Unsupervised Unknown Radar Waveform Detection

This paper proposes a high-accuracy method for detecting unknown radar waveforms using an unsupervised denoising network. The proposed method integrates a Deep Shrinkage Autoencoder (DSAE) and a Memory-Augmented Additional Autoencoder (MAAE). The increasing deployment of radar systems in military surveillance, autonomous vehicles, and medical applications has led to a significant diversification of radar waveforms. This expansion, particularly in the domains of electronic warfare and spectrum management, underscores the necessity for robust techniques capable of distinguishing between known (seen) and unknown (unseen) waveforms. The ability to detect unknown waveforms is critical for mitigating evolving military threats and efficiently managing the electromagnetic spectrum. Recent advancements in radar waveform classification have leveraged supervised deep learning. However, these approaches are inherently constrained by their reliance on large labeled datasets, which are often unavailable due to military confidentiality and the scarcity of comprehensive radar signal repositories. Furthermore, their efficacy is significantly hindered in noisy environments, where external interference degrades signal integrity, complicating waveform classification. In response to these limitations, the DSAE enhances noise robustness by extracting salient signal features while effectively suppressing irrelevant noise, eliminating the need for clean reference data. The MAAE further refines detection accuracy by preserving the structural integrity of known waveforms, thereby facilitating a clearer distinction between known and unknown signals. Empirical evaluations demonstrate that the proposed method significantly outperforms conventional methods, particularly in low signal-to-noise ratio scenarios. These findings highlight its potential as a robust and scalable solution for radar waveform detection in dynamic and contested electromagnetic environments.