Compressed Sensing Techniques for High-Speed Biomedical Signal Acquisition
Abstract
Compressed sensing (CS) is a game-changing way to acquire and reconstruct biomedical signals at sampling rates significantly below the Nyquist rate to gain powerful reduction in data size, power, and hardware. In this work, we discuss CS methods for fast biomedical signals, including ECG, EEG, and MRI. We review the theory concerning CS as it relates to sparsity, incoherent sampling, and convex optimization methods. We also propose a new adaptive reconstruction method which combines structured sparsity with iterative thresholding, providing better reconstruction under noise and real-time requirements. Simulations and experimental results show that the proposed method achieves higher (up to 70%) diminution of the sample requirements with signal fidelity above 95% compared to other methods. Overall, we build a case for CS allowing for new portable, low-power biomedical devices that are capable of non-invasive real-time monitoring and diagnosis. We also identify and discuss challenges and future work including hardware implementation, and deep learning frameworks.