An Estimation Method of Magnetic Coupling Coefficient Between Two Microstrip Lines Using Machine Learning of Near-Field Information

A novel estimation method of magnetic coupling coefficients between printed-circuit-board-level traces using a near-field information was investigated. Parallel two microstrip lines (MSLs) with different distances between the lines were used as a test bench. The current flowing in a signal line and its return current were modeled as a simple one-turn equivalent loop current model with uniform current distribution. First, a 1-D convolutional neural network (CNN) for regression prediction was trained with the theoretical values of the magnetic near-field distribution generated from the loop current model. Next, the measured magnetic near-field distributions above the parallel two MSLs at 1 GHz were input to the trained CNN to estimate the geometry of the loop current models. The magnetic coupling coefficient between two MSLs is estimated through calculating the coupled magnetic flux between the estimated loop current models. The magnetic coupling coefficients between the loop current models estimated by measured magnetic near-field distribution agreed with the coupling coefficients calculated by the full-wave finite element method (FEM) simulation within 10%, which indicates the feasibility of estimating the magnetic field coupling by the proposed method.