Independent Component Analysis by Entropy Maximization (INFOMAX)
Abstract
This thesis explores the "Infomax" method of Independent Component Analysis (ICA) to accomplish blind source separation (BSS). The Infomax method separates unknown source signals from a number of signal mixtures by maximizing the entropy of a transformed set of signal mixtures and is accomplished by performing gradient ascent in MATLAB. The thesis specifically focuses on small numbers of two types of signals: audio signals and simple communications signals (polar non-return to zero signals). The Infomax method is found to be successful and efficient only for small numbers of signals, and improvements to the gradient ascent algorithm should be made for the Infomax algorithm to succeed for more than three signal mixtures. MATLAB implementation code is included as appendices.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2007
- Accession Number
- ADA470123
Entities
People
- Jennie H. Garvey
Organizations
- Naval Postgraduate School