Researchers at Northeastern University‘s Institute for Intelligent Networked Systems (INSI) Open6G OTIC have successfully demonstrated the first open-source prototype of a massive MIMO (mMIMO) AI-RAN system. The demonstration marks a milestone in open, AI-native wireless systems by bringing together O-RAN compliant massive MIMO hardware from AmpliTech Group, with open-source software from OpenAirInterface (OAI) and NVIDIA GPU-accelerated RAN intelligence, into a fully integrated, standards-compliant platform.
The demonstration combined an Amplitech mMIMO O-RAN Category B radio unit, NVIDIA AI Aerial software for layer 1 and layer 2 RAN, and OAI’s L2+ stack into a single cohesive system using exclusively open-source software.
The INSI team showcased a two-stage precoding architecture combining a 4-layer CSI-feedback-based MIMO precoder accelerated on NVIDIA AI Aerial with a 64-antenna element codebook-based beamformer implemented in the CAT-B O-RU. This system demonstrated sustained throughput across multiple UEs under mobility conditions, enabled by proper beam management at layer 2. The results validate that a high-performance, GPU-accelerated AI-RAN architecture can be built and reproduced without reliance on proprietary, closed-stack solutions.
Massive MIMO systems — which use large antenna arrays to serve multiple users simultaneously through spatial multiplexing — have historically required tightly integrated, vendor-specific implementations. This demonstration challenges that assumption by showing that the full stack, from the physical layer up through the RAN control plane, can be assembled from open, interoperable components. The Northeastern Open6G Open Testing and Integration Center had recently certified the AmpliTech Group mMIMO CAT-B RU for O-RAN conformance, a key step toward successful interoperability.
“GPU-accelerated RAN processing has been a missing piece in the open ecosystem — powerful in principle but rarely validated end-to-end at this level of the stack” said Tommaso Melodia, Director of the Institute for Intelligent Networked Systems and William Lincoln Smith Professor with the Department of Electrical and Computer Engineering at Northeastern University. “What this demonstration shows is that openness and performance are not trade-offs. You can have a fully open, reproducible system and still push the boundaries of what massive MIMO can deliver through algorithmic control. “
The integration of NVIDIA AI Aerial GPU-accelerated L1 processing with OAI’s L2+ is a key architectural feature of the prototype, enabling the compute-intensive physical layer operations of massive MIMO to be handled efficiently on NVIDIA’s AI-RAN platform while preserving full software openness at higher layers.
“OpenAirInterface has always been committed to open, standards-compliant implementations of next-generation RAN,” said Irfan Ghauri, Director of Operations at the OpenAirInterface Software Alliance. “Seeing OAI’s CU/DU stack integrated into a full mMIMO AI-RAN prototype developed at Northeastern’s Open6G, alongside NVIDIA AI Aerial and Amplitech hardware demonstrates what becomes possible when the community builds on open foundations.”
“This is a major validation milestone for AmpliTech. Our 64T64R massive MIMO O-RAN radio unit was successfully integrated into a fully open-source, GPU-accelerated AI-RAN system, demonstrating that high-capacity massive MIMO performance and true Open RAN openness can be achieved together in real, disaggregated network environments,” added Fawad Maqbool, CEO and CTO, AmpliTech Group. “Participating alongside leading technology innovators such as NVIDIA and OpenAirInterface in Northeastern University’s first open-source prototype of a massive MIMO AI-RAN system underscores the strength and readiness of AmpliTech’s 5G radio technology. We believe this achievement elevates AmpliTech’s visibility within the AI-RAN and Open RAN ecosystem and creates a compelling foundation for additional customer engagement, strategic collaboration, and business development opportunities.”
The Northeastern INSI team led the system integration, testbed configuration, and validation measurements, providing a reproducible reference implementation that academic and industry researchers can build upon.
The demonstration aligns with growing momentum around Open RAN and AI-native wireless systems, where flexibility, vendor interoperability, and intelligent control are increasingly viewed as essential properties for future 5G and 6G deployments.
Click here to learn more about Northeastern’s Institute for Intelligent Networked Systems.