Despite their undeniable usefulness, communication networks have been known to experience delays or fail from time to time, thereby making them unreliable for use cases with zero tolerance for failure. PREDICT-6G leverages the synergies of ethernet, cellular, and WiFi, connecting them via a logical overlay to create a reliable, deterministic network capable of delivering Time Sensitive Network (TSN) services. The latter includes the connected industry, which allows for the remote control of devices on a factory floor, thereby improving operational health and safety and reducing OPEX through centralised command and control.
Key to PREDICT-6 G’s value proposition is the creation of a Multi-Domain Multi-technology data plane (MDP) and the artificial intelligence (AI)-driven Control Plane (AICP). The MDP provides the physical infrastructure that stitches together access technologies to ensure end-to-end packet delivery.
Access technologies are analogous to transport systems with different levels of reliability – wired TSN can be considered rail transport with dedicated paths, WiFi would be a country road network devoid of traffic lights, and the cellular 3GPP network would be a well-regulated road network with traffic lights at every intersection. Now, the principal aim of PREDICT-6G can be thought of as delivering a critical item, e.g. a kidney, from a donor in a remote area to a hospital far away for a scheduled transplant just in time across all these different systems.
Making sure this delivery is always on time and arrives reliably, no matter which network it travels through, is what deterministic networking is all about. It means getting guaranteed timing and reliability, like knowing exactly when a train will arrive or that your crucial parcel won’t get lost or delayed unexpectedly.
The goal is to create a seamless, predictable path for important data across these mixed networks. The delivery of the packet must be coordinated such that there are no delays introduced at the intersection of the different communication technologies.
**The MDP**
The PREDICT 6G architecture, shown in Fig. 1, leverages the IETF Deterministic Networking (DetNet) as a kind of universal translator or management layer that sits on top of these different communication networks. DetNet acts as an ‘express service’ overlay – in our analogy, it doesn’t replace the roads or train tracks but manages how high-priority ‘deterministic’ packets travel over them to ensure they meet their timing and reliability needs.
Each technology, therefore, needs to be adapted to meet these requirements.
**Wired TSN**
Wired TSN is already designed for deterministic traffic. It uses standards like IEEE 802.1Qbv for priority scheduling of multiple traffic types and 802.1AS for synchronisation. Additionally, Frame Replication and Elimination for Reliability (FRER) sends multiple copies of a packet to guarantee delivery (any duplicates are removed at the receiver).
**Wi-Fi**
WiFi is less predictable, given the many WiFi-capable devices that share the channel. To enable determinism, PREDICT-6G employs some enhancements. The first of these is Restricted Target Wake Time (rTWT, a feature allowing devices to wake up and sleep at distinct times, hence reducing channel usage). Similarly, time synchronisation aligns wake periods with transmit and receive cycles. The third crucial enhancement is scheduling channel access, thus curtailing unpredictable delays of normal random access.
**3GPP**
3GPP is also a best-effort network and requires adjustments to enable determinism. Key to these is careful radio resource allocation and scheduling traffic to keep delays low and consistent. Secondly, special components called TSN-Translators (DS-TT and NW-TT) are at the edges of the 3GPP network so that 3GPP can act as a TSN switch and connect smoothly with the other TSN-capable networks. Lastly, adapting signal quality settings (prudently selecting the number of bits per symbol transmitted) avoids errors and retransmissions that cause delays. Additionally, using adaptive de-jittering techniques curtails variations in packet arrival times.
**IETF DetNet**
This is the layer that ties these different communication technologies together. Devices or nodes at the intersection of networks (like gateways or routers) have extra DetNet functions. One such function is ‘Forwarding’, which entails sending packets along the right path. Encapsulation and Decapsulation involve enclosing the deterministic data packet inside another packet (like putting a letter inside a special express envelope) when it enters a DetNet part of the network and taking it out when it leaves. This is often done using methods like MPLS tagging or UDP/IP tunnelling. A third crucial function is Packet Replication, Elimination, and Ordering Function (PREOF), which adds robustness by sending multiple copies of a packet over different paths for reliability (Replication), discarding any duplicates that arrive (Elimination), and making sure packets are delivered in the correct sequence (Ordering). This helps smooth out delivery and reduce ‘jitter’ (variation in delay).
