Sensors, Signal Conditioning and ADCs

The measurement chain

A logger never sees the physical quantity; it sees a voltage after a sensor, some conditioning and a converter. Each link adds error, and the final number is only as good as the weakest. Work along the chain: sensor (turns temperature, pressure or load into a small electrical change), conditioning (excitation, amplification, filtering, isolation), converter (an ADC turning volts into counts), then the scaling from counts back to engineering units. The sensors themselves are described in Sensors and Field Devices; this page covers what the logger must do with them.

Choosing the sensor

Sensor	Output	Typical use	Watch out for
RTD (Pt100, Pt1000)	Resistance, 0.385 Ω/°C (Pt100)	Accurate temperature, −200 to 600 °C	Lead resistance, self heating, needs excitation
Thermocouple	Microvolts (about 41 µV/°C, type K)	Wide range, hot, fast, rugged	Cold-junction compensation, noise, wire type, drift
Thermistor (NTC)	Resistance, strongly non-linear	Cheap, sensitive, −50 to 150 °C	Non-linear, needs a curve (Steinhart–Hart) and a stable divider
4–20 mA transmitter	Current loop	Process: pressure, level, flow, temperature	Loop power, shunt resistor, broken-wire detection
0–10 V / 0–5 V sensor	Voltage	Short runs inside a cabinet	Noise pick-up and ground offsets over distance
Strain gauge / load cell	Millivolts per volt of excitation	Force, weight, torque	Needs an amplifier, bridge excitation and a low-noise ADC
Digital sensor (I²C, SPI, 1-Wire)	Numbers directly	Board-level temperature, humidity, pressure	Bus length, address clashes, built-in filtering you cannot change
Current transformer / Hall	AC voltage or current proportional to load	Energy and motor current	Burden resistor, RMS calculation and sampling rate

• Range: choose the sensor and input range so that the usual signal uses most of the span, with headroom for faults, not 5 % of it.
• Response time: a sensor in a thermowell may have a time constant of tens of seconds, which limits what the sample rate can show (see Planning a Logging System).
• Place it where the quantity is: a probe that is warmed by its cable or shaded from the air reads the wrong thing very accurately.

RTDs and wiring

A Pt100 changes by only 0.385 Ω per °C, so the resistance of the wires matters. A 2-wire connection adds both leads to the reading; a 3-wire connection cancels most of it if the leads match; a 4-wire connection measures the voltage across the sensor only, so lead resistance drops out entirely.

Thermocouples

A thermocouple makes a voltage from the difference between its hot junction and the point where it meets the copper of the logger (the cold junction). The logger must measure that cold-junction temperature and add it back, called cold-junction compensation. Without it the reading is simply wrong by the temperature of the terminals.

4–20 mA and voltage signals

The 4–20 mA loop is the industrial standard for good reasons: current is the same everywhere in the loop, so wire resistance does not matter; 4 mA, not 0, is “zero”, so a broken wire (0 mA) is distinguishable from a real zero; and the transmitter can be powered from the loop. The logger reads the current through a precision shunt resistor.

• Voltage signals (0–10 V) are easy but pick up noise and ground differences on long runs: use shielded twisted pair, a differential input and keep them short.
• Isolate the input when the sensor and logger have different grounds, or when the signal comes from a plant with ground loops; a ground loop shows as hum or a steady offset that changes when something else is switched.
• Loop power: a 2-wire transmitter needs 10 to 30 V in series with the loop, with the shunt and any other devices dropping voltage: check the supply can drive 20 mA through all of them.

Bridges and small signals

Strain gauges and load cells are Wheatstone bridges that unbalance by a few millivolts for each volt of excitation, typically 2 mV/V at full load. At 5 V excitation a 2 mV/V cell gives only 10 mV at full load, so it needs an amplifier or a 24-bit delta-sigma ADC with a programmable gain (the HX711 and ADS1232 are common). Use ratiometric measurement (the ADC reference is the bridge excitation) so that supply drift cancels, and use 6-wire connections (sense lines) over long cables.

