In this article you will read about

• how automated AI-based processes can significantly accelerate the planning and expansion of fiber-optic networks.
• what role a geospatial data infrastructure and artificial neural networks play in this process
• why operating this kind of solution in the public cloud is more advantageous than in-house operation in own data centers

Working from home and high-volume streaming for every household – fiber-to-the-home (FTTH) makes this possible. More than 33 million households in Germany already have access to speeds of up to 100 Mbit/s thanks to Telekom's fiber-optic roll-out. We don't want it to stop there: "Our schedule calls for us to put 15 times more fiber-to-the-home connections into operation each year than before," says Niko Gitzen, senior expert at Fiber Factory. As a division of Deutsche Telekom Technik GmbH, it handles the planning for nationwide fiber-optic coverage.

The ambitious schedule cannot be achieved without automated processes. The crux: Until now, large amounts of manual work have slowed down the planning and commissioning of fiber-optic networks. For example, planners had to take hundreds of photos during on-site inspections and then evaluate them manually to find the best possible route for a fiber-optic network. With such a time-consuming and error-prone process, it is highly unlikely that up to two million new fiber-optic connections per year could be achieved.

To achieve the goal, the Fiber Factory planners rely on the support of a geospatial data infrastructure (GDI) from Deutsche Telekom IT GmbH and an artificial neural network (ANN) developed by the Fraunhofer Institute for Physical Measurement Techniques (IPM). The GDI brings together geodata from a wide range of sources, including cadastral data, high-resolution aerial photographs of the roll-out areas from the Federal Agency for Cartography and Geodesy (BKG), images from the European Space Agency's (ESA) Sentinel earth observation satellite, and current images from the T-Surface Car. The latter has laser scanners and cameras with which it takes photos and 3D point clouds and uses them to generate a two- and three-dimensional image of the roll-out area. On average, it collects five gigabytes of data per kilometer; depending on the roll-out area, data of 1 terabyte and more can be collected quickly. 

The T-Surface Car plays a key role in network roll-out planning – but it’s not alone: High-performance computing resources are required to process the volumes of data generated in the GDI and quickly create new roll-out plans. These are provided flexibly and securely by the Open Telekom Cloud: "We operate our geospatial data infrastructure in Deutsche Telekom's public-cloud solution. We also prepare the collected data from the T-Surface Cars there, process it, and link it to other information," says Tobias Frechen, system architect at Deutsche Telekom IT GmbH.

How exactly is the data material evaluated? How are the roll-out plans created? All photos and 3D point clouds are loaded into a bucket of the Object Storage Service (OBS) in the Open Telekom Cloud. In order to calculate the exact surface structure of the roll-out area, aerial photos and other information from the GDI are used in addition to photos and point clouds. This process is automated in the Open Telekom Cloud using graphics processing units (GPUs).

This is where artificial intelligence comes in: The ANN evaluates the survey vehicle's data and classifies it according to 30 different surface and object types, such as asphalt, pavement, curbs, green strips, trees, and street lamps. As a result, the system generates two-dimensional images that show how and where fiber-optic routes can best be laid.

Thanks to the highly-scalable capacities of the Open Telekom Cloud, Fiber Factory employees don’t have to worry about having enough storage space. What's more, usage is dynamic and based on the pay-as-you-use principle. If, for example, less cloud capacity is required because the T-Surface Cars drive less in winter due to the weather, lower costs are also incurred. If, on the other hand, employees plan the parallel roll-out of networks for several areas, sufficient resources can always be added.

It would be expensive and uneconomical to keep such hardware and software capacities in stock in a separate data center over the long term. In the Open Telekom Cloud, scaling to meet demand is possible at any time – for example, horizontal scaling of virtual machines with GPUs for analyzing point clouds and photos, or vertical scaling of virtual machines with CPUs for analyzing aerial photos. In addition, the ANN trained by Fraunhofer IPM with over 90,000 photos can be used as a Docker container. The advantage of this technology: containers require less main memory than virtual machines, can be started and stopped more quickly, and conserve system resources.

The network planners’ experience so far with the new solution, which has been in use since January 2020, has convinced them: Fiber-optic roll-out areas can be developed significantly faster and in greater numbers than before. In addition to the automatic calculation of new routes, they also benefit from the option of virtually roaming the streets with the help of 2D or 3D models. This makes it possible, for example, to measure whether a sidewalk is wide enough to install a new network distributor there. "We have to get approval from the local authorities for this so-called site securing. There is now a standardized procedure for this in our geospatial data infrastructure, in which a form is automatically created and provided with the necessary information," says Tobias Frechen. In this way, in the future it will be possible to provide the municipalities with all the documents relating to fiber-optic roll-out in digital form.

The AI-based process in the Open Telekom Cloud makes it possible to calculate the effort required to expand fiber-optic networks much more accurately. Planners gain more transparency about the duration and costs of the roll-out. In addition, all computing resources are dynamically available to them in a transparent pay-as-you-use model. Potential routes for the fiber-optic roll-out can be calculated more quickly and construction measures can be planned more efficiently and cost-effectively. Ultimately, this speeds up the approval process. The automated processes reduce the planning period by up to 75 percent, so that planners can turn to other roll-out areas at an early stage. Last but not least, the Fiber Factory benefits from a nationwide standardized procedure that gives them significantly more flexibility. In the future, planners in northern Germany will be able to design new networks for roll-out areas in Bavaria – and vice versa.

Finally, Fiber Factory also stays on the safe side when it comes to data privacy: When recordings are made by the T-Surface Car, houses, people, or sensitive information such as license plates are automatically rendered unrecognizable. With the Open Telekom Cloud, all data can also be processed and stored in one of Telekom’s German data centers in compliance with the General Data Protection Regulation (GDPR).

Find out more about Fiber Factory and the Open Telekom Cloud.

Download