FALCON: A Diffusion Model-Empowered Semantic Communication Framework for Low-Altitude Wireless Networks

Abstract

The low-altitude wireless network (LAWN), characterized by reconfigurability and multi-functionality, is enabling diverse applications and fostering enhanced collaboration between uncrewed aerial vehicles (UAVs) and terrestrial nodes. Semantic communication, a transformative paradigm in sixth-generation (6G) networks, offers promising potential to advance the intelligence of LAWN. However, UAV-based semantic communication faces critical challenges, including heterogeneous sensing data, constrained onboard computational resources, and rapidly varying channel conditions. To address these issues, this work proposes a novel multimodal semantic communication framework, named FALCON, tailored for low-altitude UAV scenarios. Towards that end, a multi-source cooperative encoding scheme is developed to alleviate cognitive bias caused by modality heterogeneity. Subsequently, a semantic-aware resource allocation mechanism is proposed to reduce computational and spectral overhead by selectively retaining salient features under multi-dimensional contextual constraints. Additionally, we integrate diffusion models (DMs) with a range-null spatial decomposition strategy to robustly reconstruct distorted signals, significantly improving reliability under adverse transmission conditions. Extensive experiments validate that the proposed FALCON framework achieves a compelling performance in balancing task performance and transmission efficiency, highlighting its practical effectiveness for next-generation intelligent LAWN.