Expectation Propagation on Factor Graphs Based on Matrix Decomposition

In the context of the Gaussian linear model, recent works have studied factor graph modification using QR decomposition that enables the derivation of scalar Expectation Propagation (EP) based detectors. In this paper, we investigate on new factor graph representations induced by the use of the Golu...

Full description

Saved in:
Bibliographic Details
Main Authors: Mekhiche, Adam, Cipriano, Antonio Maria, Poulliat, Charly
Format: Conference Proceeding
Language:English
Subjects:
Complexity theory
Detectors
Digital communication
Expectation Propagation
Factor Graph
Heuristic algorithms
Inference algorithms
Linear Model
Matrix Decomposition
Message Passing
Signal processing
Signal processing algorithms
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In the context of the Gaussian linear model, recent works have studied factor graph modification using QR decomposition that enables the derivation of scalar Expectation Propagation (EP) based detectors. In this paper, we investigate on new factor graph representations induced by the use of the Golub-Kahan bi-diagonal Decomposition (GKD) and of the Singular Value Decomposition (SVD). New EP messages induced by the GKD or SVD underlying graphs are derived, that can be both scalar or vector messages. Complexity and performance of the resulting algorithms are studied for digital communications applications.
ISSN:2379-190X
DOI:10.1109/ICASSP49357.2023.10094970