rewrite this title Neural networks learn signal tricks like basketball, bending beams around obstacles, and unlocking extreme new levels of wireless data flow

High-frequency signals collapse when walls or people block their pathNeural networks learned beam bending by simulating countless basketball practice shotsMetasurfaces integrated into transmitters shaped signals with extreme precision

For years, researchers have struggled with some vulnerabilities in ultrahigh-frequency communications.

Ultrahigh frequencies are so fragile that signals that promise immense bandwidth can collapse when confronted with even modest obstacles, as walls, bookcases, or simply moving people can bring cutting-edge transmissions to a halt.

However, a new approach from Princeton engineers suggests those barriers may not be permanent roadblocks, although the leap from experiment to real-world deployment still remains uncertain.

You may like

From physics experiments to adaptive transmissions

The idea of bending signals to avoid obstacles is not new. Engineers have long worked with “Airy beams,” which can curve in controlled ways, but applying them to wireless data has been hampered by practical limits.

Haoze Chen, one of the researchers, says most prior work focused on showing the beams could exist, not on making them usable in unpredictable environments.

The problem is, every curve depends on countless variables, leaving no straightforward way to scan or compute the ideal path.

To make the beams useful, researchers borrowed an analogy from sports. Instead of calculating each shot, basketball players learn through repeated practice what works in different contexts.

Chen explained the Princeton team aimed for a similar process, replacing trial-and-error athletes with a neural network designed to adapt its responses.

Rather than physically transmitting beams for every possible obstacle, doctoral student Atsutse Kludze built a simulator that allowed the system to practice virtually.

This approach greatly reduced training time while still grounding the models in the physics of Airy beams.

Once trained, the system was able to adapt extremely quickly, using a specially designed metasurface to shape the transmissions.

Unlike reflectors, which depend on external structures, the metasurface can be integrated directly into the transmitter, which allowed beams to curve around sudden obstructions, maintaining connectivity without requiring clear line-of-sight.

The team demonstrated that the neural network could select the most effective beam path in cluttered and shifting scenarios, something conventional methods cannot achieve.

It also claims this is a step toward harnessing the sub-terahertz band, a part of the spectrum that could support up to ten times more data than today’s systems.

Lead investigator Yasaman Ghasempour argued that addressing obstacles is essential before such bandwidth can be used for demanding applications like immersive virtual reality or fully autonomous transport.

“This work tackles a long-standing problem that has prevented the adoption of such high frequencies in dynamic wireless communications to date,” Ghasempour said.

Still, challenges remain. Translating laboratory demonstrations into commercial devices requires scaling the hardware, refining the training methods, and proving that adaptive beams can handle real-world complexity at speed.

The promise of wireless links approaching terabit-class throughput may be visible, but the path around the obstacles, both physical and technological, is still winding.

Via Techxplore

You might also like