Network & Telecom

Core Network Evolution

Q: What is the difference between the LTE EPC and the 5G core?

The LTE Evolved Packet Core (EPC) uses a fixed set of network nodes (MME, S-GW, P-GW, HSS, PCRF) connected by point-to-point reference-point interfaces such as S1, S5/S8 and S11, with the S-GW and P-GW combining control and user-plane functions. The 5G core (5GC) decomposes these into modular network functions (AMF, SMF, UPF, AUSF, UDM, NRF, NSSF), where the control-plane functions are designed to be stateless by externalizing session state to a shared data layer and expose RESTful service-based interfaces over HTTP/2 that register with a Network Repository Function. Crucially, the 5GC fully separates the control plane (SMF) from the user plane (UPF, a stateful packet forwarder), so the UPF can be pushed to the network edge to reach the sub-1 ms latency targets that flat EPC topologies could not meet. The 5GC is also designed cloud-native from the start, running as containerized microservices, whereas EPC virtualization was retrofitted onto purpose-built node designs.

Q: What does CUPS mean and why was it introduced before 5G?

CUPS stands for Control and User Plane Separation, standardized by 3GPP in Release 14 (2017) as an evolution of the existing LTE EPC. Before CUPS, the S-GW and P-GW each bundled control-plane signaling and user-plane packet forwarding in one node, so scaling data throughput meant also scaling expensive control logic. CUPS split each into a control entity (SGW-C, PGW-C) and a user entity (SGW-U, PGW-U) communicating over the Sx interface using the PFCP protocol. Operators could then place lightweight user-plane forwarders close to the radio edge to cut latency and backhaul load, while keeping centralized control. CUPS was the architectural bridge that proved the model later formalized in the 5G core as the SMF and UPF split.

Q: How does network slicing depend on the evolved 5G core?

Network slicing creates multiple logical end-to-end networks on shared physical infrastructure, each tuned to a service class such as enhanced mobile broadband, ultra-reliable low-latency communication, or massive machine-type communication. This is only practical on the service-based 5G core because slices are instantiated as independent sets of virtualized network functions selected at registration time. The Network Slice Selection Function (NSSF) chooses the slice based on the Single Network Slice Selection Assistance Information (S-NSSAI) the device presents, and the AMF routes the session to slice-specific SMF and UPF instances. A URLLC slice might pin a UPF at the cell-site edge for 1 ms latency, while an eMBB slice uses a centralized UPF sized for 10 Gbit/s per user; the rigid EPC could not host these divergent profiles simultaneously.

/kor net-werk ev-uh-loo-shun/

Spanning four mobile generations, this is the generational redesign of the packet core that sits behind the radio access network, anchoring subscriber sessions, authentication, mobility, and policy. It tracks the migration from the circuit-switched GSM core, through the all-IP LTE Evolved Packet Core, to the cloud-native, microservices-based 5G core built on a service-based architecture. Each step traded rigid, node-centric topologies for software-defined functions that separate the control and user planes, virtualize onto commodity servers, and push the user-plane forwarder toward the network edge. That trajectory is what makes sub-1 ms latency, 10 Gbit/s per-user throughput, and per-service network slicing technically achievable on shared infrastructure.

Category: Network & Telecom

Span: 2G GSM → 5G SA

Key 3GPP releases: R99, R8, R14, R15

From Circuit Switching to a Cloud-Native Packet Core

The mobile core began as a circuit-switched system. The GSM core paired a Mobile Switching Center (MSC) with Home and Visitor Location Registers to set up dedicated voice circuits, with the GPRS packet domain (SGSN and GGSN) bolted on later for data. With LTE (3GPP Release 8, 2008) the core became flat and all-IP: the Evolved Packet Core (EPC) introduced the MME for control signaling, the Serving Gateway and PDN Gateway for user-plane bearer forwarding, the HSS subscriber database, and the PCRF for policy. Voice moved off circuits and onto IMS-based Voice over LTE. The EPC removed a hierarchy of nodes and cut control-plane latency, but its gateways still bundled control and data forwarding in monolithic, often purpose-built hardware.

