Sparse time-frequency decomposition based on dictionary adaptation

Area of Science:

• Signal Processing
• Optimization Methods
• Time-Frequency Analysis

Background:

• Traditional time-frequency analysis methods often rely on predefined basis functions.
• Adapting the decomposition basis to the signal itself is crucial for accurate analysis, especially for complex or noisy data.
• Dictionary learning typically requires a training set, limiting its application to single-signal adaptation.

Purpose of the Study:

• To propose a novel time-frequency analysis method for accurate instantaneous frequency estimation and signal decomposition.
• To develop a dictionary adaptation approach where the basis is determined concurrently with the signal decomposition.
• To address limitations of existing methods in handling signals with poor scale separation, outliers, and noise.

Main Methods:

• Formulated signal decomposition as an optimization problem with an adaptive dictionary.
• Employed the augmented Lagrangian multiplier (ALM) method for iterative dictionary adaptation.
• Accelerated the ALM method using the fast wavelet transform for enhanced computational efficiency.

Main Results:

• Successfully decomposed various signals, including those with poor scale separation and noise.
• Demonstrated accurate recovery of instantaneous frequencies for complex signal components.
• Showcased precise reconstruction of intrinsic mode functions (IMFs) from challenging datasets.

Conclusions:

• The proposed adaptive dictionary approach provides a robust method for time-frequency analysis.
• This technique offers accurate signal decomposition and instantaneous frequency estimation.
• The method is effective for real-world signals, including those with significant noise and outliers.