4 December 2024 in IEEE Sensors Journal
Non-stationary signals, characterized by time-varying statistical properties, are prevalent across various disciplines, including telecommunications, seismology, mechanical engineering, and biomedical engineering. Traditional signal processing methods often face challenges in dealing with the spectral content variations and transient phenomena inherent to these signals. This paper introduces a novel time-frequency analysis (TFA) method called the multi-reassignment S-transform (MRST), which is specifically designed to effectively characterize non-stationary signals. MRST builds upon the S-transform (ST) and integrates post-processing strategies derived from the time-frequency multisqueezing transform, yielding a highly concentrated time-frequency representation (TFR) for signals with nonlinear harmonic and impulsive-like components. Simulated and experimental examples, including rub-impact fault data and electrocardiogram (ECG) signals, illustrate MRST’s superior performance in capturing transient features and providing accurate time-frequency localization. The results demonstrate that MRST is a powerful tool for analyzing complex non-stationary signals, offering high reconstruction accuracy and robustness in noisy environments.