Eye Diagram Contour Modeling Using Multilayer Perceptron Neural Networks With Adaptive Sampling and Feature Selection

Abstract

This article presents a methodology for the modeling of high-speed systems using machine learning methods. A multilayer perceptron neural network is used to map the input–output characteristics from the design parameters to the contours of the eye diagram. In addition, an improved adaptive sampling method is applied for the effective exploration of the design space, and feature selection techniques along with self-organizing maps are used to reduce the problem dimension size. Numerical examples indicate that the proposed method is able to capture the shape and magnitude of the eye contours accurately, and the iterative nature of the algorithm allows a control to balance between accuracy and model generation time. Since well-trained neural networks are able to produce subsequent results almost instantaneously, this modeling approach would be an attractive alternative compared with traditional simulation processes involving complex electromagnetic analyses and long transient simulations.