engleski

A Realization of Adaptive Compressive Sensing System

There are numerous papers analyzing the theoretical background of compressive sensing (CS), but a few practical implementations exist due to the realization complexity. In this paper, a realization of a system for CS of analog one-dimensional signals in a shift- invariant subspace is proposed. We use statistical compressive sensing that allows efficient sampling of signals that follow some statistical distribution. In fact, principal component analysis is used to obtain a small number of principal components in which the majority of signal’s energy is compacted. The signal is demodulated using the obtained principal components, which provides optimal measurement procedure in the mean-square error sense. Orthonormality allows for a linear reconstruction which is perfectly suited for a real-time embedded system implementation. Due to hardware realization constraints, the adaptive demodulating signal is quantized. We provide qualitative and quantitative validations based on simulations of the proposed CS system realization. Furthermore, we validate performance of the system when additive white Gaussian noise is present in the measurements.

sparse representation ; statistical compressive sensing ; inverse problems ; signal reconstruction

nije evidentirano