The paper explores the behavior of Empirical Mode Decomposition (EMD) in stochastic situations, particularly when applied to fractional Gaussian noise (fGn). EMD, initially developed by N.E. Huang et al., is a method for decomposing nonstationary signals into zero-mean amplitude-frequency (AM-FM) components. The authors conduct numerical experiments to understand how EMD behaves in the presence of broadband noise. Key findings include: 1. **Filter Bank Behavior**: EMD acts as a dyadic filter bank, similar to wavelet decompositions, where the modes extracted by EMD can be interpreted as overlapping band-pass filters. 2. **Hurst Exponent Estimation**: The hierarchy of extracted modes can be used to estimate the Hurst exponent \( H \) of the fGn process. The variance progression of the modes across different IMF (Intrinsic Mode Function) indices provides a linear relationship for \( H \geq 1/2 \), indicating self-similarity in the filter bank structure. 3. **Self-Similarity**: The filter bank structure of EMD is self-similar, with the power spectra of the IMF's collapsing onto a single curve when renormalized according to a specific scaling factor. The study highlights the adaptivity and effectiveness of EMD in analyzing structured broadband stochastic processes, suggesting that it may offer a new approach to analyzing self-similar processes. The results also emphasize the need for theoretical explanations to support the observed behaviors, particularly in the context of EMD's non-analytical nature.The paper explores the behavior of Empirical Mode Decomposition (EMD) in stochastic situations, particularly when applied to fractional Gaussian noise (fGn). EMD, initially developed by N.E. Huang et al., is a method for decomposing nonstationary signals into zero-mean amplitude-frequency (AM-FM) components. The authors conduct numerical experiments to understand how EMD behaves in the presence of broadband noise. Key findings include: 1. **Filter Bank Behavior**: EMD acts as a dyadic filter bank, similar to wavelet decompositions, where the modes extracted by EMD can be interpreted as overlapping band-pass filters. 2. **Hurst Exponent Estimation**: The hierarchy of extracted modes can be used to estimate the Hurst exponent \( H \) of the fGn process. The variance progression of the modes across different IMF (Intrinsic Mode Function) indices provides a linear relationship for \( H \geq 1/2 \), indicating self-similarity in the filter bank structure. 3. **Self-Similarity**: The filter bank structure of EMD is self-similar, with the power spectra of the IMF's collapsing onto a single curve when renormalized according to a specific scaling factor. The study highlights the adaptivity and effectiveness of EMD in analyzing structured broadband stochastic processes, suggesting that it may offer a new approach to analyzing self-similar processes. The results also emphasize the need for theoretical explanations to support the observed behaviors, particularly in the context of EMD's non-analytical nature.