NVIDIA Aerial CUDA-Accelerated RAN: 5G/6G MAC-Scheduler Acceleration Library (cuMAC)


Telcos have spent billions to buy 4G/5G spectrum and will be expected to spend again to buy 6G spectrum, to meet the ever-increasing needs of wireless connected devices. While telcos are deploying 5G in new spectrum bands and using massive multiple input multiple output (MIMO) to maximize wireless performance at the air interface, technological enhancements are available that can improve spectral efficiency across all layers of the radio access network (RAN) stack, for all types of network deployments.

cuMAC, a Layer 2 MAC scheduler acceleration library, is developed to improve spectral efficiency by introducing a multi-cell scheduler with enhanced algorithms within Layer 2 of the RAN protocol stack. The MAC scheduler, controls how resources are scheduled for all the user equipments (UEs), across all cells, determining the overall spectral efficiency achieved in each cell. cuMAC will provide a net increase in overall performance per watt measured at the cell level, compared to baseline single cell scheduler approaches. NVIDIA is developing cuMAC on the GPU, to bring efficient MAC scheduler implementations to the telco world.

cuMAC will be provided as part of the Aerial CUDA-Accelerated RAN platform, and will be accelerated on the GPU to provide the best performance per watt as compared to implementations on CPU only architectures. Figure 1 shows how cuMAC fits within the overall Aerial CUDA-Accelerated RAN platform, and how L1 and L2 acceleration is enabled on the same platform.


Figure 1. Aerial CUDA-Accelerated RAN Platform

Key Features

NVIDIA Aerial cuMAC acceleration libraries has the following key aspects:

Feature 1: Multi-Cell Scheduling

  • Single MAC scheduler, controlling multiple PHY layer instances, so that scheduler can factor in all connected devices across multiple cells, to achieve best connected device experience.

Feature 2: Algorithms

  • Proportional fair algorithm for UE down-selection, at a per transmission time interval (TTI) level, based on UE priority weight.

  • Proportional fair algorithm for physical resource block (PRB) allocation, at a per TTI level, based on metrics such as channel conditions for each UE.

  • MIMO layers selection, decided on a per PRB group level, for each UE.

  • Modulation and coding scheme (MCS) selection, to meet block error rate (BLER) targets, for each UE.

  • Future algorithms cover UE grouping, dynamic beam forming and AI/ML enhanced techniques.

Feature 3: Standard MIMO and Massive MIMO Support

  • cuMAC scheduler acceleration supports various MIMO implementations from standard MIMO 4T4R to massive MIMO 64T64R and beyond.

Feature 4: Modular Architecture

  • cuMAC functions are implemented in a modular fashion, simplifying adoption and customization. Example modules include UE selection, PRB allocation and so on.

  • cuMAC will adopt functional application platform interface (FAPI) like interface, so that L2 can interface to cuMAC in a similar way to how L2 interfaces to cuPHY.

Feature 5: GPU Accelerated

  • cuMAC uses the CUDA libraries to achieve high performance and efficient GPU implementation of L2 scheduler.

Target Audience

Telecom Operators, Network Equipment Providers, Test & Measurement Equipment Providers, Network Planning ISVs, Academia & Universities

Value Proposition

cuMAC provides operators with an enhanced L2 MAC scheduler, implemented on the NVIDIA accelerated computing platform GPU, enabling better spectral efficiency per watt, compared to CPU only implementations.

© Copyright 2024, NVIDIA. Last updated on Apr 19, 2024.