Multipath Ghost Classification for MIMO Radar Using Deep Neural Networks

Multipath is a significant challenge for indoor multiple-input-multiple-output (MIMO) radar applications. It generates the so-called 'ghosts' in the radar detection, which represent the objects that do not exist. Targets and ghosts are very similar, which makes them difficult to be recognized without prior knowledge of the environment geometry. In this work, a multipath model for the indoor scenario is analyzed for a frequency-modulated continuous-wave (FMCW) MIMO radar. Based on the multipath model, spatial signals from the MIMO virtual channels are fed to a deep neural network that is proposed to classify the multipath ghost, combined with a linear pattern recognition algorithm from our previous work. Simulation and experimental results demonstrate the performance of the proposed solution.