Enhancing the Estimation of mm-Wave Large Array Channels by Exploiting Spatio-Temporal Correlation and Sparse Scattering

In order to cope with the large path-loss exponent of mm-Wave channels, a high beamforming gain is needed. This can be achieved with small hardware complexity and high hardware power efficiency by Hybrid Digital-Analog (HDA) beamforming, where a very large number $M\gg 1$ of antenna array elements requires only a relatively small $m\ll M$ number of A/D converters and modulators/demodulators. As such, the estimation of mm-Wave MIMO channels must deal with two specific problems: 1) high Doppler, due to the large carrier frequency; 2) impossibility of observing directly the M-dimensional channel vector at the antenna array elements, due to the mentioned HDA implementation. In this paper, we consider a novel scheme inspired by recent results on gridless multiple measurement vectors problem in compressed sensing, that is able to exploit the inherent mm-Wave channel sparsity in the angular domain in order to cope with both the above problems simultaneously. Our scheme uses past pilot-symbol observations in a window of length $T$ in order to estimate a low-dimensional subspace that approximately contains the channel vector at the current time. This subspace information can be used directly, in order to separate users in the spatial domain, or indirectly, in order to improve the estimate of the user channel vector from the current pilot-symbol observation.