Abstract
—This work introduces Gyre Precoding (GP), a novel linear multiuser multiple-input multiple-output (MU-MIMO) precoding approach. GP performs rotations of the symbols of each spatial layer to optimize the precoding performance. To find the rotation angles, we propose a near-optimal, gradient descent–based low-complexity algorithm. GP is constellation-agnostic and does not require significant changes to conventional receiver procedures or wireless standards. Computer evaluation results show that GP can achieve 8 dB SNR gains over linear precoding techniques and 2 dB over suboptimal symbol-level precoding (SLP) methods for a 16 × 16 MU-MIMO system. Furthermore, in a 64×12 massive-MIMO scenario in a 5G New Radio (5GNR) setup, GP achieves a 13% higher throughput gain over zero-forcing precoding. Index Terms—Multi-user multiple-input multiple-output (MU-MIMO), precoding.