Multilayered frequency selective surface design using artificial neural networks

Summary form only given. Two design approaches for a single-layered FSS using artificial neural networks were reported. The first uses a bitmap model to represent the unit-cell geometry and the second uses a parametric representation. While the former allows more freedom in choosing the unit-cell geometry, it also requires a much larger training database and more iterations for a convergent design than the latter, which only works for a fixed geometry. Both approaches require the training and inversion of a multilayer perceptron. When extending the artificial neural network approach to the design of multilayered FSS, the parametric representation can reduce the number of variables significantly and thus reduce the size of the neural network and the number of iterations required in the network inversion. Designs of multilayered FSS using dipole and tripole patches and slots were considered.<>