FPGA Enabled ECG Signal Reconstruction based on an Enhanced Orthogonal Matching Pursuit Algorithm

Abstract

The compressed sensing (CS) concept is developed to perform signal collection and compression simultaneously. At a far lower Nyquist rate, the signals are sampled with the help of signal sparsity. In the CS method, the signal is reconstructed from original data by the measured value, changing the signal from high to low dimension over a matrix multiplication process. Moreover, the bio-medical signal processing applications mostly utilize the CS technique to accomplish high bandwidth and low power consumption while performing signal reconstruction. So, in this research, a low area and high throughput CS-based orthogonal matching pursuit (OMP) algorithm is proposed, which is enhanced with sparsity regularization (SR) and an improved Gram Schmidt (IGS) based incremental QR decomposition (QRD) technique. In the index searching block, a sum-of-product (SoP) based multiplication process is utilized to obtain the inner products, and the selected atoms are stored in a regularizer unit. After that, a distributed-arithmetic (DA)-based matrix multiplication circuit is designed to perform the multiplication in the least square problem (LSP)-solving step, which balances the area overhead instigated by the multi-stage pipelining. Matrix decomposition is also performed by utilizing systolic arrays with four new processing elements (PEs). The publicly available MIT-BIH arrhythmia dataset is utilized for the reconstruction process, and the whole implementation is done using MATLAB and Xilinx Verilog coding. From the result analysis, it is shown that the proposed method takes 152 time to reconstruct an original signal from the compressed signal with less error rate. In addition, it consumes 0.059W dynamic power and accomplishes higher throughput as 0.0814506 Tera Operations per Second (TOPS).