Angle-of-Arrival Sensing Using a Machine Learning Enhanced Amplitude-Only System

A low-cost, low-complexity amplitude-only angle-of-arrival (AoA) sensor system is experimentally demonstrated. The sensor embeds a machine learning (ML)-in-the-loop procedure as the processing algorithm to extract both azimuth and elevation AoA information from a planar circular array. This has not been demonstrated in the open literature with the traditional AoA approaches, the same sensor’s dimensionality, field-of-view, or with as low system complexity. A four-channel receiver is designed, built, and assembled behind a simple four-element monopole array. The sensing components and ML algorithm are integrated to experimentally validate the hypothesis that both azimuth and elevation information can be determined from this simple, low-cost sensor. Moreover, a systematic procedure is introduced for the design of a deep neural network for the amplitude-only direction-finding application. Following this approach, major system complexity reductions are found—over 99% for the azimuth network and 98% for the elevation network, while still achieving 0.75° root mean square error (RMSE) in azimuth and 1.1° RMSE in elevation. Experimental results show up to 20% accuracy improvement in azimuth estimation, and 93% accuracy improvement in elevation, when compared to a lookup table-based method.