Latency Budgets in Real-Time Control
A control loop is only useful if it finishes in time; the latency budget is the discipline that keeps every step inside its deadline.
The deadline is physical
A plasma changes on millisecond timescales, so a control decision that arrives late is worse than useless. Real-time control lives under a hard deadline set by the physics, and the latency budget is how that deadline is divided among the steps of the loop so the whole cycle completes in time, every time.
Where time goes
- Sensing: acquiring and transporting the diagnostic signals.
- Estimation: reconstructing the plasma state.
- Decision: computing the control action.
- Actuation: applying the command to coils and heating.
Worst case, not average
Real-time systems are judged by their worst-case timing, not their average. A loop that usually finishes fast but occasionally overruns can still cause a fault. Meeting a budget means bounding the worst case, which shapes every algorithm choice: reduced-order models and surrogates are used precisely because their timing is predictable.
The split pipeline
This is why the data pipeline separates a fast path for control from a thorough path for analysis. The fast path is stripped to what fits the budget; the thorough path does the rest without the deadline.
For both machines
The Hyperion breeder and the burner each have their own latency budgets set by their physics. The tandem mirror's plug dynamics and the tokamak's shape control impose different deadlines, but the budgeting discipline is shared across both machines.
Verification
Timing is tested as rigorously as correctness, because a controller that is right but late fails just the same, connecting to verification.