Occupancy Grid Mapping for Personal Radar Applications

Next fifth generation (5G) of mobile wireless communication foresees the use of mm-wave technology to boost communication at an unprecedented scale, thanks to the large available bandwidth [1] . In addition, the move-up in the frequency spectrum allows to include a large number of antennas into a small area, thus enabling their integration into portable devices [2 , 3] . In this way, such a technological perspective can be exploited to add new functionalities in addition to communication. For example, the laser-like beamsteering allowed by massive arrays at mm-wave can be used to automatically scan and reconstruct the topology of the surrounding environment. Such an idea, namely personal radar, has been recently proposed in theory and its feasibility assessed by experiments [4 – [5]6] . In these works, the performance has been investigated through the adoption of a grid-based mapping approach relying on an extended Kalman-Filter (EKF): the environment has been discretized in a grid of cells whose root-radar cross section (RCS) values constitute the state vector to be estimated starting from the backscattered radar response [7 , 8] . To simplify the analysis, the state of the system has been modeled as a Gaussian random vector whose mean vector and covariance matrix are updated during the mapping process as soon as new measurements are collected [5 , 9] . The main limitation of this model is that the Gaussian assumption does not capture the underlaying bimodal nature of the phenomenon, i.e., each cell is empty or occupied. In laser-based mapping systems, occupancy grid (OG) methods are usually considered to model this bi-modality by exploiting the basic assumption that laser beam illuminates only one cell per time [10] . This is not the case in radio-based radars where the shape of the radiation pattern is such to illuminate an area composed of several cells, thus making existing OG methods not appropriate due to the inherent cross-correlation between cells that is not zero [11] .