The Six Thinknet 6G Focus Domains

We expect artificial intelligence (AI) and machine learning (ML) to be used natively in the design and optimisation of the wireless interface, by self-optimising senders and receivers, as well as for new forms of context-aware decision making.

AI and ML technologies have made significant advances in recent years, especially for image classification and computer vision, and are being used in systems as diverse as social media and autonomous driving. 6G researchers are also exploring how to apply AI/ML to wireless systems and networks. The plans for the current cellular generation, 5G, include using AI in at least three different ways:

• To replace some of the layer 1 and layer 2 algorithms (e.g. channel estimation, preamble detection, equalisation and user scheduling) with new AI-based algorithms that are either more performant, less complex, or both.
• In deployment optimisation, for example to configure an optimal subset of beams for a particular coverage area, taking current cell traffic patterns into account. AI/ML techniques will be used to configure and reconfigure the network on-the-fly, thereby optimizing the network without requiring human intervention.
• More accurate end-device localisation based on machine learning

In addition to these three advances in the RAN, AI/ML will also be used for comprehensive end-to-end network automation and orchestration across multiple heterogenous networks and layers. As part of the move from 5G to 6G, the use of AI will move from being an enhancement to being a fundamental element of the air interface design and optimisation.

As 6G is expected to use the 114 GHz to 300 GHz frequency range, in addition to existing spectrum assets in FR1 and dFRA, new frequency technologies need to be developed. These technologies include new high-frequency integrated circuits (RFIC) with integrated antenna arrays and phase shifters, new concepts for hybrid beam forming in combination with MIMO, as well as new receiver architectures. 

One of the major challenges in using these “sub THz” high-bands is the realisation of high-output power devices at reasonable cost. Signals with short wavelengths are easily blocked, diffraction around objects is limited, and signal absorption by water (including humidity levels in the air) is significant. These challenges are partially counterbalanced by NLOS (non-line-of-sight) coverage thanks to reflections from buildings and walls in dense urban and indoor environments. These challenges and idiosyncrasies of high-band propagation necessitate a considerable amount of research and development before deployment is viable.

We expect sub-terahertz bands from 114 GHz to 300 GHz to become available for use as backhaul networks. In addition, narrow point-to-point communication these banks can free up spectrum in mmWave bands. Additional possible uses include short-range communications between display and compute devices, or rack-to-rack communication inside data centres.

In addition to new technologies for the subterahertz bands, low-cost MIMO techniques will enable better use of the mmWave/FR2 and cmWave bands through massive-scale, multiuse MIMO with increased network density.

Research on improved spectrum utilisation in the lower-frequency bands is important since spectrum is scarce and needs to be optimized. We expect 6G operators and service providers will use AI/ML for dynamic spectrum access in time, frequency and space.

6G networks will be used not only as communication networks, but also for sensing and positioning for passive objects. Many of the 6G use cases require precise positioning and localisation, for example in industrial automation or autonomous driving. While real-time kinematics global navigation satellite system (RTK GNSS) can provide highly accurate localisation, it requires good satellite visibility, which is not possible indoors. Currently, indoor localisation is based on ultra-wideband (UWB) or Bluetooth Low Energy (BLE) which require additional access points and devices to be installed. These systems run in parallel to the regular communication system and thus generate additional infrastructure and operations costs, which could be avoided if the regular communication system was also capable of precise localisation.

5G includes capabilities to improve localisation, with a focus on industrial automation environments. As we move towards 6G, we expect the communication network to perform not only localisation, but also additional sensing tasks. The 5G solutions will be enhanced to achieve centimetre-level accuracy, through AI/ML-based channel charting as well as through data fusion from RF, camera and other sensors on robots.

6G system design will be optimized not only for communication but will also include special capabilities for sensing. For example, waveforms that are capable of sensing chirp signals can be multiplexed with waveforms that are optimised for communication. The larger signalling bandwidth in sub-terahertz and terahertz bands can increase precision sensing, possibly within millimetre accuracy. This level of precision enables new applications such as fault detection in manufacturing, or cancer diagnosis.

Combining the precise multi-modal sensing with the cognitive technologies enabled by 6G will allow advances in analysing behavioural patterns and human preferences, creating a “sixth sense” that anticipates user needs and thus permits more intuitive interaction between digital and physical worlds.

The 6G network will be able to automatically and dynamically reconfigure itself to conform to changing requirements, demand and environments.

Some of these automatic adaption techniques were already mentioned above in the discussion about AI/ML. The RAN network will be able to automatically adapt itself to meet the current requirements, for example, via:

• Improved Layer 1 and Layer 2 behaviour for channel estimation, preamble detection, equalisation, and user scheduling
• Optimised beam configuration, to configure the currently-optimal subsets of beams for a particular coverage area, taking into account the current cell traffic patterns and user behaviour

However, dynamic adaption in the RAN is only a small part of the dynamic adaption foreseen for the 6G network. Major changes to the core network architecture are required to enable comprehensive, real-time, end-to-end automation and orchestration across multiple heterogeneous networks and operators.

Sub-networks

Previous cellular generations focussed on extending the voice and data network to individual mobile end points, mainly human beings with cell phones in their hands. 5G is the first generation to extend the focus to the industrial environment, adding new architectural evolutions such as supporting time-sensitive networking (TSN) bridge functionality. 6G will extend 5G to provide full deterministic, wire-grade reliability for a large variety of connectivity scenarios, ranging from static, isolated devices, to interrelated locally interacting devices, to rapidly moving swarms that interconnect to each other and also to the network.

