v6.62
🇬🇧 EN
FPV · Data Analysis

Blackbox & Telemetry Analysis

How real flight data turned into concrete tuning decisions — including a methodology where an initial assumption was later corrected because a closer look at the data showed something else.

From blackbox log to FFT analysis

From a blackbox log with 5 chirp test flights, CSV exports were created via the official Blackbox Explorer — complete time series with ~8000 data points per second. From these, a frequency analysis (FFT) was calculated for each axis.

Gyro roll FFT spectrum
Gyro roll — frequency spectrum, longest flight (36.8 s)
Finding: 530 Hz gyro noise peak

The orange bump at ~530 Hz in the unfiltered signal has no counterpart in the filtered (turquoise) signal — the filter barely catches it. Almost identical on all 3 axes, well above the chirp test range (up to 60 Hz) — so a real noise phenomenon independent of the test.

D-term pitch FFT spectrum
D-term pitch — frequency spectrum
Corrected misjudgement

A D-term resonance peak initially suspected at ~101 Hz turned out on closer inspection to be a chirp test artefact: the elevated activity coincides with the chirp test frequency range itself (up to 60 Hz) — that is the expected system response to the test signal, not a filter problem of its own. This correction was transparently retracted rather than leaving an unfounded filter change in place.

Further statistical findings

FindingValueAssessment
Effective PID loop rate~1969 HzShould be ~8000 Hz (found: pid_process_denom=2)
Yaw noise reduction by filters1-8%Consistently the weakest of all 3 axes
Motor saturation (all motors)<5%No sign of fundamentally too-aggressive PID values

Double verification

A second, independent CLI dump was uploaded later and compared with the original — all PID and filter values matched exactly. That confirmed the analysis had been based on the correct starting configuration all along. Details in the Betaflight tuning history (v6.0/v7.0).

Downloads

Methodological principle

Not every FFT anomaly means a real tuning problem — some patterns are artefacts of the test method itself. That distinction was decisive here.