The radio access network has long been the most closed, most specialized corner of the telecom stack. That is changing fast. Since its launch earlier this year as a Linux Foundation project, OCUDU has set out to do for the RAN what Linux did for the server: deliver a complete, open source 5G/6G software stack. The Linux Foundation has assembled the entire cellular toolkit under one open-source roof, hosting projects that span everything from the SMO and O-Cloud to the core network and user equipment (UE).
The RAN Mini-Summit will take place on December 11 in Tokyo and is co-located with the ONE Summit Japan 2026. At the RAN Mini-Summit, the Open Source RAN community (including OCUDU, Duranta, and an array of other projects) and its partners are coming together in Hall B1 for a full day of talks covering everything from RAN fundamentals to AI at the physical layer, compact NPU-powered base stations and satellites working as coordinated transmission points.
In this blog we take a look at the featured talks. The complete agenda for the RAN Mini-Summit can be viewed here.
Understanding OCUDU – The “Linux of RAN”
Franciszek Maroszek, Software Radio Systems; Nils Fürste & Francisco Paisana, SRS/OCUDU TSC
Radio access networks behave very differently from the networks most engineers know: radio conditions shift constantly, users are on the move, and every decision has to respect the battery life of the devices being served. The SRS team will first build a clear mental model of how the RAN works, then show where OCUDU fills the gap in the open source ecosystem. You will get a tour of the stack’s internals, a practical look at how to deploy it, and a view of how AI can be brought to bear on classic RAN problems, whether through xApps and dApps or by building intelligence directly into the stack itself.
Beyond CPU Pinning and MIG: Telemetry for Autonomous Cloud Native AI-RAN
Sagar Arora, OpenAirInterface Software Alliance/Duranta TSC & Marco Gonzalez, Red Hat
AI-RAN promises a single platform for both radio and AI workloads, but sharing hardware with a neighboring LLM is hard when a DU faces slot deadlines measured in tens of microseconds. CPU pinning and Multi-Instance GPU partitioning are today’s go-to answer, yet memory bandwidth, last-level cache and package power still leak across partitions, sometimes badly enough to break a cell, and those metrics usually go uncollected. Sagar and Marco will show how to capture the missing hardware counters from the O-Cloud alongside DU-internal application metrics such as slot deadline margin and fronthaul lateness, join them on the node, and feed them to the SMO and rApps. The payoff is a closed loop that keeps RAN SLOs tight while freeing up everything else for general AI work.
Role of OCUDU for Private Mission-Critical Networks
Ashish Bansal, Raycom & Rami Mirsky, Raycom Communications Solutions, Inc.
When the stakes are highest, networks need to be small, autonomous and deployable at a moment’s notice. Ashish and Rami will explore how OCUDU’s features line up with the needs of private 5G for defense and mission-critical operations, enabling AI-capable networks in compact form factors. They will explain how multi-band support lets these distributed, modular networks become plug-and-play, rapidly deployable and vendor-agnostic in demanding environments.
AI-RAN Platform for ISAC: Enabling Multimodal AI at the Edge
Hidetsugu Sugiyama, Red Hat
Integrated Sensing and Communication turns the radio channel from something to be equalized away into a signal worth studying. That makes sensing models tightly bound to each cell site’s physical surroundings, demanding constant adaptation to clutter, foliage, seasons and shifting crowds. And because RF sensing data is not ordinary user traffic, it cannot simply flow through the UPF. Hidetsugu will show how dApps running on an OpenShift/Kubernetes-based AI-RAN platform, paired with an Edge Data Lake, can securely share ISAC insights with partners such as the automotive sector to power multimodal AI at scale.
Do We Really Need GPU or Accelerators for AI-RAN
Gilles Garcia, AMD
It is often taken for granted that AI-RAN requires GPUs. Gilles wants the community to test that assumption. This session kicks off a conversation about a community-driven benchmarking framework for AI-RAN workloads across CPU-only, CPU plus custom accelerator, and CPU plus GPU configurations. The goal: give operators and other decision makers the data to know where CPUs are enough, where targeted accelerators or ASIC offload make sense, and where GPUs genuinely earn their place, with OCUDU as the common ground for those comparisons.
