Wave-Phase-Constrained Neural Network for Electromagnetic Scattering Prediction

Full-wave numerical methods can provide high-fidelity radar cross section (RCS) predictions for complex targets, but their high computational cost limits their applicability in real-time or near-real-time scenarios. This paper proposes a phase-constrained neural network framework that explicitly incorporates wave-propagation phase constraints into a deep learning model, thereby improving both the prediction accuracy and physical consistency of the model. The core innovation of the framework is a decoupled dual-branch decoder, which separately predicts the amplitude response and intrinsic phase of equivalent scattering aggregation points. By further introducing round-trip propagation phase delays corresponding to the observation direction and embedding a physics-guided complex-domain coherent summation operator, the model accurately reconstructs the coherent interference of the scattered field. Validation on cylindrical and aircraft targets demonstrates that the proposed network achieves high RCS prediction accuracy while reducing inference time to the millisecond level, offering an effective solution for real-time evaluation of complex target scattering.