Implementation of an MLP-based DOA system using a reduced number of MM-wave antenna elements

It is required to know the direction of arrival (DOA) of a signal in many applications, such as car tracking or reception optimization for satellite antenna. However, several reflections of different intensities highly affect the sensor outputs and the estimation quality. The system presented here is divided into three parts: a pair of three-element antenna arrays which receives the main beam and the reflected one, a radio frequency combiner which generates power signals and multilayer perceptron neural networks used to invert the mapping between the DOA space and the combiner output space. Simulations are carried out including a model of a simple reflection over a road. They are validated by experimental dataset provided by real antenna array outputs coming from tests using an asphalt reflector in an anechoic chamber.