Why a PT100 or PT1000 reading is wrong by a degree or two

A platinum RTD is one of the most repeatable temperature sensors you can buy, yet the first board often reads a degree or two off. It also drifts when the cable moves or the board warms up. The sensor is rarely the problem. The error comes from the wires, the current pushed through the element, or the resistor the converter compares it against.

This note walks through those three sources in the order we would check them.

The sensor changes very little, so everything else matters

A PT100 is 100 Ω at 0 °C and moves by roughly 0.385 Ω for each degree near that point. A PT1000 is ten times larger and moves ten times as much in ohms, with the same fractional change. You are measuring a small change on top of a large resistance, and every stray fraction of an ohm in the path reads as temperature.

That is why lead wire resistance, which looks negligible on a multimeter, shows up directly in the result. A tenth of an ohm in a PT100 circuit is a quarter of a degree. The same tenth of an ohm in a PT1000 circuit is a much smaller error. If your cable run is long and you have a choice, the higher-resistance element is the kinder one to wire.

Lead wires: how many you need depends on the run

The wiring options are well described in Texas Instruments’ application note on universal RTD interfaces (SBAA275A). The trade-offs are these:

Wiring	What the measurement sees	Where it fits
2-wire	Element plus both leads, inseparable	Very short runs, or when you can calibrate out a fixed offset
3-wire	Compensates lead resistance if the two leads match	Most product cabling
4-wire	Separate force and sense leads, lead error mostly removed	Highest accuracy, or long and variable runs

Two-wire adds the lead resistance to the element and nothing can separate them afterward. If the leads are short and fixed, you can measure the offset once at a known temperature and subtract it. That falls apart when the cable is a customer-supplied or field-replaceable part, because the offset changes with every cable.

Three-wire works by measuring in a way that cancels one lead against another. The TI note is explicit that this compensation assumes the lead resistances match. Analog Devices makes the same point in its RTD design article: the converter sees the element plus one lead, and the other lead drops out because the converter’s inputs draw almost no current. That is a good deal on a cable with three conductors of the same gauge and length. It is a poor one if a connector contact adds resistance to only one of the three.

Four-wire carries the excitation current on one pair and senses voltage on the other. The sense leads carry almost no current, so their resistance does not appear in the reading. It costs a conductor and a few more pins, and it is the right answer when the cable is long, changes between installations, or passes through connectors you do not control.

A practical point for any wiring: the connector is part of the lead resistance. A worn or oxidized contact in series with the excitation path is a common reason a reading wanders when someone touches the cable. In a 3-wire circuit, a bad contact on one lead breaks the matching assumption. Fixed copper traces count too. Keep the traces to the connector matched in length and width, and route them together.

The ratiometric trick, and why the reference resistor sets your accuracy

Most dedicated RTD converters, such as the Analog Devices MAX31865, do not measure the element against an absolute voltage. They convert the ratio of the RTD resistance to a reference resistor on the board. The datasheet describes an external resistor that sets the sensitivity and a delta-sigma converter that digitizes that ratio. It also says the part accepts platinum RTDs from PT100 to PT1000 in 2-, 3- or 4-wire connections.

Ratiometric measurement helps in one important way. The excitation current flows through both the RTD and the reference resistor, so a change in the current affects both equally and cancels in the ratio. You do not need a precision current source. You need a precision resistor.

That moves the accuracy question to the resistor:

• Tolerance and temperature coefficient. The reference resistor’s error becomes a direct gain error in the temperature reading. A general-purpose thick-film part is the wrong choice for a precision front end. Pick a part whose tolerance and drift you have checked against the accuracy you promised, and note that it sits on a board that warms up.
• Value matched to the element. The converter’s datasheet specifies how the reference resistor relates to the RTD type you are using. Do not copy a value from a breakout board schematic. Check it against the datasheet for your exact converter and your exact element.
• Placement. Keep the reference resistor close to the converter and away from the regulator, power inductor or any part that heats the board. A reference that sits a few degrees warmer than the sensing cable’s connector is a slow, hard-to-find drift.

If you build the front end from a general-purpose delta-sigma ADC instead of a dedicated converter, the same principle applies. Derive the converter’s reference from the same current that excites the RTD, rather than from a separate reference. The TI note covers low-side and high-side reference circuits and the equations for each wiring type, and it is the place to start for that route.

Self-heating and excitation current

The excitation current dissipates power in the element, and a probe in still air or a small thermal mass has little ability to shed it. The reading rises above the true temperature by an amount that depends on the current, the element and how well the probe is coupled to what it is measuring. The TI note lists self-heating as an explicit consideration when you set the excitation current. We will not quote a number here, because it depends entirely on the probe construction. The manufacturer’s datasheet for your probe gives a self-heating coefficient, and that is the figure to use.

The trade-off is plain. A lower current heats the element less but gives the converter a smaller signal against its noise. For a slow, mostly static measurement, the better choice is usually to lower the current and average more. Some converters also let you switch the bias off between conversions, so the element is only powered while it is being read. If your firmware reads once a minute, there is no reason to leave the element energized the rest of the time.

Self-heating also explains a symptom people mistake for a bad sensor. A probe reads correctly in a stirred bath and reads high in still air. The sensor is fine. The excitation current is the cause.

Checks we would run, in order

1. Short the sensor input with a precision resistor of a known value near the middle of your range. Read the converter’s raw ratio and compare. This separates the converter and reference resistor from everything outside the board.
2. Swap the resistor for a decade box or two resistors at the ends of your range. A gain error points at the reference resistor; an offset points at the leads, the connector or the board.
3. Move the cable while watching the reading. A reading that jumps with cable movement points to a contact or a mismatched lead, not the sensor.
4. Compare a short-pulse read against a continuous one. A difference between them is self-heating.
5. Use the converter’s fault detection. The MAX31865, for example, can flag an open element, a shorted element, and the cable fault conditions, and the firmware should read those flags instead of reporting a nonsense temperature. A reading of a plausible but wrong value is worse than an error.
6. Check for noise coupling. Run the sensor wires as a twisted group, away from switching supplies and motor drive, and select the converter’s mains-frequency filter setting to match the region where the product is used.

If you are also working on a thermocouple channel, the failure modes are different, and we cover them in a separate note on why a thermocouple reading is off by a few degrees.

What goes into the calibration

A one-point or two-point calibration against a reference removes most of the remaining gain and offset error, and it is cheap to build into a test step. Store the correction in non-volatile memory with the serial number, and apply it in firmware. That also means the production fixture needs a way to present a known resistance to each RTD channel, and a resistor network or a precision decade source on the fixture does that without a temperature bath.

How we approach it

We start from the cable and connector, pick the wiring to match, then size the reference resistor and excitation current against the probe’s own datasheet. If you send us the probe part number, the cable length and the accuracy you need, we can tell you which of these errors will dominate before any board is laid out.

Sources: TI SBAA275A, universal RTD interfaces · Analog Devices, RTD measurement system design essentials · Analog Devices MAX31865 product page · Analog Devices Application Note 046 (4-wire measurement)