비정형 형상 유전체의 도파관 방식 X-Band 유전율 측정을 위한 딥러닝 모델 연구

다국어 초록 (Multilingual Abstract)

This study proposes a deep-learning model for estimating the permittivity of irregularly shaped dielectric materials employing X-band (8～12 ㎓) waveguide measurements. The proposed convolutional neural network (CNN) extends the conventional Nicolson-Ross-Weir (NRW) method to deliver accurate permittivity values, even when the dielectric samples deviate from the waveguide geometry. The model was trained on S-parameter data acquired under non-ideal conditions, including air gaps between the waveguide and sample as wells as oversized samples that exceed the waveguide aperture dimensions. To address the limitations of the NRW method with respect to thickness variations, the model incorporated S-parameter data with varying sample thicknesses. Performance validation utilizing FR-4 substrates demonstrated that the CNN model achieved highly accurate permittivity measurements with error rates within 1.2 % for samples with air gaps of up to 5 % and within 5 % error rates for oversized samples. These findings confirm that the proposed approach offers a robust and reliable solution for permittivity measurement of irregularly shaped dielectric materials in practical applications.