Architecture

OCUDU is a full 3GPP/O-RAN compliant 5G NR gNB built around a strict separation between layer logic and inter-layer wiring. Every layer (CU-CP, CU-UP, DU-High, DU-Low, RU) is an owned object tree with a public interface and internal adapter notifiers; layers never call each other directly. The same DU-High binary therefore runs unchanged whether the CU-CP sits in the same process or on a remote host - the assembly code wires adapters to either an in-process connector or an SCTP transport at construction time.

This section covers the architecture from two angles:

• The static picture what the layers are, what they own, and which standardized interface (Uu, F1-C, F1-U, E1, N2, N3, Xn, E2) carries traffic between them.
• The runtime picture how slot indications, scheduling decisions, and async procedures flow through the call graph at gNB runtime, and how the cross-cutting infrastructure (timers, executors, GTP-U, metrics, PCAP, RRM) keeps the data plane real-time.

What’s here

The Overview is the canonical end-to-end walk-through: it covers all five layers, the deployment topologies (co-located gnb binary vs. split cu_cp + cu_up + du), the CU-CP / CU-UP / DU-High / MAC-scheduler / DU-Low internals, the cross-cutting infrastructure, and a layer-by-layer correlation summary. Read this first.

Per-interface deep dives live alongside it. They focus on a single 3GPP interface, the procedures it carries, and how OCUDU implements it across the relevant layer:

• F1AP interface - F1-C control-plane between CU-CP and DU (TS 38.473).
• NGAP interface - N2 control-plane between gNB and AMF (TS 38.413).
• E1AP interface - E1 bearer context management between CU-CP and CU-UP (TS 38.463).
• RRC - Radio Resource Control: UE connection state, procedure coroutines, SRB security (TS 38.331).
• Upper PHY - DL/UL baseband processing, LDPC/polar coding, modulation, channel estimation, DFT backends.
• MAC Scheduler - Slot indication pipeline, resource grid, HARQ, PDCCH/PUCCH, RAN slicing.