ABSTRACT
Compressive Sensing (CS) has a variety of potential applications, and is currently at the forefront of signal processing research domain. It states that the signals that have a sparse or compressible representation on a dictionary can be recovered from far less linear measurements than those required by the classical Shannon-Nyquist theorem [1]. The measurements used in CS correspond to linear projections obtained by a projection matrix. Existing research has shown that nonadaptive random sensing matrices with Gaussian or Bernoulli distributions satisfy the fundamental theoretical requirements of CS [2]. On the other hand, a projection matrix that is optimally or adaptively designed based on prior measurements can further improve the reconstruction accuracy or take fewer samples. The idea behind adaptive compressive sensing is that one should focus sensing power toward the non-zero components of the spare signal in order to increase the signal-to-noise ratio (SNR). The potential advantages of an adaptive projection scheme are demonstrated in [3]-[5].