To ensure both high time and high spatial domain reliability and determinism, the 6G network will be divided into multiple semi-autonomous subnetworks. The subnetworks will need to provide at least the most critical services in the subnet, even when connectivity to the wider network is poor or non-existent. To ensure reliability, the network and/or subnets will employ multiple paths and opportunistic device-to-device connections, leading to cell-less architectures. 

Hyper-specialised Slicing

In addition to the subnet requirements and multiply-connected scenarios above, we expect the 6G network to be divided into multiple slices and virtual networks. Slices can be highly specialised and could run separate software stacks for differentiated flow treatment in each slice. The current trend in virtualisation in the RAN, especially at higher layers, will lead to the development of micro RAN services which can then be flexibly composed into slice-specific RAN implementations. For example, a slice for video services could include video optimisation microservices while a slice for low-throughput IoT devices could provide connectionless access.

We expect flexible, slice-specific functions across multiple devices (gateways, relays, cell sites, far edges, edges, regional clouds, …) and across multiple hardware platforms, depending on the needs of each slice. New concepts and innovations are needed to manage and orchestrate such highly specialized slices.

RAN/Core Convergence

In 5G we already see two trends that will affect the 6G architecture. In 5G, many of the “edge” functions in the RAN are moving towards the core, as parts of the Centralized Unit (CU) are virtualized and then implemented in the edge or metro cloud. At the same time, some of the 5G core functions are becoming more decentralized and are being virtualised and then pushed out to the regional or metro clouds, or even to the edge clouds for low-latency services.

We expect these trends to continue, with more RAN functionality moving towards the core, and more core functionality moving towards the edge. We expect to see, within the 6G timeframe, a convergence of functions into a few functional blocks that implement both a RAN and a core, leading to a “coreless” RAN.

A wireless cellular network is, in its essential nature, a highly dynamic network. Devices and their users come-and-go, they move from cell to cell at different speeds, and their data rate requirements rise and fall as they watch a video or take a call. Adding complex subnetting and network slicing, running on hardware and software blocks with diverse functionality, across multiple operators, internationally, requires a level of consensus and automated orchestration far beyond current 5G capabilities.

Performance requirements for 6G networks may include latency below 1 ms and extreme reliability (“nine 9s”). 

The main techniques to achieve low latency and high reliability in 5G are mini-slots and grant-free channel access for low latency, and multiply connected links involving multiple access points, carriers and packet duplication for reliability. However, even 1ms latency over the air is insufficient for some 6G use cases, such as replacing traditional wired connectivity for industrial communication (eg. for EtherCAT). These use cases require significantly lower radio latencies, on the order of 100 µs at Gbit/S data rates. And the target reliability can reach nine 9s for some industrial automation use cases.

6G will be designed to fulfil these requirements by using the wider bandwidth available in the mmWave spectrum to meet latency and datarate specifications. Reliability can be enhanced via simultaneous transmission over multiple paths, involving multiple hops. Cooperative relaying over device-to-device connections can create multiple separate paths from the network to a particular end device. Finally, predictive beam management based on AI/ML prediction can reduce uncertainty in link quality.

Use cases in IoT and Industrial IoT bring an additional form of extreme networking into the discussion, in the form of extreme low-energy or zero-energy end devices. There are several use cases which require low- or zero-energy sensors with extremely long operation times, spread over a wide geographical area. For ongoing sensing and inspection of bridges, for example, wireless sensors embedded during the construction phase should operate for ca. 100 years without human intervention. Solutions for such extreme situations will need to include low-power communication, extreme low-idle current, energy harvesting, and reachable energy storage.

Some of the use cases envisioned for 6G have critically-high security requirements. As the physical and biological worlds merge, the privacy solutions currently available are woefully insufficient. Since multi-modal sensing will capture nearly everything a person does and the entire environment around them, users must be able to restrict what data they share and with whom it is shared. This requires clear and easy-to-understand ways for users to express their preferences for data sharing and for users to maintain complete control over their personal data.

A variety of new signal processing techniques are emerging to address security and privacy in mixed-reality worlds, and these mechanisms will become an integral part of the 6G network.

In industrial applications, replacing wireline-grade reliability with a wireless network creates new requirements for security and privacy for that wireless network. Jamming is a new threat to industrial networks that must be addressed. Attackers could try to jam industrial networks from outside the facility, so physical security alone is inadequate. Jamming can seriously impact industrial operations, particularly in time-sensitive networks. 5G are and 6G networks will be designed to preclude such risks.

The network architecture envisioned for 6G includes segregating applications such as car2x communication, robots, and personal body sensors into separate subnets. This necessitates a change to the authorisation strategy, to move authorisation from the network level down to the subnet level. In a body-area network (BAN), for example, the assets in the subnetwork belong to that subnetwork, and so the authorization and asset management must be handled at that subnet trust boundary. The overall 6G network will connect multiple subnets with each other, so an additional level of authoritarian may be required at the network level. In addition, the subnets might belong to mutual untrusted entities, which calls not only for a clear separation between subnets but also between the network and its subnets. It is critical that subnets act as independent networks, which are empowered as their own authorisation authority and which are responsible for their own subnet asset management. The dynamic behaviour of assets in a cellular network (e.g., continuously joining/leaving the network) makes it a challenge to maintain subnet privacy and anonymity. 

Trust in the network is critical for the success of 6G.