Abstract
We consider the problem of recovering block-sparse signals with unknown boundaries. Such signals arise naturally in various applications. Recent literature introduced a pattern-coupled or clustered Gaussian prior, in which each coefficient involves its own hyperparameter as well as its immediate neighbors' hyperparameters. Some methods use a hierarchical distribution making the solution vulnerable to the parameter choice. Besides, these methods mainly rely on the expectation-maximization (EM) algorithm and either require a suboptimal solution or an approximation of the hyperparameters. To address these difficulties, we propose to solve the pattern coupling problem via fixed point iterations instead of the EM algorithm. The proposed algorithm does not require any further assumptions on the hyperparameters and provides a simple update rule for the hyperparameters. Although the fixed point iterations method is an empirical strategy, it provides a fast convergence rate. The proposed algorithm is tested on a simple direction of arrival (DOA) and amplitude estimation problem. From our simulations, we see that the proposed method achieves similar reconstruction results with the state-of-the-art; however, the proposed method is faster than the existing counterparts.
Original language | English |
---|---|
Title of host publication | Proceedings of the 2023 57th Asilomar Conference on Signals, Systems, and Computers |
Publisher | IEEE |
Pages | 1303-1307 |
Number of pages | 5 |
ISBN (Electronic) | 979-8-3503-2574-4 |
ISBN (Print) | 979-8-3503-2575-1 |
DOIs | |
Publication status | Published - 2023 |
Event | 2023 Asilomar Conference on Signals, Systems, and Computers - Pacific Grove, United States Duration: 29 Oct 2023 → 1 Nov 2023 |
Conference
Conference | 2023 Asilomar Conference on Signals, Systems, and Computers |
---|---|
Abbreviated title | ACSSC 2023 |
Country/Territory | United States |
City | Pacific Grove |
Period | 29/10/23 → 1/11/23 |
Bibliographical note
Green Open Access added to TU Delft Institutional Repository 'You share, we take care!' - Taverne project https://www.openaccess.nl/en/you-share-we-take-careOtherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.
Keywords
- sparse Bayesian learning
- block sparse signals
- fixed point iterations
- pattern coupling