Development of Near-Field Source Localization Method using Convolutional Neural Network

In EMI (Electro-Magnetic-Interference) countermeasures, it is important to accurately localize an interference source. The purpose of this research is to introduce Convolutional Neural Network (CNN), which is one of machine learning methods, to estimate source location. The proposed CNN which estimates the source location on a two dimensional plane by using the field strength of the observation surface was described. First, in order to investigate the influence of hyperparameter on estimation accuracy, source location estimation by numerical experiments was performed by changing the parameters. Next, we examined for further improvement of precision by introducing multi-condition learning. We evaluated from the viewpoint of estimation error based on mean square error and robustness against SNR. As a result, location estimation with better accuracy than the existing method was achieved, demonstrating feasibility of source localization by convolution neural network.