Efficient Pattern Synthesis Method for Series-Fed Arrays Using Generative Adversarial Networks

In this paper, a physics-informed generative adversarial network (GAN) framework is proposed for computationally efficient radiation pattern synthesis of series-fed linear arrays. In the proposed framework, the generator is a fully- connected neural network designed to predict the array excitation. Contrary to conventional GAN architectures, the proposed discriminator evaluates pattern matching probability through rigorous electromagnetic field calculation rather than neural network approximation. Adversarial training maximizes the probabilistic similarity metric between synthesized and target radiation patterns, enabling rapid convergence to optimal excitation. The efficiency and effectiveness of the proposed framework are validated through simulation results.