Dual-Stage Deep Learning Approach for Efficient Interference Detection and Classification in GNSS

Radio-frequency interference (RFI) can degrade performance or completely prevent the estimation of position, velocity, and time by the receiver. Recently, there has been an increasing trend of jamming events and at the same time, our society is relying heavily on GNSS-based solutions in a wide range of different application domains. To ensure the reliability of these solutions, it is crucial to recognize the possible interference even in low power levels and in the early phase of the signal processing. This can lead to applying the mitigation techniques correctly, by identifying the type and power of the interference. In this article, we introduce a novel approach that operates at the raw signal level (In-phase/Quadrature signal samples) and combines the strengths of two network architectures. The proposed dual-stage architecture consists of a Long Short-Term Memory based autoencoder for GNSS interference detection and Convolutional Neural Networks for classification. This selective approach minimizes the inference and processing time in the real-time monitoring system by computing a spectrogram and extracting the statistical signal features for classification only after detecting the interference in the first stage. Our proposed monitoring system is designed to recognize 16 types of RFI and further classify the power of received interference. Our detection and classification models achieved overall accuracies of 100% and 98.26%, respectively.