3GPP Release 14 (2017) introduced Control and User Plane Separation (CUPS), splitting each gateway into a control entity and a user entity linked by the Sx interface and the Packet Forwarding Control Protocol (PFCP). This let operators distribute lightweight user-plane forwarders near the radio edge while keeping control centralized, reducing backhaul and latency. Release 15 (2018) then defined the 5G core (5GC) as a set of modular network functions, AMF, SMF, UPF, AUSF, UDM, NRF, and NSSF, whose control-plane functions are designed to be stateless by externalizing session state to a shared data layer. The control-plane functions expose RESTful service-based interfaces over HTTP/2 and discover one another through a Network Repository Function, while the UPF remains a stateful user-plane forwarder. The 5GC is cloud-native by design, deployed as containerized microservices orchestrated on commodity x86 or Arm servers rather than carrier-specific blades.

This evolution is why the core network is now treated as software. Separating the SMF (control) from the UPF (user plane) means an operator can pin a UPF at a cell-site data center for ultra-reliable low-latency traffic while routing bulk broadband through a regional UPF, all from one logical core. It is also the foundation for network slicing, where the NSSF selects an isolated set of functions per service class. For an RF and antenna-systems supplier, this matters because the evolved core dictates the latency budget, synchronization tolerance, and fronthaul/backhaul interfaces that radio-side hardware must meet.

Core Capacity and Latency Relationships

End-to-end one-way latency budget:
T_e2e ≈ T_RAN + T_transport + T_UPF + T_internet

Edge UPF placement (propagation contribution):
T_transport ≈ d × (n / c) ≈ d × 5 µs/km (fiber, n ≈ 1.47)

Control-plane signaling load:
L_sig = N_UE × R_events × M_msg (transactions/s)

Where T_RAN = radio scheduling/processing delay, T_UPF = user-plane forwarding delay, d = UPF distance, n = fiber refractive index, c = speed of light, N_UE = attached devices, R_events = mobility/session events per device, M_msg = messages per event. Example: a UPF 5 km from the cell adds ≈ 25 µs of fiber latency, so a 1 ms URLLC target demands edge placement, not a regional core 200 km away (≈ 1 ms one-way on fiber alone).

Generational Comparison of the Mobile Core

Generation	Core name	3GPP release	Switching model	Control/user plane	Typical latency target
2G	GSM core (MSC/SGSN)	Phase 2 / R97	Circuit + early packet	Combined	~300 to 600 ms
3G	UMTS core (MSC/SGSN/GGSN)	R99 to R7	Circuit + packet	Combined	~100 to 200 ms
4G	Evolved Packet Core (EPC)	R8	All-IP, flat	Combined in gateways	~30 to 50 ms
4G+	EPC with CUPS	R14	All-IP, distributed	Separated (Sx/PFCP)	~10 to 20 ms
5G NSA	EPC + 5G NR (Option 3)	R15	All-IP	EPC-anchored	~10 ms
5G SA	5G core (5GC, SBA)	R15 / R16	Cloud-native, service-based	Fully separated (SMF/UPF)	< 1 to 5 ms

Common Questions

Frequently Asked Questions

What is the difference between the LTE EPC and the 5G core?

The EPC uses fixed nodes (MME, S-GW, P-GW, HSS, PCRF) joined by point-to-point reference interfaces such as S1, S5/S8 and S11, with gateways bundling control and user-plane functions. The 5GC decomposes these into modular network functions (AMF, SMF, UPF, AUSF, UDM, NRF, NSSF), where the control-plane functions are stateless (externalizing state to a shared data layer) and expose RESTful service-based interfaces over HTTP/2 that register with an NRF. The 5GC fully separates the SMF (control) from the UPF (a stateful user-plane forwarder), so the UPF can move to the edge for sub-1 ms latency, and it is cloud-native from the start rather than retrofitted.

What does CUPS mean and why was it introduced before 5G?

CUPS is Control and User Plane Separation, standardized in 3GPP Release 14 (2017). It split the EPC S-GW and P-GW into control entities (SGW-C, PGW-C) and user entities (SGW-U, PGW-U) communicating over the Sx interface using PFCP. Operators could place lightweight user-plane forwarders near the radio edge to cut latency and backhaul, while keeping control centralized. CUPS proved the architectural model later formalized as the SMF and UPF split in the 5G core.

How does network slicing depend on the evolved 5G core?

Slicing builds multiple logical end-to-end networks on shared infrastructure, each tuned to a service class such as eMBB, URLLC, or mMTC. It is practical only on the service-based 5GC because slices are sets of virtualized functions selected at registration. The NSSF picks the slice from the S-NSSAI the device presents, and the AMF routes the session to slice-specific SMF and UPF instances. A URLLC slice can pin a UPF at the cell edge for 1 ms latency while an eMBB slice uses a centralized 10 Gbit/s UPF.

Related Terms

← Core Material (Radome) Core Network Slicing →