Speaker
Description
Artificial intelligence is rapidly reshaping the radio access network, but hands-on experimentation with AI-RAN has remained out of reach for most practitioners: commercial stacks are closed, and the gap between cellular physical-layer software and the SDR tools this community knows well is wide. This 3–4 hour hands-on workshop closes that gap using OCUDU—the open-source 5G CU/DU project that evolved from srsRAN—together with USRP hardware and GNU Radio.
Participants will build and install OCUDU on their own laptops and bring up a standalone 5G network using a USRP as a Split-8 radio unit. Using pre-provisioned SIM cards supplied by the presenters, attendees will attach real commercial smartphones to the networks they have just built; a software-UE path based on OpenAirInterface is provided for participants without compatible hardware. We will then install DeepSig's OCUDU workgroup fork featuring a GPU-accelerated PHY and examine live physical-layer metrics to see where acceleration matters in a real gNB processing pipeline. Because this is GRCon, we will then connect the cellular stack to the GNU Radio ecosystem: a ZMQ tap exposes the live uplink OFDM resource grid as a streaming source, and attendees will build flowgraphs that ingest and process real 5G uplink signals—opening the L1 of a working standards-based network to the full GNU Radio toolbox. We will conclude with demonstrations of AI-RAN in practice, including a neural-network-based PUSCH receiver running inside the GPU PHY, and discuss how attendees can train and deploy models of their own in the same pipeline.
GRCon 2026 is being hosted at the site of the AERPAW lab, whose community already builds extensively on srsRAN/OCUDU for cellular research. This workshop offers directly applicable skills to this community: attendees leave with a complete, open-source 5G AI-RAN sandbox running on their own equipment and the knowledge to extend it. The software used throughout the workshop is open source and publicly available today, including OCUDU, the GPU-accelerated PHY developed in the OCUDU hardware-acceleration working group, and the OpenAirInterface soft UE. The GNU Radio integration examples developed for this workshop will be published openly alongside the workshop materials, which will be shared as public PDFs after the conference.
Code Availability
- OCUDU: https://gitlab.com/ocudu/ocudu
- CUDA-accelerated OCUDU (WG1 hardware acceleration): https://gitlab.com/ocudu/work_groups/wg1_hw_accel/cuda_accelerated_ocudu
- OpenAirInterface (soft UE): https://gitlab.eurecom.fr/oai/openairinterface5g
| Talk Length | N/A |
|---|---|
| Link to Open Source Code | - OCUDU: https://gitlab.com/ocudu/ocudu - CUDA-accelerated OCUDU (WG1 hardware acceleration): https://gitlab.com/ocudu/work_groups/wg1_hw_accel/cuda_accelerated_ocudu - OpenAirInterface (soft UE): https://gitlab.eurecom.fr/oai/openairinterface5g |
| Acknowledge | Acknowledge In-Person |