Researchers Develop Wireless Smart Edge Networks for AI Devices

The advancement of edge computing is taking a significant leap forward, thanks to researchers at Duke University who are developing an innovative approach known as Wireless Smart Edge networks (WISE). This new method enhances the capabilities of small devices, allowing them to perform complex artificial intelligence (AI) tasks more efficiently by leveraging the physics of radio waves.

As drones survey forests, robots navigate warehouses, and sensors monitor urban environments, decision-making is increasingly occurring autonomously on the edge. However, transitioning to edge computing presents challenges. While AI models continue to grow in capability, the hardware in these devices often lacks the capacity to handle such demands. Traditional options either require significant memory and energy, leading to battery drain, or rely on cloud processing, which introduces delays and security risks.

The WISE framework aims to overcome these limitations by embedding AI model weights in radio waves transmitted between devices and base stations. This approach, detailed in research published in Science Advances on January 9, 2024, allows for energy-efficient edge AI solutions without compromising speed or size. Led by Tingjun Chen, the Nortel Networks Assistant Professor of Electrical and Computer Engineering, the project also involves collaboration with Dirk Englund from the MIT Research Laboratory of Electronics.

Innovative Computing Through Radio Waves

At the core of the WISE concept is in-physics analog computing, which differs from traditional digital computing that relies on binary code. Digital systems convert data into ones and zeros, then process them through extensive calculations, a method that can be inefficient for small, battery-powered devices.

In contrast, WISE uses the natural behavior of radio waves to perform calculations directly. A base station encodes the AI model’s weight values into a radio frequency (RF) signal. When this signal reaches a nearby device, it interacts with the device’s input data in a way that enables computing without the need for a digital processor. This innovative process allows critical computations to occur in the RF domain, significantly reducing the energy and memory requirements typically associated with running AI models.

“We’re taking advantage of computations that common, miniaturized electronics already give us,” Chen explained. “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 optimized for the computation.”

The results from testing in Chen’s lab demonstrate that the WISE system can achieve nearly 96 percent accuracy in image classification while consuming significantly less energy compared to leading digital processors.

Applications and Future Potential

The implications of WISE are vast. Zhihui Gao, a PhD student in Chen’s lab and lead author of the study, highlighted the potential benefits for various devices such as drones, cameras, and traffic sensors that continuously generate data yet struggle to run advanced models for interpretation.

“Technology is moving toward smaller devices that can do more than ever before,” Gao noted. “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.”

Another significant advantage of WISE is its compatibility with existing infrastructure. Base stations already in place for technologies like 5G and emerging 6G can be adapted with minor adjustments to support this new AI delivery method. Furthermore, many wireless devices already possess the necessary hardware, such as frequency mixers, to facilitate in-physics computation.

Despite these advancements, WISE remains in its early development stages. Current prototypes operate over short distances, necessitating further research to enable longer-range capabilities. Future iterations may require enhanced transmission power or integration with next-generation wireless technologies. Additionally, efficiently broadcasting multiple AI models at once will demand improved multiplexing techniques or more available spectrum bandwidth.

Researchers see the broader applications for WISE, including its potential to support coordinated efforts in search and rescue missions or enhance the functionality of traffic cameras at busy intersections. “This is the next step in wireless technologies becoming as powerful as wired ones,” Chen asserted. “Beyond delivering data and information, these findings open a new direction, in which future networks may distribute intelligence by blending communication and computation to enable energy-efficient edge AI at massive scale.”