GPU Acceleration for 5G RAN Processing

This technical report presents a study conducted by GlobalLogic on the implementation of GPU acceleration for processing essential 5G channels within the 5G Distributed Unit (DU) as part of the gNodeB. The focus is on two key physical channels: the Physical Random Access Channel (PRACH) and the Physical Downlink Shared Channel (PDSCH). The report outlines the challenges faced by traditional CPU-based implementations, particularly in terms of performance limitations when handling massive traffic loads. It discusses the potential benefits of utilizing NVIDIA GPUs, including improved processing times, scalability, and energy efficiency. The study details the setup and system structure used for testing, which involved both simulated and real-world environments to evaluate the performance of the GPU-accelerated solution. By examining the porting process of the PRACH signal processing to GPU, the report aims to demonstrate the feasibility and advantages of GPU-based solutions in modern RAN architectures, paving the way for future advancements in AI-driven telecommunications.