MIDAR: Massive MIMO Based Detection and Ranging

Massive multiple-input multiple-output (MIMO) can increase the spectral efficiency. Besides, it can provide accurate localization. In some applications, e.g. autonomous driving, user behavior such as velocity and moving direction needs to be detected in addition to the position. In this paper, we propose Massive MIMO based Detection and Ranging (MIDAR) to offer a solution for joint localization and behavior recognition based on the power spectrum in multiple domains. An angle-delay-Doppler power spectrum (ADD-PS) is extracted from a mass of channel state information (CSI) as the fingerprint of a particular position with a certain behavior. By matching this fingerprint to a big data set of pre-collected reference points, we can obtain improved localization performance and detect the user behavior at the same time. In order to take full advantage of the geometrical structure of the multi-dimensional ADD-PS, algorithms in tensor analysis are considered and a chordal distance based kernel (CDBK) method is exploited for the large-scale fingerprint matching. Numerical results corroborate the feasibility of MIDAR for joint localization and behavior recognition and demonstrate that the CDBK approach outperforms conventional matching schemes for massive MIMO based localization.