A Predictive-Model-Assisted Moisture Content Sensor by Monitoring the DC-Unstable States of a Microwave Self-Oscillating Antenna

This work presents a portable moisture content (MC) sensor, for tree hydration control, adopting a microwave self-oscillating antenna. The operating principle exploits the dependence of the oscillator nonlinear steady-state regimes on the dispersive load, represented by a patch antenna loaded by wood in different MC conditions. A novel read-out approach that eliminates radio frequency (RF) measurements and relies solely on the oscillator's steady-state dc bias conditions is proposed. First, a free-running oscillator, adopting a pseudomorphic highly electron mobility transistor (pHEMT) is designed in the 2.4-GHz band, with a 50-Ω load. The tuning capabilities are analyzed by nonlinear/electromagnetic (NL/EM) co-simulation, with the oscillator loaded by the dispersive patch-antenna impedances corresponding to different hydration conditions. A comprehensive experimental campaign of the fabricated prototype characterizes the wood sample. By tuning the HEMT bias voltages, distinguishable dc drain currents are retrieved and correlated with the wood's hydration state. These results are fully validated by an independent method based on the standard weighting procedure. Consequently, moisture monitoring can be conducted automatically by a low-cost microcontroller (μController) and the system can be adopted as a portable device for the proposed purposes. To enable the adaptivity to new unseen data and different MC scenarios, a low-computation machine learning (ML) algorithm is developed, enabling the system to operate in real time and as a stand-alone system.