Sparse Reconstruction via The Reed-Muller Sieve
Abstract
This paper introduces the Reed Muller Sieve, a deterministic measurement matrix for compressed sensing. The columns of this matrix are obtained by exponentiating codewords in the quaternary second order Reed Muller code of length . For , the Reed Muller Sieve improves upon prior methods for identifying the support of a -sparse vector by removing the requirement that the signal entries be independent. The Sieve also enables local detection; an algorithm is presented with complexity that detects the presence or absence of a signal at any given position in the data domain without explicitly reconstructing the entire signal. Reconstruction is shown to be resilient to noise in both the measurement and data domains; the error bounds derived in this paper are tighter than the bounds arising from random ensembles and the bounds arising from expander-based ensembles.
Cite
@article{arxiv.1004.2926,
title = {Sparse Reconstruction via The Reed-Muller Sieve},
author = {Robert Calderbank and Stephen Howard and Sina Jafarpour},
journal= {arXiv preprint arXiv:1004.2926},
year = {2010}
}
Comments
To appear in ISIT 2010