Digital twins: what are they and how are they enabling future networks?

In recent decades, what we expect from our devices has changed dramatically. A phone is no longer just for calls or messages, a car likely knows the way to your destination better than you do, and our industries and cities are becoming smarter and more connected by the day, powered by 5G and IoT. We’re seeing our future being built around us – a future of highly complex networks and interconnected digital ecosystems.

With enormous volumes of real-time data constantly being generated by every device, managing these networks and ensuring they operate efficiently and sustainably will depend heavily upon artificial intelligence (AI) and machine learning (ML). But training poses a challenge. After all, how do you safely train an ML algorithm to learn and take responsive action in a live, dynamic network where downtime is not an option? That’s where digital twins come into play.

Accurate representation of an actual site in a digital twin

What is a digital twin?

A digital twin is essentially a copy – a software representation of all the assets, information and processes present in the real-world version, but based in the cloud. Digital twins open up a virtual world of possibility – a safe simulated testing environment in which you can train and play out ‘what-if’ scenarios to your heart’s (or training model’s) content, with no risk to the real-world counterpart.

Exploring digital twin use cases

While they may sound like science fiction, digital twins are already being leveraged in commercial solutions, unlocking the potential of AI, data & digitalization to enable the mobile networks of the future. Here we explore three real-world digital twin examples and discover how this technology is opening up new possibilities for optimized, automated and future-proof networks.

Use case 1 – Network Digital Twins: a safe approach to automation for 5G networks

A Network Digital Twin models what we think of as the invisible network: the signals, coverage, interference and traffic behavior, including user mobility across frequency layers. The digital twin ensures a safe approach to optimization, a vital factor when it comes to sensitive parameters, like radiated power, for example.

In Switzerland, our customer Swisscom runs the top-scoring network in the world according to umlaut’s international benchmark rankings for 2021. But they are also subject to some of the strictest regulations when it comes to radiated power. Without changes to the existing infrastructure, regulations would only allow deployment of a few new 5G sites with low power, leading to spotty coverage on a new low-band layer which would be used by both 4G and 5G New Radio (NR).

So how could we reduce the transmitted power to make headroom for the new layer, without compromising coverage or user experience?

We knew that the best approach would be using reinforcement learning (RL) – a machine learning methodology where an agent interacts with the environment by observing its state and taking iterative actions which gradually converge towards a long-term goal. Our long-term goal was to lower the transmitted power. But we couldn’t allow the agent to play around with the radiated power in the real network, as it could compromise user experience, as well as violate the very regulations which we were working to meet.

Learn more about reinforcement learning and how AI is enhancing customer experience in a complex 5G world

Download the report

Our experts developed an accurate digital twin of the network which modelled coverage, interference and traffic behavior, including user mobility across frequency layers, providing a safe environment in which the RL agent could play and learn. Deep domain knowledge was crucial in selecting what to model, and to find the balance between the twin being accurate and detailed enough, while still being simple enough to be run in a scalable commercial application.

Figure 1: A safe application to reinforcement learning powered optimization through the use of network digital twins

After thousands of rounds of learning, we implemented the final set of recommendations. Almost all the cells in the area had their power changed, both up & down, leading to a 20 percent overall reduction in transmit power with 3.4 percent lower base station power consumption. Interestingly, coverage remained unharmed and user experience actually improved, with 5 percent better download and 30 percent better upload speeds.

It turns out that, in networks, much like conversations in a busy restaurant, shouting louder will only get you so far – but if everyone lowers their voice, we can hear one another better.

The RAN digital twin 2.0 is an Ericsson Cognitive Network Solutions research initiative to create a radio access network live data-driven digital twin, based on big data computation, with multi-network, multi-layer, and multi-RAT support. It offers a modular deployment framework written in python and is based on a mixture of both analytical and ML-based data models. Read more on this initiative in our dedicated eBrief.

