Sensor Calibration Data
Calibration data maps raw sensor output to physical units; it must be versioned and stored beside measurements to keep them meaningful.
Raw counts are not physics
A diagnostic does not output a temperature or a density; it outputs voltages, counts, or photon arrivals. Turning those into physical quantities requires calibration: a mapping, often with gain, offset, and a response curve, from raw output to real units. Calibration data is what makes a measurement mean something, and it is a dataset in its own right.
Forms of calibration
- Gain and offset: linear scaling from raw counts to units.
- Response function: the wavelength- or energy-dependent sensitivity of a detector.
- Cross-calibration: aligning one diagnostic against a trusted reference.
- In-situ calibration: measuring known signals to correct drift.
Drift and recalibration
Sensors change with temperature, aging, and radiation exposure, so calibration is not one-time. Periodic recalibration produces new calibration data with a validity period. Crucially, a recalibration can change the interpretation of every measurement taken under it, so raw data must always be paired with the calibration in force at that time.
Why calibration must be versioned
If calibration is applied and the raw counts discarded, a later improvement in calibration cannot be applied, the information is gone. The durable discipline keeps raw measurements immutable and stores calibration data as a versioned dataset, so any measurement can be re-derived when calibration improves. This is versioning and provenance applied to the measurement chain.
In a fusion program
For the Hyperion breeder design, the diagnostic and calibration architecture is specified in simulation today, since the machines are design and simulation rather than built hardware. The principle carries into future operation: raw diagnostic output and its calibration are stored together and versioned, so the calibrated physics record, and its uncertainty, can be reconstructed and audited at any time.