A neural network approach to source localization

Abstract

The use of neural network techniques to localize an acoustic point source in a homogeneous medium is demonstrated. The input data are the cosines of the phase difference measurements at an array with N detectors. Only the most fundamental types of neural network systems will be considered. Use will be made of linear and sigmoid-type neurons in a single-layer network. The performance of the single-layer network is very satisfactory for a wide range of configuration parameters if the resolution and sampling conditions are satisfied. Once the parameters of the neural network are determined, the computational effort to determine a new source location is minimal. However, when a source/detector configuration is considered that does not satisfy the resolution and sampling conditions, the single-layer network will not consistently perform well.