AI-Driven RF Detection of Solutes in Water

Water quality monitoring is critical for public health and environmental sustainability. This paper introduces a novel integration of radio frequency (RF) antenna sensing with machine learning to enable low-cost, real-time monitoring of water quality. The antenna, designed in High-Frequency Structure Simulator (HFSS) and fabricated on a Rogers RO4003C substrate (εr= 3.55, thickness = 0.813 mm), is a compact wideband square loop microstrip antenna.The antenna’s copper-to-copper area is only 486.58 mm2(22 mm × 22.115 mm). The antenna operates across two distinct bands (3.88 – 8.77 GHz and 9.99 – 12.85 GHz), offering a total bandwidth of 7.75 GHz. It functions as a dielectric sensor, detecting salt and sugar concentrations in water by observing shifts in the reflection coefficient (S11) induced by solute-dependent permittivity changes. Measurements were performed using a Vector Network Analyzer (9 kHz–13.6 GHz), and the results confirmed clear correlations between solute concentration and antenna response. To improve solute classification, a hybrid stacking ensemble machine learning pipeline is proposed that fuses extracted antenna-based features with statistical water quality data. Multiple base learners including Random Forest, Logistic Regression, KNN, LightGBM, and CatBoost were optimized and stacked using meta-learner. The ensemble achieved 92% accuracy, outperforming a baseline of a neural network. This work establishes a new paradigm in antenna-based sensing by merging RF hardware with interpretable AI, advancing the development of intelligent, real-time environmental monitoring systems.