Semantic Communications for UAV Data Aggregation: A Layered Design Against Alterable Hovering Position

High-efficiency data aggregation in unmanned aerial vehicle (UAV)-assisted scenarios is highly anticipated but often restricted by link capacity. Semantic communications provide a feasible solution by focusing on image meaning and removing redundancies. However, the dynamic nature of UAVs complicates the transmission and reconstruction of semantic information. To address this challenge, a novel layered semantic communications (LSemCom) framework is designed, which consists of an image Semantic Feature Extraction (SFE) layer and a UAV Position and signal Processing Coordination (PPC) layer. Furthermore, combining with the orthogonal frequency division multiplexing systems integrating the non-orthogonal multiple access technology, i.e., OFDM-NOMA systems, we propose two algorithms for the PPC layer. Firstly, a Cascade Learning-based position-Adaptive semantic signal Power allocation (CLAP) algorithm is proposed to adapt the different UAV-user distances by training semantic signal processors (SSPs). Secondly, a joint Alternating Optimization algorithm between UAV Position and Power allocation (AOPP) is proposed. Simulation results reveal that both the CLAP and AOPP algorithms significantly outperform the codecs end-to-end training baseline by 3.0 dB and 3.1 dB w.r.t. peak signal-to-noise ratio (PSNR), respectively. Besides, the CLAP algorithm demonstrates enhanced noise resilience at low signal-to-noise ratios (SNRs), while the AOPP algorithm shows greater robustness against varying Rician fading factors.