Temperature Sensor: NTC, RTD Pt100, Infrared Thermopile

Why Temperature Measurement Matters in Medical Devices

Temperature is rarely the headline vital sign, but it is the one most likely to trigger a clinical decision on its own: a 38.5 °C reading on a neonate, a 1 °C drift in a dialysis fluid loop, or a thermopile forehead scanner disagreeing with triage staff by 0.3 °C. Getting this wrong has consequences that are both clinical and regulatory.

Clinical accuracy requirements and regulatory tolerances

Consumer thermometers tolerate ±0.3–0.5 °C error. Clinical electronic thermometers sold as medical devices are typically held to ±0.1–0.2 °C across the clinically relevant range (35–42 °C) under standards such as ASTM E1112 (patient thermometers) and ISO 80601-2-56. That tolerance band is small enough that thermistor nonlinearity, RTD lead resistance, and thermopile emissivity error all become first-order design concerns rather than footnotes.

Role of temperature in diagnostic confidence

Temperature rarely stands alone. It contextualizes other signals: a fever changes heart-rate baseline interpretation relevant to the cardiac cycle, skin temperature shifts confound galvanic skin response baselines, and peripheral vasoconstriction from cold skin alters pulse transit timing used in pulse wave velocity estimates. A temperature channel that drifts silently can corrupt the interpretation of channels that otherwise look clean.

Integration challenges in wearable and implantable systems

Wearables fight self-heating from adjacent electronics, poor thermal coupling to skin, and motion-induced contact variability. Implantables add long-term stability constraints — a sensor drifting 0.05 °C/year inside an ICP monitor housing cannot be recalibrated without surgery, so component selection and burn-in become part of the design, not an afterthought.

NTC Thermistors: Fast Response, Nonlinear Behavior

Negative Temperature Coefficient principle and B-parameter

NTC thermistors are sintered metal-oxide ceramics whose resistance falls exponentially as temperature rises. The simplified Beta (B-parameter) model is:

R(T) = R0 · exp[B(1/T − 1/T0)]

with T in kelvin, R0 the resistance at reference T0 (usually 25 °C), and B typically 3000–4500 K depending on material. Beta itself varies slightly with temperature range, which is why single-B models drift by several tenths of a degree outside a narrow span.

Resistance-temperature transfer function and linearization techniques

For wider spans or tighter accuracy, firmware should use the Steinhart-Hart equation:

1/T = a + b·ln(R) + c·(ln R)³

Fit a, b, c from three calibration points spanning your clinical range (e.g., 30 °C, 37 °C, 42 °C) using the manufacturer's resistance data or your own oven calibration. A worked example:

// Steinhart-Hart inverse: returns temperature in Celsius from resistance in ohms
float sh_a = 1.129148e-3f, sh_b = 2.34125e-4f, sh_c = 8.76741e-8f;

float ntc_temperature_c(float r_ohms) {
    float lnR = logf(r_ohms);
    float invT = sh_a + sh_b * lnR + sh_c * lnR * lnR * lnR;
    return (1.0f / invT) - 273.15f;
}

Coefficients here are illustrative; derive your own from calibration data or the thermistor's datasheet resistance table, not from generic internet values.

Self-heating effects and thermal time constant

Excitation current dissipates power in the thermistor itself (I²R), raising its temperature above the medium it measures. A thermistor with a dissipation constant of 1 mW/°C self-heating error and a 100 µA drive at 10 kΩ dissipates 1 µW — negligible — but careless designs using a 1 mA drive at the same resistance dissipate 10 mW, producing several tenths of a degree of self-heating error. The thermal time constant (seconds to reach 63% of a step change) trades off against probe size: smaller beads respond faster but have lower thermal mass margin against self-heating.

Circuit design for biomedical applications (bias current, excitation voltage)

Use pulsed excitation rather than continuous: energize the thermistor only during the ADC sample window, then power it down. This cuts average self-heating by the duty cycle factor and saves battery life in wearables. A ratiometric divider against a precision reference resistor, rather than an absolute voltage measurement, cancels supply-voltage drift.