ADC resolution and the reference

Bits	Levels	Step with a 3.3 V reference
10 bit	1 024	3.223 mV
12 bit	4 096	0.806 mV
16 bit	65 536	0.050 mV
24 bit	16 777 216	0.000 mV

• Effective number of bits (ENOB) is lower than the quoted bits, because noise, reference and layout take their share. A “12-bit” microcontroller ADC often gives 9 to 10 usable bits. Read the datasheet’s noise and ENOB figures.
• The reference sets the accuracy. A converter using the supply as its reference moves when the supply moves. For anything accurate use a precision reference or a ratiometric sensor.
• Input range: use the whole range. A 0.5 V signal on a 3.3 V ADC wastes about 85 % of the bits; amplify it or switch to a lower range.
• External ADCs (ADS1115, ADS1256, MCP3564) give better noise and stability than a microcontroller’s built-in one, at the price of an extra part and bus (see Embedded Systems).

Noise, grounding and shielding

Symptom	Likely cause	Cure
50 or 60 Hz ripple on the reading	Mains pick-up on the wires or on the supply	Shielded twisted pair, a differential input, integrate over a whole number of mains cycles (20 ms rejects 50 Hz), move cables away from power wiring
Spikes when a motor or relay starts	Switching noise coupling in	Filter or snubber at the load, separate routing, ferrites, opto-isolation
Slow wander that tracks the ambient temperature	Offset or reference drift; thermocouple junction effects	Better reference, an isothermal block, auto-zero or a periodic calibration reading
A steady offset that changes with other equipment	A ground loop	One ground point for the signal, isolation, a differential input
Readings jump between two values	A noisy value on a code boundary, or the sensor is switching	Average, add hysteresis in the display, check the sensor
A channel affects its neighbor	Multiplexer charge injection or settling	Wait for the input to settle, add a buffer, or sample slower

• Shield one end only (at the logger), unless you know the design needs both, to avoid ground-loop currents in the shield.
• Twist the pair so that interference is picked up equally by both wires and canceled by a differential input.
• Keep high-impedance inputs short; a long wire on a high-impedance node is an antenna.
• Look at the raw signal: a scope or a burst of fast samples shows noise that a slow logger hides (see Sampling, Aliasing and Filtering).

An uncertainty budget

Every part of the chain has an error. If the errors are independent, they combine as the root of the sum of the squares (RSS), not by simple addition: the worst case of all errors together is rare, and a straight sum overstates it.

u = √( u₁² + u₂² + u₃² + … )

Source	Error	Squared
Sensor accuracy (class)	±0.80 °C	0.64
Amplifier gain and offset	±0.30 °C	0.09
ADC and reference	±0.15 °C	0.02
Self heating	±0.02 °C	0.00
Resolution and noise	±0.10 °C	0.01
Installation (heat conduction)	±0.40 °C	0.16

Related

• Data Logging: the overview of this guide.
• Sampling, Aliasing and Filtering: what to do about noise and aliasing after the converter.
• Data Quality, Analysis and Reporting: calibration and validation of the channel.
• Sensors and Field Devices and the Analog Scaling Tool.
• Embedded Systems: reading an ADC on a microcontroller.

FAQ

RTD or thermocouple?

An RTD is more accurate and stable up to a few hundred degrees; a thermocouple is cheaper, faster, rugged and goes much hotter, but needs cold-junction compensation and gives smaller, noisier signals.

Why is 4–20 mA better than 0–10 V?

Current is the same along the loop, so wire resistance does not matter, noise pick-up is lower, a broken wire shows as 0 mA (a detectable fault), and the transmitter is powered from the loop.

How many bits do I need?

Enough that one step is well below your accuracy requirement, and no more than the noise allows. Check the effective bits, not the quoted ones; 16 bits of a noisy ADC may be 12 real bits.

Why do my readings drift when the room warms up?

Offset and reference drift in the electronics, cold-junction error, or the sensor itself. Test by logging a fixed precision resistor in place of the sensor; if it still drifts, the electronics are the cause.

References

• Dally, J. W., Riley, W. F. and McConnell, K. G., Instrumentation for Engineering Measurements.
• Analog Devices, Data Conversion Handbook and application notes on RTD, thermocouple and bridge measurements.
• IEC 60751 (Pt100), IEC 60584 (thermocouples), NAMUR NE 43 (4–20 mA fault levels), JCGM 100 (uncertainty).

Last reviewed: October 2026. Resistances, sensitivities and errors are typical values: use your sensor’s datasheet.