VINES: Track 2: NSF-NICT: Open, AI-RAN Massive MIMO Platform for Smart Cities and Resilient Infrastructure: A U.S.-Japan Reference Design

Modern wireless networks struggle to deliver reliable, high-capacity service where it is needed most: dense urban hubs, automated construction sites, and disaster zones where many devices compete for limited radio spectrum under harsh, rapidly changing conditions. This U.S.-Japan collaborative project advances the national interest by developing an open, programmable wireless platform that uses artificial intelligence (AI) to make the next generation of cellular networks more resilient, efficient, and adaptable. The platform is built on Massive Multiple-Input Multiple-Output (Massive MIMO) radios — antenna arrays that serve many users at once by steering signals in space — combined with Open Radio Access Network (Open RAN) standards, which let components from different vendors interoperate, preventing costly vendor lock-in. By demonstrating the technology in real settings such as smart construction, emergency response, and ultra-dense intersections, the project creates a blueprint for the secure, domestically grounded wireless infrastructure that supports public safety, economic productivity, and U.S. leadership in sixth-generation (6G) systems. The effort also trains a new generation of engineers, releases open-source software and datasets that any researcher can reuse, and feeds results into international standards bodies, broadening access to advanced wireless technology and strengthening the U.S. innovation base. This project will design, build, and field-validate an open, Graphics Processing Unit (GPU)-accelerated, O-RAN-compliant Massive MIMO platform with AI-native control. The system integrates a 64-transmit/64-receive (64T64R) O-RAN radio unit, a GPU-accelerated physical layer, and an open-source higher-layer stack. These components are coordinated through real-time dApps that observe and control the physical and medium-access layers at millisecond timescales, alongside near-real-time and non-real-time controllers for network-wide orchestration. Research will deliver: dynamic per-transmission beamforming over the O-RAN fronthaul and multi-user MIMO scheduling; AI methods for uplink-heavy verticals, including neural receivers, interference-robust neural precoding, predictive low-latency scheduling, and semantic-aware slicing that jointly allocates radio and edge-computing resources; and AI methods for downlink-heavy verticals, including lightweight learned channel-state feedback compression and interference-aware multi-user scheduling. The platform will be evaluated through phased demonstrations advancing from laboratory integration to over-the-air field pilots, targeting quantitative key performance indicators for latency, reliability, concurrent-user scalability, and spectral efficiency. Expected outcomes include open-source software, curated radio-frequency and AI-training datasets, reproducible reference designs, and contributions to O-RAN and 3GPP specifications, and AI-RAN Alliance activities, establishing a validated path from component research to deployable NextG infrastructure. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.