Analog Front End Calibration Guide

Every analog front end has errors, and calibration is how a design that is built from ordinary components achieves an accuracy that no single component could deliver. The work is to know which errors are worth calibrating and which are better removed by the circuit.

The Errors a Front End Contains

offset error is the output with a zero input, and it comes from the amplifier, the converter and the thermal electromotive forces on the board. It is the error that dominates at low signal levels.

gain error is the difference between the actual and the ideal slope of the transfer function. It comes from the tolerances of the resistors, the reference and the converter, and it dominates at high signal levels.

Non linearity is the third error, and it is the one that calibration cannot remove with two numbers. It comes from the amplifier and the converter, and the cure is a better part rather than a calibration step.

Separating offset error and gain error

A two point measurement separates them. With the input at zero, the output gives the offset; with the input at a known full scale, the difference from the expected value gives the gain. The two are then corrected independently.

The measurement needs a reference that is more accurate than the target, which in practice means a standard three to ten times better. A calibration against a meter of the same accuracy as the product transfers that meter’s error into the product.

The order matters. The offset is corrected first, because a gain correction applied to an uncorrected offset moves the zero. Doing both in one step with a matrix is possible and is easy to get wrong.

Choosing the reference

the reference is the fixed point against which everything is measured, and its accuracy and stability limit the whole system. A buried zener or a band gap reference with a low temperature coefficient is the usual choice.

The reference should be in a package that is thermally stable and placed away from heat sources. Its drift with temperature appears as a scale error on every channel, so a common reference is a common failure mode.

Where several channels share a reference, its noise is correlated between them and does not average out. Measuring the same signal on two channels and comparing them removes the reference drift but not its noise.

One Point and Two Point Calibration

A one point calibration sets the offset only, and it is enough when the gain error is small compared with the accuracy required. It is common for a channel that measures a narrow range around a fixed value.

A two point calibration sets the offset and the gain, and it is the minimum for a channel that measures across a range. The two points should be placed near the ends of the range so that the correction is well conditioned.

A multi point calibration fits a curve to the non linearity as well. It costs production time and it is used where the accuracy requirement justifies it, typically in instruments rather than in embedded measurement channels.

Calibrating the Whole Chain

Calibrating the analog front end alone is not enough if the sensor is part of the chain. The sensor has its own offset and sensitivity, and a calibration applied to the electronics only leaves the sensor error untouched.

Where the sensor can be replaced, the calibration should cover the assembled unit. Where it cannot, the electronics are calibrated separately and the sensor tolerance is handled by selection or by a compensation curve.

the reference for the whole chain is a physical standard rather than an electrical one: a known weight for a load cell, a known temperature for a thermocouple, a known pressure for a transmitter. Those standards are traceable and they are what makes the calibration meaningful.

Calibration in Production

A production calibration has to be fast and repeatable. A single point at a known input, measured with a stable source, takes a few seconds and removes the largest error in most channels.

The measurement conditions have to be recorded, because the calibration is only valid at the temperature at which it was made. A unit calibrated in a cold room and used in a warm one shows a different error, and the record explains it.

Storing the coefficients with the serial number allows a unit to be restored after repair without repeating the whole procedure. It also allows a drift in one unit to be detected against the record of the others, which is a useful early warning of a design problem.

Drift and Recalibration

Drift in a well designed front end is dominated by the reference and the amplifier offset, both of which are specified over temperature. The recalibration interval follows from those figures and from the accuracy required.

A channel that needs recalibration more often than the calculation predicts usually has a moisture or a contamination problem. The board surface leakage that changes with humidity appears as an offset drift, and coating the board is the cure. The practices that keep that leakage low are described in our guide to conformal coating board protection.

A self calibration against an internal reference is possible in an instrument that has one, and it corrects the electronics without removing the sensor error. It is a useful measure for a long installation and it does not replace an occasional calibration against a standard.

Temperature and the Calibration Record

The temperature at which the calibration is made belongs in the record, and the coefficients should be applied with a temperature term if the product works over a wide range. A calibration made in a cold room and applied in a warm one leaves an error that the record explains.

A channel calibrated at twenty degrees and used at sixty has a different gain, because the reference and the resistors both move with temperature. Measuring the error at two temperatures gives the slope that the firmware needs in order to correct it, and that measurement is the last step of a complete calibration.

Verification and Faults

Verify the calibration by measuring a third point that was not used in the fit. An error at that point that is larger than expected indicates a non linearity rather than a calibration error.

Check the offset with the input shorted to the local ground rather than left open. An open input picks up whatever is nearby and the measurement means nothing.

A channel that is accurate at the calibration point and wrong elsewhere has a gain or a linearity problem, while one that is offset everywhere has an offset problem. The distinction takes a few minutes and points to the right fix. The release checks that keep such a board consistent are collected in our PCB design release checklist, and the assembly points are listed in judging PCB quality.

FAQ

Is one point calibration enough? Only when the gain error is small compared with the accuracy required. Any channel that measures across a range needs two points.

How accurate should the calibration source be? Three to ten times better than the accuracy the product claims. A source of equal accuracy simply transfers its error.

Why does my channel need recalibrating so often? Moisture or contamination on the board changes the leakage and appears as offset drift. Clean and coat the board.