AIGC-Based Radio Map Construction for Channel Estimation in Low-Altitude Economy

Low-altitude economy (LAE) has garnered increasing interest, but its large-scale deployment faces substantial obstacles. The secure and efficient operations of unmanned aerial vehicles (UAVs) rely on reliable and high-speed wireless communication services delivered by base stations (BSs). However, high mobility, in addition to multipath effects, poses significant challenges to channel estimation. To address these challenges, we propose an AI-generated content (AIGC)-based approach to assist channel estimation for LAE. Specifically, a new generative adversarial network (GAN)-based model named continuous vector-conditioned GAN (CVCGAN) is designed to construct the radio map by establishing the mapping from UAV’s sensing information to its channel space. Subsequently, a neural network-based integrator is developed to perform Bayesian inference of channel state information (CSI), leveraging the prior distribution learned by the constructed radio map and the partial observations provided by pilot measurements. Simulation results validate the superiority of the proposed scheme, which outperforms state-of-the-art (SOTA) approaches including ChannelNet, conditional GAN (CGAN), and long short-term memory (LSTM).