Signal Processing Using Modular Struc- Tured Neural Network for Non-stationary and Nonlinear Acoustic Systems

Paying attention to realistic systems in the actual engineering fields, we must very often treat their systems as stochastic systems with non-Gaussian, nonlinear and/or non-stationary properties. In this paper, a regression analysis method for such stochastic systems is proposed by introducing reasonably a modular structured neural network. The proposed modular structured neural network is constructed by the hierarchical combination of each expert neural network for analyzing the regression relationship between input and output signals in each local stationary section, and a neural network for the prediction of weights contained in the above expert neural network. The effectiveness of the proposed method is experimentally confirmed by applying it to the simulation and actual road traffic noise data.