UCSD - University of California - San Diego
08/06/2024 | Press release | Distributed by Public on 08/06/2024 17:36
AI Software “EdgeRIC” Could Make Your Internet Experience Smoother, Faster
Story by:
Published Date
August 06, 2024Article Content
A team from the UC San Diego Qualcomm Institute(QI) and Jacobs School of Engineering, and Texas A&M University has developed an artificial intelligence (AI)-based software platform that could one day give internet users a more efficient and enjoyable experience over the cellular network.
The open-source software platform, called "EdgeRIC," uses AI to "take the temperature" of the cellular wireless environment that connects us to the internet, then redirects resources according to users' needs in real-time. While the technology is still in development, it could someday be integrated into existing hardware distributed by major telecommunications companies to reach the public.
The platform will be shared at the National Science Foundation's Open AI Cellular and EdgeRIC workshopat Mississippi State University from August 6-8, 2024. Earlier this year, it was presented in a paper accepted at the 2024 NSDI Symposium, a prominent event in the field of networked systems.
"Imagine you are on a highway about to hit a dead zone, like in a National Park, and somehow your phone can still recognize and download videos for the next 10 minutes," said Dinesh Bharadia, an affiliate of QI, associate professor in the Jacobs School of Engineering's Department of Electrical and Computer Engineering (ECE), and head of the department's Wireless Communications Sensing and Networking Group (WCSNG). "This is what EdgeRIC provides: application-aware mobile networks at every millisecond, so everything from our cars and robots which rely on the internet can be smart and safe as well."
"EdgeRIC exemplifies how AI can revolutionize cellular networks by bringing real-time intelligence from edge compute into the cellular stack," said Srinivas Shakkottai, a professor with the Department of Electrical and Computer Engineering at Texas A&M University, who co-directs the Texas A&M Initiative for Connected Intelligence. "With fast messaging pipelines to access both radio network and application-level information and using well-optimized AI models, EdgeRIC has achieved a breakthrough in reducing latency. This enables highly responsive connectivity, significantly enhancing system performance and user experience."
The study was led by UC San Diego doctoral student Ushasi Ghosh and Texas A&M University Senior Research Engineer Woo-Hyun Ko. Ghosh and Ko will present EdgeRIC to an audience of students, researchers, and industry practitioners at the Open AI Cellular and EdgeRIC workshop.
EdgeRIC is a platform for real-time AI-in-the-loop for decision and control in cellular networks. It is designed to access network and application-level information to execute AI-optimized and other policies in real-time (sub-millisecond), reducing lag, and prioritizing a smooth online experience over cellular networks.
Programming the issue of time
To date, existing software that directs resources through the radio access network (RAN)-mobile users' gateway to the internet, beginning with radio base stations and leading downstream to user devices-struggles to keep up with the fast-paced wireless environment and the needs of clients. Changes in clients' needs happen within milliseconds, but today's RAN intelligent controllers (RIC)-AI and machine learning platforms that monitor the state of wireless demand and redirect resources accordingly-can take up to 10 milliseconds to respond.
That, and the heavy lift from having to handle complex calculations, can lead to disruptions in cellular network functions and create bottlenecks in internet coverage.
Building on previous work funded by the National Science Foundation, Bharadia, Shakkottai, Ghosh, Ko and colleagues designed EdgeRIC to have a two-way connection to the RAN. Using built-in custom microapps, or μApps, EdgeRIC monitors and tracks changes in the RAN in real-time, and can choose how to respond just as quickly. EdgeRIC then sends decisions in real-time to the RAN, prioritizing users based on the specific needs of their applications.
Reducing the amount of time it takes for software to read the wireless environment and then respond within 1 millisecond was the team's greatest challenge, Ghosh said, but the problem had a surprisingly straightforward solution. The team decoupled EdgeRIC from the RAN, running it separately but beside a base station's processing and storage unit, called a compute node, where it could interact with the RAN through a standard compatible interface. This decoupling had the advantage of allowing EdgeRIC to handle complex computations without overloading and disrupting the RAN.
Building on previous work funded by the National Science Foundation, Bharadia, Shakkottai, Ghosh, Ko and colleagues designed EdgeRIC to have a two-way connection to the RAN. Using built-in custom microapps, or μApps, EdgeRIC monitors and tracks changes in the RAN in real-time, and can choose how to respond just as quickly. EdgeRIC then sends decisions in real-time to the RAN, prioritizing users based on the specific needs of their applications.
Reducing the amount of time it takes for software to read the wireless environment and then respond within 1 millisecond was the team's greatest challenge, Ghosh said, but the problem had a surprisingly straightforward solution. The team decoupled EdgeRIC from the RAN, running it separately but beside a base station's processing and storage unit, called a compute node, where it could interact with the RAN through a standard compatible interface. This decoupling had the advantage of allowing EdgeRIC to handle complex computations without overloading and disrupting the RAN.
In tests, the team found that EdgeRIC's μApps outpaced the amount of data received by existing, cloud-based, near real-time RICs by 5 to 25%. They observed a 30% increase in metrics used to measure an internet user's experience, such as the smoothness of video streaming.
"EdgeRIC was meticulously developed and tested using real-world data collected at the TAMU Innovation Proving Ground," said Ko. "By leveraging diverse scenarios involving drones, cars, autonomous robots, and human-scale mobility, we have been able to create robust and optimized AI algorithms."
EdgeRIC also includes a built-in function that allows researchers to train the software offline, allowing for improvement toward the ultimate goal of preventing lag, dropped video calls and other detractions to users' online experience.
"At the end of the day, our ultimate goal is to satisfy the end user and better understand their needs," said Ghosh.
Ushering in a new generation of the cellular networks
At NSDI 2024 in Santa Clara, California, Ghosh says EdgeRIC received a positive response, as others indicated an interest in real-time AI as the basis for a new way of interacting with RANs.
For now, Ghosh says, the team will continue to fine-tune EdgeRIC's AI algorithms and research policies in AI for a stronger product. Next week, she and Ko will lead attendees of the Open AI Cellular and EdgeRIC workshop in hands-on training sessions to integrate and use their platform with existing software and hardware. Ideally, Ghosh would like EdgeRIC to represent a "one stop shop" for intelligent solutions to problems arising in RANs, for a smoother experience in 5G.
Researchers in the lab of QI affiliate and Jacobs School of Engineering faculty member Dinesh Bharadia demonstrate how EdgeRIC, here being run on a laptop, can create real-time decisions for radio access networks. The researchers used a reinforcement learning-based policy to train EdgeRIC to maximize the overall system's throughput.