Structured Convolutional Compressed Sensing Based on Deterministic Subsamplers

摘要

As a novel method to process sparse signals, compressed sensing (CS) attracts huge attention in recent years. The sensing matrix in traditional CS is completely Gaussian random, resulting in high complexity and expensive implementation cost. To solve the problem of randomness reduction, this paper investigates and applies some partially deterministic sensing matrices like Golay families, and realizes them based on several different recovery algorithms. The core procedure of structured convolutional CS is orderly selecting sparsified samples, usually in various frequency domains, then reconstructing using the same deterministic sensing matrices based on several popular recovery algorithms. Conclusions can be drawn that with these structured sensing matrices, we can get better reconstruction quality and more stable performance with less time cost.