Network & Telecom

Centralized Unit (CU)

Q: What is the difference between CU-CP and CU-UP?

3GPP allows the CU to be further split into a Control Plane (CU-CP) and one or more User Plane (CU-UP) instances. The CU-CP hosts the RRC protocol and the control-plane part of PDCP, terminating signaling toward the core and managing radio bearers. Each CU-UP hosts SDAP and the user-plane part of PDCP, handling actual data traffic. The two are connected by the E1 interface, which carries bearer-context setup and modification messages. This separation lets an operator scale user-plane throughput independently of control-plane signaling capacity, and place CU-UP instances closer to the edge for low-latency services while keeping the CU-CP centralized.

Q: Which interfaces connect the CU to the DU and the core network?

The CU connects downward to one or more distributed units over the F1 interface, split into F1-C for control plane (carrying F1AP signaling) and F1-U for user plane (GTP-U tunnels over UDP/IP). Upward, the CU terminates the NG interface to the 5G core: NG-C toward the AMF and NG-U toward the UPF. When the CU itself is split, the internal E1 interface links CU-CP and CU-UP. The Xn interface connects neighboring gNBs at the CU level for handover and dual-connectivity signaling. All of these are packet-based and run over a routed IP transport, which is why the CU is typically virtualized on commodity servers rather than purpose-built radio hardware.

Q: How much transport latency can the F1 interface between CU and DU tolerate?

Because the CU-DU split is above the RLC layer (option 2 in 3GPP TR 38.801), the F1 interface is far more latency-tolerant than the fronthaul between DU and RU. The PDCP layer can buffer and reorder, so F1 one-way latency budgets of 1 to 10 ms are acceptable, supporting CU-DU separation distances of up to roughly 80 to 200 km over fiber. This is the midhaul segment. By contrast, the lower-layer split on the fronthaul (option 7-2x, eCPRI) demands sub-100 microsecond latency and limits the DU-RU distance to about 20 km. The relaxed F1 budget is what allows many DUs to be pooled at a single centralized CU site.

/SEE-yoo/ · sometimes "CU-CP" and "CU-UP"

Within the 5G gNB functional split, the CU is the logical node that hosts the higher-layer radio protocols, RRC, SDAP, and PDCP, while the lower layers (RLC, MAC, and high-PHY) live in the distributed unit. The two are joined by the F1 interface. Because the split sits above the latency-critical PHY, the CU can be virtualized on commodity servers and centralized at a regional data center, pooling traffic from many distributed units and the radio units behind them. 3GPP further permits a control-plane and user-plane split, giving a CU-CP that handles RRC signaling and a set of CU-UP instances that handle SDAP and user-plane PDCP, connected internally by the E1 interface. This disaggregation is the foundation of O-RAN cloud RAN deployments.

Category: Network & Telecom

Hosts: RRC, SDAP, PDCP

Southbound IF: F1 (F1-C / F1-U)

Where the CU Sits in the 5G RAN Split

The 5G New Radio base station, the gNB, is no longer a single monolithic box. 3GPP TR 38.801 defined eight candidate split points between the protocol layers, and the industry converged on a two-tier disaggregation: a higher-layer split (option 2, above PDCP/RLC) that separates the CU from the distributed unit, and a lower-layer split (commonly option 7-2x, inside the PHY) that separates the distributed unit from the radio unit. The CU owns RRC, the radio resource control state machine that manages connection setup, mobility, and bearer configuration; SDAP, which maps QoS flows to radio bearers; and PDCP, which handles header compression, ciphering, integrity protection, and in-order delivery during handover.

Because PDCP can buffer and reorder packets, the link below it tolerates milliseconds of transport delay rather than microseconds. That relaxed budget on the F1 interface is exactly what lets the CU be pulled back from the cell site into a centralized edge or regional cloud, while the distributed unit and radio unit stay near the antenna. A single CU instance commonly aggregates dozens of distributed units, which centralizes mobility decisions, simplifies inter-cell coordination, and concentrates the signaling toward the 5G core.

The control-plane and user-plane separation refines this further. The CU-CP terminates the NG-C interface toward the AMF and runs the RRC and control-plane PDCP, while one or more CU-UP nodes terminate NG-U toward the UPF and run SDAP and user-plane PDCP. Operators scale CU-UP capacity for data-heavy cells independently of CU-CP signaling capacity, and can place CU-UP instances at the edge for ultra-reliable low-latency services while keeping a centralized CU-CP for consistent mobility handling.

Interfaces and Latency Budgets

Midhaul (F1) one-way latency budget:
T_F1 ≈ 1 to 10 ms ⇒ distance ≤ 80 to 200 km over fiber

Fiber propagation delay reference:
t_prop ≈ 5 μs × L_km (n ≈ 1.47, v ≈ 2.04 × 10⁸ m/s)

F1-U user-plane stack:
F1-U = GTP-U / UDP / IP ; F1-C = F1AP / SCTP / IP

Compare fronthaul (DU–RU, eCPRI option 7-2x): T ≤ 100 μs, distance ≤ ~20 km. The CU–DU F1 link is the midhaul segment and is roughly two orders of magnitude more latency-tolerant.

CU, DU, and RU Responsibilities

Node	Protocol layers	Split point	Southbound interface	Latency tolerance	Typical hosting
CU	RRC, SDAP, PDCP	Option 2 (above RLC)	F1 (to DU)	1 to 10 ms	Virtualized, regional edge cloud
DU	RLC, MAC, High-PHY	Option 7-2x (intra-PHY)	eCPRI / Open Fronthaul (to RU)	~100 μs	Cell-site server or local edge
RU	Low-PHY, RF front end	n/a (radio)	Antenna / air interface	Real-time	Tower-mounted radio unit
CU-CP	RRC, control-plane PDCP	E1 (to CU-UP)	NG-C (to AMF)	1 to 10 ms	Centralized control cloud
CU-UP	SDAP, user-plane PDCP	E1 (to CU-CP)	NG-U (to UPF)	1 to 10 ms	Edge or regional, scaled by traffic

Common Questions

Frequently Asked Questions

What is the difference between CU-CP and CU-UP?

3GPP lets the CU split into a Control Plane (CU-CP) hosting RRC plus control-plane PDCP, and one or more User Plane (CU-UP) instances hosting SDAP plus user-plane PDCP. They are linked by the E1 interface, which carries bearer-context setup and modification. The split lets an operator scale user-plane throughput independently of control-plane signaling and push CU-UP instances toward the edge for low-latency traffic while keeping CU-CP centralized.

Which interfaces connect the CU to the DU and the core network?

Southbound, the CU reaches the distributed units over F1, split into F1-C (F1AP over SCTP/IP) and F1-U (GTP-U over UDP/IP). Northbound, it terminates NG-C to the AMF and NG-U to the UPF. When the CU is itself split, E1 links CU-CP and CU-UP, and the Xn interface connects neighboring gNBs for handover and dual connectivity. All are packet-based over routed IP, which is why the CU runs as virtualized software on commodity servers.

How much transport latency can the F1 interface between CU and DU tolerate?

The CU-DU split sits above RLC, so PDCP buffering absorbs delay and F1 one-way budgets of 1 to 10 ms are acceptable, supporting up to roughly 80 to 200 km of fiber. This is the midhaul segment. The lower-layer DU-RU fronthaul (eCPRI option 7-2x) is far stricter, demanding sub-100 microsecond latency and limiting reach to about 20 km. The relaxed F1 budget is what allows many distributed units to be pooled at one centralized CU site.

Related Terms

← CTIA CU (Central Unit) →