Optimization Of Antenna Array Configuration With A Neural Network

Abstract

The accuracy of angle of arrival (AOA) estimation with an antenna array, depends on the antenna elements positions. In this paper, we introduce a novel method for optimizing the antenna elements position that minimizes the AOAs estimation error in the case of an unknown number of sources and a single array realization (snapshot). The method utilizes a deep neural network (DNN) for estimating the number of sources and their AOAs for a given antenna elements positions, and minimizes the estimation error by jointly optimizing the antenna positions and the DNN parameters. The use of the DNN estimator in this case, enables to calculate the gradient of the estimation error with respect to the antenna elements position, and thus to minimize the estimation error with respect to the antenna positions by gradient descent. The proposed approach is unique because it determines the antenna array configuration that explicitly minimize the AOAs estimation error, while other reference methods use an optimization objective that is implicitly related to the AOA estimation error. We show that the proposed optimization method attains a significant performance advantage in the RMSE of the AOAs estimation compared to reference antenna configuration optimization methods.