基于KANsformer的IRS辅助高铁波束赋形方法OA

For intelligent reflecting surface(IRS)-assisted high-speed railway communication systems,existing beamforming algorithms still suffer from a low energy efficiency due to the impact of 5G-R fast time-varying non-stationary channels.A novel IRS-assisted high-speed railway beamforming method based on the KANsformer is proposed.First,a global-local feature extraction module is architected to capture channel-specific spatial features through sparse convolutional operations,while incorporating multi-head self-attention mechanisms for long-range dependency modeling.This integrated approach enables preliminary discrimination between the main lobe of IRS-assisted beamforming and interfering paths.Second,an enhanced KANsformer encoder architecture integrating a dynamic sparse multi-head attention mechanism is introduced.This mechanism adaptively prioritizes dominant propagation paths,exhibiting high energy characteristics based on real-time channel fluctuations while simultaneously mitigating interference inherent in the rapidly time-varying non-stationary 5G-R channel.By attenuating energy dispersion in non-critical directions,the directivity and spatial efficiency of beamforming are substantially improved.Finally,a KAN decoder is constructed for nonlinear decoding,which outputs a beamforming vector that satisfies the transmit power constraint,thereby completing the beamforming process.Simulation results demonstrate that the proposed method achieves superior energy efficiency optimization compared with existing approaches under different train velocities and IRS configurations.