Waseda AI Model Cuts Video Freezing on 5G Broadcast TV

Researchers at Waseda University in Japan have developed a lightweight AI model that predicts wireless conditions in advance and adjusts transmission settings before they deteriorate, aiming to prevent the video freezing that commonly affects television delivered over 5G broadcasting. The team, led by Professor Jiro Katto and PhD candidate Kasidis Arunruangsirilert, presented the findings on 8 September at the 2026 IEEE 104th Vehicular Technology Conference in Boston, Massachusetts.

Delivering high-definition television over local 5G networks, known as 5G Multicast/Broadcast Services (MBS), could help modernise cable TV in older apartment buildings where installing fibre-optic cable is too costly or impractical. But multicast broadcasting has a limitation: unlike ordinary two-way 5G connections, it often cannot request retransmission of a lost data packet, so the video stream can freeze momentarily when conditions worsen.

The researchers' AI model was trained on about 26 million measurements taken every 0.5 milliseconds from a commercial 5G network, letting it capture rapid fluctuations that coarser datasets miss. It relies entirely on information smartphones already collect during normal operation and is compact enough to run in real time on the device itself, without specialised hardware.

In tests on a real commercial 5G network, the AI model chose an error-free transmission setting for about 87% of video segments, compared with 32% for a conventional speed-oriented approach. It ran in under 0.07 milliseconds on smartphone chipsets released from 2020 onward, introducing no perceptible delay for viewers.

The researchers said the same approach could also benefit other forms of one-way wireless communication where retransmission is not possible, including satellite communications, autonomous vehicles, industrial systems and scientific exploration.