Fifth Generation Fixed Network (F5G); F5G Advanced Network Architecture Release 4

I. Background of F5G-A Release 4 Architecture Evolution

The F5G-A (Fifth Generation Fixed Network Advanced) Release 4 specification released by ETSI ISG F5G marks a significant evolution of fixed network architecture towards application-driven, computing-network convergence, and deep intelligence. Compared to Release 3, this version, for the first time, incorporates Data Center Network (DCN) and Data Center Interconnect (DCI) into the end-to-end architecture and introduces a cross-plane computing power collaboration mechanism, enabling end-to-end adaptation to differentiated SLAs from user terminals to cloud services.

The newly added features, such as fiber optic sensing, Agentic AI services, P2MP fgOTN, and wavelength-shared WDM architecture, enhance the network's ability to support industrial digital transformation. This architecture, centered on service elasticity and resource efficiency, achieves multi-layered collaboration and automated closed-loop management by decoupling the Underlay plane (physical connectivity), Service plane (service processing), and MCA plane (management, control, and analysis). This design not only meets the needs of flexible, on-demand service packages (such as cloud desktops, XR live streaming, and industrial control), but also provides granular QoD (Quality on Demand) assurance for each application stream through an AI-enabled FAN (Fibre Access Node) engine.

---

II. Overall Architecture: Three Planes and Cross-Plane Computing Power

The F5G-A network architecture consists of an Underlay Plane, a Service Plane, and an MCA Plane, and adds cross-plane computing capabilities, forming a four-in-one intelligent all-optical base.

As shown in Figure 1 (Figure 1 is an overview in the text), the Underlay Plane provides physical connectivity and tunneling, comprising five segments: CPN (Customer Premises Network), AN (Access Network), AggN (Aggregation Network), CN (Core Network), and DCN (Data Center Network), with each segment capable of integrating computing resources. Its protocol stack is extremely simple, supporting large-scale switching without interfering with the service plane. The Service Plane implements L2-L7 business chain processing through SAP (Service Access Point), SPP (Service Processing Point), and SMP (Service Mapping Point), and can dynamically map traffic from different applications to appropriate connections in the Underlay Plane, such as mapping enterprise leased lines to (fg)OTN pipelines, or mapping ordinary Internet traffic to IP/Ethernet structures. Service processing can be combined with AI computing power, such as identifying cloud desktop protocols in real time on the OLT and prioritizing them. The MCA Plane introduces logical components such as digital twins, intent engines, autonomous control, and generative AI analysis to form the core of intelligent operations. The QoD API exposes network capabilities to the BSS system, supporting on-demand requests in the NaaS model; the Autonomous Engine is responsible for closed-loop control of the entire lifecycle of resources and services. Cross-Plane Computing is a key innovation of R4. Computing power can be distributed across CPN's FTTR master equipment, AI cards at the AN central office, local data centers at the AggN edge, and core cloud CDCs. Collaboration between computing power is achieved through the Can, Cag, and Ccn interfaces, supporting AI training and inference, intelligent fault prediction, and real-time service optimization. For example, the AI voice assistant at the site can interact with the large model agent of the access node to achieve low-latency smart home services.

---

III. Network Topology and Interface Details

Figure 6 shows the full topology of R4, the complete path from user equipment to the cloud. The key components and interfaces of each network segment are described below.

3.1 Customer Premises Network (CPN)

For different scenarios, CPN supports multiple deployment options:

• FTTR Solution: Multiple SFUs (sub-fiber units) are connected by a passive optical splitter from an MFU (main fiber unit) to achieve whole-house fiber + Wi-Fi 7 coverage.
  The F1 interface carries ITU-T G.9941/G.9942 or the future G.Xfin series, with centralized control achieved through the FMCI/WMCI protocol of the F1 interface. FTTO Solution (Figure 7): Suitable for campuses/office buildings, it can adopt a PON method of FTTO-OLT + FTTO-ONU, or an OTN access method of (fg)O-CPE, the latter providing hard-isolated fine-grained leased lines.
• FTTM Solution (Figure 8): Targeting industrial sites, the FTTM-ONU supports industrial protocols such as EtherCAT and serial ports. Its T' interface can be converted to PON or OTN bearer.

3.2 Access Network (AN)

The main technologies are 50G-PON (G interface, ITU-T G.9804.2/3) and P2P/P2MP (fg)OTN (H interface).

The OLT can be integrated or externally coupled with AN Compute functionality, connected via a C interface (Ethernet). ROADM technology enables access nodes to connect purely optically to the λ fabric, achieving dynamic wavelength scheduling.

3.3 Aggregation Network (AggN)

The aggregation network provides three parallel fabric structures to meet the carrying requirements of different types of services. The following table compares the core differences between these three structures:

Features	IP/Ethernet Fabric	OTN Fabric	λ Fabric
Switching Granularity	Packet Data	ODU/fgODU (10M~800G)	Wavelength
Quality of Service	Statistical multiplexing, best-effort	Strictly guarantee bandwidth, latency and isolation	Physical isolation, extremely low latency
Typical Services	Ordinary residential broadband	Enterprise leased lines, mobile backhaul, high-value services	Ultra-low latency private networks, wavelength wholesale
OAM capabilities	Rich, but based on competition	Carrier-grade fine OAM	Simplified, relying on optical layer digital tags
Energy efficiency	Medium	High (reduced photoelectric conversion)	Highest (all-optical switching)

The aggregation node (AggN Edge) interacts with the core network through V (IP), Vo (OTN), and Vλ (wavelength) interfaces, and can connect to the local data center (LDC) via K/L interfaces to provide edge computing power for latency-sensitive applications.

