter with long arms, with one issue of uncommanded pitch overshoot

ArduCopter Copter 4.7
1

I am setting up the new copter (with long arms, 105 cm diagonal, with props 15 inch, and weight of 3.4 kg during this flight, and have a couple of questions.

First, I got the data for notch filters. I flew without notch filters enabled and from the graph I see that all values are pretty much below -50. My feeling is that they may be not necessary, although I can easily filter out the two spikes and their harmonics.

newplot
newplot1200×450 52.8 KB

The vibes are good, stationary are below 5, in flight at about 40 km/h, the Y vibe began to climb to about 15.

There is a small yaw imbalance.

I have run the quicktune, and the result overall seems to be very good, with desired/real pitch and roll tracking verhy closely, however, I got one strange issue, when going relatively fast in hover mode, at 40 km/h, when I release the stick, the copter initially levels off and then gives a sharp braking input. In stationary hover, the tracking is perfect.

overshoot
overshoot1068×507 7.83 KB

I am not sure what is going on…

.BIN file from the flight: www.ecoterrenos.cl/2026-06-11 11-24-25.bin

2

Fix the yaw imbalance. After that look at the dynamic notch data again.

A narrow notch with low attenuation will produce little phase lag and that might even let you increase the gyro frequency cut-off, leading to a more responsive control.

1 Like
3

Do you have any clue why there is such different response in pitch at 0 speed and 40 km/h?

4

First, yaw tracking seems terrible, like, really bad. It might be a consequence of aerodynamics at a speed paired up with high T/W. Here I’m unsure what to advise other than running a proper autotune and seeing what happens after.

newplot
newplot1915×1378 258 KB

For the pitch overshoots, you seem to be hitting MOT_SPIN_MIN on multiple motors. This is quite typical for sharp braking I guess, and one of the reasons for ATC_THR_MIX_MIN to exist.

newplot
newplot1915×1378 514 KB

You may want to decrease MOT_SPIN_ARM and MOT_SPIN_MIN to lower thrust levels, but you obviously need to avoid desyncs within possible flight envelopes and battery charge levels. Might be a darker magic than with small machines.

This will kind of go away with a heavy payload, however, as you intend it to fly both with and without payload within the same flight (judging from the sister post), you probably need to design missions to avoid sharp braking.

P.S.: and for notches - just do it. This is your noise profile with the current tuning, it will either get worse with a tighter tuning or prevent that from being reached.

5

Thanks for the review. I did not yet quick tune the yaw axis…Also I am correcting some slanted motors for yaw imbalance, but I think it is minor.

You could be very correct about MOT_SPIN values, and since even in hover the copter operates at very low level of power out values, when I release the stick at some speed, the copter becomes a glider (15 inch props are notg small!) which needs close to zero thrust.

I think I will try and fly next flight with intermediate weight and retune for that to see if the issue of overshoot changes, I have made already a dummy load of 2.5 kg.

6

Flying at 6kg immediately eliminated the issue of the overshoot, so it must be due to mot_spin… Not sure how to adjust for the low weight though, because lowering MOT_SPIN variables does not look safe to me…

Was surprised that the notch filter review for 6 kg gave totaly different profile:

newplot (1)
newplot (1)1200×450 55.9 KB

I would then use ESC telemetry, multisource, 102 Hz FQ, 20 HZ BW, ATTENUATION OF JUST 9 DB???, harmonics 1,2,3, ref 0, fm_rat 1, keep ins_gyro at 25???

I also did a flight with 7.6 kg. The good news was that the motors remained lukewarm, the ESC reached 80 C, the bad news seems that the current sensor did not show increase in current… There is either interference or something, it indicates changing current, but always at around 30 A max, i.e. at 3.4, 5.8 and 7.6 kg the current was very close to the same 22..35 A

For the next flight I plan to check the hovering current by estimating battery capacity (recharging it to get mah and then compare to hover time). I will probably risk going up to maybe 8.5 kg. 7.6 kg hover was actually quite stable.

