CU-DU Split

The CU-DU split is the decision about where to cut the gNB in two and what to put on each side. The functions are divided by latency tolerance: the DU (Distributed Unit) holds the time-critical lower layers — RLC, MAC, and high-PHY — and has to sit close to the radio because those layers can't tolerate much transport delay. The CU (Central Unit) holds the higher layers — RRC and PDCP — which are more delay-tolerant and can therefore be centralised, often serving several DUs from one site.

The payoff of centralising the CU is coordination and efficiency: one CU pooling resources across many DUs makes mobility smoother and operations simpler. The constraint is the midhaul transport between CU and DU, which has its own latency and bandwidth budget you have to respect. There's a further wrinkle worth knowing — the CU itself can be split into CU-CP (control plane, the RRC) and CU-UP (user plane, the PDCP), which lets you place control and user processing independently. This is 3GPP's framework (the higher-layer split, often called option 2 at the CU-DU boundary), and it dovetails with O-RAN, which adds the open fronthaul split between DU and radio on top of it.