A Multi-Task Bayesian Algorithm for online Compressed Sensing of Streaming Signals

Abstract

A Bayesian compressed sensing algorithm is proposed for the multi-task online recovery of streaming signals from compressed measurements. In this algorithm, a multi-task sliding window based on lapped orthogonal transform is established for the online observation of streaming signals, and statistical information obtained from previous time steps is used to suppress the error accumulation in the process of dynamic recovery, therefore the blocking effects that appeared in traditional methods is eliminated effectively. Experiments conducted on historical predicted data of evaporation duct height show that the proposed algorithm has significantly higher reconstruction accuracy, success rate and operation efficiency compared with its DCT version, and this advantage is more obvious under conditions of higher Signal-to-Noise Ratios (SNR) or larger measurement numbers.