An Efficient Design Methodology for Finite-Size Anomalous Reflectors Based on Equivalent Circuit Model and Artificial Neural Networks

Perfect anomalous reflectors (PARs) are widely being developed to extend signal coverage in 5G wireless communications thanks to their ability to reflect electromagnetic (EM) waves towards any desired direction. However, state-of-the-art design methodologies for PARs rely primarily on infinite-size periodic structures, resulting in non-negligible errors and suboptimal power efficiency and bistatic radar cross section (BRCS) results in practical applications where these reflectors are spatially confined and possess a limited number of periodic unit cells. This paper proposes an efficient design methodology for finite-size PARs based on the equivalent circuit model (ECM) and artificial neural network (ANN) techniques. An accurate ECM-based genetic algorithm (GA) of infinite-size PARs, derived from Floquet modal expansion, is implemented first to provide an efficient initial guess which reduces the data size and improves the accuracy of the ANN implemented for the finite-size PARs. As a final step, ANN and GA are combined to find finite-size PARs yielding optimal BRCS results. The surface wave analysis method has been adopted to decrease the data collection time, and a modified poles-residues tracking technique has been proposed to broaden the bandwidth and the geometrical variable ranges and improve ANN’s predictive accuracy. Three finite-size PAR cases with unitary efficiency for 70∘ anomalous reflection at 28GHz illuminated by normal incidence have been presented to demonstrate that our proposed design methodology can efficiently achieve the optimal results, outperforming the methodology under the infinite-size assumption. The PAR case, developed to tackle TE incidence, has been fabricated and measured to showcase the effectiveness of our proposed methodology further.