Japan’s SoftBank hosted a vendor team that beamed an image of a person over a 5G connection that an artificial intelligence (AI) application was able to recognize.

The team streamed an image using a 5G signal running over virtualized radio access network (vRAN) components powered by graphic processing units (GPUs) between two devices that used a mobile edge compute (MEC) AI application to detect the person in the image.

The trial was conducted in SoftBank’s AI-to-5G Lab, which it constructed last year with help from Nvidia. That lab uses Nvidia’s hardware, and its vRAN and AI processing middleware; virtualized radio signal processing software from Mavenir; and core network software, AI image processing application software on MEC, and physical radio units (RUs) provided by Foxconn Technology.

The trio touted the ability for GPUs to support both RAN and AI applications. This is important as vendors look to remove processing needs from more powerful central processing units (CPUs) to GPUs and data processing units (DPUs).

“In the future, SoftBank will utilize the GPU platform for AI application as a vRAN platform when adding RU and vRAN software, and enable the vRAN site as a MEC location that uses GPUs,” the companies noted in a statement. “SoftBank plans to realize dynamic resource allocation depending on RAN and MEC demand, and lower power consumption of vRAN in the future.”

AI is increasingly being looked at by telecom operators to both support more complex network deployments and for new services.

“There is a growing need for more automation in operator networks in the 5G era,” Malcolm Rogers, senior analyst at GlobalData, noted in a recent blog post. “Complexity from new cloud-native architectures will require a shift in how operators build, manage, and run their networks and new AI-based technologies are at the forefront of orchestration and service assurance. … AI is not a magic bullet and every operator environment is different, meaning different approaches to training AI models will be needed. The rise of 5G is adding new challenges to network management, not only from the incorporation of new cloud-based architectures and use of edge compute, but also the rise of new business cases it is expected to support.”

SoftBank rival NTT, its mobile arm DoCoMo, and Nokia recently announced work on integrating AI, ML, and sub-terahertz spectrum to power potential 6G services.

They explained one of those services was integrating AI and ML into the radio air interface, which they stated “effectively giving 6G radios the ability to learn.” Technically, the move allows the 6G radio to work through signal degradation issues, which reduced signaling overhead and supported a 30% improvement in signal throughput.

The companies also linked the integration to network slicing use cases, stating the 6G radios would gain “the flexibility to adapt to the type of connection demanded by an application, device, or user.”

“For instance, a network in a factory can be optimized for industrial sensors at one moment and then reconfigured for robotic systems or video surveillance,” the firms noted in a statement. “In the public network, an AI-enhanced network can provide an optimized connection for a pedestrian in an [extended reality] session as well as an emergency vehicle traveling at high speed.”

The SoftBank work follows what has been a busy AI week for Nvidia. The chip giant has been touting numerous AI advancements tied to its GTC Developer Conference.

SoftBank’s work with Nvidia also showed there were no lingering hard feelings tied to Nvidia’s failed attempt in acquiring SoftBank subsidiary Arm Holdings. That $40 billion deal was initially announced in 2020, but hit “significant regulatory challenges.”

In early 2021, the British government opened an antitrust investigation into the deal, and in April of that year said it would intervene on national security grounds. The acquisition was dealt another blow when the European Union launched an investigation of its own citing concerns that the deal could harm Arm customers, increase pricing, and hamper innovation.

The U.S. Federal Trade Commission (FTC) sued to block the merger in late 2021.