Machine-Learning-Enhanced Chipless RFID Tag Localization Using UWB-IR Backscatter Analysis for IoT Applications

This study presents a machine learning (ML)-enhanced framework for the detection, decoding, and localization of chipless radio-frequency identification (RFID) tags using cost-efficient ultrawideband impulse radio (UWB-IR) readers. By combining frequency-domain backscatter analysis with round-trip time-of-flight (RTOF) measurements, we propose an advanced signal processing architecture that significantly improves localization accuracy. To further enhance system robustness in noisy or multipath environments, a supervised ML classifier is integrated into the identification pipeline. Key features extracted from antenna mode backscatter signals, such as resonant amplitudes, signal-to-noise ratio (SNR) metrics, and temporal characteristics are used to train a support vector machine (SVM), achieving over 96% identification accuracy at 10 dB SNR, and maintaining high performance even under challenging conditions. This design improves resonance stability, environmental durability, and omnidirectional readability, making it highly suitable for real-time Internet of Things (IoT) applications, such as asset tracking and smart infrastructure monitoring. The proposed system is validated through comprehensive electromagnetic simulations in CST Studio Suite and MATLAB, as well as experimental measurements in anechoic chamber setups. Results demonstrate consistent performance, with a detection range of 30 cm and angular resolution of 60°, even under varying environmental conditions.