FISTA를 이용한 압축 센싱 희소 신호 복원: 임베디드용 C 구현
FISTA-Based Sparse Signal Recovery in Compressive Sensing: A Lightweight C Implementation
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This study presents a hardware-efficient implementation of both sparse signal acquisition and the Fast Iterative Shrinkage-Thresholding Algorithm (FISTA) for signal reconstruction, fully integrated on a single resource-constrained embedded system. The entire compressive sensing pipeline from sub-Nyquist sampling to sparse recovery was realized on an STM32F407 microcontroller. ECG signals were acquired using the AD8232 analog front-end and digitized ADC channels of the STM32F407 microcontroller, which were configured for random under sampling at 30–40% of the Nyquist rate. The sparse acquisition process leverages the inherent incoherence between the Discrete Fourier Transform (DFT) basis used as the sparsifying domain and the randomly selected rows of the identity matrix, which serve as the measurement matrix A. This property of high incoherence allows for accurate signal recovery without the need for additional random demodulator (RD) hardware, thereby making the system more efficient and suitable for low-power biomedical applications. Experimental results demonstrate that with 30% of the original signal measurements (M = 300), 30 iterations, and a regularization parameter λ = 0.0005, the system successfully reconstructed the signal with a PSNR of 38dB when compared to the reference signal obtained via Nyquist uniform sampling. These findings confirm the feasibility of deploying FISTA-based sparse recovery on low-power microcontroller for sparse signal processing applications.
Abstract
This study presents a hardware-efficient implementation of both sparse signal acquisition and the Fast Iterative Shrinkage-Thresholding Algorithm (FISTA) for signal reconstruction, fully integrated on a single resource-constrained embedded system. The entire compressive sensing pipeline from sub-Nyquist sampling to sparse recovery was realized on an STM32F407 microcontroller. ECG signals were acquired using the AD8232 analog front-end and digitized ADC channels of the STM32F407 microcontroller, which were configured for random under sampling at 30–40% of the Nyquist rate. The sparse acquisition process leverages the inherent incoherence between the Discrete Fourier Transform (DFT) basis used as the sparsifying domain and the randomly selected rows of the identity matrix, which serve as the measurement matrix A. This property of high incoherence allows for accurate signal recovery without the need for additional random demodulator (RD) hardware, thereby making the system more efficient and suitable for low-power biomedical applications. Experimental results demonstrate that with 30% of the original signal measurements (M = 300), 30 iterations, and a regularization parameter λ = 0.0005, the system successfully reconstructed the signal with a PSNR of 38dB when compared to the reference signal obtained via Nyquist uniform sampling. These findings confirm the feasibility of deploying FISTA-based sparse recovery on low-power microcontroller for sparse signal processing applications.