OT-ADG: Optimal-transport-driven adaptive data generation for WiFi sensing in airborne mobile networks

Airborne Mobile Networks (AMNs) require high-precision situational awareness in enclosed spaces for critical tasks like border surveillance and maritime monitoring. Compared to vision and wearable technologies, WiFi signals in AMN environments are a promising sensing medium due to their non-intrusiveness, cost-effectiveness, and privacy benefits. However, the development of WiFi-based sensing in AMNs is hindered by the scarcity of high-quality large-scale data. While Data Augmentation (DA) can address this scarcity, traditional time-series DA may distort WiFi Channel State Information (CSI)’s time-frequency properties, and deep generative models suffer from mode collapse, spectral distortion, and high computational costs. To overcome these limitations, we propose OT-ADG, an adaptive data generation algorithm based on Optimal Transport (OT) theory to synthesize high-fidelity WiFi sensing samples. It dynamically models subcarrier-specific energy distributions and computes optimal transport plans using Sinkhorn’s algorithm with entropy regularization. Furthermore, we present ViFi, the first large-scale and scenario-rich WiFi sensing dataset to fill the gap in AMN applications, encompassing 26 real-world scenarios with 20 640 CSI samples and synchronized videos. Experimental results show OT-ADG’s superior performance, with maximum improvements of up to 14.88% over baseline recognition methods, while outperforming existing DA approaches. Moreover, the robust performance of multiple WiFi-based HAR models validates the effectiveness of the ViFi dataset.