An integrated analog front-end and offline self-calibration framework for precision ADC linearization in low-cost embedded measurement systems

This work is motivated by transducers such as Hall-effect sensors and Integrated Electronics Piezoelectric (IEPE) devices, which share a common signal characteristic when measuring time-varying signals: a low-amplitude, time-varying component of interest superimposed upon a dominant, quasi-static direct current (DC) offset. While low-cost microcontrollers such as the ESP32 offer integrated analog-to-digital converter (ADC) capabilities, their utility for precision acquisition of such signals is limited by inherent nonlinearity, reference instability, and inadequate resolution. This paper presents an integrated hardware-firmware framework that strategically couples three complementary stages through a unifying dynamic range management strategy to systematically address these limitations. The analog front-end implements a phase-linear Sallen-Key Bessel filter to preserve signal morphology during anti-aliasing, followed by an instrumentation amplifier featuring programmable gain and an automated DC-offset nulling algorithm. The algorithm drives a 12-bit digital-to-analog converter (DAC) to adaptively center the signal within the ADC’s operating range via a binary search heuristic, without prior knowledge of the offset magnitude. To address the ESP32’s well-documented ADC nonlinearity, a piecewise linear calibration strategy was developed using a TL431 precision shunt reference and an internal DAC sweep across the full input range, generating a per-device, per-channel correction lookup table (LUT). The complete signal chain was characterized metrologically against a National Instruments data acquisition (NI DAQ) system used as an independent ground truth. Across three independent device units, the uncalibrated ADC exhibited errors of approximately 158.1±19.3 mV mean absolute error (MAE) (4.8 of full-scale). The self-contained piecewise linear calibration framework, requiring no external signal generator or factory calibration infrastructure beyond a single TL431 precision shunt reference, achieved an 88.8% improvement in measurement accuracy, reducing errors to 18.6±11.4 mV MAE (0.56% of full-scale), confirming that the calibration procedure is reproducible across hardware variants with per-device improvements ranging from 82.5% to 92.4%. These results demonstrate that the proposed framework effectively bridges the metrological gap between consumer-grade embedded silicon and laboratory-grade data acquisition (DAQ).