Low-Power Interference Identification Based on Convolutional Neural Networks

The inherent vulnerability of global navigation satellite system (GNSS) makes them highly susceptible to various intentional and unintentional interferences. Identifying such interferences is crucial for selecting appropriate antijamming countermeasures. Existing interference identification methods, however, primarily focus on high-power interference signals and exhibit poor performance in identifying low-power interferences. This article, therefore, initially applies time-frequency denoising to enhance the identification performance for low-power interferences. Subsequently, the time-frequency images, autocorrelation function images, and spectral flatness are combined as hierarchical features. Finally, a hierarchical identification classifier based on convolutional neural networks (CNNs) is designed. The classifier consists of a time-frequency image classifier (TFIC), an autocorrelation function image classifier (AFIC), and a spectral flatness classifier (SFC). It employs a hierarchical identification strategy, where the TFIC initially identifies interference signals with high distinguishability in the time-frequency domain. Interferences that are difficult to differentiate undergo secondary identification by the AFIC and the SFC. Experiments on simulated data show that even for low-power interference within the range from -105 to -110 dBm, the overall accuracy of the designed classifier exceeds 96% for nine common types of interference, with a low false alarm rate. In addition, data was collected in an outdoor open scenario and an outdoor obstacle scenario. The experimental results showed that even in complex environments, the CNN classifier based on hierarchical identification (HI-CNN) classifier has good identification performance. This classifier can serve as an effective tool for real-time GNSS interference identification.