Betaflight PID & filter tuning

The own blackbox/CLI analysis from v8.0 was cross-checked against an independent external analysis of the same dataset. On most points (gyro filter setup, feedforward values, Crossfire packet rate) both analyses agreed. Three safety-critical deviations were clearly identified: a failsafe delay that is too long (300 ms), a power deviation on motor 3 (+8%), and a crash recovery mode that has not yet been tested for real (turtle mode). The full comparison, including recommendations, is in its own Analysis Comparison.

A second round of real blackbox data (5 new flights, 25 Jun 2026) showed that, instead of flashing v7.0, a switch to manual tuning had been made in the meantime (simplified_pids_mode → OFF, gyro filters set up entirely anew). v8.0 documents this configuration consistently. One of the 5 logs was correctly recognised as a pure motor ground test and excluded from the analysis rather than being wrongly counted as flight data. With this configuration, the high-frequency gyro noise peak is >attenuated by 99%.

A second dump (uploaded by the user) turned out to be incomplete (no batch end, no final save). On closer inspection it also turned out that v6.0 was wrongly described as a "complete dump" — it was in fact also only a selection. v7.0 corrects this: a real diff with 27 parameters, each one verified programmatically against the original dump file (100% match).

A second, independent dump file confirmed all v4.0/v5.0 values and revealed three small documentation errors: dyn_notch_q was wrongly documented as 350 (actual: 300), vbat_min_cell_voltage as 300 (actual: 320), mixer_type as QUADX (actual: LEGACY). Was later itself wrongly described as a "complete dump" — corrected by v7.0.

From 5 CSV exports (Blackbox Explorer), a frequency analysis identified a clearly isolated gyro noise peak at ~530 Hz on all 3 axes. Methodologically important: a D-term resonance initially suspected at ~101 Hz turned out to be a chirp-test artefact — this correction was transparently retracted rather than left in place without justification.

An uploaded blackbox log (5 flights) showed that the actually flown configuration was even more aggressive than any previous estimate (P 45-47, feedforward 120-125). A motor KV error was also corrected here: 14800 KV (wrong, v1.0-v3.0) → a real ~1960 KV according to the own RPM telemetry.

Renewed research with other sources (real factory CLI dumps of comparable 65-75 mm 2S whoops such as the BetaFPV Air65/Air75 Champion) showed that the v2.0 values were systematically too high — oriented towards 5" freestyle logic that does not fit this lighter class.

First adjustment to Betaflight 2025.12 (the new calendar-based versioning scheme, also officially called "4.6" — wrongly referred to as "4.7" in this project for a long time). PID values here still estimated without own flight data.

The very first tuning recommendation, generically researched, without own flight data or factory comparisons. It served as the starting point for all further corrections.