Analog-to-Information Approach Using Adaptive Compressive Sampling and Nonlinear Affine Transformations. Analog-to-Information GMR-UW Collaboration; Final technical rept. Nov 2006-May 2008

摘要

The collaborative effort between GMR Research & Technology and the University of Wisconsin - Madison aimed at finding novel approaches in reduced rate representation and sampling. The effort concentrated on exploring data- adaptive techniques and non-adaptive structured sensing, as well as comparing randomized projection based approaches to nonlinear affine (NoLAff) approaches. The approaches explored in this work share a common theme of improving upon purely random encoding. Adaptive sampling utilizes partial information from previous observations to focus subsequent observations onto relevant signal components, and provides significant improvements in the measurement signal-to- noise ratio. Toeplitz structured matrices are effective sensing structures that are efficient to generate and implement in practice. The acquisition process of NoLAff sampling can be approximately modeled using special deterministic sensing matrices, and the inherent structure can be leveraged to reduce decoding from convex optimization to hypothesis testing, which is efficient both computationally and from a data rate perspective.