Effective Sparse Reconstruction Algorithms for Compressed ECG Sensing with Deterministic Binary Block Diagonal Sensing Matrix

For low-power, energy-efficient wearable edge health devices and health monitoring systems, compressed sensing (CS) achieves significant advances by sampling at sub-Nyquist rates for sparse signals such as electrocardiograms (ECG). The paper attempts to explore the best sparse reconstruction algorithm, along with a sparsifying matrix composed of a discrete cosine (C) and a discrete sine (S) basis [C S] and a deterministic binary block diagonal (DBBD) sensing matrix. The six different sparse reconstruction algorithms used for the recovery of CS ECG are orthogonal matching pursuit (OMP), approximate message passing (AMP), L1-minimization (L1-min), compressive sampling matching pursuit (CoSaMP), iterative hard thresholding (IHT), and iterative soft thresholding (IST). The ECG signals from 48 records of the MIT-BIH arrhythmia database (mitdb) were tested for CS reconstruction at compression ratios (CR) of 2.4, 3, 4, 4.8, 6 and 8. Further results showed that the combination of the OMP algorithm, the DBBD sensing matrix, and the basis [C S] surpassed the performance of other recovery algorithms in terms of lower average percentage root mean square difference (PRD). This combination achieved a lowest average PRD of 1.04 for the mitdb record 213, at a CR of 2.4. Although L1-min exhibits competitive performance with OMP in terms of PRD, OMP is computationally simpler than L1-min with a shorter recovery time.