Cable television (CATV) remains a cornerstone of how national broadcasts and emergency information reach thousands of households. Unfortunately, since many older apartment buildings were not built with fiber optics in mind, retrofitting them with the necessary wiring for CATV is too costly or impractical, leaving residents unable to access broadcast services. One emerging alternative is 5G MBS (Multicast/Broadcast Services), a technology that uses 5G wireless signals to deliver the same television stream to many households simultaneously. By making efficient use of the limited radio spectrum, it offers a realistic path to modernizing CATV infrastructure.
However, 5G multicast broadcasting has a fundamental limitation. Ordinary 5G connections on smartphones are two-way, so when a data packet fails to arrive, the receiving device can simply request retransmission. Multicast broadcasting may not support such return channels, which means lost data packets cannot be resent, causing the video stream to freeze momentarily. Conventional 5G communication protocols, designed around retransmission requests and maximizing speed, are not built to handle situations where retransmission is limited.
Addressing this challenge, a research team led by Professor Jiro Katto and PhD candidate Kasidis Arunruangsirilert from Waseda University, Japan, developed a lightweight AI model that predicts wireless conditions before they deteriorate and adjusts transmission settings accordingly. Their research results were presented at the 2026 IEEE 104th Vehicular Technology Conference (VTC2026-Fall) , held in Boston, Massachusetts, USA, on September 8.
This AI model was trained using approximately 26 million measurements collected every 0.5 milliseconds from a commercial 5G network, enabling it to capture rapid fluctuations in radio conditions that coarse datasets miss. Importantly, it relies entirely on information already collected by smartphones during standard operation and is compact enough to run in real time on consumer devices without requiring specialized hardware.
Tests on a real-world commercial 5G network showed that the AI model selected an error-free transmission setting for approximately 87% of video segments, compared with only 32% for a conventional speed-oriented approach. The model also operated in less than 0.07 milliseconds on smartphone chipsets released from 2020 onward, introducing no perceptible delay to viewers. As the researchers' remark, "Our study stands as a practical example of AI-native wireless communication, in which AI takes on decision-making responsibilities in next-generation networks, and represents a meaningful step toward the long-sought convergence of broadcasting and broadband on a single, spectrally efficient wireless platform." Overall, this strategy could help bridge the information gap in underserved areas, contributing to a society where everyone has equal access to broadcast services.
Beyond improving television services, the researchers believe the same approach could benefit many forms of one-way wireless communication where retransmission is impossible, including satellite communications, autonomous vehicles, industrial systems, and scientific exploration. The findings also demonstrate how local 5G infrastructure, currently used primarily by large organizations, could deliver practical public services. "We hope that this work encourages broader use of local 5G as a tool for solving social challenges and serving the public good," concludes the researchers.
