RIC

The RAN Intelligent Controller is O-RAN's bid to make the radio network programmable. Rather than every optimisation being baked into vendor firmware, the RIC gives optimisation logic a standard platform to run on, fed by RAN telemetry and able to push decisions back down. The point is to bring closed-loop, data-driven control — increasingly ML-driven — to functions that were previously fixed.

It comes in two tiers, split by how fast they need to react:

• Near-RT RIC — operates roughly on a 10 ms to 1 s timescale, hosting xApps that do things like traffic steering, load balancing, and interference management. It talks to the CU/DU over the E2 interface.
• Non-RT RIC — operates above ~1 s, lives inside the SMO, and hosts rApps for policy, analytics, and ML model training. It guides the near-RT RIC over the A1 interface.

The division of labour is the thing to internalise: the non-RT side trains models and sets policy on a relaxed clock; the near-RT side applies them fast enough to matter for live radio behaviour. Anything genuinely real-time (sub-10 ms, like scheduling) stays in the DU — the RIC is not in that loop.