Neural network based model for radiated emissions prediction from high speed PCB traces

Abstract

Printed Circuit Board (PCB) traces are one of most important PCB Radiated Emissions (RE) sources. These traces is becoming electrically long as the trace length is comparable with the wavelength resulting in higher RE. Therefore, it is essential to predict the RE to avoid out of compliance test issues. In this paper, a neural network Multi-Layer Percetron (MLP) model is developed to predict the radiated emissions of PCB traces. The MLP model is then trained and tested using data set generated based recently developed closed-form equations. Results had shown that a good estimate of the radiated emissions can be obtained using this developed model avoiding both the time-consuming simulations and expensive prototype testing in the compliance chambers. Double-layer PCB is fabricated to validate the proposed neural network model by measurement in a Semi Anechoic Chamber (SAC). Moreover, reasonable agreements are obtained between the measurement and proposed model results.