SEI-DM: Crafting Revolutionary RF Signals in Low-Resource Settings via Diffusion Model.

In the realm of low-resource environments, Specific Emitter Identification (SEI) is emerging as a paramount tool. SEI involves discerning individual devices by juxtaposing features culled from constrained Radio Frequency (RF) signals. Amidst the pantheon of advanced machine learning paradigms, deep learning-based diffusion models - a distinguished class of generative models - are carving a significant niche. Particularly in resource-tight situations, there's an escalating demand to synthesize copious amounts of RF data leveraging these models. This paper unveils SEI-DM, an innovative RF signal generator underpinned by the diffusion model paradigm. Our methodology distinctively reverses the Gaussian diffusion trajectory over multiple temporal spans, capitalizing on signal label metadata for refined guidance. Validated against authentic Automatic Dependent Surveillance Broadcast (ADS-B) RF signals from aircraft, our experimental suite attests to the prowess of our approach. The empirical evidence underscores a notable leap in SEI recognition - a 36 % uptick in accuracy even at a challenging -20dB signal-to-noise ratio (SNR). Our code and pre-trained checkpoints are available at https://github.com/ZHR-HEU/Low-resource-RF-Signal-Generator.