چكيده به لاتين
There is a pervasive need for precise material characterization, especially permittivity, in numerous engineering applications such as printed circuit board technologies, agriculture, food industry, material science, and medical treatments. A wide spectrum of materials ranging from biological tissues and chemical materials to human-made composites are dispersive, meaning that their electromagnetic properties do not remain constant over a wide frequency range. This variation makes the characterization more challenging in general compared to non-dispersive materials.
This thesis presents several inexpensive transmission line approaches to reconstruct dielectric properties of dispersive and non-dispersive materials using only power measurements, thereby removing the need for phase measurements. The proposed methods only require a signal generator and a power sensor for complex permittivity characterization without the need for either a scalar or a vector network analyzer.
First, a dielectric characterization algorithm based on transmitted power measurements is presented for characterizing non-dispersive materials. The method can retrieve the dielectric properties of materials utilizing multiobjective functions enforced at multiple frequencies and lengths without the need for phase information which requires expensive measurement setups. However, the shortcoming of this approach is its need for multiple measurements at different lengths of MUT, which makes the reconstruction process cumbersome.
Second, a semi-analytical method and a machine learning-based approach using the Bayesian regularization algorithm (BRA) are proposed to determine the complex permittivity. In both methods, a Debye model for the frequency dispersion of the MUT is used, however, the applicability extends to other models. In the semi-analytical method, the coaxial line is considered as an ideal transmission line, and by inverse solution of the forward scattering equation in different frequencies which requires numerically solving a system of nonlinear equations, the Debye parameters are extracted. The machine learning technique relies on training a multilayer artificial neural network which utilizes the full-wave simulation results of the coaxial line loaded with different MUTs as its training set. For experimental validation, a suspended coaxial line was designed and fabricated. The complex permittivities of several liquid chemicals were measured within a broad frequency band of 0.3-3 GHz. The comparison between the retrieved permittivity and the reference data demonstrates high accuracy for both methods; however, the artificial neural network approach shows higher precision.
Third, a machine learning-based approach using the BRA is proposed to determine the complex permittivity, directly, without the need for using a dispersion model. This optimal neural network can effectively retrieve the complex permittivity of several dispersive chemical materials and their mixtures with different volume ratios, within the band of 0.3-3 GHz. The retrieved results were compared with reference data (for pure materials) or those obtained by mixture formulas, demonstrating the high accuracy of the proposed method.
