Jim Lewis and Jim Springer were separated at birth as twins. Thirty-nine years later, they met again for the first time and discovered incredible, almost eerie similarities: both had married a woman named Linda in their first marriage, divorced and found a new partner named Betty, both had a dog named Toy as children, both trained as police officers, both liked math at school and spelling the least, both bite their fingernails.
“Jim & Jim” are one of the most fascinating case studies in biological twin research. And they are immensely helpful to me when I have to make small talk at a dinner party or a family celebration and tell them what I'm researching, what the subject of my work in the T-Labs is. Using their exam-ple, I can explain the idea of the "digital twin" wonderfully: A realistic digital image in the computer of complex phenomena of the real world for the purpose of investigating difficult questions - a virtual Jim for the real Jim. With the help of the digital twin, we can then conduct studies much faster, cheaper, and with less risk than in the real world – that's the value proposition of the idea. The first digital twins emerged in the manufacturing industry. Namely, as virtual replicas of 3D ob-jects such as cars or machine parts, which are now used to investigate, for example, weld seams, streamlining or material wear.
Now, when we want to build a digital twin of the telecommunications network – and that is exactly the ambition of our research projects at T-Labs – we immediately stumble across two problems. The first is the distributed nature of our networks and their processual nature. I can photograph a car or an aircraft turbine from all sides and then, as accurately as my computing technology allows, virtually recreate it. I can also "photograph" a net, i.e. record its condition at one time, but the very next moment, when a world record is set at the ISTAF in Berlin and thousands of people send the video of it from the Olympic Stadium all over the world, it has a completely different state. What-ever you want to call a digital network twin must be able to capture the dynamics in the network. This requires models that replicate the assumptions about the driving forces and the applicable laws of motion. They are the Achilles' heel of the digital twin, because assumptions and models are only as good as the current state of research. To come back from the beginning, for example, you will have to ask the real Jim about a new situation again and again in order to recreate it in the digi-tal Jim.
A second challenge lies in the layering of our network: between the electromagnetic wave that leaves the antenna of our smartphone and the result of a search query that appears on its screen a little later, lies a physical world of incredible complexity and range. No computer program will be able to replicate these processes in their entirety. Even the simulation of ten seconds of network activity in a handful of base stations at the level of detail of IP packets currently requires several hours of computing time...
Because all this is so difficult, we have become almost philosophical together with our academic partners (even though "philosophizing" is not explicitly in my employment contract). We discussed questions like these: How would a grid twin be grasped? How can it be narrowed down, cut to size, made manageable? What is its added value for the planning and optimal operation of our net-works?
The first important finding is that the digital network twin must be a modular system of model components: the modeling of wave propagation is decoupled from that of channel selection or IP packets, etc. A precise analysis of the task at hand then leads to the selection of the model compo-nents that are actually necessary: For the optimized positioning of base stations in a campus net-work, one has to model differently than for the operating cost simulation of the Germany-wide network area. To put it bluntly: For the use case that the real Jim wants to buy new clothes online, the digital Jim has to reproduce the correct body shape and can confidently do without the model-ing of his high school grades. When it comes to his skill profile, it's the other way around. School grades and that little paunch beginning to show are to be recorded in different, encapsulated model modules.
The second result of our reflection on the Digital Network Twin is its basic structure. It always con-sists of three ingredients:
- a high-resolution network condition detection,
- the model-based extrapolation of this state into the future
- and a well-defined update strategy for the parameters used.
These are the basic building blocks for our innovation projects, which we pursue together with our research partners. We have already achieved a first result: We can record the energy demand of campus networks in such a way that the two Jims would probably be quite satisfied.
But Jim & Jim also teach us a bit of humility: as astonishing as their similarities are, they remain two different people with their own distinctive inner lives – the study of twins clearly has its limits, in biology as well as in the digital world. But if we acknowledge this and do the rest as well as we and our research partners can, we will have a powerful tool for virtually analysing and optimising our telecommunications networks; it's already emerging.
