Offline Real-World Wireless Interference Signal Classification Algorithm Utilizing Denoising Diffusion Probability Model

The dependable transmission of data and the efficacy of wireless communication depend critically on the detection and classification of interference. Traditional classification algorithms may not yield the requisite precision in identifying and categorizing diverse types of interference, whereas deep learning (DL) algorithms necessitate high-quality data and training samples, which prove unfeasible in real-time communication scenarios. In addressing these challenges, we present a novel approach that utilizes the denoising diffusion probabilistic model (DDPM) for offline processing of collected signals before feature extraction and subsequently sending the signals into a predefined classifier. Our experimental analyses show that our approach delivers up to 91% accuracy without any prior information, outperforming both generative adversarial network (GAN)-based and other traditional DL algorithms, even with limited signal samples of only 5. More significantly, our approach underscores the feasibility of employing generative models in signal processing and achieves state-of-the-art performance on high-precision recognition in real-world communication scenarios.