While the use of AI and digital twins isn’t new to operators, the evolution of the former is enabling advanced twinning use cases such as improved network planning and optimisation, 6G research frameworks and training models.
A digital twin is the generation or collection of digital data representing a physical object. Most digital twins use a combination of AI and ML, sensor data and real-time 3D to generate a digital simulation of any physical entity.
With AI algorithms and closed-loop automation, operators can create AI-powered digital twins that solve complex, real-world problems at a lower risk level and with faster results.
UScellular CTO Mike Irizarry stated the operator is working with a partner to use generative AI (GenAI) to create digital twins of its cell towers.
Instead of climbing each tower or consulting tower diagrams, drones will capture images and put them in a database that is then pumped into a large language Markov model specifically tuned and optimised for image classification.
The end goal is to have a digital twin of the operator’s cell towers that will allow a person with a VR headset to correlate whether a tower’s antennas are off-line due to storms or other factors.
“It allows you to dispatch someone with the right equipment to fix it in a shorter period of time and much more cost effectively,” Irizarry explained. “That’s an application of generative AI and digital twins we’re looking at, but we’re not there yet”.
Like Google Maps, he noted GenAI and digital twins could also be used to monitor and fix data centres by creating 3D renderings of them.
“I could look at heat and power consumption in a specific data centre without actually visiting it,” he explained.
AI, digital twins for 6G research
Nvidia unveiled a platform to democratise 6G research by employing AI and digital twins. The chip vendor’s Nvidia 6G Research Cloud platform has several interconnected components that allow vendors, researchers and operators to test AI algorithms on its Aerial Omniverse Digital Twin for 6G platform.
Ansys, Arm, ETH Zurich, Fujitsu, Keysight, Nokia, Northeastern University, Rohde & Schwarz, Samsung, SoftBank Corp. and Viavi are the first members of the ecosystem.
The partners will be able to take advantage of powerful Nvidia chips to create digital twins of neighbourhoods or entire cities.
“Omniverse is a physics accurate, highly visual enhanced digital twin capability that scales from within a server or rack or a chip all the way to Earth,” explained Ronnie Vasishta, SVP of telecom at Nvidia. “We’re creating a digital twin of the Earth”.
Using ray tracing, Vasishta stated Omniverse can simulate the RF part of spectrum to help operators determine where to place their antennas and terminals while also simulating the location of moving mobile devices.
It also allows operators and vendors to accurately simulate the propagation of RF in areas where there’ is impediments such as glass, concrete or foliage in a digital twin, he stated
“You can see the effects of reflections, refractions, diffractions and absorptions of the RF at the various frequency bands,” Vasishta said.
The digital twin is designed to work across a single base station with a few mobile devices all the way up to hundreds of base stations with thousands of devices.
“I would say the RF propagation piece is perhaps one of the most exciting areas apart from the data collection,” Vasishta noted. “The ability to simulate at scale real antenna, including interface interference and other elements, data is where we’ve really spent the most time to make sure that it is an accurate implementation”.
The platform also includes a software-defined, full RAN stack to allow researchers and members to customise, programme and test 6G network components in real time. Vendors, such as Nokia, can bring their own RAN stack to the platform, but Nvidia’s open RAN compliant stack is provided.
Vasishta added users of the research platform can collect data from their digital twin within their channel model, which allows them to train for optimisation.
“It now allows you to use AI and machine learning in conjunction with a digital twin to fully simulate an environment and create site specific channel models so you can always have best connectivity or lowest power consumption, for instance,” he said.
He noted the platform could include using GenAI to create predictive models from operators’ data.
“Initially, you’ll need ground truth data from operators to validate what you’re collecting,” he said. “Operator data will be useful to enable you to build that trained model, but you can create synthetic data after that, and then you can use it to train the model itself”.
Network planning and testing
Aside of 6G research, the platform also enables network planning by letting operators run simulations of large events such as concerts or autonomous vehicles driving around the streets.
“The models that you’re training in your simulation environment will then be deployable,” Vasishta said. “Maybe you bring in an extra mobile base station or allocate more compute resource to the 5G core to service that particular area”.
Ian Wong, director of RF and wireless architecture for Viavi, stated his company realised early on the importance of using AI and ML for developing its 6G Forward Programme, which includes a partnership with Northeastern University.
The test and measurement vendor is using AI and ML to augment ray tracing for propagation modelling in a digital twin.
“We’ve used ray tracing for a long time in the industry to model the RF propagation environment,” Wong said. “But if you bring that up to a specific city scale you cannot just do brute force ray tracing. We see that as needing physics-informed AI that allows us to have accurate enough models that ultimately show operators how to optimise their networks.”
He stated Viavi uses the university’s open RAN digital twin, dubbed Colosseum, combined with AI to get the “ground truth” data for radio propagation modelling in a lab environment.
Tommaso Melodia of Northeastern University stated the AI-RAN Alliance, which includes Nvidia, Softbank and the university, is putting together a multivendor, open-RAN and 3GPP compliant blueprint for 6G using AI over the coming months. That blueprint could include work from Nvidia’s 6G Research Cloud platform.
Jane Rygaard, head of corporate partnerships for Nokia, stated the first goal for the digital twin used in Nvidia’s 6G platform is to collaborate across the industry, but she also noted it allows members to focus on specific areas of interest.
“Before we start building real [6G] systems, we need the collaboration that leads up to standardisation. We need to understand the complexity of all these systems together. That means things like twinning must be done in a completely different manner because we can no longer only look at it from a little lab in the corner”.
El artículo original puede consultarse en: