Joint Motion Prediction and Adaptive Resource Allocation for Robust ISAC Beam Tracking

Abstract

Millimeter-wave integrated sensing and communication (ISAC) roadside units need to steer narrow pencil beams toward fast-moving vehicles, and a short sensing outage can trigger a divergent tracking loop: As the tracker drifts, the beam widens, subsequent measurements degrade, and recovery becomes increasingly difficult. Addressing prediction or resource allocation in isolation does not reliably prevent this cascade. This paper proposes a closed-loop framework that couples motion prediction, beam design, and sensing-communication resource split under a unified expected-rate objective. A vehicle-to-everything (V2X)-driven Intelligent Driver Model (IDM) with online parameter calibration sustains accurate prediction through outages. The active antenna count is then selected to maximize the expected-rate under the resulting pointing uncertainty. The perframe sensing allocation is determined by jointly optimizing the current and next-epoch expected-rate, which yields regimeadaptive behavior without any hand-tuned thresholds. Monte Carlo evaluation shows an 81.0% reduction in beam alignment error and a 30.3% spectral-efficiency gain over the CV baseline.