3. Time and sources¶
Two clocks¶
Every record in a log has a time, but not always from the same clock.
- Capture time: when the ground station received the message (the 8-byte stamp in front of each
.tlogrecord). It's wall-clock time, so it can jump if the PC's clock is adjusted, and it includes radio delay. - Boot time (
time_boot_ms,time_usec): how long the flight controller has been powered on. It's precise, but it restarts from zero when the board reboots.
Epochs¶
If the board reboots mid-log, "10 s after boot" happens twice. Put both runs on one timeline and you'd mix two different flights. AnaTele avoids this by splitting the log into epochs. A new epoch starts whenever a clock clearly goes backwards. Inside one epoch, time only moves forward, so samples can be ordered, paired and compared safely.
The exact tests (per source) are:
| Clock | New epoch when… | Why the margin |
|---|---|---|
.tlog boot time time_boot_ms |
it drops by more than 10 s | Small regressions are just reordered / delayed packets |
| DataFlash boot time | it drops by more than 1 s | On-board records are written in order |
.tlog capture time |
it drops by more than 2 s | PC clock adjustments separate the whole capture |
A boot-time drop from near \(2^{32}\) ms back to near zero is recognised as a counter wrap (after ~49.7 days), not a reboot. Every epoch split is listed as a diagnostic, so you can see why it happened.
Sources¶
A capture can hold messages from more than one system: the drone, a gimbal, the ground station, a companion
computer. MAVLink identifies each sender by a system id and component id. AnaTele calls each pair a
source (for example 1:1 is the autopilot) and never mixes signals from different sources.
Pairing samples¶
Many checks compare two signals, for example desired and actual roll. They don't arrive at exactly the same instant, so AnaTele pairs each sample with the nearest sample of the other signal, within a tolerance. For attitude that tolerance is 0.25 s. If no partner is close enough, the pair is skipped rather than guessed.
Design principle
Nothing is interpolated or invented. A derived value always comes from real samples, close enough in time to be compared, and it keeps a link to the record it came from.