Estimation of varying frequency by Gabor filters and neural network

Abstract

A method of varying frequency estimation using Gabor filter bank and neural network is proposed. This method consists of two phases. First phase is the feature extraction step which decomposes a given input signal into each frequency components using Gabor filter bank. Then such components are treated as features in the frequency domain. Second phase is the estimation step which calculates instantaneous frequency from the first phase outputs using neural network. Neural network has the ability to estimate instantaneous frequency not only against artificial signal, but also against added noise signal. The aim of the proposed method is to estimate the varying frequency of non-stationary 1-D and 2-D (real) signal, where local frequency is assumed to vary smoothly.