An implementation of a direction-finding antenna for mobile communicationsusing a neural network

Direction-finding systems for radio signals are mostly used in mobile commu nications and avionics applications for antenna tracking or navigation purp oses. In general, such systems require accurate calibration and may be sens itive to noise and external interference. In this paper, we will investigat e the performance of a neural network-based direction-finding system under such conditions. The proposed topology is a hybrid one, combining a simple RF signal beamformer with a neural network. The training of the neural netw ork is accomplished experimentally with a three-element antenna array by va rying the beam's direction and the carrier frequency. The error on the esti mated direction of arrival caused by the environment and training limitatio ns are investigated.