Layer 4: Real-Time State Estimation
The layer that fuses raw diagnostics into a coherent, physically consistent estimate of the plasma state for every controller above it.
From signals to state
Controllers do not act on raw voltages and photon counts; they act on plasma quantities - current, shape, position, density, pressure, safety factor. State estimation is the layer that turns dozens of noisy, partial diagnostic streams into a single best estimate of those quantities, with uncertainty attached, fast enough to be useful within a control cycle.
What it produces
- Plasma current and current profile estimate
- Boundary shape and gap distances to the wall
- Vertical and radial position
- Line-averaged and profile density
- Stored energy and normalized pressure estimates
- Safety-factor profile where diagnostics allow
Methods
Real-time equilibrium reconstruction (a fast constrained fit to the Grad-Shafranov equation) provides the magnetic picture. Kalman-type filters fuse successive frames and enforce temporal consistency, rejecting outliers and smoothing over dropped samples. Where full reconstruction is too slow, reduced observers estimate only the few quantities a given loop needs.
Consistency is the point
Different loops must not act on contradictory pictures of the plasma. By centralizing estimation, the stack guarantees that the shape controller, the current controller, and the disruption predictor all see the same state at the same timestamp. Disagreements are resolved once, here, rather than fought out through the actuators.
Honesty about uncertainty
A good estimator reports not just a value but a covariance. Downstream controllers and the disruption predictor use that uncertainty to widen safety margins when diagnostics are degraded. When too many sensors drop out, the estimator declares low confidence and the stack becomes conservative rather than pretending to see clearly.