Estimating Sparse Signals Using Integrated Wideband Dictionaries

Maksim Butsenko, Johan Svard, Andreas Jakobsson

Research output: Contribution to journalArticlepeer-review

Abstract

In this paper, we introduce a wideband dictionary framework for estimating sparse signals. By formulating integrated dictionary elements spanning bands of the considered parameter space, one may efficiently find and discard large parts of the parameter space not active in the signal. After each iteration, the zero-valued parts of the dictionary may be discarded to allow a refined dictionary to be formed around the active elements, resulting in a zoomed dictionary to be used in the following iterations. Implementing this scheme allows for more accurate estimates, at a much lower computational cost, as compared to directly forming a larger dictionary spanning the whole parameter space or performing a zooming procedure using standard dictionary elements. Different from traditional dictionaries, the wideband dictionary allows for the use of dictionaries with fewer elements than the number of available samples without loss of resolution. The technique may be used on both one- and multi-dimensional signals, and may be exploited to refine several traditional sparse estimators, here illustrated with the LASSO and the SPICE estimators. Numerical examples illustrate the improved performance.

Original languageEnglish
Pages (from-to)4170-4181
JournalIEEE Transactions on Signal Processing
Volume66
Issue number16
Early online date2018 May 14
DOIs
Publication statusPublished - 2018

Subject classification (UKÄ)

  • Signal Processing

Keywords

  • Computational complexity
  • Dictionaries
  • Estimation
  • Frequency estimation
  • Optimization
  • Tensile stress
  • Wideband

Fingerprint

Dive into the research topics of 'Estimating Sparse Signals Using Integrated Wideband Dictionaries'. Together they form a unique fingerprint.

Cite this