Detail publikace

Broadband Design of Planar Transmission Lines: Feed-Forward Neural Approach Versus Recurrent One

Originální název

Broadband Design of Planar Transmission Lines: Feed-Forward Neural Approach Versus Recurrent One

Anglický název

Broadband Design of Planar Transmission Lines: Feed-Forward Neural Approach Versus Recurrent One

Jazyk

en

Originální abstrakt

In the paper, exploitation of recurrent neural networks for broadband modeling of EM structures is discussed. As a representative of EM structures, a shielded microstrip transmission line in layered media is elected. The structure is numerically modeled by finite-element method in the frequency range from 10 GHz to 80 GHz. The numerical model acts as a teacher for a neural network representing the behavior of the structure. For modeling purposes, both the feed-forward neural networks (a static mapping of an input pattern to an output target) and the recurrent neural networks (a mapping of an input pattern to the series of output targets) are exploited. Both the feed-forward neural models and the recurrent ones are in detail compared from the point of view of accuracy, approximation abilities, CPU-time demands and elaborateness of the development. The winner of the comparison is used for a broadband genetic optimization of the structure in order to demonstrate computational efficiency of the neural model.

Anglický abstrakt

In the paper, exploitation of recurrent neural networks for broadband modeling of EM structures is discussed. As a representative of EM structures, a shielded microstrip transmission line in layered media is elected. The structure is numerically modeled by finite-element method in the frequency range from 10 GHz to 80 GHz. The numerical model acts as a teacher for a neural network representing the behavior of the structure. For modeling purposes, both the feed-forward neural networks (a static mapping of an input pattern to an output target) and the recurrent neural networks (a mapping of an input pattern to the series of output targets) are exploited. Both the feed-forward neural models and the recurrent ones are in detail compared from the point of view of accuracy, approximation abilities, CPU-time demands and elaborateness of the development. The winner of the comparison is used for a broadband genetic optimization of the structure in order to demonstrate computational efficiency of the neural model.

BibTex


@inproceedings{BUT8194,
  author="Zbyněk {Raida}",
  title="Broadband Design of Planar Transmission Lines: Feed-Forward Neural Approach Versus Recurrent One",
  annote="In the paper, exploitation of recurrent neural networks for broadband modeling of EM structures is discussed. As a representative of EM structures, a shielded microstrip transmission line in layered media is elected. The structure is numerically modeled by finite-element method in the frequency range from 10 GHz to 80 GHz. The numerical model acts as a teacher for a neural network representing the behavior of the structure. For modeling purposes, both the feed-forward neural networks (a static mapping of an input pattern to an output target) and the recurrent neural networks (a mapping of an input pattern to the series of output targets) are exploited. Both the feed-forward neural models and the recurrent ones are in detail compared from the point of view of accuracy, approximation abilities, CPU-time demands and elaborateness of the development. The winner of the comparison is used for a broadband genetic optimization of the structure in order to demonstrate computational efficiency of the neural model.",
  address="Polytecnico di Torino",
  booktitle="Proceedings of the International Conference on Electromagnetics in Advanced Applications",
  chapter="8194",
  institution="Polytecnico di Torino",
  year="2003",
  month="september",
  pages="155",
  publisher="Polytecnico di Torino",
  type="conference paper"
}