Aerial CUDA-Accelerated RAN
Aerial CUDA-Accelerated RAN 24-1 (Archive)

Future Development

  • GPU accelerated multi-user MIMO user pairing/grouping.

    The optimization of MU-MIMO user pairing/grouping involves a large amount of complex matrices computations for determining the most suitable set of UEs to be multiplexed on each time-frequency resource via MU-MIMO transmissions. These complex matrix computations are needed for evaluating the orthogonality among each possible group of UEs’ MIMO channels and/or their achievable data capacity. For a pure CPU-based MAC scheduler, such computation is a major computational bottleneck due to the limited computing resources available on CPU. The burden becomes even higher when multiple cells are considered for joint optimization. The cuMAC library will provide GPU-accelerated MU-MIMO user pairing/grouping algorithms to offload such optimization computation from the CPU scheduler to GPU. Both single-cell and multi-cell versions of algorithms will be provided.

  • GPU accelerated beamforming weights computation.

    Similar to the optimization of MU-MIMO user pairing/grouping, the computation of optimal beamforming weights also involves a large amount of complex matrices operations and adds burden on the CPU-based MAC scheduler. The cuMAC library will provide GPU-accelerated beamforming algorithms to offload this computation to GPU. Both single-cell and multi-cell versions of algorithms will be provided.

Previous Examples
Next Aerial Data Lake
© Copyright 2024, NVIDIA. Last updated on Apr 26, 2024.