2026 International Conference on Networks Computers and Communications (ICNCCom 2026)

Low-altitude unmanned aerial vehicles (UAVs) that transmit sensing data or video streams over millimeter-wave (mmWave) links incur a received-power loss when a narrow beam is misselected. Top-1 beam prediction reduces beam-training overhead but is unreliable near beam-transition regions. Fixed-top-$K$ beam probing improves link reliability at the cost of a constant probing overhead per frame. This paper studies a controlled synthetic UAV beam-link problem in which the receiver selects the candidate-beam probing set size and the UAV controller applies bounded trajectory smoothing. The proposed framework combines ranked beam prediction, future beam-loss risk estimation, risk-triggered adaptive top-$K$ beam probing, and finite-horizon model predictive control (MPC). The objective accounts for estimated beam loss, future beam-loss risk, probing overhead, trajectory smoothness, and mission-path deviation. For the pilot split with 765/153/306 training/validation/test samples, adaptive top-3 risk-aware MPC reduces the 95th-percentile (P95) beam loss from 2.3180 dB with top-1 prediction and 1.1735 dB with smoothing-only control to 0.5829 dB, while requiring 2.105 probed beams per frame on average. A larger five-seed synthetic diagnostic confirms top-$K$ beam containment and probing-overhead reduction, but shows weak risk-score discrimination across unseen trajectories. The simulation results support an overhead-aware reduction in P95 beam loss under controlled synthetic beam-link conditions.

UAV communications,millimeter-wave communications,beam prediction,adaptive beam probing,trajectory optimization