Deep Learning Predictive Band Switching in Wireless Networks

In cellular systems, the user equipment (UE) can request a change in frequency band when its rate drops below threshold on current band. The UE is then instructed by base station (BS) to measure quality of candidate bands, which requires measurement gap data transmission, thus lowering rate. We propose an online-learning based switching approach that does not require any gap. Our proposed classifier-based policy instead exploits spatial and spectral correlation between radio signals different bands knowledge location. focus lower (e.g., 3.5 GHz) millimeter wave 28 GHz), design evaluate two classification models are trained ray-tracing dataset. A key insight gaps overkill, only relative order necessary for selection, rather than full channel estimate. machine learning-based policies achieve roughly 30% improvement mean effective rates over those industry standard policy, while achieving misclassification errors well 0.5% maintaining resilience against blockage uncertainty.