Exact recovery of structured block-sparse signals with model-aware orthogonal matching pursuit

Thomas Wiese, Lorenz Weiland, Wolfgang Utschick

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

3 Scopus citations

Abstract

We provide recovery guarantees for the model-aware OMP algorithm, which is an extension of the known block-OMP algorithm. The new algorithm exploits additional structural information about the unknown signal. This leads to successful recovery under weaker conditions on the sensing matrix, namely, a restricted isometry property with respect to structured signals, only. A possible application of this algorithm is for channel estimation in wireless communication systems where nonlinear delay and angle estimation problems need to be solved. For this application, our method only discretizes the delay parameter and not the angular parameter. As such, it is situated between very complex nonlinear maximum-likelihood estimation and very efficient compressive sensing based methods that discretize all parameters.

Original languageEnglish
Title of host publicationSPAWC 2016 - 17th IEEE International Workshop on Signal Processing Advances in Wireless Communications
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781509017492
DOIs
StatePublished - 9 Aug 2016
Event17th IEEE International Workshop on Signal Processing Advances in Wireless Communications, SPAWC 2016 - Edinburgh, United Kingdom
Duration: 3 Jul 20166 Jul 2016

Publication series

NameIEEE Workshop on Signal Processing Advances in Wireless Communications, SPAWC
Volume2016-August

Conference

Conference17th IEEE International Workshop on Signal Processing Advances in Wireless Communications, SPAWC 2016
Country/TerritoryUnited Kingdom
CityEdinburgh
Period3/07/166/07/16

Keywords

  • Union of subspaces
  • model-aware compressive sensing
  • model-based compressive sensing
  • orthogonal matching pursuit
  • restricted isometry

Fingerprint

Dive into the research topics of 'Exact recovery of structured block-sparse signals with model-aware orthogonal matching pursuit'. Together they form a unique fingerprint.

Cite this