Hybrid Data- and Physics-Informed Electromagnetic Inversion for Nonuniform Plasma Diagnostics Using Antenna Reflection Coefficients

This paper presents a hybrid data- and physics-informed network (HDPNet) for the electromagnetic inversion of nonuniform plasma parameters using antenna reflection coefficients. The network integrates an antenna decoupling network (ADNet) to transform measured antenna reflection cofficients into equivalent plane-wave reflection coefficients, thereby mitigating the modeling challenges associated with complex antenna structures. Meanwhile, an initial estimation network (IENet), trained on the same antenna reflection cofficients inputs, provides preliminary estimates of plasma parameters. The outputs of ADNet and IENet, serving as physical constraints and initialization, are jointly incorporated into a data-enhanced physics-informed neural network (DE-PINN). To address the performance degradation of PINN in high-frequency regimes, a composite enhancement strategy—combining phase factor separation with multi-scale coordinate mapping—is proposed to improve the convergence and accuracy. The framework is validated through both full-wave simulations and ground-based experiments under uniform and nonuniform plasma conditions. Results demonstrate that HDPNet enables accurate, robust, and physically consistent plasma diagnostics, offering the solution for electromagnetic sensing in complex scenarios such as hypersonic flight.