A Deconvolution Method for Impulsive Signals using Neural-Networks

Abstract

This work describes a proposal method for impulsive signal deconvolution. The method uses a simple one-layer neural-network configured for implementing an adaptive filter that models the unknown inverse impulse response of the propagation media. The Levenberg-Marquardt (LM) algorithm is used for finding the coefficients of the neural-network. The method has been tested on simulated channels, and on a mathematical model from a real cantilever beam. Results show that with proper training deconvolution can be achieved