I had also issues with vertical controller. 4.7 uses different parameters and also different values. One issue I had was that in HOVER mode, when I tried to take off, it lifted off and then settled, after that I was pushing the throttle up to about 75 % but it did not want to lift off. I switched to STABILIZE, lifted off, then switched to HOVER, and then it hoverred at 50 % throttle without issues. Not sure what happened, This happened a couple of times. I.e. if I lift off at about 50 % throttle and keep it hovering, it stays in the air (in hover mode), but if I allow it to settle, the throttle seems to have no effect at all. I did not try to push full throttle (scared that it might shoot up like a rocket)…

7

These should be better:

MODE=3
ATT=9
FREQ=120
BW=12
HMNC=3
FM_RAT=0.2
INS_GYRO=50
REF=1

1 Like
8

He reaches as low as 25 Hz sometimes - without payload of course. I would set FM_RAT = 0.2 for what it’s worth.

2 Likes
9

thanks a lot!

10

AMC 4.0.4 supports the ArduCopter 4.7 vertical controller parameters.

1 Like
11

x0.1 of 4.6. That’s it. Everyone will have to get with the program for 4.7, the earlier the better.

12

The only reason I went with 4.7 because I had H7A3, (which I bought by mistake, thinking it was H743). Scripting is available for H7A3 from 4.7 only… If I had a spare H743 I would have stayed with 4,6. Ha ha…

13

Poor yaw authority can be corrected by canting motors 3-5°, CCW ones for CW torque and CW ones for CCW torque.

14

Did another hover test with 8.6 kg. Did not go exactly well… The ESC gave up after 90 seconds. The copter fell from 1 meter, just one leg dislodged. But the temperature logs showed 120 C for ESC, with temperature tendency rising.

So it seems that I will have to divide the ESC in two, i.e. install two 4x 1 ESCs, each one working with 2 channels only. I thought that I could get away with just one ESC under the prop, but I was wrong.

The notch filter graph also changed, compared to 5.8 kg. The base frequency spike went up from 102 to to 116 Hz. Not sure if it is just RPM related. Initially I thought so, but an increase in RPM proportional to 116/102 would correspond to a thrust increase from 5.8 kg to 7.5 kg, and this was 8.6 kg. So here I am not sure for which weight one should optimize the notch filters, or wheter one should make them scalable by means of lua script.

15

Does the ESC have a radiator attached? You could try putting a fan on it.

1 Like
16

The thust is not linear withe rpm speed/frequency.
It is quadratic.

A linear regression does not apply, just like you found out.

1 Like
17

Yes, but, 116/102 = 1.13; 1.13*1.13 = 1.27 , i.e. 27 % thrust increase assuming that the frequency correlates directly to RPM. However, the thrusts were 5.8 kg and 8.6 kg, i.e. 8.6 / 5.8 = 1.48, i.e. the thrust increased 48 %.

18

It is exposed directly under the tip of the propeller, it has the maximum airflow possible… considering that the it was a test at 8.6 kg., and ideally I have to push it to close to 10.0 kg, the power increase will be very significant. So far, I would say takeoffs under 60 seconds could be feasable at 8.6 kg, but still risky.

The problem is that the second ESC, second power bank, and cables will add more weight. About 60 grams.

19

Can you show a picture of this setup. Which ESC (brand/type) you are using

1 Like
20

Speedybee BLH 4 x 1 60 A. I think that one could marginally improve cooling by making a scoop for pushing air UNDER the ESC by the prop, but the heat up rate is overall too large to be able safely give a solution. With 5.8 kg the ESC was operating well below 80, so splitting the ESC in two would bring it to safe temperature operation.

WhatsApp Image 2026-06-10 at 20.54.34 (1)
WhatsApp Image 2026-06-10 at 20.54.34 (1)1600×1200 402 KB