Impact of Varying Neurons and Hidden Layers in Neural Network Architecture for a Time Frequency Application

Abstract

In this paper, an experimental investigation is presented, to know the effect of varying the number of neurons and hidden layers in feed forward back propagation neural network architecture, for a time frequency application. Varying the number of neurons and hidden layers has been found to greatly affect the performance of neural network (NN), trained via various blurry spectrograms as input over highly concentrated time frequency distributions (TFDs) as targets, of the same signals. Number of neurons and hidden layers are varied during training and the impact is observed over test spectrograms of unknown multi component signals. Entropy and mean square error (MSE) is the decision criteria for the most optimum solution.