Extraction of time varying information from noisy signals: An approach based on the empirical mode decomposition

A windowed average technique is designed as an efficient assistance of empirical mode decomposition, aimed especially at extracting components with temporally variant frequencies from heavily noisy signals. Unlike those relying on detection of such points as local extrema that are highly sensitive to noise interference, the present method evaluates a local mean curve that reflects the slow variation of a signal in longer time scales by locally integral average over a sliding window. It adapts to variation of signal component in a broad frequency range by making the window width variable in response to the variation. The enhanced performance and robustness of the new algorithm with respect to noise resistance are demonstrated in comparison with other EMD-based methods, and examples of processing both speech and underwater acoustic signals are given to show the success of extracting time varying information.