Optimal machine learning and signal processing synergies for low-resource GNSS interference classification

Interference signals degrade the performance of a global navigation satellite system (GNSS) receiver. Classification of these interference signals allow better situational awareness and facilitate appropriate counter-measures. However, classification is challenging and processing-intensive, especially in severe multipath environments. This article proposes a low-resource interference classification approach that combines conventional statistical signal processing approaches with machine learning (ML). It leverages the processing efficiency of conventional statistical signal processing by summarizing, e.g., a short-time Fourier transform (STFT), with statistical measures. Furthermore, the ML design space is bounded as the signal is pre-processed. It results in fewer opportunities for ML but facilitates faster convergence and the use of simpler architectures. Therefore, this approach has lower ML training complexity and lower processing and memory requirements. Results show competitive classification capabilities to more complex approaches. It demonstrates that more efficient architectures can be developed using existing signal-processing approaches