A Novel Utilization of NARX for Antenna Array Adaptive Beamforming

In this paper, we investigate the use of a nonlinear autoregressive network with exogenous inputs (NARX) for adaptive beamforming on smart antennas. As a beamformer, NARX receives the angles of arrival of incoming signals to extract the complex feeding weights that produce the appropriate antenna radiation pattern. In order to demonstrate the potential of such an implementation, we test our model on a realistic linear antenna array composed of 16 microstrip elements. We use the null steering beamforming technique to produce the datasets needed for training and testing of our model and then we evaluate the accuracy of the radiation patterns produced by this model. To further demonstrate the efficiency of the NARX implementation, we also make a comparison with a feed-forward neural network that has the same architecture with that of NARX.