[B-5A] Radio Communication systemA (mobile communication)

2:30 PM - 2:45 PM JST(5:30 AM - 5:45 AM UTC)

[B-5A-27] Latency Exceedance Prediction Using Machine Learning with Throughput Difference Features

〇Mitsuki Nakamura¹, Motoharu Sasaki¹, Keisuke Wakao¹, Genki Tajima¹, Kenichi Kawamura¹ (1. NTT)

Keywords:

Wireless communication quality,Machine Learning

For mobile networks, 3GPP TS 22.104 defines performance requirements for industrial control use cases, including an End-to-End (E2E) latency below 100 ms and a communication success probability exceeding 99.99%. Furthermore, for advanced use cases such as remote operation and cooperative control, predictive control mechanisms are becoming increasingly important, where future communication quality degradation is proactively estimated and utilized for communication path selection and computation offloading. This paper focuses on short-term throughput fluctuations and investigates a Random Forest (RF)-based method for predicting latency violations, using the current throughput and its deviation from the median throughput observed over the preceding few seconds as predictive features.