基于大语言模型的低空无人机波束预测OA

In the context of the rapidly developing low-altitude economy,the high mobility of unmanned aerial vehicles(UAVs)in three-dimensional space causes severe channel state information(CSI)aging and beam misalignment in integrated sensing and communication(ISAC)systems.To address these issues,this paper proposes an end-to-end beamforming prediction framework based on a large language model(LLM).Specifically,we first establish the channel and measurement models for a monostatic shared transceiver scenario and derive the Cramér-Rao lower bound(CRB)for multi-dimensional sensing parameters,including elevation angle,azimuth angle,delay and Doppler shift.We construct an unsupervised loss function based on the derived integrated sensing CRB and fine-tune a pre-trained LLM to extract the spatial-temporal correlations of historical CSI,thereby directly outputting the optimal beamforming vector.Simulation results demonstrate that,when confronting complex,random 3D flight trajectories and high flight speeds,the proposed scheme achieves end-to-end high-precision tracking of the highly maneuverable and non-linear UAV trajectories.Furthermore,its sensing accuracy,robustness against high-speed-induced fading,and system communication rate significantly outperform those of traditional time-series networks and baseline algorithms.