Hilbert-Huang Transform (HHT)

The foundational, fully data-adaptive algorithm of the HHT. EMD decomposes any complex signal into a finite set of Intrinsic Mode Functions (IMFs) through an iterative sifting process. Unlike Fourier or wavelet transforms, it requires no a priori basis functions, making it ideal for analyzing nonlinear and non-stationary processes. The sifting algorithm identifies local maxima and minima, constructs upper and lower envelopes via cubic spline interpolation, and subtracts the mean envelope to isolate the highest-frequency oscillation. This process repeats until the residue becomes a monotonic function, ensuring each IMF satisfies the narrow-band condition necessary for meaningful instantaneous frequency calculation.

The elementary building blocks extracted by EMD. An IMF is a function that satisfies two strict conditions:

• The number of extrema and the number of zero-crossings must either be equal or differ at most by one.
• At any point, the mean value of the envelope defined by the local maxima and the envelope defined by the local minima is zero. These constraints ensure each IMF represents a simple oscillatory mode with a well-behaved instantaneous frequency, free from riding waves and amplitude modulation asymmetries that would corrupt the subsequent Hilbert spectral analysis.

The second stage of the HHT, applied to each extracted IMF. The Hilbert transform is used to construct the complex analytic signal from each real-valued IMF, enabling the unambiguous calculation of instantaneous amplitude, instantaneous phase, and instantaneous frequency as functions of time. Plotting the instantaneous amplitude against time and frequency yields the Hilbert spectrum, a high-resolution energy-time-frequency distribution. Unlike the spectrogram, which smears energy across fixed frequency bins due to the uncertainty principle, the Hilbert spectrum provides a sharp, physically meaningful representation of non-stationary dynamics, revealing intra-wave frequency modulation.

The EMD process guarantees completeness, meaning the original signal can be perfectly reconstructed by summing all extracted IMFs and the final residual trend. While not theoretically guaranteed, the decomposition achieves near-orthogonality in practice, with the orthogonality index typically falling well below 1%. This quasi-orthogonal property ensures minimal energy leakage between modes, allowing each IMF to represent a physically distinct oscillatory component. The residual trend captures the slowest time-scale variation, representing the signal's underlying mean behavior without imposing a polynomial or exponential model.

The HHT's primary advantage over Fourier and wavelet methods is its ability to handle signals generated by nonlinear and non-stationary physical processes. Traditional methods decompose signals into linear superpositions of predefined harmonic functions, which can produce mathematically valid but physically meaningless harmonics for nonlinear systems. The HHT's adaptive basis, derived directly from the signal's local time scales, captures intra-wave frequency modulation—frequency variations within a single oscillation cycle—which is a hallmark of nonlinear dynamics. This makes it indispensable for analyzing real-world phenomena like ocean waves, structural vibrations, and biomedical signals.

A known limitation of standard EMD is mode mixing, where a single IMF contains signals of widely disparate scales or similar scales appear across different IMFs, often triggered by intermittent noise. To address this, Ensemble Empirical Mode Decomposition (EEMD) was developed. EEMD adds finite-amplitude white noise to the signal across multiple trials, performs EMD on each noisy version, and averages the resulting IMF ensembles. The added noise populates the entire time-frequency space, providing a uniform reference frame that forces the sifting process to lock onto the true physical scales, effectively eliminating mode mixing without requiring subjective intervention.