ECG Sensor Design Assessment with Variational Autoencoder-Based Digital Watermarking.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Digital Watermarking

Background:

• Designing effective Electrocardiogram (ECG) sensor circuits is crucial for accurate heart signal detection and conditioning.
• Digital watermarking is essential for data security, authenticity, and copyright protection in healthcare.

Purpose of the Study:

• To evaluate ECG sensor performance using watermarked signals generated by a Variational Autoencoder (VAE).
• To investigate and compare different watermarking strategies for ECG signals.

Main Methods:

• Generation of diverse ECG signals with embedded watermarks using a VAE framework.
• Implementation and comparison of three watermarking techniques: mean (μ) latent space, latent variable (z), and frequency domain.
• Assessment of signal fidelity using Mean Squared Error (MSE) under varying conditions.

Main Results:

• Watermarking in the mean (μ) and frequency domain yielded low MSE and stable signal fidelity.
• These methods demonstrated robustness across different watermark strengths, latent space dimensions, and noise levels.
• Performance was minimally affected by parameter variations, indicating suitability for ECG signal preservation.

Conclusions:

• The mean (μ) and frequency domain watermarking methods are effective for ECG sensor applications, ensuring data security and signal quality.
• VAE-based watermarking enhances ECG signal resilience and authenticity for modern healthcare.
• This approach supports reliable signal acquisition and protects digital ECG content.