Enforcing Statistical Orthogonality in Massive MIMO Systems via Covariance Shaping

TL;DR

This paper introduces a novel covariance shaping technique for massive MIMO systems that improves downlink data transmission by exploiting non-Kronecker channel structures and multiple antennas at UEs, especially under high spatial correlation and pilot contamination.

Contribution

It proposes a statistical beamforming method using covariance shaping at the UE side and a low-complexity algorithm for optimizing this shaping, addressing practical channel correlation issues.

Findings

Achieves significant sum-rate performance gains over traditional spatial multiplexing.

Converges to a stationary point for the two-UE case with a low-complexity algorithm.

Effective in scenarios with limited spatial separation of UEs.

Abstract

This paper tackles the problem of downlink data transmission in massive multiple-input multiple-output (MIMO) systems where user equipments (UEs) exhibit high spatial correlation and channel estimation is limited by strong pilot contamination. Signal subspace separation among UEs is, in fact, rarely realized in practice and is generally beyond the control of the network designer (as it is dictated by the physical scattering environment). In this context, we propose a novel statistical beamforming technique, referred to as MIMO covariance shaping, that exploits multiple antennas at the UEs and leverages the realistic non-Kronecker structure of massive MIMO channels to target a suitable shaping of the channel statistics performed at the UE-side. To optimize the covariance shaping strategies, we propose a low-complexity block coordinate descent algorithm that is proved to converge to a limit point of the original nonconvex problem. For the two-UE case, this is shown to converge to a stationary point of the original problem. Numerical results illustrate the sum-rate performance gains of the proposed method with respect to spatial multiplexing in scenarios where the spatial selectivity of the base station is not sufficient to separate closely spaced UEs.