Narrowband direction finding using complex EKF trained multilayered neural networks

A technique using a multilayered neural network has been developed for the narrowband direction finding problem that involves array processing of non-Gaussian signals. The complex extended Kalman filtering algorithm is derived for training the networks with complex input signals. Two networks were implemented, one with third order cumulants and the other with traditional correlations of the received signal vector evaluated at different combinations of directions of arrival (DOAs) as training inputs. Simulation results show that the network trained with cumulants outperforms the network trained with the correlations