Radio access network (RAN) digital twins

Read our ebrief to learn about the importance of applying computational irreducibility to RAN digital twins and what distinguishes Ericsson RAN digital twin from industry-standard simulators.

Download the paper

Use case 2 – Site Digital Twins: a new era for site and equipment management

A Site Digital Twin models the visible network – the towers, equipment and all other assets included in a physical site. As 5G technology accelerates, we need to make sure we can expand and maintain networks quickly and efficiently. But in reality, the lifecycle management of on-site equipment is often far from agile. There can be more than 20 documents outlining what is installed in a single physical site – from CAD designs and images to spreadsheets and product data sheets. This manual documentation makes the process slow and prone to errors, and often ends in unnecessary site revisits and mast climbs.

Figure 2: Site Digital twin

By using a digital twin, we are bringing the IKEA kitchen planner to the telecom industry, enabling fully digitalized site and equipment design and management. We have a single digital twin for each site, with an accurate 3D model captured with laser scanners (LiDAR), cameras and drones. The twin includes all the key metadata necessary for effective and efficient lifecycle management, including constraints such as weight, power and compatibility between components.

We have also developed a 40,000-strong component library, with every component available to easily ‘drag and drop’ into place. This has reduced design time by 50 percent and improved maintenance, reducing the need for site revisits from one in ten to one in one thousand. That’s less travel to the site and less people having to climb masts – for safer, more predictable and sustainable operations overall.

Ericsson Site Digital Twin (ESDT) is an open standards (BIM) based application, which digitalizes telecom sites, physical and logical aspects of all its components, in a 3D environment. It enables analytics/AI/ML in network services. Currently ESDT is utilized as one of the cornerstone capabilities of Ericsson’s Intelligent Deployment in multiple project deliveries across the globe. There is also major interest from customers in different segments, and engagements are ongoing.

Use case 3 – Subscriber Digital Twins: bringing 5G into the Omniverse

We’ve got a twin for the network, we’ve got twins for the sites, but there’s still a key third dimension missing in our digital twin trifecta – the subscriber. Our research team have been collaborating with NVIDIA Omniverse to bring game and movie CGI technology to the telecom industry, enabling the real-time modeling of subscribers using the Unity gaming engine.

This game-changing crossover will involve the evolution of in-house network models with a never-before-seen accuracy in real-world measurements. Essentially, we’re taking 3D gaming technology – and its extremely high computational complexity of physically accurate models – as a baseline, then deploying propagation models for 5G on top.

Welcome to the Omniverse: Ericsson’s radio network simulation expertise meets NVIDIA’s technologies in rendering and collaborative design.

The technology allows high-resolution complex city and indoor geometry for modeling, including bridges, tunnels, foliage and the detailed modeling of surface materials that influence radio frequency (RF) propagation, and modeling of the mobility of users and dynamic scene features such as automotive traffic. It also uses Pixar’s open Universal Scene format, which enables reuse of detailed city meshes & geodata, which is sometimes one of the biggest challenges to model an environment accurately.

As we know, future networks will only become more complex, so models will need extensive visualization support to be meaningful. NVIDIA Omniverse Create integrates a state-of-the-art ray tracing engine with the interactive tools to manipulate and explore complex scenes, allowing us to experiment with the placement of Ericsson products and explore their impact in real time – a true enabler for top-performing product development.

The future of 5G is bright – and brimming with virtual worlds full of possibilities.

Learn more

Discover Ericsson AI-driven cognitive network software and solutions

Find out more about AI and reinforcement learning in telecoms.

Discover how AI is applied to achieve efficiency and performance in networks

See what else is possible with Ericsson’s intelligent site engineering.

Explore our collaboration into 5G simulation on the Omniverse platform.

Discover how digital twins are modernizing the oil and gas industry and transforming port operations.

Read more about the future of digital twins in mobile networks in our blog post.

Read more about Digital twins: bridging the physical and virtual worlds