Non-linear adaptive techniques for DOA estimation-a comparative analysis

Linear model based beamforming techniques (e.g. MUSIC, MLM, MVDR, etc.) have been widely used for direction-of-arrival (DOA) estimation which, in terms of statistics, only make use of the first and second order moment information (e.g. the mean and the variance) of the data. In these techniques, the higher order statistics (3rd and 4th order “cumulants”) that provide the information regarding deviation from Gaussianity and presence of phase relations of a signal have been discarded. In the sequel, the performance of these techniques is limited. Recently, artificial neural network techniques based on non-linear function and also independent of signal model have been proposed in the literature. A comparative analysis is carried out in this paper for a high resolution MLM and three ANN techniques. The Hopfield neural network, backpropagation neural network and radial basis function networks are described. Computer simulation results have demonstrated that nonlinear adaptive (ANN) techniques have more superior performance