Hardware implementation of optimal threshold singular spectrum analysis for real-time denoising of biomedical signals

Singular Spectrum Analysis (SSA) has been shown in the literature to be an effective approach in biomedical denoising applications, however, it can be computationally expensive, especially in its sliding window format. In this paper, we have developed a method to efficiently denoise biomedical signals, using a real-time application running on an inexpensive ARM based embedded system. The algorithm uses a novel Optimal Threshold technique applied to traditional SSA to exploit low-rank matrix structures. This Optimal Threshold SSA (OT-SSA) recovers a low-rank matrix from noisy data by retaining singular values above the threshold during the grouping stage. We evaluated OT-SSA using quantitative metrics with both synthetic and real mixtures of biomedical signals. The results shows that OT-SSA achieves significant interference suppression, with a low reconstruction error. OT-SSA enables real-time biomedical signal processing on an embedded system, capable of efficiently removing ECG interference from diaphragmatic EMG signals, and removing EOG interference from EEG signals. The resultant application can operate in real-time with a low latency of 125 m s that employs short non-overlapping SVD window sizes of 100 × 50.