A Deep Learning-Based Method for Open-set Wireless Signal Detection in HF Band

Wireless signal detection (WSD), which aims to detect and identify signals of interest in the wideband spectrum, is an important part of non-cooperative high frequency (HF) receivers. However, in HF band, WSD remains a challenging task due to the complexity and openness of HF environments, and the presence of unknown signal types may severely affect detection performance. To improve detection performance and robustness, this paper focuses on the deep learning-based methods and proposes a multi-signal detection neural network (MDNN) for open-set WSD in HF band. The proposed network utilizes multi-resolution spectrograms (MRS) of the received wideband signal to estimate the frequency range and signal type of each narrowband signal in the wideband spectrum. In addition, prototype learning and metric learning are also applied to learn discriminative feature representations of different signal types for distinguishing between known and unknown signal types. The experimental re-sults indicate that the proposed method outperforms the baseline methods in both closed-set and open-set HF environments.