DOA Estimation of Wideband Acoustic Target Based on Group FNN

Aiming at the estimation of the direction of arrival (DOA) of underwater sound sources, this paper proposes a grouped fully connected structure and a loss function based on feed-forward neural networks, decoupling the information between each frequency point so that they do not affect each other. The network is composed of a 4-layer feed - forward neural network with grouped fully connected layers. Through comparative analysis of simulation experiments and the SWellEx-96 experiment, the proposed structure in this paper can improve the environmental adaptability and antiinterference ability of the model compared with the coherent network structure.