A Machine Learning Approach-based Chipless RFID System for Robust Detection in Real-world Implementations

This work describes a novel method based on a machine learning (ML) approach for highly accurate and robust identification of chipless RFID applications in realistic environments. For this purpose, effective transponder reading is achieved not only for a wide variety of ranges and contexts but also for different commercial objects attached to the chipless RFID tag, while providing tag-ID detection accuracy of up to 98.5 %. A UWB transmitting tag antenna, four tags encoding the four 2-bit IDs, and a UWB receiving tag antenna were inkjet-printed onto flexible low-cost substrates and interrogated without cross-talk or clutter interference de-embedding at ranges from 2 to 50 cm, with different objects attached to the tag (non-conductive, aluminum can, and plastic bottle filled with water), for both configurations with and without the presence of scattering objects in the vicinity of the tags and reader. Finally, a Support Vector Machine (SVM) using the information of the measured transmission coefficients (S
21
) outperforms the other methods displaying reading accuracies 98.5 %.