DeepMeasurement: Complex Permittivity Characterization of Liquid Crystals Using Transfer Learning

Liquid crystal (LC) is a kind of anisotropic material with tunable permittivity, giving it significant application in the microwave field. Therefore, accurate characterization of the complex permittivity of LC materials at microwave frequency is of great importance for LC-based tunable microwave devices. In this article, a Gaussian process regression (GPR)-based deep learning network model, named DeepMeasurement, is generated and trained using transfer learning, which can be used to characterize the complex permittivity of LCs from transmission parameters. First, to simply and efficiently obtain the transmission response of the LC material under test (MUT), a measurement setup based on artificial coaxial structure has been designed and fabricated. In addition, to train the initial network parameters of GPR-based DeepMeasurement, large data of the source domain is rapidly generated by a simple multi layer model and transmission matrix theory. Furthermore, the transfer learning is performed by small data of the target domain generated by full-wave simulation to calibrate the model parameters. The introduction of transfer learning has effectively improved the accuracy of measurement results and the efficiency of network parameter generation. Finally, one prototype is fabricated and measured, respectively. Experimental results demonstrate that the DeepMeasurement can effectively establish a mapping from transmission parameters to the complex permittivity of LC-MUT, and the relative permittivity and loss tangent values can be accurately extracted, thereby proving the effectiveness of the proposed method.