flight controller

1,024 consecutive simulated control-loop periods at 4 kHz, 249.58–250.48 µs around the 250 µs target, all inside the ±1% band. Simulated, not measured: a model of the bare-metal schedule on a 168 MHz STM32F405, the 8 kHz gyro interrupt and the 4 kHz control task chained after every second one, with representative execution times and jitter. In the model, jitter is the period error of each control-task start, as the DWT cycle counter would time it: for every start-to-start interval T_i of 1,024 consecutive loops, J_i = 100 · |T_i - 250 µs| / 250 µs. The simulated worst case, max J_i, is 0.192% (0.48 µs off target); 99% of intervals are within 0.358 µs (0.143%). The modeled control task, DShot frame included, ends at most 71.26 µs after its gyro edge, leaving 53.74 µs before the next one. These periods are the model, not the board’s measurement, which is stated on its own: measured on the board, control-loop jitter stays under 1% at 4 kHz.

Open-loop injections into the control core, simulated on the host; nothing is fed back. Step: the roll-rate setpoint goes from 0 to 200 °/s at 20 ms with the gyro still. P jumps to 0.20, I ramps to its 0.15 clamp by 113.75 ms, D on measurement stays within 0.007, where D on the error would kick to 1.09; the mixer's DShot commands split and the low pair saturates at 48. Noise: a synthetic gyro with a 400 Hz tone; the disturbance's rms falls from 10.382 °/s raw to 3.978 after the pt2 low-pass at 250 Hz and 1.535 with the notch. Sweep: 50 frequencies from 10 to 1600 Hz; the notch at 400 Hz (Q 2.5) reaches -60 dB, and the filter chain is at -26.22 dB by 1600 Hz.