Microwave-Based Parameter Reconstruction of Inhomogeneous Plasma Using Full-Wave Simulation and Deep Learning

This paper presents a novel microwave-based parameter reconstruction method integrating full-wave simulation and deep learning inversion for diagnosing inhomogeneous plasmas. A computationally efficient parametric model for three-dimensional non-uniform plasmas is developed, ensuring accuracy and reduced computational complexity. The Finite Element-Boundary Integral-Multilevel Fast Multipole Algorithm (FE-BI-MLFMA) is utilized to generate high-fidelity datasets. A Plasma-Inversion Network (PINet) is proposed, reconstructing plasma electron density by converting antenna reflection coefficients into image-like representations. This approach facilitates robust feature extraction, improves inversion accuracy. Ground-based experiments are conducted with a cascaded arc plasma source, producing large-scale, high-density, stable plasma sheaths. Probe-based comparative analysis confirms the accuracy of the proposed diagnostic method. Experimental results demonstrate the method’s robustness across various plasma distributions, highlighting its potential for real-time in-situ diagnostics under hypersonic flight conditions. The proposed method significantly enhances understanding of the electromagnetic environment around hypersonic vehicles and supports crucial communication and detection systems.