3.4 Core Network (CN) and Data Center Network (DCN)

The core network adopts all-optical OTN and λ switching, supporting 400G~1.2T single wavelength. The DCA interface connects to the data center gateway, while the DCI interface enables cross-DC interconnection, requiring zero packet loss protection (1+1 hot standby) and RDMA congestion control. The data center intranet (AI DCN) tends towards OXC (optical cross-connect) all-optical networking to reduce power consumption and support millions of GPU interconnects.

---

IV. Key Enabling Features Explained

4.1 AI-Enabled Fixed Access Network Architecture

This feature extends R3's FIE (FAN Intelligent Engine), introducing AI computing power into the OLT and CPN nodes to form an application-aware, fine-grained flow control system. Core components include:

• QoD API: Provides on-demand quality prediction, user self-service portal, and real-time SLA control. Allows cloud service providers to request network performance for specific application flows.
• App-Flow: Goes beyond traditional slicing, automatically identifies applications (such as video conferencing and industrial control) through AI, and creates flows with individual QoS guarantees for them, supporting dynamic bandwidth adjustment and failover.
  **Agentic AI Service:** Supports multiple AI agents working collaboratively across FTTR nodes, OLTs, and the cloud. For example, a home AI agent handles local voice commands, while complex tasks are seamlessly offloaded to edge or central AI, while the FTTR network provides deterministic latency for inter-agent communication. **Self-Optimizing Network:** Utilizes AI to learn Wi-Fi coverage patterns and traffic patterns, automatically adjusting channels and power, and combining optical layer digital tags to achieve precise ODN fault location. **Figure 22 shows the AI Access Node architecture. The OLT service system imports data plane information into the computing service platform through the Export interface, where the AI framework performs in-depth analysis, and then issues optimization commands via the OLT Service interface.

  4.2 Fine-grained Optical Transport Network (fgOTN) Structure

  OTN structure is upgraded to (fg)OTN in R4. Its key advancements lie in supporting 10Mbit/s granular fgODU containers and lossless bandwidth adjustment, greatly improving bandwidth utilization and service flexibility. Simultaneously, the newly introduced P2MP fgOTN technology allows for the construction of OTN leased lines on point-to-multipoint ODNs, reducing the complexity and cost of leased line access for SMEs. At the control level, the C1 interface is responsible for the automatic creation, adjustment, and rapid recovery of (fg)OTN network connections (rapid rerouting of a large number of connections), while the C2/C2' interfaces implement MAC/IP address learning and mapping rule distribution on the service plane, enabling SMP to correctly encapsulate user traffic into the corresponding fgODU connections. Combined with the OCN (Optical Cloud Network) protocol stack, dedicated channels from users to multi-cloud services can be dynamically established. 4.3 Wavelength-Shared WDM Structure This structure utilizes ROADM technology (especially colorless ROADM and SC-FDM-based subcarrier multiplexing) to construct a multi-ring interconnected mesh aggregation layer. As shown in Figure 26, the M*N WSS node supports wavelength sharing between different access rings: subcarriers can be used for P2MP traffic (such as OLT uplink), while single carriers meet the needs of P2P leased lines. Optical digital tag technology enables end-to-end automated deployment, plug and play, greatly reducing operation and maintenance costs.

  4.4 Fiber Optic Sensing Architecture

  Figures 29 and 30 show two sensing modes: dual-end active measurement and single-end reflection measurement. By analyzing changes in Rayleigh scattering or Fresnel reflection in the optical fiber, the system can detect environmental parameters such as vibration, temperature, and bending, and is used for railway perimeter intrusion detection (use case #5), pipeline monitoring, and its own fiber health management. Sensing data can coexist with communication traffic on the same optical fiber, enabling facility reuse.

  4.5 PON Industrial Network In the FTTM scenario (Figure 31), the PON system is extended into an industrial intranet, supporting network slicing, deterministic latency (target <0.5ms), redundancy protection, and local edge computing. The FTTM-ONU provides various industrial interfaces (such as EtherCAT and RS485) and is adaptable to harsh environments. This architecture conforms to the ETSI GS F5G 022 specification and communicates with the FTTM-OLT via the M interface. The latter can simultaneously connect to the FTTO network in the office, forming a unified all-optical factory network. V. Implementation Recommendations and Business Value When adopting the F5G-AR4 architecture, operators should focus on the following points: **Segmented Evolution:** First, upgrade the access network to 50G-PON and (fg)O-CPE, then gradually introduce OTN structure or wavelength-shared WDM based on business needs, ultimately achieving end-to-end slicing. **Computing Power Collaboration:** Initially, small AI modules can be integrated into the OLT to achieve application recognition, later expanding to edge LDC and connecting with the cloud AI platform to build distributed intelligence. Automated Operations: Prioritize the deployment of digital twins and intent engines, and reduce manual intervention and shorten service activation time by opening up capabilities through NaaS/QoD APIs. Energy Efficiency Optimization: Utilize hibernation mechanisms and all-optical networks to reduce photoelectric conversion, meeting green standards; simultaneously, use fiber optic sensing to reduce line inspection costs. From a business perspective, this architecture enables operators to launch flexible "package + X" services, such as home AI safes, cloud private lines for SMEs, and industrial robots as a service (RaaS), opening up new revenue streams. At the same time, the minimalist Underlay protocol stack and automated maintenance significantly reduce OPEX.

  ---

  VI. Summary and Outlook

  ETSI GS F5G 036The defined R4 architecture is a key milestone in the transition of fixed networks to computing-network convergence and service-oriented operation. Through three-plane decoupling, cross-plane computing power injection, and full-scenario optical technology, it provides a deterministic, low-latency, and highly reliable connection foundation for various industries. In the future, with the maturity of 50G-PON and fgOTN, and the lightweight deployment of AI big data models in set-top boxes and home gateways, F5G-A will truly realize a ubiquitous intelligent all-optical network.