Volterra kernels extraction from neural networks for amplifier behavioral modeling

Abstract

In wireless communication systems the amplifier non-linear distortion problems are extremely challenging. The linearization techniques based on behavioral models of amplifiers, seem to be very promising, therefore developing a suitable non-linear model is of the crucial importance. A rigorous approach for non-linear modeling is using the of Volterra series, however the calculation of the Volterra coefficients is a complex and time-consuming task. In this paper, an easy and advanced approach for extraction of the Volterra kernels will be presented. The third order Volterra kernels are derived from the parameters of a feed-forward time delay neural network with a suitable activation function.