When you think of cellular networks, you probably envision radio towers and invisible data streams. But AT&T, by necessity, needs to see everything in between: buildings, trees and the multitude of obstacles that can interfere with wireless signals getting to your phone.

The cellular provider is turning to a key technology from gaming and computer graphics to get an accurate picture. AT&T Wireless Geo Modeler is a new system that uses ray tracing and AI to generate detailed representations of the areas covered by its network and improve connectivity. In doing so, AT&T says it can react to service interruptions quickly and also better predict how its network can be configured in response to large social events or during natural disasters.

Don’t miss any of our unbiased tech content and lab-based reviews. Add CNET as a preferred Google source.

How ray tracing works in a cellular context

In computer graphics, ray tracing is a technique for rendering three-dimensional scenes. Software simulates light beams emanating from a virtual camera and calculates how the light affects objects and materials in the scene. Ray tracing is notable for rendering shadows and reflections, leading to more realistic-looking environments.

In the past, ray tracing was computationally challenging. Early examples, such as the original Toy Story movie, required rooms full of processing hardware and up to 24 hours to render a single frame of footage. The graphics processor in high-end smartphones can now render photorealistic, ray-traced scenes in games in real time.

According to Velin Kounev, lead inventive scientist at AT&T Labs, the technology’s cellular application works the same way. «Whatever Nvidia is doing for games, whatever Disney is doing… we are doing at a much bigger scale,» he said.

In the context of AT&T’s Geo Modeler, Kounev explained, radio propagation from cellular towers is high-frequency light that our eyes cannot see. The towers measure how the rays react to the surrounding environment, such as colliding with structures or reflecting off surfaces. That collected data is processed and analyzed by several internal AT&T systems and machine learning models to determine if changes or optimizations need to be made, in what AT&T calls «near scale time.»

Those changes can include everyday adjustments to the angle of nearby antennas or compensating for a tower that has gone offline during a natural disaster. Modifications can be deployed automatically in seconds or minutes, ideally in a way that doesn’t impact customers.

«We don’t want [customers] to notice,» said Jennifer Yates, assistant vice president of inventive science, network and service automation at AT&T Labs. «The network is self-healing [and] autonomous behind the scenes so they don’t have to think about it.»

The benefit can also be a technical challenge that you would never notice as an AT&T subscriber. «When you hit Lincoln Tunnel traffic at 5 o’clock in the afternoon and you can get your website loaded, that’s when we come in,» said Kounev. «We’re optimizing the network traffic … in rush hour, where you’re able to get your connection.»

Predicting where to deploy resources

Although day-to-day network optimization is one advantage of using the Geo Modeler, it’s also a tool for determining how and where the company should deploy resources during situations such as weather events. For instance, if a prominent tree is blown over during a storm, ray tracing can quickly build a new representation of how towers should compensate.

For large events like music festivals, where tens of thousands of phones are accessing the network or impending natural disasters, the technology can be used to predict upcoming changes that are needed; Kounev mentioned Geo Modeler was applied in April at the Coachella festival. 

Kounev also explained that if a hurricane is coming, for example, knowing its estimated size and timing, «we can go in and within two minutes remove [within the model] the towers that we think are going to be affected, and then see what the network coverage is going to look like.» Knowing where to expect holes in the network allows AT&T to position resources, such as generators or mobile cellular towers, in place before the hurricane strikes.

Most predictive tools, said Kounev, rely on existing measurement data. «Because we use ray tracing, we can predict in places where there’s no measurement data.»

AT&T has been building the Geo Modeler for a year and has accumulated enough data from different use cases over that time to be confident about deploying it more broadly. Yates said that AT&T has performed extensive validation of data, comparing the modeler’s results with measurements in the field.

«Over the last year,» said Kounev, «we had to convince people that this thing can actually work in real time with the many tower stations they have.»