Werner Weiershausen
The 2025 Optical Fiber Communications Conference and Exhibition (OFC) will take place from 30 March to 3 April in San Francisco, CA, USA. This year’s conference will feature a Market Watch, Network Operator Summit and Data Center Summit on the exhibit floor. Ahead of the event, OPN spoke with the subcommittee chair for these programs, Werner Weiershausen, Deutsche Telekom Technik GmbH, Germany, to get a preview of what they will entail and his thoughts on the industry more broadly.
What is your role at Deutsche Telekom?
I'm a lead architect for transport networks and platforms at Deutsche Telekom, Germany. My role focuses very much on the evaluation of next-generation networks to be deployed by Telekom, regarding, for example, IP networks and some ethernet or optical networks, as well as wavelength-division multiplexing and optical transport networks (OTN) and services. So my work is not only related to production on the platforms, but also to the services, end-to-end, for certain customers.
Can you tell us a bit about the programs you helped organize for OFC?
We followed a democratic process at the beginning to decide what would be the best program, enlightening different aspects of network operation. One area we decided to focus on is the data center domain. Another focus is the networking domain from incumbent carriers, and also from hyperscalers. Hyperscalers are in the middle of both, having data centers and also telecommunications. In addition, we plan to discuss market watch.
“... the overall focus will be on industry trends, the direction the market is going, and then bringing all of these topics together. I think we have made up a very nice, interesting program.”
—Werner Weiershausen
We will have two panels during the Network Operator and Data Center Summits and keynotes at the beginning of each session. And we have a big Market Watch section with six panels spread over different days. There, we will tackle parts of the topics I mentioned before, but the overall focus will be on industry trends, the direction the market is going, and then bringing all of these topics together. I think we have made up a very nice, interesting program.
What do you think are some of the overarching trends in network operators?
One thing to look at is the overall drivers for traffic increases in our networks, and the need for future technologies that enable these high bandwidth demands. And I think there we see different things: On one side, you have the incumbent telecommunications providers, and on the other, the data centers and hyperscalers, like Google. And it is interesting how things change very quickly, compared with in the past. It’s all about AI now.
Typically, with customer traffic behavior we always had this 35% annual growth. Now, this growth has reduced to lower rates. People aren’t using any new killer applications yet. For example, they don't wear special AR or VR glasses or whatever, and so this trajectory between residential customers and our network is flattening, at the time being. On the other hand, though, we see a strong worldwide increase—more than the 35% per year—in traffic between data centers, related to AI and machine learning. There is a lot going on with how these two worlds are evolving right now.
Another topic, which is very important for my own business, is IP optical integration. Today, we have classical WDM systems, we have an OTN network and we have IP on top. In some parts of the network, we now integrate IP and the optical transport by the application of pluggable optics directly into the routers. But the main problem here is the carrier-grade end-to-end management, which we are solving with proprietary multi-vendor overarching solutions.
For the future, we are striving for an open ecosystem based on standard interfaces and architectures. However, this is not yet available for deployment, as neither the needed system vendors nor the major standards have achieved a mature status to reach this target. But the overall community trend is going in this direction, and we see progress in worldwide trials as well as in our own proof-of-concept tests at DT. Within such an open ecosystem, you will be able to plug and play different optics from different vendors in a nonproprietary way.
So, for example, you can use optical plugs from one company that are independent from IP router vendors, and transmit the data via an optical line system (with optical amplifiers and ROADM switches) from another company. And you can also mix the transmitter from one company with the receiver of another. With this, customers benefit from a broad optical transceiver market, thus saving costs. Moreover, skipping grey client interfaces by directly plugging the colored long-haul interfaces into the IP routers allows you to further save investment costs and power.
What are other barriers to this open ecosystem?
I think the physics is easy with the technical achievements of today. The barrier is the control and management plane, but really this is more or less due to companies’ strategic positions—because this could also be done quite easily in principle. Technical solutions are there, but the standards have to be driven further to reach a mature and overarching momentum for the whole industry.
“For the future, we are striving for an open ecosystem based on standard interfaces and architectures. ”
—Werner Weiershausen
Do you predict new killer applications for the mass market, driving bandwidth demand?
Well, where will the killer applications come from? This is very difficult for us to answer. One option would be, for example, consumers having these AR or VR glasses. Then in combination with AR/VR, there is the discussion about the Metaverse application. We saw hype around this in the beginning, but then people backed off from that and said no, it was only hype. But now you hear new voices that this is again getting traction.
Does anything from this come to fruition? Is this the future for us? If so, it will drive the traffic demand, and we will need new optical interfaces, and even may need to install new fiber types, like multi-core fibers for spatial multiplexing, to really carry that traffic. When will this need to arise? We really don’t know; this is an unknown variable. This makes it hard to plan for the future.
So it's still a difficult market where it's not opening as fast as it should, but we’re working on this.
What do you think are some needs of the market, going into 2025?
One thing to watch is all of these AI applications and AI support. There will be a lot of pressure to develop technology that is low power consuming. Interfaces are getting lower and lower with power consumption per bit, but it's not getting lower in total because you have this increase of traffic. And I think the data centers have some strategies to use cheaper energy or renewable energy sources, but in total, we still see an increase.
Also, these interfaces are evolving, not only with regard to power efficiency but also with respect to their speed in Gbit/s increasing with each generation. However, this does not mean that we can relax the bandwidth limit of the transmission fiber, as we have already reached the limit placed by the Shannon–Hartley theorem. That means the product of distance versus bandwidth has hit a certain physical limit. If you want to increase the speed, respectively broaden the bandwidth of an optical interface, you may need to shrink the transmission distances. But in most cases, network topologies can’t or won’t be adapted.
An alternative solution is to extend the bandwidth so that the frequency window per interface is broader. But this does not relax the bandwidth limitation of the optical fiber. As a result, there's a lot of work now on multi-core fibers or multi-mode fibers, which can now be applied due to the enormous advances in digital signal processing (DSP). If you have multi-mode communication and the signals completely smear and interfere with each other, unidentifiable optical signals seem to exit the line, however state-of-the-art DSP can resolve this with mathematical algorithms and get the right information back out from the signals.
“Interfaces are getting lower and lower with power consumption per bit, but it's not getting lower in total because you have this increase of traffic. ”
—Werner Weiershausen
How are you seeing AI integrated in network operators for internal operations?
We’re definitely working hard to bring AI to internal processes. One aspect of that is making AI usable for employees, through a large language model to support their work on the computer. For example, you can use it to make slides with PowerPoint. You can do some interesting analyses, like of the market, with kind of ChatGPT. To enable this, we installed our own AI, because you need to have a break between the outside and inside of a company—we cannot use our private data with a program open to the outside, because the AI absorbs it and then it becomes open to the public. This solution is still evolving, however, and I'm not totally satisfied yet. But it's in the early stages, so we are working on it and it is quickly improving.
Another element is that we want to modernize and automate our processes, like service activation from the customer going down to the operational layer. For production, you have the end-to-end production of services—a lot of this may be automated. And we work on orchestration architectures, such as hierarchical orchestration, where the top layer is the service orchestration, and then we have the inventory and production orchestration below. We’re working hard on all this and also taking AI into consideration.