The performance of neural network for high resolution direction-of-arrival estimation

Abstract

Neural network (NN) method for DOA estimation, recently suggested by Y.L. Ma et al. (1990), is a novel approach. It circumvents problems encountered in DOA estimation efficiently, especially for fully correlated signals. Asymptotically the algorithm has been shown to yield unbiased estimates (the standard deviation of the estimate error approaches the CR bound). This paper gives an analysis of the performance of the NN method, comparing it with eigenstructure algorithm (MUSIC). The article introduces certain measures to characterise its performance. Results of computer simulations are also presented.<>