The Hilbert spectrum is a crucial component of the Hilbert-Huang Transform (HHT), a data-adaptive signal analysis method particularly well-suited for analyzing non-linear and non-stationary signals. It represents the instantaneous frequency and amplitude of a signal as a function of time, providing a time-frequency-energy distribution that reveals the embedded energy-frequency-time relationship.
Overview
Traditional spectral analysis methods, such as the Fourier transform, provide a global frequency representation of a signal but assume linearity and stationarity. The Hilbert spectrum, by contrast, offers a local and adaptive decomposition, allowing for the characterization of signals whose frequency content changes over time. It is derived from the application of the Hilbert transform to the Intrinsic Mode Functions (IMFs) obtained through Empirical Mode Decomposition (EMD), the first step of HHT.
Methodology
The construction of the Hilbert spectrum involves two main steps:
-
Empirical Mode Decomposition (EMD): The original signal is first decomposed into a finite and often small number of Intrinsic Mode Functions (IMFs). Each IMF is an oscillatory function that satisfies two conditions:
- The number of extrema and the number of zero crossings must either be equal or differ at most by one.
- The mean value of the envelope defined by the local maxima and the envelope defined by the local minima is zero. Each IMF represents a simple oscillation mode embedded in the data, typically ranging from high-frequency components to low-frequency trends.
-
Hilbert Transform: Once the IMFs, denoted as $c_j(t)$, are extracted, the Hilbert transform is applied to each IMF to obtain its analytic signal, $Z_j(t)$: $Z_j(t) = c_j(t) + i \hat{c}_j(t) = a_j(t) e^{i \theta_j(t)}$ where $\hat{c}_j(t)$ is the Hilbert transform of $c_j(t)$, $a_j(t)$ is the instantaneous amplitude, and $\theta_j(t)$ is the instantaneous phase. The instantaneous frequency, $\omega_j(t)$, for each IMF is then derived as the time derivative of the instantaneous phase: $\omega_j(t) = \frac{d\theta_j(t)}{dt}$
The Hilbert spectrum, denoted as $H(\omega, t)$, is then constructed by plotting the instantaneous amplitudes $a_j(t)$ as a function of time $t$ and instantaneous frequency $\omega_j(t)$. It can be visualized as a 3D plot or a 2D intensity map where the intensity represents the energy or amplitude.
Characteristics and Advantages
- Time-Frequency Resolution: Unlike the fixed resolution of traditional Fourier analysis, the Hilbert spectrum offers a high and adaptive time-frequency resolution, as the frequencies are instantaneous and local.
- Adaptive Nature: The decomposition into IMFs is entirely data-driven, without relying on pre-defined basis functions, making it suitable for a wide range of signals.
- Non-linear and Non-stationary Analysis: It is particularly effective for analyzing signals that exhibit non-linear and non-stationary characteristics, where both frequency and amplitude can vary significantly over time.
- Energy-Frequency-Time Representation: It provides a direct physical interpretation of the energy distribution across different frequencies and over time.
Marginal Hilbert Spectrum
By integrating the Hilbert spectrum over time, one can obtain the Marginal Hilbert Spectrum, $h(\omega)$: $h(\omega) = \int_0^T H(\omega, t) dt$ This spectrum provides a measure of the total energy (or amplitude) contribution from each frequency value over the entire duration of the signal, offering a global energy-frequency distribution analogous to the power spectrum in Fourier analysis, but often more meaningful for non-stationary signals.
Applications
The Hilbert spectrum and the broader HHT framework have found extensive applications in various fields, including:
- Geophysics: Analysis of seismic waves, atmospheric data, and oceanographic signals.
- Biomedical Engineering: Analysis of electroencephalograms (EEGs), electrocardiograms (ECGs), and other physiological signals.
- Mechanical Engineering: Diagnosis of rotating machinery faults, vibration analysis.
- Finance: Analysis of market trends and economic time series.
- Speech Processing: Analysis of speech signals.
Relationship to Other Transforms
While the Hilbert spectrum also provides a frequency representation, it fundamentally differs from the Fourier spectrum. The Fourier spectrum is based on a global decomposition using sinusoidal basis functions, resulting in a single frequency for a given component. The Hilbert spectrum, conversely, uses instantaneous frequencies derived from adaptive decomposition, allowing for multiple frequency components to exist at a single time instant, reflecting the varying oscillatory behavior of non-stationary signals.