Combined CS and DL techniques for DOA with a Rotman Lens

Rotman lenses are useful devices commonly utilized within multi-beam antenna array networks. They are generally used in radar surveillance systems to detect targets in multiple directions simultaneously without physically moving the antenna front-end. Nowadays, the communications sector (5G) also has great interest in this technology. Due to the use of a free-space true-time delay network, for instance attached to an Uniform Linear Array (ULA) consisting of broadband Vivaldi antenna elements, this type of microwave lens support wide-band operation with low-phase error estimation, and wide-angle scanning combined with simultaneous spatial beams (beamspace) for fast coverage. In particular the multi-beam feature makes the lens very attractive for Direction of Arrival (DoA) applications. This paper combines the aforementioned advantage with a dedicated Neural Network (NN) for an efficient wideband Direction of Arrival (DoA) and frequency estimation technique based on a single snapshot from such a multi-beam antenna configuration. The proposed approach uses machine learning techniques to establish the NN with a training set obtained from measurements in an anechoic chamber enriched superimposing different noise levels. 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.