Khanna, Saurabh and Murthy, Chandra R (2017) Renyi Divergence Based Covariance Matching Pursuit of Joint Sparse Support. In: 18th IEEE International Workshop on Signal Processing Advances for Wireless Communications (SPAWC), JUL 03-06, 2017, Sapporo, JAPAN.
PDF
Ieee_Int_Wor_SPAWC_2017.pdf - Published Version Restricted to Registered users only Download (179kB) | Request a copy |
Abstract
In this work, we consider the joint sparse support recovery problem where the goal is to recover the common support of multiple joint sparse vectors from their compressive, linear measurements. We propose a Renyi Divergence based Covariance Matching Pursuit (RD-CMP) algorithm which recovers the common support of the joint sparse signals as the hyperparameters of a joint sparsity inducing Gaussian signal prior. The support hyperparameters are learned as a set variable by solving a reverse information projection problem based on the alpha-Renyi information divergence. We show that the alpha-Renyi divergence objective can be expressed as a difference of two submodular functions, and propose an iterative majorization-minimization procedure to minimize the objective, with each iteration involving a greedy optimization. Through simulations, we demonstrate that the proposed RD-CMP algorithm is capable of recovering k-sparse support from fewer than k measurements per signal. Compared with existing covariance matching based joint sparse support recovery methods, RD-CMP is empirically shown to be several times faster in execution.
Item Type: | Conference Proceedings |
---|---|
Series.: | IEEE International Workshop on Signal Processing Advances in Wireless Communications |
Publisher: | IEEE, 345 E 47TH ST, NEW YORK, NY 10017 USA |
Additional Information: | Copy right for the article belong to IEEE, 345 E 47TH ST, NEW YORK, NY 10017 USA |
Department/Centre: | Division of Electrical Sciences > Electrical Communication Engineering |
Date Deposited: | 13 Apr 2018 19:57 |
Last Modified: | 13 Apr 2018 19:57 |
URI: | http://eprints.iisc.ac.in/id/eprint/59549 |
Actions (login required)
View Item |