RLS Filter Optimization for Non-Invasive Fetal Electrocardiogram Extraction Using the PSO Algorithm

Abstract

Non-invasive fetal electrocardiogram (fECG) extraction is still a challenging task due to the overbearing preponderance of the maternal ECG (mECG) and the presence of noise and interferences. Adaptive filtering techniques, particularly the Recursive Least Squares (RLS) algorithm, have been shown to work well for this issue. However, RLS performance largely depends on a few of its parameters (filter order, forgetting factor, and regularization term), typically tuned empirically, thus limiting robustness and generalizability. In this work, we introduce an automatic parameter optimization process based on the Particle Swarm Optimization (PSO) algorithm. The proposed method was validated using simulated signals generated on MATLAB, including abdominal recordings (aECG), maternal thoracic signals (mECG), and a reference fECG for comparison. Quantitative outcomes, using Mean Squared Error (MSE) and Signal-to-Noise Ratio (SNR) metrics, indicate that PSO-based optimization improves the quality of the resultant fECG compared to the optimal empirical settings, eliminating residual maternal interference. These findings show the potential of PSO for robust fECG extraction and its potential feasibility for real clinical data in the future.