Accurate and Computationally Efficient Lookup-Table-Based Thermistor Temperature Conversion for Embedded Systems

Lookup-table (LUT)-based temperature conversion can reduce the computational cost of nonlinear thermistor calibration, but full-resolution LUTs require substantial memory in resource-constrained embedded systems. This study systematically evaluates a reduced ADC-indexed LUT with direct interval indexing and linear interpolation for temperature conversion using four commercial negative temperature coefficient (NTC) thermistors. A dense reference resistance–temperature relationship was generated from manufacturer data, and the voltage-divider reference resistance was optimized considering ADC quantization. The reported errors therefore represent numerical approximation errors relative to the nominal manufacturer-derived R–T characteristics rather than absolute sensor measurement accuracy. Reduced LUT entries were reconstructed by least-squares optimization, and the effects of LUT resolution and operating temperature range were evaluated using a 12-bit ADC baseline. Over a wide temperature range of −38–123 °C, an 8-bit LUT maintained RMSE, maximum absolute error, and 95th-percentile error comparable to or lower than those of a five-parameter Hoge-3 calibration equation. Over the narrower 45–103 °C range relevant to thermal cycling for real-time polymerase chain reaction, the LUT could be reduced to 6 bits while maintaining comparable numerical approximation error relative to the manufacturer-derived reference characteristic. For the floating-point implementations evaluated in this study, the reduced LUT achieved execution-time speed-ups of 10.6×, 12.7×, and 9.2× on Cortex-M4F, Cortex-M0+, and 8-bit RISC microcontrollers, respectively, relative to the five-parameter equation. These results demonstrate that LUT resolution can be selected according to the operating temperature range to achieve a practical trade-off among numerical approximation error, memory usage, and computational cost.