Computing Library › AI Plasma Control
AI Plasma Control

Layer 2: Fast Feedback Loops

Sub-millisecond feedback that stabilizes the fastest-growing plasma instabilities, chiefly vertical position and resistive-wall modes.

The fast unstable modes

Elongated, vertically unstable plasmas grow off-axis on a timescale set by the surrounding conducting structures - often a few milliseconds, but effectively unstable and requiring active correction faster than that. Resistive-wall modes (RWMs) grow on the wall's resistive timescale. Both must be caught by a loop that closes far faster than the mode grows, which puts them below the shape and current loops in latency.

Loop budget

Kronos motion — wall loading

Fast feedback typically runs at tens of microseconds to about one millisecond per cycle. The budget covers sensor sampling, a compact linear controller, and command output to fast power supplies. There is no time for full equilibrium reconstruction here; the loop uses a small number of directly measured signals and a fixed, pre-tuned response matrix.

Control law

These loops are usually linear and model-based. A plasma-response model linearized about the operating point yields a state-space description; the controller is designed for guaranteed stability margins against that model plus uncertainty. Simplicity and provable stability outrank optimality. A loop that is fast, linear, and robust beats a cleverer loop that occasionally hesitates.

Handoff to slower layers

Fast feedback keeps the plasma alive on the millisecond scale but does not decide where the plasma should be. It tracks references handed down by the shape and scenario layers. If those references become infeasible, fast feedback still tries to follow them - which is why the actuator-management and scenario layers must never command a target the fast loop cannot reach.

Failure behavior

If fast feedback saturates its actuators or loses a key sensor, it signals distress upward and the reflex layer stands ready to trigger a controlled termination. Loss of vertical control is a leading cause of disruptions, so this loop's health is monitored continuously and treated as a first-class safety signal.