Betaflight Blackbox Setup: Record a Log You Can Actually Diagnose

Blackbox records the gyro, sticks and setpoint, PID terms, motors, RPM, voltage and current. But a log on out-of-the-box settings often shows very little. Real examples from logs we've analysed: a 3.5-inch on Betaflight 4.5.2 logging at 1 kHz while its motors spin up to 644 Hz; a Betaflight 2026.6 log at 247 Hz with no raw gyro and no RPM; a 7-inch on 4.6 logging 2 kHz to an SD card with 182 gaps; an INAV 7.1 log at 58 Hz that says nothing about vibration.

What each check needs

Thresholds come straight from the analyzer's code; below them a check stays silent or flags too little data.

Under 10 seconds in the air, every tuning finding is downgraded to “info” and loses its CLI commands, and bounce-back and punch-outs aren't measured. If over half the flight was in Angle or Horizon, step response and bounce-back are skipped: there the stick commands an angle, not a rotation rate.

Logging rate and aliasing

Why 1 kHz misleads on small quads

A log recorded at N Hz only shows vibration up to N/2, the Nyquist frequency. Anything faster doesn't vanish; it mirrors back below that limit. A motor at 644 Hz in a 1000 Hz log shows up at 1000 − 644 = 356 Hz. That leaves a tidy spectral peak where nothing is happening: an easy target for a misplaced notch or filter.

Three logs from a VX3.5 on Betaflight 4.5.2 (4 kHz PID loop, 1/4, so 1 kHz) had the motors peaking at 583, 605 and 644 Hz, above 500 Hz every time. The figure shows the same effect on another quad: a 2 kHz log with every fourth frame kept, exactly what blackbox_sample_rate does, skipping frames with no filtering first.

bbdoctor measures per-motor vibration from each motor's rotation phase (eRPM), so it knows where the mirrored copy lands and that check still works at 1 kHz; the noise spectrum and resonances can't be rescued. When peak RPM in a Betaflight log goes above 95% of the Nyquist frequency, the report adds “Logging rate is below the motor frequency” with the right divider as a command.

Picking blackbox_sample_rate

From 4.3 on, the rate is a divider of the PID loop: 1/1, 1/2, 1/4, 1/8 or 1/16, with 1/4 as the default. bbdoctor picks the largest divider that still leaves at least 2 kHz:

Your loop rate is on the Configuration tab. Not every build needs 2 kHz: bbdoctor only asks once the motors outrun the logging rate. A 7-inch 6S in our set never went past 315 Hz, so 1 kHz would have done. A 3 to 3.5 inch on 4S at 1 kHz usually goes over, and not only those: in four 1 kHz logs from a 6S quad on 4.5.1 the motors reached 529–554 Hz. On a BMI270, 1/1 eats flash fastest; if space is tight, 1/2 (1.6 kHz) still covers motors up to about 760 Hz. 1/8 and 1/16 are useless for analysis (500 and 250 Hz on a 4 kHz loop), and the config check flags them.

Flash budget

File sizes in our set: at 1 kHz with all fields, about 34 KB/s without debug fields and 41–43 KB/s with a debug mode on; at 2 kHz, 58–70 KB/s. An 8 MB flash lasts about 3 to 4 minutes at 1 kHz and 2 minutes at 2 kHz. That VX3.5 (debug_mode = GYRO_SCALED) filled its 8 MB to the last byte in 195 seconds at 1 kHz and stopped mid-flight. bbdoctor spots this (a cut-off end, or a file of exactly 2, 4, 8 or 16 MB) and warns alongside the rate advice that recording time will halve.

So erase the flash before testing (Erase flash on the Blackbox tab, or flash_erase in the CLI), fly 1–2 minute test packs and drop fields you don't need. An SD card has room, but a slow one drops frames: the 7-inch on 4.6 logging 2 kHz to SD had 182 gaps. Use a decent, freshly formatted card; if gaps persist, step the rate down.

