AFundamentallyNewKindofgNB
CNS is building the Radio Access Neural Network — a uniquely disaggregated gNB architecture that adds a dedicated intelligence plane to the Central Unit. Our Central Unit - Intelligence Plane (CU-IP) is a novel third element in the CU split architecture, peer to both the CU-CP and the CU-UP. Built on OCUDU.
How the RA-NN works
The CU-IP is a dedicated intelligence plane — peer to the CU-CP and the CU-UP, inside the gNB as a native function. One function, one representation, all tasks.
Internal, not external
The CU-IP receives the same decoded L3 data the control plane uses for its own decisions — directly, via an internal interface. No E2. No protocol boundary.
Dual interface
One interface to the control plane — receiving L3 data and returning learned decisions. One interface to the platform — reading cell configuration and writing learned state. The CU-IP is the only CU function with both.
One representation, all tasks
One continuously-updated representation of the radio environment. PCI, neighbor discovery, coverage, mobility — all consume the same representation. Intelligence compounds instead of fragmenting.
Validated Over the Air
The RA-NN is operating on real RF signals with our three-plane CU split.
Cell Identity Coordination
The CU-IP resolves cell relationships from live UE reports, detects conflicts, and coordinates assignments. The network organizes its own cell structure without manual planning.
RF Spatial Reasoning
Directly from per-UE observations, the CU-IP learns the spatial geometry of the radio environment — understanding where users exist relative to surrounding cells without explicit propagation models.
No RIC. No xApps. No rApps.
Built on OCUDU.
When intelligence is a native CU function, the external systems become architecturally unnecessary. We'd like to show you how.