Artificial neural network-based standalone tunable RF sensor system.

Area of Science:

• RF sensing
• Artificial Neural Networks (ANNs)
• Electromagnetics

Background:

• Conventional resonant sensors often require multiple structures for broad characterization.
• Existing tunable sensors may have limited tuning ranges and rely on active tuning components.

Purpose of the Study:

• To propose a novel, automated, tunable Radio Frequency (RF) sensor system.
• To address the non-linear inverse characterization problem with improved accuracy.
• To eliminate the need for multiple resonant structures in sensor design.

Main Methods:

• Development of a unified design topology integrating a modified complementary split-ring resonator (CSRR) and microstrip line.
• Training an ANN using numerically generated S-parameters with Levenberg-Marquardt backpropagation and Bayesian regularization.
• Experimental validation using standard dielectric samples to extract complex permittivity.

Main Results:

• Achieved a significantly higher tuning range (1900 MHz) compared to previous designs (580 MHz).
• Demonstrated a dielectric sensitivity of approximately 8.8%.
• The ANN-based system accurately extracted complex permittivity across the operational frequency range.

Conclusions:

• The proposed ANN-based tunable RF sensor system offers an accurate and automated solution for material characterization.
• The novel design alleviates the need for active tuning circuitry and multiple resonant structures.
• This approach enhances the efficiency and applicability of RF sensing technology.