Energy Efficient Compressed Sensing of Bio-Signals with Sparse Binary Matrix

Abstract

A variety of constraints exist on the design of a wireless tele-monitoring system of bio-signals. In order to reduce on-chip energy consumption or extend sensor life, recorded signals are usually compressed before transmission. Compressed sensing (CS) is promising to as a low-power compression framework to improve energy efficiency. Its performance is largely determined by the characteristic of sensing matrix. The zero-one binary sensing matrices can be adopted for its relative low complexity and competitive performance. However, most of the bioelectric signals is non sparse in the time domain and also non sparse in transformed domains. Thus, it is extremely difficult for conventional CS algorithms to recover bioelectric signals with the quality that satisfies the requirements of applications. Due to better recovery quality, the block sparse Bayesian learning (BSBL) can be provided for this reconstruction of bioelectric signals. Considering the complete signal chain from acquisition to reconstruction, experiments on fetal ECG signals and epilepsy EEG signals showed that the BSBL algorithm with sparse binary matrix has good balance between speed and data reconstruction fidelity. Focusing on the encoding stage and its supporting circuit functions, it has also been shown by analysis that the CS based data compression with spare binary matrix can largely save energy and on-chip computing resources.