Typical medical accuracy and stability over time

Well-calibrated NTC circuits reach ±0.1 °C over a narrow clinical span (35–42 °C) but can drift 0.1–0.2 °C per year from oxide-layer aging, which is why single-use disposable probes (oral, axillary) favor thermistors while long-service devices often don't.

RTD Pt100: Precision and Long-Term Stability

Platinum resistance thermometer (PRT) fundamentals and Callendar-Van Dusen equation

A Pt100 is 100 Ω of platinum wire or film at 0 °C, with resistance rising nearly linearly (+0.385 Ω/°C nominal, per IEC 60751). The Callendar-Van Dusen equation models the slight curvature precisely enough for lookup-table firmware implementations.

2-wire, 3-wire, and 4-wire measurement configurations

A 2-wire RTD includes lead resistance directly in the measurement — a few meters of 26 AWG wire can add 0.1–0.3 Ω, equivalent to roughly 0.3–0.8 °C error. A 3-wire configuration cancels lead resistance assuming both leads match. A 4-wire (Kelvin) configuration drives current through one pair and senses voltage through a separate high-impedance pair, eliminating lead resistance error entirely — the standard for bench-grade and reference-grade medical measurement.

Excitation current selection and lead resistance compensation

Typical excitation is 100–1000 µA, chosen low enough to keep self-heating under ~0.1 °C in air, and switched or alternated in polarity to cancel thermocouple-effect offsets at connector junctions.

Repeatability, hysteresis, and aging characteristics

Pt100 sensors show minimal hysteresis and drift under 0.01 °C/year in properly packaged designs, which is why they anchor calibration-lab reference baths and devices needing long recalibration intervals.

When to choose Pt100 over thermistors: accuracy vs. cost trade-off

Choose Pt100 when you need absolute accuracy better than ±0.1 °C sustained over years, or wide-range linearity. Choose NTC when fast response, small size, and low cost dominate and a narrow, well-calibrated span suffices — most disposable patient-contact probes.

Infrared Thermopile Sensors: Non-Contact Measurement

Thermopile detector array architecture and thermal isolation

A thermopile is a series of thermocouple junctions on a thermally isolated membrane; absorbed infrared radiation heats the "hot" junctions relative to a reference, generating a microvolt-to-millivolt signal proportional to incident power.

Blackbody radiation principles and Planck's law in sensor design

Output follows the Stefan-Boltzmann relationship integrated over the sensor's spectral bandpass — net signal is proportional to (Tobject⁴ − Tsensor⁴), which is why sensor-body temperature tracking matters as much as the target.

Emissivity correction and surface property dependencies

Human skin emissivity is approximately 0.98, close enough to blackbody that most clinical thermometers hardcode this value, but any air gap, moisture, or non-skin surface (plastic housing, hair) will skew readings; firmware should expose an emissivity coefficient as a configurable calibration parameter rather than a fixed constant.

Distance-to-spot ratio and optical field of view constraints

The sensor's field of view defines how much of the measured spot is actual target versus background — exceeding the rated distance-to-spot ratio lets background radiation (room air, surrounding tissue) dilute the reading, a common source of forehead-thermometer complaints.

Reference junction (cold junction) compensation

Because thermopile output measures a differential temperature, an onboard thermistor or silicon bandgap sensor measures the sensor die temperature, and firmware adds it back: T_object = f(V_thermopile) + T_reference. Skipping this step is the single most common thermopile integration bug — the sensor appears to work at room temperature and fails whenever ambient shifts.

Applications: tympanic, forehead, and non-contact fever screening

Tympanic probes benefit from proximity to the tympanic membrane (close to core temperature); forehead and screening devices trade accuracy for speed and zero cross-contamination risk, at the cost of higher environmental sensitivity.

Signal Conditioning and Analog Front-End Design

Excitation sources for resistance-based sensors (constant voltage vs. constant current)

Constant-current excitation is preferred for RTDs because it makes the voltage output directly proportional to resistance change; constant-voltage dividers are simpler and adequate for NTC circuits where ratiometric measurement already cancels supply variation.

