An Efficient Antenna Beampattern Synthesis Method Based on a Knowledge-Driven Graph Convolutional Network

Due to the mutual influence of the amplitude and phase distribution of the elements in series-fed microstrip antennas (SFMAs), the complexity of its radiation pattern synthesis increases. To address this issue, a knowledge-driven graph convolutional network (GCN) framework is proposed. By projecting the physical topology of the series-fed antenna array into a graph, the relationship between the amplitude and phase of the array elements is directly embedded into the update mechanism of the graph node features. Additionally, an attention mechanism is introduced based on the correlation between the radiation pattern error and the array excitation, enabling the synthesis process to focus on the main sources of error. Numerical results show that the proposed framework achieves high efficiency and accuracy in pattern synthesis of SFMAs.