LLM–empowered wireless channel quality prediction for reliable cell handover

In high-mobility and ultra-dense wireless networks, frequent handovers necessitate continuous beam scanning and measurements, incurring substantial measurement overhead and prolonged handover latency. Existing artificial intelligence (AI)-assisted handover solutions typically rely on lightweight, task-specific models and often generalize poorly across rapidly changing mobility conditions. This paper proposes a large language model (LLM)-empowered wireless channel quality prediction framework that forecasts the temporal evolution of inter-cell reference signal received power (RSRP) to enable proactive handover. Specifically, the framework integrates a beam-adaptive RSRP embedding module with a fine-tuned GPT-2 backbone, effectively bridging numerical RSRP and LLM representations while supporting arbitrary input beam configurations. This architecture enables the effective capture of long-range spatiotemporal dependencies within multi-cell RSRP measurements, achieving high-fidelity prediction accuracy across arbitrary beam configurations. Leveraging these channel quality predictions, we further develop a proactive handover mechanism that significantly reduces latency while enhancing reliability. Simulation results show improved prediction accuracy and generalization over baseline methods, and yield the lowest handover failure rates, indicating the promise of LLM-assisted proactive mobility management.