Space mapping based neuromodeling

Abstract

A powerful concept in neuromodeling of microwave circuits based on Space Mapping technology is described. The ability of Artificial Neural Networks (ANN) to model high-dimensional and highly nonlinear problems is exploited in the implementation of the Space Mapping concept. By taking advantage of the vast set of empirical models already available for many microwave structures, Space Mapping based neuromodels decrease the number of EM simulations for training, improve the generalization and extrapolation performance and reduce the complexity of the ANN topology with respect to the conventional neuromodeling approach. Five innovative techniques are proposed to create Space Mapping based neuromodels for microwave circuits: Space Mapped Neuromodeling (SMN), Frequency-Dependent Space Mapped Neuromodeling (FDSMN), Frequency Space Mapped Neuromodeling (FSMN), Frequency Mapped Neuromodeling (FMN) and Frequency Partial-Space Mapped Neuromodeling (FPSM). Excepting SMN, all these approaches establish a frequency-sensitive neuromapping to expand the frequency region of accuracy of the empirical models already available for microwave components that were developed using quasi-static analysis. We contrast our approach with the conventional neuromodeling approach employed in the microwave arena, as well as with other state-of-the-art neuromodeling techniques. We use Huber optimization to efficiently train the simple ANN that implements the mapping in our SM-based neuromodels. The five space mapping based neuromodeling techniques are illustrated by two case studies: a microstrip right angle bend and a high-temperature superconducting (HTS) quarter-wave parallel coupled-line microstrip filter.