Recently, the research group led by Professor Weibing Lu from the Center for Flexible Radio Frequency Technology, Southeast University, presented two papers titled Rapid Analysis of Scattering Problem for LFPSs with Physics‑informed Attention Network and Improved Hybrid Quantum Models for More Accurate Phase Prediction of Expansion Coefficients in Large Finite Periodic Structures at the 2026 IEEE MTT‑S International Microwave Workshop Series on Advanced Materials and Processes for RF and THz Applications (IMWS‑AMP).

In this work, a physics-informed attention network (PIAN) is proposed for the rapid analysis of scattering problems in large-scale finite periodic structures (LFPSs). The encoder captures the relationship between array features and global coupling with a self-attention mechanism, which guides the decoder to generate the final current coefficients of LFPSs based on the NASED initial current coefficients with a cross-attention mechanism. The introduction of initial current coefficients enables the characterization of cell structure for different cells. The well-trained network can generalize to larger-scale arrays, which significantly reduces the cost of dataset generation. Numerical experiments validate the accuracy, generalization capability, and efficiency of the proposed method.

图1 PIAN网络结构图

Another paper presented at the conference, entitled “Improved Hybrid Quantum Models for More Accurate Phase Prediction of Expansion Coefficients in Large Finite Periodic Structures,” proposes an enhanced quantum-assisted framework for efficient electromagnetic analysis of LFPSs. Accurately and efficiently analyzing the electromagnetic characteristics of large finite periodic structures (LFPSs) remains a significant challenge due to the high computational cost of conventional full-wave methods. Although the SED basis function method reduces computational complexity, efficiently constructing reduced matrices with mutual coupling remains difficult. In this work, the previously developed hybrid quantum model is improved by incorporating a residual architecture to enhance the prediction of expansion coefficient phases in SED basis functions. Numerical experiments demonstrate that the proposed model achieves faster learning capability and higher prediction accuracy, offering an effective approach for accelerating electromagnetic analysis of LFPSs.