Detecting Interference in GNSS using Machine Learning

Global Navigation Satellite Systems have truly established themselves as part of the critical infrastructure in today’s world. GNSS positioning, navigation, and timing services have a significant role in various applications. These GNSS signals are transmitted from satellites that take place over 22000 km far from the earth. The power of the received signal in the antenna is weak and vulnerable to interference, causing reduced positioning accuracy or even the complete lack of position availability. Thus, it is critical and essential for several applications to detect interference and recognize their type to mitigate and guarantee a high accuracy positioning. Various types of intentional interferences disturb the GNSS signal on the way of receivers. In this paper, we focused on the CHIRP signal, one of the frequent jamming signals. We created a dataset composed of the original GNSS signal, and we combined it with different shapes of simulated CHIRP signals to implement a smart decision to unfold the CHIRP’s presence in the received signal using machine learning algorithms. The proposed algorithms are K-Means clustering, unsupervised machine learning, and Convolutional Neural Network(CNN). Features of signals extracted for K-Means solution are in the time domain, frequency domain using Fourier Transform, and time-frequency domain applying Wavelet Transform and Wigner-Ville Transform. Different approaches are then implemented, such as Principals Component Analyses, Correlation matrix, and unsupervised feature selection to select the most relevant and informative features to increase model performance and decrease model creation’s computational time. The other studied classifier was the CNN, based on multi-layer neural networks as it has excellent performance. The images of Time-Frequency analysis using Wigner-Ville Transform fed the CNN algorithm to classify the different shapes of CHIRP signals.