Co-Design of Sparse Coding and Dictionary Learning for Real-Time Physiological Signals Monitoring

Compressive sensing (CS) is a novel technique to reduce overall transmission power in wireless sensors. For physiological signals telemonitoring of wearable devices, chip area and power efficiency need to be considered simultaneously. There are many prior studies aim to develop algorithms that applied to CS reconstruction chips with reconfigurable architecture. However, representative dictionaries are also important when these CS reconstruction chips are verified in real-time physiological signals monitoring tasks. That is, a more representative dictionary can not only enhance the reconstruction performance of these chips but also alleviate memory overhead. In this paper, we apply the concept of co-design between sparse coding algorithms and learned dictionaries. We also explore the representativeness and compatibility of each learned dictionary. In addition, the computational complexity of each reconstruction algorithm is provided through simulations. Our results show that the dictionaries trained by fast iterative shrinkage-thresholding algorithm (FISTA) are more representative according to the quality of reconstruction for physiological signals monitoring. Besides, FISTA reduces more than 90% of the computational time compared with other hardware-friendly reconstruction algorithms.