RTD Sensor Measurement Circuit Design
An RTD sensor is the reference choice when a temperature measurement has to be accurate over a wide range, because its resistance changes with temperature in a way that is stable, repeatable and well characterised. The sensor itself is rarely the limiting factor. What limits the measurement is the wiring, the excitation current and the board layout that surrounds it.
How an RTD Responds to Temperature
The sensing element is a coil or a thin film of platinum whose resistance rises with temperature in a nearly linear way. A platinum element of 100 ohms at zero degrees Celsius, the standard Pt100, reaches about 138.5 ohms at 100 degrees and about 100 ohms again when it returns to the starting temperature, so the resistance itself is the measurand and everything else in the circuit exists to measure it precisely.
The sensitivity is only about 0.385 ohms per degree at the common ratings. A one degree error corresponds to less than four tenths of an ohm, which is a small change to resolve in a circuit that also has to tolerate cable resistance, contact resistance and the resistance of its own traces. That ratio between the signal and the unwanted resistance is what makes the wiring scheme so important.
Thin film elements are smaller and respond faster than wire wound types, while wire wound elements are more stable over long periods and at high temperature. Both follow the same resistance curve within their tolerance class, so the circuit can be designed once and used with either, provided the tolerance class and the temperature range are checked.
Lead Resistance and Wire Count
A two wire connection puts the lead resistance in series with the element, and the measurement includes it. Ten metres of a thin copper pair can easily add an ohm, which is nearly three degrees of error, and that error changes with the ambient temperature of the cable run rather than with the temperature being measured.
A three wire connection removes most of the error by measuring the resistance of one lead and subtracting it. The circuit assumes that both leads are identical and at the same temperature, which is a good assumption inside a single cable. It is the standard compromise for industrial measurements where four wires are inconvenient.
A four wire connection is the accurate solution. Two wires carry the excitation current and two carry the sense signal, so no current flows in the sense pair and the lead resistance cancels completely. The scheme costs one more conductor and a slightly more complex front end, and it is the right choice whenever the measurement has to be traceable or the cable is long.
Whichever scheme is used, the lead resistance has to be included in the error budget. Writing down the expected resistance of the cable, the connector and the traces, then converting that number into degrees, is a five minute exercise that prevents an unpleasant surprise during commissioning.
Excitation Current and Self Heating
The excitation current turns the resistance into a voltage, and its value sets the sensitivity of the measurement. One milliampere through a Pt100 produces about 100 millivolts at zero degrees and 38.5 microvolts per degree, which is a workable signal for a 16 bit converter with a little gain.
Current also heats the element through resistive dissipation, and that is the self heating error. The temperature rise is the dissipated power multiplied by the thermal resistance of the sensor package to its surroundings, and it can reach a degree in a poorly coupled probe at five milliamperes. Reducing the excitation current reduces the error in proportion to the square of the current, so a small reduction buys a large improvement.
The trade is a smaller signal and a greater sensitivity to noise. A common compromise is around 0.5 to 1 milliampere in still air, with periodic pulsed excitation at a higher current where the sensor is well coupled to a liquid and the thermal time constant is short. Pulsed excitation also avoids the error entirely if the measurement is taken before the element warms.
Signal Conditioning and Converter Choice
Most RTD front ends use either a delta sigma converter with a differential input and a programmable gain amplifier, or a precision instrumentation amplifier followed by a separate converter. Both approaches work; the integrated version saves board area and is easier to calibrate, while the discrete version gives more freedom in filtering and input protection.
Ratiometric operation is the key idea in either case. If the converter reference is derived from the same current source that excites the element, the excitation current cancels in the ratio, and the accuracy of the current source no longer matters. Only the reference and the resistance of the sensor remain in the transfer function, which is a much easier set of components to specify.
Offset and gain errors in the converter are handled by calibration rather than by component tolerance. A two point calibration against a known resistance at the bottom and top of the range removes both, provided the amplifier and converter are linear over that range. Linearity, not absolute accuracy, is the specification that calibration cannot repair.
Noise, Filtering and Cable Effects
Thermal noise sets the floor. A 100 ohm source at room temperature produces about 1.3 nanovolts per root hertz, which is negligible for a slow measurement once the bandwidth is limited to a few hertz. The noise that actually matters comes from the cable, the supply and the digital circuitry nearby, all of which can couple far more energy than the thermal floor.
A differential input with a first order filter at a few hertz removes most of the pickup, and a common mode choke at the connector reduces the current that the cable can carry into the board. Twisted pair wiring for the element leads is the most effective single measure, because it makes the loop area of the measurement path small and the coupling to external fields nearly equal on both conductors.
Mains frequency rejection deserves attention in industrial installations. A filter corner below 10 hertz attenuates both 50 and 60 hertz components, and a converter with a sinc filter that notches the line frequency does better still. Where the measurement must be fast, taking an integer number of line cycles and averaging is a simple alternative that avoids the filter settling time.
Layout of the Measurement Front End
Keep the front end small and quiet. The current source, the reference and the converter should occupy one area over a solid ground plane, with no digital trace crossing beneath them. The traces from the connector to the front end carry the excitation current, so they should be routed as a pair with the return, sized generously, and kept away from clock and switching nodes.
Trace width and resistance are part of the error budget here, not just a thermal question. A long thin trace in the excitation path adds resistance that appears directly in the measurement, so the technique described in trace width and current applies with an accuracy motive as well as a thermal one. Keeping the current path short is usually more effective than making it wide.
Cable screens and board grounds need a deliberate decision. Bond the screen at the board end only, provide a separate quiet ground for the analogue section, and return the sensor currents to that ground rather than to the digital plane. Our guide to mixed signal board design describes the partition in more detail, and it applies directly to a temperature channel that shares a board with a microcontroller.
Conformal coating is worth considering for the analogue area once the layout is final. Conformal coating keeps humidity off the high impedance nodes and prevents surface leakage from adding an error that changes with the weather, which is otherwise a fault that appears only in the field and never on the bench.
Calibration and Long Term Stability
Calibrate against a known resistance rather than a known temperature. A precision resistor of the nominal zero degree value gives a reference point that is traceable and stable, while a temperature bath introduces its own uncertainty. Two resistors, one at the bottom of the range and one at the top, are enough to set offset and gain for most applications.
Long term drift comes from three places: the sensor, the reference and the amplifier. The sensor drifts slowly and predictably, the reference drifts with the initial tolerance and the temperature coefficient, and the amplifier contributes offset drift. Specifying a reference with 10 parts per million per degree and an amplifier with 1 microvolt per degree keeps the electronics well below the sensor contribution.
A final verification step is to measure the same temperature with two channels that share a reference. Any discrepancy comes from the front end rather than from the sensor, and comparing them over a day of operation separates a warm up effect from a genuine drift. That test costs one extra channel on a prototype and answers questions that would otherwise be investigated for weeks.
FAQ
Can I use a two wire RTD in a short cable? Yes if the cable is short and the error budget allows it. Half a metre of a thick pair contributes about 0.02 ohms, which is roughly 0.05 degrees and may be acceptable.
Is a Pt1000 sensor better than a Pt100? The higher resistance gives a larger signal for the same current and reduces the relative effect of lead resistance, but it also dissipates more power for a given current. It is a good choice where lead resistance is hard to control.
How often should the channel be calibrated? Most industrial RTD channels hold their calibration for a year or more. Recalibrating at the annual service interval, and after any work that disturbs the wiring, is enough for typical process measurements.
