Distributed time alignment on multi-vehicle test cards
The clock-domain count on a modern multi-vehicle test card is higher than most data-engineering plans assume. A short field guide to what we now require before the first flight of a new test card.
For a single-vehicle card, the clock domains are typically: the airborne acquisition stack (one or two boards, IRIG-B or PTPv2 disciplined), the ground-station receiver (its own PTP master or a free-running clock that gets disciplined post-flight against IRIG bursts in the telemetry stream), the test-range TSPI source (independent), and the operator-action logger on the GCS (which is almost always on the GCS host clock, which is itself loosely synchronized to the rest of the world). That is four clock domains, three of which are typically managed.
For a two-vehicle card, the count doubles plus shared infrastructure. The two vehicles' clocks are not synchronized to each other in flight unless someone designed the GPS-disciplined PTP master into the program; in most programs no one has. Post-flight reduction has to align the two airborne stacks against a common time, which requires a common reference point that both observed (a TSPI handshake, a synchronized squib firing, a coordinated transponder pulse). Programs that do not plan for this discover it when the two vehicles' tracks do not overlap correctly in the post-flight reconstruction.
For a swarm experiment, the alignment problem becomes a graph rather than a chain. Our practice is to require, before first flight of a new multi-vehicle card: a documented common-time strategy with named reference signals, a planned reduction sequence with explicit alignment checkpoints, and a successful dry-run of the alignment reduction against synthetic data with known offsets. The dry run is the part everyone wants to skip; it is also the part that catches the integration errors that would otherwise consume the first flight.
None of this is exotic. It is just discipline that programs reliably underspend on early because the cost of getting it wrong is invisible until the post-flight review when the data does not assemble.