Narrowband DOA Estimation using EKF Trained Multilayered Neural Networks with Cumulants as Training Inputs

Abstract

A technique using multilayered neural network has been developed for narrowband direction finding problem that involves in array processing of non-Gaussian signals. Two networks were implemented, one with the third order cumulants and the other with the traditional correlations of received signal vector evaluated at different combinations of direction of arrivals (DOA's) as training inputs. In training these networks, the back-propagation and the Extended Kalman Filtering (EKF) based learning algorithms were used. Simulation results show that the network trained with cumulants outperforms the network trained with the correlations.

Indexing terms:

• DOA estimation
• Neural networks
• HOS
• EKF learning algorithm