BECAS BGI2026 - GPU‑Accelerated 5G Systems & Intelligent Frequency Optimization Internship- ESTADOS UNIDOS DE AMERICA (Boston)
- Código Oferta:14828
- Proyecto: IMFAHE_NORTHEASTERN UNIVERSITY INSTITUTE FOR INTELLIGENT NETWORKED SYSTEMS (INSI) / BECAS BGI2026 - GPU‑Accelerated 5G Systems & Intelligent Frequency Optimization Internship- ESTADOS UNIDOS DE AMERICA (Boston) / ESTADOS UNIDOS DE AMERICA / Boston
- País: ESTADOS UNIDOS DE AMERICA
- Municipio: Boston
- Entidad: IMFAHE_NORTHEASTERN UNIVERSITY INSTITUTE FOR INTELLIGENT NETWORKED SYSTEMS (INSI)
- Tipo de contrato: Beca
- Duración: Mínimo 6 meses
- Salario mensual: 1.785 euros brutos
- Observaciones: Project Title: Intelligent Frequency Management for GPU-Accelerated 5G Open RAN: From Empirical Characterization to Adaptive Optimization Background and Motivation: The virtualization of 5G Radio Access Networks (vRAN) on GPU-accelerated platforms such as NVIDIA Aerial enables flexible, software-driven deployments. However, prior work by this research group has revealed a critical and previously unexplored challenge: the interaction between CPU and GPU operating frequencies has a non-monotonic and complex effect on throughput stability. Specifically, for a given GPU frequency, only certain CPU frequencies yield stable performance, creating frequency sweet spots separated by regions of high variability. This phenomenon, hypothesized to arise from timing synchronization between CPU scheduling, PCIe data transfers, and the 0.5 ms slot boundaries of 5G New Radio, has significant implications for real-world ORAN deployments where standard power management policies (DVFS) can inadvertently degrade service quality. Research Objectives: This internship will extend the foundational work in three directions: (1) Validate the timing synchronization hypothesis using NVIDIA Nsight Systems profiling to capture fine-grained CPU-GPU interaction traces, correlating kernel launch latencies, PCIe transfer timing, and FAPI interface deadlines with observed throughput instability. (2) Extend the experimental characterization to multi-UE scenarios and additional GPU hardware (NVIDIA L40S, H100) to assess generalizability across deployment configurations. (3) Design and prototype a machine-learning-based frequency controller, implemented as an O-RAN RIC xApp, capable of dynamically selecting optimal CPU-GPU frequency pairs based on real-time system telemetry.
- Nivel Académico: Predoc / estudiante de doctorado
- Fecha Alta: 2026-07-01
- Fecha Inicio: 2026-07-23
- Titulación: Ingeniería Telecomunicaciones, Ingeniería Informática, Ingeniería eléctrica
- Observaciones Idioma:
- Informática: 4G/5G, Python, C/C++, Linux, GPU (CUDA), PyTorch, scikit-learn, Docker, Git
© 2026 Fomento de San Sebastián.
