A High-Throughput Subspace Pursuit Processor for ECG Recovery in Compressed Sensing Using Square-Root-Free MGS QR Decomposition

Sensor nodes in wireless body-area networks are expected to use compressed sensing (CS) for ultralow-power consumption. For signal recovery in CS systems, compared with the software implementation, the hardware processor has superiority on power, real-time ability, and so on. Although orthogonal matching pursuit (OMP) is conventionally selected in hardware implementations, its low recovery performance is still a problem. We propose to use the subspace pursuit (SP) algorithm that can achieve higher recovery performance with less iteration to implement the hardware processor. First, a new SP by a square-root-free method and a computation reuse scheme are proposed to reduce the SP’s complexity. Then, the fully paralleled architectures and elaborate pipelines are proposed to improve the processing throughput. Finally, we give an FPGA implementation prototype. Simulations compared with OMP show that the traditional SP can improve more than 8 dB on recovery performance or it can reduce more than 31% sampling paths in sensor nodes. These benefits are kept by our proposed SP because our improvement does not modify the algorithmic nature. Nevertheless, the significant complexity reduction is achieved by our proposal. The proposed SP processor can recover ECG signals with 28.2 dB and achieve the processing throughput of 44.4 K vectors/s.