A multiband microwave sensor integrated within a compact RF transceiver system has been developed and experimentally validated for non-invasive glucose monitoring. The proposed architecture employs a pair of identical fractal-slotted patch antennas operating across multiple resonant frequencies—2.78, 6, 6.8, 8.62, and 9.08 GHz—allowing the system to probe the dielectric response of biological tissue over a broad spectral range. Utilizing a realistic five-layer finger phantom (skin, fat, muscle, blood, and bone) and modeling blood permittivity based on the Cole–Cole equation, the sensor’s transmission response S21 was analyzed under varying glucose concentrations from 0 to 500 mg/dL. The system demonstrated measurable and repeatable shifts in S21 magnitude, phase, and resonance frequency across the operating bands. The key sensitivity outcomes include: A peak frequency-shift-based sensitivity of 0.561 MHz/(mg/dL) in the 9.08–9.28 GHz band, a maximum magnitude sensitivity of 0.041 dB/(mg/dL) at 9.08 GHz, and a notable maximum phase sensitivity of 0.176°/(mg/dL) at 8.62 GHz. These results emphasize the multiband nature of the sensor and its ability to leverage different RF features—magnitude, frequency shift, and phase—for robust and versatile non-invasive glucose detection. The integration of such RF-based multi-dimensional sensing into a single system provides a promising pathway toward compact, calibration-free, and wearable microwave biosensors. Moreover, experimental validation confirmed the agreement between simulated and measured responses, demonstrating the practical feasibility of the proposed sensor design. To further exploit the rich multiband and multi-parameter RF responses, a deep learning framework is employed to model the nonlinear relationship between glucose concentration and the extracted electromagnetic features, improving estimation robustness and noise resilience without explicit per-band calibration. Experimental validation demonstrates good agreement with full-wave simulations, confirming the practical feasibility of the proposed multi-dimensional RF sensing and learning-based glucose monitoring platform. All full-wave electromagnetic simulations were performed using CST Studio Suite with a frequency-domain finite element solver.
Hassain et al. (Thu,) studied this question.