The 8-Layer Control Architecture
A layered control stack that spans microsecond hardware reflexes to a predictive digital twin and human oversight, each layer with its own latency budget.
Why layers
A control problem that spans microseconds to minutes cannot be solved by one controller on one clock. Fast phenomena demand deterministic, hard-real-time responses with no room for a general-purpose scheduler; slow strategy demands models, optimization, and judgment that cannot run in a microsecond. Layering separates these concerns so each runs at its natural rate and hands well-defined signals to its neighbors.
The eight layers
- Layer 1 - Deterministic hardware reflexes (microseconds): trip lines and protective interlocks in logic
- Layer 2 - Fast feedback loops (tens of microseconds to ~1 ms): vertical stabilization, RWM control
- Layer 3 - Magnetic and kinetic feedback (~1-10 ms): shape, position, current control
- Layer 4 - Real-time state estimation (~1-10 ms): equilibrium reconstruction and profile estimation
- Layer 5 - Actuator management and allocation (~10 ms): mapping demands to limited hardware
- Layer 6 - Scenario and trajectory supervision (~10-100 ms): scheduling references through the discharge
- Layer 7 - Predictive digital twin (~100 ms-seconds): forward simulation, disruption forecasting, what-if
- Layer 8 - Human oversight (seconds and up): intent, approval of exceptions, off-normal decisions
Signal flow
Sensors feed the fast layers directly and feed state estimation, which publishes a coherent plasma state to the higher layers. The scenario layer sets references; actuator management turns competing demands into commands the hardware can honor; the fast loops close around the plasma. The predictive layer runs alongside, forecasting where the trajectory leads and requesting changes before limits are reached.
The contract between layers
Each layer trusts the layer below to meet a latency and correctness budget, and exposes a small, well-typed interface upward. A higher layer may be late or wrong without endangering the machine, because a lower, faster, simpler layer always guarantees safety. This inversion - safety lives in the simplest, fastest layer - is the core design principle.
The pages that follow describe each layer in turn.