Enhancing Microwave Sensor Performance With Ultrahigh Q Features Using CycleGAN

In this work, a microwave planar sensor is used for liquid material characterization. Two identical complementary split ring resonators (CSRRs) operating at 3 GHz are coupled to create a highly sensitive capacitive region. The moderate quality factor of the sensor
$\approx 230$
is significantly improved up to
$\approx 5040$
with loss compensation using a regenerative amplifier. The moderate quality factor restrains the passive mode sensor from distinguishing low concentrations of 1%–4% water in ethanol, while considerably distinct profiles are achievable using the active-mode sensor. The measured passive mode sensor response is then processed using CycleGAN, a machine-learning algorithm conventionally used for image-to-image translation. This strongly enhances the quality factor of the responses, effectively translating them to the active domain. This improvement reduces the limit of water detection down to 1% for the water-in-ethanol mixture. In addition, the sensor is used for noninvasive monitoring of glucose levels, in both passive and active modes. The resolution of the CycleGAN-boosted response approaches that of the active sensor (
$\approx 20$
mg/dL), showing a considerable enhancement when compared to the resolution of the passive sensor (
$\approx 70$
mg/dL).