Single Snapshot DoA Estimation from a Rotman Lens using Machine Learning Techniques

In this paper an efficient wideband Direction of Arrival (DoA) and frequency estimation technique is presented which uses only a single snapshot from a multi-beam antenna array. The simultaneous spatial beams (beamspace) are created by a Rotman lens which is composed of a free-space true time delay network attached to a Uniform Linear Array (ULA) of broadband Vivaldi antenna elements. The proposed technique uses machine learning techniques to establish an optimal Neural Network (NN) configuration obtained with a training set. To improve the spatial and frequency space resolution a further estimation stage follows to the NN topology. This results in a lower computational load during the training phase and finally a very fast estimation of direction and frequency of the impinging signal. The performance is evaluated by measurements obtained in an anechoic chamber.