Radio Anomaly Detection Based on Improved Denoising Diffusion Probabilistic Models

Detection of radio anomalies is critical for wireless communication security. In this letter, we propose an algorithm for radio anomaly detection based on improved denoising diffusion probabilistic models (DDPMs). First, the signal’s power spectral density (PSD) is obtained. Next, an autoencoder-DDPMs (AE-DDPMs) framework is proposed by incorporating an autoencoder into DDPMs to learn the distribution of normal signals and their PSD. Finally, the signal is determined to be anomalous or not based on the error between the reconstructed data and the input data. Simulation results demonstrate that the performance of the AE-DDPMs-based algorithm outperforms state-of-the-art algorithms.