Examination of TinyML Approach in ESP32-Based NFC Applications

This paper presents a comparison between the efficiency of non-learning and machine learning approaches on edge devices, using a typical NFC system centred around commercial ESP32 microcontroller as an example. NFC tag data is read by the NFC reader and forwarded to two ESP32 modules, one using rule-based non-learning approach, and the other one using pretrained model for text correction. Read success rate as well as self-heating are analyzed on the basis of 100 unsuccessful data processing attempts. Machine learning method shows $4 \%$ better read success rate, but also suffers from more self-heating.