Semi-Blind Detection in Hybrid Massive MIMO Systems via Low-Rank Matrix Completion

In massive multiple-input multiple-output (MIMO)systems with hybrid analog/digital architectures, large trainingoverhead is required for conventional pilot-only methods toestimate channel accurately before detecting data. To reduce thetraining overhead, a semi-blind detection method is proposedfor data detection without knowing channel in an uplink multiuser system. The main idea is to exploit the received signalcorresponding to both the pilot and data payload for channelestimation or data detection via a low-rank matrix completionformulation. The leveraged low-rank property stems from thefact that the number of active users K is typically much smallerthan the number of antennas N_a at a base station and thenumber of time slots T_c in a coherence interval. Comparedwith the pilot-only method, the number of pilots required isreduced from an order of N_a to K. Two iterative algorithmsare introduced to solve the low-rank matrix completion problem:regularized alternating least squares and bilinear generalizedapproximate message passing. We further extend the semiblind detection method to systems with low-resolution analogto-digital converters. Simulation results show that the proposedmethods achieve significant performance gain over the pilotonly method with reduced training overhead for hybrid massiveMIMO systems in various settings.