Realizations of Space Mapping based neuromodels of microwave components

Cargando...
Miniatura

Fecha

2000-04

Autores

Bandler, John W.
Rayas-Sánchez, José E.
Zhang, Qi J.
Wang, F.

Título de la revista

ISSN de la revista

Título del volumen

Editor

AP2000 Millennium Conf. on Antennas & Propagation

Resumen

Descripción

Artificial Neural Networks (ANN) are suitable in modeling high-dimensional and highly nonlinear elements, such as those found in the microwave arena. In modeling microwave components, the learning data is obtained from a detailed or “fine” model (typically an EM simulator), which is accurate but slow to evaluate. This is aggravated because simulations are needed for many combinations of input parameter values. This is the main drawback of conventional ANN modeling. We use available equivalent circuits or “coarse” models to overcome this limitation. In the Space Mapping (SM) based neuromodeling techniques an ANN is used to implement a suitable mapping from the fine to the coarse input space. The implicit knowledge in the coarse model not only allows us to decrease significantly the number of learning points needed, but also to reduce the complexity of the ANN and to improve the generalization performance. We present novel realizations of SM based neuromodels of practical passive components using commercial software. An SM-based neuromodel of a microstrip right angle bend is developed using NeuroModeler, and entered into HP ADS as a library component through an ADS plug-in module.

Palabras clave

Microwave Components, Neural Modeling, Space Mapping, CAD, RF, Artificial Neural Networks (ANN)

Citación

Bandler, J.W; Rayas-Sánchez, J.E; Wang, F. and Zhang, Q.J. (2000) “Realizations of Space Mapping based neuromodels of microwave components,” in AP2000 Millennium Conf. on Antennas & Propagation, Davos, Switzerland, Apr., vol. 1, pp. 460.