Sensor Callibration

When using temperature sensors, you are actually measuring a voltage, and relating that to what the operating temperature of the sensor must be. If you can avoid errors in the voltage measurements, and represent the relationship between voltage and temperature more accurately, you can get better temperature readings. How much effort is worthwhile depends on the application's error tolerances.

There are four adjustments that a good calibration can provide.

• OFFSET
  All voltages are measured with respect to a reference. All devices operate at some operating voltage. Any displacements in these voltages, or any consistent errors during measurement, will produce consistent errors that affect all measurements. Offset corrections make these errors as small as possible.
• GAIN
  The voltage that you measure is not really the voltage present on the sensor device. Amplifiers and attenuation between the sensor and the digitizing converter change the signal level. To recover the sensor information, you must restore the data to the original level accurately. Uncorrected gain errors tend to produce measurement errors that change consistently across the operating range.
• LINEARIZATION
  The relationship between measured voltage and sensed temperature is in general nonlinear and dependent on the physical properties of each sensor type. Over a limited range, a simple linear function is often a sufficient approximation, but a more complicated curve is necessary to describe the relationship accurately. Generalized curves defined by standards help, but they won't match any individual device perfectly. For best accuracy, you need to calibrate, and adjust the coefficient values of the conversion function.

After applying the offset, gain, and linearization corrections, the results might not be in the most useful form. A good follow-up step is the following:

• UNIT SCALING
  Convert the results to a common and useful representation. For example, present all temperature measurements in degrees C.

Calibrating linearization curves

Calibration is a process of aligning what your formulas say with what real devices actually do. This involves taking some accurate measurements.
You can't control what a sensor does directly. It responds based on its physical properties. But you can to some extent control what your sensor measures. You can establish a set of temperature levels that span the operating range, and measure those levels with a laboratory-grade temperature standard. For each of those measured temperature points, observe the response level of the sensor. If you construct a curve that passes through these points, you will have a very good calibration specialized for the individual sensor.
Given the temperature vs. voltage data set, treat the sensor readings as noisy input values, and the temperature measurements as the corresponding output values to be produced. Taking multiple measurements and averaging them helps to obtain the best possible data quality for calibration.
From here, there are two ways that you can go.

1. Explicit function. Use a processing command that applies a "calibrated conversion function" for each measurement. This can be a very generic calculation like a DAPL expression or a GENPOLY command, or it can be a specialized conversion such as the RTD command.
2. Piece-wise linear approximation. Select a set of representative points along the curve and approximate the curve with straight lines "point to point." Code the "breakpoints" as vectors and supply them to the DAPL command INTERP, which can then locate the appropriate line section and evaluate the function at intermediate points.