Element Failure Detection of Array Antenna using Near-field Measurement with Shallow Neural Network

Abstract

In this report, the element failure detection of array antenna is performed with a minimum number of measurement points while maintaining sufficient accuracy by learning the relationship between excitation coefficients of array antenna and the electric near-field distribution by a shallow neural network. When training the neural network, the massive number of training data are generally required. For increasing the training data, we use each element-fed near-field distribution multiplied by a number of random excitation coefficients. In the case of dipole array antennas, the estimation error of excitation coefficients of array antenna less than 1% are achieved by our trained neural network with a minimum number of near-field measurements.