Betaflight settings by version

4.5 and 2025.12+

For a 4 kHz PID loop (use 1/4 on 8 kHz and 1/1 on 3.2 kHz):

set blackbox_mode = NORMAL
set blackbox_sample_rate = 1/2
set blackbox_disable_gyrounfilt = OFF
set blackbox_disable_rpm = OFF
set blackbox_disable_setpoint = OFF
set blackbox_disable_pids = OFF
set blackbox_disable_motors = OFF
set debug_mode = NONE
save

• blackbox_device is left out so the paste can't break SD logging. Check it yourself: SPIFLASH for onboard flash, SDCARD for a microSD slot, NONE records nothing.
• blackbox_mode = NORMAL records from arm to disarm; with a BLACKBOX mode assigned on the Modes tab, only while that switch is on.
• Raw gyro and RPM have their own fields here, so no debug mode is needed; with debug_mode = NONE no debug fields are written.
• The RPM field only exists with bidirectional DShot on (see below).

A Betaflight 2026.6.1 log was recorded at 1/16 (247 Hz on a 4 kHz loop) with raw gyro and RPM off, though bidirectional DShot was running. The report gave both commands to turn them back on. A 4.5.1 log had setpoint off: bbdoctor rebuilds it from sticks and rates, with a rougher PID estimate, and says so.

4.3–4.4

set blackbox_sample_rate = 1/2
set debug_mode = GYRO_SCALED
save

Pick the divider from the table above. There are no unfiltered gyro or RPM fields yet. The gyro goes through debug, and only one debug mode runs at a time, so DSHOT_RPM_TELEMETRY is out (bbdoctor doesn't read RPM from debug anyway). Even with bidirectional DShot working it recommends 4.5, since without RPM it can't name the unbalanced motor. The blackbox_disable_… settings exist from 4.3 (rpm and gyrounfilt from 4.5): check that setpoint isn't off.

4.0–4.2

set blackbox_p_ratio = 64
set debug_mode = GYRO_SCALED
save

These versions use blackbox_p_ratio. The rate is roughly p_ratio × 31.25 Hz: 32 gives about 1 kHz, 64 about 2 kHz, capped at the PID loop rate. A 4.2.11 log in our set was recorded at just 534 Hz and had 36 gaps on top.

Fields you can switch off, and ones you can't

blackbox_disable_gps, blackbox_disable_mag and blackbox_disable_alt (baro) are safe to turn ON unless you analyse navigation. Keep:

• gyro and motors: a log without the gyro or motor commands isn't accepted at all;
• setpoint and pids: needed for an accurate step response and to split bounce-back into P, I, D and FF;
• rc: gives the stick-smoothing delay (rcCommand → setpoint) and rebuilds the setpoint if that field is off;
• acc: impacts in crash analysis;
• bat: voltage sag, battery resistance, power loss as a crash cause;
• rssi, if RSSI is set up: link loss before a crash.

RPM in the log

RPM comes from bidirectional DShot: set dshot_bidir = ON, save, with the DShot protocol and ESCs on Bluejay, AM32 or BLHeli_32. Then take the props off, spin the motors with the sliders on the Motors tab and check that RPM shows up with errors near zero. Check motor_poles too: motor frequency is computed from eRPM through the pole count. Most 5-inch motors have 14 poles, some small ones 12; get it wrong and every motor frequency in the log is off. More in the RPM filter and bidirectional DShot guide.

INAV

set blackbox_rate_num = 1
set blackbox_rate_denom = 2
blackbox GYRO_RAW
blackbox MOTORS
save

In INAV the logging rate is the fraction blackbox_rate_num/blackbox_rate_denom of loop cycles. INAV 8 and later default to a 2 kHz loop (looptime = 500), where 1/2 gives 1 kHz; INAV 7 and earlier default to 1 kHz (looptime = 1000), where 1/2 gives 500 Hz. Vibration and filter checks need 500 Hz or more. Three INAV 7.1.2 logs in our set, from a 7-inch and a 5-inch, were recorded at 1/32, or 58–60 Hz: all they can show is motor balance and the battery. An INAV 8.0 log at 1/2 came in at 995 Hz, with noise and filters plain to see.

