Cooperation of multi-layer perceptrons for angles of arrival estimation

A novel method for estimating the Angles Of Arrival (AOA) of multiple plane waves with a narrow-band array of sensors is presented here. This method is based on the cooperation of two Multi-Layer Perceptrons (MLP). The elements of the covariance matrix of the signals are input to the first MLP (a classical three-layer perceptron). Then, taking profit of the gross estimations provided by this first neural net, we refine them by using a new kind of perceptron: the Constrained Structured Weights (CSW) perceptron, which we demonstrate finds the Maximum Likelihood Estimation (MLE) of the AOA. Unlike the classical MLE approach, this method presents the great advantage of being computationally inexpensive. Finally some results and comparisons are presented.