Fast ICA Algorithm Based on Independent Component Analysis for Maternal and Fetal ECG Signal Separation

This article introduces the Fast ICA algorithm based on Independent Component Analysis (ICA), which effectively separates maternal and fetal electrocardiogram (ECG) signals. We begin by explaining the fundamentals of Independent Component Analysis and its applications in signal processing. The core implementation involves centering the mixed signals, whitening them using eigenvalue decomposition, and iteratively updating the separation matrix through contrast function optimization.

We then detail the Fast ICA algorithm's principles and computational steps, including: 1) Signal preprocessing and normalization, 2) Weight vector initialization and orthogonalization, 3) Nonlinear contrast function selection (like tanh or exp(-x^2/2)), and 4) Convergence criteria for separation vector updates. The algorithm employs fixed-point iteration to maximize non-Gaussianity through negentropy approximation.

The implementation for maternal-fetal ECG separation specifically handles multi-channel abdominal recordings where the mixing matrix represents different signal propagation paths. Key functions include covariance matrix computation, eigenvalue decomposition for whitening, and iterative separation using approximate Newton method. The algorithm's advantage lies in its computational efficiency and stability compared to traditional ICA methods.

We present practical cases demonstrating successful separation with signal-to-interference ratios above 15 dB, using Python implementations with NumPy for matrix operations and SciPy for optimization. Research results confirm the algorithm's potential for non-invasive fetal monitoring, with code snippets illustrating critical steps like Kurtosis calculation and separation vector normalization.

In summary, this article comprehensively explores the Fast ICA algorithm's application in maternal-fetal ECG separation. Through detailed algorithm explanation, implementation steps, and practical validations, we demonstrate its superiority in handling biomedical signal separation tasks, providing valuable references for researchers and practitioners in biomedical engineering and signal processing.