Abstract
—Multi-User Multiple-Input, Multiple-Output (MU-MIMO), and massive-MIMO (mMIMO) have been central technologies to the evolution of the latest mobile generations since they promise substantial throughput increase and enhanced connectivity capabilities. Still, the type of signal processing that will unlock the full potential of MU-MIMO developments, has not yet been determined. Existing realizations typically employ linear processing that, despite its practical benefits, can leave capacity and connectivity gains unexploited. On the other hand, traditional non-linear processing solutions (e.g., sphere decoders) promise improved throughput and connectivity capabilities but can be computationally impractical, with exponentially increasing computational complexity, and with their implementability in 5G-NR systems still being unverified. At the same time, emerging new Open Radio Access Network (Open-RAN) designs call for physical layer (PHY) processing solutions that are also practical in terms of realization, even when implemented purely on software. In this work, we present a first, purely software-based, Open-RAN compliant, 5G-NR MIMO PHY that operates in real-time and over-the-air, encompassing both linear and non-linear MIMO processing, and achieving support for 8 concurrently transmitted MIMO streams at a 10MHz bandwidth with just 12 processing cores. Here, we not only demonstrate that implementing non-linear processing is feasible in software within the stringent real-time latency requirements of 5G-NR, but we also compare it side-by-side in a real-time and over-the-air environment against traditional linear methods. We show that the gains of non-linear processing include substantially enhanced throughput with insignificant computational power overhead, the halving of the base-station antennas without performance degradation, and overloading factors of up to 300%.