Planar Transmission Lines of Atypical Cross-Section: Shape Optimization Exploring Low Accuracy Neural Networks

In the paper, optimization of the shape and of the size of the shielded microstrip line is described. Reaching prescribed dispersion characteristics is the goal of the optimization. CPU-time demands of the optimization procedure are minimized using efficiently trained neural networks. Genetic algorithms estimate regions of the optimization space, which might contain a global minimum. Detailed investigation of the suspicious regions is performed by Newton’s method.