• blackbox GYRO_RAW adds the raw gyro, blackbox MOTORS the motor commands. Two INAV 7 logs with MOTORS off were rejected outright.
• INAV doesn't log per-motor RPM, so per-motor vibration isn't available.
• An INAV 8.0 log flown 78% in Angle/Horizon got no PID verdict.
• An INAV 6.1 log had 27% corrupted frames. bbdoctor drops them and flags its PID and noise findings as unreliable; re-fly a log like that.

Details are in the INAV blackbox guide.

ArduPilot and PX4 in brief

ArduPilot. Without the fast bits in LOG_BITMASK (the default has none), ArduCopter logs the gyro at 25 Hz, as in all ten ArduCopter logs in our set, 4.1 to 4.8-dev. For analysis, set bit 0 (ATTITUDE_FAST) and bit 19 (IMU_RAW) in LOG_BITMASK, plus INS_RAW_LOG_OPT = 9 on 4.5+ for the gyro before and after filtering. RPM comes from ESC telemetry or bidirectional DShot (SERVO_BLH_BDMASK). Use the .bin from the SD card (APM/LOGS), not a .tlog. For ArduCopter PIDs the built-in AUTOTUNE is more reliable; the log is for vibration and filters. See the ArduPilot log analysis guide.

PX4. Add bits 4 (High rate) and 8 (Raw FIFO gyro) to SDLOG_PROFILE: from the default of 1 that's 273, or param set SDLOG_PROFILE 273 in the QGroundControl MAVLink Console, then reboot: the profile is read at startup. The .ulg is under Analyze Tools → Log Download. See the PX4 ULog guide.

The test flight

Freestyle hits these events by chance; a 1.5–2 minute test pack collects 6–8 of each on purpose:

1. Before takeoff. Fresh props, a charged pack, erased flash, BLACKBOX switch on if you use one.
2. Acro mode. Angle and Horizon rule out step response and bounce-back.
3. 30 seconds of hover and smooth cruising, then, if there's room, a slow 5-second throttle roll from hover to 100% for the noise-versus-throttle map.
4. 6–8 flips or rolls per axis, plus yaw spins. Snap the stick back to centre in under 0.2 s and hold it there for half a second. Moves count from 150 deg/s on roll and pitch and 120 deg/s on yaw; a slow return to centre is thrown out.
5. 6–8 punch-outs from hover (about 30% throttle, always below 70%): more than a quarter of throttle travel in 0.2 s. Don't touch roll and pitch; above 80 deg/s of setpoint the punch doesn't count.
6. 6–8 straight chops: climb, chop to zero, sticks centred for about a second. A chop counts if throttle was above 35% and fell by more than 25% within 0.15 s. No flips straight after: in one 2 kHz log, 12 of 15 chops went straight into a flip or roll, leaving 3 usable for propwash.
7. Landing. Land gently, without a chop just above the ground: a chop that ends in an impact is left out of the propwash check.

A big rig doesn't need flips; short, sharp stick tilts at a safe height will do. Length alone guarantees nothing: five minutes of ordinary flying on a 7-inch gave 5 chops (one clean), 1 punch-out and 1 sharp roll stop.

Getting the log off

In Betaflight and INAV, use Save flash to file on the Blackbox tab, or Mass Storage mode so the board mounts as a drive. With an SD card, copy the file. INAV files often end in .TXT, ArduPilot uses .bin, PX4 .ulg.

One file can hold many flights, one per arm; our set includes a Betaflight 2026.6 file with 66. bbdoctor takes the longest by default (number 58 there), and the report lets you pick any other.

Common mistakes

Check the log automatically

Drop the file into bbdoctor; it's free and needs no account. First it checks the log itself: flight time, gaps and dropped frames, a cut-off end, raw gyro, RPM and setpoint, the logging rate against peak motor frequency, and time in self-level. Anything missing is named, with commands for your firmware version. See the sample report.

You can check the settings before flying: paste your diff all into the config check. It flags blackbox being off, a 1/8 or 1/16 divider, disabled fields, missing GYRO_SCALED on 4.0–4.4, and a low rate or missing GYRO_RAW on INAV. For the finished log, see how to read a blackbox log.

Updated: 2026-10-22