Turbo-HB: A Sub-Millisecond Hybrid Beamforming Design for 5G mmWave Systems

Abstract

Hybrid beamforming (HB) architecture has been widely considered for 5G mmWave systems. It reduces hardware complexity by allowing the number of RF chains to be far fewer than the number of antennas. A major practical challenge for HB is to obtain a beamforming solution in real-time. In 5G NR, new frame structures with short TTIs are employed to support mmWave communications. Under such frame structures, it is necessary to obtain a beamforming solution with a time resolution varying from 1 ms to 125 us -- an extremely stringent time requirement considering the complexity involved in HB. In this paper, we present the design and implementation of Turbo-HB -- a novel beamforming design under the HB architecture that is capable of offering the beamforming matrices in less than 500 us. The key ideas of Turbo-HB include: (i) reducing the complexity of computation-intensive SVD operations by exploiting channel sparsity at mmWave frequencies, and (ii) achieving large-scale parallel computation with minimal memory access. We implement Turbo-HB on an off-the-shelf Nvidia GPU and conduct extensive experiments. Our experimental results demonstrate that Turbo-HB can obtain a beamforming solution in 500 us for up to 100 RBs and 10 MU-MIMO users on each RB while offering competitive throughput performance compared to state-of-the-art (non-real-time) algorithms.