Detection of Stationarity Loss in Passive Sonar Signals

Abstract

This study addresses the detection of stationarity loss in passive sonar signals, caused by changes in data distribution over time. For this purpose, we utilized statistical tests such as the Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) tests, along with probability distribution similarity estimates, including the Kullback-Leibler Divergence, Jensen-Shannon Divergence, and Wasserstein Distance. The results indicate that similarity estimates between distributions outperform traditional tests in detecting subtle transitions in Probability Density Functions (PDF). Validation was conducted with synthetic data generated noises with different PDFs and real data from the ShipsEar dataset, demonstrating the efficiency of the proposed technique. To facilitate the reproducibility of the experiments, the codes used are available at https://github.com/gabrielhblisboa/stationarityanalysis.