Covariance Shaping for Massive MIMO Systems

The low-rank behavior of massive multiple-input multiple-output (MIMO) channel covariance matrices and its
exploitation for pilot decontamination and statistical beamforming are well documented. Existing algorithms, however, rely
on signal subspace separation among user equipments (UEs) and, as such, they tend to fail when the distance between UEs
becomes small. This paper proposes a solution to this problem via covariance shaping at the UE-side in the case where the
UEs are equipped with (a small number of) multiple antennas. The key resides in: i) exploiting general non-Kronecker MIMO
channel structures that allow the transmitter to suitably alter the channel statistics perceived by the base station, and ii) sacrificing
some spatial degrees of freedom at each UE so as to improve the statistical orthogonality between closely spaced UEs. Numerical
results illustrate the sum-rate performance gains of the proposed covariance shaping method with respect to existing ones.