AINEWS
Japan's largest mobile carrier and Samsung ran a January 2026 simulation on a 5G trial field. The share of users hit by speed degradation fell from 13.1% to 7.2%.
On a commuter train, a video call stutters just as the train slips between cell towers. The base station could compensate, but it cannot tell that the person on the call is moving, watching a stream, or about to lose service. On August 10, 2026, NTT DOCOMO, Japan's largest mobile carrier, and Samsung Electronics published the result of a January 2026 simulation in which an AI-based radio access network (AI-RAN) cut the share of users hit by communication-speed degradation from 13.1% to 7.2%, a 5.9-percentage-point drop compared with the same network without the technology (DOCOMO and Samsung press release).
Today, the radio settings on a 5G base station are a compromise. One set of parameters has to work for a commuter on a Shinkansen video-calling, a pedestrian browsing, and a stationary 5G hotspot, all at the same time. The DOCOMO-Samsung system profiles each subscriber and applies a configuration tailored to that user's radio conditions and the service they are using (DOCOMO official PR). When the network detects early signs of throughput trouble, like a stream starting to downshift or a video call renegotiating resolution, it can preemptively switch frequency band or radio parameters to keep the session steady.
The data stream underneath is aggregated Minimization of Drive Tests, or MDT, a telemetry feed phones already send back to the network, originally designed so operators could plan radio coverage without driving test vans around. The DOCOMO-Samsung system reads MDT per user rather than per cell, and ties the radio response to the service being consumed: video streaming, voice, gaming, or a video call gets a different tuning (Samsung Research blog). The framing in the press release leans on customer experience, keeping an uninterrupted video stream alive, rather than the older telco-internal goal of predicting radio link failures.
The simulation ran in January 2026 in Japan, on a 5G trial field, against data collected from DOCOMO's commercial network. The 5.9-percentage-point improvement is the comparison against a baseline of the same network without the technology, not against a different vendor's setup (TelecomTV coverage). It is a vendor-issued validation, not an independent benchmark. No external reproduction is visible in the source set.
The press release reaches further than the trial. It frames the result as a step toward predictive network operations in 6G, the next-generation cellular standard that is still being defined. The trial itself ran on 5G. The simulation's value is in showing the per-user mechanism works at all. The 6G framing is forward-looking language, not a delivered capability.
The release does not address one question: data. Reading each subscriber's movement and service use well enough to preempt a buffering event means the network is building a per-user radio-behavior profile, even if the underlying telemetry stream is already in production. A commercial deployment will need to clear that bar in public.
The next read is whether either operator moves the system onto a live commercial 5G network, with the same per-user metric measured on a larger population.