Bayes-Optimal Convolutional AMP
Information Theory
2021-05-11 v2 math.IT
Abstract
To improve the convergence property of approximate message-passing (AMP), convolutional AMP (CAMP) has been proposed. CAMP replaces the Onsager correction in AMP with a convolution of messages in all preceding iterations while it uses the same low-complexity matched filter (MF) as AMP. This paper derives state evolution (SE) equations to design the Bayes-optimal denoiser in CAMP. Numerical results imply that CAMP with the Bayes-optimal denoiser--called Bayes-optimal CAMP--can achieve the Bayes-optimal performance for right-orthogonally invariant sensing matrices with low-to-moderate condition numbers.
Keywords
Cite
@article{arxiv.2101.11185,
title = {Bayes-Optimal Convolutional AMP},
author = {Keigo Takeuchi},
journal= {arXiv preprint arXiv:2101.11185},
year = {2021}
}
Comments
accepted for presentation in ISIT2021