Accurate Characterization of Graphene Reconfigurable Reflectarray Antenna Element by SVR

Optimal design of transmitarray antennas via low-cost surrogate modelling

“…At this point, it should be emphasized that although SVRM and other benchmark surrogates have been demonstrated successful in modeling of different microwave components 19 , 24 , 40 , 44 , the results presented above suggest otherwise in the context of transmitarrays. The main reason for this is that performance of SVRM and other benchmark surrogate modeling techniques is contingent upon the variation ranges of the input parameters (which are normally set rather narrow), and the total number of samples (270 + 100) to generate a globally accurate surrogate model 5 , 45 .…”

“…Fast surrogates can be used to determine-at negligible cost-the geometry and material parameters of the elements that ensure the required transmission phase and magnitude characteristics, as well as the distance between the unit element and the feed antenna 16,17 . The literature offers the abundance of surrogate-assisted methods, oriented towards improving both the computational efficiency and reliability of the optimization processes [18][19][20][21][22][23][24] . Still, data-driven surrogate modelling is a challenging problem in the context of high-frequency design, primarily due to the high initial cost related to the acquisition of the training and testing data sets.…”

“…To work around direct handling of expensive EM models, surrogate-assisted optimization has been studied by many researches over the last decades. The literature offers a number of algorithmic tools in the context of reflectarray and transmitarray design (e.g., [22][23][24] ). Notwithstanding, the high initial cost of surrogate model construction, pertinent to the acquisition of the training and testing data samples (typically, 4000 samples or beyond [25][26][27] ) is still a major practical issue.…”

“…"Low-cost surrogate modeling of unit elements".C, an accurate surrogate model of TA unit cell is obtained only using 370 data points, obtained using a linear sampling technique with the step sizes of 6, 3, 15, for ε r , W, H respectively. For the sake of benchmarking, the proposed M2LP modelling approach is compared to the methods commonly used in the context of surrogate modelling of similar microwave antenna designs: (i) Support Vector Regression 19,24,40 , (ii) Ensemble Learning 41 , and (iii) Generalized Regression Neural Network (GRNN) 42 . The hyper-parameter configuration of the considered methods are presented in Table 2.…”