Instrument calibration and end-to-end transfer functions

An affine calibration uses an offset b and sensitivity K so that the output voltage is b+Kx. Two distinct known inputs determine these two parameters exactly in an idealized noiseless example. Real measurements require uncertainty in reference values and indications, repeated observations or other information about variability, and checks at additional inputs. Two points cannot establish that a nonlinear instrument is linear everywhere between or beyond them.

Applying a calibration means inverting the relation to infer x from a later voltage. The reference conditions matter: excitation, temperature, loading, range, and time since calibration can change b or K. A correction can be positive or negative. Its uncertainty describes incomplete knowledge of the needed correction, and is not the same thing as the error that the correction is intended to remove.

The first-order law of uncertainty propagation weights each input standard uncertainty by the derivative of the reported result with respect to that input. Independent contributions combine in variance. Correlated quantities require covariance terms, which can be important because offset and gain estimated from the same calibration data are often correlated. Treating them as independent without justification can overestimate or underestimate the final uncertainty.

NIST Technical Note 1297, in its 1994 edition, distinguishes standard uncertainty from expanded uncertainty and emphasizes corrections for recognized significant systematic effects. A coverage factor multiplies the combined standard uncertainty, but a chosen factor two does not automatically prove exactly ninety-five-percent coverage. Distributional assumptions, effective degrees of freedom, and the adequacy of the linear propagation model determine how an interval should be interpreted. The method and coverage statement should accompany the numerical result.

NIST's metrological-traceability policy, consulted on 18 October 2026, treats traceability as a property of a measurement result linked to a reference through a documented unbroken calibration chain, with each link contributing uncertainty. A calibration label or a claim of NIST-traceable equipment is insufficient by itself to establish that an arbitrary later experiment has the required uncertainty. The user still needs a suitable chain, conditions, and uncertainty budget for the actual measurand.

A DC calibration does not determine phase lag, bandwidth, timing error, or channel crosstalk. A dynamic end-to-end test supplies known signals at the physical or electrical input and measures amplitude and phase through the connected chain. Record stimulus amplitude, source impedance, frequency, clock relation, and filtering. Large signals can violate linearity even when a small-signal transfer is accurate. Reporting these limits makes a calibrated result reproducible and separates a calculated model from a verified instrument performance claim.