L'IA face aux grands problèmes du monde · Informatique théorique · arXiv · 2026-10-20
Generative Site-Specific Beamforming for Next-Generation Spatial Intelligence
Site-specific beamforming mitigates the channel acquisition bottleneck in multiantenna systems by exploiting environmental priors, but classical discriminative deep learning struggles to represent the multimodality of wireless propagation and the structural features of beamformers. The proposed GenSSBF uses conditional generative models that learn a conditional distribution over feasible beamformers, synthesizing diverse, high-fidelity beam candidates from coarse channel sensing measurements. Case studies in indoor and outdoor scenarios show it attains near-optimal beamforming gain with ultra-low channel acquisition overhead.