Expand description
Two or more recordings of one flight on one timeline, read side by side and never averaged.
What it is for. A high-power rocket often carries two altimeters, and a staged one may carry a logger in each stage. Each logger counts from a zero of its own: when it was switched on, or when it saw the boost. To lay the logs side by side their clocks must be lined up.
How. By a shared liftoff, the method Debrief uses: every board in a stack leaves
the pad at the same instant, so the time from one log’s liftoff to the other’s is the offset
between their clocks. There is no cross-correlation and no sweep over lags. What a timeline says
of a log is its liftoff’s reading, with that reading’s bound and interval. The offset’s bound
is the sum of the two liftoffs’ bounds, the most their errors can add to, and its 90% interval
is OFFSET_SHARES of that bound (ADR-257 §1).
Nothing is averaged. Two boards in one airframe are two measurements. A SideBySide holds
each log’s reading in its own cell, and each row’s Row::spread, the largest less the least,
describes how far apart they lie. It is not a check, and it isn’t an error on either: two boards
can agree and both be wrong by the day’s air. No field holds a mean, a weighted mean or a
fused track, and a test holds that to the serialized form.
Stages (Flew::Stages) are laid out in the same way, each stage’s logger leaving the pad
with the stack, and report no spread: a booster’s apogee and a sustainer’s are different
heights, not two measures of one. A logger that was still when the stack lifted off, or that
started in the air, has no liftoff, and sits on no timeline.
How far to trust it. On 2,000 synthetic flights carrying two loggers, on seeds of their own,
1,383 gave an offset (a log with no barometer has no liftoff); every one stayed within its
bound, by 0.13 of it at most and 0.03 s in the median, and its interval passed the coverage test
(cargo xtask two-logs, validation/reports/two-logs.md). Each of the 135 of 200
two-stage flights that gave one stayed within its bound too; the interval missed none of them,
so it is wider than it need be there. One real pair checks it, weakly, against another clock: on
RocketPy’s Prometheus flight the two boards’ GPS time gives an offset about 0.08 s from this one,
where an offset of zero would miss by 0.13 s. The offset is one number per pair: a clock that runs fast or slow
isn’t corrected, by about 15 ms over a 300 s flight at a crystal’s 50 ppm. Nothing here can tell
that two logs are of one flight: that is the caller’s word.
Structs§
- Given
- A reading in a cell.
- Offset
- One log’s clock against the first’s.
- Placed
- A log’s place on the timeline.
- Row
- One reading across the logs.
- Side
BySide - The logs’ readings on one timeline.
Enums§
- Cell
- One log’s cell in a row.
- Flew
- What the logs recorded.
- What
- A row of a
SideBySide.
Constants§
- METHOD
- How the clocks are lined up, in the result itself rather than implied.
- OFFSET_
SHARES - The clock offset’s shares of its bound (
Offset::bound_s), fitted on the two-logger calibration set: its 5% and 95% points of(offset − truth)/bound, on 2,000 flights on a seed of their own.cargo xtask two-logsfits them again and fails if they differ.
Functions§
- offset
- The clock offset of
otheragainstfirst, from their places. - place
- A log’s place, from its readings.
- side_
by_ side - Reads each of
logsand lays them side by side. - side_
by_ side_ read - Lays
logsside by side, with the readings already taken from each.