Air65 — Blackbox Review of 16 Logs

Pulled off a completely full 16 MB flash. Sixteen logs, fifteen readable, fourteen with flight. The flight-controller clock restarts in each pack, so this is not one session. Device UID 002900023235510535303333 identifies the board.

Parameter	Value
Gyro loop	3.2 kHz (`looptime 312`, `pid_process_denom 1`)
Logging	811 Hz, 1/4 decimation
Hover	~43 % stick against `thr_hover = 44`
Mean hover RPM	~27 000, a 455 Hz fundamental
Throttle limit	80 %

Healthy: the airframe is balanced

Front against rear averages 2.3 % and changes sign between flights, which no standing bias does. Pitch I-term averages +1.1 and crosses zero freely.

Metric	Air65	Crafty (12 Sep)
Front/rear RPM split	2.3 % (sign varies)	12.8 % (always front)
Pitch I-term, mean	+1.1	−26.6
Left/right RPM split	+2.4 %	+0.7 %
Roll I-term, mean	+5.6	+1.4

The one small asymmetry is roll: left runs 2.4 % faster in thirteen of fourteen flights. A slight lean to the right, an order of magnitude below the Crafty pitch problem.

The blackbox samples below its own motor frequency

A 455 Hz fundamental against an 811 Hz log gives a Nyquist limit of 406 Hz, so everything above folds back. A real motor peak moves with RPM; this one pins near 400 Hz regardless:

Log	Mean RPM	True fundamental	Apparent peak
`btfl_004`	26 752	446 Hz	393 Hz
`btfl_005`	26 836	447 Hz	400 Hz
`btfl_013`	26 907	448 Hz	394 Hz

This is a logging limitation only; the controller filters at the full 3.2 kHz. But these logs cannot be used to judge filter performance, and the 5.7–9× attenuation figure is not comparable to Crafty’s 22×. To tune filters from logs on this craft, lower the decimation: blackbox_sample_rate 1/2 gives ~1.6 kHz, 1/1 the full 3.2 kHz.

`dyn_notch_max_hz` sits below the fundamental

It is 400, under the 455 Hz the motors produce, so the dynamic notch never reaches the fundamental. The RPM filter does (bidirectional DShot, 3 harmonics), so this is lost redundancy rather than a fault. Betaflight’s default is 600.

Voltages: the decoder, again

The same blackbox_decode scaling bug as in the Crafty review applies: CSV volts read 12.8 % low. Corrected, four logs open on a fresh pack at 4.19–4.20 V, each pack works down to about 3.7 V before a swap, and in-flight minima reach 2.88 V under hard punches. vbat_scale = 113 is correct on both craft.

The crashes were hiding the tune

Nine of fourteen flights end within 1.5 s of their acceleration peak, with peaks to 13.8 g. Masking one second either side of every sample above 4 g changes the picture:

Metric	Raw	Impacts masked
Tracking error, roll	~60 °/s	5 °/s
Tracking error, pitch	~55 °/s	9 °/s
D-term RMS, roll	93	19

Noise floor and motor balance are unchanged after every impact, so nothing broke.

Open items

motor_kv = 1960 here too, against ~27 000 RPM observed. Harmless.
btfl_003 is a 6 kB fragment with no timing. Not recoverable.
The flash was completely full (16 777 216 of 16 777 216 bytes), so logging had stopped.