Amplifier noise and offset considerations for millivolt-level signals

Thermopile signals in the tens-of-microvolts-to-millivolts range demand low-offset, low-drift instrumentation amplifiers (offset drift under 1 µV/°C) — a generic op-amp's own thermal offset can exceed the signal you're trying to measure.

ADC resolution and sampling rate requirements for medical-grade temperature

A 0.1 °C resolution target over a Pt100's ~0.385 Ω/°C slope requires sub-milliohm resolution in the measurement chain — practically, 16–24 bit delta-sigma ADCs with on-chip PGA and reference-ratiometric measurement are standard, sampled at a few Hz since thermal signals change slowly.

Filtering strategies: anti-aliasing and noise rejection

A simple RC low-pass ahead of the ADC, combined with the ADC's own sinc filter and 50/60 Hz notch rejection (via integration time selection), handles most mains-coupled noise without adding measurable lag to a signal with multi-second time constants.

Thermal drift compensation in AFE circuits

Place voltage references and critical resistors away from heat-generating components (regulators, radios), and consider a chopper-stabilized or auto-zeroing amplifier topology to cancel the AFE's own temperature-dependent offset — ironic but necessary in a temperature-measurement circuit.

Comparison Matrix and Selection Criteria

Accuracy, repeatability, and long-term drift across sensor types

NTC: ±0.1–0.2 °C narrow-span, moderate long-term drift. Pt100: ±0.05–0.1 °C wide-span, minimal drift. Thermopile: ±0.1–0.3 °C, dependent on emissivity and optics rather than element aging.

Response time and measurement lag in dynamic thermal environments

NTC beads respond in 1–10 s depending on packaging; Pt100 elements in metal sheaths respond in 5–30 s; thermopiles respond in under 1 s since they sense radiated power with no thermal mass to equilibrate.

Cost, miniaturization, and power consumption trade-offs

NTC is cheapest and smallest; Pt100 costs more and needs more board area for 4-wire routing; thermopile modules cost more per unit but eliminate patient contact entirely, which has its own value in infection control.

Environmental robustness: humidity, sterilization, biocompatibility

Contact sensors need biocompatible, sterilizable encapsulation (autoclave-rated RTDs exist; thermistor beads are typically single-use disposables); thermopile optics require a clean, non-fogging window and are sensitive to condensation.

Integration with medical device architecture (wearable vs. bench vs. contact)

Wearables favor NTC for size and power; bench/reference instruments favor Pt100; triage and infection-control-sensitive settings favor thermopile non-contact designs.

Practical Implementation: Firmware and Calibration

Curve-fitting and lookup table approaches for nonlinear sensors

For resource-constrained MCUs, a precomputed lookup table with linear interpolation between points (every 0.5–1 °C) is faster than evaluating Steinhart-Hart in real time and avoids floating-point log/exp calls on cores without an FPU.

Factory calibration vs. field recalibration protocols

Factory calibration against a traceable reference bath at 2–3 points, with coefficients stored in device EEPROM, is standard; field recalibration is rarely practical for implantables but common for bench instruments with removable probes.

Handling sensor faults and out-of-range conditions

Detect open or shorted sensors by checking raw ADC codes against physically impossible bounds (e.g., resistance below the sensor's theoretical minimum), and flag thermopile readings as invalid when the reference junction sensor itself is out of its rated range.

Real-time temperature trending and alarm thresholds

Apply a short moving average (3–5 samples) to suppress ADC noise without masking genuine fever onset, and set alarm thresholds with hysteresis (e.g., trigger at 38.0 °C, clear at 37.7 °C) to avoid chattering near the boundary.

Validation during clinical validation studies

Validation protocols should include side-by-side comparison against a reference thermometer across the full clinical population range, not just a single calibration bath temperature, since nonlinearity and emissivity errors often only appear away from the calibration point.