Immersive AR and VR
Needs: multi-Gbps throughput with sub-millisecond latency. A stalled frame breaks immersion.
FORMULA3: steers beams and surfaces ahead of the user, so movement and blockage don't interrupt the stream.
NSF × Research Council of Finland VINES · Track 1
FORMULA3
Framework for Optimal Resource Management Using GeneraLizable AI in FR3 Dynamic Programmable Wireless Environments
Self-adapting FR3 networks that combine learning, sensing, and programmable environments to deliver reliable, high-throughput, low-latency 6G connectivity.
Try it Illustrative model
What do people moving through a room do to a 6G link?Can a 6G link get between the street and indoors?What do traffic and trees do to a 6G street link?
An interactive illustration of a programmable wireless environment appears here when JavaScript is enabled.
Throughput now
– Gbps
| Band | None | RIS | MRA | Both |
|---|---|---|---|---|
| Sub-73.5 GHz | ||||
| FR315 GHz | ||||
| mmWave28 GHz |
- Throughput of the selected option
- 2 Gbps target
- Below target
- ▲▼Person enters / leaves
- Vehicle passes
The result
Programmable surfaces keep FR3 above multi-Gbps rates far more of the time
The interactive scenario above compares sub-7, FR3, and mmWave, with and without RIS and MRA surfaces.
From the illustrative scenario above, not from measurements. Measured results will appear in publications.
Who benefits
Applications that cannot wait for a dropped link
Emerging 6G applications need consistent multi-Gbps throughput and sub-millisecond latency indoors, not just good peak numbers.
Remote healthcare
Needs: links that stay up, without surprises, for remote monitoring and tele-surgery.
FORMULA3: forecasts outages before they happen and reroutes through surfaces. The team's earlier work forecast outages about 100 ms ahead.
Advanced manufacturing
Needs: reliable wireless in crowded spaces where machines, people, and layouts keep changing.
FORMULA3: senses occupancy and layout changes from the channel itself, and adapts its AI models when the floor is rearranged.
The question
How can FR3 (7–24 GHz) programmable wireless environments, built from RIS and metallic reflector arrays, keep communication fast and reliable as people move, rooms fill, and furniture shifts?
~10× more bandwidth than sub-7 GHz
~2× the range of mmWave, with better wall penetration
80%+ of mobile data is consumed indoors, where FORMULA3 focuses
Approach
Measure. Model. Predict. Adapt.
Reacting to a lost link is too late for AR/VR or tele-surgery. FORMULA3 makes the network proactive. It learns from real FR3 measurements, predicts outages before they happen, and reconfigures surfaces using context it senses from the channel itself.
-
01
Measure
Characterize dynamic FR3 channels on new hybrid RIS and reflector-array testbeds, indoors and outdoor-to-indoor.
-
02
Model
Build validated channel models and generate rich datasets of mobility, occupancy, and layout dynamics.
-
03
Predict
Use physics-informed AI to forecast channel gains and outages before they disrupt service.
-
04
Adapt
Reconfigure surfaces and reassign users using context sensed from the channel (ISAC), and keep learning as rooms change.
Every step is validated against measurements and simulation in a unified evaluation platform.
Research thrusts
Four integrated thrusts
Measurements feed the models, the models train the AI, and context sensing keeps the AI accurate in rooms it has never seen.
T1
Channel characterization & dataset generation
A first-of-its-kind hybrid FR3 testbed with fabricated RIS panels and metallic reflector arrays, plus a calibrated simulation suite for dynamic data.
Read more → T2AI-native proactive resource management
Physics-informed AI that predicts surface channels, excludes shadowed elements, and configures surfaces to suppress LoS/NLoS interference.
Read more → T3Generalization via ISAC & continual learning
Sensing occupancy, mobility, and layout from the channel itself to pick the right model, and retraining when the environment drifts.
Read more → T4Validation & evaluation
Benchmarking every model and strategy on a unified measurement and simulation platform under realistic, changing conditions.
Read more →What we are aiming for
From today's limits to FORMULA3's targets
Where conventional approaches fall short today, and what the project sets out to achieve.
AI facing conditions it wasn't trained on
Today>50%performance loss
Target<2%QoS degradation
Sensing occupancy, mobility, and layout from the channel picks the right model; continual learning handles rearranged rooms.
Signaling overhead
TodayAllsurface elements managed
Target−50%overhead
Shadowed RIS tiles and reflector elements are excluded instead of reconfigured, so the network spends effort only where it helps.
Channel prediction with few sensors
TodayHigherror from sparse data
Target104×lower error
Physics-informed AI builds propagation knowledge into training and densifies sparse channel data.
Interference between direct and reflected paths
TodayCappedQoS from LoS/NLoS interference
Target10×better QoS
Users are assigned to surfaces, and RIS, SIM, and MRA settings are chosen so direct and reflected signals stop canceling each other.
"Today" reflects conventional approaches and the team's earlier studies, mostly at mmWave and in optical wireless. Targets are research goals set at the start of the project, not measured results; results will be reported in publications.
Partners
Five institutions, two countries
FORMULA3 combines expertise in FR3 measurement, reconfigurable surfaces, reflector arrays, generalizable AI, and integrated sensing across the United States and Finland.
Updates
Latest news
-
FORMULA3 officially begins
The NSF awards supporting FORMULA3 started on October 1, 2026, launching a three-year U.S.–Finland collaboration that runs through September 2029.
-
FORMULA3 selected for NSF–Research Council of Finland VINES funding
A collaborative effort of about $1.54 million, spanning five universities in the United States and Finland, will develop AI-native FR3 programmable wireless environments.
