Angle of arrival estimator based on artificial neural networks

This paper presents the approach to the design of angle of arrival estimator for narrow-band noise-like signal based on artificial neural network (ANN). The multilayer perceptron type ANNs are trained to minimize the sum of squared errors or maximize the likelihood function using the deterministic approach with the data samples generated by the single station model. The special type of output neuron processing unit is developed to perform sensible angle estimation by the signals obtained from previous network level. The results of numerical simulation show the significant increase in the estimation procedure speed for trained ANN in comparison with direct maximum likelihood estimator. The cost of the performance boost is accuracy deterioration which is no more than 10 percent at moderate SNR values.