Adaptive DOA estimation using a radial basis function network

DOA (direction-of-arrival) estimation problems have considerable importance in the fields of radar, sonar, and high-resolution spectral analysis. Many methods have been proposed for solution of DOA estimation by using array antennas. In general, the computational complexity is significant. In the present paper, adaptive estimation of the direction of arrival is proposed by using the training network called the RBF (Radial Basis Function). In this method, the problem is treated as one of deriving the mapping from the autocorrelation matrix to the angle of arrival and then the DOA estimation problem is solved. Specifically, the angle of arrival and the signal power are independently discretized. The autocorrelation matrix corresponding to these discretized data is used as the training data for the RBF network. In this processing, the basis functions are reduced by means of clustering. Also, by sensitivity analysis, a network that is robust to the estimation error of the autocorrelation matrix is constructed. In this way, the DOA can be estimated at high speed. (c) 2005 Wiley Periodicals, Inc.