Implementation of antenna array beamforming by using a novel neural network structure

Abstract

The present study introduces the implementation of antenna array beamforming based on a new neural network (NN) structure. The NN comprises two hidden layers, which use different interconnectivity patterns. The first one is divided in sublayers, which are equal in number to the inputs of the NN. Each sublayer communicates only with the respective input but is fully interconnected with the second hidden layer. The NN training is performed by using data sets derived by a well-known beamforming technique called minimum variance distortionless response. The trained NN is capable of serving as adaptive beamformer that makes a linear antenna array steer the main lobe towards a desired signal and place nulls towards respective interference signals in the presence of additive zero-mean Gaussian noise. The performance of the trained NN is tested by estimating the mean absolute deviation of main lobe and null directions from their respective desired directions.