5G has brought us to the age of softwarisation: by moving parts that could only be managed physically to the cloud, operators can now perform tasks such as network slicing and digital twinning, something they could not do some years ago. And that new era will be critical for 6G.

At least, this is what Chris Dick, Wireless Architect at chipset manufacturer NVIDIA, envisions for the future of telecommunications. In a session during the 6GSymposium Fall 2023, he shared some scenarios the company is working with.

According to the expert, the International Telecommunications Union (ITU) expanded three fundamentals from 5G to 6G (MBB, UILLC, and MMTC) and added three more (ubiquitous communication, integrated sensing and communication, and native AI in wireless) to support evolving use cases. Some examples are interconnected intelligent machines, the internet of senses, and the digitalisation and programmability of the physical world.

“We believe that the key enabler for bringing [these fundamentals] to reality in a cost-effective way is that everything is software-defined,” he said.

“The whole network is software-defined RAN. Machine learning is a fundamental principle in all layers of the wireless network, and digital twins are a central part of the deployment, operation, and optimisation of the network.”

Chris explained that the thinking at NVIDIA is that everything is “all-software.” Not just software running on CPUs but on what they call accelerated compute – GPUs. “This is the key to going beyond Moore’s Law and Cooper’s Law.”

While Chris acknowledges that some people might be critics of machine learning and wireless because fundamental limits are known, and the industry is already close to them, there are too many good ideas in the making to ignore.

“We’re kind of limited by the typical approach to the construction of systems [like Gaussian and Stationarity],” he said. “We can pick up further gains in energy efficiency and compute density just through the parallelisation in neural networks you can do on current generation GPUs.”

A World That 5G Can’t Deliver Today

According to the NVIDIA expert, the road to 2030 networks should include the following steps:

• Going from a traditional receiver architecture and being able to replace some of the blocks with machine learning equivalents.
• A second stage, in which it will be possible to take subsystems of a channel estimator, channel equaliser, and soft mapper, for example, and combine them into a neural network.
• And the era of 6G, when we should have machine learning design the physical layer itself.

This evolution path will enable industries to operate in a way they cannot today. “We can’t do end-to-end optimisations if we want to encompass learning that extends from the base station through the channel,” Chris said.

“When we are in the 6G era, we might be able to open the door and exchange models between the base station infrastructure equipment. The user equipment can’t do that in 5G,” he added.