A Deep Generative Model for Synthesizing Labeled Wireless Signals

Acquiring labeled wireless signals for sensing tasks remains an expensive, measurement-intensive bottleneck. Addressing the limits of conventional propagation models that require heavy hyper-parameter tuning, this paper introduces IIns-GAN, a generative adversarial network designed to synthesize realistic, position-labeled wireless waveforms. Evaluated on public Ultra-Wideband (UWB) benchmarks, the generated data mirrors real physical properties and reliably supports downstream tasks like distance estimation and environment identification, presenting a pragmatic augmentation method for wireless learning pipelines.