EEGDTF: Time-Frequency Disentangled Diffusion for High-Fidelity EEG Signal Generation
IEEE Journal of Biomedical and Health Informatics
|May 19, 2026
Summary
This study introduces EEGDTF, a novel diffusion model for generating high-quality electroencephalogram (EEG) signals. EEGDTF enhances time-frequency modeling and improves generalization across subjects for EEG data augmentation.
Area of Science:
- Neuroscience
- Signal Processing
- Artificial Intelligence
Background:
- Electroencephalogram (EEG) signal generation faces challenges including complex time-frequency structures, limited data, and poor generalization.
- Existing methods lack explicit spectral modeling and robust cross-subject performance.
Purpose of the Study:
- To develop a diffusion-based generative framework, EEGDTF, for synthesizing high-fidelity EEG signals.
- To improve time-frequency modeling and generalization capabilities of EEG signal generation.
Main Methods:
- Proposed EEGDTF framework utilizing a multi-scale residual encoder for temporal representation.
- Implemented a dual-branch encoder-decoder architecture for time-frequency disentanglement.
- Employed a frequency-guided cross-attention mechanism and joint waveform/spectral loss for optimization.
Main Results:
- EEGDTF achieved state-of-the-art performance in both time and frequency domains across four benchmark datasets.
- Demonstrated superior robustness and generalizability, especially under cross-subject conditions.
- Successfully synthesized high-fidelity EEG signals with improved time-frequency characteristics.
Conclusions:
- EEGDTF offers a reliable solution for EEG data augmentation, addressing limitations of existing generative models.
- The framework's robust performance positions it as a valuable tool for Brain-Computer Interface (BCI) applications.
- Highlights the potential of diffusion models in advancing EEG signal synthesis and analysis.