Portable Real-Time AI-RAN in OCUDU for Open, AI-Native 5G and 6G
Matt Pennybacker, DeepSig & Tim O’Shea, DeepSig Inc
Open RAN standards already let AI plug into slower management loops, but the places where AI could matter most, the physical layer and the microsecond-scale scheduler, remain largely out of reach, with every algorithm tied to a particular stack. Matt and Tim will present how OCUDU’s AI-RAN Working Group is changing that by mapping 37 use cases onto three classes of plug-in contract sharing a common control channel. These dApps are signed, loaded into a live base station, tested in shadow against the proven implementation, then switched on or rolled back with no code change and no outage. Because open and proprietary implementations meet the same contracts, a university prototype can be judged on equal terms with a commercial product, opening a path toward the AI-native air interfaces expected in 6G.
NPU Accelerated Radio Access Network on a 100mm Board
Raymond Knopp, OpenAirInterface Software Alliance & Douglas Knisely, Qualcomm Technologies, Inc./OCUDU Governing Board
How small can a complete 5G RAN get? Raymond and Douglas will report on porting Duranta/OpenAirInterface as an O-RAN distributed unit onto Qualcomm’s DragonWing IQ-9075, a 100 x 100 mm board with eight Cortex-A78C cores and a Hexagon DSP. They will walk through the hardware, the O-RAN 7.2 fronthaul setup to a commercial O-RU, and the real-world lessons of the port, from aarch64 cross-compilation and ARM SIMD work to the real-time tuning and per-slot instrumentation that kept overturning their assumptions. The board already carries 700 Mbps downlink at 100 MHz TDD with no steady-state loss, with 1 Gbps+ expected by December. They will also explain why uplink at scale is the next frontier, their plan to move the receive chain onto the Hexagon, and benchmarks of hand-written kernels against Zig- and Mojo-generated code.
Extending OCUDU for NR Dual Connectivity and 6G–NTN Multi-Connectivity
Stefano Taborelli, CNAM
Stefano will present a proposed extension of OCUDU that brings NR Dual Connectivity, with full dual-stack support across both the Central Unit and the User Equipment. He will cover the architecture and software changes that coordinated multi-connectivity requires, spanning control and user plane handling, bearer management, scheduling coordination, mobility, state synchronization and end-to-end observability. He will also sketch an extensible design path toward future multi-connectivity between terrestrial 6G networks and non-terrestrial networks.
Satellites as Transmission Points: Enabling Reliability and Capacity for NTN Multi-TRP in Open CU/DU
Peng Tan, Skylo Technologies
Close out the day looking skyward. Satellite direct-to-cell is becoming part of 5G and 6G, but each satellite still operates as an isolated link. Peng will demonstrate how the open CU/DU can serve as a reference platform for satellite access, using standard 3GPP Rel-17 to Rel-19 mechanisms to bring two-satellite non-coherent joint transmission multi-TRP to NTN: two transparent-payload LEO satellites, fed by a single ground gNB, serving one ordinary handset. The approach supports a diversity mode for reliability and a spatial-multiplexing mode for capacity, with link-level results showing roughly 3.7 dB of macro-diversity gain and up to about 64% higher rank-2 sum-rate. Since the best mode depends on changing geometry, Doppler and blockage, mode selection is framed as an open rApp, a natural fit for AI-native control and the open-source roadmap.
Join Us this December at the RAN Mini-Summit
From first principles to satellites in orbit, the RAN Mini-Summit shows just how much ground the open RAN community is covering, and how central OCUDU has become to that work in its very first year. Across the day, a common thread emerges: open, shared foundations make it possible to compare ideas fairly, bring AI to every layer of the network, and build toward 6G together rather than in silos.
Whether you come for the fundamentals, the AI-RAN debates or the hardware deep dives, there is a session for you. Add the talks to your schedule, bring your questions, and come meet the developers, researchers and operators shaping the future of the radio access network. We’ll see you in Hall B1.
Check out the complete agenda and register here.