A Sparsity-Aware Large AI Model-Enabled Framework for Joint Channel Estimation and Location Sensing in Low-Altitude Networks

Abstract

The emergence of low-altitude networks (LANs) necessitates high-rate communication and high-accuracy localization to overcome the channel dynamics caused by unmanned aerial vehicle (UAV) mobility. To address these requirements, this paper proposes a novel sparsity-aware learning framework for joint channel estimation and location sensing. The proposed framework treats channel state information (CSI) acquisition and sensing enhancement as a unified problem solved via delay-Doppler estimation. Specifically, we design a sparsity-aware encoder that identifies and extracts dominant scattering components to exploit the inherent sparse structure of the considered aerial millimeter wave channel. To enhance estimation precision, we introduce a hybrid learning-based estimator that integrates a specialized embedding module with a pre-trained large artificial intelligence model (LAM). This architecture improves accuracy while significantly reducing the training overhead compared to conventional deep learning approaches. Through numerical results, we demonstrate that our framework accurately estimates delay and Doppler parameters, outperforming both traditional heuristic methods and state-of-the-art learning-based algorithms in terms of robustness and computational complexity. This work provides valuable insights into supporting simultaneous high-rate communication and sensing in future aerial wireless systems.