High-Capacity Chipless RFID System Enabled by Machine Learning Predictive Models

This paper presents a robust, high-capacity chipless encoding solution based on mapping resonance frequencies to Euclidean space. We introduce, for the first time, a methodology that exploits a usually undesired phenomenon, i.e., mutual coupling, as a method to improve the encoding capacity of chipless RFID tags. A deep analysis of the proposed method’s capability to mitigate fabrication tolerance and measurement uncertainties is performed. To address the computational challenges of the decoding phase, we employ machine learning to predict resonant frequencies of non-calibrated tags. A forecast accuracy of 100% without tolerance, and 97.6% with a tolerance of 50 MHz, is achieved using the trained model, which shows great potential to address concerns about simulation cost. A reliable space encoding efficiency of 22.5 bits/cm and spectrum encoding efficiency of 2.1 bits/GHz are achieved. As a proof of concept, we designed tags using periodically arranged, middle-notched planar dipole resonators, chosen for their fabrication-tolerant characteristics, enhanced Radar Cross Section (RCS) level and back-shielding properties.