Evaluation of Machine Learning Models for a Chipless RFID Sensor Tag

Radar cross section (RCS) is a measure of the reflective strength of a radar target. Chipless RFID tags use this principle to create a tag that can be read at a distance without needing a power-hungry radio transceiver chip and/or battery. A chipless tag consists of a pattern of conductive and dielectric materials that backscatter electromagnetic (EM) waves in a distinctive pattern. A chipless tag can be read and identified by analysing the reflected waves and matching it with a predefined EM signature. In this paper, for the first time, several regression-based machine learning (ML) models are evaluated to detect identification and sensing information for an RCS-based chipless RFID tag. The simulated EM RCS signatures containing an 8-bit identification code and six capacitive sensing values are evaluated. The EM RCS signatures are evaluated within the UWB frequency band from 3.1 to 10.6 GHz. A dataset of 1,530 simulated signatures with relevant features are utilised for model training, validation, and testing. Root mean square error (RMSE) is used as the quantitative metric to evaluate their performance. It is found that Support Vector Regression (SVR) models provide the minimum RMSE for the identification code. At the same time, the Gradient Boosted Trees (GBT) regression model performed better in detecting the sensing information.