This multi-network integration has been successfully tested in real lab setups using commercially available equipment. This demonstrates that it’s possible to make these different networks work together to provide end-to-end deterministic services.
Furthermore, networks or devices that don’t have built-in deterministic features can potentially be included in this system by adding MDP Agents to provide the necessary deterministic capabilities on top.
**The AICP**
The AICP is a smart, central management system that sits above the MDP. Its main goal is to manage these different networks in a unified way to provide guaranteed, end-to-end deterministic services.
The AICP is hierarchical. On the one hand, the Local Management Domain (MD) manages just one type of network technology (e.g. WiFi network or the 3GPP network). On the other hand, the End-to-End (E2E) Management Domain is responsible for overseeing the entire route of a packet across all the different networks.
The AICP creates a simplified, common view of each network domain. Instead of knowing every single road or train track, the E2E level sees simplified connections (like ‘network slices’) between the main entry and exit points of each network domain, along with details about what kind of express service capabilities that domain offers (like guaranteed low delay or high reliability). The Local MDs help provide this simplified view.
Like a complex organization, the AICP has different departments, each with a specific job.
Service ingestion: acts as the front desk where requests for a deterministic service (like sending that important parcel) are received and checked.
Exposure services: are ‘departments’ that gather information from the different network domains (e.g. their topology, available capacity, and special features) and make this information available to other parts of the AICP.
Path computation: is a route planner. The E2E level plans the overall trajectory (which sequence of network domains to use), and the Local MD level plans the specific route within each network domain.
Resource configurator: is the team that actually sets up the transport systems. It tells the wired TSN switches, WiFi access points, and 3GPP network functions how to handle each time-sensitive packet, ensuring it gets the guaranteed service it needs.
Service automation: plays the role of the project manager who oversees the whole process of setting up (provisioning), maintaining, and eventually shutting down (decommissioning) a service. It coordinates all the other departments. It works in a closed loop, relying on control via algorithms, not humans.
Data collection and management (monitoring): gathers real-time performance data (like how long the parcel took, if any copies were lost) from all the networks.
Time synchronization: guarantees that all involved equipment operate simultaneously across the different networks. This is crucial for time-sensitive traffic.
Additionally, the AICP uses AI and Machine Learning (ML), along with Digital Twins (DTs – essentially detailed virtual models of the networks). Digital Twins can help simulate what-if scenarios to determine whether the network should admit new flows. AI/ML can use the real-time data collected and the simulations from the DTs to make smarter decisions about planning routes, allocating resources, and quickly reacting if the service isn’t meeting its guarantees. An MLOps framework helps manage these AI tools.
In essence, the AICP is a sophisticated control system that uses a hierarchical structure, data abstraction, and AI/ML with Digital Twins to manage the complexity of providing guaranteed, deterministic communication services across diverse network technologies like wired TSN, WiFi, and 3GPP. It ensures that critical data packets arrive reliably and on time (no matter the combination of networks they traverse) by intelligently planning, configuring, and monitoring their end-to-end paths.
**What comes next**
In order to meet the demanding key performance indicators of emerging 6G vertical use cases, it will be necessary to harness the capabilities of myriad access technologies, leveraging the strengths of one to compensate for the weaknesses of the other. PREDICT-6G focuses on the connected industry and how devices such as robots can be remotely controlled to eliminate the need to have workers on the factory floor. The deterministic networking features developed by PREDICT-6G can be harnessed to allow such crucial services as remote surgery in remote areas using high precision surgical robots connected via myriad access technologies to a specialist in a remote area. The possibilities are boundless.
Please note, this article will also appear in the 23rd edition of our quarterly publication.
