Duke University engineers have developed a method to offload complex artificial intelligence tasks to the airwaves, potentially enabling small, battery-powered devices to make autonomous decisions without depleting their power reserves.
In a study published in Science Advances, the researchers unveiled “WISE” (WIreless Smart Edge networks), a technique that bypasses traditional computer chips by using the physical properties of radio waves to perform calculations.
The breakthrough addresses a critical bottleneck in the development of autonomous technology: while drones, robots, and sensors are becoming increasingly sophisticated, the hardware required to run advanced AI models is often too heavy or energy-intensive for small devices.
“Technology is moving toward smaller devices that can do more than ever before,” said Zhihui Gao, a PhD student at Duke and lead author of the paper. “In order to achieve that, we need new improvements in edge computing. With WISE, we have shown how devices can run on powerful AI without relying on heavy chips or distant servers.”
Reliable but inefficient
Traditional computers process information using binary code — converting data into streams of ones and zeros that are crunched by a digital processor. This method, while reliable, is highly inefficient for battery-powered edge devices.
The WISE system uses “in-physics analog computing” instead. A nearby base station broadcasts a radio signal that encodes the “weights” (the numerical values) of an AI model. When this signal hits a device — such as a drone — simple internal hardware mixes the radio wave with the device’s own data.
This physical mixing process naturally performs the complex multiplication required by AI models, effectively using the radio wave itself as a processor.
“We’re taking advantage of computations that common, miniaturised electronics already gives us,” said Tingjun Chen, assistant professor of electrical and computer engineering at Duke. “Instead of running every step of the model on a chip designed for digital computing, the radio waves themselves help carry information in a way optimised for the computation.”
Smarter swarms
In laboratory experiments, the WISE system achieved nearly 96 per cent accuracy in image classification tasks while consuming more than an order of magnitude less energy than leading digital processors.
The researchers believe the technology could eventually enable a single base station to support swarms of drones during search-and-rescue missions, or help smart traffic cameras coordinate signals without requiring heavy internal computers or a slow connection to the cloud.
“This is the next step in wireless technologies becoming as powerful as wired ones,” Chen added. “Future networks may distribute intelligence by blending communication and computation.”
